Proceedings of the MODFLOW and More 2017 Conference

May 21 - 24, 2017

Organizing Committee

Reed Maxwell, IGWMC; Colorado School of Mines

Mary Hill, University of Kansas
Chunmiao Zheng, SUSTC; University of Alabama
Matt Tonkin, S.S. Papadopulos & Associates
 Table of Contents

Section 1: Natural and Managed Aquifer Recharge

Water Banking on the Kern River Alluvial Fan 16

David M. Bean, J. Parker, and M. Anderson

Surface Water Accounting Model, SWAcMOD. A MODFLOW-USG Pre 17

Processor for Recharge and Surface Water Systems
Alastair Black

Horizontal Flow Models that are Not 23

Henk Haitjema

Estimating Evapotranspiration and Recharge Rates Using a Remote 24

Sensing Algorithm
Alan Laase

Analysis of Infrequent (Quasi-Decadal) Large Groundwater Recharge 25

Events: A Case Study for Northern Utah
Melissa D. Masbruch, C.A. Rumsey, S. Gangopadhyay, D.D. Susong,
and T. Pruitt

Solving the Water Supply Puzzle: MODFLOW and Uncertainty in the Context 26
of Mitigated Water Rights
Aaron Pruitt, P. Bannister, and T. Flynn

Analysis of Aquifer Recharge by Environmental Flows in a Semiarid Zone 27

Using Diffusion Wave Package for MODFLOW
Eliana Rodriguez, M. Shanafield, and J. Ramirez-Hernandez

Vertically Integrated Flow for Complex Aquifer Systems as a Modeling Tool 28

Otto D.L. Strack

Simulating Managed Aquifer Recharge and San Pedro River (Arizona) 29

Baseflow Augmentation Using MODFLOW-NWT
Robert Wyckoff, J. Weaver, and J. Wright

Section 2: Case Studies of Depleted Aquifers

Equation for Stream Accretion Timing from a Line Source Oriented 30

Perpendicular to a Stream
Jacob Bauer

Case Study: Surface-Groundwater Modelling to Guide the Revocation or 31

Reduction of Abstraction Licences in the UK’s Primary Sandstone Aquifer
Alastair Black
Implementation of the Subsidence and Aquifer-System Compaction 36
Package to Model Land Subsidence in Virginia’s Chesapeake Bay Area
Using a Hydrogeologic-Unit Flow Based SEAWAT Model
Thomas Griffiths, S. Kudlas, N. Jones, and A. Lemon

Simulations Supporting Management of Saltwater Movement in the Aquifers 37

of the Baton Rouge Area, Louisiana
Charles Heywood, M. Lindaman, and J. Lovelace

Evaluating Potential Groundwater Pumping Effects in Pahrump Valley on 38

Water Resources in the Amargosa Wild and Scenic River, Nevada and
California
Tracie Jackson and K. Halford

Living Groundwater Models: Case Study for a Continually Updated 39

Groundwater Models
Alan M. Lemon, N. Jones, S. Kudlas, T. Griffiths, and H. Tran

Preliminary-Assessment and Upgrade of a Groundwater Flow Model of the 40

Seacoast Bedrock Aquifer, New Hampshire
Thomas J. Mack

Numerical Modeling of the Agricultural-Hydrologic System in Punjab, India 45

Madeline Nyblade, T. Russo, L. Zikatanov, and K. Zipp

Grid Generation and Model Calibration for Complex Fluvial Aquifer System 46
Including Geological Faults (Cover Figure)
Hai V. Pham, F. T-C. Tsai, and K.F. Pohlmann

Future Baseline Modeling for the Sustainable Groundwater Management 52

Act: Prediction Uncertainty Resulting from Uncertainty in Climate Change
and Land Use Change
Andrew Rich

Addressing Non-linear Model Response in the Application of the Republican 53

River Compact Groundwater Model to Estimate Relative Contributions to
Streamflow Depletion in a Highly Stressed Aquifer
James C. Schneider, D. Ahlfield, C. Spalding, and B. Dunnigan

Conjunctive Use Analysis with the Enhanced Snake Plain Aquifer Model 54
(ESPAM)
Sophia C. Sigstedt

Assessment of Management Strategies to Help Mitigate Aquifer Depletion 55

and Subsidence in the Bicycle Basin, Fort Irwin National Training Center,
California
Linda R. Woolfenden and J. Densmore
Section 3: Sustainable Groundwater Quality: Challenges and Tradeoffs

How Watershed Characteristics Affect Transit Time Distributions 56

Daniel B. Abrams and H.M. Haitjema

Short-term, Large-scale, Multi-objective Optimization for Water-Resources 57

Management, Santa Barbara, CA
Tracy Nishiwaka, Z. Stanko, and S. Paulinski

Influences of a Weir Construction on Freshwater Resources of a Coastal 58

Aquifer in Germany Under Sea Level Rise
Jie Yang, T. Graf, and T. Ptak

Section 4: Past, Present, and Future of Mining and Groundwater

Fast Assessment of Pore Pressures and Inflows in Open Pit Slopes Using 59
Smart Models
Eduardo De Sousa

Simulation of High Quality Wetland Protection Measures Concurrent with 60

Surface Mine Dewatering Utilizing GMS-MODFLOW
Christopher J. DeMarco, M.S. Weikel, and L.F. Vittorio

Comparison on Groundwater Model Simulations of Open-Pit Mining Using 64

Three Numerical Codes: MODFLOW-USG, MODFLOW-SURFACT, and
MINEDW
Alan Jang, H. Liu, J. Xiang, and S. Meyerhoff

Assessing the Cumulative Effects of Groundwater Withdrawals for Oils 66

Sands Production on a Watershed Scale
Dirk Kassenaar, E.J. Wexler, P.J. Thompson, and M.G.S. Takeda

Groundwater Modeling to Support Wetland Restoration of a Former Peat 67

Mine
Seann McClure and P. Bannister

Use of MODFLOW-USG and Connected Linear Networks (CLN) to Represent 68

Deformation and Fracturing above a Longwall Mine
Will Minchin and S. Brown

Bridging the Gap: Coupling of Geotechnical and Groundwater Models in the 69

Mining Industry
Christopher Pantano, D. Edington, and G. Meza-Cuadra

Past, Present, and Future Predictions of Mine Dewatering Rates in the 70

Pilbara Region of Western Australia
Gareth Price
Section 5: Past, Present, and Future of MODFLOW

Upgrading Your Model to MODFLOW 6 74

Ned Banta

Integration of Solute Transport and Water Flow Models for Unsaturated and 75
Saturated Soil Zones Using the HYDRUS Package for MODFLOW and
MT3DMS
Sahila Beegum, J. Simunek, A. Szymkiewicz, K.P. Sudheer, and
I.M. Nambi

MODFLOW-OWHM v2: New Features and Improvements; The Next 76

Generation of MODFLOW Conjunctive Use Simulation
Scott E. Boyce, R.T. Hanson, I. Ferguson, T. Reimann, W. Henson,
S. Mehl, W. Schmid, S. Leake, and T. Maddock

Free MODFLOW-based Web Modeling Framework for Planning and 80

Assessment of Managed Aquifer Recharge Schemes
Jana Glass, C. Stefan, and R. Junghanns

Use of the Advanced Packages and Demand-Based Boundary Flows in the 85

MODFLOW 6 Groundwater Flow Model
Joseph D. Hughes, C. Langevin, S. Panday, E. Banta, A. Provost, and
R. Niswonger

MODFLOW 6: An Object-Oriented Version of the U.S. Geological 86

Survey’s MODFLOW Model
Christian Langevin, J.D. Hughes, E.R. Banta, A.M. Provost, R.G.
Niswonger, and S. Panday

Rescaling Riverbed Conductance to Improve Approximations of 87

Stream/Aquifer Leakage
Steffen Mehl, H. Morel-Seytoux, and C. Miller

Application and Challenges of MODFLOW-USG to Simulate Tunnel Inflows 88

Jaco Nel, A. Johnstone, and N. Rapantova

Modeling Water Supply Effects on the Distribution and Consumption of 89

Water in Developed River Basins
Richard Niswonger, E. Morway, E. Triana, and W. Kitlasten

A Hydraulic Head Formulation for Density Dependent Flow and Transport 90

Sorab Panday, C. Langevin, and A.M. Provost

The “XT3D” Option for Simulating Fully Three-dimensional Anisotropy on 91

Regular and Irregular MODFLOW 6 Grids
Alden Provost, C. Langevin, and J. Hughes
 Generation of Source Reliable Output Diagrams of Adited Groundwater 92
Sources Using Models: An Example from the UK
Azucena Rodriguez Yebra, A. Hughes, A. Butler, D. Peach, and
C. Jackson

Borehole-scale Simulations from a Regional Groundwater Model Using 93

Connected Linear Networks and Ghost Node Correction
William R. Witterick

Section 6: GUIs and You

FREEWAT, a HORIZON 2020 Project to Build Open Source Tools for Water 94
Management: a European Perspective
Laura Foglia, R. Rossetto, I. Borsi, M. Cannata, S.W. Mehl, and
E. Vasquez-Sunez

PEST.cloud: Cloud-based Model Calibration 95

Douglas A. Hayes, J. Doherty, and C. Muffels

AlgoCompute: Large-scale Calibration and Uncertainty Analysis Made Easy 96

in the Cloud
Damian P. Merrick

StreamExtractor: A New Tool to Help Generate Topographically Consistent 101

Input for the Stream and Streamflow Routing Packages in MODFLOW
Richard B. Winston and J.P. Pope

Section 7: Parameterization, Sensitivity Analysis, and Uncertainty

Tools for the Hands-on MODFLOW Modeler 102

Daniel B. Abrams and G.S. Roadcap

Capture Functions in a Coupled Surface Water-Groundwater Model Using 103

MODFLOW-ADJOINT Code
Ramarao S. Banda, V.A. Kelley, and M. Lavenue

Modelling Uncertainty Analysis for Contaminant Risk Assessment 104

Martinus Brouwers, P. Martin, and D. Abbey

Parameter Values in the DRAIN Package for a “Groundwater Weir” Near a 105
Canal in Reconstruction
H. Kooi, J.H. Hoogewoud, and Wim J. de Lange

PEST HP: A Version of PEST Optimized for Highly Parallelized Computing 107
Environments
John Doherty

Quantifying Predictive Uncertainty of Groundwater/Surface-water Exchange 108

in the Lower San Antonio River Basin, Texas
Linzy K. Foster and J. White
Data Drought to Data Surplus – the Value of Traditional Groundwater Data 109
Sets and Extensive Hydrogeophysics to Make Model Predictions Including
Uncertainty
Phil Hayes, C. Nicol, and V. Puech

Use of Transmissivity Observations with MODFLOW and PEST: Application 114

to the Floridian Aquifer System
Eve L. Kuniansky, J. Bellino, and K. Halford

Promise and Peril of Calibrating a High Resolution Gridded Regional 115

Watershed Model
Andrew Leaf, R.J. Hunt, D.E. Christiansen, and A.E. Haj

Comparison of Predictive Uncertainty Methods to Quantify Iron Ore 116

Dewatering Impacts
Neil Manewell and V. Shapoori

Efficient Quantification of Uncertainty in Integrated Surface and Subsurface 121

Hydrologic Simulations
Killian L. Miller, Y-J. Park, S. Berg, and E. Sudicky

Soil and Geologic Controls on Recharge and Groundwater Flow Response 122
to Climate Change: A Case Study of the Yakima River Basin
Tung T. Nguyen and J.C. Adam

Generating Initial Fresh/Saline Groundwater Distributions for a 3D 123

MODFLOW SWI Model Using 2D Cross-sectional Seawat Model as Proxy
Philip Nienhuis and P. Kamps

The Ensemble Kalman Filter for Groundwater Plume Characterization: 128

A Pilot Study
James L. Ross and P.F. Andersen

Identifying Controls on Carbon Exchanges in High Altitude Headwaters to 132

Improve Representation in Earth System Models
Anna Ryken, R. Maxwell, D. Gochis, and K. Williams

Demystifying Information and Fisher Information: Observation Data Reveals 133

the Good and the Bad of Our Models
Judith Schenk, E. Poeter, and W. Navidi

Global Sensitivity Analysis for Supporting Long-Term Monitoring During 138

Sustainable Remediation
Haruko M. Wainwright

The Ensemble Smoother: Jacobians for Nothing and Uncertainty for Free 139
Jeremy T. White

PESTPP-OPT: Model-independent Groundwater-management Optimization 140

Under Uncertainty
Jeremy T. White
 Automated Transient Element Activation Plug-in for FEFLOW 141
William L. Wingle and P. Sinton

ZoneTable Plug-in for FEFLOW 145

William L. Wingle and P. Sinton

Improving Parameterization of Top Layer for Groundwater Models by Means 149

of a 3D Geologic Voxel Model
Willem J. Zaadnoordijk and J. Gunnink

Section 8: Using Models to Identify Data Needs and Guide Data

Acquisition

The Evolution of a Groundwater Flow and Transport Model Over Two 150
Decades of Updates and Applications
Peter Andersen and J. Ross

Importance of Characterizing River-aquifer Interaction in Building 154

Groundwater Flow and Transport Model
S.M. Helalur Rashid, J. Ewing, T. Jones, and G. Ruskauff

Physical Heterogeneity Controls on Geochemical Reaction Rates: Insights 155

from Reactive Transport Models
Alexis Navarre-Sitchler, H. Jung, J. Beisman, and R. Maxwell

Mexico Valley Aquifer’s Transition from a University Simulation Model to 156

MODFLOW for Water Resource Assessment and Subsidence Modeling
Adriana Palma Nava, F. Gonzalez Villarreal, R.T. Hanson, and S.E. Boyce

One Model to Rule Them All: The Illinois Groundwater Model 157
George S. Roadcap, D. Abrams, D. Hadley, and D. Mannix

Section 9: Simulation Code Verification, Benchmarking, and

Intercomparisons

Does MODFLOW-USG/AlgoMesh Give the Same Results as 158

MODFLOW-SURFACT?
Noel P. Merrick and D. Merrick

Integrated Hydrology Model Intercomparison Using High-Resolution Data at 161

the East River Colorado
David Moulton, E. Siirila-Woodburn, D. Dwivedi, C. Steefel, D. Svyatskiy,
R. Maxwell, and L. Condon

mod-PATH3DU Version 2: New Features, Verification and Intercomparisons 162

Christopher Muffels

Benchmarking and Inter-comparison of MODFLOW-USG’s Block Centred 163

Transport (BCT) Package
Chris Nicol
 Building a Global Groundwater Model – Case Study: Comparison to the 169
Central Valley Hydrologic Model
Robert Reinecke, L. Foglia, S. Mehl, and P. Döll

Effective Utilization of Parallel Resources 170

Willem Schreuder

Intercomparison of Hybrid Multiscale Methods in Simulating Subsurface 171

Flow and Reactive Transport
Xiaofan Yang, Y. Mehmani, D. Barajas-Solano, M.T. Balhoff,
A.M. Tartakovsky, and T.D. Scheibs

Application of the Partition Stress Boundary Capability with 172

MODFLOW-NWT in a Large-Scale Regional Groundwater Flow Model
Kurt Zeiler, J. Weaver, A. Moore, and M. Halstead

Performance Analysis of the Chi-MD Matrix Solver Package for MODFLOW 176
Yiding Zhang, M. Ibaraki, R.G. Niswonger, and S. Panday

Section 10: Simulating the Agriculture-Groundwater Connection

Legal and Hydrologic Considerations to Curtail Agricultural Pumping 181

During Extended Drought
Rosemary Carroll, G. Pohll, J. Benedict, and R. Felling

Assessing the Impact of Groundwater Pumping on the Integrated 182

Hydrologic Cycle
Laura E. Condon and R.M. Maxwell

Stream-Aquifer Interaction in a Regional Alluvium Aquifer 183

John E. Ewing and M. Jigmond

Simulating Water Availability and Sustainability in California’s Central Valley 189

Claudia C. Faunt and J. Traum

Imagining a Maximum Future Irrigation Footprint to Evaluate Water 190

Resources in a Wisconsin County
Michael N. Fienen, P. Juckem, M. Parsen, M. Gotkowitz, S. Ebel, and
D. Masterpole

Forecasting Nitrate in Agricultural Groundwater 191

Christopher Green and L. Liao

Examples of Integrated Modeling of Agricultural Conjuctive-Use from the 192

Lower Rio Grande Valley, New Mexico and Texas to the Salinas Valley,
California
Randall T. Hanson, A.B. Ritchie, S.E. Boyce, A. Galanter, and W. Henson

Drought on the North American High Plains: Modeling the Effects of 193
Vegetation, Temperature, and Rainfall Perturbations on Regional Hydrology
Annette Hein and R.M. Maxwell
 A Food-Energy-Water Calculator, with Initial Application to Western Kansas 194
Mary C. Hill, R.D. Miller, D. Rogers, and J. Roundy

Improving Simulation and Understanding of Third-Party Impacts Resulting 195

from Water-Right Transfers
Eric D. Morway, R. Niswonger, and E. Triana

Development of a Wellfield Operating Plan for a Vineyard in Peru Using 196

MODFLOW’s Groundwater Management Process: A Case Study
Christopher Peters

A New Model to Assess Groundwater Availability of the Northern High 197

Plains Aquifer
Steven M. Peterson and J. Traylor

River Basin-scale Integrated Surface-Subsurface Hydrologic Modeling to 198

Support Agricultural Risk Management
S. Frey, M. Callaghan, H-T. Hwang, Y-J. Park, R. McConnell, S. Berg,
and Edward A. Sudicky

Estimating Future Irrigation and Recharge for the Northern High Plains 199
Aquifer Using Multiple Climate and Land Use Forecasts
Jonathan Traylor

A Custom Modeling Tool to Support Evaluation of the Agriculture- 200

Groundwater Connection
Enrique Triana, C. Garner, G. Pohll, and R. Carroll

Development and Application of an Irrigation Demand Module for the USGS 201
GSFLOW Model
E.J. Wexler, P.J. Thompson, M.G.S. Takeda, S. Malott, S.J. Shifflett, and
J.D.C. Kassenaar

Updates and Improvements to Colorado’s South Platte Decision Support 202

System Alluvial Groundwater Model
Kurt Zeiler, A. Moore, M. Halstead, Z. Wengrovius, J. Weaver, M. Lindburg
and P. Dupre

Section 11: Unsaturated Zone and Multiphase Flow Modeling

Using an Integrated Hydrologic Model to Assess the Ecohydrologic Impacts 206

of Change on a Mountain Headwaters Critical Zone
Caitlin Collins, R. Maxwell, and A. Visser

Coupled Porous-Medium and Free-Flow Across Wavy Surfaces 207

Bo Gao and K. Smits

Evaluating a Mixed-Dimensional Spatial Structure for Integrated Thermal 208

Hydrology Models of Permafrost-Affected Landscapes
Ahmad Jan, S. Painter, and E.T. Coon
 Upscaling Thermal Properties in Heterogeneous Soils 209
Zhen Li and K. Smits

The Soil Moisture Velocity Equation: Reliable, Accurate, and Fast 1-D 210
Unsaturated Zone Flux Calculations
Fred L. Ogden, M.B. Allen, C.C. Douglas, W. Lai, M. Seo, and J. Zhu

Section 12: Advancements in Data Discovery, Retrieval, and Visualization

Novel Modeling Approach for Sites with Complex, Well-Characterized 211

Geology
Jessica Barros, M. Sellwood, A. Wilson, E. Schwartz, and L. Hovey

HydrogeoCloud 3D Groundwater Modeling, a Cloud Computing 212

Hydrogeology Application
Benny Bian and W. Li

Canada’s Oil Sands Innovation Alliance Regional Groundwater Solutions 216

Project for the Southern Athabasca Oil Sands – Spatial Visualization of a
Global Transient Misfit Quality Indicator
Louis-Charles Boutin, M. Claprood, G. MacMillan, M. Brewster, and
A. Walter

HydroShare: A Community Repository for Sharing MODFLOW Models 217

Norm Jones, K. Liang, J. Goodall, M. Morsy, J. Sadler, D. Tarboton,
A. Lemon, and M. Kennard

MFtools: An R Library for Reading MODFLOW and MT3DMS Files 218

Dan Puddephatt

Development of Colorado’s South Platte Decision Support System Alluvial 219

Groundwater Model Using a Data-Centered Approach
Kurt Zeiler, A. Moore, M. Halstead, Z. Wengrovius, J. Weaver,
M. Lindburg, and P. Dupre

Section 13: Effective Presentation of Modeling Results

Groundwater Development and Surface Water: Capture Maps of Streamflow 223

Depletion
Lynette E. Brooks

The Use of Non-Groundwater Software in Visualization and Presentation of 224

Groundwater-Flow Modeling Studies
Brian Clark

Modeling Fresh Submarine Groundwater Discharge Across the Heeia 225

Coastal Shoreline in Hawaii
Kariem A. Ghazal, O.T. Leta, A.I. El-Kadi, and H. Dulai
 Complex Models Require Looking and Comparing 228
Keith Halford

Breaking the Mould: Using Streamlined, Stochastic Groundwater Models to 229

Improve Acceptance by Regulators, Clients, and the General Public
Martin F. Helmke, T. Farney, D. Lynch, P. Girafalco, L. Cole, and
A. Pfahler

Regional Groundwater-Flow Modeling to Assess Tradeoffs and 230

Consequences of Groundwater Development
Howard Reeves

Beyond the Model: Communicating Results to the Decision Maker 231

Blythe Reiha

Using MODFLOW and MODPATH to Access the Changes in Hyporheic 232

Flow During Pre and Post-Restoration Periods
Harsh Vardhan Singh, B. Faulkner, A. Keeley, K. Forshay

Interactive Pathline Simulation 233

Peter Vermeulen

Difficulties in Modeling Systems with Extreme Climate Records: Is a Model 234

of a Groundwater System During a Drought Unrepresentative or
Conservative?
Sophia A. Wolfenden Lee and B. Bader

Section 14: Contaminant Transport Modeling: Developments and

Case-studies

Upscaling the Equilibrium Reactions in the Fractal Radial Domains 235

Ahmed Ali, T. Ginn, and T. Le Borgne

Eliminating the Scale Effect from Reactive Transport: Particle-Tracking 236

Simulation of Batch- to Field-Scale Bioremediation Experiments at the
Schoolcraft, MI Site
David Benson and D. Ding

A Practical Modeling Framework for Non-Fickian Transport and First-Order 237

Reaction
Daniel K. Burnell, S. Hansen, J. Xu, C. Faust, and A. Walter

Strategies for Simulating the Complete Transport Pathways of 238

Atmospherically Dispersed Contaminants
Evan Christianson, D. Dahlstrom, A. Janzen, J. Carter, and R. Wuolo

When do we Really Need to be Modeling Transience? Examples from 242

Residence Time Distributions and Solute Transport
Nicholas B. Engdahl
Aurora: A Partner Package for MODFLOW that Enables Non-Fickian 243
Transport Modeling of Real Aquifers
Scott K. Hansen, B. Berkowitz, and D. Burnell

System-Level Modeling for Waste Management Area C Performance 244

Assessment at the Hanford Site in Southeast Washington, U.S.A.
Nazmul Hasan and S. Mehta

Engineered Streambeds for Improved Water Quality: Comparison of 245

MODFLOW and STAMMT-L Representations of Performance Data from
Constructed Flume Experiments
Skuyler Herzog, C.P. Higgins, and J. McCray

Using Uncertainty Analysis for Groundwater Flow and Transport Modeling 246
to Inform Remedial Design and Monitoring
Katrina Marini, K. Lindstrom, D. Dahlstrom, and J. Mohr

Evaluating Enhanced Attenuation of Uranium at the 300 Area of Hanford 247

Site, Washington
Amena M. Mayenna and S. Mehta

Understanding CH4 Leakage from Pipelines as Affected by Soil 248

Heterogeneity and Moisture
Melissa K. Mitton

Integrated Hydrology Model Intercomparison Using High-Resolution Data at 249

the East River Colorado
David Moulton, E. Siirila-Woodburn, D. Dwivedi, C. Steefel, D. Svyatskiy,
R. Maxwell, and L. Condon

Partial Ages: Defining Domain Specific Residence Time Distributions 250

Carlos A. Rivera Villarreyes, F. Cornaton, and V. Clausnitzer

Using MODFLOW and a Coupled RT3D-OTIS Model to Simulate Best 251

Management Practices for Water Quality Improvement in an Irrigated
Agricultural River Valley
Christopher D. Shultz, T. Gates, and R. Bailey

Embracing Hydrogeologic Uncertainty: Rapid Robustness Assessment of 252

the Capture Well System for the EDB Plume from the Kirtland Air Force
Base Fuel Release Using Analytic Element Modeling
John Sigda, W. Linderfelts, R. Shean, and E. Marcillo

Characterization of Seasonal Thermal Plume Migration from a Below-Water- 258

Table Aggregate Extraction Operation
Michael G.S. Takeda, E.J. Wexler, P.J. Thompson, and J.D.C. Kassenaar

Computationally Efficient Methods for Modeling Diffusion and Reaction in 259

Low Permeability Zones in Field-Scale Simulations using
MODFLOW/MT3D/RT3D
Albert J. Valocchi
 Use of Groundwater Modeling in Assessment of Corrective Actions for 260
Contaminated Groundwater Short-Circuit
Assaf Wunsch and S. Leppert

Section 15: The Analytic Element Method: Developments and Applications

Handling Dewatered Domains in Analytic Element Models 261

Charlie Fitts

Modeling Vertical Leakage and Storage Changes with Spatially-Variable 266

Area Sinks
Charlie Fitts

Analytic Element Modeling on the Cloud: Tethys and TimML 272

Jacob Fullerton, N. Jones, and M. Bakker

AEM Modeling to Support Site Conceptual Model Development and 273

Remediation
Charles McLane, M. Kauffman, T. Pryshlak, and M. Moore

Section 16: Fracture Flow and Transport Modeling

Generalized Responses of Fractured Rock Aquifers to Pumping and 277

Episodic Recharge
Rishi Parashar and D.M. Reeves

Linking FracMan and HydroGeoSphere to Simulate Watershed-Scale Nitrate 281

Transport in Fractured Till
Nathan L. Young, W. Simpkins, R. Horton, R. McLaren, and R. Therrien

Section 17: Unconventional Applications of Groundwater Models

Simulating Complex Surface Water/Groundwater Interactions During Flood 282

Events with a Fully-Integrated Physics Based Hydrologic Model
Steven J. Berg, H-T. Hwang, Y-J. Park, S. Frey, E. Sudicky, N. Grosso,
and M. Sherrier

Design of Infiltration Basins for Cooling Tower Water Bleed and Pool Water 283
Backwash
Loring W. Crowley and E. Rehwoldt

What Makes a MODFLOW-Derived Metamodel Useful in Decision Support? 284

Daniel Feinstein, M. Fienen, and H. Reeves

Boundaries, Interfaces and Discontinuities: Where Things Happen! 285

Stefan Finsterle
The Role of Hydrology in Meteorological Modeling: A Case Study on Insect- 286
Induced Forest Disturbance and its Feedbacks from Groundwater to
Atmosphere
Mary M. Forrester, R. Maxwell, L. Bearup, and D. Gochis

Incorporating Insights from Time Series Analysis in Groundwater Modelling 287

for the Urban Area of the City of Amsterdam
Pytrik Graafstra, F. Smits, and T. Janse

Coupling MODFLOW and Distributed Hydrologic Model BTOP in the 292

Fujikawa River Basin
Maksym A. Gusyev, D. Abrams, J. Magome, and Y. Tokunaga

Using MODFLOW-USG to Simulate the Rise of an Existing Tailing Storage 298

Facility and Estimate Transient Seepage
Rodrigo Herrera, J. Lagos, and T. Opazo

Beyond Hydrology: Application and Uncertainty Analysis of Watershed 302

Heat Transport
Randy Hunt, D. Feinstein, and W. Selbig

Perspectives on the Potential Migration of Fluids Associated with Hydraulic 303

Fracturing in Southwest Florida
William C. Hutchings

Conceptual Model Simulation for the Establishment of Remedial System 308

Performance Criteria
Shawn C. Leppert, A. Wunsch, and R. Farnsworth

Evaluating Non-potable Water Usage for Oil and Gas Purposes in the 309
Permian Basin
Kara Marsac and A. Navarre-Sitchler

Using Fine-scale Simulations to Develop Subgrid Representations of 310

Microtopography Effects in Integrated Hydrology Models
Scott L. Painter, A. Jan, and E.T. Coon

Metamodel Analysis of California’s Coastal Groundwater Basins to Identify 311

Undesirable Results of Management Actions
Zachary P. Stanko, M. Fienen, and C. Faunt

MODFLOW-based Meta-modeling for Estimating Regional Residence-time 312

Distributions
Jon J. Starn, K. Belitz, C. Carlson, L. Desimone, and L. Kauffman

Large Scale High Resolution Modelling 313

Peter Vermeulen, B. Minnema, J. Verkaik, and J. Hughes
Corresponding author: David Bean
Amec Foster Wheeler
[email protected]

Water Banking on the Kern River Alluvial Fan

David M. Bean, Amec Foster Wheeler

J. Parker
M. Anderson

Abstract: Amec Foster Wheeler was retained by KWBA to develop a regional scale ground-
water flow model to evaluate artificial recharge and recovery pumping operations on the
Kern River Alluvial Fan. The model utilizes MODFLOW-NWT to evaluate the impacts
of 6 independently operated recharge and recovery Projects on groundwater levels beneath
the Kern River fan. The Projects consist of 118 recharge basins of approximately 13,600
acres spread over an approximately 36 square mile area. The Projects have recharge nearly
5 million acre-ft of applied water and recovered approximately 4.4 million acre-ft of water
since 1988. Over 260 geophysical logs and 126 aquifer pumping tests were utilized to develop
a 3-dimensional model of hydraulic conductivity distribution in the upper 1,000 feet of the
alluvial aquifer. The model simulates the period from Oct 1988 through December 2016 us-
ing 339 monthly stress periods. Inflow to the model was primarily via intentional recharge,
leakage from the Kern River and irrigation canals, and agricultural return flow. Outflow
was primarily via 750 recovery wells and water supply wells. The model was calibrated to
over 26,500 water level observations in 172 monitoring and water supply wells within the
model domain. The model has been used to evaluate the impacts of recovery operations
on the surrounding aquifer, calculate the net benefit of water banking operations, optimize
future recharge and recovery operations, maximize storage and recovery within the bank,
and minimize outside losses.

16
Corresponding author: Alastair Black
Groundwater Science Ltd
[email protected]

Surface Water Accounting Model, SWAcMOD. A MODFLOW-USG Pre

Processor for recharge and surface water systems

Alastair Black, Groundwater Science Ltd.

Abstract: Representation of topologically routed surface water systems, distributed recharge

and riparian demand in groundwater models is frequently desired, but challenging to im-
plement, in a conceptually robust, mass conservative and time efficient manner. Various
tools exist to assist in the generation of MODFLOW recharge and stream files however
are typically closed-source and limited in the processes which are represented. SWAcMOD
is distributed model to aid the evaluation of how water is partitioned through processes
between rain and recharge, surface flows and AE. Output include MODFLOW-USG com-
patible Recharge, Streamflow-Routing and/or Evapotranspiration files alongside tabulated
water balances for defined areas. To optimise runtimes parallelization, vectorization with
NumPy and use of Cython have been implemented. SWAc is developed to be an accessible,
open-source tool, of modular-design and for multiple climatic regions. The program is devel-
oped in python, is free and is released under the GPLv3 licence model to meet accessibility
needs. The intent of SWAcMOD is to provide an efficient means to build surface water
systems, distributed recharge and riparian demand into MODFLOW.

Extended abstract: Methodology, findings, and implications

Groundwater models are expected to represent observed groundwater level fluctuation and
in many instances additionally simulate groundwater interaction with surface water sys-
tems and wetlands. The representation of groundwater level trends and seasonal fluctuation
relies upon the appropriate magnitude and timing of recharge inputs to the groundwater
model, which typically are highly variable spatially. Within groundwater modelling stud-
ies the use of basic estimates of recharge directly from rainfall records is not uncommon
due to: a) relative difficulty in creating these files; and, b) The uncertainty in the param-
eterisation of processes which delay, attenuate and partition rainfall between groundwater
recharge, surface flow and actual evapotranspiration. Without robust inputs an inability for
the groundwater model to represent observed responses is likely and potentially erroneous
calibration of aquifer properties may be undertaken to compensate for recharge timing and
magnitude deficiencies.

The Surface Water Accounting Model (SWAcMOD) program is presented herein. In an

effort to overcome the above challenges to comprehensively represent the surface and near
surface systems in MODFLOW groundwater models: 1. The program directly produces
the bespoke input files for the industry standard groundwater modelling tool MODFLOW,
including Unstructured Mesh versions (Panday et al 2013). 2. SWAcMOD incorporates
a collection of peer reviewed processes which can be optionally enabled depending on the

17
conceptual system which the modeller wishes to simulate. 3. Topologically routed river
networks can be simulated. 4. The modelling package is designed to take advantage of
increasingly parallel computing architectures reducing runtimes. 5. SWAcMOD is released
under the GNU GPLv3 open source licence and written in the Python language to maximise
accessibility.

Overview

SWAcMOD comprises a workflow of sequential processes. Each process is derived from

existing studies referenced in literature, many of these are described in greater detail in
Heathcote et al 2004 and Markstrom et al. 2015. Processes can be optionally enabled or
disabled depending on the needs of the site being modelled. The program is of modular
design permitting additional processes to be added or adapted within the existing workflow.

The workflow constitutes a 1D model from ground surface down through the soil zone
and unsaturated zones. The 1D model requires, as a minimum, rainfall and potential evapo-
transpiration (PE), with potentially additionally temperature and solar flux time series data
alongside parameters controlling how water is partitioned through each process. Figure 1a
summarises the 1D workflow.

Represented processes include:

• Canopy / vegetation interception and evaporation.
• Temperature dependant precipitation as snowfall or rainfall.
• Snowpack development and melt.
• Partitioning of rainfall/melt water as rapid runoff.
• Soil Moisture Accounting and actual evapotranspiration (AE) calculation.
• Additional impervious zone enhancements to the Soil Moisture Accounting.
• Urban water main or canal leakage.
• Interflow and combined near surface lateral flow processes.
• Attenuation of recharge as passes through the unsaturated zone.
• Surface flow time of concentration attenuation.
• Surface water abstractions and discharges.
• Bypass mechanisms of rejected recharge, macropore recharge and runoff recharge.

The 1D workflow is duplicated for each and every groundwater model cell, however the
spatially variable parameters controlling each property can be spatially implemented either
by zones of identically parameterised cells or via specification of unique values per cell. By
this system the 2D parameterisation of the model can be readily setup. Figure 1b is an ex-
ample of zonal input of model properties in SWAcMOD at the scale of the MODFLOW-USG
groundwater model mesh. Each of the hexagonal cells hosts a 1D workflow as presented in
Figure 1a.

Figure 2 is an example catchment summary for a model located in the centre of Finnish
Lapland. Processes enabled in this model include snowfall-pack and melt, rapid runoff, Soil
Moisture Deficit and AE accounting via an adapted form of the FAO56 methodology (Allen

18
et al 1998), interflow, macropore bypass recharge (rapid recharge), secondary recharge from
runoff (important for representation of observed increased melt recharge as approach surface
water courses), surface water attenuation and groundwater recharge lag and attenuation.
All processes down to the interflow store have time variant properties to permit soil freeze.

Uncertain model properties have been constrained through calibration to river flow gaug-
ing, measured snow thickness (green dots in Figure 2) and Sentinel Satellite (ESA 2017) de-
rived snow cover spatial distributions over time. In addition, several properties, for example
rapid runoff characteristics and vegetation rooting depths, have been estimated based on
experience from past studies and reference material respectively.

The bottom three plots represent time series inputs for MODFLOW-USG and are the
primary outputs for the model. These are the ‘surface component of total river flow’ (all
excluding baseflow), ‘Unutilised PE’ (the input to control riparian evapotranspiration) and
‘recharge to the saturated zone’. These three outputs are spatially distributed however this
plot summarises the average daily time series values for this aggregated catchment area.

All units in Figure 2 are millimetres, the temporal resolution is daily.

Closing

Surface Water Accounting Model (SWAcMOD) has been developed to be an open, acces-
sible, adaptable and fast tool to enable better representation of time variant and spatially
heterogeneous recharge, actual evapotranspiration and river flows. Alongside a ground-
water modelling package like MODFLOW-USG this integrates catchment water balances.
SWAcMOD is designed to work in sequence with MODFLOW and is compatible with the
Unstructured Grid versions. The processes included within the SWAcMOD program are
not new and are based on peer reviewed methods. However, the holistic framework to pull
the processes together, generate river topology in unstructured models and, particularly,
the methods to invoke parallelization–reducing run-times-are considered advancements. In
addition, barriers to robust representation of recharge are considered to result from data
scarcity in many studies. The SWAcMOD program has been designed to make use of newly
available high resolution and free satellite data-sets which help to constrain several spatially
and temporally variable processes. This, in combination with conventional ground station
based data, is a significant improvement in what has been openly available in recent years.

19
Figure 1a Top: SWAcMOD 1D process workflow. Figure 1b Bottom: Parameterization by zones on a
MODFLOW-USG mesh (as an alternative to unique values per cell)

20
 21
Figure 2 Break down of the outputs from SWAcMOD processes for a catchment in Finnish Lapland.
 References

Allen et al. 1998 Crop evapotranspiration - Guidelines for computing crop water requirements - FAO
Irrigation and drainage paper 56
ESA 2017 https://earth.esa.int/web/sentinel/user-guides/sentinel-2-msi/resolutions/spatial
Heathcote et al. 2004 Rainfall routing to runoff and recharge for regional groundwater resource models.
Quarterly Journal of Engineering Geology and Hydrogeology, 37, 113-130
Panday, Sorab, Langevin, C.D., Niswonger, R.G., Ibaraki, Motomu, and Hughes, J.D., 2013, MODFLOW–
USG version 1: An unstructured grid version of MODFLOW for simulating groundwater flow and tightly
coupled processes using a control volume finite-difference formulation: U.S. Geological Survey Techniques
and Methods, book 6, chap. A45, 66 p
Markstrom, S.L., Regan, R.S., Hay, L.E., Viger, R.J., Webb, R.M.T., Payn, R.A., and LaFontaine, J.H.,
2015, PRMS-IV, the precipitation-runoff modeling system, version 4: U.S. Geological Survey Techniques
and Methods, book 6, chap. B7, 158 p., https://dx.doi.org/10.3133/tm6B7

22
Corresponding author: Henk Haitjema
Haitjema Consulting, Inc.
[email protected]

Horizontal Flow Models that are not

Henk Haitjema, Haitjema Consulting, Inc.

Abstract: We strive for realism and predictive power when developing a groundwater
model. We also try to keep that modeling effort affordable. A one-layer MODFLOW model
is, of course, more affordable than a fully 3D model, but is such a “horizontal flow model”
realistic and has it any predictive power? Or is it more likely to mislead us? Only a
few years after Henry Darcy presented his experiment that lead to Darcy’s law, his fellow
countryman Jules Dupuit observed that groundwater flow on a regional scale is mostly hor-
izontal and proposed to ignore what little vertical flow is there. Later Philipp Forchheimer
incorporated this concept in a governing differential equation for unconfined flow that only
involves the horizontal coordinates x and y. Thus was born the “horizontal flow model”
also known as a “Dupuit-Forchheimer model.” In the mid 1960’s Don Kirkham redefined
the Dupuit-Forchheimer paradigm as ignoring resistance to vertical flow – thus not ignoring
vertical flow itself. In the 1980’s Otto Strack presented equations for three-dimensional
streamlines in Dupuit-Forchheimer models, albeit approximate ones. Later work compared
complex 3D models to Dupuit-Forchheimer models often obtaining very similar insights (and
streamlines) from each of them. However, multi-layer models are needed when two or more
aquifers are stacked with spatially varying exchanges of water between them. But even then,
multi-layer MODFLOW models can be considered as stacked Dupuit-Forchheimer models
with approximate 3D flow in each layer. Dupuit-Forchheimer models, therefore, are ideal
candidates for a step-wise modeling approach to maximize results and minimize cost.

23
Corresponding author: Alan Laase
Navarro Research and Engineering, Inc
[email protected]

Estimating Evapotranspiration and Recharge Rates Using a Remote Sensing

Algorithm

Alan Laase, Navarro Research and Engineering, Inc

Abstract: Assigning representative recharge rates to groundwater flow models is inexact

and usually involves rough estimates based on professional judgement. To improve recharge
estimates, the US Department of Energy Office of Legacy Management applied a remote
sensing algorithm derived from ground evapotranspiration (ET) measurements and satellite
imagery, to determined spatially variable ET discharge and precipitation recharge rates.
These rates were for used in large scale groundwater flow models. At a former uranium
mill site the algorithm was used to determine ET rates for nine distinct plant communities
within the model domain at a former uranium mill site. For areas where the annual average
precipitation rate exceeded the individual annual plant community ET rate, recharge was
assigned at a rate equal to the difference between the two rates. In the plant communities
where the ET rate exceeded annual average precipitation, groundwater was removed from
the model at a rate equal to the difference between the two rates. The algorithm predicted
that groundwater recharge occurred in five of the nine different plant communities at rates
ranging from 2 to 50 mm/yr. Discharge (ET) was predicted to occur in two plant communi-
ties at rates between 24 to 151 mm/yr. Zero recharge occurred in the two remaining plant
communities where ET equaled annual average precipitation. The validity of the remote
sensing algorithm was evaluated by comparing the net volumetric discharge rate for the
model domain with base flow measurements collected from a downgradient creek. The two
values compared favorably, validating the plant community based recharge and ET rates
derived using the remote sensing algorithm.

24
Corresponding author: Melissa Masbruch
U.S. Geological Survey
[email protected]

Analysis of infrequent (quasi-decadal) large groundwater recharge events: A

case study for northern Utah

Melissa D. Masbruch, U.S. Geological Survey, Utah Water Science Center, Salt Lake City, Utah, USA
Christine A. Rumsey, U.S. Geological Survey, Utah Water Science Center, Salt Lake City, Utah, USA
Subhrendu Gangopadhyay, U.S. Bureau of Reclamation, Technical Service Center, Denver, Colorado, USA
David D. Susong, U.S. Geological Survey, Utah Water Science Center, Salt Lake City, Utah, USA
Tom Pruitt, U.S. Bureau of Reclamation, Technical Service Center, Denver, Colorado, USA

Abstract: There has been considerable research linking climatic variability to hydrologic
responses in arid and semi-arid regions such as the western United States. Although much
effort has been spent to assess and predict changes in surface-water resources, little has
been done to understand how climatic events and changes affect groundwater resources.
This study focuses on quantifying the effects of large quasi-decadal groundwater recharge
events on groundwater in the northern Utah portion of the Great Basin for the period
1960 to 2013. Groundwater-level monitoring data were analyzed with climatic data to
characterize climatic conditions and frequency of these large recharge events. Using observed
water-level changes and multivariate analysis, five large groundwater recharge events were
identified within the study area and period, with a frequency of about 11 to 13 years.
These events were generally characterized as having above-average annual precipitation and
snow water equivalent and below-average seasonal temperatures, especially during the spring
(April through June). Existing groundwater ow models for several basins within the study
area were used to quantify changes in groundwater storage from these events. Simulated
groundwater storage increases per basin from a single event ranged from about 115 Mm3
to 205 Mm3 . Extrapolating these amounts over the entire northern Great Basin indicates
that even a single large quasi-decadal recharge event could result in billions of cubic meters
(millions of acre-feet) of groundwater recharge. Understanding the role of these large quasi-
decadal recharge events in replenishing aquifers and sustaining water supplies is crucial for
making informed water management decisions.

25
Corresponding author: Aaron Pruitt
Aspect Consulting
[email protected]

Solving the Water Supply Puzzle: MODFLOW and Uncertainty in the Context
of Mitigated Water Rights

Aaron Pruitt, Aspect Consulting

P. Bannister
T. Flynn

Abstract: Water rights law in the State of Washington requires that new or junior appro-
priations do not impair more senior appropriations. In some cases, Washington’s Depart-
ment of Ecology has closed stream basins to further appropriations to protect endangered
species habitat. In these closed basins, new appropriations of groundwater rights must
be water budget neutral (i.e. cannot alter stream flows). A recent State Supreme Court
ruling requires that mitigation proposed to offset new appropriations be in-kind, in-time,
and in-place. Satisfying these requirements often requires numerical modeling to quantify
impacts and demonstrate the effectiveness of proposed mitigation. This case study will
discuss how a large municipal water supplier in the Pacific Northwest, seeking to meet the
stateâ s mitigation criteria above, investigated the viability of using Aquifer Storage and
Recovery (ASR) in a complex of glacially-deposited aquifers. A MODFLOW-SURFACT
groundwater flow model was used to simulate the potential impacts various ASR scenarios
might have to the closed basin overlying the aquifer. Model results show that drawdown
from pumping could affect stream flows, and that the seasonal storage could not address
all stream flow losses. The modeling analysis was able to identify alternative mitigation
strategies to demonstrate water budget neutrality and support permitting of new ground-
water supplies. This project illustrates the need to understand and properly communicate
model limitations, assumptions, and uncertainty. In a stringent regulatory framework, any
modeled impactâ no matter how smallâ can affect the ultimate goals of the project.

26
Corresponding author: Eliana Rodriguez
Universidad Autonoma del Estado de Baja California
[email protected]

Analysis of aquifer recharge by environmental flows in a semiarid zone using

Diffusion Wave package for MODFLOW

Eliana Rodriguez, Universidad Autonoma del Estado de Baja California

M. Shanafield
J. Ramirez-Hernandez

Abstract: Along his course the Colorado River is dammed and diverted, before this reg-
ulations the river fed one of the world largest delta, reduced to a few remaining wetlands.
In 2014, environmental flows through the Colorado River in Mexico were delivered as a
part of an international agreement between Mexico, United States and nongovernmental
organizations denominated Minute 319. During one event, the pulse flow, 130 x 106 m3
were released during the period of March 23 - May 18, 2014. The assessment of the hydro-
logical response to these discharges is a fundamental part of the commitments established,
but it is also indispensable information for the maintenance, management and planning of
restoration efforts. A coupled groundwater and surface water model has being developed
and calibrated using MODFLOW with a diffusion wave routine along 43 km of the dry
riverbed. This model is capable of simulating flood routing, spatially-variable infiltration,
and flood-wave attenuation, essential components for the design of future flows. The dry
river was discretized in 114 segments and 558 reaches, incorporating detailed topographic
geometry, 43 discharges were simulated which varied from 3.99 to 88.54 m3 /s, calibrating
the riverbed hydraulic conductivity using the water front as calibration target. The model
obtained results of vertical flows up to 1.25 m3 /s. Two scenarios were simulated, it is con-
cluded that for a longitudinal river connection opt for a hydrograph of variable flows of less
than 20 m3 /s, in order to avoid flooding of high terraces and thereby reduce the infiltration.

27
Corresponding author: Otto Strack
University of Minnesota
[email protected]

Vertically integrated flow for complex aquifer systems as a modeling tool

Otto D. Strack, University of Minnesota

Abstract: In a vertically integrated groundwater model, the discharge represents the verti-
cally integrated discharge. Modeling aquifer systems in a vertically integrated sense, either
in addition to creating a complete model of an aquifer system, or as a stand alone model,
is attractive for the following reasons. First, complex systems of aquifers can be dealt with
effectively, even if the aquifers have vastly different properties. Second, the data collection
effort reduces in scope, and becomes similar to that required to model a single aquifer. The
analytic element method is a logical choice for vertically integrated modeling; this method
was originally developed for single aquifers. At present, the method of vertically integrated
modeling is capable of dealing with any number of aquifers, or strata, with horizontal inter-
faces between the aquifers, and vertical boundaries. For such systems, the method provides
accurate discharge components throughout the system, even if the aquifers are partially con-
fined, unconfined, or even empty of water. Wells can be included regardless of the aquifer
or stratum that contains the well screen. Research is underway that makes it possible to
include rivers and other boundaries in any one of the aquifers or strata. This requires that
resistances be determined that represent the total resistance between the river and those of
the aquifers that the river does not penetrate. Eventually, a full set of parameters must be
determined for each partially penetrating feature to include it in the vertically integrated
model. A vertically integrated storage coefficient is an example.

28
Corresponding author: Robert Wyckoff
Brown and Caldwell
[email protected]

Simulating Managed Aquifer Recharge and San Pedro River (Arizona)

Baseflow Augmentation using MODFLOW-NWT

Robert Wyckoff, Brown and Caldwell

J. Weaver
J. Wright

Abstract: The City of Sierra Vista, Arizona, developed an Environmental Operations Park
(EOP) designed to treat up to 4.0 million gallons per day of reclaimed water. Wastewater
is treated in approximately 50 acres of constructed wetlands before being recharged via
11 artificial recharge basins. The system provides recharge to the regional aquifer, offsets
deficit groundwater pumping, and increases baseflows to the nearby San Pedro River. Un-
derstanding the effects of long-term EOP recharge is important for maximizing benefits to
local riparian habitat, surface water resources, and for groundwater planning efforts by area
governmental agencies. A groundwater model was developed in 2009 to evaluate the ef-
fects of managed recharge on coupled shallow groundwater and surface water resources. For
the 2015 model update (MODFLOW-NWT), BC added EOP recharge volumes and regional
pumping in the model through 2013. Following model calibration, BC developed projections
for future EOP recharge and groundwater supply pumping rates using population growth.
These data were then used to inform a predictive model that simulates future potential
effects of EOP recharge. Model results predict that by 2040, increases in groundwater water
levels resulting from the program extend approximately 8 miles to the west and south of
the EOP facility, helping to mitigate long-term groundwater level declines resulting from
regional pumping. Discharge from groundwater to area springs and the San Pedro River
are predicted to limit groundwater rise to the east but serves to directly augment baseflows
in the San Pedro River.

29
Corresponding author: Jacob Bauer
Leonard Rice Engineers
[email protected]

Equation for Stream Accretion Timing from a Line Source Oriented

Perpendicular to a Stream

Jacob Bauer, Leonard Rice Engineers

Abstract: Groundwater return flow timing from irrigated parcels or aerial recharge sources
are frequently estimated using equations that express stream depletion timing as a function
of a point, often at the centroid of the source, that is used to represent return flows from
the entire source. The poster presents an equation for stream accretion timing from a line
source oriented perpendicular to a stream. The equation solves for the ratio of the stream
accretion rate (q) to the constant source rate (Q), also known as the q/Q ratio. The equation
is derived by integrating the Glover (1954) equation along every point on the line source
oriented perpendicular to the stream. The accuracy of the equation is confirmed using a
groundwater flow model of an equivalent system that honors the assumptions built into the
equation. Additionally, the applicability of the equation to estimate return flow timing for
an entire polygon source, based on an assumption that flow changes are oriented primarily
perpendicular to the stream, is explored. Finally, test cases showing appropriate uses of the
equation are presented showing suggested parameter ranges where the analytical expression
provides a meaningful improvement over traditional (centroid) approaches for characterizing
the line source or polygon source.

30
Corresponding author: Alastair Black
Groundwater Science Ltd.
[email protected]

Case study: Surface-Groundwater modelling to guide the revocation or

reduction of abstraction licences in the UKs primary Sandstone aquifer

Alastair Black, Groundwater Science Ltd.

Abstract: The East Midlands Sandstone aquifer is the second largest groundwater resource
in the UK and in some catchments, has a designation of ‘over-abstracted’ and in unfavourable
condition within the EU Water Framework Directive. The metrics for determining aquifer
status and sensitivity to groundwater abstraction are derived from river flows where surface
water is in connection with the groundwater system. Most metrics concentrate on river low-
flow conditions. The purpose of the study presented here is to identify the abstractors which
influence the river flows at the time, and locations which lead to unfavourable designations
for the wider area. This paper is a case study demonstration of the methodology applied. A
description of the scheme to classify depletion and stress across the sub-catchments of the
aquifer is introduced. A phased approach to identify significant stresses follows.

Extended abstract: Methodology, findings, and implications

PROCESS OVERVIEW

The East Midlands Sandstone aquifer is utilised for potable, agricultural and industrial
purposes and abstraction sustainability measures are considered to protect river environs.
The purpose of the study presented here is to identify the abstractors which influence the river
flows at the time, and locations, which lead to unfavourable designations for the wider area.
These analyses are used to limit the number of licence reductions or revocations to a minimum
and halt deterioration of the aquifer. Prior to model simulations, surface waterbody sub-
catchments are identified and confirmed to be consistent with MODFLOW (Harbaugh 1996)
routed Stream topology. River Flows are extracted from MODFLOW models at the outflow
points of each waterbody, the data are formatted as Flow Duration Curves (FDCs) and
Environmental Flow Indicators (EFIs) are calculated. The Environmental Flow Indicator is
a percentage deviation from the natural river flow and is a guide as to how much water is
available for anthropogenic use, further detail can be found in (EA 2013). This percentage
deviation is different at different flows, typically invoking the highest restrictions at low flow
periods. It is also dependant on the ecological sensitivity of the river to changes in flow. The
EFI provides a metric to compare the licensing status across space.
With EFI metrics derived initial MODFLOW simulations are setup to demonstrate the
effect of each of the ‘sectors’ of groundwater abstractor collectively, for example: potable,
industrial or spray irrigator’s. With focus on the timing and magnitude of impact, the effect
of each abstraction type on groundwater levels, stream flows and the waterbody designations
can be demonstrated. At this level of studies individual licences which the river shows
greatest sensitivity to are not identified. This first pass summarises the relative importance
of sectors and may eliminate large numbers of abstractors from future phases.

31
 Individual abstraction apportionment to sub-catchments follows providing further detail
and ranking. Volumetric backtracking of groundwater is used to identify the provenance of
abstracted volumes. Data are tabulated for each and every abstraction and by waterbody.
This volumetric backtracking method has the advantage of not requiring large numbers of
runs to identify the effect of each well substantially reducing time and computational burden.
There are around 1000 licenced abstractors, above a deregulation threshold, within this single
model area therefore water source delineation from a single run is preferable to conventional
simulations for each well.
The tabulation of abstracted volumes from each waterbody typically identifies the small
number of abstractors to review in detail and bespoke model runs can be initiated to consider
the maximum permissible volume to be licenced in the future which will lead to the desired
‘no deterioration’ or local aquifer status improvements. These data are used to support
stakeholder consultation.

ADDITIONAL DETAIL AND WORKED EXAMPLE COMMENTARY

Environmental Flow Indicators, Abstraction Sensitivity Bands and Compliance

Banding

In the development of the EFI each surface waterbody has a surface flow outflow point
and the corresponding MODFLOW Stream cell identified. Flow timeseries are extracted
from a ‘Naturalised’ MODFLOW model which excludes all groundwater wells. These data
are summarised as a flow duration curve for each of the waterbodies and represent the best
estimate of Natural river flows. The overall intent of the below described methodology is to
ascertain whether the proportion of the natural river flow that is taken-up by abstraction
exceeds surpluses therefore may cause an undesirable effect on river habitats and species.
As such an estimate of what is required by the environment is required, the EFI is that
demarcation. In many areas, detailed surveys of ecological flow requirements will exist,
out with these sites, ‘Abstraction Sensitivity Bands’ (ASB) are applied which estimate the
requirements based on local geology and river typology, Macroinvertebrate typology and
Fish typology. The ASBs exist for all waterbodies in England and Wales (EA 2016) and
apply different proportions of water available at each flow percentile. For example, in an
area of high sensitivity only 10% of river flow at the Q95 may be accessible to groundwater
abstractors. In a low sensitivity area, this may be 20%. Similarly, at periods of greater flow,
e.g. Q50 the bands may be relaxed to range between 20 and 26% of flow which would be
available for abstraction usage between high and low sensitivity areas respectively.
The Environmental Flow Indicator (EFI) is the ‘Natural’ Flow Duration Curve multiplied
by the ASB’s and indicates where water may need to be recovered for each surface waterbody
sub-catchment. It is relative to this volume at various flow percentiles that waterbodies and
abstractors are assessed.
Figure 1 denotes the waterbodies of the East Midlands Sandstone aquifer MODFLOW
model. Colouring and posted values are the ML/d surplus and deficits of each waterbody at
Q95. These values are the difference between the EFI (the volume estimated as the needs
of the environment to maintain Good Ecological Status) and the flows from a simulation of
the current abstraction regime. A negative value represents human abstraction beyond the
EFI demarcation.

32
 Figure 2 classifies waterbodies as ‘compliant’ and Bands 1 to 3 ‘non-compliant’. At Q95
complaint (green) denotes Q95 flows exceed the EFI. Yellow, Orange and Red are Bands 1
to 3 where current Q95 flows breach the EFI by <25%, <50% and >50% of remaining flows
respectively.
These plots are developed to show the catchments to focus licensing review.

First Pass Modelling – Sectoral Analysis

Figures 3 through 7 show a selection of model runs where a sector (potable, industrial,
spray irrigators etc.) have increased rate from current ‘Recent Actual’ usage to their ‘Fully
Licenced’ permissible rate. The contours in each plot show the Sandstone groundwater
piezometric surface. The colour scale shows the additional drawdown induced by increasing
abstraction to the fully licenced rate. These plots can be useful on a regional scale how-
ever the sectoral simulations are undertaken to reproduce the band classifications as seen
in Figure 2. On figure 2 some water bodies have a stipple. This denotes the catchment
would deteriorate into the next more severe banding if abstractors utilised their headroom
and increased pumping to their Fully Licenced rate. Detailed Identification of individual
abstractors influencing status outcomes
A single model run can be used to backtrack the pathway and source zone of abstracted
water for each and every licenced well in the model. The backtracking analysis is required
to be volumetric and fast such that it can be scaled to 100s or 1000s of discrete abstractors.
FlowSource (Black and Foley 2013) or MODALL (Potter 2008) can be applied to track
abstracted volumes and the spatial distributions of abstraction source zones output. These
datasets are too comprehensive to review manually and a simple macro is utilised to aggregate
the volume of each abstraction which is sourced from each waterbody. These volume data
can be tabulated by waterbody against well.
As an example, a large catchment in the south of the model has been singled out and the
tabulated wells which source water from within that catchment (only) are summarised in the
chart in Figure 9. The abstraction with the greatest volume from the waterbody also derives
water from 3 other waterbodies as seen in Figure 8. The posted values are average ML/d
per catchment for that well. On initial inspection, the impacted catchments are surprising.
a) the abstraction is to the east of the catchment, not within; and, b) there are unaffected
(white) catchments seemingly in between. At this level of investigation, the comprehensive
datasets formerly extracted can be reviewed to show in Figure 10 the volumetric ‘pathway’ of
the water and Figure 11 the volumetric ‘source’ zone of the abstractions water. Combined,
the Figures 10 and 11 plots show why the simulation implies the uncoloured catchments
are not within the source zone and how that water is distributed to be sourced within the
waterbody. Figure 11 demonstrates a very low volume of water is derived at the location
of the well and the source has a sharp boundary. This coincides with a confining tight clay
Formation (white in Figure 11). The Sandstone Unit which hosts the abstraction is confined
to the east and in the location of the well. This is showing the well to have a dislocated
impact sourcing the majority of its water from the unconfined Sandstone (blue). Dark green
contours denote the piezometric surface, light green show simulated drawdown of the well.

Closing

33
 A consistently applied scheme to classify the availability of water at river sub-catchments
scale is introduced. Large scale sectoral analysis runs are initially undertaken to obtain an
understanding of the spatial and temporal influence of each group for the aquifer. This
assists in reducing the scope of the following stage. The final stage identifies the individual
abstractors which the river reaches are simulated to be most sensitive to. This can be rapidly
undertaken by tabulation of abstraction source zones. These data are used to justify why
individual abstractors are identified and more in-depth bespoke model simulations can be
undertaken to assess need and magnitude of license changes.

Figures 1 through 7 Left to Right: Surplus and deficit by catchment at Q95, Compliance Band classification,
5 sets of hydraulic head impact scenarios (Non-Public Water Supplies and various Time Limited Licences
groupings)

34
Figures 8 through 11 Top Left: Catchments influenced by Abstraction Licence 13, chart of abstractors influ-
encing waterbody 32, Bottom Left: FlowSource volumetric ‘Pathway’, Bottom Right: FlowSource volumetric
abstraction ‘Source’ and surface geology base mapping.

References

Black & Foley 2013. FlowSource: A program to efficiently delineate volumetric capture areas, pathways
and source areas in groundwater models. MODFLOW and More 2013: Translating Science into Practice -
Conference Proceedings, this volume - igwmc.mines.edu.
EA 2013 http://webarchive.nationalarchives.gov.uk/20140328084622/http:/cdn.environment-agency.gov.uk/
LIT_7935_811630.pdf
EA 2016 https://data.gov.uk/dataset/abstraction-sensitivity-bands-cycle-2
Harbaugh, A.W. & McDonald, M.G., 1996. User’s documentation for MODFLOW-96, an update to the
U.S. Geological Survey modular finite-difference ground-water flow model. U.S. Geological Survey Open-
File Report 96-485
Potter, S. T., Moreno-Barbero,E. and Divine, C. E. 2008. MODALL: A Practical Tool for Designing and
Optimizing Capture Systems. Vol. 46, No. 2 GROUND WATER (pages 335–340)

35
Corresponding author: Thomas Griffiths
Aquaveo
[email protected]

Implementation of the Subsidence and Aquifer-System Compaction Package to

Model Land Subsidence in Virginia’s Chesapeake Bay Area Using a
Hydrogeologic-Unit Flow Based SEAWAT Model

Thomas Griffiths, Aquaveo

S. Kudlas
N. Jones
A. Lemon

Abstract: Areas in and around the southern Chesapeake Bay region of Virginia are experi-
encing land subsidence. This subsidence has been observed since 1940. Subsidence, together
with global sea-level rise, poses a serious risk to infrastructure, homes, and the environ-
ment. The Virginia Department of Environmental Quality (VDEQ) manages groundwater
resources for the State of Virginia. Two regional groundwater models are used by the VDEQ
in managing the groundwater resources of the state. Of these, the VAHydro-GW model en-
compasses the southern Chesapeake Bay region. The VAHydro-GW is a SEWAT model
(MODFLOW+MT3DMS) which utilizes the Hydrogeologic-Unit Flow (HUF) package to
define aquifer properties at each model cell. The aquifer properties of individual model cells
are synthesized from the properties of layer-independent hydrogeologic units that intersect
the cell. The VDEQ implemented the MODFLOW Subsidence and Aquifer-System Com-
paction (SUB) package within the VAHydro-GW. To do so, custom scripts were created to
synthesize HUF aquifer parameters and combine them with subsidence parameters. The
SUB package was calibrated to historic extensometer data. Various land subsidence scenar-
ios due to aquifer-system compaction were then forecasted using the SUB Package within
the VAHydro-GW.

36
Corresponding author: Charles Heywood
U.S. Geological Survey
[email protected]

Simulations supporting management of saltwater movement in the aquifers of

the Baton Rouge area, Louisiana

Charles Heywood, U.S. Geological Survey

M. Lindaman
J. Lovelace

Abstract: Groundwater withdrawals since the 1940s have lowered water levels, altered
groundwater-flow directions, and caused saltwater to intrude within some freshwater-containing
sands of the fluvial-deltaic Southern Hills regional aquifer system beneath Baton Rouge,
Louisiana. Groundwater investigations in the 1960’s identified a freshwater-saltwater in-
terface located at the Baton Rouge fault, where abrupt changes in water levels also occur.
Generally, aquifers south of the fault contain saltwater and aquifers north of the fault con-
tain freshwater, although by 2005 saltwater intrusion had been detected in seven of the ten
sand aquifers north of the fault. The aquifers underlie East and West Baton Rouge Parishes,
Pointe Coupee Parish, and East and West Feliciana Parishes, and provided about 184 million
gallons per day (Mgal/d) for public supply and industrial use in 2012. A modeling investi-
gation ongoing since 2007 utilizes refinements of the hydrogeologic framework for updates
to evolving models of variable-density groundwater flow and saltwater transport within the
aquifer system. This study periodically reports on simulations of management scenarios for
prioritized ”Sands” that compose the aquifer system. Recent simulations of saltwater trans-
port in the ”1,200-Foot”, ”1,500-Foot”, ”2,400-Foot”, and ”2,800-Foot” Sands include the
effects of saltwater intrusion-abatement strategies, such as operation of aquifer-connector
and saltwater-scavenger wells.

37
Corresponding author: Tracie Jackson
U.S. Geological Survey
[email protected]

Evaluating Potential Groundwater Pumping Effects in Pahrump Valley on

Water Resources in the Amargosa Wild and Scenic River, Nevada and
California

Tracie Jackson, U.S. Geological Survey

K. Halford, U.S. Geological Survey

Abstract: Pahrump Valley is an over-allocated groundwater basin in southwest Nevada

and eastern California. Groundwater pumping began in 1913, continued to increase up to
the mid-1970s for predominantly agricultural use, then declined and stabilized from 1980 to
present due to the shift from agricultural to municipal use. About 1.2 million acre-feet of
groundwater has been withdrawn from the basin-fill aquifer between 1913 and 2016. Down-
gradient of Pahrump Valley is the Amargosa Wild and Scenic River (AWSR), a national
conservation land in the Death Valley region, California. Federal, State, and local entities
are interested in potential effects of current and future groundwater pumping in Pahrump
Valley on the capture of discharge from springs that provide perennial flow to the AWSR.
A groundwater-flow model was developed to simulate transient (pumping) conditions in
Pahrump Valley. Pumping between 1913 and 2010 was simulated, and transmissivity and
storage of the basin-fill aquifer were estimated. Calibration targets were measured water lev-
els in wells, spring discharge, groundwater evapotranspiration, and transmissivity estimates
from specific capacity. The calibrated model delineates the maximum extent of drawdown
from cumulative pumping between 1913 and 2010 in Pahrump Valley. The calibrated model
then was run in prediction mode to estimate water-level and spring-discharge declines near
the AWSR from 2010 and 2100, assuming 2010 pumping rates.

38
Corresponding author: Alan Lemon
Aquaveo
[email protected]

Living Groundwater Models: Case Study for a Continually Updated

Groundwater Models

Alan M. Lemon, Aquaveo

N. Jones
S. Kudlas
T. Griffiths
H. Tran

Abstract: The Virginia Department of Environmental Quality (VDEQ) manages ground-

water for the State of Virginia. Two regional groundwater models are used by the VDEQ
in managing the groundwater resources of the state. Historically, the state would build new
models or update old models only every 7 to 10 years due to the high cost of model develop-
ment. However, the VDEQ and others are collecting groundwater data every year, including
new wells, pumping data from existing wells, water level data, and water quality data which
could be used to improve the models. To improve the efficiency of the process, the VDEQ
has implemented an enterprise system (web portal) for entry of this data and each year the
VDEQ publishes new ”reported use” versions of the groundwater models and new ”total
permitted” versions of the models using the updated data. Tools have been developed as
part of the web portal to generate updated model input files with ”one click”. VDEQ also
has plans in place for periodic (every 2 to 4 years) model updates that are not yet fully
automated such as an update to the geologic framework of the model or recalibrating the
model based on newly available field data. The system has resulted in more useful models
because each year the model has incorporated the latest available field data and planners
are able to use the newly updated models for their forecasts. Further, the incremental cost
to update the models is much less than building new models every 7 to 10 years.

39
Corresponding author: Thomas Mack
U.S. Geological Survey
[email protected]

Preliminary-Assessment and Upgrade of a Groundwater Flow Model of the

Seacoast Bedrock Aquifer, New Hampshire

Thomas J. Mack, P.G., P.HG., U.S. Geological Survey

Abstract: In 2003 and 2004, the U.S. Geological Survey investigated the availability of
groundwater resources in a 160-square mile area of coastal New Hampshire (Figure 1) using
a regional groundwater flow model (Mack, 2009). At that time, population growth and in-
creasing water demand prompted concern for the sustainability of the region’s groundwater
resources in a fractured-crystalline bedrock-aquifer with little storage. The groundwater
flow model developed for the previous study incorporated detailed water-use information
for 2003-4 and simulated the effects of projected increases in water use. However, poor
stream representation may reduce the effectiveness of the original model head simulations.
Improvements to the model, made by incorporating the USGS’s MODLFOW-2005 Newton
formulation (MODFLOW-NWT, Niswonger and others, 2011) and by more accurately repre-
senting stream characteristics, are presented in an example simulating approximate changes
in water use. Groundwater heads in an area of relatively larger population change, near
the center of the Seacoast’s fractured bedrock aquifer, were simulated with the upgraded
model using published 2004, and approximated 2015, water use rates. This area is situated
at a local topographic high point and near the junction of three towns, where drainages flow
westward, toward Great Bay, and eastward, toward the Atlantic Ocean (Figure 1).

Extended abstract: Methodology, findings, and implications

MODEL UPGRADES AND ASSESSMENT

The original Seacoast groundwater flow model was developed in the early 2000’s based
on head and flow data for 2003 and 2004 (Mack, 2009). Since that time improvements in
elevation data and software have become available that were utilized here. The original
model used less accurate stream and digital elevation data, which led to difficulties in proper
streamflow routing. For this example, the model was upgraded by incorporating a stream
network generated from the National Elevation Dataset (NED, nationalmap.gov) and 1/3
arc-second (10 m) NED digital elevation model (DEM) data, processed using ModelMuse
(Winston, 2009). The model top and stream stage were assigned the lowest DEM elevation in
the 200 square foot model cell area to avoid streams being incorrectly assigned as flowing up
elevations. Additionally, the MODFLOW model code used was changed from MODFLOW-
2000 (Harbaugh and others, 2000) to MODFLOW-NWT (Niswonger and others, 2011).
Although this example presents confined simulations, the new code will allow for improved
unconfined simulations. Model datasets were formatted for MODFLOW-NWT but all other
model design, features, and parameters were kept the same as used in the original model
(Mack, 2009).
The updated steady-state model used the original model parameter values–therefore an
unchanged calibration–and shows an improved simulation for the Fall 2004 low-flow cali-
bration period (Mack 2009). The standard error of regression (s, dimensionless), calculated

40
using UCODE-2014 (Poeter and others, 2014), was reduced from 5.6 to 2.8 for the simulation
as a whole. Additionally, 95 percent of the simulated heads are within the original s (5.6)
as opposed to 91 percent in the original model. In the updated model, all simulated flows,
except for the smallest basin (1.53 square miles), were within one s (2.8) whereas in the
original model only 9 of the 14 basin flows were within the larger error (5.6) of the original
model.

SIMULATION OF APPROXIMATED 2015 WATER USE

The previous Seacoast groundwater flow study (Mack, 2009) projected groundwater re-
sources for the years 2017 and 2025, based on average per person water use and projected
population increases in the Seacoast area (Mack, 2009). The local regional planning com-
mission has provided revised population values for 2015 by town for the Seacoast area (New
Hampshire Office of Energy and Planning, 2016). Population increase factors for the 2004
to 2015 period are shown by town on Figure 1. One town with a large population change, a
20 percent increase, immediately west of the topographic high in Figure 2. For this analysis,
the 2004 simulated municipal and industrial groundwater withdrawals, and distributed do-
mestic withdrawals and returns, in each town were adjusted by the factor shown on Figure
1 to approximate 2015 conditions. The original 2017 water use projection was more a much
detailed analysis therefore no attempt was made to compare the current 2015 approxima-
tion to the previous 2017 simulation. Groundwater heads were simulated in model layer
3 representing the upper bedrock aquifer. Simulated groundwater heads for the 2004 and
2015 scenarios (Figure 2) show small changes in head contours but those changes are most
apparent in the town to the northwest where the water use change was greater than other
areas. In the 2015 scenario however, with increased water use in the town to the west, the
simulated heads at the well W-1 showed the greatest change in the Figure 2 area of the
model. These hypothetical scenarios illustrate the potential changes in groundwater heads
near a watershed divide due to changes in water use.

DISCUSSION AND LIMITATIONS

This example illustrates the value of the improved stream network in the Seacoast ground-
water flow model and the sensitivity of the model to stream stage. Additionally, approxi-
mated water use changes, shown for illustration, affect simulated groundwater heads near
watershed divides. Several notable limitations of the model and simulations in this exam-
ple, include (1) because of the changes made this model would need to be re-calibrated
and fully assessed prior to use for planning or management purposes; (2) the steady-state
groundwater flow model averages seasonal water use which is greater during summer months,
2015 head changes may be much greater seasonally; (3) the steady-state simulation depicts
does not account for transition time between water use changes; and (4) the 2015 water use
change simulated in this example is a simple approximation of actual changes. The original
Seacoast model was constructed using detailed 2004 water use data. However, since 2004
there are many new developments and large groundwater withdrawals that are not incorpo-
rated in this example. The location, magnitude, and timing of such water uses would need
to be incorporated into a calibrated model to provide realistic groundwater head and flow
simulations.

41
 70°50'0"W 70°40'0"W

0.96

0.99
70°40'0"W
1.03
43°0'0"N

Figure 2 area
Great Bay 1.04
1.2

Atlantic
Ocean

1.15
1.06

43°0'0"N 1.01
1.04

71°0'0"W

1.19

1.12 1.11
1

1.34
42°50'0"N

0.96
1
70°50'0"W

Hillshade base derived from National Elevation Dataset, 2011,

Simulated streams
NAVD 88, 1/3 arc-second, http://nationalmap.gov
Original model
0 3 6 Miles Revised model
Town outline and 2004-2015
population change factor
0 3 6 Kilometers
Model outline
(inactive dashed)
Figure 1. Simulated streams and approximate water-use changes in the Seacoast
model area, New Hampshire.

42
 70°50'0"W 70°48'0"W

Great Bay

43°2'0"N

Well

k
Broo
s
rry
Be

43°0'0"N

Hillshade base derived from National Elevation Dataset, 2011,

NAVD 88, 1/3 arc-second, http://nationalmap.gov
Simulated steady-state groundwater heads.
Model layer 3. 10-foot contour interval.
0 1 2 Miles 2004 (plotted over 2015
pathlines)
2015
0 1 2 Kilometers Stream
Town outline

Figure 2. Simulated steady-state groundwater heads in model layer 3 for 2004 and 2015
approximated water-use conditions, Seacoast model area, New Hampshire.

43
 References

Harbaugh, A.W., Banta, E.R., Hill, M.C., and McDonald, M.G., 2000, MODFLOW-2000, the U.S. Geological
Survey modular ground-water model – User guide to modularization concepts and the Ground-Water Flow
Process: U.S. Geological Survey Open-File Report 00-92, 121 p.
Mack, T.J., 2009, Assessment of ground-water resources in the Seacoast region of New Hampshire: U.S. Geo-
logical Survey Scientific Investigations Report 2008–5222, 188 p. Available online at http://pubs.usgs.gov/sir/2008/5222.
New Hampshire Office of Energy and Planning, 2016, State of New Hampshire County Population Pro-
jections, By Municipality. Accessed at http://www.nh.gov/oep/data-center/documents/2016-subcounty-
projections-final-report.pdf
Niswonger, R.G., Panday, Sorab, and Ibaraki, Motomu, 2011, MODFLOW-NWT, A Newton formulation
for MODFLOW-2005: U.S. Geological Survey Techniques and Methods 6–A37, 44 p. Available online at
http://pubs.usgs.gov/tm/tm6a37/
Poeter, E.P., Hill, M.C., Lu, Dan, Tiedeman, C.R., and Mehl, S., 2014, UCODE_2014, with new capabilities
to define parameters unique to predictions, calculate weights using simulated values, estimate parameters
with SVD, evaluate uncertainty with MCMC, and more: Integrated Groundwater Modeling Center Report
Number GWMI 2014-02, 172 p.
Winston, R.B., 2009, ModelMuse-A graphical user interface for MODFLOW-2005 and PHAST: U.S. Geo-
logical Survey Techniques and Methods 6-A29, 52 p.

44
Corresponding author: Madeline Nyblade
Pennsylvania State University
[email protected]

Numerical Modeling of the agricultural-hydrologic system in Punjab, India

Madeline Nyblade, Pennsylvania State University

T. Russo
L. Zikatanov
K. Zipp

Abstract: The goal of food security for India’s growing population is threatened by the
decline in freshwater resources due to unsustainable water use for irrigation. The issue
is acute in parts of Punjab, India, where small landholders produce a major quantity of
Indiaâ s food with declining groundwater resources. To further complicate this problem,
other regions of the state are experiencing groundwater logging and salinization, and are
reliant on canal systems for fresh water delivery. Due to the lack of water use records,
groundwater consumption for this study is estimated with available data on crop yields,
climate, and total canal water delivery. The hydrologic and agricultural systems are mod-
eled using MODFLOW-OWHM including the Farm Process. To our knowledge, this is the
first state-wide hydrologic numerical model of Punjab that accounts for multiple aquifer
layers, agricultural water demands, and interactions between the surface canal system and
groundwater. To more accurately represent the drivers of agricultural production and there-
fore water use, we couple an economic crop optimization model with the hydrologic model.
These tools will be used to assess and optimize crop choice scenarios based on farmer in-
come, food production, and hydrologic system constraints. The results of these combined
models can be used to further understand the hydrologic system response to government
crop procurement policies and climate change, and to assess the effectiveness of possible
water conservation solutions.

45
Corresponding author: Hai V. Pham
Division of Hydrologic Sciences, Desert Research Institute, 755 E. Flamingo Rd., Las Vegas, NV 89119
[email protected]

Grid generation and model calibration for complex fluvial aquifer system
including geological faults
Hai V. Pham, Division of Hydrologic Sciences, Desert Research Institute,
Frank T.-C. Tsai, Department of Civil and Environmental Engineering, Louisiana State University,
Karl F. Pohlmann, Division of Hydrologic Sciences, Desert Research Institute.

Abstract: This study introduces a framework to conduct groundwater modeling for com-
plex faulted fluvial-in-origin aquifer systems with a focus on grid generation and model
calibration – the two most challenging steps during groundwater model development. We
introduce a grid generation technique to create MODFLOW structured grids from complex
hydrostratigraphy. The study first uses well logs to construct a hydrostratigraphy with fine
vertical discretization through indicator kriging. Then, an upscaling procedure is introduced
to convert the hydrostratigraphy into a MODFLOW grid. To expedite model calibration,
we adopt the Covariance Matrix Adaptation-Evolution Strategy (CMA-ES) as it has many
advantages, such as global-local search, parallelization, and uncertainty quantification. The
framework is applied to developing a groundwater model for the Baton Rouge area, south-
eastern Louisiana that includes the Baton Rouge fault and the Denham Springs-Scotlandville
fault. By using 583 wireline well logs and the grid generation technique, a MODFLOW grid
of 162 layers were constructed, which preserves geological features, including pinch-outs and
narrow hydraulic connections through the

Extended abstract: Methodology, findings, and implications

1 Introduction
The Baton Rouge aquifer system in southeastern Louisiana (Figure 1a) resulted from fluvial
deposition, erosion, and faulting is a complex and non-uniform aquifer system [Chamber-
lain et al., 2013]. The study area includes two geological faults. The Denham Springs-
Scotlandville (DSS) fault is generally thought to have no significant effect on groundwater
flow. The Baton Rouge (BR) fault has been known as a horizontal flow barrier that sepa-
rates freshwater to the north and saline water to the south. Groundwater naturally flows
southward. However, excessive groundwater withdrawals in the area between two faults have
caused significant decline in groundwater level, saltwater intrusion, and land subsidence.
Developing a groundwater model has never been an easy task for the Baton Rouge
aquifer system. Fortunately, there are abundant well logs to help delineate the aquifer
system (https://sites.google.com/site/louisianawelllogportal/). This study aims to address
two questions: (1) How to utilize a large amount of well logs with highly irregular sand-clay
sequences to generate a proper computational grid for groundwater modeling; and (2) How to
calibrate such a complex groundwater model to ensure model integrity before a groundwater
model can be used for its applications.

46
2 GRID GENERATION
This study proposes four steps to generate a MODFLOW grid:
Step 1: Interpret wire-line well log data (spontaneous potential, electrical resistivity, and
gamma ray) and classify sand and clay facies into a binary sequence (sand facies is 1 and
clay facies is 0) for each well log.
Step 2: Perform indicator kriging to construct three-dimensional hydrostratigraphy to a
grid with a high vertical resolution.
Step 3: Upscale the hydrostratigraphy by merging the same hydrofacies in each vertical
column.
Step 4: Develop a MODFLOW grid based on the upscaled hydrostratigraphy.
The four steps is applicable to generating MODFLOW structured grids [Harbaugh, 2005]
or MODFLOW-USG unstructured grid [Panday et al., 2013]. This study focuses on con-
structing structured grid. Readers are referred to [Pham and Tsai, 2017] for more details on
the grid generation technique.

3 MODEL CALIBRATION
For complex aquifer systems, model calibration can be time consuming. This study adopts
the Covariance Matrix Adaptation-Evolution Strategy (CMA-ES) [Hansen and Ostermeier,
2001; Hansen et al., 2003] to calibrate a groundwater model and estimate model parameters
because CMA-ES has the capability of obtaining a near global solution and avoiding entrap-
ment in local optima. In addition, CMA-ES provides a full covariance matrix of estimated
parameters, which can be used to assess parameter estimation and prediction uncertainty.
Another advantage of CMA-ES is that it can be implemented in a high-performance comput-
ing system to overcome the prohibitive computational cost. The parallel CMA-ES [Elshall
et al., 2015] superiorly improves the calibration speed over the sequential CMA-ES.

4 APPLICATION
The methodology was applied to developing a groundwater model for the Baton Rouge
aquifer system utilizing 583 boreholes (Figure 1a). The resulted binary well log interpretation
(Figure 1b) shows complex sand and clay deposition in the area. It is impractical to manually
build correlations between boreholes. Using the aforementioned method, the model domain
in the planar direction is discretized into 93 rows and 137 columns. Cell size is 200 by 200 m.
Then, indicator kriging was used to construct three-dimensional sand-clay hydrostratigraphy
with a vertical resolution of one foot (nominally 0.304 m). The vertical range is from land
surface to -2890 ft (nominally -881 m) below the NGVD29 that covers the 11 sands in
the model domain. Finally, an upscaled hydrostratigraphy is obtained shown in Figure 1c.
The hydrostratigraphy indicates unconformity, interbeds, isolated beds, coalescences, and
pinch-outs.
A three-dimensional MODFLOW structured grid of 162 model layers and 808,078 active
computational cells was generated with layer thickness ranging from 3.05 to 30.5 m as shown

47
in Figure 2a. The model grid accurately matches the hydrostratigraphy and preserves layer
continuity.
Parallel CMA-ES was implemented to calibrate the groundwater model using 10,393
transient groundwater heads from 42 USGS observation wells (Figure 1a) from January 1,
1975 to December 31, 2014. Since the number of model parameters to be estimated was 38,
the study used a population size of 380. To maximize the efficiency of parallel computing,
380 cores were used to conduct model calibration. Given that the algorithm parallelization
time is less than 1 second per iteration, the speedup of the parallel CMA-ES is roughly
equal to the population size. A groundwater model run took about 18 hours. Using 380
processors, model calibration was completed in 75 days with 100 iterations. It would need
78 years without parallel computing.
The calibrated model showed a good match between observed and calculated groundwater
heads by the root mean square error (RMSE) of 3.76 m. The simulation result shows excessive
pumping in deeper aquifers (Figure 2a). The “2,000-foot” sand has the lowest groundwater
level owning to heavy industrial pumping. As a result, upward groundwater flow from
the “2,400-foot” sand to the “2,000-foot” sand is created through pumping wells that were
screened at both sands.
The water budget analyses for the sands between the two faults indicate that 70% of the
total inflow comes from the east and west boundaries of the model domain and about 30% of
total inflow comes through the two faults. It is estimated that 13% of the total inflow comes
from the Baton Rouge fault. Groundwater pumping is estimated about 59% of total outflow.
The water budget analyses also indicate that groundwater storage in 2014 was significantly
depleted as much as 4.9 billion gallons (nominally 18 million mϯ ) of groundwater since 1995
due to excessive groundwater pumping.
The flow paths (particle tracking) across the Baton Rouge fault indicate fault leaky areas.
The result shows the possibility that salty groundwater south of the Baton Rouge fault can
reach some pumping wells near the fault within 40 years as presented in Figure 2 (b, c and
d).

5 Acknowledgements
The study was supported in part by the U.S. Geological Survey (Grant No. G16AP00056),
Louisiana Board of Regents (Contract No. LEQSF(2015–18)-RD-B-03) and the Capital
Area Ground Water Conservation Commission. LSU High Performance Computing and LSU
Center for Computation and Technology are acknowledged for providing a supercomputer
for this study.

48
Figure 1. (a) The study area and the model domain (the thick black box), (b) sand and clay facies in
boreholes; and (c) hydrostratigraphic architecture. The vertical exaggeration is 20.

49
Figure 2. (a) Conceptual groundwater model for the Baton Rouge aquifer system and simulated groundwater
level in December 31, 2014 in 11 aquifers. Clay facies are blanked. Vertical lines are pumping wells with
screens (white segments); (b), (c), and (d) are 40-year particle tracking simulations from January 1, 1975
to December 31, 2014 for the “1,500-1,700-foot” sands, the “2,000-foot” sand, and the “2,400-foot” sand,
respectively. Contours are the simulated groundwater level in December 31, 2014.

References

Chamberlain, E. L., J. S. Hanor, and F. T.-C. Tsai (2013), Sequence stratigraphic characterization of the
Baton Rouge aquifer system, southeastern Louisiana, Gulf Coast Assoc. Geol. Soc. Trans., 63, 125–136.
Elshall, A. S., H. V. Pham, F. T.-C. Tsai, L. Yan, and M. Ye (2015), Parallel inverse modeling and uncertainty
quantification for computationally demanding groundwater-flow models using covariance matrix adaptation,
J. Hydrol. Eng., 20(8), 04014087, doi:10.1061/(ASCE)HE.1943-5584.0001126.
Hansen, N., and A. Ostermeier (2001), Completely derandomized self-adaptation in evolution strategies,
Evol. Comput., 9(2), 159–195, doi:http://dx.doi.org/10.1162/106365601750190398.
Hansen, N., S. D. Müller, and P. Koumoutsakos (2003), Reducing the time complexity of the derandomized
evolution strategy with covariance matrix adaptation (CMA-ES), Evol. Comput., 11(1), 1–18.
Harbaugh, A. W. (2005), MODFLOW-2005, The U.S. Geological Survey Modular Ground-Water Model -
the Ground-Water Flow Process, in Book 6. Modeling techniques, Section A. Ground Water, p. 253, US
Dept. of the Interior, US Geological Survey, Virginia.

50
Panday, S., C. Langevin, R. Niswonger, M. Ibaraki, and J. Hughes (2013), MODFLOW–USG Version 1: An
Unstructured Grid Version of MODFLOW for Simulating Groundwater Flow and Tightly Coupled Processes
Using a Control Volume Finite-Difference Formulation, U.S. Geological Survey Techniques and Methods,
book 6, chap. A45.

Pham, H. V., and F. T.-C. Tsai (2017), Modeling complex aquifer systems: a case study in Baton Rouge,
Louisiana (USA), Hydrogeol. J., 25(3), 601–615, doi:10.1007/s10040-016-1532-6.

51
Corresponding author: Andrew Rich
Sonoma County Water Agency
[email protected]

Future Baseline Modeling for the Sustainable Groundwater Management Act:

Prediction Uncertainty Resulting from Uncertainty in Climate Change and
Land Use Change

Andrew Rich, Sonoma County Water Agency

Abstract: The California Sustainable Groundwater Management Act passed in 2014 re-
quires a 50-year planning horizon for groundwater basins subject to the law. Future growth
in groundwater demands will continue to be influenced by socio-economic, technological, and
other factors. Producing a single estimate of change in future demands is therefore com-
plex and unreliable. A similar situation exists for future climate conditions. Our approach
to modeling the combined impact of future climate and growing groundwater demands on
groundwater resources in the Sonoma Valley groundwater basin is to combine a suite of
demand and climate scenarios to produce a total of 12 scenarios. This method attempts
to bracket the expected uncertainty inherent in both future demands and climate change.
These scenarios are implemented in MODFLOW-OWHM as integrated models that incor-
porate feedbacks between changing runoff and precipitation regimes, agricultural pumping,
surface water diversions, and evapotranspiration. Agricultural expansion for three growth
scenarios was achieved through the use of a logistic regression ranking system, and concomi-
tant demands are calculated internally in MODFLOW-OWHM. Output from a regional eco-
nomic model was used to forecast rural domestic growth and concomitant demands. Of the
12 scenarios, total groundwater storage change is negative for 5 scenarios, with a range of
+10,500 to -27,000 acre-feet. Results suggest that in hotter and drier climates the ground-
water system cannot capture greater inflows to offset increases in demands as effectively as
it does in the cooler and wetter climate scenarios. The diminished availability of streamflow
drives increased groundwater pumping more than the increased transpirative demands do.

52
Corresponding author: James Schneider
Olsson Associates
[email protected]

Addressing non-linear model response in the application of the Republican

River Compact groundwater model to estimate relative contributions to
streamflow depletion in a highly stressed aquifer

James C. Schneider, Olsson Associates

D. Ahlfield
C. Spalding
B. Dunnigan

Abstract: The Republican River Basin, located in Colorado, Kansas, and Nebraska, is
largely underlain by the Northern High Plains Aquifer. While this portion of the High
Plains Aquifer does receive significant amounts of recharge, the average rate of groundwater
withdrawals are much greater than the average rate of recharge. As a result, the aquifer and
baseflow to the Republican River have both been substantially depleted. In fact, in some
locations and during some years, stream baseflows are fully depleted, creating a significantly
non-linear hydrologic system due to the stream-drying and re-wetting that occurs over time.
The Republican River Compact requires the states of Colorado, Kansas, and Nebraska
to limit their use of Republican River streamflow, an issue that has generated significant
conflict between the states. Nebraska law requires Natural Resources Districts (political
subdivisions of the state) to form plans to manage the impact of the groundwater pumping
in their regions to both the aquifer and hydrologically connected streams. The Republican
River Compact groundwater model is used to estimate these relative impacts. Without an
modeling application that recognizes the non-linear nature of the system, estimates of these
relative impacts do not match the total impact and vary depending on the baseline pumping
level chosen for the analysis. This presentation will explore the various manifestations
of these misestimations and describe the procedures that have been adopted by the U.S.
Supreme Court to address the interstate dispute and the procedures adopted by Nebraska
for their intrastate water management goals.

53
Corresponding author: Sophia Sigstedt
Lynker Technologies
[email protected]

Conjunctive Use Analysis with the Enhanced Snake Plain Aquifer Model
(ESPAM)

Sophia C. Sigstedt, Lynker Technologies

Abstract: The Eastern Snake Plain Aquifer is a vital resource to the State of Idaho. The
Snake River Plain is a desert turned green by the flow of the Snake River and the ground-
water stored in the underlying basalts. The development of both groundwater and surface
water use across the Eastern Snake Plain and recognition of the hydrologic connection be-
tween the aquifer and Snake River has led to conjunctive management rules in Idaho. The
Snake River region supports several municipalities, agribusiness giants, extensive aquacul-
ture, hydropower production, numerous dairies and feedlots, as wells as millions of acres of
agricultural fields all supplied by the common water resource. The pressure on the regional
water resource has resulted in a decline of 10 million acre-feet in aquifer storage volume since
the 1980’s. The ESPAM is a MODFLOW model developed by the State of Idaho as a tool
to help quantify the impacts of current water use practices and/or proposed alternatives on
the common water resource of the State across the Eastern Snake Plain. This presentation
will highlight the use of ESPAM in analysis for a delivery call to the State by senior surface
water right holders against junior groundwater rights. The ESPAM was utilized to analyze
a reduction in groundwater diversions of 240,000 acre-feet and State sponsored recharge of
250,000 acre-feet annually to provide information on potential groundwater level increase
and a timeline for an ultimate groundwater level target and intermediary benchmark goals.
After more than a decade of litigation these parties came to a historic settlement agreement
framed around an adaptive management plan to stabilize and enhance Eastern Snake Plain
Aquifer levels and Snake River flows.

54
Corresponding author: Linda Woolfenden
U.S. Geological Survey
[email protected]

Assessment of management strategies to help mitigate aquifer depletion and

subsidence in the Bicycle Basin, Fort Irwin National Training Center,
California

Linda R. Woolfenden, U.S. Geological Survey

J. Densmore

Abstract: Groundwater pumpage in the Bicycle Basin, Fort Irwin National Training Cen-
ter, California exceeds natural recharge in the basin causing water-levels to decline by as
much as 100 feet and, as a result, subsidence. A groundwater-flow model of Bicycle Basin
was used to assess groundwater management strategies to help mitigate future water-level
declines and subsidence. Management strategies included managed aquifer recharge, re-
duction of pumping, and redistribution of pumping to parts of the basin without current
subsidence issues. For all strategies assessed, water levels continued to decline and the more
productive upper layers of the aquifer were dewatered; lower layers were forced to satisfy
withdrawals, resulting in higher rates of water-level declines and larger vertical gradients in
the future. Subsidence decreased slightly (4 to 5 percent) with managed aquifer recharge
compared with historical pumping conditions without managed recharge. The reduction of
pumpage in the area of subsidence and redistribution of the amount of the reduction to
wells in other parts of the basin resulted in slight (5 to 6 percent) decreases in projected
subsidence compared with continuation of historical pumping rates. However, including a
basinwide reduction of pumpage of 3 percent with reduction and redistribution of pumpage
or discontinuation of pumping in the area of subsidence demonstrated that subsidence could
be reduced substantially (22 to 26 percent and 62 to 68 percent, respectively) compared with
the continuation of historical pumping rates. The redistribution of pumping did not cause
subsidence in other areas because of the absence of fine-grained interbeds in those areas.

55
Corresponding author: Daniel Abrams
University of Illinois
[email protected]

How Watershed Characteristics Affect Transit Time Distributions

Daniel B. Abrams, University of Illinois

Henk M. Haitjema, Haitjema Consulting, Inc.

Abstract: The groundwater transit time distribution (TTD) for a watershed provides
insight into the migration of non-point source contaminants such as nitrate through an
aquifer. The most commonly used TTD is the gamma lumped parameter model (GLPM),
which requires calibration of shape and scale parameters to data (typically tritium). Setting
the shape parameter to one yields the exponential lumped parameter model (ELPM). The
ELPM does not require calibration, but depends on the average aquifer thickness, porosity,
and recharge rate. Whether the ELPM or GLPM is more appropriate appears to be related
to watershed characteristics. For watersheds where the water table mimics the topography,
MODFLOW simulations indicate that the GLPM can be calibrated to match the TTD. Such
topography-controlled systems are generally associated with relatively thick, low permeable
aquifers. In contrast, relatively thin, high permeable aquifers are more often recharge-
controlled with the water table mounding between streams. Here it is found that the ELPM
is a good approximation for the TTD. However, weak sinks (streams that do not receive
water from the entire aquifer depth) distort this ELPM by introducing a greater frequency of
very short as well as very long transit times. Unfortunately, the shape and scale parameters
of the GLPM can only be modified to match either the very young or old range of transit
times, but not both. So even in the presence of weak sinks, the ELPM is generally the best
choice for recharge-controlled watersheds, particularly at the regional scale where the effects
of weak sinks are muted.

56
Corresponding author: Tracy Nishiwaka
U.S. Geological Survey
[email protected]

Short-Term, Large-Scale Multi-Objective Optimization for Water-Resources

Management, Santa Barbara, CA

Tracy Nishiwaka, U.S. Geological Survey

Z. Stanko
S. Paulinski

Abstract: The city of Santa Barbara, located in coastal southern California, is concerned
that excessive groundwater pumping will lead to chloride contamination from seawater
intrusion (SWI). In addition, the city wishes to estimate the effect of continued pump-
ing on groundwater availability under a variety of climatic conditions. A SEAWAT-based
groundwater-flow and solute-transport model of the Santa Barbara groundwater basin was
developed and coupled with a multi-objective evolutionary algorithm. The optimization
problems were formulated as multi-objective so that the tradeoffs between maximizing
pumping, minimizing SWI, and minimizing drawdowns can be examined by the city. The
coupled simulation-optimization model identified Pareto-optimal management strategies for
3 different climate scenarios, which vary in their initial conditions and problem formulation.
Using the results, the city can make decisions on a pumping schedule based on operational
preferences and climatic conditions. The results for one of these scenarios allows the city to
investigate the effects of short-term climate variations on optimal pumping schedules. Using
the recent precipitation record, three specific 2-year simulations were optimized: typical-
to-dry; dry-to-typical; and dry-to-dry climates. The most notable result from the scenarios
tested is the significance of a typical versus dry year in the first year of the simulation, re-
gardless of the second year climatic condition. In general, it was found that using a typical
first year will result in schedules with larger pumpage volumes and lower maximum draw-
downs. In addition, it was found that SWI was relatively insensitive to short-term climate
variability when a dry first year is simulated.

57
Corresponding author: Jie Yang
GmbH - UFZ
[email protected]

Influences of a weir construction on freshwater resources of a coastal aquifer in

Germany under sea level rise

Jie Yang, Helmholtz-Zentrum fuer Umweltforschung GmbH - UFZ

T. Graf
T. Ptak

Abstract: A significant volume of an aquifer along the coastline in the German Bight has
been salinized due to seawater intrusion. A mean sea level rise (MSLR) is expected to
continue in future due to global climate change, subsequently degrading the fresh ground-
water resources. To stop extra salinization in future, we propose a solution based on a weir
construction in an existing canal hydraulically connected to this aquifer. The effect can
be twofold: (i) the groundwater level can be elevated due to the elevated water level in
the canal, expanding surface water ponds, (ii) the elevated groundwater level can upgrade
present fresh groundwater resources by shifting the saltwater-freshwater interface position
further seaward, or by inhibiting its landward movement. We employed a fully coupled
3D numerical surface-subsurface model, using a modified HYDROGEOSPHERE code, to
investigate the effects of variable weir construction heights under different MSLR rates,
and to help finding an optimal relation of fresh aquifer volume gains to costs of weir con-
struction and land loss due to surface ponding. A higher weir construction height increases
the desalinized aquifer volume, or decreases the new salinized aquifer volume under future
MSLR. Depending on the MSLR rate, a minimum height of a weir was calculated, such that
the present freshwater resource can be maintained. Both weir construction and MSLR can
cause new ponding on land surface, representing loss of land usage. A simulated loss-gain
ratio can be utilized to determine the optimal weir height, meeting economical requirements
of coastal land management under future MSLR.

58
Corresponding author: Eduardo De Sousa
Pells Sullivan Meynink
[email protected]

Fast assessment of pore pressures and inflows in open pit slopes using smart
models

Eduardo De Sousa, Pells Sullivan Meynink

Abstract: Analytical solutions for estimation of pore pressures in open pit slopes and pit
inflows are an attractive solution for preliminary assessments and studies of limited scope.
However, oversimplification due to inherent assumptions of these methods, particularly in
regards to storage terms and transient behaviours, commonly fail to identify peak inflows and
pore pressures throughout the mining excavation period. The use of numerical flow models,
in relation to their analytical counterparts, requires a substantial effort during model setup
and running times. In order to overcome this, this paper presents a methodology for faster
implementation of numerical flow models using smart models, i.e., flow models that can be
built automatically based on basic information provided by the user. A general FEFLOW
mesh has been generated to be capable of simulating a large variety of pit geometries and
aquifer configurations. With the use of a FEFLOW plugin developed for this purpose,
the user is only required to provide details on pit geometry (such as slope angles and pit
depths) and aquifer parameters in a simple text file. The plugin reads this file and setup
the model accordingly, creating an entire model instantaneously. Auxiliary Python scripts
were also developed to execute multiple runs in context of sensitivity analysis. While this
approach does not fully utilize the flexibility provided by numerical models, it provides
a more comprehensive solution in relation to analytical methods, by taking into account
transient behaviour, particularly in relation to excavation rates. Furthermore, this general
models can be manually modified to refine particular hydrogeological elements of interest.

59
Corresponding author: Christopher DeMarco
EarthRes Group
[email protected]

Simulation of High Quality Wetland Protection Measures Concurrent with

Surface Mine Dewatering Utilizing GMS-MODFLOW

Christopher J. DeMarco, MS, EarthRes Group

Matthew S. Weikel, MS, PG, EarthRes Group
Louis F. Vittorio, Jr., MS, PG, EarthRes Group

Abstract: Impacts to wetlands and surface water sources are a common concern when
dealing with active quarries and the resulting aquifer drawdown. This study focuses on the
steady state MODFLOW simulation of a Greenfield quarry site and the proposed measures
necessary to preserve hydrology in and around nearby streams and wetlands. The aquifer
was modeled utilizing field measurements of: groundwater elevation; hydraulic character-
istics from aquifer testing; intra-seasonal wetland and channel flow measurements; vertical
head gradients from piezometers placed in and around wetlands, and overburden infiltration
rates. Field studies focused on characterizing gaining and losing conditions along stream and
wetland extents in order to model ground and surface water interaction to a high degree of
accuracy. The steady state, pre-development model simulation was calibrated using PEST
by varying recharge and conductivity parameters. The calibrated model was then used to
simulate quarry pumping, aquifer drawdown and the effects on surface water budgets. Mit-
igation of impacts was simulated by infiltration of quarry discharge and captured stormflow
into the aquifer via a system of infiltration trenches. The infiltration trenches were located
by assessing surface water budgets and aquifer drawdown in order to truncate the draw-
down cone produced by quarry pumping. By distribution of stored site water amongst the
wetland systems throughout the site, the model simulations demonstrate that adjacent sur-
face hydrology can be preserved during quarry dewatering operations. The model assessed
the need for ongoing monitoring points and helped establish criteria for measurement of
potential impacts.

Extended abstract: Methodology, findings, and implications

Impacts to wetlands and surface water sources are a common concern when dealing with
actively pumping quarries and the resulting aquifer drawdown. This study focuses on the
steady state MODFLOW groundwater simulation of a proposed quarry site in southeastern
Pennsylvania and the measures developed to preserve the hydrology in and around nearby
streams and wetlands.
The modeled aquifer lies within a Triassic fanglomerate along the edge of an intrusive
Triassic diabase sill. The aquifer was modeled utilizing field measurements of: groundwater
elevation; hydraulic characteristics from aquifer testing; seasonal wetland and channel flow
measurements; vertical head gradients from wetland piezometers; and overburden infiltration
rates derived from site testing. Field studies focused on characterizing gaining and losing

60
conditions along the stream and wetland extents in order to model ground and surface water
interaction to a high degree of accuracy.
The steady state, pre-development model simulation (Figure 1.) was calibrated using
PEST by varying recharge and conductivity parameters. The calibrated model was then
used to simulate quarry pumping, aquifer drawdown and the effects on surface water bud-
gets. Mitigation of impacts was simulated by infiltration of quarry discharge and captured
stormflow into the bedrock aquifer via a system of infiltration trenches. Although overburden
infiltration rates were greater than bedrock infiltration rates, the modeled infiltration was set
to the lower conductivity of the bedrock as that would be the limiting factor governing water
recharge to bedrock. The infiltration trenches were located by assessing surface water bud-
gets and aquifer drawdown in order to mitigate the drawdown produced by quarry pumping.
By distribution of stored site water to the wetland systems throughout the site, the model
simulations demonstrated that adjacent surface hydrology can be preserved during quarry
dewatering operations using infiltration trenches. The model assessed the need for ongoing
monitoring points and assisted in establishing criteria for measurement of potential impacts.
Following generation of the model, additional field data was collected to better character-
ize vertical flow and hydrology within site overburden and site wetlands. With this new data
and the previous model, a more refined site scale model was generated (Figure 2.) by re-
building the model grid to simulate an overburden unit. This was completed upon regulatory
request to see if modeling of infiltration to the higher conductivity overburden would produce
a different result than modeling infiltration to the more restrictive rate of the bedrock. The
scale model was developed using the head output from the larger initial model simulation
by setting the boundaries to specified head. The site scale model was recalibrated to reflect
adding an overburden of higher conductivity than the underlying bedrock. Simulations of
quarry deepening and infiltration trenching were generated in an identical manner as the
previous model and results were compared. The refined model showed nearly identical head
and flow values as the larger scale model. However, the refined model allowed for comparison
of observed and computed overburden water levels in and around site wetlands. The results
show that, in both scenarios, infiltration of quarry discharge water can be used to mitigate
aquifer drawdown impacts and preserve stream and wetland features.

61
Figure 1. Steady State, Pre-development Model Simulation.

Figure 2. Refined Site Scale Model

References

62
Aquaveo, LLC. Groundwater Modeling System (OMS). Version 10.0, Provo, UT 84602, 2015. http://www.aquaveo.com/softwa
modflow
Harbaugh, A.W., Edward R. Banta, Mary C., and Michael G. McDonald, 2000, MODFLOW- 2000, The
U.S. Geological Survey Modular Ground-Water Model-User Guide To Modularization Concepts And The
Ground-Water Flow Process. U.S. Geological Survey- Open File Report 00-92.
Hill, M.C., et al. MODFLOW-2000, The U.S. Geological Survey Modular Ground-Water Model – User Guide
to the Observation, Sensitivity, and Parameter-Estimation Processes and Three Post-Processing Programs.
U.S. Geological Survey Open-File Report 00-184, 2000.
Longwill, S. M., and C.R. Wood. Groundwater Resources of the Brunswick Formation in Montgomery and
Berks Counties, Pennsylvania. Pennsylvania Geologic Survey, 4th Series, Water Resources Report 22, 1965.
Low, D.J., et al. Geohydrology of Southeastern Pennsylvania. Water Resources Investigation Report, PA
DCNR Bureau of Topographic and Geology Survey, 2002.
Shultz, Charles. The Geology of Pennsylvania. Pennsylvania Geological Survey and Pittsburgh Geological
Survey, Special Publication 1, 1999.

63
Corresponding author: Alan Jang
Itasca Denver, Inc.
[email protected]

Comparison on Groundwater Model Simulations of Open-Pit Mining Using

Three Numerical Codes: MODFLOW-USG, MODFLOW-SURFACT, and
MINEDW

Alan Jang, Itasca Denver, Inc.

Houmao Liu
Jianwei Xiang
Steve Meyerhoff

Abstract:

Extended abstract: Methodology, findings, and implications

Groundwater flow modeling is extensively used to estimate dewatering requirements and

pore-pressure distribution for slope-stability analyses, to design water-management plans,
and to assess potential environmental impacts for mining operations. The widely-used
groundwater codes include MODFLOW-SURFACT, MODFLOW-USG, and MINEDW.
In this study, these three codes are utilized to simulate seepage during open-pit excavation
and the subsequent formation of a pit lake for a hypothetical open-pit. The model domain
and grids for the codes are shown in Figures 1 and 2, respectively. Comparisons of these three
codes were made by simulating pit excavation, the development of a zone of relaxation due
to the stress unloading of excavated rock mass, pore-pressure distribution, and the formation
of a pit lake.
All three codes are capable of producing comparable results; however, MINEDW can sim-
ulate the groundwater conditions to meet the requirements of dewatering, inputs to stability
analyses, environmental impact assessments, and pit-lake geochemistry more efficiently than
the other two codes. The main advantages of MINEDW over the other two codes are that
it is capable of simulating temporal changes in pit-shell configuration and the associated
changing permeability value of the zone of relaxation over time. It can also easily provide
the pore-pressure distribution in both the two-dimensional section and in three-dimensional
volume.

64
Figure 1. Schematic Diagram of the Model Domain and Boundary Conditions

Figure 2. Numerical Model Grids

References

65
Corresponding author: Dirk Kassenaar
Earthfx Incorporated
[email protected]

Assessing the Cumulative Effects of Groundwater Withdrawals for Oils Sands

Production on a Watershed Scale
Dirk Kassenaar, Earthfx Incorporated
E.J. Wexler
P.J. Thompson
M.G.S. Takeda

Abstract: The MacKay River Watershed, northwest of Fort McMurray, Alberta, Canada,
is a diverse oil sands development region with numerous steam-assisted gravity drainage
(SAGD) operations planned or in production. Freshwater withdrawals (diversions) from
channel aquifers and deeper confined formations are needed to support the SAGD opera-
tions. Land modification for access roads, well pads, and processing facilities will also alter
surface water flow and groundwater recharge. Earthfx developed an integrated surface wa-
ter/groundwater model using the USGS GSFLOW code to assess the cumulative impacts of
SAGD operations on the MacKay watershed. The entire watershed, including 17 subsurface
hydrogeologic units, over 4000 km of streams, and all mapped lakes, wetlands, land alter-
ation and water diversions, was represented. The GSFLOW code was modified by Earthfx
to include representation of ground freeze/thaw processes that significantly affect spring
runoff and groundwater recharge processes. Multiple scenarios representing groundwater
diversions and land alteration under pre-development, current and future development con-
ditions were simulated on a daily basis over a representative 25-year climate period. Large
scale drawdowns were noted in the upper part of the watershed in the deep confined aquifers.
Significant, but localized, drawdowns developed due to cumulative pumping from the shal-
lower channel aquifers. Induced leakage from the stream network and inflows from adjacent
aquifers were found to mitigate these effects; however, the induced reduction in stream-
flow exceeded 15ύ in some reaches. The application of an integrated model provided key
insights into the overall function of the watershed and the complex interactions between
components of the hydrologic system. The analysis indicated that projected water use in
the study area is broadly sustainable on a watershed scale; however, ongoing efforts will be
needed to mitigate effects.

66
Corresponding author: Seann McClure
Aspect Consulting
[email protected]

Groundwater Modeling to Support Wetland Restoration of a Former Peat Mine

Seann McClure, Aspect Consulting

P. Bannister

Abstract: A former peat mine-turned-suburban lake that has historically received stormwa-
ter discharge impacted by dioxin/furans is being restored to a scrub-shrub wetland to re-
mediate contaminated lake sediments through capping and filling in the lake. A secondary
goal of the project is to remove open-water lake conditions to deter water fowl and meet
Federal Aviation Agency (FAA) requirements. Because of the project’s location within the
Seattle-Tacoma International Airport, the water fowl are a hazard to the safe operation
of commercial aircraft. Project goals required that multiple and wide-ranging hydrologic
conditions be considered, including minimizing effects on protected stream flows, demon-
strating efficacy of contaminant remediation, and promoting ecological function to meet
Agency requirements. Therefore, a site-scale MODFLOW 2005 groundwater model with
NWT solver was employed to inform design. The model utilized regional modeling results
to inform boundary conditions, and incorporated the effects of peat consolidation due to
fill placement. Simulating critical surface water/groundwater interaction was dependent on
detailed topographic and bathymetric data. Baseline hydraulic monitoring results were used
in model calibration. Groundwater modeling identified fill material characteristics and en-
gineered wetland topography that would maintain upward hydraulic gradients through the
impacted sediment and cap, and promote a scrub-shrub wetland without open-water con-
ditions that might attract water fowl. The sediment cap and wetland restoration is part of
a much larger environmental remediation and construction effort that dictated tight sched-
ule and budget constraints. The presentation of groundwater modeling results supported a
streamlined regulatory approval process.

67
Corresponding author: Will Minchin
Watershed HydroGeo
[email protected]

Use of MODFLOW-USG and Connected Linear Networks (CLN) to represent

Deformation and Fracturing above a Longwall Mine

Will Minchin, Watershed HydroGeo

S. Brown

Abstract: Longwall mining results in ground subsidence and fracturing of overlying strata.
A series of disturbance zones in the overburden are home to a variety of changes to local
groundwater systems, including changes to permeability and storage. Groundwater models
must represent these changes to hydraulic properties in order to predict the effects of longwall
mining. Several methods are currently used to represent these changes to strata properties
particularly the enhancement of vertical permeability, e.g. MODFLOW Drains, ”time-slice”
models or time-varying properties - TMP/TVM. The development of MODFLOW-USG has
allowed the authors to use Connected Linear Networks (CLN) to broadly simulate the net-
work of connected sub-vertical fractures above longwalls. The method employed involved
setting thousands of CLNs, one per cell, extending upward to the top of the zone of con-
nective fracturing. This method has the conceptual advantage of allowing the model to
simulate both matrix and fracture flow in the overburden. However, it is complicated by
several factors - primarily that a single CLN per cell must approximate the many fractures
present, and that vertical fracturing is not the only mode of disturbance; bedding plane
partings and inclined fractures occur throughout the overburden. Additional CLN elements
may be required to fully represent mining-related fracturing. Our experience is that model
instability and numerical artefacts may occur, however the method has been broadly suc-
cessful in representing mine inflow and enhanced drawdown above mine panels, as well as
simulating the age of water entering the mine, and is conceptually more appropriate than
alternative methods.

68
Corresponding author: Christopher Pantano
WSP | Parsons Brinckerhoff
[email protected]

Bridging the gap: Coupling of geotechnical and groundwater models in the

mining industry

Christopher Pantano, WSP | Parsons Brinckerhoff

D. Edington
G. Meza-Cuadra

Abstract: The mining industry presents unique challenges for groundwater modeling
projects given the frequent and continuous changes to surface and subsurface ground condi-
tions. Mining practices, such as underground tunneling, blasting, block cave mining, back-
filling, etc., alter rock and change the hydraulic properties of the affected rock mass over
time. The TMP package (time-varying material properties), developed within MODFLOW-
SURFACT, provides a valuable tool for simulating the transient effects of mining activities
and determining the impacts of these changes. Although simplistic assumptions regarding
hydraulic property changes can be approximated, collaboration with geotechnical engineers
allows for more detailed estimation of bulk rock-mass hydraulic behavior changes through
time. Utilization of these data is valuable in defining the extent and timing of hydraulic
property changes with the goal being to improving model predictions. A hypothetical exam-
ple is presented for discussion regarding the integration of 4D geotechnical model data into
4D material property changes, which can then be easily incorporated into the MODFLOW-
SURFACT TMP package. Specifically, a sample geotechnical dataset, providing plastic
strain and rock-mass volume changes, is converted into TMP multipliers and applied to
hydraulic conductivity and specific yield values at numerous model times.

69
Corresponding author: Gareth Price
BHP Billiton
[email protected]

Past, present and future predictions of mine dewatering rates in the Pilbara
region of Western Australia

Gareth Price, BHP Billiton, Perth, Australia

Abstract: The Central Pilbara region of Western Australia is home to several active and
proposed below water table iron ore mines. Since the decision was taken to mine one of
these deposits 20 years ago, numerous groundwater models have been produced to predict
dewatering requirements. The models used to undertake these assessments and the predic-
tions that they produced have evolved considerably in that time, although the basic building
block (MODFLOW), has remained unchanged. The predictions have varied considerably,
along with data availability, the conceptual model, mine plans and attitudes to uncertainty.
The discussion will describe this evolution and explore the reasons behind the changes.

Extended abstract: Methodology, findings, and implications

Introduction
BHP Billiton (BHPB) operate numerous below water table (BWT) open pit iron ore
mines in the Pilbara region of Western Australia. Numerical groundwater flow models have
been, and are continued to be used, to provide:

• dewatering predictions for:

– water balance calculations (i.e. mine water surplus and deficit)

– planning of infrastructure (i.e. pipe sizes, wellfield layout and size) and
– guidance to the mine plan

• predictions of any changes to the local and regional groundwater system from ground-
water management activities for:

– government approval, and

– guidance to the mine plan

The subject of this short paper is a 20 year history of dewatering predictions at a single
mine. Both the modelling methodology and the model predictions have evolved significantly
over that time. In fact, the latest predictions have changed beyond recognition from those
first produced in 1997. Understanding the reasons behind these changes provides a unique
opportunity to improve our future modelling methods and expectations.

The mine
The orebodies that make up the mine have been deposited along one limb of an east –
west trending anticline. The orebodies occur over a length of about 20 km. They are several
hundred metres in width and form local aquifers. A laterally extensive regional aquifer is

70
found to the north of the orebodies. The local and regional aquifers are separated by shales.
Several north – south trending dykes cut through the entire system and may or may not
impede groundwater flow, which is from west to east.
The pits vary in depth but several are planned to be in excess of 100 m BWT. Mining
started in 2003 with the first BWT mining in 2010.

The predictions
A model was first developed in 1997. This has been updated, rebuilt and recalibrated
many times since. This analysis will focus on only one output; the total volume of dewatering
over the life of the mine. The major predictions since 1997 are shown Figure 1. This shows
that the most recent prediction is about 200 times greater than the first.
Uncertainty in these predictions has been assessed using judgement alone. Only in the
case of the most recent predictions was the model itself used to test the uncertainty and this
was done by manually producing calibrated alternatives.

The conceptual model

For the first 13 years, the conceptual model assumed a set of disconnected local bathtub
orebody aquifers. Dewatering was therefore required to remove groundwater storage prior
to mining, but with limited inflow from the regional aquifer. As the years progressed, the
connection of the local aquifers to the regional aquifer system has been revised. The latest
conceptual model assumes connection between all the aquifer systems in the area.
The bathtub conceptual models were developed when the only available data was pseudo-
steady state, complemented gradually by short-duration test pumping and some water supply
abstraction. Only after dewatering commenced did truly revealing data become available to
critically test, and reject, the original hypothesis.
Another aspect of the conceptual modelling was that, until the latest iteration, only one
version was progressed at a time.

The numerical model

The numerical model started off relatively simple using a telescopic mesh with smallest
cells 100 m2 and with two layers (with a total of 13,000 cells). The model was completely
rebuilt in 2005. It was extended to cover more area, had two layers added and a smaller
(now uniform) grid of nearly 2 million 100 by 50 m cells. However, the 2005 predictions were
roughly the same order of magnitude as in 1997.
For both versions the major calibration data inputs were steady state heads. The 1997
model achieved a Scaled Root Mean Squared (SRMS) error of 3% and was judged by external
peer review as being “well calibrated and fit for purpose”. The 2005 version achieved a 2%
SRMS.
Insights from this include:

• the acceptable SRMS error was entirely misleading,

• single predictions early on in the project life are extremely uncertain,

• creating a far more complex numerical model did not reduce the uncertainty and made
robust uncertainty analysis difficult.

71
 The mine plan
Mining hydrogeologists are well aware that mine plans change. However, the changes that
occurred to this mine plan were considerable (including mine life, pit depth and breadth and
mining rate). This goes a long way to explaining some of the changes in predicted dewatering.
For example the planned BWT mining duration was:

• 1 year in 1997

• 22 years in 2005

• 45 years in 2014

Economically driven changes are hard to anticipate. However, more predictable changes
include the size and extent of orebodies, which tend to increase with time. In this case about
30 m extra BWT depth was added to the mine plan between 1997 and 2008.

The present
BHPB’s present modelling methodology is based on the following:

• Models simplified in time and space and running in a few minutes

• Automated uncertainty analysis used to:

– Investigate a wide range of hydraulic parameters of the key hydrostratigraphic

units and rainfall recharge rates.
– Generate hundreds of calibrated model versions which are then used to produce
hundreds of predictions.

• Predictions presented as ranges and percentiles

Our ambition is now twofold:

1. To provide predictive outcomes that reflect the hydrogeological uncertainty.

2. To reduce that range as more information is collected.

The future
From a hydrogeological perspective, the uncertainty in hydrostratigraphy and high level
conceptual components such as impermeable structures, boundary inflows and 3rd party
activities needs to be included in our assessments.
From a mine plan perspective it is clear that the analysis of uncertainty might benefit
from including some upside to the contemporary geological appreciation of the orebody
geometry, especially early on in the mine life. However, the apparent unpredictability of
economic factors that control commodity prices and therefore mine plans, mean that we
should be more cautious when considering these types of uncertainty.

72
 Figure 1. The evolution of life of mine total dewatering volume predictions

References

73
Corresponding author: Ned Banta
U.S. Geological Survey
[email protected]

Upgrading Your Model to MODFLOW 6

Ned Banta, U.S. Geological Survey

Abstract: The form of input for the U.S. Geological Survey’s new MODFLOW 6 ground-
water model is substantially different from the input format for all earlier versions of MOD-
FLOW. In the new MODFLOW 6 format, input will be easier to understand for model users
preparing and reading the input files; however, reformatting will be required to update input
for a model prepared for an earlier version.
A converter utility is available to convert an existing model based on earlier versions of
MODFLOW to MODFLOW 6. This poster will address such questions as:
• What MODFLOW versions can be converted?
• How do packages of MODFLOW 6 correspond to packages of earlier versions?
• Which MODFLOW packages can be converted?
• What can I do if my model is based on a version of MODFLOW not supported by
the converter utility or if my model uses packages that are not supported?

74
Corresponding author: Sahila Beegum
University of California, Riverside/Indian Institute of Technology Madras
[email protected]

Integration of Solute Transport and Water Flow Models for Unsaturated and
Saturated Soil Zones using the HYDRUS Package for MODFLOW and
MT3DMS
Sahila Beegum, University of California, Riverside/Indian Institute of Technology Madras
Jirka Šimůnek, University of California, Riverside
Adam Szymkiewicz, Gdańsk University of Technology
K P Sudheer, Indian Institute of Technology Madras
Indumathi M Nambi, Indian Institute of Technology Madras

Abstract: Deterioration of the ground water quality is a global concern that adversely af-
fect the environment and living beings. There are many ways in which soil and groundwater
can get contaminated. These include residential, municipal, commercial, industrial and agri-
cultural activities which disposes waste water to the soil. The movement of the contaminant
take place through unsaturated and saturated zone of the ground water, and hence there
will be an interaction between the two zones. Because of the complexity associated with
modelling the interaction between saturated and unsaturated soil zone, modelling of fate
and transport of contaminant in soil and groundwater has been a research focus for many
years. Also, many of the existing mathematical models, that are capable of simulating the
water transport through saturated and unsaturated zone, do not have the facility to simulate
the solute transport. A promising approach for comprehensive study of movement of water
and solute through unsaturated zone and saturated zone, is possible linking of indepen-
dent models for these two zones. MODFLOW (Modular three-dimensional finite-difference
groundwater model - Harbaugh et al.,2000) developed by the U.S Geological Survey is one
of the most widely used groundwater flow model. In this model, the major input is the
recharge to the groundwater. Owing to the complexity associated with the estimation of
ground water recharge, most of the time, this major input is highly simplified and general-
ized to a uniform value for the whole study area. This consideration is an oversimplification
of the process, and may affect the impact of unsaturated flow on the groundwater. A cou-
pled model ’HYDRUS-based flow package for MODFLOW’ was developed by Simunek et
al. (2008), which used HYDRUS 1D as the unsaturated zone model, and was linked to
the MODFLOW. The water flow part of HYDRUS-1D has been used in the coupled model
to represent the effects of unsaturated zone processes. Being fully incorporated into the
MODFLOW program, the HYDRUS package provides MODFLOW with recharge fluxes at
the water table, while MODFLOW provides HYDRUS with the position of the groundwater
table that is used as the bottom boundary condition. The performance of this model was
analyzed by Twarakavi et al. in 2008, and validated with observed field data. The coupled
model is effective in addressing spatially variable saturated-unsaturated hydrological pro-
cesses at regional scale with complex layering in the unsaturated zone and with alternating
recharge and discharge fluxes (Twarakavi et al., 2008).

75
Corresponding author: Scott Boyce
U.S. Geological Survey
[email protected]

MODFLOW-OWHM v2: New Features and Improvements; The Next

Generation of MODFLOW Conjunctive Use Simulation

Scott E. Boyce, U.S. Geological Survey

Randall T. Hanson, U.S. Geological Survey
Ian Ferguson, U.S. Bureau of Reclamation
Thomas Reimann, Technische Universitaet Dresden
Wesley Henson, U.S. Geological Survey
Steffen Mehl, California State University, Chico
Wolfgang Schmid, CSIRO Land and Water
Stanley Leake, U.S. Geological Survey
Thomas Maddock, University of Arizona

Abstract: The MODFLOW-2005 (MF) family of hydrologic simulators has diverged into
multiple versions designed for specific needs, thus limiting their use to their respective de-
signs. The One-Water Hydrologic Flow Model (One-Water or MF-OWHM) is an integrated
hydrologic flow model that is an enhanced fusion of multiple MF versions. While main-
taining compatibility with existing MF versions, One-Water includes: linkages for coupled
heads, flows, and deformation; facilitation of self-updating models, additional observation
and parameter options for higher-order calibrations; and redesigned code for faster simula-
tions. The first version was selected by The World Bank Water Resource Software Review,
in 2016, as one of three recommended simulation programs for conjunctive use and man-
agement modeling. It also was selected as the primary simulation engine for FREEWAT,
a European Union sponsored open-source water management software environment. This
release of One-Water incorporated MODFLOW-2005 and the Farm Process (MF-FMP2),
with new features, combined with Local Grid Refinement (MF-LGR), Streamflow Routing
(SFR), Surface-water Routing Process (SWR), Seawater Intrusion (SWI), Riparian Evapo-
transpiration (RIP-ET), the Newton Formulation (MF-NWT), and more. The next version
of One-Water v2 will include a new surface-water operations module that simulates dynamic
reservoir operations, a conduit-flow process for karst aquifers and leaky pipe networks, a
soil zone process that adds an enhanced infiltration process, interflow, deep percolation and
soil-moisture budgets, and a new subsidence and aquifer compaction package. The Farm
Process base code has been significantly rewritten for improved speed, easier user input,
easier feature extensibility, and new features. One-Water represents a complete integrated
hydrologic model that fully links the movement and use of groundwater, surface water, and
imported water for consumption by agriculture and natural vegetation on the landscape,
and for potable and other uses. By retaining and keeping track of the water during simula-
tion of the hydrosphere, One-Water accounts for “all of the water everywhere and all of the
time.” This provides the foundation needed to address integrated hydrologic problems such
as evaluation of conjunctive-use alternatives and sustainability analysis.

Extended abstract: Methodology, findings, and implications

76
The simulation of sustainable conjunctive management of water resources by using MODFLOW-
2005 (Harbaugh, 2005) (MF) required extending its capabilities of to analyze the movement
and use of water throughout the hydrologic cycle in a process-based context. These ex-
tensions led to the transformation of MF to a fully-coupled, integrated hydrologic model
(IHM) that can simulate the complete movement and use of water across the land surface
in conjunction with the surface-water and groundwater systems. This type of MF-based
simulation was first attempted by the development GSFLOW (Markstrom and others, 2008)
and MF-FMP (Schmid and others, 2006, 2009). While GSFLOW connects MODFLOW to
the precipitation-runoff model PRMS (Leavesley and others, 1983) to simulate the coupling
between groundwater flow and surface flows, it lacks a connection to human infrastructure
and related landscape processes and a supply-and-demand structure needed for sustainable
management and conjunctive-use analysis. This shortcoming prompted the development
of MF-FMP, which provided a connection to both natural and anthropogenic water uses
throughout the simulated hydrosphere in the context of a supply-and-demand framework.
The comprehensive simulation of water use includes the application, consumption, and move-
ment of water for natural vegetation, agriculture, and urban settings across the land surface.
Typically, integrated water use simulations need to account for the movement and use of
all of the water in the hydrosphere and needs to simulate any associated secondary effects.
By account for all the water provides a more complete simulation and understanding of each
flows inter-dependence. The secondary effects include groundwater-dependent ecosystems
(GDEs), seawater intrusion, land subsidence, and salt and nutrient transport related to
agricultural practices. Any of these effects may become the controlling factors to resource
development, conjunctive use, and sustainable management.
To analyze a broader class of hydrologic settings and related water-resource issues many
specialized versions of MF have evolved in the last ten years. Although these multiple
versions of MF are currently available, the analysis of conjunctive use and other water-
management methods was prohibitive without running multiple models that contained some
of the features needed to adequately simulate these methods. This deficiency led to the
development of the One-Water Hydrologic Flow Model (One-Water or MF-OWHM) that
includes the Farm Process version 3, the Riparian ET Package (RIP; Maddock and others,
2012), and combines the features from MF-2005 (Harbaugh, 2005), MF-NWT (Niswonger
and others, 2011), MF-SWR (Hughes and others, 2012), MF-SWI (Bakker and others, 2013),
and MF-LGR (Mehl and Hill, 2005). One-Water maintains backward compatibility with the
MF versions, so that models built in these previous versions can run within the One-Water
framework.
The first One-Water version, released in 2014, was selected by The World Bank Water
Resource Software Review in 2016 (Borden and others, 2016) as one of three recommended
simulation programs for conjunctive use and management modeling. One-Water is also
being used as the primary simulation engine for FREEWAT, a European Union sponsored
open-source water management software environment (De Filippis and others, 2017).
The next version of One-Water will include a new surface-water operations module that
simulates dynamic reservoir operations, a new sustainability analysis package that facilitates
the estimation and simulation of reduced storage depletion and captured discharge, a conduit-
flow process for karst aquifers and leaky pipe networks, a soil zone process that adds an
enhanced infiltration process, interflow, deep percolation and soil-moisture budgets, and a

77
new subsidence and aquifer compaction package. It will also include additional features to
facilitate easier model updates, faster execution, better runtime error messages and reporting,
and more integration/cross communication between the traditional MODFLOW packages.
The new structure also helps facilitate the new integration into a “Self-Updating” structure
of data streams, simulation, and analysis needed for modern water resource management.
By retaining and tracking the water within the hydrosphere, One-Water accounts for
“all of the water everywhere and all of the time.” This philosophy provides more confidence
in the water accounting to the scientific community and provides the public a foundation
needed to address wider classes of problems. Ultimately, more complex questions are being
asked about water resources, requiring tools that more completely answer conjunctive-use
management questions.

References

Borden, J.C., Gaur, A., Singh, C. R., 2016. Water resource software: application overview and review. Wash-
ington, D.C.: World Bank Group. http://documents.worldbank.org/curated/en/279391467995445292/Water-
resource-software-application-overview-and-review
De Filippis, G., Borsi, I., Foglia, L., Cannata, M., Velasco Mansilla, V., Vasquez-Suñe, E., Ghetta, M., Ros-
setto, R., 2017. Software tools for sustainable water resources management: the GIS-integrated FREEWAT
platform. Rendiconti Online Società Geologica Italiana, Vol. 42 (2017), pp. 59-61, doi:10.3301/ROL.2017.14
Harbaugh, A.W., 2005. MODFLOW-2005. The U.S. Geological Survey modular ground-water model — the
Ground-Water Flow Process: U.S. Geological Survey TM6-A16, v.p.
Leavesley, G.H., Lichty, R.W., Troutman, B.M., Saindon, L.G., 1983. Precipitation-Runoff Modeling
System–User’s Manual: U.S. Geological Survey Water Resources Investigations Report 83-4238.
Hanson, R.T., Boyce, S.E., Schmid, Wolfgang, Hughes, J.D., Mehl, S.M., Leake, S.A., Maddock, Thomas,
III, and Niswonger, R.G., 2014. One-Water Hydrologic Flow Model (MODFLOW-OWHM): U.S. Geological
Survey TM6-A51, 120 p., http://dx.doi.org/10.3133/tm6A51.
Niswonger, R.G., Panday, Sorab, and Ibaraki, Motomu, 2011. MODFLOW-NWT, A Newton formulation
for MODFLOW-2005: U.S. Geological Survey TM6-A37, 44 p.
Hughes, J.D., Langevin, C.D., Chartier, K.L., and White, J.T., 2012. Documentation of the Surface-Water
Routing (SWR1) process for modeling surface-water flow with the U.S. Geological Survey Modular Ground-
Water Model (MODFLOW-2005): U.S. Geological Survey TM6-A40, 113 p.
Bakker, M., Schaars, F., Hughes, J.D., Langevin, C.D., and Dausman, A.M., 2013. The seawater intrusion
(SWI2) package for modeling vertically-integrated variable-density groundwater flow in regional aquifers
using the U.S. Geological Survey Modular Groundwater Flow Model (MODFLOW-2005): U.S. Geological
Survey Techniques in Water Resources Investigations, TM6-A46, 47 p.
Mehl, Steffen, and Hill, M.C., 2005. MODFLOW-2005, The U.S. Geological Survey Modular Groundwater
Model— Documentation of shared node local grid refinement (LGR) and the boundary flow and head (BFH)
package: U.S. Geological Survey TM6-A12, 68 p.

78
Schmid, W., Hanson, R.T., Maddock III, T.M., and Leake, S.A., 2006. User’s guide for the farm process
(FMP) for the U.S. Geological Survey’s modular three-dimensional finite-difference ground-water flow model,
MODFLOW-2000: U.S. Geological Survey TM6-A17, 127 p.
Schmid, Wolfgang, and Hanson R.T., 2009. The Farm Process Version 2 (FMP2) for MODFLOW-2005 -
Modifications and Upgrades to FMP1: U.S. Geological Survey TM6-A32, 102p.
Schmid, W., Hanson, R. T., Leake, S. A., Hughes, J. D., and Niswonger, R. G., 2014. Feedback of land
subsidence on the movement and conjunctive use of water resources. Environmental Modelling and Software,
62, 253-270.

79
Corresponding author: Jana Glass
Technische Universität Dresden
[email protected]

Free MODFLOW-based web modeling framework for planning and assessment

of managed aquifer recharge schemes
Jana Glass, Junior Research Group INOWAS, Department of Hydrosciences,Technische Universität
Dresden, Germany
Catalin Stefan, Junior Research Group INOWAS, Department of Hydrosciences,Technische Universität
Dresden, Germany
Ralf Junghanns, Junior Research Group INOWAS, Department of Hydrosciences,Technische Universität
Dresden, Germany

Abstract: INOWAS DSS is an open source web-service developed to solve groundwater re-
lated issues with a focus on the planning; management; and optimization of managed aquifer
recharge (MAR) schemes. The platform is easily accessible via standard web browsers and
includes tools of varying complexity with MODFLOW being the numerical simulation core.
Besides the description of the system’s architecture, a case study of a groundwater flow
model of the city centre of Hanoi, Vietnam is presented showing the applicability and par-
ticular features of the web-based modeling platform for the sustainable management of
groundwater resources.

Extended abstract: Methodology, findings, and implications

The web-based implementation of the INOWAS DSS integrates a unified, user-friendly and
modern graphical user interface (GUI) that provides a structural framework for all the
functionalities of the platform. The actual solution combines modern design elements with
powerful server capabilities for a smooth browser-based modeling experience and reliable
simulation outcomes. The technical infrastructure applied (Figure 1) consists of an inte-
grated approach based on three (physically separated) main components; the CLIENT – the
terminal for the user with internet access and web browser, the SERVER - a standard Linux
Server hosted by the Centre for Information Services and High Performance Computing
(ZIH) of Technische Universität Dresden (TUD), and the WORKER MICROSERVICES -
scalable applications running decentralized on one or more nodes. The communication with
the WEBSERVER takes place via a centralized Message Broker Software over the Advanced
Message Queuing Protocol (AMQP). The WORKER MICROSERVICES are easy scalable,
which enables a large quantity of parallel simulation runs and faster results regarding to
model calibration and scenario analysis.
The simulation core of the system is represented by a numerical groundwater flow model
based on MODFLOW (Harbaugh, 2005) and the open-source library FloPy (Bakker et al.,
2016). FloPy is a Python Library with the goal of making MODFLOW scriptable. It is able
to generate model input files, to run the MODFLOW executables, including MODFLOW-
2005 (Harbaugh, 2005), MODFLOW-NWT (Niswonger et al., 2011), MODPATH (Pollock,
2016), MT3DMS (Zheng and Wang, 1999) and SEAWAT (Langevin et al., 2008) and also
to load, read or plot simulation results of models that are part of the MODFLOW family

80
(Bakker et al., 2016). The use of FloPy brings an easy configurable and scriptable object
oriented abstraction layer to the application, which helps to describe and calculate models
remotely.
The INOWAS platform enables the user to set up and calculate a new groundwater flow
model. The integration of MODFLOW packages was realized modular, so that besides the
default configuration which includes basic packages (BAS6, DIS, LPF, UPW, OC), vari-
ous boundary conditions (CHD, GHB, RCH, WEL, RIV) as well as different solvers (PCG,
NWT), additional ones can be easily added in the future. Furthermore, a scenario analyzer
helps to vary boundary conditions to evaluate new management options as well as future
development induced e.g. by climate or land use change and urbanization. Different analy-
sis options such as time series analysis, cross sections of groundwater levels or drawdowns,
volumetric budget evaluation, and the calculation of head differences between different sce-
narios help to identify best management options. Besides the MODFLOW-based core, simple
tools solving various analytical equations are ported on the server, providing the user with
workflows of different levels of complexity. A dynamic environmental information system
provides additional support in model parametrization and data management.

Case Study: Groundwater flow model of Hanoi, Vietnam to test the feasibility
of MAR for sustainable management of groundwater resources
The INOWAS DSS was used to evaluate the extent of the decline of groundwater levels in
Hanoi, Vietnam and the suitability of various managed aquifer recharge (MAR) techniques.
A transient numerical groundwater flow model (MODFLOW-NWT) was setup and calibrated
in order to understand the groundwater flow system and to suggest solutions for a sustainable
water resources management (Ringleb et al., 2016). The results indicate vast groundwater
depletion cones in parts of the study area (Figure 2). For scenario analysis, riverbank
filtration was simulated by relocating some of the pumping wells towards the Red River and
new injection wells were implemented into the groundwater flow model. The wells’ relocation
scenario demonstrates that groundwater levels can be increased, especially in the depression
cones (Figure 2, Scenario 1). The infiltration of surplus water into the shallow aquifer via
injection wells during the rainy season shows that groundwater levels can also be improved,
but it is not effective to raise the depression cones (Figure 2, Scenario 2). Furthermore, water
needs to be pretreated before it can be recharged into the injection wells which is cost- and
labor-intensive. The combination of the above mentioned scenarios riverbank filtration and
injection wells leads to the highest increase of groundwater levels but the same disadvantage
with water pretreatment applies as for the second scenario (Figure 2, Scenario 3). Therefore,
the implementation of riverbank filtration wells is recommended. Before the implementation,
a detailed hydrogeological study including water quality aspects is needed.
The results of groundwater flow modeling support that MAR can be a feasible tool to
restore the groundwater balance and to sustainably manage the aquifer in Hanoi. The case
study showed the applicability of the INOWAS DSS platform and demonstrated its usefulness
to easily visualize model results and compare different simulation scenarios among each other
(Figure 2).

81
Conclusion
The new framework, which is easily accessible via standard web browsers, includes several
advantages over conventional simulation approaches: it provides best accessibility of project
data and multi-institutional collaboration through web-based implementation; makes use of
a combination of widely available open-source tools; and promotes the case-based reasoning
approach as additional support for parameter estimation and solution finding.

Figure 1: INOWAS DSS architecture.

82
 Figure 2: Screenshot of INOWAS DSS Scenario Analyzer.

References

Bakker, M., Post, V., Langevin, C.D., Hughes, J.D., White, J.T., Starn, J.J., Fienen, M.N., 2016. Scripting
MODFLOW Model Development Using Python and FloPy. Groundwater 54, 733–739. doi:10.1111/gwat.12413
Harbaugh, A.W., 2005. MODFLOW-2005, the US Geological Survey modular ground-water model: the
ground-water flow process (No. 6-A16), U.S. Geological Survey Techniques and Methods. US Department
of the Interior, US Geological Survey Reston, VA, USA.
Langevin, C.D., Thorne, D.T., Drausman, A.M., Sukop, M.C., Guo, W., 2008. SEAWAT Version 4: A
Computer Program for Simulation of Multi-Species Solute and Heat Transport (No. Chapter A22), U.S.
Geological Survey Techniques and Methods Book 6.
Niswonger, R.G., Pandey, S., Ibaraki, M., 2011. MODFLOW-NWT, A Newton Formulation for MODFLOW-
2005 (No. 6-A37), Techniques and Methods. U.S. Geological Survey.

83
Pollock, D.W., 2016. User guide for MODPATH Version 7—A particle-tracking model for MODFLOW (No.
2016–1086), USGS Open-File Report.
Ringleb, J., Via Rico, D.A., Stefan, C., Tran Thi Viet Nga, 2016. Transient flow modelling of an exploited
aquifer in the city of Hanoi, Vietnam and simulation of managed aquifer recharge measures, in: Proceedings
of 9th International Symposium on Managed Aquifer Recharge (ISMAR9). Presented at the ISMAR9, June
20-24, 2016, Mexico City, Mexico.
Zheng, C., Wang, P.P., 1999. MT3DMS: A Modular Three-Dimensional Multispecies Model for Simulation
of Advection, Dispersion and Chemical Reactions of Contaminants in Groundwater Systems, Documentation
and User’s Guide, Contract Report SERDP-99- 1. U. S. Army Engineer Research and Development Center:,
Vicksburg, MS, USA.

84
Corresponding author: Joseph Hughes
U.S. Geological Survey
[email protected]

Use of the Advanced Packages and Demand-Based Boundary Flows in the

MODFLOW 6 Groundwater Flow Model
Joseph D. Hughes, U.S. Geological Survey
C. Langevin, U.S. Geological Survey
S. Panday, GSI Environmental, Inc.
E. Banta, U.S. Geological Survey
A. Provost, U.S. Geological Survey
R. Niswonger, U.S. Geological Survey

Abstract: MODFLOW 6 is the latest version of the U.S. Geological Survey’s modular
hydrologic model. MODFLOW 6 includes a number of advanced packages that solve a
continuity equation for a hydrologic process. Advanced packages available for the MOD-
FLOW 6 groundwater flow model include packages for simulating wells open to multiple
aquifers (MAW8), streamflow routing (SFR8), lakes (LAK8), and unsaturated groundwater
flow (UZF8). The continuity equation for the advanced packages are solved as a separate
system of equations in the SFR8, LAK8, and UZF8 Packages and implicitly coupled with
the groundwater flow system of equations in the MAW8 Package. MODFLOW 6 also in-
cludes a generic mover (MVR8) Package that provides native functionality for moving water
from standard boundary conditions that remove water from the groundwater model and the
advanced packages (water providers) to advanced package elements (water receivers). The
amount of water from a boundary package delivered though the MVR8 package to a water
receiver can be controlled using an amount that is (1) a fraction of the provided water, (2)
in excess of a specified value, (3) in excess of a threshold value, or (4) less than or equal
to a maximum value. Another generic package (DMD8) is currently being developed to
allow demands calculated by other models incorporated in the MODFLOW 6 framework to
define simulated extraction rates for elements in the advanced packages. The combination
of the MVR8 and DMD8 Packages will allow native simulation of demand-based advanced
boundary conditions in MODFLOW 6. Examples of the use of (1) the MVR8 to move water
between the advanced packages and (2) a combination of the SFR8, MAW8, MVR8, and
DMD8 Packages to simulate demand-based irrigation withdrawals from surface water and
groundwater will be presented.

85
Corresponding author: Christian Langevin
U.S. Geological Survey
[email protected]

MODFLOW 6: An Object-Oriented Version of the U.S. Geological Survey’s

MODFLOW Model
Christian Langevin, U.S. Geological Survey
Joseph D. Hughes, U.S. Geological Survey
Edward R. Banta, U.S. Geological Survey
Alden M. Provost, U.S. Geological Survey
Richard G. Niswonger, U.S. Geological Survey
Sorab Panday, GSI Environmental, Inc.

Abstract: MODFLOW 6 is the latest version of the U.S. Geological Survey’s modular hy-
drologic model. MODFLOW 6 was developed to synthesize many of the recent MODFLOW
advances into a single program, improve they way different process models are coupled, and
to provide an object-oriented framework for adding new types of models and packages. The
first release of MODFLOW 6 contains the Groundwater Flow (GWF) Model, which com-
bines most of the capabilities of MODFLOW-2005, MODFLOW-NWT, MODFLOW-USG,
and MODFLOW-LGR. The GWF Model supports three different types of grids: regular
MODFLOW grids consisting of layers, rows, and columns; layered grids that are defined
using vertices; and, generalized unstructured grids. Layered grids support a new capa-
bility to remove thin or inactive cells from the simulation. When thin cells are removed
from the simulation, the model automatically connects overlying and underlying cells. The
Node Property Flow (NPF) and Storage (STO) Packages replace the Block-Centered Flow,
Layer-Property Flow, and Upstream Weighting flow packages of previous MODFLOW ver-
sions. NPF includes a new formulation, called XT3D, for representing full, three-dimensional
anisotropic groundwater flow and for improving the flow approximation for certain grid
types. The GWF Model in MODFLOW 6 contains the six, traditional MODFLOW stress
packages (Well, Drain, River, General-Head Boundary, Recharge, and Evapotranspiration)
and four advanced hydrologic stress packages (Multi-Aquifer Well, Streamflow Routing,
Lake, and Unsaturated Zone Flow). MODFLOW 6 also includes a new and generalized way
to move water between these hydrologic stress packages. This new capability can be used
to route water from drains into a stream, for example, or to send rejected infiltration into
a nearby lake. The object-oriented framework also makes it possible to tightly couple any
number of parent, child, and grandchild models, or to couple an unstructured GWF Model
of a shallow aquifer with a deep aquifer GWF Model that uses a regular grid. Future plans
for MODFLOW 6 include the addition of variable-density groundwater flow, transport, and
a landscape hydrology model.

86
Corresponding author: Steffen Mehl
California State University, Chico
[email protected]

Rescaling Riverbed Conductance to Improve Approximations of

Stream/Aquifer Leakage

Steffen Mehl, California State University, Chico

H. Morel-Seytoux
C. Miller

Abstract: Accurate simulation of stream/aquifer exchanges can be difficult in large-scale

integrated modeling studies where the grid resolution is typically much larger than the
stream features. Locally, the flow lines contract and change direction from horizontal to
vertical causing additional resistance to flow, but representation of these three-dimensional
hydraulic details requires a highly refined grid in the vicinity of the stream, which can be
impractical in large-scale models. In practice, the riverbed conductance is often consid-
ered as a calibration parameter that accounts for both unknown hydraulic properties of the
streambed and artifacts of representing a three-dimensional flow process as a one-dimension
head-dependent flux boundary condition. This causes a grid-size dependency in stream leak-
age which is problematic when grid resolution changes (for example, going from regional to
local-scale models, one has to recalibrate riverbed conductance). This is even more problem-
atic considering the flexibility in gridding options that are becoming more prevalent in recent
versions of Modflow. This work examines approaches for rescaling riverbed conductance,
given the streambed and aquifer hydraulic properties and stream geometry, to account for
the three-dimensional hydraulic details and lead to better accuracy in stream leakage rates.
The eventual goal is to provide a systematic way to adjust riverbed conductance for cases
where the stream system might be represented with varying grid sizes.

87
Corresponding author: Jaco Nel
GCS Water and Environmental
[email protected]

Application and challenges of MODFLOW USG to simulate tunnel inflows

Jaco Nel, GCS Water and Environmental

A. Johnstone
N. Rapantova

Abstract: A ’HYDRUS-based flow package for MODFLOW’ was developed by Seo et al.
(2007) and Twarakavi et al. (2008) to simultaneously simulate transient flow in both un-
saturated and saturated zones. The package, which is based on the HYDRUS-1D model
(Šimůnek et al., 2016) simulating unsaturated water flow in the vadose zone, was linked to
MODFLOW simulating saturated groundwater flow. The water flow part of HYDRUS-1D
has been used in the coupled model to represent the effects of unsaturated zone processes.
The coupled model is effective in addressing spatially-variable saturated-unsaturated hy-
drological processes at the regional scale with complex layering in the unsaturated zone,
spatially and temporarily variable water fluxes at the soil surface and in the root zone, and
with alternating recharge and discharge fluxes (Twarakavi et al., 2008). One of the major
limitations of the coupled model was that it could not simulate solute transport. However,
solute transport is highly depended on water table fluctuations due to temporal and spa-
tial variations in groundwater recharge. This is an important concern when the coupled
model is used for analyzing groundwater contamination due to transport through the un-
saturated zone. The objective of this study is to integrate the solute transport model (the
solute transport part of HYDRUS-1D for the unsaturated zone and MT3DMS for the sat-
urated zone) into an existing coupled flow model. The unsaturated zone component of the
coupled model can consider solute transport involving many biogeochemical processes and
reactions, including first-order degradation, volatilization, linear and/or nonlinear sorption,
one-site kinetic sorption, two-site sorption, and two-kinetic site sorption (Simunek and van
Genuchten, 2008). Due to complex interactions between the coupled models at the ground-
water table, certain modifications of the pressure head profiles (compared to the original
coupling) and solute concentrations have been incorporated into the HYDRUS package.

88
Corresponding author: Richard Niswonger
U.S. Geological Survey
[email protected]

Modeling water supply effects on the distribution and consumption of water in

developed river basins
Richard Niswonger, U.S. Geological Survey
E. Morway, U.S. Geological Survey
E. Triana, U.S. Geological Survey
W. Kitlasten, U.S. Geological Survey

Abstract: Hydrologic models are used to evaluate past water budgets and future water
supplies in developed river basins that are affected by a changing climate. Models also are
used to evaluate management change options for enhancing water supply or to estimate
the impacts of land-use change or development. Principal to these analysis is the realis-
tic representation of the interplay between water supply and water use. Most hydrologic
studies develop models with specified water use by agricultural, municipal, and industrial
sectors. This approach works well for historical simulations that incorporate metered water
use. However, where water-use data have not been measured, or for simulating future water
resources scenarios, specified water use that is incongruent with water supply conditions is
unrealistic, and likely does not provide meaningful results for evaluating the impacts of cli-
mate change or water resources management scenarios. An important example for the case
of the agricultural sector is the problem of simulating the distribution of irrigation amounts
when water demands are greater than supply. This case arises often in over-allocated basins
or when simulating drought scenarios. Realistic representation of water limited systems
requires complex decisions about how water scarcity changes allocation according to water
rights priority, groundwater pumping, water consumption, and downstream water availabil-
ity. In an effort to more realistically simulate water supply and water use dynamics, we
propose a new type of integrated model that couples a river/reservoir operations model
to an integrated hydrologic model. This type of integrated model provides a framework
for simulating surface water distribution according to water rights priorities and rules that
realistically represent the impacts of water scarcity. Furthermore, it provides an approach
for estimating groundwater pumping that tends to be focused where water users have ju-
nior surface water right priority and require supplemental groundwater pumping to satisfy
their water demands. This talk will discuss the benefits and challenges of simulating water
distribution and consumption in developed basins where demands are often greater that
supply.

89
Corresponding author: Sorab Panday
GSI Environmental, Inc.
[email protected]

A Hydraulic Head Formulation for Density Dependent Flow and Transport

Sorab Panday, GSI Environmental, Inc.

Christian Langevin, U.S. Geological Survey
Alden M. Provost, U.S. Geological Survey

Abstract: Density dependent flow and transport solutions for coastal saltwater intrusion
investigations, analyses of fluid injection into deep brines, or studies of fingering and insta-
bilities of denser fluids moving through lighter ones traditionally use the equivalent fresh-
water head (EFH) formulation for solution of the flow equation. A hydraulic head (HH)
formulation is presented here as an alternative. The HH formulation may be a beneficial
approach for adapting existing constant-density groundwater flow software to include the
density driven flow components since it avoids conversion from EFH to HH and vice versa
in various sections of a code along with the associated difficulties. The formulation there-
fore readily accommodates complexities such as unsaturated flow, turbulent flow, Newton
Raphson solutions, and complex head-dependent boundary conditions. Test cases will also
be presented.

90
Corresponding author: Alden Provost
U.S. Geological Survey
[email protected]

The ”XT3D” option for simulating fully three-dimensional anisotropy on

regular and irregular MODFLOW 6 grids

Alden Provost, U.S. Geological Survey

C. Langevin, U.S. Geological Survey
J. Hughes, U.S. Geological Survey

Abstract: The ’XT3D’ option in the Node Property Flow (NPF) Package of MODFLOW
6 enables simulation of fully three-dimensional (3D) anisotropy on regular and irregular
grids in a way that properly takes into account the full, 3D conductivity tensor. Because
it can also improve the accuracy of groundwater-flow simulations in cases in which the cell
connections and shared faces violate certain geometric requirements, the ’XT3D’ option can
serve as an alternative to the Ghost-Node Correction (GNC) Package. Example problems
demonstrate the use of the ’XT3D’ option to simulate groundwater flow on irregular grids
and through 3D porous media with anisotropic hydraulic conductivity.

91
Corresponding author: Azucena Rodriguez Yebra
British Geological Survey
[email protected]; [email protected]

Generation of Source Reliable Output diagrams of Adited Groundwater

Sources using models: An example from the UK

Azucena Rodriguez Yebra, British Geological Survey

A. Hughes
A. Butler
D. Peach
C. Jackson

Abstract: Drought and water scarcity pose a threat to the environment and society in
the UK, particularly under climate change. Securing sustainable groundwater sources is a
legal requirement achieved through Water Resources Management Plans. These plans as-
sess the total amount of water available for supply under drought conditions by quantifying
the deployable output of all groundwater and surface water sources. They are encapsu-
lated in Source Reliable Output (SRO) diagrams relating abstraction rate to borehole level.
Typically constructed at the end of the 19th century, adit systems in the Chalk aquifer (a
fine-grained Cretaceous limestone with significant fracturing) were dug in cases where yields
were poor. Adits are horizontal tunnels with diameters of approximately 2 m dug radially
out from large diameter wells or shafts to increase yield. They can run for a number of kilo-
metres along valleys and beneath ecologically important Chalk streams. There are about
100 adit systems in the whole of the UK and many are still in operation as major public
supply sources. The availability of data in adit systems is often limited and not always
reliable, particularly for drought conditions, therefore, there is a need to use computational
models to produce SRO diagrams. Using MODFLOW-CLN, we describe a methodology to
produce a SRO diagram for a typical adit system using a site informed groundwater model
for a catchment in South-East England. Simulation of source yields for historic droughts and
climate change scenarios has the advantage of providing increased confidence in calculating
reliable yields under extreme drought.

92
Corresponding author: William Witterick
Amec Foster Wheeler
[email protected]

Borehole-scale simulations from a regional groundwater model using Connected

Linear Networks and Ghost Node Correction

Willam R. Witterick, Amec Foster Wheeler

Abstract: For over 20 years, the UK’s environmental regulator has developed regional scale
MODFLOW models for the purposes of sustainable water resource management. These have
typically been built on coarse regular meshes (around 200 m), and use the WEL package
to simulate groundwater withdrawals; often resulting in multiple wells being located within
the same model cell. A methodology has been developed to re-purpose an existing regional
groundwater model into a borehole-scale tool to help investigate and optimise borehole yield
in situations where this is constrained by the pumped water level. Simulations to determine
source yield are carried out using current MODFLOW-USG functionality, additional anal-
ysis of source-specific data and operational knowledge, and new code developments. The
Connected Linear Network (CLN) process has been used to simulate borehole-aquifer inter-
actions and allowed the incorporation of borehole specific geometries and conductances. The
Ghost Node Correction (GNC) package places these CLN boreholes in their true locations,
including multiple boreholes within the same cell without the need for time consuming and
expensive grid refinement. Additional code enhancements have been introduced to incorpo-
rate source-specific level - yield profiles so that the simulations provide a time series of the
yield which can be delivered by individual or aggregated boreholes under a range of design
drought conditions. Alternative source operational strategies allowing for conjunctive use
and seasonal rest - peak cycles can then be explored to maximise the combined yield.

93
Corresponding author: Laura Foglia
University of California, Davis
[email protected]

FREEWAT, a HORIZON 2020 project to build open source tools for water
management: a European perspective

Laura Foglia, University of California, Davis

R. Rossetto
I. Borsi
M. Cannata
S.W. Mehl
E. Vasquez-Sunez

Abstract: FREEWAT is an HORIZON 2020 EU project. FREEWAT is an open source

and public domain GIS integrated modelling environment for simulation of water quantity
and quality in surface water and groundwater with an integrated water management and
planning module. FREEWAT aims at promoting water resource management by simplifying
the application of the Water Framework Directive and related Directives. Specific objectives
of the project are: to coordinate previous EU and national funded research to integrate ex-
isting software modules for water management in a single environment into the GIS based
FREEWAT and to support the FREEWAT application in an innovative participatory ap-
proach gathering technical staff and relevant stakeholders (policy and decision makers) in
designing scenarios for application of water policies. The open source characteristics of the
platform allow to consider this an initiative ”ad includendum”, as further institutions or
developers may contribute to development. Main expected impact of FREEWAT is to help
produce scientifically and technically sound decisions and policy making based on data and
innovative data analysis tools and to support participatory approach not only in the final
stage of discussion but during the phase of scenario generation. The platform currently
includes specific modules for: 1) water management and planning to help managing and ag-
gregating the distributed data coming from simulation scenarios; 2) calibration, uncertainty
and sensitivity analysis; 3) solute transport in unsaturated zone; 4) crop growth and water
requirements in agriculture; 4) tools for groundwater quality issues; 5) tools for analysis,
interpretation and visualization of time series and hydrogeological data.

94
Corresponding author: Douglas Hayes
S.S. Papadopulos and Associates
[email protected]

PEST.cloud: Cloud-based Model Calibration

Douglas A. Hayes, S.S. Papadopulos and Associates

J. Doherty
C. Muffels

Abstract: Calibration and uncertainty analysis of groundwater models requires the use of
high-end inversion software. This, in turn, requires that many model runs be undertaken.
By parallelizing these runs, model and parameterization complexity are readily accommo-
dated. High end inversion packages such as BEOPEST and PEST++ support parallel run
management in inverse problem solution. A new version of PEST, named PEST_HP, opti-
mizes inversion efficiency in highly parallelized computing environments. To an increasing
extent, PEST users are turning to the cloud for cheap and instant access to many comput-
ing nodes. A new web-based service named PEST.cloud has been developed to facilitate
use of PEST_HP in the cloud environment. Through a web-based interface, it automates
upload of a modeler’s files to the Azure cloud platform, initiates nodes used for model run
parallelization, distributes files to these nodes, and terminates these nodes after completion
of the PEST_HP run. Prior to cloud transfer, it lists options for the number and types
of computers that a modeler can employ (options include the latest and fastest available
hardware), and estimates the total cost of the PEST_HP run associated with each choice.
A user can set a limit on how much he/she will spend on a particular PEST_HP run. If that
limit is reached it can be easily extended, so that nothing is lost. Meanwhile, as PEST_HP
is running, PEST.cloud provides up-to-the-minute PEST_HP run-time stats. The user is
notified by email when PEST_HP has completed the inversion process.

95
Corresponding author: Damian Merrick
HydroAlgorithmics
[email protected]

AlgoCompute: Large-scale Calibration and Uncertainty Analysis Made Easy in

the Cloud

Damian Merrick, HydroAlgorithmics

Abstract: This paper presents AlgoCompute, a new web-based platform that makes it
easy to perform simulation runs in the cloud. It allows scaling out to any desired number
of compute nodes, which is particularly valuable when undertaking complex uncertainty
analysis, calibration or optimization tasks. AlgoCompute is accessed through a web browser:
you log in, provide it with your model files, initiate new runs, and monitor ongoing runs.
The system is designed to integrate well with PEST, and the platform automates the setup
and pull-down of appropriate processing resources, distribution of model data and results,
and coordination of PEST slaves. Continuous reports of run progress are provided on the
web interface and optionally via email notifications.
AlgoCompute’s access to cloud resources is entirely self-contained: users do not need to
maintain their own accounts with a public cloud provider, nor to provision their own virtual
machines for running models or write scripts to start, stop and monitor progress. Addition-
ally, being completely web-based means that no custom software installations are required,
allowing modelers to access the system with minimal support from IT personnel.

Extended abstract: Methodology, findings, and implications

There is a growing trend away from manual, trial-and-error calibration of groundwater and
integrated hydrological models, and toward inversion for automated calibration. This is
made possible by parameter estimation software such as PEST (Doherty, 2015), which au-
tomatically performs model runs to explore the parameter space of a model and produce
parameter sets that minimize the discrepancy between field data and model results.
Beyond calibration, it is important to note one of the primary purposes of these models:
to make predictions of how a physical system is likely to behave in the future. Will resource
usage be sustainable? Will groundwater dependent ecosystems be impacted? Will land
be further degraded by waterlogging and salinization? The answers to these questions are
inherently uncertain: there is uncertainty in our conceptualization of the real system, in
our knowledge of aquifer property values, in the boundary conditions imposed on the finite
domain of a model, in anticipating the climate regime, in the representation of natural
processes within algorithms in standard software packages, and in numerous other aspects
of our models.
Indeed, automated calibration processes highlight the fact that there are often many
different parameter sets that lead to similar model results. Modelers often either ignore
or are not aware of the extent of non-uniqueness lurking beneath a model calibrated by
trial-and-error; this brings about the need for uncertainty analysis to explore the effect on
predictions of many possible combinations of plausible model properties.
Performing automated calibration and uncertainty analysis typically necessitates a large
number of model simulation runs. For complex models, each simulation may itself take

96
hours or days to run. This leads to a requirement for large amounts of computing time,
which often renders the application of such processes infeasible in practice. This may be
mitigated somewhat by parallelizing the processes over multiple machines, but for many
modelers, limited availability of local computing resources typically restricts this to a few
dozen processes at a time.
AlgoCompute is a new platform that aims to overcome this limitation by providing a
convenient way to access cloud computing resources for simulation and parameter estimation
workloads. It has a web-based user interface allowing easy access from any modern web
browser, anywhere in the world.
Using AlgoCompute to perform a parameter estimation run with PEST proceeds as
follows:

1. Log in to the AlgoCompute web site.

2. Upload a file set for the run, including a PEST control file, static model files, tem-
plates to produce parameter-dependent model files, a model run batch file, and PEST
instruction files for reading results. Files may be uploaded in the web browser, or
transferred directly from a Dropbox account.

3. Instruct AlgoCompute to start the run, specifying important details: what model
software is needed; how many compute nodes (virtual machines) should be utilized,
and of what type (number of CPU cores, amount of RAM and disk space); and how
many CPU cores should be reserved for each model process.

4. Monitor the run once it has been launched, as desired, until it completes.

5. Download the completed run results from AlgoCompute to a local machine, either
directly in the web browser or via Dropbox, for review and post-processing.

AlgoCompute automatically provisions the requested compute nodes and initiates the
PEST master and slave processes to start the run. The time taken to create and start
cloud compute nodes can vary; in practice, this process usually completes within three to
five minutes. Following this, PEST executes and AlgoCompute gives continuous updates
on the run’s progress via its web interface; see Figure 1. Once the run has completed,
AlgoCompute copies result files to its cloud storage area from the PEST master process and
from the slave process that executed the most recent model run, before terminating and
releasing the allocated compute nodes.
While the run is executing, the console outputs of the PEST master and slave processes
may be viewed from AlgoCompute’s console monitor; see Figure 2. This is useful for getting
a detailed picture of the run’s progress and identifying any potential problems prior to the
run completing. If any problems are identified, the run may be terminated at any time,
either immediately or after copying any files created by the run in its intermediate state.
AlgoCompute optionally sends out email notifications when a run has started executing,
when it completes, and each time a new PEST optimization iteration is reached. This allows
the progress of very long runs to be periodically checked without the need to keep a web
browser open constantly.

97
 The system is flexible with regards to the model simulation software used. Various
versions of MODFLOW are provided as part of the system, in addition to HydroGeoSphere
for fully-integrated surface and subsurface modeling (Aquanty Inc., 2017). Alternative model
executables or post-processing utilities may be provided by the user by including them with
the uploaded set of model files.
In addition to PEST runs, AlgoCompute supports the HydroGeoSphere Uncertainty
Quantification tool (Miller et al., 2017), which provides methods based on polynomial chaos
expansion for uncertainty quantification, in addition to a number of random sampling meth-
ods including Monte Carlo, Latin Hypercube and quasi-Monte Carlo. HGSUQ utilizes a
similar master/slave arrangement to PEST.
Finally, the system may also be used to execute manually-created scenario runs of a
model. In this case, a set of model files is provided as normal, but multiple alternative
versions of one or more of these files may be provided such that each version represents a
different model scenario. This is useful for scheduling a number of alternative runs of the
same model where, for example, a different set of material parameters is being evaluated in
each. The compute resources used to run the scenarios are specified independently from the
scenarios themselves, meaning that some or all of the scenarios may be run concurrently,
with scenarios being automatically executed as compute nodes become available.
In summary, AlgoCompute is a new platform facilitating access to cloud computing
resources for both forward and inverse modeling workloads. It allows these workloads to be
scaled out to hundreds or thousands of concurrent processes as needed, in correlation with
project scope and budget. As it only requires a web browser, modelers may access the system
from their regular workstation or laptop, with minimal support from their organization’s IT
personnel.
AlgoCompute is available as a commercial Software-as-a-Service (SaaS) offering to inter-
ested modelers; readers are encouraged to contact the author for more information.

98
 99
Figure 1: AlgoCompute displays real-time progress information for a PEST run
 Figure 2: Monitoring the console outputs of an AlgoCompute PEST run

References

Aquanty Inc., 2017. Hydrogeosphere: A Three-Dimensional Numerical Model Describing Fully-Integrated

Subsurface and Surface Flow and Solute Transport. Waterloo, ON.
Doherty, J., 2015. Calibration and Uncertainty Analysis for Complex Environmental Models. Watermark
Numerical Computing, Brisbane, Australia. ISBN: 978-0-9943786-0-6.
Miller, K. L., Park, Y-J., Berg, S. J., Sudicky, E. A., 2017. Efficient Quantification of Uncertainty in
Integrated Surface and Subsurface Hydrologic Simulations. Proc. MODFLOW and More, Golden, CO.

100
Corresponding author: Richard Winston
U.S. Geological Survey
[email protected]

StreamExtractor: A New Tool to Help Generate Topographically Consistent

Input for the Stream and Streamflow Routing Packages in MODFLOW

Richard B. Winston, U.S. Geological Survey

Jason P. Pope, U.S. Geological Survey

Abstract: StreamExtractor uses previously described methods to extract stream networks

from Digital Elevation Models (DEMs) and exports then as a PolyLineZ shapefile. Z-
coordinates of each vertex are included in the shapefile. Shapefile attributes can be used
to assign stream segment numbers in the Stream and Streamflow Routing packages and to
define stream routing. ModelMuse can import the Z coordinates in PolyLineZ shapefiles and
to use those data to help define the input for the Stream and Streamflow Routing packages.
By using the Z coordinates, these input files can be created where stream elevations always
decline in the downstream direction. This is in contrast to using cell elevations to define
stream elevations, which can lead to model streams that flow uphill. The new method was
tested in a model of groundwater flow in the Coastal Plain in Virginia, Maryland, Delaware,
and New Jersey. StreamExtractor was used to identify over 6,000 streams in the model
area. For comparison, the heads in the streams were specified in two different ways. When
the head in each stream reach was set equal to the head in the cell containing the reach,
there were over 1,000 cases in which the head in a reach was lower than the head in the
next downstream reach. When the elevations stored in the shapefile by StreamExtractor
were used to define the heads in each stream reach, there were no cases in which the head
in a reach was lower than the head in the next downstream reach.

101
Corresponding author: Daniel Abrams.
University of Illinois
[email protected]

Tools for the hands-on MODFLOW modeler

Daniel B. Abrams, University of Illinois

George S. Roadcap, IL State Water Survey - U of Illinois

Abstract: As models become larger and simulated processes become more complex, auto-
mated approaches for model development are essential. However, many aspects of the initial
modeling process, such as achieving convergence or an initial calibration, benefit from a
hands-on approach, particularly for large regional MODFLOW models. The deluge of data
involved in manually improving convergence and calibration can mask important trends or
insights. To this end, freely available toolboxes have been developed to assist modelers with
manually improving calibration and convergence. Although PEST is a widely accepted for
parameter estimation, modelers must still develop appropriate conceptualizations based on
geologic insight. As such, manual calibration is necessary in the initial phases of model
development. However, manual calibration to temporal datasets is difficult in large regional
models with widely variable geology. The second toolbox reads the simulated head file as
well as a target shapefile. The calibration results can then be queried by stress period, layer,
and three user-defined fields. The toolbox also creates a residual error map based on the
query. Large MODFLOW models will often be slowed by a few cells that have difficulty
converging on a user-defined head change or residual criteria. The list file provides informa-
tion on these ”bad actors”, but can run into severe memory limitations in many text editors
or word processors. This precludes the modeler from fully exploring the range of errors.
The first toolbox reads the entirety of the list file and outputs the cells with the largest and
the most frequent errors.

102
Corresponding author: Ramarao Banda
INTERA Incorporated
[email protected]

Capture Functions in a Coupled Surface Water-Groundwater Model Using

MODFLOW-ADJOINT Code

Ramarao S. Banda, INTERA Incorporated

Van A. Kelley, INTERA Incorporated
Marsh Lavenue, INTERA Incorporated

Abstract: A sustainable groundwater development strategy calls for managing aquifer pro-
duction, to ensure that the depletion of the hydrologically connected surface water sources
(rivers, streams, wetlands, etc.) necessarily balances the considerations of existing water
rights and the potential degradation of the eco-systems. Capture functions quantify the
surface water depletion associated with pumping the aquifers, and they denote the fraction
of the pumped water that is drawn from the surface water sources. Capture maps provide
useful insight for optimal location of the pumping wells, their rates of withdrawals, and their
timing. MODFLOW is being used for the calculation of capture functions via a parameter
perturbation approach with finite differences, requiring large number of repeated simula-
tions with MODFLOW. This paper presents an Adjoint Sensitivity method, which requires
just one simulation for the capture functions at all nodes in the model, leading to an im-
mense computational efficiency. Within MODFLOW-2005 the STR package, implements a
non-linear boundary condition for the river stage using a coupled surface water-groundwater
model. As a result, the flows to/from the aquifer depend not only on the river stage and
aquifer head at that location (as it would in RIV package), but also on the river stages
and aquifer heads at all the upstream locations in the complex network of tributaries and
diversions. An exact mathematical analysis for the STR Package in MODFLOW-ADJOINT
has been formulated, implemented, verified, and validated by a case study of the San Pedro
Model of the United States Geological Survey; and is presented here.

103
Corresponding author: Martinus Brouwers
Matrix Solutions Inc.
[email protected]

Modelling Uncertainty Analysis for Contaminant Risk Assessment

Martinus Brouwers, Matrix Solutions Inc.

P. Martin
D. Abbey

Abstract: Models are often relied upon to extrapolate from our existing site knowledge to
make predictions regarding potential future impacts associated with contaminant plumes,
or the effectiveness of proposed remediation systems. Predictions made with models are
always subject to uncertainty. That uncertainty stems from having incomplete conceptual
understanding of the hydrogeology and groundwater flow system, incomplete characteriza-
tion of the contaminant source and distribution in the subsurface, sparse observation data to
calibrate the model, and incomplete measurements of hydrogeologic parameters. Develop-
ments in cloud computing and numerical analysis tools have made meaningful assessment of
model prediction uncertainty more cost-effective and informative. With such advancements,
detailed models that accurately depict our understanding of the hydrogeologic setting and
site data can be readily applied. The uncertainty analysis provides insights regarding the
range of potential exposure levels, the most likely levels, and the timing of expected expo-
sures. The goal of such analysis is to increase the transparency of the analysis and thus
increase confidence regarding risk predictions or remediation measures designed to mitigate
risks. Further, this process also highlights data gaps that control the predictions and the
value of filling such data gaps with respect to reducing exposure level uncertainty. Project
stakeholders, including regulators, will find uncertainty analysis to be an especially valuable
tool because it provides a higher degree of confidence in predictions made regarding human
health and ecological risk assessment. This presentation will illustrate how modelling tools
are being applied to evaluate contaminant prediction uncertainty, and evaluate remediation
alternatives for selected sites.

104
Corresponding author: Wim de Lange
Deltares
[email protected]

Parameter values in the DRAIN package for a “groundwater weir” near a canal
in reconstruction

Henk Kooi, Deltares

Jacco H. Hoogewoud, Deltares
Wim J. de Lange, Deltares

Abstract: In the southern part of The Netherlands, a canal is reconstructed to facilitate

large ships passing sluices and bridges. As a part of this the water level of the canal is aimed
to be lowered by 2.5 meter over a distance of about 2 kilometer. Lowering of the surface
water level lowering will cause lowering of the groundwater level over a significant area if no
counter measures are taken.

Extended abstract: Methodology, findings, and implications

One of the counter measures under investigation concerns the application of an impermeable
foil layer several meters below the canal bottom (Figure 1) along the sides of which ground-
water may seep into the canal. The flux to the canal depends on the shape of the foil, the
groundwater level and the aquifer properties.
We have developed a set of parameter values for the MODFLOW DRAIN package (Figure
2) for the seepage of the groundwater as it occurs in the upper layers of the aquifer system.
The parameter values have been computed from a conceptual model including unsaturated
flow based on finite element technique. A set of discrete parameter values has been used
to compute the seepage to the canal under different conditions. For each case, the relation
between the phreatic head and the flux has been used to derive parameter values for use in
the DRAIN package in the MODFLOW model.
The results have been verified by use in a MODFLOW model. In our presentation we
will discuss, the problem, our analysis using conceptual models and the results. The actual
list of parametersets is used in the MODFLOW modeling of the reconstruction of the canal
including the lowering of the surface water level.

Figure 1. Scheme of aquifer system (yellow) with aquitard (green) and silty toplayer (orange) with a canal
(blue) isolated from the groundwater by foil (blue staight lines) on top of which groundwater may seep to
the canal (arrows).

105
 Figure 2. Scheme of detail of “Groundwater Wear” (top); Model scheme of section in Figure 1 (bottom) .

References

106
Corresponding author: John Doherty
Watermark Numerical Computing
[email protected]

PEST HP: A Version of PEST Optimized for Highly Parallelized Computing

Environments

John Doherty, Watermark Numerical Computing

Abstract: PEST_HP is a version of PEST that is built for use on the cloud or on clusters
with many nodes. It consistently achieves better performance than BEOPEST because of
its ability to make use of nodes that would otherwise be idle when calculating and test-
ing parameter upgrades, at the same time as it minimizes incidences of unused computing
resources. This same strategy results in greater tolerance for poor model numerical perfor-
mance that would otherwise compromise calculation of finite-difference derivatives. Other
features of PEST_HP include superior reporting of calibration statistics and node efficien-
cies, improved Broyden updating of the Jacobian matrix, and more comprehensive storage of
model run outcomes, this supporting construction of proxy models for use in difficult inver-
sion contexts. A ”PEST whisperer” is also provided that reads all PEST_HP output files,
giving advice to a modeller on how PEST setup can be improved in his/her particular situa-
tion. Use of PEST_HP is demonstrated in calibration of a large regional MODFLOW-USG
model used for coal bed methane impact assessment, and in calibration of a TOUGH2 model
built for geothermal reservoir management. The first of these models employed over 5000
parameters while the second, employed over 500. Calibration of models such as geothermal
models is further enhanced by PEST_HP’s ability to transfer files between manager and
agent nodes so that initial conditions can be upgraded as parameter values change.

107
Corresponding author: Linzy Foster
U.S. Geological Survey
[email protected]

Quantifying predictive uncertainty of groundwater/surface-water exchange in

the lower San Antonio River basin, Texas

Linzy K. Foster, U.S. Geological Survey

J. White

Abstract: A novel groundwater/surface-water model of the lower San Antonio River basin
and surrounding areas is in development and is being paired with linear-based uncertainty
analysis to quantify groundwater/surface-water exchange under uncertainty. The main goal
of this study is to develop a tool which can be used to assess spatial and temporal variations
in exchange occurring between the lower San Antonio River (and its major tributaries)
and the different aquifer units these streams traverse. This area has experienced drought
conditions in recent years, which has increased interest in simulation of potential drought
scenarios with this model; therefore, a predictive scenario of streamflow response to the
combination of reduced rainfall and increased groundwater withdrawals will be investigated.
The Texas coastal uplands aquifer system and the coastal lowlands aquifer system are being
simulated within the study area to depths of approximately 5500 ft below land surface for
the time period of 2006-2013. The predictions made with this model will include volumetric
totals with associated uncertainty at the confluence of the San Antonio and Guadalupe
rivers, which will provide improved understanding of streamflow impacts from groundwater
pumping within the basin.

108
Corresponding author: Phil Hayes
U.S. Geological Survey
[email protected]

Data drought to data surplus - the value of traditional groundwater data sets
and extensive hydrogeophysics to make model predictions including uncertainty

Phil Hayes, Jacobs

Chris Nicol, Groundwater Logic Pty. Ltd.
V. Puech

Abstract: To address supply reliability in the desert city of Broken Hill, NSW, Australia, a
5-year program of studies by Geoscience Australia (GA) identified groundwater targets for
wellfield applying advanced hydrogeophysics. Groundwater modelling aimed to define the
environmental impacts of abstraction, including aquifer recovery time within an uncertainty
framework. Hydrogeophysics data proved reliable for model structure, including erosional
holes and neo-tectonic fault pathways through aquitards that enable surface water recharge
to deeper aquifers. Hydrogeophysical extrapolation geophysics to predict aquifer properties
provided indications of spatial facies and hydraulic property changes, but provided insuffi-
cient information to reproduce key conceptual model aspects. Re-balancing the calibration
to favour borehole groundwater level observations enabled representation of the conceptual
model and reduced the overall uncertainty in predictions of wellfield drawdown, environ-
mental impacts and time to recovery. With larger, more complex datasets increasingly
available, modellers must weigh the value against traditional observations and craft model
designs and workflows to best integrate new information.

Extended abstract: Methodology, findings, and implications

Data Availability

An unusually high volume of data was available due to the Broken Hill Managed Aquifer
Recharge Study (BHMAR; GA 2012/2). Following the BHMAR study, two borefields were
drilled on the dry bed of Lake Menindee and at Talyawalka (Figure 1). The BHMAR study
comprised comprehensive investigations over 4 years including conventional hydrogeologi-
cal techniques supplemented by an extensive airborne and downhole geophysics program
of 30,000+ km of Airborne Electromagnetic (AEM) surveys. Study reports are over 3,000
pages with 42 Gb of spatial and other datasets that include: bores; hydrochemistry; ground
geophysics; borefield targets; topography; rivers and bathymetry; geology; physiography; ge-
omorphology; hydrogeology; AEM data and interpretations; surface materials; flood extent;
vegetation and interpreted groundwater-dependence; and basemaps. Data post-processing
included key hydrostratigraphic surface delineation (aquifers and aquitards), and estimates
of hydraulic parameters and groundwater quality. A further aspect revealed by the detailed
3D information is the presence of complex neotectonic faults (current or recent tectonic
motions and deformations) around the Talyawalka area.

Model Design

The model design, process and overall philosophy included:

109
 • Focus on key predictions: assessment of wellfield pumping and recovery impacts on
riparian vegetation, lakes, and river.

• Honour BHMAR datasets and conceptual model, unless evidence suggests otherwise.

• Use of widely variable pertinent field data types to inform calibration.

• Accounting for lack of observed wellfield operation or recovery.

• Use of PEST for calibration and uncertainty analysis to make the most of the compre-
hensive BHMAR data BHMAR.

• Defining likely range of pumping impacts, using the Null Space Monte Carlo (NSMC)
method.

• Calibration to cover as long a period of historical record as practical.

• Detail only where needed, to facilitate practical PEST application.

• Follow good practice, and broadly follow the Australian Modelling Guidelines (Barnett
et al., 2012).

Model Build

Model build used:

• MODFLOW-USG (Panday et al., 2013), with quadtree refinement (from 1280- to 40m;
Figure 1).

• 3 layers from BHMAR geophysics-based 3D model, including substantial processing to

upscale units and represent erosional ‘holes’ through the key aquitard (layer 2). Layers
represent: a shallow unconfined aquifer and clay drape that restricts areal recharge;
Blanchetown Clay (regional aquitard); and Calivil sand aquifer.

• Hydraulic properties and uncertainties derived from downhole and airborne geophysics
(3-4 orders of magnitude often separated minimum and maximum estimates). PEST
was provided this spatially-varying range, with geometric mean as its starting point
and preferred value.

• Evapotranspiration (EVT) implemented from BHMAR datasets (Normalized Differ-

ence Vegetation Index (NDVI)), pan evaporation records and vegetation classification
that includes phreatophyte trees with extinction depths of 10+m.

• Surface water boundary conditions included the River Darling (and weirs), and ephemeral
lakes. Substantial time was spent ensuring correct river and lake wetting and recession
locations, with stress periods designed around flow variability. As lakes dried out,
EVT was implemented in lake beds.

• Model extent defined either by no-flow boundaries, or General Head Boundaries (GHBs)
for flow to or from surrounding formations.

110
 • Abstraction via standard WEL package

• 404 neo-tectonic faults with significant offset represented by enhanced vertical hy-
draulic conductivity between the shallow unconfined and Calivil aquifers, see Figure
2.

• Transient PEST calibration (1985-2015).

• Predictions were hind-cast, assuming pumping started in 2006 with observed climate
from 2006-2016.

Calibration

Observation groups for calibration included groundwater levels, vertical and lateral head
differences and temporal head differences. Lateral head differences included those to/from
surface water features. For identified Groundwater Dependent Ecosystems (GDEs), EVT
ratios and variability between drought and flood from NDVI were also included. A total
of 409 calibratable parameters and 31,953 calibration targets were ultimately used. During
calibration, objective function re-balancing was necessary to favour traditional borehole ob-
servations of groundwater level fluctuation ahead of geophysics derived hydraulic properties.
Spatially constant factors of GA’s minimum and maximum hydraulic conductivity rasters
were initially applied, but spatial flexibility was ultimately required, using pilot points.

Results

A series of 10 abstraction scenarios represented pumping from the two wellfields. NSMC
methods of uncertainty analysis were applied to two scenarios and probabilistic results pre-
sented showing maximum drawdowns, time to recover, and recovery mass balances.

Conclusion

With cheaper spatial data acquisition, and airborne geophysics the modeller’s decades-
long refrain of ‘we need more data’ will more frequently be replaced by a tidal wave of
information. With increasingly large datasets available, greater account is needed of key
model predictions, the worth of large datasets, and detailed model design that adapts to the
constraints imposed by parameter estimation and uncertainty analysis.

Acknowledgements

We acknowledge Water NSW (Tony Paull and Josh White) for their contributions and
permission to publish, the inputs of the GA team and the model reviewers, Hugh Middlemis,
Brian Barnett and Dr John Doherty.

111
 Figure 1. Model domain, USG grid and location of wellfields and surface water features.

Figure 2. GA fault trace data set with estimated overlap of shallow and Calivil aquifers (in m).

References

112
Barnett B, Townley LR, Post V, Evans RE, Hunt RJ, Peeters L, Richardson S,. 2012. Australian groundwater
modelling guidelines. Canberra : Waterlines report, National Water Commission, 2012.
Doherty, J. 2016. PEST, Model-Independent Parameter Estimation User Manual, v6. Brisbane : Watermark
Numerical Computing, 2016.
Geoscience Australia 2012/11. 2012. Securing Broken Hill’s Water Supply: Assessment of Conjunctive
Water Supply Options Involving Managed Aquifer Recharge and/or Groundwater Extraction at Menindee
Lakes: Data acquisition, processing, analysis and interpretation methods. s.l. : Geoscience Australia, 2012.
Geoscience Australia 2012/13. 2012. Assessment of Conjunctive Water Supply Options Involving Managed
Aquifer Recharge Options at Menindee Lakes. s.l. : Geoscience Australia, 2012.
Geoscience Australia 2012/14. 2012. BHMAR Project: Assessment of Potential Groundwater Resources and
Underground Storage Options in the Darling Floodplain near Menindee. Canberra : Geoscience Australia,
2012.
Geoscience Australia 2012/2. 2012. BHMAR Project: Geological and Hydrogeological Framework and
Conceptual Model. Canberra : Geoscience Australia, 2012.
Geoscience Australia. 2012a. BHMAR Project GIS - User Guide. Canberra : GA, 2012a.
Panday, Sorab, Langevin, C.D., Niswonger, R.G., Ibaraki, Motomu, and Hughes, J.D. 2013. MODFLOW–
USG version 1: An unstructured grid version of MODFLOW for simulating groundwater flow and tightly
coupled processes using a control volume finite-difference formulation. s.l. : U.S. Geological Survey Tech-
niques and Methods, book 6, chap. A45, 66 p, 2013.

113
Corresponding author: Eve Kuniansky
U.S. Geological Survey
[email protected]

Use of transmissivity observations with MODFLOW and PEST: Application to

the Floridan Aquifer System

Eve L. Kuniansky, U.S. Geological Survey

J. Bellino
K. Halford

Abstract: A database of over 1,000 transmissivity results from aquifer pumping tests in the
Floridan aquifer system was utilized to develop transmissivity observations for constraining
the calibration of this 100,000 square mile regional aquifer system. The process allows
one to compare the transmissivity of volume of modeled aquifer with the same volume as
determined from individual aquifer tests. These transmissivity observations along with head
and flow observations help constrain the final set of parameters of the model. This is far
superior to using the aquifer pumping test results to extrapolate hydraulic conductivity
values (K) for hydrogeologic units and develop statistic to constrain the range of K for
a zone or pilot point and is more appropriate use of the actual result of the aquifer test
(transmissivity not hydraulic conductivity). Users of the model have more confidence that
the model reflects the aquifer system, when it can be shown that the final spatial distribution
of transmissivity matches field data. The development of transmissivity observations is not
trivial and requires some quality assurance of the aquifer pumping test data. Additionally.
preliminary radial models at each test site must be developed for determination of cells of
the model domain to be compared. Some experimentation with weighting the transmissivity
observations for use with the set of head and flow observations is required.

114
Corresponding author: Andrew Leaf
U.S. Geological Survey
[email protected]

Promise and peril of calibrating a high resolution gridded regional watershed

model

Andrew Leaf, U.S. Geological Survey

R.J. Hunt
D.E. Christiansen
A.E. Haj

Abstract: A high-resolution gridded approach to the simulation of surface water runoff

offers several advantages, including a higher level of representative detail in watershed
characteristics, and in the context of integrated modeling, simplified coupling between the
groundwater and surface-water systems. Moreover, aligning discretization of the surface-
water and groundwater systems allows for similar conceptualizations of important processes
such as ponding and surface runoff. Added parameter flexibility can also reduce structural
error and associated parameter surrogacy in the model, resulting in reduced uncertainty in
model forecasts. Realization of these advantages, however, can increase the runtime of the
forward model such that it can be computationally prohibitive for parameter estimation of
a regional-scale model. In this study, a grid-based integrated watershed model was con-
structed for the 1,000 square mile Bad River Watershed in northern Wisconsin using the
U.S. Geological Survey code GSFLOW. The purpose of the model is to inform management
of water resources that include upland cold-water ecosystems, a deep aquifer important for
water availability, and the largest wild rice estuary in the Great Lakes Basin. The model
employs more than 460,000 hydrologic response units (HRUs) on a uniform grid with a
76.2 meter spacing, commensurate with the underlying MODFLOW model. A step-wise
approach utilizing high-throughput computing was applied to the transient calibration to
overcome both a large number of parameters and long model runtime. This presentation will
highlight the challenges encountered and insights gained from this high-resolution approach.

115
Corresponding author: Neil Manewell
Australasian Groundwater and Environmental Consultants Pty Ltd (AGE)
[email protected]

Comparison of predictive uncertainty methods to quantify Iron Ore dewatering

impacts
Neil Manewell, Australasian Groundwater and Environmental Consultants Pty Ltd (AGE)
Vahid Shapoori, Australasian Groundwater and Environmental Consultants Pty Ltd (AGE)

Abstract: Groundwater models are often overly simplistic representations of complex sys-
tems. Predictive parameter uncertainty of simple groundwater flow models can be quanti-
fied using an array of non-linear techniques. A comparison of formal Bayesian approaches
including GLUE (Generalised Likelihood Uncertainty Estimation), Markov Chain Monte
Carlo using DREAM (DiffeRential Evolution Adaptive Metropolis), and Null-Space Monte
Carlo is made. A large scale numerical model was developed to simulate groundwater
impacts from iron ore mining in Western Australia and was assessed using three predic-
tive uncertainty techniques. Our results demonstrate that GLUE and DREAM produce
broader groundwater impacts when compared to Null-Space Monte Carlo results. GLUE
and DREAM analysis reveals multi-normal and non-normal parameter distributions using
information from the observation dataset. The key is to improve the understanding of the
methodology, and to analyze posterior parameter distributions to ensure worse case impacts
are fully explored. This study provides a framework to quantify uncertainty using regional
models when highly parameterised inversion is not possible.

Extended abstract: Methodology, findings, and implications

Introduction

Groundwater numerical models are often used to predict potential drawdown extents
and pit dewatering rates. Given the simplification of hydrogeology and subsurface processes
implemented in groundwater models, it is crucial to explore the uncertainty of model pre-
dictions. In literature, several methods have been proposed for quantifying the uncertainty
of groundwater models. However, each uncertainty method potentially produces different
prediction ranges. For a rigorous uncertainty investigation, an assessment is required to
investigate the range that each uncertainty method can produce.
To quantify the uncertainty associated with groundwater dewatering and drawdown ex-
tent from a proposed iron ore proposed mine in Western Australia, a 3D numerical groundwa-
ter flow model was first developed in MODFLOW-USG (Panday, 2015). The model consists
of 7 layers at 50 x 50 m cell resolution, with a total of 75 adjustable parameter zones.
The model was calibrated using field test data and two different calibration methods:
Levenberg gradient based optimization with singular value decomposition, and global opti-
mization using Shuffled-Complex Evolution (SCEUA_P) (Watermark Numerical Comput-
ing, 2010). The field test data includes a series of Constant Rate Tests (CRT) and a Hydrody-
namic Trial (HDT), and were simulated in the model by the WEL package. The calibration

116
results showed that the SCEUA_P calibration provided the best replication of measured
drawdown response. The calibrated model was then used to predict the dewatering rate and
drawdown extent of a proposed pit development with an 11 year mine life.

Research Method

To quantify the predictive uncertainty of regional drawdown and dewatering volumes to

the mine workings, GLUE (Generalised Likelihood Uncertainty Estimation), Markov Chain
Monte Carlo using DREAM (DiffeRential Evolution Adaptive Metropolis), and Null-Space
Monte Carlo were applied.

GLUE (Generalised Likelihood Uncertainty Estimation)

Often there is sparse knowledge of the probability distributions to apply to the parame-
ters in the model. Hence, the model needs to be run numerous times to explore groundwater
impacts using different combinations of parameters across a wide range of values. A GLUE
methodology was applied to generate parameter sets using a latin-hypercubic sampling tech-
nique for horizontal and vertical hydraulic conductivity, specific yield, and specific storage
values. Two thousand randomly distributed realizations were produced. Each realization was
simulated and tested against the best realization achieved through calibration and rejected
if the error was exceeded by more than 100%. Using this method, Two hundred realizations
were deemed suitable for the predictive uncertainty analysis.

DREAM (DiffeRential Evolution Adaptive Metropolis)

Creating random realizations using GLUE to meet strict calibration criteria can be ex-
tremely inefficient, resulting in high “hit-to-miss” ratios, given the unguided nature of pa-
rameter distribution. The DREAM algorithm (Vrugt, 2015) is more efficient as it employs a
genetic algorithm to derive parameter sets sequentially closer to the optimal phi value with
each evolution. In this study, the DREAM Fortran package was modified to work in a par-
allel high performance computing environment, and to apply a logarithmically distributed
density function.
Similar to the GLUE analysis, a flat distribution of prior parameter sets were assumed.
DREAM was then run with 30 chains using initial parameter sets derived from prior distri-
butions. Ten thousand runs were completed before model convergence was achieved, with
the last 10% of samples in each chain used to derive the posterior parameter distributions.
Using this method, 475 runs were deemed suitable for predictive analysis.

Null-space Monte Carlo

Null-space Monte Carlo uncertainty analysis requires the covariance matrix c(K), which
represents the relationship between the model parameters and the sensitivities to the ob-
servation dataset. This approach is typically suited to highly parameterised environments
where a vast array of adjustable parameters is employed via pilot point multiplying fields
(Doherty, 2015).

117
 Posterior parameter distributions were defined using the posterior standard deviations de-
fined in the Jacobian at the optimal parameter set, and were not recalibrated per realization.
Instead, a phi rejection process was employed, resulting in a total 300 model realizations.
Parameter distributions were restricted to the same bounds as defined in the GLUE and
DREAM exercises, using log-normal distributions centered on the optimal parameter value.

Results

The uncertainty of the maximum drawdown experienced during the predictive mine life
significantly varied across the three methods. The results indicate that the optimal solution
(base) experienced drawdowns 2 m at a distance of up to 4.7 kilometers (km) to the im-
mediate west from the mining area (Figure 1). The P95 exceedance probabilities for GLUE
and DREAM analysis revealed maximum drawdown extents up to 11.2 km and 10.2 km,
respectively. Null-space analysis predicted P95 drawdowns up to 2.7 km from the mine, a
likely result of highly constrained parameter distributions.
With regards to predicted inflow rates, the result shows that all three methods provide
similar P5 and P95 percentiles and suggests a maximum likely inflow rate of ̃50 ML/day.
Posterior parameter distributions varied substantially between the three methods. Due
to the random sampling of the GLUE approach and the generalized nature of the phi rejec-
tion process, refinement of distribution of the calibrated parameter sets was poorly defined
(Figure 2). Parameter distributions defined by the DREAM approach was better defined,
and produced multi-normal parameter distributions for parameter sets that were sensitive
to the observation dataset. Null-space analysis revealed significant constraints to parame-
ters displaying high values of identifiability. As a result, certain sensitive parameters to the
predictive impacts displayed a narrow range.

Conclusions

The results of the three different methodologies demonstrate that DREAM is the most
efficient approach to quantify predictive uncertainty using computationally intensive regional
models. GLUE produces similar results, although parameter distributions are poorly defined.
Comparison between the prior and posterior parameter sets is crucial to instill confidence in
the range of predictions.
However, for highly parameterized models, Null-space Monte Carlo is still the recom-
mended approach as both GLUE and DREAM require many thousands of model runs to
properly replicate likely parameter interactions to derive reliable uncertainty predictions.

Acknowledgements

Thanks to Gareth Price at BHP Iron Ore Australia for initiating and supporting the
background study this paper is based upon.

118
 Figure 1. Model grid showing P95 maximum predicted drawdown extents.

Figure 2. Posterior distribution of sample parameters sets versus the optimum (calibrated) value (black
line).

References

119
Doherty, J., 2015. Calibration and Uncertainty Analysis for Complex Environmental Models. Watermark
Numerical Computing, Brisbane, Australia. ISBN: 978-0-9943786-0-6.
Panday, Sorab, Langevin, C.D., Niswonger, R.G., Ibaraki, Motomu, and Hughes, J.D., 2015, MODFLOW-
USG version 1.3.00: An unstructured grid version of MODFLOW for simulating groundwater flow and tightly
coupled processes using a control volume finite-difference formulation: U.S. Geological Survey Software
Release, 01 December 2015, http://dx.doi.org/10.5066/F7R20ZFJ
Vrugt, J., 2015, Markov chain Monte Carlo Simulation Using the DREAM Software Package: Theory,
Concepts, and MATLAB Implementation
Watermark Numerical Computing, 2010, PEST model independent parameter estimation user’s guide (5th
ed.). Watermark Numerical Computing

120
Corresponding author: Killian Miller
Aquanty Inc.
[email protected]

Efficient Quantification of Uncertainty in Integrated Surface and Subsurface

Hydrologic Simulations

Killian L. Miller, Aquanty Inc.

Y-J. Park
S. Berg
E. Sudicky

Abstract: As a result of today’s high performance computers, sophisticated numerical mod-

els are capable of incorporating many complex processes. Partial or incomplete knowledge
of these processes and the input parameters that describe them necessitates the modeller to
make various approximations, and consequently introduces uncertainty into the numerical
model. The aim of parametric uncertainty quantification is to measure the variability in
model responses propagated by uncertain model inputs, thus providing the modeller with a
certain level of confidence regarding the predictions made by their model. Sampling methods
such as Monte Carlo simulation are the workhorse methods for uncertainty quantification,
however, when a simulation requires on the order of days or even weeks to finish, the ap-
plication of such methods is largely infeasible. In this work we investigate the probabilistic
collocation method (PCM) for quantification of uncertainty in fully-integrated, physics-
based hydrologic simulations using the HydroGeoSphere model. In contrast to sampling
methods, PCM represents a model response of interest as a weighted sum of orthogonal
polynomials over a space of random inputs. Once a PCM expansion has been constructed,
histograms of the PDF and CDF as well as local and global sensitivity parameters may
be obtained with minimal further computation. Numerical experiments confirm that PCM
computes the mean and variance to a similar accuracy as Monte Carlo simulation often
with significantly fewer model runs and thus has the potential for significant speedup over
traditional methods. We demonstrate the efficiency and accuracy of our approach through
extensive numerical experiments involving direct comparison with sampling methods.

121
Corresponding author: Tung Nguyen
Washington State University
[email protected]

Soil and geologic controls on recharge and groundwater flow response to

climate change: A case study of the Yakima River Basin

Tung T. Nguyen, Washington State University

Jennifer C. Adam, Washington State University

Abstract: The Yakima River Basin (YRB) is one of the most important agricultural basins
in Washington State with annual revenues in excess of ;3.2 billion. This intensively irri-
gated basin is, however, one of the state’s most climatically sensitive water resources system
as it heavily relies on winter snowpack and its limited reservoir storage. Water shortages
and droughts are expected to be more frequent with climate change, population growth and
increasing agricultural demand. This could result in significant impacts on the groundwater
system and subsequently the Yakima river. The goal of this study is to assess how soil
and geologic characteristics affect catchment recharge and groundwater flow across three
catchments within the YRB using a coupled framework including a physically based hydro-
ecological model, the Regional Hydro-Ecologic Simulation System (RHESSys) and a ground
water model, MODFLOW. Soil and geologic-related parameters were randomly sampled to
use within the Distributed Evaluation of Local Sensitivity Analysis (DELSA) framework to
explore their roles in governing catchment recharge and groundwater flow to climate per-
turbation. Preliminarily results show that soil field capacity is the most important factor in
controlling catchment recharge response to climate perturbations while transmissivity is the
main control governing flow response to changes in low flow conditions. Interestingly, best
calibrated parameter sets do not follow the same trend in sensitivity analysis experiment.
The modeling framework developed in this study will be used to investigates the impacts of
both climate and drought-relief supplemental pumping on potential recharge, groundwater
and streamflow changes in the YRB.

122
Corresponding author: Philip Nienhuis
Waternet
[email protected]

Generating Initial Fresh/Saline Groundwater Distributions for a 3D Modflow

SWI Model using 2D Cross-Sectional Seawat Model as a Proxy

Philip Nienhuis, Waternet

Pierre Kamps, Waternet

Abstract: The fresh/saline groundwater flow system under the Amsterdam Water Supply
Dunes (AWD) at large has a response time on the scale of centuries; present movements
of the fresh/saline interface (FSI) are still affected by transient stresses that occurred in
the 19th and 20th century. Long-term model predictions depend on adequately simulated
future FSI movements. Fresh/saline groundwater flow in AMWADU, the transient semi-
3D groundwater model of the AWD, is simulated the SWI/SWI2 packages. Its transient
calibration period starts around 1980 as only then the first overall, yet locally patchy, in-
ventory was made of the FSI position. That inventory has been reworked into a modeled
FSI starting position. The reliability of the simulated FSI movements obviously depends on
the quality of that constructed initial position. The development of the fresh/saline ground-
water distribution under the AWD since 1500 has been simulated with Seawat in transient
2D cross-sections extracted from AMWADU. Provisional validation against chloride data of
simulated FSI positions and movements through time showed good agreements especially
after 1975. As the 2D simulations provide continuous coverage of the FSI through time,
albeit in a strip, the results of the computed FSI movement in the 2D models have been
used to augment and improve the initial interpolated FSI position for the 3D model. Com-
paring the results with field data illustrates the importance of a properly defined initial FSI
starting position.

Extended abstract: Methodology, findings, and implications

Introduction

The fresh/saline groundwater flow system in the Amsterdam Water Supply Dunes (AWD;
Figure 1) and its surroundings has a response time to geohydrological stresses on the scale
of centuries (Nienhuis et al., 2013, 2014). As a consequence present-day movements of the
fresh/saline interface (FSI) are still affected by transient stresses that occurred in the 19th
and 20th century.
The most important historical geohydrological stresses that acted on the groundwater in
the AWD and surroundings have been identified as follows:

• S1: Build-up of the fresh water lens from precipitation. How this actually occurred is
only known qualitatively at best. Yet whatever happened, around 1850 AD a stable
fresh water lens in dynamic equilibrium with the hydraulic / hydrological boundary
conditions at the time must have existed, as far as we know undisturbed by hydraulic
stresses resulting from human interventions.

123
 • S2: Reclamation of the relatively large Haarlemmermeer (“Lake Haarlem”) in 1850,
some 3 to 4 km to the East of the AWD. This led to a sudden drop in local hydraulic
potential from around sea level to 6 m below sea level, in turn inducing a strong,
eastward regional saline/brackish groundwater flow system below the fresh water lens,
from the North Sea to the Haarlemmermeerpolder.

• S3: Start of the drinking water supply from the AWD in 1853, initially tapping the
phreatic aquifer using a system of abstraction canals that was gradually expanded.
Obviously this meant capturing an increasing portion of the natural recharge that was
otherwise used to sustain the dynamic equilibrium position of the fresh water lens.

• S4: Directly tapping the deeper aquifer(s) as of 1903, invoking an ever increasing
number of deepwells, in later stages leading to severe overexploitation of the fresh
water lens.

• S5: Implementation of large-scale artificial infiltration of pretreated water from the

Rhine river during the period 1957-1967. Since 1967 about 80% of the total drinking
water abstraction from the AWD is being recharged by pretreated Rhine water. As a
result the once overexploited fresh water lens has largely recuperated to date.

3D transient model simulations with Modflow/SWI

Fresh/saline groundwater flow in AMWADU, the Modflow-based transient semi-3D ground-

water model of the AWD (Kamps et al, 2005), has been simulated with the SWI package
(Bakker et al., 2003) because when AMWADU was built around 2005 Modflow and SWI
provided the only practical option for modeling the complicated geohydrological situation in
and around the AWD. AMWADU’s transient calibration period starts only around 1984 for
a number of reasons:

• A first overall interpretation of the FSI position for the entire AWD was made in the
early 1980s.

• In the first half of the 20th century the fresh/saline groundwater system has undergone
severe overexploitation of the fresh water lens, leading to generation of large amounts
of brackish water. This cannot be simulated using the SWI package.

The reliability of the simulated FSI movements in AMWADU obviously depends on the
quality of the constructed initial FSI position (taken to be the 10,000 mg/L isosurface).
Most chloride field data only cover the AWD proper; outside the AWD only scarce chloride
data are available. Between data locations the FSI had to be interpolated. As a result the
constructed initial FSI position in 1984, especially where at some distance from the AWD,
is the best what could be obtained at the time.

2D transient model simulations with Seawat

124
 Recently Nienhuis et al. (2013, 2014) have simulated transient fresh/saline groundwater
flow under the AWD in 2D cross-sections with Seawat and mflab (Olsthoorn, 2010). The
original objective of these simulations was to get an impression of possible effects of prolonged
future seasonal fresh water storage in the fresh water lens (Nienhuis and Olsthoorn, 2012).
Avoiding mixing up these effects with artefacts due to erroneous initial FSI input was a major
point of concern. The idea was to generate a stable fresh/saline groundwater distribution
sufficiently representative for the year 1850 AD based on a hypothetical development of the
fresh/saline groundwater distribution under the AWD since Medieval times (see stress (1)
above); and from then on simulate the stresses (2)-(5) above to obtain an initial transient
situation representative of 2020, followed by simulating seasonal storage until 2070. To be
able to obtain a reasonably realistic fresh water lens for 1850 AD an initial fresh/saline
groundwater distribution in a wide strip along the Dutch coast during Medieval times had
to be inferred from indirect information, viz. historical maps and general hydrogeological
insights. A palaeohydrological reconstruction on a much large scale by Delsman et al. (2013)
proved very helpful for filling in unknown boundary conditions in the easternmost parts of
our cross-section models.
Provisional validation of this cross-section model against available chloride data of the
simulated FSI positions and movements through time showed good agreements especially
for the period after 1975. As this and additional 2D simulations elsewhere in the AWD
provide continuous coverage of the FSI through time, albeit in a strip, the results of the
computed FSI movement in the 2D models have been used to augment and improve the
initial interpolated FSI position for the SWI/SWI-2 based 3D model.
As an example the FSI-positions for AMWADU in 1984 (starting position) and 2009 (end
of simulation run) have been plotted in a cross-section simulated with Seawat in Figure 2. As
can be seen the constructed FSI only marginally conforms to the fresh/saline groundwater
distribution simulated with Seawat, especially in the east where field data are lacking. As
the simulation continues, the FSI gradually moves into a position that matches with the
Seawat simulation. This implies that for several years the SWI-simulated FSI movements
are mere artefacts introduced by lack of data.

Conclusions and discussion

Seawat-based transient 2D cross-section simulations are useful to help construct the initial
FSI position for the AMWADU model for places where no data are available. In between
data locations the cross-section model is expected to yield better results as it also takes into
account salinization / flushing of aquitards.

125
Figure 1. Location of Amsterdam Water Supply Dunes and some modelled 2D cross-sections (see text). The
lowermost dark cross-section is referred in Figure 2.

Figure 2. Comparing initial constructed FSI position in 1984 with 2D Seawat cross-section model (top) and
the simulated position in 2009 (bottom). White dotted: constructed/simulated FSI position; colored dots:
chloride concentrations at vertical positions inferred by interpolation between measured piezometer depths.
The location of the cross-section is shown in Figure 1.

126
 References

Bakker, M, Schaars, F.W., 2003. The Sea Water Intrusion (SWI) Package manual. Part I Theory, User
Manual and Examples. http://www.engr.uga.edu/people/mbakker/swi.html
Delsman, J.R., K.R.M. Hu-a-ng, P.C. Vos, P.G.B. de Louw, G.H.P. Oude Essink, P.J. Stuyfzand and M.F.P.
Bierkens (2013) Palaeo-modeling of coastal salt water intrusion during the Holocene: an application to
the Netherlands. Hydrol. Earth Syst. Sci. Discuss., 10, 13707–13742. Kamps, P.W.T.J, Schaars, F.W.,
Olsthoorn, T.N., 2005. Calibration of the model of the Amsterdam Dune Area using multiple stresses. Poster
presented at ModelCare 2006, Scheveningen, June 2-6, 2006
Nienhuis, P.R. and T.O. Olsthoorn (2012) Modelling AS(T)R in the Amsterdam Water Supply Dunes to
Explore Options to Cope with Future Intake Water Shortages Caused by Climate Change. Proc. 21th
SWIM, Buzios (Brazil), June 17-22, 2012, p.138.
Nienhuis, P.R., Olsthoorn, T.O. and Kamps, P.T.W.J. (2013) 160 Years of History of the Amsterdam Water
Supply Dune Area Modeled with Variable Density, Outlook into the Future. Proc. Modflow and More 2013,
Golden (CO), June 2-5, 2013, p.443-447.
Nienhuis, P.R., Olsthoorn, T.O. and Kamps, P.T.W.J. (2014) Hydrological history of Amsterdam Water
Supply Dunes simulated with a 2D cross-section model. Proc. 23rd Salt Water Intrusion Meeting, Husum
(Germany), June 16-20, 2014, p. 265-268.
Olsthoorn, T.N. 2010. http://code.google.com/p/mflab

127
Corresponding author: James Ross
Tetra Tech, Inc.
[email protected]

The Ensemble Kalman Filter for Groundwater Plume Characterization: A

Pilot Study

James L. Ross, Ph.D., Tetra Tech, Inc.

Peter F. Andersen, PE, Tetra Tech, Inc.

Abstract: The Kalman filter is an efficient tool to refine an estimate of a state variable using
measured data and the variable’s correlations in space and/or time. The ensemble Kalman
filter (EKF) (Evensen, 2009) is a Kalman filter variant that employs Monte Carlo analysis to
define the correlations that help to refine the updated state. While use of EKF in hydrology
is limited, it has been applied in other fields of engineering (e.g. oil reservoir modeling,
weather forecasting). Here, EKF is used to refine a simulated groundwater TCE plume that
underlies the Tooele Army Depot-North (TEAD-N) in Utah, based on observations of TCE
in the aquifer. The resulting EKF assimilated plume is simulated forward in time to predict
future plume migration. The correlations that underpin EKF updating are predicated on
multiple realizations of a well-calibrated SEAWAT groundwater flow and transport model
that simulates the historical plume emanation from former sources at TEAD-N. The EKF
methodology is compared to an ordinary kriging assimilation method with respect to the
accurate representation of plume concentrations in order to determine the relative efficacy
of EKF for water quality data assimilation.

Extended abstract: Methodology, findings, and implications

Data Assimilation

Calibration of the TEAD-N groundwater flow and transport model to thousands of flow
and transport targets culminates in the simulation of a January 2015 TCE plume that reflects
observed conditions with reasonable accuracy (Figure 1). In order to best define initial plume
conditions for predictions of TCE plume migration, modelers refine the simulated plume by
incorporating observed TCE concentrations. Two methods of such data assimilation have
been evaluated: a kriging-based method and an EKF-based method

Kriging-Based Method

In the kriging-based method, modelers combined observed and simulated TCE concentra-
tions via ordinary kriging. In this approach, observed TCE concentrations were interpolated
to all model grid cells to produce a 3D representation of the plume (Figure 1) and an es-
timate of the kriging standard deviation. Near observation wells, the standard deviation is
near-zero, while standard deviation increases with increasing distance from TCE observation
locations. Using this standard deviation, modelers calculated plume assimilation weights for
interpolated and simulated TCE concentrations at each model grid cell. These weights range
from zero to one; at grid cells proximate to monitoring wells, the kriged TCE concentration
weight is at or near one and the simulated TCE concentration is at or near zero. Away
from monitoring wells, the magnitude of the respective weights invert. At each grid cell

128
location the assimilated plume TCE concentration is a weighted sum of the interpolated and
simulated TCE concentration. This tool was employed to define the TCE plume for January
2015, using simulated and observed TCE concentrations from around that time (Figure 2).
Note that though the assimilated plume shares many characteristics with the interpolated
plume, there are features that are informed by the simulated present-day plume.
While this method is straightforward, kriging is a static tool that does not consider
aquifer characteristics (including extensive faulting throughout the site), TCE source history,
historical hydrology, or any other stresses that may have impacted the nature and location
of the present day TCE plume.

EKF-Based Method

An alternative approach evaluated by modelers defines correlations between concentra-

tions throughout the TCE plume based on the simulation of historical migration of the TCE
plume. This EKF-based tool seeks to update the simulated present-day TCE plume with ob-
served present-day TCE concentrations. How TCE concentrations throughout the simulated
plume are refined by the observed TCE concentrations is governed by the aforementioned
correlations. These correlations are informed by a Monte Carlo-based construction and
simulation of a multitude of model realizations (an ensemble) that are predicated on the cal-
ibrated groundwater flow and transport model. The resulting ensemble of simulated model
results are used to define how TCE concentrations throughout the present-day plume are
related to aquifer conditions and source characteristics, as well as how they are related to
each other.
Based on the TCE plume correlations and present day observed TCE concentrations,
the simulated TCE plume (Figure 1) was updated via EKF. Like the kriging-based plume,
the EKF assimilated TCE plume (Figure 2) shares many characteristics with the simulated
TCE plume. However, the EKF-assimilated plume is both larger than the simulated plume
and has generally higher concentrations than both the simulated and kriging-assimilated
plumes, particularly in areas with no monitoring wells. Residuals between observed TCE
concentrations and those produced by the two methodologies show that the EKF-based
method generally over-simulates TCE concentrations at monitoring well locations.

Bootstrapping Analysis

The higher concentrations between monitoring wells in the EKF-assimilated plume raised
questions about accuracy of the plume in areas with no observed concentrations. So, a
bootstrapping analysis was conducted, wherein individual wells’ data were withheld from
the observation dataset and the two data assimilation methods executed, in order to see how
well each method estimated the TCE concentration at the omitted monitoring well locations.
This was analysis was conducted for eight different wells located throughout the plume. The
assimilated plume TCE concentration coincident with each monitoring well location was
compared to the observed concentration, and residual calculated. The residuals showed that
EKF tends to over-simulate relatively low TCE concentrations, but better estimates elevated
concentrations. This statistical test also revealed the potential value of EKF at identifying
redundancies in the existing monitoring network and sampling plan, as the impact of omitting
different data produced variable changes to the updated TCE plumes.

129
 Conclusions

Results of the EKF assimilation of simulated and observed data demonstrated that this
approach produced a reasonable, yet slightly less accurate TCE plume based on available
data. Additionally, a bootstrapping analysis illustrated that while the kriging-based method
was able to predict TCE concentrations at monitoring well locations with slightly greater
accuracy than the EKF method, the EKF approach was better at predicting higher con-
centrations. Further evaluations of the EKF methodology will be conducted to gauge its
appropriateness for assimilating data at this site. Whereas, EKF’s utilities as a model cali-
brator and monitoring network optimization tool were not evaluated, the value of this tool
in such capacities were became apparent during the data assimilation evaluations.

Figure 1. Model-simulated (left) and interpolated (right) January 2015 TCE plume. Colored circles represent
monitoring well locations and the TCE concentrations observed at those wells. Note that in these figures, as
well as those in Figure 2, the plume color flood at each location is the maximum concentration throughout
the depth of the model so as to represent a three-dimensional plume in two-dimensions.

130
 Figure 2. Assimilated January 2015 plumes produced by the kriging (left) and EKF (right) methods.

References

Evensen, Geir. 2009. Data Assimilation: The Ensemble Kalman Filter, 2nd Edition. New York: Springer,
2009.

131
Corresponding author: Anna Ryken
Colorado School of Mines
[email protected]

Identifying Controls on Carbon Exchanges in High Altitude Headwaters to

Improve Representation in Earth System Models

Anna Ryken, Colorado School of Mines

R. Maxwell, Colorado School of Mines
D. Gochis, University Corporation for Atmospheric Research
K. Williams, Lawrence Berkeley National Lab

Abstract: The hydrology of high-elevation, mountainous regions is poorly represented in

Earth Systems Models (ESMs). In addition to regulating downstream water delivery, these
ecosystems play an important role in the storage and land-atmosphere exchange of carbon
and water. In an effort to resolve this hydrologic gap in ESMs, this study seeks to better
understand the interactions between groundwater, carbon flux, and the lower atmosphere
in these high-altitude environments through integration of field observations and model
simulations. Here we use observations from a meteorological station co-located with an
eddy covariance tower in the East River Basin - a Colorado River headwaters basin, which is
emblematic of other high-elevation basins. The meteorological and carbon flux data collected
from these instruments over water year 2017, coupled with snow surveys, will be used
to force an integrated single-column hydrologic model, ParFlow-CLM. These observations
will be used to better constrain the water, carbon, and energy fluxes in the coupled land-
atmosphere model to increase our understanding of high-altitude headwaters. Through
obtaining more accurate and higher resolution evapotranspiration and carbon cycling data
and applying it to a coupled hydrologic model, this study can lead to better representation
of mountainous environments in all Earth Systems Models.

132
Corresponding author: Judith Schenk
Colorado School of Mines
[email protected]

Demystifying Information and Fisher Information: Observation Data Reveals

the Good and the Bad of Our Models

Judith Schenk, Colorado School of Mines, Golden, CO, USA

Eileen Poeter, Professor Emeritus, Colorado School of Mines, Golden, CO and Poeter Engineering, USA
William Navidi, Colorado School of Mines, Golden, CO, USA

Abstract: Observation data reveals information about groundwater models that can be
used to quantitatively compare uncertainty on parameters and predictions between mod-
els of the same system. Information (I ) is quantified as the determinant of the weighted
Jacobian (sensitivity) matrix. Fisher Information (FI ) is quantified as the determinant of
the weighted sensitivity matrix divided by the residual variance. One-dimensional models
are used to demonstrate the relationship between I and FI and the resulting model uncer-
tainty on parameter estimates and model predictions for: 1) increasing model complexity;
2) differing model structure; 3) differing boundary conditions; and 4) over-fitted models.
Greater model complexity results in increased I, but more complex models produce greater
uncertainty in parameter estimates. FI generally increases with increasing model complex-
ity. FI may decrease with increasing model complexity if a model structure is inconsistent
with the underlying true structure of an aquifer system. Models with lower FI have a
higher level of disorder, produce greater uncertainty of parameter estimates, and generally
produce greater uncertainty of model predictions. A constant-head boundary condition can
result in lower I and FI as compared to a constant-outflow boundary. The constant-head
boundary constrains model response, thus reducing sensitivity of simulated equivalents to
changes in parameter values. Such models produce lower I and FI, and have a higher level
of disorder and a larger confidence interval on predictions as compared to models with a
constant-outflow boundary. Complex, over-fitted models with high correlations between pa-
rameter estimates lose both I and FI because highly-correlated parameter estimates have
greater uncertainties.

Extended abstract: Methodology, findings, and implications

Observation data, used to calibrate groundwater models, provide information for comparison
of quantitative uncertainty associated with both estimated parameter values and model
predictions for different models of a groundwater system calibrated with the same observation
data set. The natural log of the determinant of the information (I ) matrix demonstrates how
increasing model complexity results in increasing information unless the model is deficient.
The Fisher Information Matrix includes I and the residual variance term, −𝑝 ln (𝜎2 ) where
p is the number of estimated parameters and 𝜎2 is the residual variance (sum of weighted-
square residuals divided by the number of observations). The natural log of the determinant
of the Fisher Information matrix (FI ) provides information about the relative disorder of a
model within a set of models (Frieden, 2000).
One-dimensional synthetic models with a complex distribution of hydraulic conductivity
are created to represent two true, but unknown, systems. Both systems have an up-gradient
lake boundary. SynM1 has a down-gradient stream boundary (Figure 1a) and SynM2 has

133
a down-gradient discharge boundary (Figure 1b). SynM1 and SynM2 have the same distri-
bution of hydraulic head. These systems are sampled for multiple head observations and an
observation of flow into the system at the up-gradient lake boundary to provide observations
for calibrating experimental models. SynM1 and SynM2 are stressed by injecting water to
the system in the up-gradient portion of the model to provide a true prediction of change
in head (Figure 1a,b). True predictions are never known, but synthetic systems SynM1
and SynM2 can be used to compare experimental model predictions to a “true”, but always
unknown, prediction.
Experimental models are created to serve as constructed models of the true but unknown
systems SynM1 or SynM2 (Figure 1c). M1 models correspond to system SynM1, and M2
models correspond to system SynM2. M1 models have a down-gradient constant head to
represent the stream boundary, and M2 models have a down-gradient constant flow boundary
to represent the known discharge. Both model sets have an up-gradient constant head
boundary to represent the lake. A set of six M1 or M2 models with different levels of model
complexity (Figure 1b) are used to demonstrate how I and FI are affected by: (1) increasing
model complexity in terms of increasing number of estimated parameter values, (2) different
model structures, and (3) different down-gradient boundary conditions. M1 and M2 models
are identified by the number of parameters (2K, 3K, 4K, 5K, 6K, 8K). Each M1 and M2
model is first calibrated using non-linear regression with a data set consisting of one flow
observation and ten head observations (blue circles in Figure 1a,b). Each calibrated model
is stressed by a flux to a portion of the model (Figure 1a,b), and a prediction of change
in head is made in the same location in each experimental model where the prediction is
generated in SynM1 and SynM2 (Figure 1a,b). The sensitivity of each simulated equivalent
observation to each estimated parameter produces the Jacobian matrix (Hill and Tiedemen,
2007) . I and FI are calculated for each experimental model.
For both M1 and M2 conditions, the constructed 3K and 5K models produce a larger
residual variance because their structure is inconsistent with the underlying structure of
the true system (Figure 2a,b). Increasing model complexity results in increasing I for M1
(component ln |𝑋 T 𝜔𝑋| Figure 2c) and M2 (Figure 2d) models, but increasing I results in
greater uncertainty of parameter estimates and commonly greater uncertainty of predictions.
Models 3K and 5K have a higher level of disorder than the other models in the set because
of the larger error associated with model fit reflected in the value of −𝑝 ln (𝜎2 ) (Figure 2c,d).
FI is the sum of I and −𝑝 ln (𝜎2 ). Models 3K and 5K, therefore, have lower FI (ln |𝐹 𝐼|
Figure 2c,d) resulting in greater uncertainty associated with parameter estimates and model
predictions (Figure 2e,f). M2 models with a down-gradient constant-outflow boundary allow
for heads in the down-gradient portion of the model to adjust more during calibration than in
M1 models. Therefore, simulated equivalents in the down-gradient area of the model exhibit
greater sensitivity to estimated parameters as compared to models with a down-gradient
constant-head boundary. The greater sensitivity of observations to estimated parameters
for this constant-outflow condition in M2 models results in improved model fit (Figure 2b),
higher I, and higher FI (Figure 2d) compared to M1 models (Figure 2c,d), indicating a
higher level of order for M2 models as a result. The higher level of order for these models
is reflected in lower uncertainty of parameter estimates and predictions as compared to M1
models, and the 95% individual confidence interval on the prediction is narrower (Figure
2e,f).

134
 Models are sometimes over fit to observation data giving the appearance that the model
is reproducing the true system response, but closer examination may indicate these models
are deficient in that one or more parameters cannot be estimated given the observation
data. A set of six M1 models with different model complexities (2K, 4K, 6K, 8K, 10K
and 12K) are calibrated with 18 head observations and one flow observation (blue circles
plus blue x observations, Figure 1a). All models reached convergence during the non-linear
regression. Model 12K has the lowest model error (Figure 2g). Both I and FI increase
with increasing model complexity for all models except model 12K. The lower FI for model
12K as compared to model 10K indicates a higher level of disorder for the more complex
model. The determinant of the correlation matrix, which is an indicator of how well model
parameters can be estimated, is 3x10-32 , indicates one or more of the parameters cannot
be well estimated (Figure 2h). The condition number of the Jacobian matrix is 3x108 ,
indicating small errors in model input (observation data) can result in large error of model
output (parameter estimates). Parameters K9 and K10 in model 12K are highly correlated
indicating these parameters cannot be estimated independently, thus 12K is a deficient model
and should not be included if the model set is evaluated using multi-model analysis.

Comprehensive Explanations of Figures

(Figure 1.)

Two synthetic models represent true but unknown systems. Both systems have an up-
gradient lake boundary, and a complex distribution of hydraulic conductivity. SynM1 has
a down-gradient stream boundary (a), and SynM2 has a down-gradient discharge boundary
(b). Each system is sampled for one flow observation at the up-gradient end, and ten head
observations (blue circles) that are used to calibrate experimental models. SynM1 and SynM2
are stressed with an additional flux of water over a portion of each system (light blue arrows),
and a prediction is made of the change in head. A set of experimental models with varying
hydraulic conductivity structure and coarser grid spacing than SynM1 or SynM2, represent
constructed models of the “unknown” systems (c). M1 models correspond to SynM1 with a
down-gradient constant-head boundary, and M2 models correspond to SynM2 with a down-
gradient constant-outflow boundary. SynM1 is additionally sampled for eight more head
observations for a second example presented in Figure 2g,h,i (a, blue x’s).

(Figure 2.)

Figure 2. Six M1 and six M2 models are calibrated with the same data set. Models 3K
and 5K have a poor model structure that results in a higher model error (a,b). Information
increases with increasing model complexity, but Fisher Information decreases for models
3K and 5K because these models have a higher level of disorder (c,d) resulting in higher
uncertainty of predictions (e,f). M1 models have a higher level of disorder as compared to
M2 models resulting in comparatively lower Information and Fisher Information (c,d) and
greater uncertainty of predictions (e,f). A second example consists of six M1 models with
increasing model complexity calibrated with eight additional head observations. The most
complex model produces the best fit to the data (g). Information and Fisher Information
increase with increasing complexity for five of the models, but decrease for model 12K because

135
the observation data do not support that model structure (h). Model deficiency is reflected
in a low value of the determinant of the correlation matrix (i), a high value of the condition
number (not shown) and near perfect correlation between parameters K9 and K10.

Figure 1.

136
 Figure 2.

References

Frieden, B.R., 2000, Physics from Fisher Information – A unification; Cambridge University Press, Cam-
bridge, 328 p.
Hill, M.C. and Tiedeman, C.R. (2007), Effective groundwater model calibration with analysis of data, sen-
sitivities, predictions, and uncertainty, New York, John Wiley & Sons, 455 p.

137
Corresponding author: Haruko Wainwright
Lawrence Berkeley National Laboratory
[email protected]

Global Sensitivity Analysis for Supporting Long-Term Monitoring during

Sustainable Remediation

Haruko M. Wainwright, Lawrence Berkeley National Laboratory

Abstract: ”Sustainable remediation” has recently emerged as a key concept for better re-
mediating contaminated soil and groundwater. The sustainable remediation considers not
only contaminant removal but also cost effectiveness, greenhouse emission, waste production
and energy usage. It highlights the use of passive remediation technologies to immobilize
contaminants or natural attenuation with longer institutional control, both of which reduce
remediation cost and waste significantly. However, such passive remediation requires long-
term monitoring to ensure that those immobilized contaminants would be stable and not
compromise public health. In this study, we first introduce an innovative long-term monitor-
ing technology based on in situ automated sensors, aiming to reduce groundwater sampling
and monitoring cost, and also to serve as an early warning system for detecting changes in
plume mobility. In situ sensors can measure the key properties (such as pH, groundwater
level, redox potential) that control plume mobility and its distributions. We have developed
a machine learning method to continuously estimate contaminant concentrations based on
those in situ datasets. In addition, we have developed a flow and contaminant transport
model to predict the long-term efficacy of this monitoring approach. In particular, we use a
global sensitivity method to identify the most effective in situ data for estimating contami-
nant concentrations in a long period of time, and also to quantify the effect of climate change
and hydrological shifts on the long-term monitoring and plume mobility. We demonstrated
our approach at the uranium-contaminated Savannah River Site F-Area.

138
Corresponding author: Jeremy White
U.S. Geological Survey
[email protected]

The Ensemble Smoother: Jacobians for nothing and uncertainty for free

Jeremy T. White, U.S. Geological Survey

Abstract: Many groundwater modeling analyses use some form of automated parameter
estimation, where the Marquardt-Levenburg algorithm (MLA) has been shown to work well
on a wide range of problems. However, implementation of the non-linear MLA requires re-
peated calculation of a Jacobian (sensitivity) matrix. Typically, the Jacobian matrix is filled
by a computationally-expensive perturbation-based derivative calculation process, where the
model is run once for each adjustable parameter. Recently, ensemble-based approximations
to the Jacobian have been proposed in the petroleum industry, where derivatives are approx-
imated by the ensemble spread about the mean for observations and parameters, requiring
only one model for each ensemble member. The ensemble approximation yields an algo-
rithm, the Ensemble Smoother (ES), which is closely related to the Ensemble Kalman filter
used widely in numerical weather prediction. The ES formulation frees practitioners from
computation constraints of filling the Jacobian matrix by repetitive parameter perturba-
tion processes, rendering zonation-based and pilot-point-based parameterization methods
obsolete. The ES formulation instead allows for direct use of node-scale parameterizations,
such as those based on geostatistically-generated fields, where adjustable parameters can
number in the millions. Furthermore, non-linear parameter uncertainty quantification is
an outcome of the ES because the result of the process is an ensemble of parameter sets
that reproduce the observation data acceptably well. A prototype ensemble smoother has
been implemented in the pyEMU module and a full-scale, performant implementation in
the PEST++ framework is under development. Some initial results will be presented to
demonstrate the concepts and workflow of applying the ensemble smoother to groundwater
models.

139
Corresponding author: Jeremy White.
U.S. Geological Survey
[email protected]

PESTPP-OPT: model-independent groundwater-management optimization

under uncertainty

Jeremy T. White, U.S. Geological Survey

Abstract: Environmental resources can be effectively managed by coupling simulation

models and optimization-under-uncertainty methods to shed new insights on uncertainty
and risk given what is known about the model parameters. Here we present an optimization
tool (PESTPP-OPT) for the management of environmental resources within the familiar
PEST(++) model-independent framework. PESTPP-OPT uses the proven and efficient
simplex algorithm to repeatedly solve the sequential linear programming problem. The Yet
Another Run Manager (YAMR) parallel run manager is used to distribute the calculation of
the response matrix for repeated linearization. We use the integrated PEST++ first-order,
second-moment (FOSM) linear-based uncertainty-estimation capabilities to estimate vari-
ance in model-based constraints arising from incomplete knowledge of model parameters.
For hydrologic models, examples of model outputs for which there might be incomplete
knowledge include water levels, streamflow, streamflow depletion, and spring discharge. Us-
ing an optional user-specified ”risk” parameter, PESTPP-OPT can use a form of stochastic
programming known as chance-constrained programming, where model-simulated constraint
values are ”shifted” using the estimated constraint variance towards a risk-averse or risk-
tolerant optimal solution. Incorporating uncertainty into the optimization process yields
a single optimal solution that includes model parameter uncertainty. Furthermore, using
data-worth concepts, users can easily estimate the value, in objective function units, of new
and existing observations to reduce constraint uncertainty and ultimately allow for a more
favorable optimal solution.

140
Corresponding author: William Wingle
AquaGeo Ltd.
[email protected]

Automated Transient Element Activation Plug-In for FEFLOW

William L. Wingle, AquaGeo Ltd.
Peter Sinton, AquaGeo Ltd.

Abstract: The ability to efficiently simulate the transient expansion of large engineered
features is now possible using FEFLOW 7. Such engineered features include mine exca-
vations (mine pits, underground workings, etc.) and tailings impoundments. Simulation
of these features can result in numerical instability if the changes in element state result
in large variations in saturation. While FEFLOW 7 has interactive tools that enable the
modeler to specify changes in element state over time, the difficulty of set up and edit of
such changes rises dramatically in cases where many changes occur over short periods of
time. Therefore, AquaGeo has developed a plug-in for FEFLOW 7 that automates the
gradual activation or deactivation of elements in a three-dimensional model over time. This
plug-in, called Scheduled Element State (SES), allows the modeler to efficiently specify and
edit numerous and complex changes in element state. The SES plug-in also records changes
in water mass associated with changes in element state, which is otherwise difficult to ob-
tain using FEFLOW’s interactive tools. Used in conjunction with AquaGeo’s plug-in that
automates transient changes in specified-head boundary conditions, the transient expansion
of mine pits and tailings impoundments have been successfully modeled. The ability to ef-
ficiently simulate the transient expansion of large engineered features is now possible using
FEFLOW 7. Such engineered features include mine excavations (mine pits, underground
workings, etc.) and tailings impoundments. Simulation of these features can result in nu-
merical instability if the changes in element state result in large variations in saturation.
While FEFLOW 7 has interactive tools that enable the modeler to specify changes in element
state over time, the difficulty of set up and edit of such changes rises dramatically in cases
where many changes occur over short periods of time. Therefore, AquaGeo has developed a
plug-in for FEFLOW 7 that automates the gradual activation or deactivation of elements in
a three-dimensional model over time. This plug-in, called Scheduled Element State (SES),
allows the modeler to efficiently specify and edit numerous and complex changes in element
state. The SES plug-in also records changes in water mass associated with changes in ele-
ment state, which is otherwise difficult to obtain using FEFLOW’s interactive tools. Used
in conjunction with AquaGeo’s plug-in that automates transient changes in specified-head
boundary conditions, the transient expansion of mine pits and tailings impoundments have
been successfully modeled.

Extended abstract: Methodology, findings, and implications

PLUG-IN CONCEPTS

Instantaneous switching of the active state of an element can affect model setup and the
numerical stability of a model. Very small time steps or numerous switching-time arrays to
limit the number of elements activated in any one time step may be required to maintain
reasonable stability and mass balance. At the beginning of a time step FEFLOW will
instantly add or subtract the material properties of the specified elements and also some or

141
all of the calculation points (nodes) connected to those elements. Depending on simulation
mode, newly activated nodes will have a water level assigned by FEFLOW that may not
be ideal which the modeler must consider in configuring a simulation. In simulations of
mass and heat transport, the solute and thermal energy, and associated properties are also
involved, and FEFLOW 7.0 does not account for the hydraulic effects of consolidation or
transient changes in material properties (e.g. transient reduction in hydraulic conductivity
during the consolidation process).
If the modeler uses just FEFLOW’s interactive tools, some situations cannot be efficiently
modeled and some simulation data will not be accessible. For example, when activating
elements, FEFLOW may assign a water level to its nodes that are not compatible with the
situation being modeled. Also, pore-water mass either removed or added during a simulation
is not tracked by FEFLOW.
Because SES gives the user access to FEFLOW’s intrinsic functions, SES expands the
type of situations that can be simulated and provides detailed information needed to assess
simulation results. For example, SES allows the user to specify the water level that will
be assigned to newly activated nodes, and, in so doing, the saturation of the associated
elements. SES also monitors and records pore-water mass added or removed due to changes
in element state. SES activates or deactivates elements based on user-specified temporal and
elevational information. For an excavation through 10, 100-foot thick model layers over 10
years, the rate of element deactivation would be one layer of elements per year. The modeler
can specify multiple SES groups to create complex schedules with minimal input.

CASE HISTORY

At a mine site, transient buildup of a tailings impoundment was simulated to evaluate the
impoundment’s hydraulic effects on the underlying hydrogeologic system. Figure 1 shows the
dam and impounded tailings when the facility reaches maximum capacity. At the beginning
of the simulation, all elements above the underlying (pre-existing) materials were inactive.
As the tailings dam was raised and the impoundment was filled, elements were activated
according to a specified SES schedule. Figure 2 shows the rate of activation and volume of
water added over time, as computed by SES. In this example, the tailings were assumed to
be saturated at the time of deposition. Therefore, the mass of water added as elements of
tailings were activated grows over time.

SUMMARY

Element activation and deactivation is useful for simulating the effects of many engineered
features, for example:

• Excavation of mine pits (as the pit deepens, elements are progressively deactivated),

• Backfilling of excavations (as backfill is added, elements are activated), or

• Expansion of impoundments or dams (as an impoundment grows, elements are acti-

vated).

142
 The SES plug-in allows a modeler to efficiently simulate such situations, some of which
cannot be set up using FEFLOW’s interactive tools alone. Future work on SES will allow
more complex methods of activation and more integration with other AquaGeo tools.

Figure 1. Dam and impounded tailings when the facility reaches maximum capacity.

Figure 2. Rate of activation and volume of water added over time, as computed by SES.

References

143
DHI, 2016, FEFLOW User’s Manual.

144
Corresponding author: William Wingle
AquaGeo Ltd.
[email protected]

ZoneTable Plug-In for FEFLOW

William L. Wingle, AquaGeo Ltd.

Peter Sinton, AquaGeo Ltd.

Abstract: Entering, defining, and updating material properties across a layered FEFLOW
(DHI, 2016) model can be challenging even with FEFLOW’s intrinsic tools. The ZoneTable
plug-in ties a table of named material properties (zone names) to a user-defined elemental
array of integer identifiers that specifies the 3D distribution of named zones across the
model. Additionally, the 3D distribution of material zones can be defined by the modeler
using two-dimensional surfaces that define hydrogeological contacts. The contacts are not
required to correspond to the computational layering because the plug-in computes hybrid
property values according to user-specified settings. For visualization of property zonation,
the most dominate material is shown. Areas that require focused edits can co-exist with the
zonation and contact-surface approaches.

Extended abstract: Methodology, findings, and implications

PLUG-IN CONCEPTS

Elements that share the same material properties can be thought of as a hydrogeologic
unit, referred to as a zone. Although the intrinsic FEFLOW editing tools are flexible, errors
can evolve into problems that consume project time. For example, element selections can
overlap or may exclude elements erroneously, resulting in unexpected property values and
possibly poor model results. In addition, FEFLOW also has no built-in ability to manage
zones that do not conform to the model mesh.
The core of ZoneTable is a listing of named material types (zones) with corresponding
hydraulic conductivities and storage properties. For every element on which ZoneTable
operates, this table is used to “look up” the modeled material properties and assign them to
the element. The lookup table is then tied to specific 3D distributions of elements to allow
property assignment. ZoneTable currently allows assignment of hydraulic conductivity (Kx,
Ky, and Kz), specific storage (compressibility, Ss) and specific yield (drain / fillable porosity
or unsaturated-flow porosity, Sy).
ZoneTable currently has two methods, which can be used separately or together, to define
the distribution of materials. Method 1 employs a user-defined element array of integers to
specify the zone to which every element is assigned. Every element is assigned to one and
only one zone, and the modeler uses FEFLOW’s intrinsic editing tools to manipulate this
elemental array.
Method 2 employs 2D surfaces that define contacts between zones. The elevation of the
surface must be defined at the horizontal location of each node on Slice 1. The 2D surfaces
must intersect the 3D FEFLOW mesh to have any effect and can extend below or above the

145
computational mesh. Multiple surfaces can be used to create complex geometries. Pinch-
out of zones can be modeled, but overlapping surfaces will be trimmed. Hybrid properties
are assigned to elements that are bisected by one or more surfaces using volume-weighted
averaging approaches. The method of averaging is defined by the modeler.
To use both assignment methods, the modeler can exclude specified zones from the
operations of method 2, and any element or groups of elements can also be explicitly excluded
from ZoneTable’s calculations. However, with flexibility comes responsibility: ZoneTable will
not capture all possible errors in model input.

MATHEMATICS

Under Method 1, assignment of material properties is straight forward. For every ele-
ment in the model, if ZoneTable can find the element’s ID in the lookup table, the plug-in
will simply assign the user-defined material properties. If the element’s ID is not found,
ZoneTable does not alter the material properties of the element.
Under Method 2, assignment of material properties is more complicated. Every element
in the model is evaluated, and, using the 2D contact surface information supplied by the user,
ZoneTable determines to which zone or zones an element belongs. If the element belongs
to more than one zone, ZoneTable computes hybrid material properties for the element as
follows:

• A simple weighted arithmetic mean of the user-defined zonal material values is assigned
to the element for Kx , Ky , Ss and Sy . For example, the hybrid hydraulic conductivity of an
element in the x-direction, Kx is:
𝑖=𝑛
vi
𝐾x = ∑ 𝐾
𝑖=1
vt x,i
where:
Kx,i = Kx of the i th ZoneTable zone in the element,
n = number of ZoneTable zones in the element,
v t = volume of the element, and
v i = volume of i th ZoneTable zone in the element.

• For Kz, ZoneTable computes a geometric-weighted mean value for material properties:
𝑖=𝑛
vi
𝐾 z = 𝑒𝑥𝑝[ ∑ log (𝐾 z,i )]
𝑖=1
vt
In practice, the modeler edits ZoneTableID, defines material properties for zones in
Zonetable’s lookup table, imports 2D contacts, etc. Then ZoneTable computes and fills
FEFLOW’s Kx , Ky , Kz , Ss and Sy arrays when the modeler launches a simulation or ex-
its ZoneTable’s interface by clicking “OK”. ZoneTable first executes Method 1 and then
Method 2. ZoneTable does not alter the material property of an element that the modeler
has completely excluded from ZoneTable’s operations.

146
 Data required by ZoneTable is stored in FEFLOW’s FEM file, and can easily be imported,
exported and stored in simple “comma-space-variable” ASCII files. Such external files are
easy to edit using spreadsheet software, text editors, contouring software, etc.

CASE HISTORY

A geologic model was developed using topographic and drill-hole data. The geologic
model, developed in FEFLOW, is shown in the Figure 1, and consists of 6 geologic units:
four sedimentary units and two crystalline bedrock units. The geologic units were assigned
integer values of hydraulic conductivity.
Using the interpreted two-dimensional surface data for contacts between the units, ZoneTable
was used to translate the geologic model into the computational mesh depicted in Figure 2.

SUMMARY

ZoneTable provides FEFLOW users a new and efficient approach to model complex
hydrogeologic settings. The modeler can easily view and edit material properties of elements
that belong to a particular hydrogeolgic unit, and apply those changes all at once rather
than by tedious step-wise application. The ZoneTable approach is therefore less prone to
errors. Complex hydrogeologic settings, including faults, tunnels, and discontinuous lenses
of material can be efficiently modeled.
Future work on the plug-in includes the addition of recharge and all other element-based
model properties, allow transient materials, and the ability to work with partially or fully
“unstructured” finite-element meshes.

Figure 1. Geologic model developed in FEFLOW consisting of 6 geologic units: four sedimentary units and
two crystalline bedrock units.

147
 Figure 2. Computational mesh translated from the two-dimensional surface data using ZoneTable.

References

DHI, 2016, FEFLOW User’s Manual.

148
Corresponding author: Willem Zaadnoordijk
Geological Survey of the Netherlands / Delft University of Technology

Improving parametrization of top layer for groundwater models by means of a

3D geologic voxel model

Willem J. Zaadnoordijk, Geological Survey of the Netherlands / Delft University of Technology

J. Gunnink

Abstract: The Geological Survey of the Netherlands makes available 3D models of the
subsurface at various resolutions. GeoTOP is a high resolution voxel model of the upper
tens of meters of the subsurface. It consists of voxels of 100m x 100m x 0.5m for which
the stratigraphy, lithology, and hydrogeological parameters are identified with their uncer-
tainties. The first step in the creation of GeoTOP is a geological schematization of the
borehole descriptions into lithostratigraphical units. During the second step, maps are cre-
ated of the top and base of each unit based on the elevation in the boreholes. The third
step is a 3D stochastic interpolation of lithofacies and lithology within each unit, result-
ing in probabilities for each voxel. In the fourth step, voxels get hydraulic conductivities
from stochastic upscaling of values from laboratory tests on undisturbed samples from high
quality cores from boreholes. Hydraulic parametrization of GeoTOP has been carried out
for the Province of Zeeland in the Netherlands. The resulting hydraulic resistance of the
Holocene confining layer compared reasonably well to existing information. However, no
data was available for an area with very high resistances according to GeoTOP. Therefor, a
pumping test has been designed to obtain data for the validation of these high resistances.
The integration of the geology in GeoTOP gives better a priori hydraulic parametrization
of (complex) confining layers for groundwater modeling. Still unresolved is how the 3D
voxel model can be improved with the a posteriori values of groundwater model calibration.
Corresponding email: [email protected]

149
Corresponding author: Peter Andersen
Tetra Tech
[email protected]

The Evolution of a Groundwater Flow and Transport Model Over Two Decades
of Updates and Applications

Peter Andersen, Tetra Tech

James Ross, Tetra Tech

Abstract: Groundwater models developed for specific sites generally become obsolete
within a few years due to changes in: 1) modeling technology, 2) site/project personnel,
3) project funding, and 4) modeling objectives. Consequently, new models are developed
using the latest technology and data, but without potential knowledge gained from the prior
models. As a contrary situation, we present the unique case of a numerical flow and TCE
transport model that was developed in 1993 and updated annually. Updates are prompted
by an increase in the amount of data, exposure to a wider range of hydrologic conditions
over increasingly longer timeframes, technological advances, and revised modeling method-
ologies. The history of updates shows smooth, incremental changes in the conceptual model
and modeled aquifer parameters that result from both increases and decreases in complex-
ity. The continued success of this application emphasizes an original tenant of groundwater
modeling: iterative adjustment of the conceptual model based on observations of actual
versus model response.

Extended abstract: Methodology, findings, and implications

Introduction

Advances in modeling technology and administrative/funding considerations often drive

the creation of new models in favor of resurrecting or updating previously created models.
This practice may result in failure to glean lessons from the prior models that would have
advanced the conceptual model. A unique contrary case is presented wherein a numerical
model has been annually updated with new data and conceptual understanding since its
inception over 20 years ago. Many positive results have been obtained from adopting this
evolutionary approach.
The initial development of the Tooele Army Depot-North (TEAD-N) groundwater flow
model took place in 1993 with the primary objective of providing a tool for determining sci-
entifically based optimum pumping rates and locations, to ensure the hydrodynamic isolation
of a large TCE plume. A secondary objective was to ably reflect hydrogeologic responses of
future stresses placed upon the site under various scenarios (HEC, 1993).

Updates

In addition to a regulatory (Utah Department of Environmental Quality) requirement

to update the model annually, updates to the model have resulted from addition of data,
changes to the conceptual model, and improvements in technology. The model has been used

150
to identify locations for additional monitoring wells, which have increased from 45 to 133
since 1993. The progression of time from 1993 to 2016 has increased the temporal domain
of the data from 45 to 14949 measurements of groundwater levels and TCE concentrations,
and, consequently, the understanding of temporal response of the hydrogeologic system. This
new data has necessitated alterations to the conceptual model based on an improved under-
standing of temporal responses (water level changes and plume movement) and resulted in
an expansion of the model domain (spatially and vertically) and the improved representation
of hydrogeologic features (faults, surface water interactions). Technological advances include
faster computers and more sophisticated analysis techniques, such as automated parameter
estimation, the ensemble Kalman filter, and uncertainty analysis. Faster computers have
allowed an increase in model size (12,840 to 246,240 cells) and consideration of transience
and solute transport, and the use of more robust calibration techniques. Overall, the number
of computations for the base case flow model run has increased from 12,840 to 536,803,200
when all factors are considered.

Conceptual Model Changes

Table 1 shows changes to the modeled hydraulic parameters as a function of time (5 year
intervals). While the model has grown in terms of size and resolution, the number of zones of
hydraulic conductivity has decreased from 21 to 11. The range in horizontal hydraulic con-
ductivity has also generally decreased (excluding fault gouge, which has been a model feature
since 2002). These decreases may be indicative of an improved representation of the sys-
tem, requiring less variability to adequately describe it. The range in hydraulic conductivity
anisotropy and porosity has steadily increased and may correspond to the inclusion of solute
transport in the analysis. Table 2 shows changes to modeled solute transport parameters as a
function of time. As expected, the table reveals that the understanding and characterization
of solute transport lags that of groundwater flow. In general, the most significant changes
to both the flow and transport models appear to correlate to inception of consideration of
transience, solute transport, and the use of parameter estimation techniques.

Model Predictions

The model has served myriad roles, such as: 1) supporting pump-treat-reinjection system
construction, 2) providing basis for computations in a USEPA demonstration to evaluate po-
tential cost savings at pump and treat sites (Greenwald, 1999), 3) serving as a subject of
USEPA and DoD (Becker et al., 2006) transport model optimization study, 4) rationaliz-
ing a planned 3-year shutdown of the extraction/injection system, 5) supporting continued
shutdown of the extraction/injection system, and 6) supporting a Corrective Measures Study
that recommended MNA for the distal parts of the plumes. As a part of the MNA remedy,
the model forecasts the probability of exceedance of regulatory standards at the property
boundary.

Conclusions

This application represents a unique case where the same groundwater model has been
applied and updated by the same team of modelers over a 20-year period. Every year of
the model’s history has brought advances in model sophistication, quality, and capability,

151
without a need to “reinvent the wheel.” Conceptual understanding of the model has con-
sistently increased, while the complexity required has generally decreased. Accordingly, the
model has been useful in a wide array of objectives. The primary advantage of this approach
is that there is a logical progression and understanding of the conceptual model. Contin-
ual improvement of the conceptual model is critical when it is considered that results of
post-audits reveal that most errors in predictions are attributable to errors in the conceptual
model (Anderson and Woessner, 1992).

Table 1. Summary of groundwater flow model features at 5 year increments.

152
 Table 2. Summary of solute transport model features at 5 year increments

References

Anderson, M. P. and Woessner, W. W. (1992). Applied groundwater modeling: Simulation of flow and
advective transport, Elsevier, San Diego, CA.
Becker, D., Minsker, B., Greenwald, R., Zhang, Y., Harre, K., Yager, K., Zheng, C., and Peralta, R. (2006)
Reducing long-term remedial costs by transport modeling optimization. Ground Water, 44(6), 864–875.
Greenwald, R. (1999). Hydraulic optimization demonstration for groundwater pump-and-treat systems, HSI
GeoTrans, Inc., Prepared for Dynamac Corporation and U.S. Environmental Protection Agency.
HEC (1993) Hydrogeologic Flow Model for Pump-and-Treat System at Tooele Army Depot, U.S Army Corps
of Engineers- Hydrologic Engineering Center Draft Project Report, July 1993.

153
Corresponding aut hor: S.M. Helalur Rashid
INT ERA Incorporat ed
hrashid@int era.com

I mport ance of Charact erizing R iver-aquifer I nt eract ion in Building

G roundwat er Flow and Transport M odel

S.M. Helalur Rashid, INT ERA Incorporat ed

J. Ewing
T . Jones
G. Ruskauff

A bst ract : At t he former B and C react or sit es along t he Columbia River on t he Hanford Sit e
accurat ely represent ing t he river-aquifer int eract ion in t he groundwat er flow and t ransport
model is key t o a credible numerical model. Columbia River st age is cont rolled by t he dams
and reservoir upst ream t o t he sit e, and hydraulic head at select ed inland wells were used t o
ident ify t he response t o aquifer for t he t ransient river st age. T he flow model was calibrat ed
wit h PEST using t he aut omat ed wat er level network (AWLN) dat a, hydraulic conduct ivity
dat a from slug/ pump t est s, and groundwat er velocity measurement s. A flow and t ransport
model was const ruct ed t o: 1) simulat e groundwat er head, hydraulic gradient , and flows
t o t he Columbia River for use in general flow syst em underst anding as well as pot ent ial
remediat ion syst em analysis, 2) est imat e fut ure groundwat er concent rat ions of hexavalent
chromium and st ront ium-90 t o support risk screening and evaluat ing remediat ion opt ions,
3) est imat e cont aminant discharge t o t he Columbia River due groundwat er upwelling in t he
riverbed, and 4) est imat e pot ent ial concent rat ions for ext ract ed groundwat er. Confidence
in t he result s was built by not ing t hat t he aquifer property dist ribut ion inferred from cal-
ibrat ion agree wit h t he locat ion of an inferred paleochannel, and t hat t ransport modeling
agreed wit h areas of observed groundwat er upwelling in t he riverbed.

154
Corresponding author: Alexis Navarre-Sitchler
Colorado School of Mines
[email protected]

Physical heterogeneity controls on geochemical reaction rates: Insights from

reactive transport models

Alexis Navarre-Sitchler, Colorado School of Mines

H. Jung, Colorado School of Mines
J. Beisman, Colorado School of Mines
R. Maxwell, Colorado School of Mines

Abstract: Physical heterogeneity in geologic systems induces spatially variable fluid and
solute transport rates, which can lead to spatially variable geochemical behavior. Our un-
derstanding of the coupling of physical heterogeneity and geochemical behavior is limited,
however. Reactive transport models allow for exploration and elucidation of the coupling
of physical heterogeneity and geochemical reactions and reaction rates. We present three
examples of using reactive transport models to quantify changes in geochemical reaction
rates as a function of physical heterogeneity: mineral dissolution and precipitation at the
column scale and biologic nitrogen reduction at the floodplain scale. In one case we ex-
plore the impact of numerical methods used for transport in a reactive transport model on
geochemical reaction rates and their spatial distributions. This research demonstrates that
heterogeneity in physical structure can induce spatially variable distributions of geochemi-
cal reaction rates and impart orders of magnitude changes in rates of geochemical processes
compared to often-used, simplified, homogeneous representations. More work is still needed
to map out and quantify any consistent correlations between physical heterogeneity and
geochemical reaction rates that could be used to inform data collection and simulation of
natural systems. This work also suggests that physical heterogeneity contributes to some of
the long-observed differences between laboratory measured and field calculated geochemical
reaction rates.

155
Corresponding author: Adriana Palma Nava
Instituto de Ingenieria, Universidad Nacional Autonoma de Mexico
[email protected]

Mexico Valley Aquifer’s Transition from a University Simulation Model to

MODFLOW for Water Resource Assessment and Subsidence Modeling

Adriana Palma Nava, Instituto de Ingenieria, Universidad Nacional Autonoma de Mexico

F. Gonzalez Villarreal
R.T. Hanson
S.E. Boyce

Abstract: The aquifer system of the Mexico Valley, located at Mexico City, Mexico, is
an important source of water for agriculture land and domestic consumption. The valley
is contains large, areal extents of highly compressible clay materials that originated from
the original system of lakes that covered the area. The Mexico Valley main water supply
originates from a deep aquifer that is connected to a shallow, phreatic aquifer through the
lake-deposited clay layer (aquitard). Originally the water table of the aquifer system in-
tercepted the ground surface resulting spring flows that supplied agriculture and domestic
supplies. Over the past 100 years the spring flows have diminished due to groundwater
extraction. This resulted in a positive feedback loop where groundwater extraction would
reduce spring flows necessitating more groundwater extraction. The secondary effect of
groundwater extraction from the Mexico Valley is the compaction of the clay materials
resulting in significant subsidence. The compaction of the clay materials resulted in land
surface elevation changes up to 15 meters. This has resulted in significant damage to in-
frastructure, including the water conveyance and sewer system. The Universidad Nacional
Autonoma de Mexico (UNAM) developed a specialized subsidence and groundwater flow
software tailored to the Mexico Valley. The software served as the simulation engine for
understanding the relationship between pumpage with reduced spring flow and incurred
subsidence. The software also assisted in the understanding ways to increase potable water
supplies with the focus being on using agricultural water and reducing system transit losses,
which are estimated at 34ύ. The UNAM custom software integrates the groundwater flow
equation using a finite volume mesh in space and an implicit finite difference scheme in time.
The interaction between aquifer system layers and aquitard is computed with the integration
of the vertical flow differential equation. The solution of the flow through aquitard is then
used to infer its compaction and subsidence. UNAM is now working with the United States
Geological Survey (USGS) to develop a MODFLOW-OWHM (One-Water) model of the
Mexico Valley and comparing its results to the original UNAM custom software. This inter-
national cooperation is brought about by the need to address a broader spectrum of urban
and domestic future planning and developing a sustainable groundwater management plan.
This presentation will discuss the transition to One-Water, the comparison between the two
simulation codes, and the interpretation of the new, additional information generated by
switching to One-Water.

156
Corresponding author: George Roadcap
IL State Water Survey - U of Illinois
[email protected]

One Model to Rule them All: The Illinois Groundwater Model

George S. Roadcap, IL State Water Survey - U of Illinois

D. Abrams
D. Hadley
D. Mannix

Abstract: Modelers at the Illinois State Water Survey have created many different models
for different aquifers and purposes over the last 50 years. As our understanding of the
interconnections between the different aquifers has advanced through time, many of these
models have unfortunately gone out of date. To better address the growing number of
water supply issues that impact multiple aquifers and regions, the existing models needed
to be combined into one overarching MODFLOW model with a common grid system and
input database. The new model is essentially merging the details from several finer-scale
grids of the glacial aquifers and a coarser-scale grid of the deep sandstone aquifers that
covers the northern half of the state and extends into Wisconsin, Iowa, and Indiana. The
database for the model keeps all the details, such as geologic surfaces, at a fine resolution
so that new analyses of can be run with telescopic mesh refinement (TMR) or unstructured
grids. Calibration checks were necessary after adjusting the grids and layers of the finer
models, such as those for the Mahomet Aquifer and St Charles Aquifer, to match those
of the larger model. Tools have been developed to efficiently update both regional and
local scale models when necessary, as well as to check any changes in calibration that may
have taken place. An additional advantage of the one model approach is eliminating the
problem of changing personnel or models not functioning in the future because of a software
or computer upgrades.

157
Corresponding author: Noel Merrick
[email protected]

Does MODFLOW-USG/AlgoMesh Give the Same Results as

MODFLOW-SURFACT?

Noel Merrick, HydroAlgorithmics

Damian Merrick, HydroAlgorithmics

Abstract: AlgoMesh is a tool that generates meshes with triangular or Voronoi (polyg-
onal) cells. It writes MODFLOW-USG input files and allows import of binary results for
subsequent export to other post-processing software. For MODFLOW-USG/AlgoMesh to
be adopted with confidence, the veracity of the code and the equivalence of model outputs
with more traditional modeling tools require demonstration. Two synthetic models have
been designed to allow comparative testing against MODFLOW-SURFACT.
The first model is a 7-layer model with uniform properties except for a corridor of alluvium
which hosts a river, with coal seams at 100 m and 170 m depth mined by two open cut
pits. The second consists of 11 flat layers with a similar river and alluvium corridor to the
first, and a coal seam 450 m below ground that is mined by two longwall panels. Temporal
changes for spoil and fracture zone properties are modeled with the Time-Varying Material
Property (TMP1) package and the Time-Variant Materials (TVM) package respectively in
MODFLOW-SURFACT and MODFLOW-USG.
Comparative results between MODFLOW-SURFACT and MODFLOW-USG are presented
for mine inflows, river baseflow and groundwater hydrographs.

Extended abstract: Methodology, findings, and implications

This paper presents a validation of MODFLOW-USG (USG) against MODFLOW-SURFACT

(MS) for identical structured grids and for an unstructured grid using AlgoMesh software.
For MODFLOW-USG/AlgoMesh to be adopted with confidence, the veracity of the code
and the equivalence of model outputs with more traditional modeling tools require demon-
stration.
Separate synthetic models have been set up for open cut (OC) mining, using a struc-
tured grid, and longwall (LW) mining, using both structured and unstructured grids. The
conceptual models are illustrated in Figure 1. The model extent is approximately 40 km
east-west by 40 km north-south. The structured grid has a maximum cell size of 200 m and
a minimum cell size of 50 m, with total cell counts of 280,000 (OC) and 440,000 (LW). The
unstructured grid uses a Voronoi mesh with cells ranging in size from 25 m to 500 m, with a
total cell count of about 410,000 (LW). At ground surface, the model has a uniform regolith
interrupted by a corridor of alluvium that hosts a linear river. The river occupies the central
50 m portion of the alluvial corridor of 450 m width. The initial water table has a change
in elevation of 10 m from west to east and the river stage drops 5 m from north to south.
All layers are horizontal and have laterally uniform transmissive and storage properties, al-
though they vary from layer to layer, alternating vertically between consolidated aquifer and
aquitard layers at depth. In each model, mining is simulated as occurring for a period of
eight years after a pre-mining period of one year, with post-mining recovery for 50 years.
Hydrographs are examined in each layer at seven sites (P1-P7) along a central transect.

158
 The OC model of seven layers has coal seams in layers 4 and 6. The coal seams lie at
depths of 100 m and 170 m below ground and are each 20 m thick. They are to be mined
in two parallel strips which are each 200 m wide and 2,000 m long. An earth barrier of 50
m width separates the two pits. Spoil emplacement is simulated only during the recovery
period, with no final void.
The LW model of 11 layers has a coal seam in layer 9. The coal seam lies 450 m below
ground and is 3 m thick. It is to be mined by two longwall panels which are each 200 m wide
and 2,000 m long. A coal pillar of 50 m width separates the two longwall panels. Fracturing
occurs concurrently with excavation using a log-linear increase in hydraulic conductivity
and specific yield. The fractured zone (for connective vertical fracturing) is 200 m above
the roof of the mined coal seam and spreads across layers 5 to 7 with a caved zone in layer
8 immediately above the seam (layer 9) (Figure 1). The fracture height is calculated as
the 95th percentile of the A-height of the Ditton geology model as defined by Ditton and
Merrick (2014). Enhanced vertical hydraulic conductivity (Kz) is applied progressively above
the mined longwall panels to create the fractured zone (layers 5-7) and the caved zone (layers
8-9). Enhanced horizontal hydraulic conductivity (Kx) is applied to the same layers but also
in the constrained zone (layer 4) and a relaxed zone (layer 10) below the mined coal seam.
Specific yield (Sy) is also increased but only in the created caved zone and mine void (layers
8-9) within and immediately above the mined longwall panels (Figure 1).
For cell desaturation simulation, MS uses the pseudo-soil option with upstream weight-
ing and USG uses equivalent upstream weighting. In addition, the USG model uses the
CONSTANTCV option for the Layer Property Flow (LPF) package. Different VC options
have been tested: VC1 (CONSTANTCV NOVFC), VC2 (default), VC3 (CONSTANTCV),
VC4 (NOVFC) and VC5 (NOVCCORRECTION).
For the base OC case, USG and MS gave essentially the same mine inflow and river
baseflow (within 0.01 percent). Groundwater hydrographs were identical during mining,
though MS water levels recovered slightly faster in layer 6. The MS simulation ran 4 times
faster than USG (XMD) and 2.5 times faster than USG (PCGU).
When the USG model was run with different VC options, some difference in mine inflow
was observed (by 3 percent). Options VC1 and VC3 agreed perfectly. Options VC2, VC4
and VC5 agreed perfectly with each other. No significant difference in baseflow was observed.
During mining, slight differences in hydrographs were observed at bores P2-P7 between the
two VC groups. Larger difference was observed in hydrographs at P1 for layers 2-3 (maximum
15-60 m). VC1 and VC3 gave earlier recovery of deep water levels.
For the base LW case, MS gave higher mine inflow (by 10 percent) but both MS and
USG gave essentially the same baseflow. Hydrographs agreed perfectly during mining and
during recovery. The MS simulation ran 2 times faster than USG (XMD) and 4 times faster
than USG (PCGU).
When USG used an unstructured Voronoi grid, a rainfall recharge difference of 0.3 per-
cent resulted from differing shapes of CHD edge cells. MS gave lower mine inflow (by 4
percent) but both MS and USG gave essentially the same baseflow (0.2 percent difference).
Hydrographs differed often by up to 10 percent (Figure 2). Spatial head patterns agreed
very well.
It is concluded that the MODFLOW-USG code provides a valid simulation of the effects
of mining operations on a groundwater system. The use of mesh generation software (Algo-

159
Mesh) for finer spatial resolution and generation of corresponding USG input files has led to
essentially equivalent results to those that are traditionally obtained through application of
MODFLOW-SURFACT groundwater modelling software.

Conceptual Models.JPG Conceptual Models.bb

Figure 1. Synthetic conceptual models

Hydrographs.JPG Hydrographs.bb

Figure 2. Hydrographs for longwall mining for MS (structured grid) and USG (Voronoi grid)

References

Ditton, S. and Merrick, N., 2014. A New Subsurface Fracture Height Prediction Model for Longwall Mines
in the NSW Coalfields. Australian Earth Sciences Convention (AESC) 2014. Newcastle, NSW Australia.

160
Corresponding author: David Moulton
Los Alamos National Laboratory
[email protected]

Integrated Hydrology Model Intercomparison using High-Resolution Data at

the East River Colorado
David Moulton, Los Alamos National Laboratory
Erica Siirila-Woodburn, Lawrence Berkeley National Laboratory
Dipankar Dwivedi, Lawrence Berkeley National Laboratory
Carl Steefel, Lawrence Berkeley National Laboratory
Daniil Svyatskiy, Los Alamos National Laboratory
Reed Maxwell, Colorado School of Mines
Laura Condon, Syracuse University

Abstract: Growing interest in understanding and predicting climate impacts and feedbacks
on terrestrial ecosystems is driving an increased interest in models that represent and couple
all relevant processes. But even as field and laboratory campaigns rise to the challenge of
collecting diverse and heterogeneous datasets that can inform and constrain these models,
questions remain about the model’s process representation and implementation. To address
these questions, the community is increasingly turning to model intercomparision studies
where multiple codes solve the same problem to expose and understand differences in re-
sults. Recent model intercomparisions in integrated hydrology (e.g., Kollet et al., 2017),
have made significant progress but have utilized simple geometries and small domains. In
this work, we take advantage of the growing set of data being collected in the East River
watershed, Colorado as part of Lawrence Berkeley National Laboratory’s Scientific Focus
Area. Specifically, we highlight our progress in developing a model intercomparison that
uses high-resolution (0.5m) Lidar data and bathymetry data along a meandering section
of the East River. We compare two open-source integrated hydrology codes that partici-
pated in earlier model intercomparison studies, ParFlow and Amanzi/ATS. Although, both
codes solve variably saturated subsurface flow using Richards equation coupled to the diffu-
sive wave model of surface water flow, they use different assumptions about the underlying
mesh, discretizations, and time evolution strategies. Here we focus on integrated hydrology
to explore these differences, and describe future plans to explore the impact of differing land
model representations and coupling to reactive transport models.

161
Corresponding author: Christopher Muffels
S.S. Papadopulos and Associates
[email protected]

mod-PATH3DU Version 2: New Features, Verification and Intercomparisons

Christopher Muffels, S.S. Papadopulos and Associates

Abstract: It has been over 2 years since the original release of mod-PATH3DU. It was the
first particle tracking program released for the unstructured grid version of MODFLOW. It
is still the only one available that supports all of the cell geometries available to MODFLOW-
USG. This release was built upon the MODPATH 6 framework to facilitate development.
Recently, version 2 of mod-PATH3DU was released. It is a complete rewrite with a focus on
input and runtime improvement. The input file format is new; it follows JSON, a commonly
used data-interchange format that is human readable. This new release is multithreaded
- when processing the MODFLOW binary files and particle tracking, making this version
substantially faster than its predecessor. New functionality has also been introduced, in-
cluding a Random-Walk module to simulate the effects of dispersion. Additionally, the
program can be used to track particles on an arbitrary surface, specifically a head surface
generated using a kriging application. This presentation highlights the improvements to
version 2, demonstrates the new functionality with examples and compares the results with
its predecessor and MODPATH 7.

162
Corresponding author: Chris Nicol
Groundwater Logic Pty Ltd
[email protected]

Benchmarking and Inter-comparison of MODFLOW-USG’s Block Centred

Transport (BCT) Package

Chris Nicol, Groundwater Logic Pty Ltd

Abstract: The conceptual and computational functionality and accuracy of MODFLOW-

USG’s (MF-USG’s) Block-Centred Transport (BCT) package were assessed through bench-
marking and inter-comparison. Three benchmarks from the FEFLOW and MODFLOW-
SURFACT (SURFACT) documentation were tested, with a focus on those of relevance to
Managed Aquifer Recharge (MAR) simulation – i.e. injectant transport and mixing. Quali-
tative inter-comparison with a more complex (‘real-world’) predictive FEFLOW model was
also undertaken. The BCT package produces similar results in a practical sense to the
tested benchmark cases, and broadly similar results to the real-world FEFLOW case, but
that comparison was hampered by mesh-related numerical issues in the FEFLOW model.
For higher Courant / Peclet number cases, the BCT package appears to suffer from slightly
greater numerical dispersion than does FEFLOW and SURFACT, hence mesh refinement
and temporal discretisation appropriate to each simulation case is, as always, required.

Extended abstract: Methodology, findings, and implications

Introduction

A suite of inter-linked local- and regional-scale FEFLOW (DHI-WASY, 2009) models

of groundwater flow and solute transport for a Managed Aquifer Recharge (MAR) scheme
were to be modified, expanded and re-calibrated. The expansion was to include simulated
operations at several other MAR schemes located within the regional model extent. This
demanded a more considered approach to code selection, model design and workflow than
the existing models, as the models were (already) cumbersome, prone to user error, and slow
to build and run.
MF-USG beta is an enhanced version of MFUSG 1.3 (Panday et al., 2013) distributed
with several commercial Graphical User Interfaces (GUIs). It offers the BCT package, and a
range of other enhancements such as adaptive time stepping, in addition to the unstructured
grid capabilities of the United States Geological Survey’s (USGS’s) release of MF-USG 1.3.
The pre-existing MAR models were streamlined into two models (calibration and predictive),
taking advantage of the unstructured grid capabilities of MF-USG, after having assessed the
suitability of its BCT package for MAR simulation relative to the more established FEFLOW
and SURFACT codes. This paper describes that assessment.

Method

163
 Two primary benchmark cases were used to assess the applicability of MF-USG’s BCT
package to simulating MAR operations. The first was that of Wendland (1996), as docu-
mented and benchmarked by DHI-WASY (2009). This benchmark simulates two-dimensional
advective-dominant transport at an oblique flow. It’s geometry and parameterisation are doc-
umented in detail by DHI-WASY (2009). In summary, the benchmark consists of a square
domain with diagonal flow across it from lower-left to upper-right. In the centre of the left
side boundary is a fixed concentration boundary. Two variants of this case were modelled:
a high Courant / low Peclet number case (Case 4), and a low Courant / high Peclet number
case (Case 5). A finer mesh version of each variant was also used for comparison of effects
of the high Peclet / Courant numbers.
The second was that of Gelhar and Collins (1971), as used to benchmark SURFACT.
This benchmark simulates an injection/pumping cycle for a fully penetrating bore within a
flat flow field. The injection cycle simulates injection of contaminant, whilst the pumping
cycle abstracts from the contaminant plume. It is described in detail by HydroGeoLogic
(1996), who benchmarked SURFACT against Gelhar and Collins’ approximate analytical
solution to the problem. For this benchmark, a FEFLOW-equivalent was also built and
assessed, using a quadrilateral mesh.
For both benchmarks, the MF-USG equivalent models were built to replicate as closely
as possible, if not exactly, the corresponding FEFLOW and SURFACT models. In both
cases, inclusion of cross-dispersion terms was activated in the BCT package. In addition
to the above benchmarks, modelled solute concentrations from the pre-existing FEFLOW
model of the MAR scheme that instigated this assessment was also compared to a MF-USG
equivalent using the Voronoi dual of the triangular FEFLOW mesh.

Results

Selected results for Wendland Case 4 (high Courant / low Peclet number) and Case 5 (low
Courant / high Peclet number) are compared between FEFLOW and MF-USG in Figure 1.
The MF-USG BCT package’s simulated concentration field is similar to the range of cases
presented by Wendland (1989) and DHI-WASY (2009), but it does appear to suffer from
slightly more numerical dispersion across the flow field, particularly in the high Courant
number Case 4. In practice, such a high Courant number (1.5) for Case 4 would not be
applied – standard practice is to design a grid that provides values less than 1 to avoid such
issues. For Case 5, such high Peclet numbers (200) would also be avoided in practice. The
MF-USG benchmark models with finer grid resolution provide very similar results to both
the FEFLOW and Wendland cases.
Figure 2 shows the simulated breakthrough curve at the well for the various solutions
to the Gelhar and Collins benchmark. The MF-USG simulation is most similar to the com-
parable (areal) SURFACT case. However, MF-USG appears to suffer from slightly greater
numerical dispersion than SURFACT, but this is insignificant in a practical sense; it would
be unwise to use a similarly excessive time step ( 3 months), cell size and longitudinal disper-
sivity in practical applications, and hence this issue would be unlikely to arise in practice.
Increasing the number of time steps by a factor of 10 resulted in very similar results to
SURFACT, as did a finer mesh. In the real-world FEFLOW MAR model case, MF-USG
also displayed slightly greater numerical dispersion, but this comparison was hampered by

164
numerical errors in the FEFLOW model, thought to primarily arise from a poor-quality
mesh. In a practical sense, the results were sufficiently similar.

Conclusion

MF-USG’s BCT package produces broadly similar results to the two benchmarks tested,
and the real-world MAR case, although MF-USG appears to exhibit slightly greater numeri-
cal dispersion for high Peclet / Courant number cases. Hence, mesh refinement and temporal
discretisation appropriate to each simulation case is, as always, required.

165
Figure 1. Selected FEFLOW and MF-USG Wendland (1996) Benchmark Results

166
 Figure 2. Gelhar and Collins (1971) Benchmark Results

References

DHI-WASY, 2009. DHI-WASY Software FEFLOW, Finite Element Subsurface Flow & Transport Simulation
System. White Papers. Vol. 1.
Gelhar, L.W., and Collins M.A., 1971. General analysis of longitudinal dispersion in non-uniform flow,
Water Resources Research, 7(6):1511-1521.
HydroGeoLogic Inc., 1996. MODFLOW-SURFACT Software (Version 3.0). Overview: Installation, Regis-
tration and Running Procedures. Herndon, VA, USA.
Panday, Sorab, Langevin, C.D., Niswonger, R.G., Ibaraki, Motomu, and Hughes, J.D., 2013, MODFLOW–
USG version 1: An unstructured grid version of MODFLOW for simulating groundwater flow and tightly
coupled processes using a control volume finite-difference formulation: U.S. Geological Survey Techniques
and Methods, book 6, chap. A45, 66 p.

167
Wendland, E. C., Numerische Simulation von Strömung und hochadvektivem Stofftransport in geklüftetem,
porösem Medium. Dissertation, Mitteilung Nr. 96-6, Ruhr-Universität Bochum, 1996.

168
Corresponding author: Robert Reinecke
Goethe University
[email protected]

Building a Global Groundwater Model - Case Study: Comparison to the

Central Valley Hydrologic Model

Robert Reinecke, Goethe University

L. Foglia
S. Mehl
P. Döll

Abstract: In order to represent groundwater-surface water interactions as well as the im-

pact of capillary rise on evapotranspiration in global-scale hydrological models, it is neces-
sary to simulate the location and temporal variation of the groundwater table. This requires
replacing simulation of groundwater dynamics using groundwater storage variations in indi-
vidual grid cells (independent from the storage variation in neighboring cells) with hydraulic
head gradient-based groundwater modeling. Based on the work of two research groups who
have published different approaches for global-scale groundwater modeling, we present the
first results of our effort to develop a transient global groundwater model to replace the
simple storage-based groundwater module of the global hydrological model WaterGAP. To
this end we developed the model code from scratch while adapting MODFLOW principles
for the global scale. The software is tailored to maximize performance and extensibility, and
eventually will be released as an OpenSource project. In order to evaluate the model results
we present a comparison to the well known Central Valley Hydrologic Modell (CVHM).
Furthermore, we discuss current data deficits on the global scale and present an outlook on
future developments.

169
Corresponding author: Willem Schreuder
Principia Mathematica
[email protected]

Effective Utilization of Parallel Resources

Willem Schreuder, Principia Mathematica

Abstract: Parallel processing has significantly advanced our ability to solve larger problems
for the same amount of elapsed time. However, achieving full utilization of computing
resources is difficult to achieve. When dealing with a finite resource such a cluster you own,
or an effectively infinite resource such as cloud computing, the effective utilization of the
resource is often illusive. This talk explores the issues involved in more optimally using the
available resources, and balancing the competing objectives of reducing the total elapsed
time and the cost of the resources. In particular the the talk explores matching the size
of the problem being solved, to the number of parallel processes utilized. For tasks with
an uneven load such as PEST, the talk explores strategies for doing large tasks such as
solving the Jacobian and smaller tasks such as a parallel Lambda search. The talk touches
on techniques for reducing run times for the Jacobian, and more exactly solving the lambda
search. Effective strategies for utilization are different when paying by the CPU hour vs.
owning a cluster with a finite number of cores. Additionally the strategies are different
when atomic actions such as model runs are parallelizable or not, and the length of time
each atomic action takes.

170
Corresponding author: Xiaofan Yang
Pacific Northwest National Laboratory
[email protected]

Intercomparison of Hybrid Multiscale Methods in Simulating Subsurface Flow

and Reactive Transport
Xiaofan Yang, Pacific Northwest National Laboratory
Y. Mehmani
D. Barajas-Solano
M.T. Balhoff
A.M. Tartakovsky
T.D. Scheibs

Abstract: Hybrid multiscale simulations that couple models across scales are critical to
advance predictions of the larger system behavior using understanding of fundamental pro-
cesses. In the current study, three hybrid multiscale methods are intercompared: multiscale
loose-coupling method, multiscale finite volume (MsFV) method and multiscale mortar
method. The loose-coupling method enables a parallel script-based workflow that man-
ages the complex process of executing coupled micro- and macro-scale models sequentially
without being intrusive to the at-scale simulators. The MsFV method applies micro- and
macro-scale models over overlapping subdomains and enforces continuity of concentration
and transport fluxes between models via restriction and prolongation operators. The mor-
tar method is a non-overlapping domain decomposition approach capable of coupling all
permutations of micro- and macro-scale models with each other. In doing so, Lagrange mul-
tipliers are used at interfaces shared between the subdomains so as to establish continuity of
species/fluid mass flux. All the above methods have been proven to be accurate and efficient
to couple pore- and Darcy-scale flow and transport in porous media. However, there has
not been any field-scale applications and benchmarking among various hybrid multiscale
approaches. To address this challenge, we apply all three hybrid multiscale methods to sim-
ulate water flow and transport in a conceptualized 2D modeling domain of the hyporheic
zone, where strong interactions between groundwater and surface water exist across multiple
scales. In all three multiscale methods, fine-scale simulations are applied to a thin layer of
riverbed alluvial sediments while the macroscopic simulations are used for the larger subsur-
face aquifer domain. Different numerical coupling methods are then applied between scales
and inter-compared. Comparisons are drawn in terms of velocity distributions, reactive
transport behavior, algorithm-induced numerical error and computing cost. The intercom-
parison work provides support for confidence in a variety of hybrid multiscale methods and
motivates further development and applications.

171
Corresponding author: Kurt Zeiler-
Brown and Caldwell
[email protected]

Application of the Partition Stress Boundary Capability with

MODFLOW-NWT in a Large-Scale Regional Groundwater Flow Model

Kurt Zeiler, Brown and Caldwell, Lakewood, Colorado

Jeff Weaver, Brown and Caldwell, Durango, Colorado
Andy Moore, Colorado Water Conservation Board, Denver, Colorado
Mary Halstead, Colorado Division of Water Resources, Denver, Colorado

Abstract: The South Platte Decision Support System Alluvial Groundwater Model (model)
was developed by the Colorado Water Conservation Board (CWCB) and the Colorado Divi-
sion of Water Resources (DWR) and is designed to simulate groundwater flow and ground-
water/surface water interactions at a regional scale. One issue in managing model inputs
and outputs in a large-scale regional model is the variety of water budget components that
require tracking. The initial version of the model lumped several flow components into a
single WEL file; these include agricultural, municipal/industrial, and recharge augmenta-
tion pumping, plus lateral boundary flows, bedrock flows, and alluvial underflow. Similarly,
recharge from precipitation, irrigation return flows, canal and reservoir seepage, and recharge
ponds were lumped into a single RCH file. This approach presents difficulties in both data
input preparation and interpretation of model predicted flows. The Partition Stress Bound-
ary (PSB) capability was originally developed by Banta (2011) and was incorporated into
MODFLOW-NWT for this model to improve the simulated water budget accounting. PSB
facilitates “partitioning” of selected MODFLOW packages; each partition is defined by a
separate input file and each partition’s volumetric water budget is tracked and reported
individually. The use of the PSB capability allowed for separate tracking of three pumping
components, six recharge components, three underflow components, and three evapotran-
spiration (ET) components. Use of the PSB capability allowed for overall reduced file sizes
(from over 40 gigabytes to less than 1 gigabyte per file), and greater flexibility in developing
and testing regional-scale management options related to specific flow components.

Extended abstract: Methodology, findings, and implications

The South Platte Decision Support System (SPDSS) Alluvial Groundwater Model (model)
was originally completed in 2012 by the Colorado Water Conservation Board (CWCB) and
the Colorado Division of Water Resources (DWR) as a planning-level tool for the alluvial
aquifer system associated with the South Platte River and its tributaries (Brown and Cald-
well, 2017; CDM-Smith, 2013; Zeiler et al., 2017a). SPDSS is a component of the broader
Colorado’s Decision Support Systems (CDSS) effort managed by CWCB and DWR. This
model was recently updated to use MODFLOW-NWT and extend the simulation period
(Niswonger et al., 2011). The model was designed to simulate groundwater flow and ground-
water/surface water interactions with streams at a regional scale in the South Platte River
Basin.
The South Platte River Basin provides water used for a wide variety of economic activi-
ties including agriculture, municipal and industrial (M&I) uses, and recreation. As such, the

172
model includes several water budget components that are commonly simulated using speci-
fied flow boundaries including agricultural pumping, M&I pumping, precipitation recharge,
irrigation return flow recharge, canal seepage recharge, alluvial underflow/lateral boundary
flows, reservoir seepage, water rights augmentation pumping, recharge ponds, and bedrock-
alluvial flows. Evapotranspiration (ET) is simulated in the model for subirrigated alfalfa,
subirrigated meadow, and phreatophyte vegetation. The initial version of the model used the
MODFLOW WEL and RCH packages to simulate the specified flows and the ETS package
to simulate ET from the three vegetation types (Banta, 2000). Simulated water budgets
were reported with the outputs mixed together in the MODFLOW package used for each
component.
To improve the tracking of multiple water budget components within MODFLOW, the
Partition Stress Boundaries (PSB) capability was developed by the United States Geolog-
ical Survey with funding from CWCB as part of MODFLOW-CDSS (Banta, 2011). The
PSB capability tracks water budget components (“partitions”) within MODFLOW based on
user-specified categories rather than being constrained to categories based on MODFLOW
package types. The PSB capability can track recharge by source, well pumping rates by wa-
ter use type, ET by vegetation type, as well as other types of commonly simulated stresses.
Each partition has an individual input file using the standard input format of the underly-
ing package, and the PSB capability is invoked by entering a 16-character identifier in the
MODFLOW Name file after the standard entry of package type, unit number, and filename.
This identifier internally replaces the standard TYPE variable for the package name in
MODFLOW. The simulated water budget results are printed with the user-specified TYPE
identifiers to the output List file and cell-by-cell files as specified by the user in the OC
package input file. Many graphical user interfaces for MODFLOW will skip non-standard
TYPE values in cell-by-cell file headers, but ZONEBUDGET or other utilities can be used
(Harbaugh, 1990).
During the recently completed model update effort, MODFLOW-CDSS source code for
the PSB capability was added to MODFLOW-NWT and support for PSB was added as
an option to the State Pre-Processor (StatePP) data-centered approach tool (Zeiler et al.,
2017b). StatePP was used to develop separate input files for agricultural pumping, M&I
pumping, precipitation recharge, irrigation return flow recharge (separated into surface water
and groundwater sources), canal seepage recharge, alfalfa ET, subirrigated meadow ET, and
phreatophyte ET. Input files representing other components were created using additional
data-centered approach tools, but kept separate rather than being combined as previously.
Simulation with PSB-enabled MODFLOW-NWT produced budget outputs that were easily
analyzed for each component.
The PSB capability provides several benefits in addition to improved tracking of water
budget components. Performing predictive scenarios is simplified with PSB for any case
involving simulation of an additional stress because an input file for the additional stress
can be developed separately, added to the model, and the water budget for this additional
component is automatically tracked. Using individual input files improves the model input
file construction and quality control (QC) processes by avoiding the need to enter and check
inputs for multiple components in a single file. For several MODFLOW package types,
additional calculations for inputs can be avoided (e.g., summing recharge from different
sources in a single RCH file).

173
 The PSB capability greatly simplified the inputs to the ETS package for the model
with three vegetation types having different extinction depths, ET rate and depth profiles,
and maximum ET rates. The previous implementation of the model required that these
extinction depths, depth profiles, and maximum rates be calculated for each model cell using
area-weighted averages. The arrays for these inputs resulted in an approximately 40 gigabyte
(GB) size ETS input file, which made it virtually impossible to perform QC activities without
text-parsing tools. The extinction depths and the ET rate and depth profiles for each
vegetation type are constants in each stress period, so by separating the vegetation types
into individual PSB partitions those values can be specified as single constant values rather
than arrays. The PSB approach to the ETS package combined with use of external binary
files for arrays reduced the overall ETS input file size to approximately 2 GB, and the main
text input files are easily read with any editor.
Using the PSB capability increases the overall input and output for any given model
simulation, but there are several strategies to help mitigate the increased data storage re-
quirements. Using PSB to separate components may reduce array inputs to constant values;
otherwise, external binary files can be employed. Recharge sources that are not broadly spa-
tially distributed (e.g., canal seepage, recharge ponds) can be represented with much smaller
input files using equivalent inputs to the WEL package rather than the RCH package. Out-
put cell-by-cell flow files can be reduced using the COMPACT BUDGET keyword in the
OC package. These strategies reduced the overall SPDSS model input file size total from
approximately 48 GB to 7.5 GB, and allowed the output file size total to remain about the
same.
See the MODFLOW-CDSS documentation for a full description of the PSB capabilities
(Banta, 2011).
See the CDSS website (http://cdss.state.co) for more information on acquiring CDSS
datasets, tools, models and documentation.

References

Banta, E.R. 2000. MODFLOW-2000, the U.S. Geological Survey Modular Ground-Water Model - Docu-
mentation of Packages for Simulating Evapotranspiration with a Segmented Function (ETS1) and Drains
with Return Flow (DRT1): U.S. Geological Survey Open-File Report 00-466, 127 p.
Banta, E.R. 2011. MODFLOW-CDSS, a version of MODFLOW-2005 with modifications for Colorado
Decision Support Systems: U.S. Geological Survey Open-File Report 2011-1213, 19 p.
Brown and Caldwell. 2017. South Platte Alluvial Groundwater Model Update Documentation: prepared
for the Colorado Water Conservation Board and Division of Water Resources, May.

174
CDM-Smith. 2013. South Platte Decision Support System Alluvial Groundwater Model Report: prepared
for the Colorado Water Conservation Board and Colorado Division of Water Resources, April.
Harbaugh, A.W. 1990. A computer program for calculating subregional water budgets using results from
the U.S. Geological Survey modular three-dimensional ground-water flow model: U.S. Geological Survey
Open-File Report 90-392, 46 p.
Niswonger, R.G., Panday, S., and Ibaraki, M. 2011. MODFLOW-NWT, A Newton formulation for MODFLOW-
2005: U.S. Geological Survey Techniques and Methods 6-A37, 44 p.
Zeiler, K., Moore, A., Halstead, M., Wengrovius, Z., Weaver, J., Lindburg, M., Dupre, P. 2017. Updates
and improvements to Colorado’s South Platte Decision Support System Alluvial Groundwater Model. In
MODFLOW and More 2017: Modeling for Sustainability and Adaptation. Golden, Colorado, USA: Colorado
School of Mines.
Zeiler, K., Moore, A., Halstead, M., Wengrovius, Z., Weaver, J., Lindburg, M., Dupre, P. 2017. Use of a Data-
Centered Approach in updating Colorado’s South Platte Decision Support System Alluvial Groundwater
Model. In MODFLOW and More 2017: Modeling for Sustainability and Adaptation. Golden, Colorado,
USA: Colorado School of Mines.

175
Corresponding author: Yiding Zhang
The Ohio State University
[email protected]

Performance analysis of the Chi-MD matrix solver package for MODFLOW

Yiding Zhang, The Ohio State University

Motomu Ibaraki, The Ohio State University
Richard G. Niswonger, U.S. Geological Survey
Sorab Panday, GSI Environmental, Inc.

Abstract: The 𝜒MD matrix solver package is incorporated in to MODLFOW-NWT and

MODFLOW-USG to solve both symmetric and unsymmetric matrices assembled through
numerical discretization of groundwater flow equation and Newton-Raphson linearization.
It has demonstrated its higher robustness, faster execution speed and more efficient mem-
ory usage compared to the existing solvers. The 𝜒MD matrix solver package consists of
preconditioning and acceleration modules. Because the 𝜒MD matrix solver uses a variety
of features including level-based incomplete LU (ILU) factorization method with a drop tol-
erance scheme in the preconditioning module, the users have to choose parameters for the
solver in order to obtain better execution speed and robustness. In order to examine how
the ILU level and drop tolerance values affect the overall performance of the 𝜒MD matrix
solver, we conducted a series of simulations for five scenario cases with different numerical
complexities. For those five cases, the domain size and number of nodes vary from 8.2 km2
to 59,108 km2 and from 10,000 to 730,300, respectively. From the analysis, we found that
execution times for preconditioning and acceleration procedures greatly changed, however,
a combination of ILU level equal to 6 or 7 and the drop tolerance value between 0.01 and
0.001 usually result in shorter overall execution time. Our study suggests that the users
can elicit higher performance of the 𝜒MD matrix solver using this combination and solve
groundwater problems more efficiently.

Extended abstract: Methodology, findings, and implications

Introduction

In most numerical simulation models, more than 80% of memory and execution time
are used in matrix solver. Improving the performance of matrix solver, therefore, is a key
to improving the overall performance of the simulation models. The 𝜒MD matrix solver
package is incorporated into MODLFOW-NWT and MODFLOW-USG to solve matrices.
The solver has demonstrated higher robustness, faster execution speed, and more efficient
memory usage compared to the existing solvers (Ibaraki et al. 2011).
The 𝜒MD matrix solver package consists of preconditioning and acceleration modules.
The preconditioning module includes level-based incomplete lower-upper (ILU) factoriza-
tion with a drop tolerance scheme, which aims to minimize computational memory cost
and reduce execution time (Niswonger et al. 2011). The acceleration module uses iter-
ative Krylov-subspace methods to solve a transformed matrix based on the result of the
preconditioning procedure.
During level-based ILU factorization, a parameter of level is assigned to each potential
fill entry so that the decomposition is terminated based on the setting of the level. With
a higher level of factorization, fewer iterations are required. However, cost-per-iteration is

176
much higher than that of using lower level factorization because a larger number of new
entries are filled in the matrix during the process of ILU factorization (Hysom & Pothen
2002). In order to achieve a higher level factorization without increasing cost-per-iteration,
a drop tolerance strategy is applied. With drop tolerance, all the entries which are smaller
than the given tolerance with certain criteria will be discarded. By using this strategy,
cost-per-iteration will be reduced even if a higher level factorization is selected.
In order to examine how the values of ILU level and drop tolerance, which determine
the quality of the preconditioned matrix, affect the overall performance of the 𝜒MD matrix
solver, we conducted a series of simulations for five scenario cases with different numerical
complexities.

Results and Discussion

Performance of Preconditioning and Acceleration Processes

Figure 1 illustrates the execution time with different levels and drop tolerance values
for the (a) preconditioning, (b) acceleration, and (c) entire processes. We only show the
results from Case 2 for a demonstration purpose. It can be seen from Figure 1(a) that the
execution time increases as the level of ILU factorization increases for smaller drop tolerance
value cases (e.g., 10-6 case). On the other hand, the execution time for larger drop tolerance
value cases (e.g., 10-1 case) is constant over the level. With larger drop tolerance values,
more non-zero entries in the factorized matrix are discarded so that the matrix size does not
increase even if the level of ILU factorization increases. On the other hand, more non-zero
entries in the factorized matrix are preserved for smaller drop tolerance value cases, and
accordingly, the execution time increases as the level of ILU factorization increases.
The execution time for the acceleration process shows different behavior than that of the
preconditioning process (Figure 1(b)). Smaller drop tolerance cases (10-4, 10-5, 10-6) shows
that the execution time decreases as the level of ILU factorization increases up to the level
equal to two. However, further increases in the level increase the execution time. Other
cases show a decrease of execution time when level increases up to two and no changes in
the execution time even if the level value become greater than two.
In order to understand this behavior, we investigate the acceleration process in details.
Figure 2 illustrates (a) the number of iterations, (b) the average execution time per iteration,
and (c) the entire execution time for the acceleration process for a single Picard iteration
loop. The number of iteration decreases as the level of ILU factorization increases for most of
the cases (Figure 2(a)) because the quality of the preconditioned matrix increases. Because
more non-zero entries are discarded for the 10-1 case in the preconditioning process, the
quality of factorized matrix becomes lower, and this results in higher iteration numbers.
The execution time per iteration, on the other hand, increases as the level of ILU factor-
ization increases for most of the cases (Figure 2(b)). This is because the number of non-zero
entries in the factorized matrix increase as the level of ILU factorization increases. Higher
drop tolerance value cases (10-1, 10-2, 10-3), in contrast, show little changes in the execution
time due to limited numbers of non-zero entries are preserved in the factorized matrix.
The entire execution time for the acceleration process, which is equal to the multiplication
of the number of iterations and the average execution time per iteration, shows that the 10-2,

177
10-3, and 10-4 cases with higher levels of ILU factorization achieves shorter execution time.
On the other hand, smaller drop tolerance value cases (10-5 and 10-6) illustrate an increase
in the execution time as the level increases. This is because an increase in the cost per
iteration (Figure 2(b)) offsets a decrease in the number of iteration (Figure 2(a)).
Combining the preconditioning and acceleration processes together, Figure 1(c) presents
the entire execution time of the 𝜒MD solver. It can be seen from the figure that smaller
drop tolerance value cases (10-5, and 10-6) show a significant increase in the execution time
with a higher level. On the other hand, medium drop tolerance cases (10-2, 10-3, and 10-
4) illustrate shorter execution time for higher level as well. Although we do not show the
results from other cases, we find that the shortest overall execution time usually occurs in
the combination of ILU level equal to 6 or 7 and the drop tolerance value between 0.01 and
0.001.

Conclusion

The effects of matrix solver parameters that control the quality of preconditioning and
overall performance of the 𝜒MD matrix solver package are examined through a series of
numerical analysis. The analysis shows that both preconditioning and acceleration processes
are greatly affected by the level of ILU factorization and drop tolerance values. It also shows
that a combination of ILU level equal to 6 or 7 and the drop tolerance value between 0.01
and 0.001 usually result in shorter overall execution time.

178
Figure 1. Execution time with different levels and drop tolerance values for the (a) preconditioning, (b)
acceleration, and (c) entire processes.

179
Figure 2. (a) The number of iterations, (b) the average execution time per iteration, and (c) the entire
execution time for the acceleration process for a single Picard iteration loop.

References

Hysom, D. & Pothen, A., 2002. Level-based incomplete LU factorization: Graph model and algorithms. In
SIAM Journal On Matrix Analysis and Applications. pp. 1–19.
Ibaraki, M. et al., 2011. Improvement of performance and applicability of MODFLOW-2005: new NWT
solver and �MD matrix solver package. IAHS-AISH publication, pp.100–105.
Niswonger, R., Panday, S. & Ibaraki, M., 2011. MODFLOW-NWT, a Newton formulation for MODFLOW-
2005.

180
Corresponding author: Rosemary Carroll
Desert Research Institute
[email protected]

Legal and Hydrologic Considerations to Curtail Agricultural Pumping during

Extended Drought

Rosemary Carroll, Desert Research Institute

G. Pohil
J. Benedict
R. Felling

Abstract: Many arid and semi-arid agricultural systems of the Great Basin in the west-
ern United States depend on supplemental groundwater pumping to augment diminished
surface water flows during periods of drought. As droughts become longer and more severe
in the region, unprecedented drawdown in these aquifer systems has occurred with legal
and environmental implications on both surface and groundwater. The Walker River in the
Great Basin supports extensive agriculture in the region and is the sole perennial stream
to one of the few desert terminal lakes in North America. Continuous declines in the lake
have spurred extensive research into management options to balance demands of agriculture
and increase water deliveries to the lake. Smith and Mason Valleys are important agricul-
tural centers within the Walker Basin. In 2015 the region entered its fifth year of drought
and both valleys were the focus of curtailment orders to restrict the use of supplemental
groundwater rights. To aid management decisions, hydrologic models were developed that
simulate complex feedbacks between surface diversions, crop consumptive needs, ground-
water recharge, return flow, and groundwater-surface water interactions. Demand-driven
pumping that incorporates priority dates and maximum duty allocations are directly input
to the hydrologic model to allow an assessment of groundwater curtailment options under
a variety of drought scenarios to meet targeted water levels and downstream conveyance of
surface water in a legally defensible framework. Hydrologic results using a sliding scale ap-
proach to priority based curtailment are presented in the arena of stakeholder participation
and response.

181
Corresponding author: Laura Condon
Syracuse University
[email protected]

Assessing the impact of groundwater pumping on the integrated hydrologic

cycle

Laura E. Condon, Syracuse University

R.M. Maxwell, Colorado School of Mines

Abstract: Unsustainable groundwater pumping has caused widespread groundwater de-

clines and threatened the long-term sustainability of water resources in many locations.
Although there has been significant work to try to quantify water table changes and re-
maining aquifer storage volumes, the impact of pumping on the dynamics of interconnected
groundwater surface water systems remains poorly quantified. Previous research has shown
that groundwater can buffer surface water variability in natural systems and provide a sta-
ble water supply to managed systems; more work is needed to understand how interactions
change in systems with sustained groundwater depletions. We use an integrated hydrologic
model to simulate the impacts of groundwater pumping on system dynamics and water
management decision-making. Changes in land energy fluxes and streamflow variability are
evaluated across the majority of the continental US by comparing a predevelopment ground-
water scenario with a developed scenario that incorporates the groundwater depletions of
the 20th century. Differences between these simulations isolate the impact of groundwater
development from other water management changes over this time period and identify sen-
sitive regions where groundwater losses can fundamentally alter surface water behavior. We
also use long term simulations of a regional domain in the High Plains aquifer to simulate
moisture dependent groundwater supported irrigation. Multi-year analysis shows the po-
tential for positive feedbacks between groundwater drawdown, soil moisture variability and
irrigation demand. Results from both simulations demonstrate the importance of historical
groundwater pumping in understanding previous hydrologic change as well as the role of
ongoing pumping in integrated system dynamics and irrigation demand.

182
Corresponding author: John Ewing
INTERA Incorporated
[email protected]

Stream-Aquifer Interaction in a Regional Alluvium Aquifer

John E. Ewing, INTERA Incorporated

Marius Jigmond, INTERA Incorporated

Abstract: A regional groundwater flow model was developed for the Brazos River Allu-
vium Aquifer in east-central Texas. The aquifer is approximately 220 miles long but only
several miles wide and is in hydraulic communication with the Brazos River - the largest
river in Texas - along the entire length of the aquifer. Specific attention was paid to inter-
action between the Brazos River and the Alluvium Aquifer and the effects of agricultural
pumping in the aquifer. To address both regional and local phenomena in a computation-
ally efficient manner, an unstructured quadtree grid was developed for MODFLOW-USG.
The Alluvium Aquifer transects numerous underlying formations including three previously
developed Groundwater Availability Models developed for the state. A software tool was
developed to connect the Brazos River Alluvium model temporally and spatially to the
three underlying models so that the effects of pumping in the underlying aquifers could
be accounted for in this model. Although water levels in the alluvium have not declined
significantly over time, simulations indicate that agricultural pumping in the alluvium is
indirectly capturing baseflow to the Brazos River.

Extended abstract: Methodology, findings, and implications

The Brazos River Alluvium Aquifer is a shallow, surficial aquifer in southeast Texas which
provides a source of groundwater used primarily for agricultural irrigation purposes, with
smaller quantities used for rural domestic, livestock, and municipal purposes. A numerical
groundwater model was developed as part of the Texas Water Development Board’s ground-
water availability model program to provide a tool for managing the groundwater resources
in the aquifer.
The aquifer consists of the floodplain deposits and hydraulically connected terrace de-
posits of the Brazos River. Sediments comprising these deposits range from clay to large
cobbles and occur in lenses that grade both laterally and vertically. The transition from one
type of material to another, both laterally and vertically, can be either sharp and distinct or
gradual. The Brazos River Alluvium Aquifer is unconfined with potentially locally confined
conditions where clay lenses overlie lenses of sand or gravel. From northwest to southeast,
the aquifer overlies the Carrizo-Wilcox, Queen City, Sparta, Yegua-Jackson, and Gulf Coast
aquifers. The shallow portions of these aquifers are assumed to be hydraulically connected
to the Brazos River Alluvium Aquifer since they are conceptualized to discharge regionally
to the Brazos River.
The code used to implement the numerical model was MODFLOW-USG. The model
consists of three layers, and the model grid is composed of square grid cells ranging from
1/8-mile to 1-mile in size. The first two model layers were used to represent the upper

183
and lower portions of the Brazos River Alluvium, respectively, and a third layer was used to
represent the shallow portions of the underlying formations within the Brazos River Basin. A
quadtree mesh was used to transition laterally from the more refined cells within the footprint
of the Brazos River Alluvium Aquifer to the coarser cells used within the remainder of the
Brazos River Basin (see Figure 1). This allowed for both the local and shallow portions of
the regional groundwater flow systems to be simulated in a computationally efficient manner.
Because the Brazos River Alluvium Aquifer overlies a series of major and minor aquifers,
for which three numerical groundwater models already exist, a software tool was developed
to account for the hydraulic communication between this model and the models of the un-
derlying aquifers. In this way, the regional-scale flow systems of the underlying models were
incorporated into the local and intermediate-scale flow systems simulated in the alluvium
aquifer model. In addition, the temporal and spatial effects of pumping in the underlying
aquifers on the cross-formational flows to and from the alluvium aquifer were directly ac-
counted for in the model. Figure 2 depicts the simulated cross-formational flow between this
model and the underlying models at the end of the simulated historical period. This is of
significant importance when any of the underlying models are used to simulate large quanti-
ties of groundwater development in the vicinity of the Brazos River Alluvium for predictive
scenarios involving those aquifers in groundwater management planning. The software tool
provides an easy means of simulating those effects on the alluvium aquifer.
In the development and calibration of the model, particular attention was paid to the
effects of the significant temporal fluctuations in the streamflow (and associated stage) in
the Brazos River on stream-aquifer interaction. The model was tested at daily, monthly, and
annual stress period lengths over the same time period in 2007 and the results were compared.
This comparison indicated that, while the magnitude of stream-aquifer interaction varies
dramatically among the three time scales investigated, a monthly time scale is accurate
for simulating seasonal water-level variations within the aquifer that may be important to
planning for irrigation purposes and the annual time scale is appropriate for simulating
long-term aquifer management planning.
The model simulates the historical time period from 1950 to 2012, with an initial steady-
state stress period that represents pre-development conditions. The model was primarily
calibrated to observed heads in the Brazos River Alluvium Aquifer. It was calibrated to
both steady-state and transient conditions. Both the steady-state and transient calibration
statistics are well within acceptable ranges. The model was also calibrated to steady-state
estimates of baseflow to the streams in the model domain and to steady-state observed heads
in the formations underlying the Brazos River Alluvium Aquifer.
In the steady-state calibration, recharge is the major source of inflow to the Brazos River
Alluvium Aquifer followed by cross-formational flow from the underlying units. Discharge
to perennial rivers is the largest source of outflow. In the transient model, perennial rivers
continue to be the largest source of outflow on average but, in individual months, pumping
may eclipse outflow to perennial rivers. Recharge and cross-formational flow continue to be
the major sources of inflow to the aquifer in the transient model. It should be noted that the
Brazos River can, at any given day, month, or year, act as either a major source of inflow
or a major source of outflow to or from the Brazos River Alluvium Aquifer based on the
variations in stream stage in the river.

184
 Cross-formational flow from the underlying formations into the Brazos River Alluvium
Aquifer remains relatively constant at around 50,000 acre-feet per year throughout most of
the transient record although it can fluctuate significantly on a monthly basis. The largest
change in the transient water budget over the historical period is an increase in irrigation
pumping in recent years.
A sensitivity analysis was performed, which indicated that heads in the Brazos River
Alluvium Aquifer were most sensitive to the horizontal hydraulic conductivity of the upper
portion of the aquifer. The heads in the aquifer were also sensitive, in decreasing order, to
recharge to the alluvium, the horizontal hydraulic conductivity of the lower portion of the
aquifer and to the streambed conductance of perennial streams.
All groundwater models have limitations with respect to data support, scale, and the
assumptions used in their development. However, the work described here resulted in a well-
calibrated model of the Brazos River Alluvium Aquifer that can be used to support water
availability planning at a regional scale, while accounting for local scale processes.

185
Figure 1. Quadtree grid allowing refinement within the alluvium and coarser cells elsewhere.

186
Figure 2. Cross-formational flow to and from the underlying models.

187
 References

Ewing, J.E., M. Jigmond, 2016. Numerical Model Report for the Brazos River Alluvium Aquifer Ground-
water Availability Model. Final report prepared for the Texas Water Development Board.
Ewing, J.E., J.J. Harding, T.L. Jones, C. Griffith, J.S. Albright, B.R. Scanlon, 2016. Conceptual Model
Report for the Brazos River Alluvium Aquifer Groundwater Availability Model. Final report prepared for
the Texas Water Development Board.

188
Corresponding author: Claudia Faunt
U.S. Geological Survey
[email protected]

Simulating Water Availability and Sustainability in California’s Central Valley

Claudia C. Faunt, U.S. Geological Survey/CAWSC

J. Traum

Abstract: California relies heavily on groundwater to meet its water needs. It provides a
crucial buffer against land-use change effects, water restrictions, drought, and the impacts of
climate change, including the depletion of mountain snowpack, all of which affect surface wa-
ter supplies. Despite its essential role, the state’s groundwater system is under considerable
strain and until recently has been largely unregulated. California’s Sustainable Groundwa-
ter Management Act of 2014 (SGMA) provides a framework to comprehensively measure
and manage groundwater and empowers local agencies to assess hydrologic issues that man-
ifest as ”undesirable results”. California’s Central Valley has many basins with ”undesirable
results”. The Central Valley covers about 20,000 mi2 and is one of the most productive
agricultural regions in the world. Because the valley is semi-arid, surface-water availability
varies substantially. Agricultural demand for irrigation is heavily reliant on surface wa-
ter and groundwater. Starting in the 1950s, state and federal water distribution systems
have eased the reliance on groundwater as dependence shifted to diverted surface water. In
the last 20 years, however, land-use changes and limitations to surface-water availability-
including drought and environmental flow requirements - have increased pumping, causing
groundwater-level and groundwater-storage declines, subsidence, decreased stream flows,
and changes to ecosystems. As these recent trends continue, monitoring and modelling
are critical to understanding the dynamics of groundwater use and developing better man-
agement strategies. Modeling tools, such as the USGS Central Valley Hydrologic Model,
enable managers to develop strategies to mitigate adverse impacts while also optimizing
water availability.

189
Corresponding author: Michael Fienen
U.S. Geological Survey
[email protected]

Imagining a maximum future irrigation footprint to evaluate water resources in

a Wisconsin County

Michael N. Fienen, U.S. Geological Survey

P. Juckem
M. Parsen
M. Gotkowitz
S. Ebel
D. Masterpole

Abstract: For many groundwater modeling projects, forecasts of responses to future con-
ditions drive the analysis. A challenge is to adequately represent such future conditions
with balancing specificity and flexibility of an unknown future. An example solution to this
challenge may be drawn from the atmospheric sciences, where ensembles of future climate
forcings informed by global circulation models can be carried forward to provide a range of
likely forecasts. We took a similar approach for a county scale groundwater modeling project
in Chippewa County, Wisconsin USA. Stakeholders want to understand base flow and water
level impacts in the face of uncertain future land use pressures. Industrial sand mining for
hydraulic fracturing and irrigated agriculture are key land use changes occurring now and
into the future. In this presentation, we discuss how to determine a reasonable maximum
impact case of base flow changes due to increasing irrigated agriculture. Distributed mapped
datasets were examined incorporating stakeholder expertise in determining a footprint of
potentially irrigable land based on soil, existing land use, slope, and other mappable char-
acteristics. Stakeholders indicated that 80ύ of that footprint was likely to be irrigated,
so a stochastic approach was adopted using 1,000 random samples of the 80ύ footprint,
assigning recharge and pumping changes based on agriculture in other parts of Wisconsin,
and maintaining the proportion of the three most common crops throughout the irrigated
land. Each realization was run through the groundwater model resulting in an ensemble of
base flow and water level changes including estimates of their uncertainty.

190
Corresponding author: Christopher Green
U.S. Geological Survey
[email protected]

Forecasting nitrate in agricultural groundwater

Christopher Green, U.S. Geological Survey

Lixia Liao, U.S. Geological Survey

Abstract: Agricultural nitrate (NO3 - ) has degraded shallow groundwater and surface wa-
ter quality across extensive areas of farmland. Uncertainties in the sources, reactions, and
transport of nitrate pose challenges to water quality forecasts that are essential for balancing
agricultural demands with sustainable water resources. The Vertical Flux Model (VFM) is a
simple, adaptable modeling tool that uses chemical data to estimate nitrate fluxes and reac-
tions in the subsurface. Previous studies applied the VFM locally at 14 sites with intensive
data sets across the US to estimate the depth of nitrate extinction (the eventual depth of
the nitrate front) in various settings. Nitrate extinction depth was found to depend mainly
on aquifer dimensions and denitrification rates. An ongoing study has applied the VFM
regionally in the Fox-Wolf-Peshtigo river basins of Wisconsin. Simulated concentrations
of NO3 - , N2 from denitrification, O2 , and environmental tracers of groundwater age were
matched to observations by adjusting parameters for recharge, source concentration, O2
reduction rate, O2 threshold for denitrification, denitrification rate, and dispersivity. Cor-
relations between calibrated parameters and GIS parameters (land use, drainage class, soil
properties, etc.) were evaluated and revealed that relationships between VFM parameters
and landscape characteristics were consistent with expected relationships. For example, the
estimated source concentrations were lower in the vicinity of wetlands and greater near crop
lands. Overall, the results demonstrate applicability of the VFM at a regional scale, and pro-
vide a framework for a hybrid statistical-VFM model to estimate NO3 - fluxes continuously
across the study area.

191
Corresponding author: Randall Hanson
U.S. Geological Survey
[email protected]

Examples of Integrated Modeling of Agricultural Conjunctive-Use from the

Lower Rio Grande Valley, New Mexico and Texas to the Salinas Valley,
California

Randall T. Hanson, U.S. Geological Survey

A.B. Ritchie
S.E. Boyce
A. Galanter
W. Henson

Abstract: Agricultural development, including shifts to more water-intensive crops, and

climate variability are placing larger demands on the sustainability of water resources in
many regions across the United States. Conjunctive use has resulted in surface-water and
groundwater depletion and related secondary effects such as reduced surface-water deliveries
and flows as well as land subsidence and seawater intrusion. The Lower Rio Grande provides
a good example where transboundary-conjunctive use and reuse of water provides new
constraints on sustainability. Similarly, the Salinas Valley provides a good example where
competing interests for environmental and agricultural demands is creating new demands
on sustainability through increased crop rotation of high value crops, salinity demands, and
environmental flows, as well as new augmentations such as the reuse of recycled water,
dual reservoir operation, and controlled extraction permits near the Salinas River. The
’One Water’ model (MODFLOW-OWHM) provides the simulation framework needed to
simulate and analyze these complex and changing sets of supplies and demands in both of
these agricultural settings by use of features that address conjunctive use. The ’One Water’
modeling framework also facilitates the assessment of sustainability within the context of
climate variability and climate change and related adaptation analysis by integrating water
management. These examples illustrate the utility of simulating fully linked conjunctive
use.

192
Corresponding author: Annette Hein
Colorado School of Mines
[email protected]

Drought on the North American high plains: modeling the effects of

vegetation, temperature, and rainfall perturbations on regional hydrology
Annette Hein, Colorado School of Mines
R.M. Maxwell, Colorado School of Mines

Abstract: The North American high plains are currently one of the most productive agri-
cultural regions, but they are particularly vulnerable to drought. Over the past decades,
this region has seen some of the largest groundwater declines in the world. As groundwater
supplies are depleted, farmers may stop pumping and irrigating, leaving the region more
vulnerable to varying rainfall year to year. If there is a year with too little rainfall for
agricultural needs and farmers cannot use groundwater instead, they would not be able to
produce crops that year. The Dustbowl drought of the 1930s is a worst-case example of
possible consequences. In this study, we examined existing temperature and precipitation
records to characterize the Dustbowl and compare it to other droughts in the 20th century.
Three major factors caused the Dustbowl: increased temperature, decreased precipitation,
and a change from native grasses to dryland wheat farmingâ vegetation that could not toler-
ate these climate perturbations. Using this background information, we built an integrated
hydrologic model of the region to run numerical experiments. The model was constructed
following previous work on modeling the hydrology of the continental US using existing
public domain datasets as input. The simulation used ParFlow-CLM, a fully-coupled code
that includes surface and subsurface flow together with land surface processes such as snow
and surface energy budgets. The model domain has a resolution of 1 km and covers Ok-
lahoma, Nebraska, Kansas, and portions of neighboring states. This was the region most
impacted during the Dustbowl, and suffers from a shortage of groundwater today. The nu-
merical experiments start with a baseline run using typical climatology of the region. Three
additional runs explore the individual effects of raising temperature, lowering rainfall, and
removing vegetation. Finally, a worst-case run combines all three perturbations, as in the
historic Dustbowl drought. To evaluate the effect of these changes on regional hydrology,
we examine water table levels, stream flows, and soil moisture content in the model. We
compare the effects found to historic measurements where data is available. The results of
this modeling exercise should be relevant to water management in the High Plains and areas
with similar water shortages such as California and India.

193
Corresponding author: Mary Hill
University of Kansas
[email protected]

A Food-Energy-Water Calculator, with initial application to Western Kansas

Mary C. Hill, University of Kansas

Ruth Douglas Miller, Kansas State University
Danny Rogers, Kansas State University
Joshua Roundy, University of Kansas

Abstract: Many agricultural areas are facing difficult circumstances. Kansas is one exam-
ple, with problems typical of other places. Past agricultural and hydrologic data document
how irrigation in western Kansas has produced a multi-billion-dollar agricultural economy
that is now threatened by pumping-induced declines in groundwater storage. A disincentive
of reducing pumping is the economic impact of diminished irrigation. One alternative is a
water-energy tradeoff that seeks to enable enduring use of the Ogallala aquifer while main-
taining local income levels through development of the rich regional wind and solar energy
resources in a way that focuses on local economic benefits. In considering this alternative,
most citizens are currently unable to address a key question, “What could this mean for
me?” Answers require knowledge of agriculture, energy, water, and drought probabilities,
knowledge that is available at Kansas universities. This talk presents a KU-KSU effort to
address this need through development of the Food-Energy-Water Calculator. This talk
will present the idea and discuss how the calculator would work. It is suggested that the
framework created provides a powerful way to organize data and analysis results, and thus
to seek solutions to difficult problems in many regions of the US and the world.

194
Corresponding author: Eric Morway
U.S. Geological Survey
[email protected]

Improving simulation and understanding of third-party impacts resulting from

water-right transfers

Eric D. Morway, U.S. Geological Survey

Rich Niswonger, U.S. Geological Survey
Enrique Triana, Research Triangle Institute

Abstract: Water resource investigations typically rely on river-operations models to in-

corporate water rights and regulations in the analysis of alternative management scenarios.
However, these models crudely approximate interactions between the surface water and
groundwater, and can be inaccurate in systems where surface water/groundwater exchanges
are important. Changes in river management and land use can alter groundwater storage,
recharge, and discharge, and ultimately affect the outcome of water management strate-
gies. Furthermore, changes in water management strategies, such as a transfer of a water
right between two parties, often impacts a third party that is not directly involved in the
water transaction. For example, water-rights transfers between two parties can result in
third-party impacts where a water transfer changes the place of use and results in a loss of
return flows that historically supplied the third party’s water right. Third-party impacts
also can result from irrigation efficiency improvements, such as canal sealing, switching from
food irrigation to sprinkler or drip irrigation, or changes to reservoir management. Inher-
ently, third party impacts are dependent on both surface water and groundwater processes
that are difficult to assess without a model that integrates explicit representation of these
processes. Accordingly, it is important to assess the outcome of water rights transfers,
and other changes in water management with models that simulate both river/reservoir
operations and surface water and groundwater interactions. A new modeling tool, referred
to as MODSIM-MODFLOW, implicitly-couples hydrology and river-operations and simu-
lates the complex feedbacks that occur among hydrologic and water management processes,
including changes in water rights, water infrastructure, and reservoir operations. MODSIM-
MODFLOW can be applied over a broad range of water management problems in developed
river basins because it allocates water from rivers and reservoirs among a distributed set
of water demands in order of their priority, while also representing the physically-based
GW-SW response to that distribution. Thus, it is uniquely suited for directly assessing
third-party impacts resulting from water transfers. A hypothetical model patterned after
Carson Valley, NV, but typical of irrigated systems throughout the semi-arid West, is used
to demonstrate the downstream impacts of upstream water transfers on third parties using
the new MODSIM-MODFLOW simulation software.

195
Corresponding author: Christopher Peters
Montgomery and Associates
[email protected]

Development of a Wellfield Operating Plan for a Vineyard in Peru using

MODFLOW’s Groundwater Management Process: A Case Study

Christopher Peters, Montgomery and Associates

Abstract: A table-grape vineyard in Peru is irrigated with groundwater pumped from an

unconfined alluvial aquifer. Water from each well is pumped to a central reservoir and then
distributed via a drip-irrigation system. The vineyard’s permitted groundwater pumping
capacity exceeds the projected irrigation demand. The grower sought to establish a wellfield
operating plan that would minimize well-to-well interference and drawdown impacts to ad-
jacent users while meeting the irrigation demand and water-quality targets. A MODFLOW
model was developed and calibrated to groundwater contour levels in a government report
as no water-level records were available. A null space Monte Carlo uncertainty analysis was
coupled with MODFLOW’s Groundwater Management (GWM) Process to develop a well-
field operating plan. The uncertainty analysis considered both the aquifer transmissivity and
the water quality (total dissolved solids (TDS) concentration) of each well. A stochastic field
generator was used to create 100 transmissivity realizations; each of these was calibrated
using PEST (Doherty, 2016). TDS concentrations were generated based on the statistical
properties of the observed TDS values in a subset of the vineyard’s wells. The GWM pro-
cess was used to optimize pumpage for each combination of calibrated transmissivity and
TDS concentrations. The result is a wellfield operating plan that minimizes drawdown and
blended-water TDS concentrations and meets the vineyard’s irrigation demand.

196
Corresponding author: Steven Peterson
U.S. Geological Survey
[email protected]

A New Model to Assess Groundwater Availability of the Northern High Plains

Aquifer

Steven M. Peterson, U.S. Geological Survey

J. Traylor

Abstract: The U.S. Geological Survey’s Water Availability and Use Science Program is
conducting an assessment of groundwater availability of principal aquifers to gain a clearer
understanding of the status of the Nation’s groundwater resources and the natural and
human factors that can affect those resources. One principal aquifer being studied is the
High Plains aquifer, a nationally important water resource that underlies about 174,000
square miles in parts of eight western states, of which the Northern High Plains constitutes
more than half. Droughts across much of the Northern High Plains from 2001 to 2007
have combined with recent (2004) legislative mandates to elevate concerns regarding future
availability of groundwater and the need for additional information to support science-
based water-resource management. The High Plains Groundwater Availability Study will
address these needs by providing tools and information for water-resource managers and
other stakeholders. The MODFLOW-NWT model of the Northern High Plains aquifer was
calibrated to 1940-2009 hydrologic data, using novel programming due to data handling
requirements. The largest simulated water budget components of the calibrated model
were recharge from precipitation, and outflows to stream base flow, though groundwater
withdrawals for irrigation rivaled other outflow terms near the end of the calibration model
period. Groundwater availability was subsequently analyzed with the model by assessing the
potential 2010-2050 impacts of selected climate or anthropogenic changes on groundwater
resources and inter-related stream base flows. This analysis revealed different sensitivities
of various major stream basins to the changes in land cover or climate.

197
Corresponding author: Edward Sudicky
Aquanty Inc.
[email protected]

River Basin-scale Integrated Surface-Subsurface Hydrologic Modeling to

Support Agricultural Risk Management
S. Frey
M. Callaghan
H-T Hwang
Y-J Park
R. McConnell
S. Berg
E. Sudicky

Abstract: Providing a scientific basis for water management policy, and assessing the phys-
ical characteristics underlying hydrologic risk, typically requires watershed-scale assessments
that encompass a few hundred km2 at a minimum. However, agriculture-focused water re-
sources challenges often encompass much larger areas, and can easily extend to major river
basins (>100,000 km2 ). Given the recent increase in crop losses attributed to large-scale ex-
treme climate related events (i.e. overland pluvial flooding, excess moisture, and drought),
and the concern that the frequency of these events will progressively increase in response
to climate change, there is growing demand for large-scale hydrologic risk assessments. Be-
cause of complex interactions between climate, surface water, groundwater, and soil moisture
across much of the agricultural landscape, robust physically-based 3-D integrated hydrologic
models provide a holistic means of performing water related risk assessment for these types
of applications. In this presentation we discuss a large-scale modelling-based agricultural
risk assessment project whereby fully-integrated surface/subsurface water models are be-
ing developed using the HydroGeoSphere platform for the South Saskatchewan River Basin
(SSRB, 150,000 km2 ) located in the predominately arid region of Western Canada. The
hydrologic responses within its major sub-basins are nested seamlessly within the full-basin
scale model in order to capture additional details at an increased resolution. Once complete,
the SSRB modelling platform will facilitate large-scale spatially-distributed projections of
flood, drought, and other water-related risks to crop production at unprecedented detail, over
temporal intervals ranging from days to decades. Results from this work demonstrate that
comprehensive physically-based hydrologic simulation platforms are becoming increasingly
relevant and feasible tools for addressing water related challenges facing global agriculture.

198
Corresponding author: Jonathan Traylor
U.S. Geological Survey
[email protected]

Estimating future irrigation and recharge for the Northern High Plains aquifer
using multiple climate and land use forecasts

Jonathan Traylor, U.S. Geological Survey

Abstract: The Northern High Plains aquifer underlies about 62 million acres in parts
of Colorado, Kansas, Nebraska, South Dakota, and Wyoming. The aquifer is an important
water resource that helps sustain the region’s agricultural economy, accounting for about half
of the groundwater use for irrigation from the High Plains aquifer. Future climate and land
use changes will affect the hydrologic system and may result in future water supply issues if
the system is managed with an inadequate understanding of these effects. In this study, a
Soil Water Balance model was used to assess the potential effects of future climate and land
use scenarios on crop irrigation requirements and groundwater recharge in the Northern High
Plains aquifer. Simulations from 2010-2050 included a base forecast of constant 2004 climate
and 2009 land use, projections from two climate models, and two land use change forecasts.
Future precipitation trends were not homogeneous within each climate forecast: there were
drier and wetter periods when compared to base conditions. The land use forecasts featured
increases in irrigated land. Results show that average crop irrigation requirements and
recharge, under the two climate forecasts, decreased up to 22 percent from base conditions,
and, under the two land use forecasts, increased up to 5 percent. Regression analysis showed
that precipitation has a strong influence on the temporal variability of recharge rates and to
a lesser extent irrigation rates, whereas changing land use has a small influence on recharge
and irrigation rates but a strong influence on irrigation volume.

199
Corresponding author: Enrique Triana
MWH
[email protected]

A Custom Modeling Tool to Support Evaluation of the

Agriculture-Groundwater Connection

Enrique Triana, MWH

C. Garner
G. Pohll
R. Carroll

Abstract: With advances in computation and specialized modeling platforms, Decision

Support Systems (DSS) that combine different types of models have become more frequently
developed and applied to inform water resource planning and management decisions. The
Walker Basin Decision Support Tool (DST) is a custom DSS designed to provide compre-
hensive modeling of the irrigation-stream-aquifer system. In the DST, two MODFLOW
groundwater flow models, a MODSIM surface water allocation model, and a field-based wa-
ter balance module are coupled to simulate interactions between surface water availability,
agricultural activities, aquifer response, and the stream-aquifer interaction. Because the
DST simulates the connection between agriculture, and surface and groundwater processes,
changes in agricultural activities can be simulated to assess the impact on natural flow
available to water right holders, reservoir operations, groundwater pumping, and aquifer
water levels. The DST modeling approach, calibration techniques and unique features will
be highlighted, as well as, discussion of the importance and potential applications of the
DST.

200
Corresponding author: E.J. Wexler
Earthfx Incorporated
[email protected]

Development and Application of an Irrigation Demand Module for the USGS

GSFLOW Model
E.J. Wexler, Earthfx Incorporated
P.J. Thompson
M.G.S. Takeda
S. Malott
S.J. Shifflett
J.D.C. Kassenaar

Abstract: The Whitemans Creek subwatershed in southwestern Ontario, Canada was iden-
tified as being ”moderately stressed” under drought conditions due to high agricultural water
demand. Agricultural activity comprises over 75ύ of the land use in the 400 km2 subwater-
shed. Total permitted pumping for irrigation exceeds 80,000 m/d; and while self-reporting
of water use indicates that actual takings are substantially lower, some under-reporting is
suspected based on summer streamflow response especially under drought conditions. An
integrated groundwater/surface water model of the Whitemans Creek subwatershed was
constructed using the USGS GSFLOW code (Markstrom et al., 2007) to evaluate avail-
able resources and assess drought sensitivity. To better capture the cumulative effects of
irrigation demand on groundwater and surface water flow, Earthfx modified the GSFLOW
code to simulate daily irrigation demand based on soil moisture conditions and crop re-
quirements at each identified farm parcel in the study area. Irrigation water, supplied by
wells or surface water diversions, was applied to farm parcels. Return flows are simulated
directly as increased overland runoff or interflow and are routed to streams or recharge to
groundwater. The module was tested in Whitemans Creek subwatershed. Over 4000 farm
parcels were represented, each with unique crop type and installed equipment based on data
from site surveys and satellite imagery. Irrigation module parameters were selected by an-
alyzing reported water takings for each crop type under average and dry year conditions.
A good match between reported and simulated irrigation demand was observed across the
study area. The calibrated model will be used to assess long-term drought effects within
the subwatershed, evaluate ecological flows, and gage the impact of favoring groundwater
pumping over surface water diversions on subwatershed sustainability.

201
Corresponding author: Kurt Zeiler
Brown and Caldwell
[email protected]

Updates and Improvements to Colorado’s South Platte Decision Support

System Alluvial Groundwater Model

Kurt Zeiler, Brown and Caldwell, Lakewood, Colorado

Andy Moore, Colorado Water Conservation Board, Denver, Colorado
Mary Halstead, Colorado Division of Water Resources, Denver, Colorado
Zach Wengrovius, Brown and Caldwell, Lakewood, Colorado
Jeff Weaver, Brown and Caldwell, Durango, Colorado
Matt Lindburg, Brown and Caldwell, Lakewood, Colorado
Paul Dupre, Brown and Caldwell, Lakewood, Colorado

Abstract: The South Platte Decision Support System Alluvial Groundwater Model (model)
was developed by the Colorado Water Conservation Board (CWCB) and the Colorado Divi-
sion of Water Resources (DWR) to be a planning-level tool for the alluvial aquifer associated
with the South Platte River and its tributaries. The model is designed to simulate ground-
water flow and groundwater/surface water interactions at a regional scale in the Basin and
was developed using a “Data-Centered Approach” to process information from DWR’s Hy-
droBase database, geographical information system (GIS) datasets, and other information.
The model was recently updated to extend the simulation period through 2012 with im-
proved setup and computation. MODFLOW-NWT was used rather than MODFLOW-2000
(which included the Doherty dry cell methodology). The Partition Stress Boundary (PSB)
capability of MODFLOW-CDSS was incorporated into MODFLOW-NWT to improve the
simulated water budget accounting, (Banta 2011). PSB facilitates “partitioning” of selected
MODFLOW packages; each partition is defined by a separate input file and each partition’s
volumetric water budget is tracked and reported individually. Several custom software codes
used as part of the Data-Centered Approach were improved in their efficiency, functionality,
and accuracy. These changes resulted in reducing the overall size of MODFLOW input files
from approximately 48 GB to 7.5 GB. The model calibration to observed groundwater levels
and estimated stream baseflows was improved in the update process.

Extended abstract: Methodology, findings, and implications

The South Platte Decision Support System (SPDSS) Alluvial Groundwater Model (model)
was originally completed in 2012 by the Colorado Water Conservation Board (CWCB) and
the Colorado Division of Water Resources (DWR) as a planning-level tool for the alluvial
aquifer system associated with the South Platte River and its tributaries (Brown and Cald-
well, 2017; CDM-Smith, 2013). SPDSS is a component of the broader Colorado’s Decision
Support Systems (CDSS) effort managed by CWCB and DWR. The model was designed
to simulate groundwater flow and groundwater/surface water interactions with streams at a
regional scale in the South Platte River Basin.
The model was recently updated with the primary objective of extending the simulation
period from 1950-2006 to 1950-2012 and to incorporate additional data collected since the
initial construction of the model. Additional objectives of the model update included upgrad-
ing the MODFLOW version to MODFLOW-NWT from MODFLOW-2000 with a previous
dry cell correction, updates to software tools used in the “Data-Centered Approach” to

202
model input file development, improved tracking of water budget components, and reducing
the model input file sizes to enhance the usability of the model (Niswonger et al., 2011;
Harbaugh et al., 2000, Doherty, 2001).
From its inception, the model has been developed using a Data-Centered Approach
to process information from DWR’s HydroBase database, GIS datasets, and other infor-
mation (Zeiler et al., 2017a). The Data-Centered Approach provides a consistent, repro-
ducible framework for producing model input files based on standard State-administered
data sources, easing the processes of extending the simulated period or updating the model
with additional or improved underlying data. The Data-Centered Approach includes several
custom software tools that are designed to map information from these data sets to model
input files. The Data-Centered Approach facilitates and largely automates the construc-
tion of boundary condition input files for several water budget components simulated in the
model.
Several of the Data-Centered Approach software tools, referred to as Data Management
Interfaces (DMIs), were updated during the model update process to improve consistency,
speed, efficiency, and functionality, including options for reducing the file sizes of MODFLOW
inputs (Brown and Caldwell, 2017). The geoprocessing scripts of the CDSS Toolbox and
the code of the SFR2 Generator were converted to ArcPy, and updated ArcToolbox user
interfaces for these ArcPy scripts were updated to include validation of user inputs before
execution. The Lateral Boundary Processor’s Fortran source code’s flow was improved and
made more efficient, reducing processing time by approximately one order of magnitude.
The State Pre-Processor (StatePP) produces MODFLOW input files and was updated to
reduce the sizes of these input files by specifying constant values instead of arrays where
appropriate and to optionally produce external binary files for array inputs. (Note that
additional software tool update descriptions can be found in the model update documentation
[Brown and Caldwell, 2017].)
Groundwater and surface water in the South Platte Basin are used for agricultural,
municipal and industrial (M&I) and other uses. The model includes several water budget
components that are commonly simulated using specified flow boundaries including agricul-
tural pumping, M&I pumping, precipitation recharge, irrigation return flow recharge, canal
seepage recharge, alluvial underflow/lateral boundary flows, reservoir seepage, water rights
augmentation pumping, recharge ponds, and bedrock-alluvial flows. Evapotranspiration
(ET) is simulated in the model for alfalfa, subirrigated meadow, and phreatophyte vegeta-
tion. Tracking these individual water budget components within the traditional construct of
MODFLOW package types is difficult because total well pumping is reported in the WEL
package outputs, total recharge is reported in the RCH package outputs, and total ET flow
is reported in the ETS package output (Banta, 2000).
To improve the tracking of multiple water budget components within MODFLOW, the
Partition Stress Boundaries (PSB) capability was developed by the USGS with funding from
CWCB as part of MODFLOW-CDSS (Banta, 2011). The PSB capability tracks water budget
components (“partitions”) within MODFLOW based on user-specified categories rather than
being constrained to categories based on MODFLOW package types. The MODFLOW-
CDSS source code for the PSB capability was added to MODFLOW-NWT and support for
optionally producing PSB-enabled input files was added as an option to StatePP (Zeiler
et al., 2017b). Simulation with PSB-enabled MODFLOW-NWT produced budget outputs

203
that were easily analyzed for each component. Additionally, the use of the PSB capability
simplified model input file construction, improved QC processes, and contributed to the
reduction of the input file sizes from approximately 48 GB to 7.5 GB through the ability
to assign constant values to some water budget components’ inputs rather than needing to
specify large input arrays and the ability to use MODFLOW package types strategically for
certain PSB partitions (Zeiler et al., 2017b).
The model is calibrated to observed groundwater-level data, total streamflows, and esti-
mated stream baseflows. Traditional model calibration activities (e.g., hydraulic conductivity
value adjustments) were limited during the model update. However, model calibration met-
rics generally improved with the model update. Of note is that prior to this update the
model included agricultural well pumping reduced to 80% of the values previously estimated
due to solution convergence issues caused by desaturation of model cells. DWR GIS staff
improved the data linking irrigation wells to irrigated parcels during the period between
the original completion of the model and this model update. The updated model now uses
the entire estimated agricultural pumping from the Data-Centered Approach without those
solution stability issues and with continued acceptable calibration.
Some preliminary model scenarios have been performed with the updated model. These
scenarios include simulations to assess the potential impacts of lined storage reservoirs in
the alluvium of the South Platte River and tributaries on groundwater levels and base-
flow, assess potential strategies for mitigating high groundwater conditions, and compare
using the model to assess streamflow depletion from well pumping versus those analytical
methods traditionally used in Colorado water rights administration. See the CDSS website
(http://cdss.state.co) for more information on acquiring CDSS datasets, tools, models and
documentation.

References

Banta, E.R. 2000. MODFLOW-2000, the U.S. Geological Survey Modular Ground-Water Model - Docu-
mentation of Packages for Simulating Evapotranspiration with a Segmented Function (ETS1) and Drains
with Return Flow (DRT1): U.S. Geological Survey Open-File Report 00-466, 127 p.
Banta, E.R. 2011. MODFLOW-CDSS, a version of MODFLOW-2005 with modifications for Colorado
Decision Support Systems: U.S. Geological Survey Open-File Report 2011-1213, 19 p.
Brown and Caldwell. 2017. South Platte Alluvial Groundwater Model Update Documentation: prepared
for the Colorado Water Conservation Board and Division of Water Resources, May.
CDM-Smith. 2013. South Platte Decision Support System Alluvial Groundwater Model Report: prepared
for the Colorado Water Conservation Board and Colorado Division of Water Resources, April.
Doherty, J. 2001. Improved calculation for dewatered cells in MODFLOW. Ground Water, v. 39, No. 6, pp.
863–869.

204
Harbaugh, A.W., Banta, E.R., Hill, M.C., and McDonald, M.G. 2000. MODFLOW-2000, the U.S. Geological
Survey modular ground-water model–User guide to modularization concepts and the Ground-Water Flow
Process: U.S. Geological Survey Open-File Report 00-92, 121 p.
Niswonger, R.G., Panday, S., and Ibaraki, M. 2011. MODFLOW-NWT, A Newton formulation for MODFLOW-
2005: U.S. Geological Survey Techniques and Methods 6-A37, 44 p.
Zeiler, K., Moore, A., Halstead, M., Wengrovius, Z., Weaver, J., Lindburg, M., Dupre, P. 2017. Use of a Data-
Centered Approach in updating Colorado’s South Platte Decision Support System Alluvial Groundwater
Model. In MODFLOW and More 2017: Modeling for Sustainability and Adaptation. Golden, Colorado,
USA: Colorado School of Mines.
Zeiler, K., Weaver, J., Moore, A., Halstead, M. 2017. Application of the Partition Stress Boundary capability
with MODFLOW-NWT in a large-scale regional groundwater flow model. In MODFLOW and More 2017:
Modeling for Sustainability and Adaptation. Golden, Colorado, USA: Colorado School of Mines.

205
Corresponding author: Caitlin Collins
Colorado School of Mines
[email protected]

Using an Integrated Hydrologic Model to Assess the Ecohydrologic Impacts of

Change on a Mountain Headwaters Critical Zone

Caitlin Collins, Colorado School of Mines

R. Maxwell
A. Visser

Abstract: The critical zone is the region of the Earth’s crust where hydrogeology, ecology,
and climate interact. As many critical zone processes are fundamental, the significance
of studying critical zone processes goes beyond understanding the local ecohydrological
setting. Therefore studying critical zone governing processes requires an interdisciplinary
approach that integrates simulation and observation. In this study, a high-resolution inte-
grated hydrologic model, ParFlow-CLM, was developed for the Providence Creek watershed.
Providence Creek is a highly instrumented critical zone observatory (CZO) located in the
southern Sierra Nevada Mountains, a region currently experiencing a range of short-term
responses (i.e. tree mortality) to a severe four-year drought. Sources of plant water use,
pathways and residence times of water through the subsurface are identified using a suite
of isotopic signatures and numerical particle tracking. Implications of using a fully coupled
integrated hydrologic model accompanied by tracer analysis include better understanding
of water partitioning and water storage in the regolith and vegetation water use during
drought time conditions. The importance of subsurface storage, plant available water and
lateral flow during the 2012-2015 drought to mitigate vegetation stress are addressed and
verified against observed tree mortality. The stream flow response to tree mortality in the
aftermath of the drought, analogous to the Colorado Mountain Pine Beetle case, provides
insight into the potential effects of proposed forest management practices.

206
Corresponding author: Bo Gao
Colorado School of Mines
[email protected]

Coupled Porous-Medium and Free-Flow Across Wavy Surfaces

Bo Gao, Colorado School of Mines

K. Smits, Colorado School of Mines

Abstract: The interaction of fluid flow in the porous medium and free-flow regions has
broad applications in various fields, e.g. water-gas management in fuel cells, multi-phase
flow through fractured-vuggy reservoirs, etc. Among the few previous studies of this coupled
porous-medium and free-flow behavior applied to evaporation from soil, only flat soil surfaces
were taken into account. However, wavy surfaces affect the airflow behavior and local mass
exchange between the atmosphere and the subsurface significantly. The goal of this work
is to give a preliminary exploration of the influence of the large-scale roughness to the flow
behavior near to the soil surface and evaporation by coupled porous-medium and free-flow
model. To do this, we modified a previously developed theory for two-phase porous-medium
flow, coupling it to single-phase free flow to simulate and analyze the evaporation from wavy
surfaces directly. In the free-flow region, both a laminar Navier-Stokes model and turbulent
RANS model are considered. Through the comparison, the phenomenon of flow separation
and reattachment can be clearly observed for the turbulent airflow across wavy surface.
Thus, it is important to consider a stepwise distribution of boundary layer thickness to
properly capture the evaporation rate for wavy surfaces. Vapor diffusion plays a significant
role in the mass exchange adjacent to the surface. Simulation results demonstrate the effects
of turbulence on the diffusive flux for wavy surfaces, and thus in turn to the evaporation. In
addition, the effect of aspect ratio of the wavy surface to the evaporation is also considered
in this study.

207
Corresponding author: Ahmad Jan
Oak Ridge National Laboratory
[email protected]

Evaluating a mixed-dimensional spatial structure for integrated thermal

hydrology models of permafrost-affected landscapes
Ahmad Jan, Oak Ridge National Laboratory
S. Painter, Oak Ridge National Laboratory
E.T. Coon, Los Alamos National Laboratory

Abstract: In general, a mixed-dimensional modeling approach connects a family of mutu-

ally independent one-dimensional vertical columns to a two-dimensional lateral flow system.
We have implemented a variant of this computationally advantageous modeling strategy
in the Advanced Terrestrial Simulator (ATS) to study permafrost dynamics near Barrow,
Alaska. In our sequential approach, we first solve an overland thermal hydrology system
with no sources and then a family of one-dimensional columns representing multiphase ther-
mal hydrology. The overland thermal hydrology system mainly acts as a spatial distributor
of the mass and energy, and updates the column systems before they advance in time.
The output of the column systems updates the overland thermal hydrology system for the
next iteration. This is a first attempt to couple recently developed representations of phase
partitioning in freezing, unsaturated soils with overland flow and surface energy balance
at scales of 100s of meters. We demonstrate the accuracy and efficiency of this scheme.
The scheme is highly scalable, supports subcycling of different processes, and compares well
with fully 3D representations, but is computationally less expensive. Further, it provides
a path to tractably represent thaw-induced subsidence. Although, we focus on permafrost
thermal hydrology near Barrow, Alaska, the spatial structure is applicable to many inte-
grated surface/subsurface hydrology problems at field scale. This work was supported by
the Interoperable Design of Extreme-scale Application Software (IDEAS) project and the
Next-Generation Ecosystem Experiments-Arctic (NGEE Arctic) project. NGEE-Arctic is
supported by the Office of Biological and Environmental Research in the DOE Office of
Science. This manuscript has been authored by UT-Battelle, LLC under Contract No. DE-
AC05-00OR22725 with the U.S. Department of Energy. The United States Government
retains and the publisher, by accepting the article for publication, acknowledges that the
United States Government retains a non-exclusive, paid-up, irrevocable, world-wide license
to publish or reproduce the published form of this manuscript, or allow others to do so, for
United States Government purposes. The Department of Energy will provide public access
to these results of federally sponsored research in accordance with the DOE Public Access
Plan (http://energy.gov/downloads/doe-public-access-plan).

208
Corresponding author: Zhen Li
Colorado School of Mines
[email protected]

Upscaling Thermal Properties in Heterogeneous Soils

Zhen Li, Colorado School of Mines

K. Smits, Colorado School of Mines

Abstract: Mass and energy transfer between the atmosphere and shallow subsurface near
the land surface is of critical importance to many natural and industrial applications. Be-
cause soil is naturally heterogeneous, soil hydraulic and thermal properties vary, resulting in
spatial and temporal differences in soil evaporation and heat transfer. However, capturing
this heterogeneity through discretization of areas with fine resolutions to capture the phys-
ical behavior on the larger scale in not feasible. Scaling is therefore needed to capture the
complexity of the behavior of the heterogeneous domain, especially in applications where
the sizes of grid cells are often much larger than the size of the laboratory scale samples on
which soil hydraulic and thermal properties are developed. Although there is much research
on the upscaling of hydraulic properties, the scaling of thermal properties has often been
overlooked. The goal of this work is to develop reliable upscaling methods for soil thermal
properties and verify these methods using numerical models. Estimates of thermal proper-
ties from readily available data at the lab scale was combined with techniques that allow
for upscaling at the larger scale, thus incorporation the rich information in the smaller scale
relationships or functions in larger scale correlations. Simulations were conducted on a va-
riety of heterogeneous domains including soil layers and random cells. Results indicate that
the evaporation behavior predicted by the model involving upscaled thermal conductivity
is consistent with the one accounting for heterogeneity, demonstrating the validity of this
method.

209
Corresponding author: Fred Ogden
University of Wyoming
[email protected]

The Soil Moisture Velocity Equation: Reliable, Accurate, and Fast 1-D
Unsaturated Zone Flux Calculations

Fred L. Ogden, University of Wyoming

M.B. Allen
C.C. Douglas
W. Lai
M. Seo
J. Zhu

Abstract: Coupling of the land-surface to groundwater table through the vadose zone
affects many important processes such as evapotranspiration, precipitation and energy par-
titioning at the land surface. The standard method for computing vadose zone water fluxes
is by solving the computationally expensive and unreliable Richards’ partial differential
equation, especially when sharp wetting fronts are present. For this reason, large-scale
hydrological and Earth System models frequently rely on conceptualizations to calculate
vadose zone fluxes. We have transformed the 1-D Richards’ equation into a new form that
describes the velocity of propagation of moisture content values in an unsaturated soil under
the actions of capillarity and gravity. In this transformed Richards’ equation, the position
of the wetting front is the dependent variable, and we call this equation the Soil Moisture
Velocity Equation (SMVE). The SMVE contains two distinct terms. The first advection-like
term represents the flux due to gravity and capillarity. The second diffusion-like term rep-
resents the flux due to capillarity divided by the vertical gradient of the capillary pressure
head. The power of the SMVE comes from the fact that the advection-like term can be
converted into an easy to solve ordinary differential equation (ODE). We compare against
exact analytical solutions of the Richards equation, which allows us to avoid questions re-
garding the validity of the Richards’ equation numerical solution. This presentation shows
that fluxes calculated using the method are 99ύ accurate while neglecting the diffusive
flux term using the ODE solution of the advection-like term with a finite moisture-content
solution method.

210
Corresponding author: Jessica Barros
TRC Solutions
[email protected]

Novel Modeling Approach for Sites with Complex, Well-Characterized Geology

Jessica Barros, TRC Solutions

M. Sellwood
A. Wilson
E. Schwartz
L. Hovey

Abstract: A groundwater model was generated to inform the design of a groundwater

remedy involving a slurry wall and extensive groundwater extraction well network. The
site is underlain by a gently sloping alluvial plain, where hydrogeologic zones consist of
mostly fine-grained low permeability units with localized high permeability regions that form
a heterogeneous and interbedded aquifer system. C Tech’s Mining Visualization System
(MVS) software was used to develop a comprehensive geologic model from 233 highly-
detailed boring logs. Initially, a 3D finite difference grid was created in MVS and imported
into Modflow USG. This version consisted of 56 layers, each representing a zone of uniform
hydraulic conductivity. An attempt was made to simulate the complex geology using the
MVS pinchout feature; however, this resulted in many discontinuous lenses and associated
model computation/converging challenges. Ultimately, it was determined that this approach
was not the best fit for modeling the site’s well-characterized, heterogeneous geology. The
final version of the model was constructed using a 3D rectilinear grid, generated in MVS,
which consists of approximately 120,000 active cells. MVS was used to assign one of four
hydrologic conductivity zones (sand with fines, clay, sand, or gravel) to each cell in the
model based on imported boring log data. Exporting the hydraulic conductivity zone data
from MVS and importing this information into Groundwater Vistas presented particular
challenges. Additionally, a boundary condition data gap was identified through model use
and resulted in additional field investigation and model refinement.

211
Corresponding author: Benny Bian
Questedge Systems International Corp.
[email protected]

HydrogeoCloud 3D Groundwater Modeling, a Cloud Computing Hydrogeology

Application

Benny Bian, Questedge Systems International Corp. 18-7091 Blundell Rd, Richmond, BC V6Y 1J5,
Canada
Wenpeng Li, China GeoEnvironmental Monitoring Institute, 20 Dahuisi Rd, Haidian District, Beijing,
China

Abstract: The China National Groundwater Monitoring Project will establish a nation-
wide groundwater monitoring network. 20,000 data loggers are being deployed, most of
which can automatically transmit up to 12 monitoring data points daily to data collection
systems. The project will generate large volumes of real-time groundwater monitoring data.
More frequently, large volume and multi-sourced data creates an unprecedented challenge to
traditional data processing tooling, including groundwater modeling in future hydrogeology
studies. A Scientific Computing Cloud Platform called HydrogeoCloud is developed which
takes advantage of the latest advancements of cloud computing and big data technology.
The platform wraps traditional software modules in any programming languages, such as
MODFLOW in FORTRAN code from USGS, into a standard cloud API and links it to
hydrogeological domain data. Computing steps are connected to form a computing work-
flow. The platform manages links between data and algorithms, and handles computing
resources allocation in the cloud environment. A 3D Groundwater Modeling Application
was developed on the platform. HydrogeoCloud is being used in a preliminary hydrogeo-
logical study in Heihe basin, Gansu Province, China. The cloud platform approach demon-
strates numerous advantages by filling gaps between database-oriented information systems
and traditional desktop applications. An integrated data and extendable tooling platform
makes it possible to create near real-time data processing, thus to speed up hydrogeologi-
cal decision making. In addition, HydrogeoCloud automatically produces provenance. The
platform can be used as a research platform to develop and to explore different algorithms
and as a project collaboration portal by sharing data and computing project in the cloud.

Extended abstract: Methodology, findings, and implications

Introduction

Current tools in scientific computing domain are incapable of dealing with nowadays data
exploration reality. Traditional desktop applications, either backed by a file structure or a
database, suffer from its proprietary and rigid data models. Consequently, it is difficult to
extend such applications, to incorporate additional data, or to integrate with other applica-
tions. Additional data sources such as real-time data must be transformed externally and
then imported to these applications. On the other hand, traditional enterprise information
systems are designed around databases. They suffer from lack of analytical tools beyond
build-in database functionalities. They are difficult to be integrated with existing analytical
tools due to those tools may be in different programming languages and thus requiring dif-
ferent execution environments. Furthermore, even for those tools with same programming
languages and compatible execution environments, they are likely needed to be customized

212
or be re-implemented to fit predefined data model. As a consequence, current enterprise
data workflows frequently force users to move data among different systems. Such workflows
may delay decision making or even cause errors in the processes.
This paper introduces HydrogeoCloud, a scientific computing platform. HydrogeoCloud
platform provides data integration in cloud, flexible semantic data modelling, algorithm as
a service, and computing workflow services. These services allow app developers quickly
developing new apps or extending existing apps, and also provide tools for the algorithm de-
velopers to easily add or explore new or existing algorithms into workflows without worrying
underlying computing infrastructures.

HydrogeoCloud Key Services

1. Application Domain Data modeling

A data model, According to Wikipedia, “is an abstract model that organizes elements
of data and standardizes how they relate to one another and to properties of the real world
entities”. All Applications organize their input and output datasets in certain structures.
These structures are essentially application data models. All sophisticated data processing
applications require well-defined data models, either implicitly or explicitly. For example,
a groundwater modeling application needs to make sure a piece of data, such as simulation
modeling area, is the right data, in right format, and with right units to be used in the
application. Applications logics are then linked with their data models.
Most software applications implicitly define their proprietary data models. Hydrogeo-
Cloud platform allows application developers defining domain data model explicitly. A do-
main data in HydrogeoCloud is defined by a domain data type and its optional data schema.
For example, in our 3D Groundwater app on HydrogeoCloud, a set of domain data types
like model domain, river boundary conditions, and MODFLOW package files are defined.
Not only individual domain data types can be defined, relationships among data types can
be created in logical or dependent structures.
Explicit domain data model defined for one application may be re-used for extending the
application and even for other applications in a similar problem domain. For example, a
pumping well data type (defined by its geometry and pumping schedule) may be used in a
groundwater modeling application, and in a pumping test application. Using an open data
model increases interaction possibilities among similar applications.

2. Algorithm as service API

API stands for Application Programming Interface. In HydrogeoCloud, platform APIs

are documented in open standard and are open to application developers. These APIs can
be called from services in web backend or from web applications in a browser.
HydrogeoCloud refers Algorithm as a piece of code, either as binary or in source code,
being runnable in pre-defined environments with specific inputs and outputs. HydrogeoCloud
platform can wrap almost any programming code into a web API. Once an API is called by
applications, HydrogeoCloud manages its life cycle and its computing resources in a cloud
environment. HydrogeoCloud platform currently has algorithms implemented in FORTRAN,
python, and Node js as web APIs.

213
 MODFLOW-USG code from USGS is a good example. MODFLOW-USG, written in
FORTRAN programming language, is compiled into target execution environment. A light
weight container (www.docker.com) is used to provide its execution compartment. The
HDFS (Apache Hadoop Distributed File System, hadoop.apache.org) is used as the comput-
ing file system. On top of containers and file systems, HydrogeoCloud cloud engine provides
additional services including runtime resource allocation, computing efficiency optimization,
and life cycle management.
Wrapping an algorithm as service API can be used to extend the platform with existing
and new algorithms.

3. Computing steps and computing workflow

On HydrogeoCloud, a computing step is formed by linking an algorithm (a service), and

its input and output datasets. Domain data model is used to semantically connect datasets
with an algorithm. Multiple computing steps are connected to form a computing workflow.
A computing step or an entire workflow can be exposed as platform API.
For example in 3D Groundwater modeling, the name file of MODFLOW-USG specifies all
input packages (files) and the instruction for simulation results. MODFLOW-USG program
requires a list of package files before execution. Each package file preparation can be regarded
as a computing step. To run MODFLOW-USG, one or more package translators service
components are responsible for generating package files from project datasets. An API
called “Translate-and-Run” is essentially a computing workflow which connects package file
translators and eventually run MODFLOW-USG simulator.
In addition, HydrogeoCloud workflow engine records data provenances for each comput-
ing step and workflow execution. Computing workflow breaks a complex application into
a set of easily manageable steps. It is possible, often quite desirable in practice, to replace
individual algorithm in a computing workflow. Thus the best algorithms are chosen to form
a computing workflow to meet computing objectives.

Results and Discussion

HydrogeoCloud 3D Groundwater modeling platform was demonstrated with datasets

from Heihe river basin in China (Figure 2). A real world project is currently carried out for
a large groundwater model in Beijing area. Through initial testing of integrating different
data in cloud and creating a complex groundwater model using HydrogeoCloud platform, we
believe HydrogeoCloud, as an integrated cloud data management, extendable and open API
platform and a collaborative app ecosystem, offers clear advantages than desktop applications
and traditional information systems.

214
 Figure 1. HydrogeoCloud Platform Key Services

Figure 2. Heihe River Basin Model Results Visualization

References

215
Corresponding author: Louis-Charles Boutin
Matrix Solutions Inc.
[email protected]

Canada’s Oil Sands Innovation Alliance Regional Groundwater Solutions

Project for the Southern Athabasca Oil Sands - Spatial Visualization of a
Global Transient Misfit Quality Indicator

Louis-Charles Boutin, Matrix Solutions Inc.

M. Claprood
G. MacMillan
M. Brewster
A. Walter

Abstract: The Canada’s Oil Sands Innovation Alliance is composed of 13 leading companies
with a common goal to accelerate innovation to reduce the impact of the oil sands devel-
opment on the environment. The Regional Groundwater Solutions project was established
to evaluate the potential range of change in aquifer pressures resulting from groundwater
withdrawals and disposal associated with future in situ bitumen production within Alberta’s
Southern Athabasca Oil Sands region. In 2016, Matrix Solutions Inc. was retained to under-
take FeFlow model update and computationally intensive coupled steady-state and transient
calibration using PEST software. Calibration targets for the inversion included; water table
depth, 209 measured hydraulic heads in industrial groundwater wells, 724 hydraulic heads
inferred from industry data, and 13 years of transient hydraulic head data that was reduced
to 21,782 monthly calibration targets. The calibration process required harnessing the power
of cloud computing, allowing for up to 100 model runs to be solved simultaneously for a
total of 3,310 hours. Through the inversion process, measured observations were compared
to simulated values (i.e. residual). During steady-state calibration, residuals’ mean and
standard deviation statistics are often used to quantify the precision and accuracy of the
calibration misfit quality. These statistics can also be used for transient calibration, al-
though, additional insight can be gained when exploring statics related to residual trends
versus time. This presentation describes an innovative approach to computing an arbitrary
Global Transient Misfit Quality Indicator (GTMQI) which allows visualization of the spatial
distribution of transient misfits, based on statistics.

216
Corresponding author: Norm Jones
Brigham Young University
[email protected]

HydroShare: A Community Repository for Sharing MODFLOW Models

Norm Jones, Brigham Young University

K. Liang
J. Goodall
M. Morsy
J. Sadler
D. Tarboton
A. Lemon
M. Kennard

Abstract: Groundwater modeling is a critical part of long-term water resource manage-

ment. The time and effort involved in building and calibrating a groundwater model can
be considerable, and there is value in sharing and re-using models developed by other
modelers whenever possible, or creating a new model by adapting an existing one. In
this paper we present a new framework for sharing MODFLOW models. HydroShare
(www.hydroshare.org) is an online collaborative environment for sharing hydrologic data
and models, established with NSF funding and operated by the Consortium of Universi-
ties for the Advancement of Hydrologic Science Inc. The supported resource types include
MODLOW model executables and MODFLOW model instances comprising all the files for
the application of a model at a site. Users can manage who has access to the resources they
share. HydroShare also supports an application programming interface (API) that allows
developers to build custom tools for sharing and accessing resources. As a case study, we
present a utility in GMS that allows users to publish MODFLOW models to HydroShare
directly from the application with a short series of clicks. The utility also allows users to
explore available HydroShare MODFLOW resources either in a map layout or a list. Once
a model is selected, it may be downloaded and opened in the application for execution or
editing.

217
Corresponding author: Dan Puddephatt
GHD
[email protected]

MFtools: An R Library for Reading MODFLOW and MT3DMS Files

Dan Puddephatt, GHD

Abstract: R is an open-source statistical programming language. Advancements in the

language are steering users towards a paradigm of tidy data: that in general terms data
should be organized in data-frames consisting of columns of variables and rows of obser-
vations. This paradigm has improved readability of code scripts and R library developers
are creating tools that enhance analysis and visualization. The MODFLOW family of code
input and output files are not written in this tidy format but rather in variously sized arrays
of values. The MFtools library was developed for reading the input and output files from
MODFLOW and MT3DMS model simulations so that they can be evaluated using the suite
of utilities available through R. This means that users no-longer need to develop dedicated
one-off fortran/C processors for evaluating their models in detail. Once a file has been read
into R using MFtools features of the MODFLOW/MT3D can be queried rapidly and iter-
atively using tidy-based R packages. These queries can be used to isolate zones of interest,
calculate statistics, and transform variables. Users can use these tools to prepare figures il-
lustrating model properties such as predicted concentration, groundwater elevation changes,
or vertical/horizontal gradient changes over time. These tools are particularly useful when
evaluating continuous distributions of hydraulic properties following pilot point parameter
estimation. MFtools is designed to simplify model analysis so that more time can be spent
interpreting results. These tools are also particularly useful for quickly comparing various
versions of a model including the respective model results.

218
Corresponding author: Kurt Zeiler
Brown and Caldwell
[email protected]

Use of a Data-Centered Approach in Updating Colorado’s South Platte

Decision Support System Alluvial Groundwater Model
Kurt Zeiler, Brown and Caldwell, Lakewood, Colorado
Andy Moore, Colorado Water Conservation Board, Denver, Colorado
Mary Halstead, Colorado Division of Water Resources, Denver, Colorado
Zach Wengrovius, Brown and Caldwell, Lakewood, Colorado
Jeff Weaver, Brown and Caldwell, Durango, Colorado
Matt Lindburg, Brown and Caldwell, Lakewood, Colorado
Paul Dupre, Brown and Caldwell, Lakewood, Colorado

Abstract: The South Platte Decision Support System (SPDSS) Alluvial Groundwater
Model (model) was developed by the Colorado Water Conservation Board (CWCB) and the
Colorado Division of Water Resources (DWR) as a planning-level tool for the alluvial aquifer
associated with the South Platte River and its tributaries. SPDSS is a component of the
broader Colorado’s Decision Support System (CDSS) effort managed by CWCB and DWR.
The model is designed to simulate regional-scale groundwater flow and groundwater/surface
water interactions and was developed using a “Data-Centered Approach” to process infor-
mation from DWR’s HydroBase database, geographical information system (GIS) datasets,
and other information. The Data-Centered Approach is a key concept used throughout
development of all CDSS surface water and groundwater models. The Data-Centered Ap-
proach is designed to map the datasets that are managed by the State to the groundwater
flow models, including this model. The system is designed to allow for efficient periodic
updates to the model as State databases are updated. Several custom software codes are
used as part of the Approach, and many of these codes were updated and improved as part
of the most recent model update effort. These codes take data from the State databases and
GIS systems and format them for use in the model, which is based on MODFLOW-NWT.
Updates to pumping, recharge, and evapotranspiration (ET) input files are streamlined to
allow for efficient testing of regional management options.

Extended abstract: Methodology, findings, and implications

The South Platte Decision Support System (SPDSS) Alluvial Groundwater Model (model)
was completed in 2012 by the Colorado Water Conservation Board (CWCB) and the Col-
orado Division of Water Resources (DWR) as a planning-level tool for the alluvial aquifer
system associated with the South Platte River and its tributaries (Brown and Caldwell, 2017;
CDM-Smith, 2013; Zeiler et al, 2017a). SPDSS is a component of the broader Colorado’s
Decision Support System (CDSS) effort managed by CWCB and DWR. This model was
recently updated to use MODFLOW-NWT and extend the simulation period (Niswonger et
al., 2011). The model was designed to simulate groundwater flow, interaction with agricul-
tural processes and other uses, and groundwater/surface water interactions with streams at
a regional scale in the South Platte River Basin.
From its inception, the model has been developed using a “Data-Centered Approach”
to process information from DWR’s HydroBase database, geographical information system

219
(GIS) datasets, and other information. The Data-Centered Approach is a key concept used
throughout development of all CDSS surface water and groundwater models. The Data-
Centered Approach provides a consistent, reproducible framework for producing model input
files based on standard State-administered data sources, easing the processes of extending the
simulated period or updating the model with additional or improved underlying data. Several
data sources are used for the developing model inputs, and DWR’s HydroBase database is
the primary repository of these data. HydroBase is a central database that stores historic and
geographic data related to climate, water rights, stream diversions, well pumping, streamflow,
and groundwater levels. Additional geographic data are housed in State-administered GIS
files, including historic “snapshots” of irrigated agricultural lands and related data, soils,
land use, streams, and canals.
The Data-Centered Approach includes several custom software tools that are designed
to map information from these data sets to model input files. The Data-Centered Approach
facilitates and largely automates the construction of boundary condition input files for sev-
eral water budget components simulated in the model. These tools, referred to as Data
Management Interfaces (DMIs), include:

• Proc_rainfall and Proc_runoff – Fortran-based utilities that process precipitation

data for estimates of precipitation-based groundwater recharge and surface runoff to
streams.

• State Consumptive Use Model (StateCU) – Fortran-based tool that estimates

historic agricultural consumptive use based on climate, diversions/pumping, irrigated
area, irrigation practice, and cropping patterns.

• State Data-Centered Ground Water Interface (StateDGI) – A geodatabase

that houses GIS data sets that have been intersected with the model’s MODFLOW
grid and contains several linked queries that produce files to be used as input to
StatePP.

• State Pre-Processor (StatePP) – Fortran-based utility that reads inputs generated

by StateCU, StateDGI, and other utilities, processes the data, and produces MOD-
FLOW input files formatted for the WEL, RCH, ETS packages (Banta, 2000).

• CDSS Toolbox – ArcToolbox file containing links to several ArcPy-enabled Python

scripts for processing and uploading GIS data to a StateDGI geodatabase.

• SFR2 Generator – ArcPy script that writes an SFR package input file based on the
stream network represented in a geodatabase, including flows and diversions.

• Lateral Boundary Processor – Fortran-based utility that estimates flows at the

edges of the active MODFLOW domain for flows across the active model domain from
connected materials (e.g., groundwater in lower permeability alluvium and colluvium).

• Bedrock Flux Processor – Set of Fortran-based utilities, including Grid2Grid (Banta

et al., 2008), that estimates flows between the alluvium and bedrock based on the USGS
Denver Basin Model (Paschke, 2011).

220
 While these tools provide a consistent framework for translating data to MODFLOW
input files, they are not static and rigid. During the latest model update, several of these
tools were improved:

• StateDGI – queries were updated to remove references to older HydroBase data fields,
to improve consistency in apportioning irrigated acreage and well pumping for situ-
ations where more than one well provides irrigation water to a single parcel or more
than one parcel receives irrigation water from a single well, and to provide a consistent
set of queries to divide well pumping between multiple model layers.

• StatePP was updated to optionally produce binary arrays or specify constant array
values as appropriate to reduce the sizes of RCH and EVT/ETS package input files
and to produce multiple WEL, RCH, EVT/ETS package input files for use with the
Partition Stress Boundary (PSB) capability including representing spatially discrete
specified-flux recharge sources (e.g., canal seepage or recharge ponds) using the WEL
package to reduce input file size (Banta, 2011; Zeiler et al. 2017b).

• CDSS Toolbox – the original arcgisscripting scripts were converted to the more mod-
ern ArcPy syntax, and user input validation routines were added to the ArcToolbox
interfaces.

• SFR2 Generator – converted from VB.NET code to an ArcPy script with an Arc-
Toolbox interface.

• Lateral Boundary Processor – code flow was made more efficient, reducing pro-
cessing time by approximately one order of magnitude.

Prior to this update the model included agricultural well pumping reduced to 80
The Data-Centered Approach provides a consistent and reproducible framework for per-
forming simulations based on openly accessible data sets. The Data-Centered Approach is
applied throughout the CDSS process for both groundwater and surface water modeling ef-
forts to provide accepted and dependable results. While the approach provides a framework
and a set of tools, the mechanics of those tools are not static and rigid. During the re-
cent update of the SPDSS Groundwater Model, several Data-Centered Approach tools were
improved to provide additional functionality and performance (Brown and Caldwell, 2017;
Zeiler et al., 2017a).
See the CDSS website (http://cdss.state.co) for more information on acquiring CDSS
datasets, tools, models and documentation.

221
 References

Banta, E.R. 2000. MODFLOW-2000, the U.S. Geological Survey Modular Ground-Water Model - Docu-
mentation of Packages for Simulating Evapotranspiration with a Segmented Function (ETS1) and Drains
with Return Flow (DRT1): U.S. Geological Survey Open-File Report 00-466, 127 p.
Banta, E.R. 2011. MODFLOW-CDSS, a version of MODFLOW-2005 with modifications for Colorado
Decision Support Systems: U.S. Geological Survey Open-File Report 2011-1213, 19 p.
Banta, E.R., Paschke, S.S., Litke, D.W. 2008. User guide and documentation for five MODFLOW ground-
water modeling utility programs: U.S. Geological Survey Techniques and Methods 6–A28, 24 p.
Brown and Caldwell. 2017. South Platte Alluvial Groundwater Model Update Documentation: prepared
for the Colorado Water Conservation Board and Division of Water Resources, May.
CDM-Smith. 2013. South Platte Decision Support System Alluvial Groundwater Model Report: prepared
for the Colorado Water Conservation Board and Colorado Division of Water Resources, April.
Niswonger, R.G., Panday, S., and Ibaraki, M. 2011. MODFLOW-NWT, A Newton formulation for MODFLOW-
2005: U.S. Geological Survey Techniques and Methods 6-A37, 44 p.
Paschke, S.S. (Ed.). 2011. Groundwater Availability of the Denver Basin Aquifer System, Colorado: U.S.
Geological Survey Professional Paper 1770, 306 p. Model data available at: http://pubs.usgs.gov/pp/1770/
Zeiler, K., Moore, A., Halstead, M., Wengrovius, Z., Weaver, J., Lindburg, M., Dupre, P. 2017. Updates
and improvements to Colorado’s South Platte Decision Support System Alluvial Groundwater Model. In
MODFLOW and More 2017: Modeling for Sustainability and Adaptation. Golden, Colorado, USA: Colorado
School of Mines.
Zeiler, K., Weaver, J., Moore, A., Halstead, M. 2017. Application of the Partition Stress Boundary capability
with MODFLOW-NWT in a large-scale regional groundwater flow model. In MODFLOW and More 2017:
Modeling for Sustainability and Adaptation. Golden, Colorado, USA: Colorado School of Mines.

222
Corresponding author: Lynette Brooks
U.S. Geological Survey
[email protected]

Groundwater Development and Surface Water: Capture Maps of Streamflow

Depletion

Lynette E. Brooks, U.S. Geological Survey

Abstract: Groundwater and surface-water resources are often connected. Much has been
written about the effects of groundwater development on surface water, and new tools are
available to simulate and quantify these effects. In this study, a numerical groundwater flow
model was used to prepare capture maps to illustrate the connection between groundwater
development and surface water. Capture maps show the amount of surface-water depletion
caused by a well at any location in the model domain and can be used by water managers
to assess the impacts of groundwater development. The first part of this presentation gives
a brief synopsis of the source of water to wells and common misconceptions about surface-
water depletion. The second part presents the development and analysis of capture maps
for a basin dominated by surface-water features. Rather than analyzing depletion in only
one river or spring, this work uses one set of capture simulations to determine the effects
of groundwater development on 11 segments along two rivers, 30 springs, 13 areas of field
drains, and 3 areas of evapotranspiration of groundwater. The presentation concludes with
a discussion of the implications of the capture maps and suggestions on how this type of
analysis can be used by water managers.

223
Corresponding author: Brian Clark
U.S. Geological Survey
[email protected]

The use of non-groundwater software in visualization and presentation of

groundwater-flow modeling studies

Brian Clark, U.S. Geological Survey

Abstract: Groundwater-flow models are often presented in a building-block manner to tell

the story of a groundwater system. These building blocks begin with a background of the
study area, in-depth description of the hydrogeology, and a detailed outline of the conceptual
model inflows and outflows of the system. The model report culminates in the technical
aspects and assumptions of transforming the conceptual model into computer code, followed
by a description of how closely the code mimics reality, and presents the results of some,
hopefully, useful scenarios. A question arises from this building block approach: are there
examples of other model-based results that more succinctly relay the important outcomes
to stakeholders yet still retain enough documentation for scientific reliability? Alongside
ongoing development of various models, exploration of such outlets are explored, along with
open-source software more often used in video game creation or visualization of interstellar
cloud data. Through these other software solutions, it is possible to evaluate groundwater
model input data, often in a rapid manner to uncover potential flaws. Likewise, the results
from the groundwater models, particularly transient results, may be revealed in much more
intuitive ways to those not so familiar with subsurface-fluid flow. A simple example is
a three-dimensional animation of a given water surface as it changes through time from
pumping stresses. Another benefit is the ability to easily distribute the animations, images,
and other formats to multiple media outlets. Advances and standardization of web-based
protocols assist this effort as well, allowing interactive user experience to explore selected
facets of the model.

224
Corresponding author: Kariem A. Ghazal
Department of Natural Resources and Environmental Management, University of Hawaii at Manoa,
Honolulu, HI 96822, USA
[email protected]

Modeling fresh submarine groundwater discharge across the Heeia coastal

shoreline in Hawaii
Kariem A. Ghazal, Dept. of Natural Resources and Environmental Management, University of Hawaii at
Manoa, Honolulu, HI, 96822, USA
Olkeba Tolessa Leta, Water Resources Research Center, University of Hawaii at Manoa, Honolulu, HI
96822, USA
Aly I. El-Kadi, Dept. of Geology and Geophysics, and Water Resources Research Center, University of
Hawaii at Manoa, Honolulu, HI 96822, USA
Henrietta Dulai, Dept. of Geology and Geophysics, and Water Resources Research Center, University of
Hawaii at Manoa, Honolulu, HI 96822, USA

Abstract: Assessing the fresh submarine groundwater discharge (FSGD) in the Heeia
Coastal shoreline of Hawaii via integrated hydrological modeling approach, is needed to un-
derstand the comprehensive relationships between coastal hydrological processes and ecosys-
tems. For this, the MODFLOW and SWAT models were used and integrated whereby the
SWAT model provided estimated recharge input to MODFLOW model to estimate FSGD.
The FSGD was assessed under three scenarios: Heeia wetland restoration (LU), climate
change (CL), and sea level rise (SLR). The current daily average of Heeia coastal FSGD
was about 0.43 m3/d/m, but expected to decrease by about 20 percent by the end of 21st
century due to the combined effect of LU, CL, and SLR. Overall, findings showed that
FSGD was significantly influenced by the CL and SLR especially by the end of 21st cen-
tury. Moreover, the FSGD fluxes will decline more during the dry season compared to the
wet season. This study demonstrated the ability of integrated modeling approach to assess
the temporal distribution of FSGD across coastal shoreline for various scenarios.

Extended abstract: Methodology, findings, and implications

The Hawaiian coastal zones represent the essential natural resources for Hawaiian economy
and overall ecological health. The Heeia coastal zone in Hawaii is a typical example of
groundwater dependent ecosystems due to the presence of boundary interface among the
Island of Oahu’s largest fishpond, the largest federally designated wetland, and the Kaneohe
Bay’s greatest sheltered water body of coral reefs (Dulai et al., 2016; Jokiel, 1991). Assess-
ing the fresh submarine groundwater discharge (FSGD) across the Heeia Coastal shoreline
via integrated hydrological modeling approach, is needed to understand the comprehensive
relationships between coastal hydrological processes and ecosystems. Fresh water flows play
vital roles in preserving the native adjacent ecosystems. The positive impact of FSGD on
marine ecosystems is achieved through decreasing the salinity level and altering the pH
gradient in coastal waterways, which enables the organisms to maintain their sustainable
productivity (Duarte et al., 2010; Libes, 2011; McGowan, 2004). In addition, FSGD is a
significant hydrological process and an important component of water budget that trans-
ports considerable amount of fresh water and dissolved nutrients from coastal aquifer to the

225
nearshore marine environment. The transport of nutrients highly depends on the amount of
aquifer groundwater storage and head that determine the hydraulic flow gradient from land
to ocean (Burnett et al., 2006; Dulaiova et al., 2015; Moore et al., 2009). Estimating FSGD
is very challenging due to the fact that the flow magnitude can vary spatially and temporally,
which in turns depend on the site hydrogeological framework features, climate variability,
and human activities. For this, the MODFLOW and SWAT models were used and integrated
whereby the SWAT model provided estimated recharge input to MODFLOW model. Then
MODFLOW estimated FSGD and assessed the temporal distribution of FSGD across the
coastal shoreline of Heeia. The FSGD was assessed under three scenarios: Heeia wetland
restoration (LU), climate change (CL), and sea level rise (SLR) effects. The current daily
average of Heeia coastal FSGD was about 0.43 cubic meter per day per meter of coastal line,
which increased by 15 percent during the wet season but decreased by 16 percent during the
dry season due to the temporal variation of FSGD (Figure 1). The results showed that the
ocean got fresh groundwater more than the surface water. The average FSGD comprised 18
percent of baseflow, 11 percent of recharge, and 3 percent of rainfall. Under transient condi-
tions, the findings showed that the FSGD was significantly influenced by the SLR, recharge
rate, groundwater head at the coastal aquifer, and CL, especially at the end of the twenty
first century. The respective average annual decline in FSGD fluxes during the scenarios of
sea level effect, were 0.5 percent, 2 percent, and 5 percent due to increasing the sea level by
0.12 m and 0.4 m for the midcentury and 1.1 m for the end of 21st century. The average
decline in FSGD fluxes during the scenarios of LU, CL and combined LU, CL, and SLR (1.1
m) effects, were 1.6 percent, 10 percent, and 20 percent, respectively (Figure 2). Overall,
findings showed that FSGD was significantly influenced by the CL and SLR especially by
the end of 21st century. Moreover, the FSGD fluxes will decline more during the dry season
compared to the wet season. This study demonstrated the ability of integrated modeling
approach to assess the temporal distribution of FSGD across coastal shoreline for various
scenarios.

Figure 1: The daily estimated FSGD across the Heeia coastal shore line.

226
2.jpg 2.bb

Figure 2: The daily percent change relative to the baseline in FSGD duration curve under different scenarios
of LU, CL, and combined of LU, CL, and SLR by 1.1 meters effects.

References

Burnett, W.C. et al., 2006. Quantifying submarine groundwater discharge in the coastal zone via multiple
methods. Science of the Total Environment, 367(2): 498-543.
Duarte, T.K., Pongkijvorasin, S., Roumasset, J., Amato, D., Burnett, K., 2010. Optimal management of a
Hawaiian Coastal aquifer with nearshore marine ecological interactions. Water Resources Research, 46(11).
Dulai, H., Kleven, A., Ruttenberg, K., Briggs, R., Thomas, F., 2016. Evaluation of Submarine Groundwater
Discharge as a Coastal Nutrient Source and Its Role in Coastal Groundwater Quality and Quantity, Emerging
Issues in Groundwater Resources. Springer, pp. 187-221.
Dulaiova, H., Kleven, A., Ruttenberg, K., Briggs, R., THOMAS, f., 2015. Evaluation of submarine ground-
water discharge as coastal nutrient source and its role in coastal groundwater quality and quantity.
Jokiel, P.L., 1991. Jokiel’s illustrated scientific guide to Kaneohe Bay, Oahu. Hawaiian Coral Reef Assess-
ment and Monitoring Program, Hawaii Institute of Marine Biology, Kaneohe, Hawaii.
Libes, S., 2011. Introduction to marine biogeochemistry. Academic Press.
McGowan, M.P., 2004. Submarine groundwater discharge: freshwater and nutrient input into Hawaii’s
coastal zone.
Moore, W., John, H., Steve, A., 2009. Submarine Groundwater Discharge Encyclopedia of Ocean Sciences.
Academic, Oxford: 4530-4537.

227
Corresponding author: Keith Halford
U.S. Geological Survey
[email protected]

Complex Models Require Looking and Comparing

Keith Halford, U.S. Geological Survey

Abstract: Increasing complexity of groundwater models demands greater scrutiny of re-

sults for reasonableness. Simulated water levels, discharges, and flow budgets are typical
model results that can be sampled discretely and compared directly to measured quanti-
ties. These quantities typically are examined and reported because they are included in
an objective function for parameter estimation. Model input also is a result because esti-
mated parameters interact and require manipulation to become model input. For example,
hydraulic conductivities assigned to pilot points must be interpolated to model nodes to
become model input. Interactions between many pilot points and less visible specifics of the
interpolation process make comprehension difficult without mapping distributed hydraulic
conductivities. Processing into transmissivity further aids comprehension because thickness
variations are integrated in mapped distributions. Vertical variations are differentiated by
mapping transmissivities by hydrogeologic units and model layers in addition to total trans-
missivity. Mapped distributions and graphical representations reveal misfits between con-
ceptual and numerical models that are difficult to quantify in objective functions. Mapped
distributions, pilot points, residuals, hydrographs, well construction, lithologic columns, and
links to supporting data can be distilled and displayed rapidly in Google Earth. Model re-
sults are better understood by scientists and managers where results are geo-referenced in
a relatively simple interface.

228
Corresponding author: Martin Helmke
West Chester University
[email protected]

Breaking the Mould: Using Streamlined, Stochastic Groundwater Models to

Improve Acceptance by Regulators, Clients, and the General Public

Martin F. Helmke, West Chester University

T. Farney
D. Lynch
P. Girafalco
L. Cole
A. Pfahler

Abstract: Computer simulations of groundwater flow and contaminant fate and transport
have revolutionized our understanding of hydrogeology. The utility of these sophisticated
tools is undeniable, yet professional hydrogeologists rarely employ the full power of these
techniques, opting to use antiquated 1-dimensional solutions instead. Although complex
3-dimensional models are generally accepted for water supply and wellhead protection stud-
ies, regulators appear reluctant to admit such simulations for contaminant transport and
remediation. Moreover, clients are hesitant to commit funds for detailed modeling stud-
ies and the general public struggles to understand model results. We propose the use of
streamlined, stochastic simulations to encourage acceptance of modeling for hydrogeologic
investigations. Limiting models to under 300 cells provides simulations that are efficient to
run and construct (i.e., require fewer billable hours), easier to validate, yet significantly more
powerful than 1-dimensional deterministic solutions. Stochastic simulations are appealing
to regulators and the general public who are familiar with probabilistic risk assessment and
weather reports, respectively. We applied this technique using MODFLOW and the Monte
Carlo method to a variety of case studies, including production well influence on stream flow,
water table mounding beneath a wastewater drip field, and contaminant fate and transport
from a hazardous waste landfill. Clients, regulators, and the public appreciated the efficient
and transparent nature of these simulations, opening the door to increased use of these
methods.

229
Corresponding author: Howard Reeves
U.S. Geological Survey
[email protected]

Regional groundwater-flow modeling to assess tradeoffs and consequences of

groundwater development

Howard Reeves, U.S. Geological Survey

Abstract: Research teams within the U.S. Geological Survey (USGS) have been developing
regional groundwater-flow models focused on principal aquifers of the USGS to help ana-
lyze regional groundwater availability. These flow models are usually fairly coarse, cover
from 20 to over 400 thousand square kilometers, and tend to have long stress periods that
span pre-development to recent conditions. The regional models are designed to quantify
the development of groundwater systems and their response to land-use changes and cli-
mate variability. Although these models are usually too coarse to directly inform local
decisions, they can be a powerful way to illustrate the consequences of human activities on
the groundwater system and to inform decision makers about the tradeoffs or constraints
that may be considered when evaluating groundwater sustainability. Notably, examina-
tion of results within and between studies often contrasts the capture of surface water to
storage depletion that may lead to consolidation or migration of poor-quality water. In
addition to the direct benefit of quantifying groundwater systems, these projects have pro-
duced multiple data sets including regional estimates of recharge, time series estimates of
groundwater withdrawals, and enhanced hydrogeologic framework information. Stakehold-
ers may use these data in the context of tradeoff information for decision making. Finally,
these regional models have contributed testing opportunities for many recent MODFLOW
developments including history-matching and forecasting, python-based model development
tools, and data-worth evaluation.

230
Corresponding author: Blythe Reiha
The Regional Municipality of York
[email protected]

Beyond the Model: Communicating Results to the Decision Maker

Blythe Reiha, The Regional Municipality of York

Abstract: Evolving the numerical hydrogeologist’s role as ”modeler” to ”big-picture com-

municator” is paramount for municipal water supply decision makers. Opportunities for
completing environmental decisions (involving groundwater flow systems), through sim-
ple application of standard engineering and scientific principles, still exist; however, many
present-day challenges go beyond qualified rationalization and reasoning, and are best ad-
dressed in conjunction with a numerical hydrogeologic modeling approach. Over the last
several decades, both the private and public sectors have applied hydrogeologic models for
interpreting, assessing, and predicting the dynamics of complex groundwater flow systems.
The scientific connection between anthropogenic usage and environmental impacts (aquifer
depletion, water quality impacts, etc.), require decision makers (i.e., the ”non-modeler”) to
consider all facets of the water balance system. This phenomenon, in parallel with increased
computing capabilities, has played a role in advancing the frequency and extent of ground-
water modeling application in the industry. As the modeling processes, including its tools,
increase in complexity, often multi-disciplined in nature, it is imperative to ensure that
model results are translated into understandable and applicable ”messages” for the decision
maker. This paper highlights modeling project challenges experienced by municipal water
supply decision makers. Effective communicative approaches are emphasized, in addition
to lessons learned.

231
Corresponding author: Harsh Vardhan Singh
Physical Scientist
[email protected]

Using MODFLOW and MODPATH to access the changes in hyporheic flow

during pre and post-restoration periods

Harsh Vardhan Singh, Physical Scientist

B. Faulkner
A. Keeley
K. Forshay

Abstract: A groundwater model was setup using MODLFOW to study the changes in
hyporheic flow due to floodplain restoration activities (specifically levee setback) in the Gap-
to-Gap region of Yakima, WA. The model was developed as a single layer unconfined aquifer
discretized into 220 rows and 660 columns. The packages used to setup the model included
hydrogeologic unit-flow, river, barrier, drain, recharge and evapotranspiration. The model
was calibrated in steady and transient states using head data from five observation wells
located near the Yakima River. The Light Detection and Ranging (LiDAR) dataset from
2008 and 2013 were used to represent the topographic conditions before and after the levee
setback. The pathlines for hyporheic flow were developed using MODPATH by generating
particles at the cells adjacent to the Yakima River. The results from model simulations
indicated that the flow directions of the hyporheic flow pathlines changed after the levee
setback. The results also showed that during both steady and transient states, length of
the hyporheic pathlines increased significantly (p =< 0.001) after the levee setback. This
presentation does not necessarily reflect EPA policy.

232
Corresponding author: Peter Vermeulen.
iMod Software
[email protected]

Interactive Pathline Simulation

Peter Vermeulen, iMod Software

Abstract: Complex models are constructed using sophisticated software packages which
make it possible to enhance the model input for each grid cell and create complex patterns
of groundwater flow behavior. A common method to determine insight in flow patterns is
the usage of particle tracking algorithms in which particles are released and traced through
a 3-D velocity field. It results in a static 3-D image of the flow pattern often colored by
travel times. We developed another type of presentation to visualize their instant results
in a more dynamic way. We load the velocity fields and the subsoil characteristics into
the 3-D environment of iMOD (http://oss.deltares.nl/web/imod). Here, we can allocate
and visualize the start locations for particles in a variety of manners, such as around wells,
spatial distributed within model layers and/or within a 3-D plane that can be stretched and
moved in 3-D. Groups of particles can be organised separately such that they can be turned
on or off or change in trace direction independently of other groups. By simply clicking the
”Start” button, particles are released and computed. Instead of writing the tracking results
to file the results are transferred to OpenGL and presented in 3-D, instantaneously. The
particles act as moving objects than can repeat itself endless whenever captured by a sink or
boundary. The interesting part of this is that differences in flow velocities are clearly visible,
or more-over, unexpected behavior can be discovered efficiently as the particle movement
can be inspected from all directions.

233
Corresponding author: Sophia Wolfenden Lee
Lytle Water Solutions, LLC
[email protected]

Difficulties in Modeling Systems with Extreme Climate Records: Is a Model of

a Groundwater System During a Drought Unrepresentative or Conservative?

Sophia A. Wolfenden Lee, Lytle Water Solutions, LLC

B. Bader, Lytle Water Solutions, LLC

Abstract: The hydrology of an alluvial groundwater system can be heavily dependent

on the temporal climatology of a region. In particular, agricultural basins that have been
impacted by regional drought conditions, evolving irrigation techniques, and water use re-
appropriations can require a recalibration of groundwater models-both conceptually and
quantitatively-before future predictions can be attempted. Determining the appropriate
conditions for such recalibrations can result in complications with a models ability to both
represent a fixed past and predict a changing future. Utilizing a historically agricultural
region in Colorado, USA as a case study, an investigation was undertaken as to how different
groundwater modeling techniques can affect the operations of future use change and its
impacts on regional water users. Utilizing a steady-state MODFLOW model, an averaged
transient MODFLOW model, and a climatically variable transient MODFLOW model, an
attempt was made to determine if modeling a system in flux should be considered merely
conservative or unrepresentative for the purposes of future demands and impacts.

234
Corresponding author: Ahmed Ali
University of California, Davis
[email protected]

Upscaling the equilibrium reactions in the fractal radial domains

Ahmed Ali, University of California, Davis

T. Ginn
T. Le Borgne

Abstract: Non-horizontal flow in radial domains (i.e. well injection case) can result from
aquifer structure non-uniformity and/or non-fully penetrating well screen. The flow in such
a case is neither 2-D nor 3-D but rather may be described as a fractional radial flow (i.e.
2.4-D for example). The focus here is on upscaling the bimolecular calcite precipitation
equilibrium reaction in the fractal radial domains by means of the lamella approach. This
approach counts on identifying the time-dependent geometry of the mixing front on which
the effective reaction takes place. The fractal radial flow problem is reviewed, the lamella-
based solution is summarized, and comparative simulations between explicit solution of the
reactive transport in the fractal radial domain, and the effective upscaled solution using
the lamella approach, are presented. The match between the theoretical results here when
verified against the numerical simulation results reinforces the power of the lamella approach
in upscaling the mixing limited reactions in the fractal radial domains.

235
Corresponding author: David Benson
Colorado School of Mines
[email protected]

Eliminating the Scale Effect From Reactive Transport: Particle-Tracking

Simulation of Batch- to Field-Scale Bioremediation Experiments at the
Schoolcraft, MI, site

David Benson, Colorado School of Mines

D. Ding, Colorado School of Mines

Abstract: It is commonly observed that the measured (or empirically fitted) reaction rates
at groundwater remediation field-scale sites are much lower than those found in the same
material at the batch- or laboratory-scale. The reduced rates are typically attributed to
poorer mixing at the larger scale. A variety of methods have been proposed to capture this
scaling effect in reactive transport. Most studies focus on simulating simple reactions, such
as a bimolecular forward reaction. Here we move to a much more complex setting and use the
Lagrangian particle tracking and reaction (PTR) method to simulate a eld bioremediation
project at the extensively studied Schoolcraft (MI) site. A denitrifying bacterium, Pseu-
domonas Stutzeri strain KC (KC), is injected to the aquifer, along with sufficient substrate,
to degrade the contaminant, Carbon Tetrachloride (CT), under anaerobic conditions. The
PTR method simulates the occurrences of chemical reactions through probabilistic rules
of particle collisions, interactions, and transformations to address the scale effect (lower
reaction rate in the eld scale than batch- or lab- scale). The comparison between field mea-
surements and simulations shows that the PTR method is able to match the measurement
concentrations in the field experiment using batch rate coefficients, while previous Eulerian
methods (using RT3D) needed to reduce CT reaction rates by factor up to about 30. Our
results indicate that truly accurate predictions can be made if the correct data is collected
and the proper methods are used.

236
Corresponding author: Daniel Burnell
Tetra Tech
[email protected]

A Practical Modeling Framework for Non-Fickian Transport and First-Order

Reaction

Daniel K. Burnell, Tetra Tech

Scott Hansen
Jie Xu
Charles Faust
A. Walter

Abstract: Many studies indicate that small-scale heterogeneity and/or mobile-immobile

mass exchange may produce transient non-Fickian field-scale plume behavior that is not
well captured by the advection-dispersion equation (ADE). In such circumstances, time-
nonlocal governing equations have been seen to capture the correct behavior. Here we
discuss the problem of such non-Fickian transport coupled with simultaneous first-order
decay (e.g. biodegradation or radioactive decay). Using the time-nonlocal continuous time
random walk (CTRW) theory, we derived transient analytical solutions in Laplace space for
a variety of injection and detection conditions, which we numerically inverted using the de
Hoog algorithm to study their behavior. Reaction was found to have a qualitatively different
effect on breakthrough curves than on whole-plume spreading. Compared to conservative
transport, first-order decay selectively removes longer-resident (slower) particles, leading to
break!through curves at fixed points that appear Fickian as well as mean arrival times that
scale linearly with distance. However, reaction was found not to affect the non-Fickian
spreading rate and spatial distribution of plumes as a whole. A few instructive analytic
solutions are presented and practical implications of our results are discussed.

237
Corresponding author: Evan Christianson
Barr Engineering
[email protected]

Strategies for simulating the complete transport pathways of atmospherically

dispersed contaminants
Evan Christianson, Barr Engineering
Dave Dahlstrom, Barr Engineering
Adam Janzen, Barr Engineering
Jonathon Carter, Barr Engineering
Ray Wuolo, Barr Engineering

Abstract: Simulation of the complete transport pathways for atmospherically dispersed

contaminants presents a number of challenges. No single model exists in standard practice
that can account for transport through air, the unsaturated zone, and groundwater flow
systems. Additionally, the scale of the impacted area associated with many aerially deposited
contaminants is well beyond that of most typical, point-sourced, groundwater contaminant
plumes. This paper describes a framework that was developed to address these challenges,
allowing for simulations to be conducted that account for spatial variation of the physical
components controlling transport through the unsaturated zone at a computational scale
that is tractable using widely-available computing resources.

Extended abstract: Methodology, findings, and implications

Simulation of the complete transport pathways for atmospherically dispersed contaminants

presents a number of challenges. No single model exists in standard practice that can account
for transport through air, the unsaturated zone, and groundwater flow systems. Additionally,
the scale of the impacted area associated with many aerially deposited contaminants is well
beyond that of most typical, point-sourced, groundwater contaminant plumes; ranging up to
global scales. This paper addresses regional-scale aerial deposition (on the order of hundreds
to thousands of square kilometers).
A framework was developed to address these challenges by linking a series of com-
monly used fluid flow and solute transport models, each designed for a specific medium.
These include: AERMOD (USEPA, 2004), SWB (Westenbroek et. al., 2010), MODFLOW-
UZF/UZT (Bedekar et. al., 2016; Niswonger et. al., 2011), and MODFLOW-NWT/MT3D-
USGS (Bedekar et. al., 2016; Niswonger et. al., 2011). The models are linked by passing
output from one or more models as input for the model simulating the next segment of the
transport pathway (Figure 1).
Within this framework, simulation of transport through the unsaturated zone is typically
the most computationally-demanding simulation component. The scale of the impacted area
associated with aerially deposited contaminants and the level of discretization necessary to
accurately simulate flow and transport processes in the unsaturated zone make the problem
computationally intractable at full scale. However, because flow in the unsaturated zone can
be assumed to be only in the vertical direction, adjacent cells within the same model layer
do not interact with each other. It is therefore possible to reduce the number of cells in the

238
unsaturated zone model using a binning and averaging approach without compromising the
physical aspects of the flow and transport simulation.
A compartmentalization approach similar to Hydrological Response Units (HRU’s), which
have been a common practice for surface water routing models for decades, was implemented
for the unsaturated zone model. This methodology allows for a simulation that accounts for
spatial variation of the physical components controlling transport through the unsaturated
zone at a computational scale that is tractable using widely-available computing resources.
The binning scheme is based on estimates of unsaturated zone thickness, infiltration rates,
and solute concentrations in the infiltrating water throughout the model domain. Histograms
of each component are developed and the bin numbers for each component are mapped to
the row-column locations of the full scale model (Figure 1). A “binned” MT3D-USGS model,
representing only the unsaturated zone, is then developed with each row-col of the binned
model representing a unique combination of the bins mapped to the full scale model. The
concentrations exiting the unsaturated zone of the binned model are tracked and mapped to a
separate, full-scale MODFLOW/MT3D model as recharge inputs for simulation of transport
in the saturated zone.
This approach, developed during investigations of widespread groundwater contamina-
tion from air emission sources, has been shown to successfully reduce the total spatial dis-
cretization (total number of cells) of the unsaturated-flow and -transport model by factors
over 95%. Figure 2 shows the tradeoff in runtime along with cumulative and annual mass
exiting the unsaturated zone for different binning schemes. In this example, the difference
between a 90% reduction and 97% reduction is minimal and insignificant in regard to the
timing of mass to the saturated zone for most applications. However, the difference in model
runtimes between these realizations is a factor of 4.
The reduced model discretization does come with some tradeoffs regarding the ability
to capture the timing and quantity of mass exiting the unsaturated zone and entering the
saturated flow system (simulated at full scale with no dimensional reduction). For example,
steady-state infiltration rates were adopted over the transient simulations that would be
required to simulate time-varying infiltration rates based primarily on consideration of the
size of the FTL file linking MODFLOW-NWT and MT3D-USGS. The influence of time-
varying deposition rates was captured within the binning framework by computing an average
within each bin for each year of the simulation.
Compartmentalization of the unsaturated zone transport makes it feasible to conduct
computationally demanding routines on the full simulation framework, such as predictive
analysis and model calibration. It is also possible to conduct most of these computationally
demanding routines at greater dimensional reduction and then simulate with less dimensional
reduction for final model simulations.

239
Figure 1. a.) Schematic of modeling framework. b.) Model binning approach for unsaturated zone flow and
transport

240
Figure 2. a.) Cumulative mass exiting the unsaturated zone for different binning schemes. b.) Annual mass
exiting the unsaturated zone for different binning schemes. c.) Model run time for different binning schemes

References

Bedekar, V., Morway, E.D., Langevin, C.D., and Tonkin, M., 2016b, MT3D-USGS version 1.0.0: Ground-
water Solute Transport Simulator for MODFLOW: U.S. Geological Survey Software Release, 30 September
2016, (http://dx.doi.org/10.5066/F75T3HKD).
Cimorelli, A. J., Perry, S. G., Venkatram, A., Weil, J. C., Paine, R. J., Wilson, R.B., Lee, R.F., Peters,
W. D., Brode R.W., & Paumier, J.O., 2004. AERMOD–Description of model formulation. EPA Report
EPA-454-03-004
Niswonger, R.G., Panday, S., and Ibaraki, M., 2011. MODFLOW-NWT, A Newton formulation for MODFLOW-
2005. U.S. Geological Survey Techniques and Methods 6-A37, 44 p.
Westenbroek, S.M., Kelson, V.A., Drips, W.R., Hunt, R.J., and Bradbury, K.R., 2010, SWB-A modified
Thornthwaite-Mather Soil-Water-Balance code for estimating groundwater recharge: U.S. Geological Survey
Techniques and Methods 6-A31, 60 p.

241
Corresponding author: Nicholas Engdahl
Washington State University
[email protected]

When do we really need to be modeling transience? Examples from residence

time distributions and solute transport

Nicholas B. Engdahl, Washington State University

Abstract: The steady-state approximation is often invoked as a simplifying assumption in

models of solute transport and residence time because of the immense complexity and data
requirements of time-dependent models. At a minimum, the steady-state assumption re-
quires that the flow field driving fluid and solute migration is steady over time, though some
cases also assume steady sources for solute fluxes. Given long enough, all natural systems
are transient so the fundamental question explored in this talk is, under what circumstances
is the steady-state approximation reasonable? The question is investigated in detail for res-
idence time distributions, which also govern solute transport, and we consider analytical
and numerical models of confined saturated systems as well as variably saturated integrated
models. In both cases, the role of transience is significant when the sample location is near
a transient hydrologic boundary and the effects are greatest when the sampling location is
adjacent to a boundary. However, regions located far from the boundaries show minimal
variability in their residence time distributions and can be approximated quite well from a
steady-state model, even when transient fluctuations in the flow boundary conditions are
large. The counter-example to this is when sources are transient, such as solute transport
loading rates, or if there is seasonal disconnection and reconnection of flowpaths. Determin-
ing a generalized response in these systems is not straightforward and it is recommended
that fully transient models be constructed to study the complex behaviors and hysteresis
that may be present. An important point is that most of the examples considered here use
homogeneous flow properties to isolate the role of transient boundaries, but the effects are
expected to be compounded in more realistic, heterogeneous flow systems.

242
Corresponding author: Scott Hansen
Los Alamos National Laboratory
[email protected]

Aurora: a partner package for MODFLOW that enables non-Fickian transport

modeling of real aquifers

Scott K. Hansen, Los Alamos National Laboratory

B. Berkowitz
D. Burnell
-
-
-
-

Abstract: It is now widely accepted in the academic world that non-Fickian transport
is ubiquitous in the subsurface, on account of both heterogeneous advection and mobile-
immobile mass transfer. Unfortunately, it has been difficult to apply the knowledge that
has been gained in these areas to real sites because of a lack of available tools. This problem
is addressed by Aurora, a new particle tracking software package that partners with MOD-
FLOW. For large-scale advection, Aurora uses the flow field computed by MODFLOW. It
then conceptually implements a continuous time random walk (CTRW) on each streamline
as a subordination technique to model (Fickian or non-Fickian) dispersion due to advective
heterogeneity at the sub-block scale. A variety of transition time distributions are available
to the user for this purpose, including shifted Dirac (plug flow), inverse Gaussian (Fick-
ian), and power law distributions. Mobile-immobile mass transfer due to, e.g., adsorption
is also optionally implemented, and so is Fickian diffusion transverse to streamlines. We
present the theory underlying the software, information about its usage, and some example
numerical results.

243
Corresponding author: Nazmul Hasan
INTERA Incorporated
[email protected]

System-Level Modeling for Waste Management Area C Performance

Assessment at the Hanford Site in Southeast Washington, U.S.A

Nazmul Hasan, INTERA Incorporated

S. Mehta

Abstract: A performance assessment of Single-Shell Tank Waste Management Area C

located at the U.S. Department of Energy’s Hanford Site in southeastern Washington is
being conducted to satisfy the requirements of the Hanford Federal Facility Agreement and
Consent Order, as well as State-approved closure plans and permits. The performance
assessment of Waste Management Area C evaluates the fate, transport, and impacts of
radionuclides and hazardous chemicals within residual wastes left in tanks and ancillary
equipment in their assumed closed configuration and the subsequent risks to humans into
the far future. The analysis for releases to groundwater has been structured to provide two
complementary approaches to the analysis. A detailed process model for flow and transport
has been implemented in the Subsurface Transport Over Multiple Phases (STOMP) code.
However, the model implemented in STOMP is computationally intensive, limiting the
potential for evaluation of parameter uncertainties in the performance assessment. Further-
more, it is focused solely on the processes associated with groundwater flow and transport.
Consequently, a system-level model has been implemented in GoldSim. It includes models
for release of contaminants from the near field, transport through the vadose and saturated
zones into the biosphere, and dose calculations and uncertainty analyses to evaluate the
performance of the overall system. The uncertainty analysis is conducted over 10,000 year
time period and the highest calculated peak dose of 2.5 mrem/yr results primarily from
99Tc. The most influential uncertain parameters that affect peak dose are the vadose zone
hydraulic properties and Darcy flux in the saturated zone.

244
Corresponding author: Skuyler Herzog
Colorado School of Mines
[email protected]

Engineered streambeds for improved water quality: Comparison of

MODFLOW and STAMMT-L representations of performance data from
constructed flume experiments

Skuyler Herzog, Colorado School of Mines

C.P. Higgins, Colorado School of Mines
J. McCray, Colorado School of Mines

Abstract: The hyporheic zone (HZ) is a hotspot for biogeochemical processing that can
attenuate a variety of nonpoint source contaminants in streamwater. However, HZ in urban
and agricultural streams are often degraded and poorly connected with surface water. To in-
crease hyporheic exchange and improve water quality, we introduced engineered streambeds
as a best management practice. Modifications to streambed hydraulic conductivity and re-
activity are termed Biohydrochemical Enhancements for Streamwater Treatment (BEST).
BEST are subsurface modules that utilize low permeability sediments to drive efficient hy-
porheic exchange, and reactive geomedia to increase reaction rates within the HZ. Previous
contaminant mitigation potentials of BEST were estimated based on hyporheic flow and res-
idence time outputs from MODFLOW and MODPATH models and reported contaminant
removal rate constants. Simulations showed that BEST structures in series can substantially
improve water quality in small streams along reaches of tens of meters. Here we present
new physical data from two artificial stream flumes at the Colorado School of Mines in
Golden, CO: a BEST vs. all-sand control condition. NaCl breakthrough curves were ana-
lyzed using STAMMT-L, a mobile-immobile exchange model, which showed ~50ύ greater
hyporheic exchange and longer residence times in the BEST stream relative to the control.
Calibrated models were validated using the indicator compounds resazurin and its daughter
product resorufin. These new observed and modeled data show reasonable agreement to
previous MODFLOW results, and demonstrated that BEST may be an effective novel best
management practice for improving streamwater quality in urban and agricultural settings.

245
Corresponding author: Katrina Marini
Barr Engineering
[email protected]

Using uncertainty analysis for groundwater flow and transport modeling to

inform remedial design and monitoring

Katrina Marini, Barr Engineering

K. Lindstrom
D. Dahlstrom
J. Mohr

Abstract: Groundwater flow and solute transport models can be effective tools for under-
standing and managing environmental risks throughout a project’s life, including predicting
effects of potential remedial alternatives on future site conditions. Such predictions should
include analysis of model uncertainty. At a former manufactured gas plant (MGP), remedial
alternatives were evaluated to mitigate MGP-related impacts in sediments within a river
adjacent to the site. Predictive uncertainty analysis was completed for the calibrated model
using Null Space Monte Carlo (NSMC) techniques. PEST utilities generated 50 parameter
sets by simultaneously varying all parameter values within their calibration ranges. Param-
eter sets were refined to reduce adjustments to values in the solution space of the inverse
model while leaving adjustments to values in the null space. Up to two calibration iterations
partially recalibrated the parameter sets. Steady-state flow and transient transport predic-
tive scenarios were developed to represent natural recovery and three remedial alternatives.
The predictive scenarios were run with 50 parameter sets and concentration and mass flux
results were compared between scenarios. The modeled alternative of dredging in the river
followed by application of an impermeable cap was selected for design refinement. After
additional site investigation for remedy design, the groundwater flow and solute transport
model was recalibrated and 50 new parameter sets were generated with NSMC techniques.
Predictive scenarios were limited to natural recovery and remediation with dredging and in-
stallation of an impermeable cap. Modeled groundwater flow and solute concentrations from
the NSMC uncertainty analysis informed design decisions and guided the post-remediation
monitoring plan.

246
Corresponding author: Amena Mayenna
INTERA Incorporated
[email protected]

Evaluating Enhanced Attenuation of Uranium at the 300 Area of Hanford Site,

Washington

Amena M. Mayenna, INTERA Incorporated

Sunli Mehta, INTERA Incorporated

Abstract: Large volumes of liquid waste containing uranium were discharged to the soil
column in the 300 Area of the U.S. Department of Energy’s Hanford Site located in south-
east Washington State resulting in groundwater contamination. Part of the remedy for
enhanced attenuation of uranium involves contacting the sediments with high concentra-
tions of phosphate solutions to form in-situ calcium phosphate bearing precipitates that can
sequester the uranium and reduce its leachability. Stage A of the remedy was completed
in November 2015 by infiltrating and injecting polyphosphate solutions within an area of
approximately 0.3 ha (0.75 ac). The target zone for sequestering uranium is the portion
of the vadose zone that undergoes periodic rewetting due to nearby Columbia River stage
fluctuations that results in leaching of uranium mass to aquifer. A variably saturated 3-D
numerical model was used to simulate fate and transport of uranium in the vadose zone
and unconfined aquifer to evaluate the effectiveness of the Stage A remedy. Hydraulically
the model was calibrated to rapid Columbia River stage changes and resulting changes in
groundwater levels, and mixing of river and aquifer water. Transport modeling involved
simulating uranium concentrations over a 20-year period prior to the treatment, the treat-
ment period that lasted about two weeks, and the post-treatment period to predict the
longer-term uranium groundwater concentrations. The modeling results predict reductions
in groundwater uranium concentrations in the near future and in the longer term because
of the Stage A remedy.

247
Corresponding author: Melissa Mitton
Colorado School of Mines
[email protected]

Understanding CH4 leakage from pipelines as affected by soil heterogeneity

and moisture

Melissa K. Mitton, Colorado School of Mines

Abstract: Sub-surface transport of leaked natural gas from pipelines occurs through com-
plex transport pathways due to soil heterogeneities and changes in soil moisture and is
affected by variable atmospheric conditions such as winds, frontal passages and rain. For
example, gas migration in a texturally heterogeneous soil system will be markedly different
from that of a homogenous soil system due to the different texture- (or porosity-) induced
tortuosity effects. Although industry has made progress on detection technologies them-
selves, limited progress has occurred on leak detection and quantification from pipelines.
This is, in part, due to the limited availability of experimental studies investigating subsur-
face methane migration under different subsurface and atmospheric conditions. To better
understand fugitive emissions from underground natural gas infrastructure, we are develop-
ing a field testing facility. The test site includes four test beds designed to study different
aspects affecting subsurface methane diffusion including; soil type, temperature, soil mois-
ture, leak size, depth and direction. The site is equipped with subsurface and surface
sensors to continuously monitor changes in these variables. As soil moisture increases, a
decrease in spreading width and volumetric flux rate is expected. We also expect larger
effects from emission points directed along the horizontal plane than from those directed
vertically. Accompanying work will include numerical modeling of the field experiments to
relate subsurface concentrations, to above ground concentrations, measured by most de-
tection technologies. These findings will better inform leak detectors of the leak severity
before excavation, aiding with safety precautions and work order categorization for improved
efficiency.

248
Corresponding author: David Moulton
Los Alamos National Laboratory
[email protected]

Integrated Hydrology Model Intercomparison using High-Resolution Data at

the East River Colorado
David Moulton, Los Alamos National Laboratory
Erica Siirila-Woodburn, Lawrence Berkeley National Laboratory
Dipankar Dwivedi, Lawrence Berkeley National Laboratory
Carl Steefel, Lawrence Berkeley National Laboratory
Daniil Svyatskiy, Los Alamos National Laboratory
Reed Maxwell, Colorado School of Mines
Laura Condon, Syracuse University

Abstract: Growing interest in understanding and predicting climate impacts and feedbacks
on terrestrial ecosystems is driving an increased interest in models that represent and couple
all relevant processes. But even as field and laboratory campaigns rise to the challenge of
collecting diverse and heterogeneous datasets that can inform and constrain these models,
questions remain about the model’s process representation and implementation. To address
these questions, the community is increasingly turning to model intercomparision studies
where multiple codes solve the same problem to expose and understand differences in re-
sults. Recent model intercomparisions in integrated hydrology (e.g., Kollet et al., 2017),
have made significant progress but have utilized simple geometries and small domains. In
this work, we take advantage of the growing set of data being collected in the East River
watershed, Colorado as part of Lawrence Berkeley National Laboratory’s Scientific Focus
Area. Specifically, we highlight our progress in developing a model intercomparison that
uses high-resolution (0.5m) Lidar data and bathymetry data along a meandering section
of the East River. We compare two open-source integrated hydrology codes that partici-
pated in earlier model intercomparison studies, ParFlow and Amanzi/ATS. Although, both
codes solve variably saturated subsurface flow using Richards equation coupled to the diffu-
sive wave model of surface water flow, they use different assumptions about the underlying
mesh, discretizations, and time evolution strategies. Here we focus on integrated hydrology
to explore these differences, and describe future plans to explore the impact of differing land
model representations and coupling to reactive transport models.

249
Corresponding author: Carlos A. Rivera Villarreyes
DHI WASY GmBH
[email protected]

Partial ages: defining domain specific residence time distributions

Carlos A. Rivera Villarreyes, DHI WASY GmBH

F. Cornaton
V. Clausnitzer

Abstract: Age and travel time properties have been routinely used to quantify the transport
rate of tracers in surface and subsurface water bodies, and to make estimates on pore velocity
magnitudes and internal mixing phenomena. Water age is today generating more interest for
it can be associated to an ideal, conservative tracer assessable from field measurements, lab
analysis and numerical modelling. Age is subject to both advection and dispersion/diffusion
processes and relates to the time used to reach a given location of the domain while ignoring
other aspects of the flow paths. Lifetime expectancy is this complementary travel time that
measures the time remaining at a given location prior to exiting the domain. In order to
be able to keep track of the subregions visited by a tracer - the fundamental information
disregarded by total age - age quantities can be partitioned in a composite way allowing them
to relate to the times spent (or still to be spent) in the different subregions of the domain.
Such composite (or partial) travel times can be computed in an Eulerian framework very
similar to that used to assess the global quantities. The technique is implemented in the
state-of-the-art FEFLOW software, which is then used to demonstrate the versatility of
the concept of travel times and their decomposition, and the potential new insights that
can be gained, by means of a series of theoretical and practical examples. Targeted fields
of application range from subsurface, overland and oceanic contexts, to industrial porous
media.

250
Corresponding author: Christopher Shultz
Colorado State University
[email protected]

Using MODFLOW and a Coupled RT3D-OTIS Model to Simulate Best

Management Practices for Water Quality Improvement in an Irrigated
Agricultural River Valley

Christopher D. Shultz, Colorado State University

T. Gates
R. Bailey

Abstract: Over recent decades, many water quality problems related to nutrients and trace
elements have been reported. These constituents can be harmful to plants and animals de-
pendent on aquatic ecosystems and even to humans in cases of elevated concentrations. This
study develops a model of the stream-aquifer system of a representative region of the Lower
Arkansas River Valley in Colorado, an area with a long history of irrigated agriculture, by
linking MODFLOW (with SFR) to simulate flow conditions and RT3D-OTIS to simulate
reactive chemical transport. In modeling solutes, an enhanced version of RT3D is used for
cycling and transport in saturated and unsaturated zones in the subsurface; whereas, an
enhanced OTIS accounts for reactive transport within the river system. Calibration of a
previously-developed MODFLOW model was enhanced, in conjunction with the incorpo-
ration of the SFR module for stream flow routing to match field data. The RT3D-OTIS
model was calibrated and tested to accurately represent field selenium (Se) and nitrate
(NO3 ) groundwater concentrations, surface water concentrations, and estimates of mass
loading to the Arkansas River. Oxidation-reduction reactions have been accounted for in
the interaction of Se and NO3 . The model is applied to assess best management practices
(BMPs) by comparing their resulting concentrations and mass loadings with the baseline
condition, with the aim of lowering Se and NO3 in-stream concentrations and comply with
the state of Colorado’s standards. Results indicate the potential for significant lowering of
Se concentrations through adoption of BMPs, but with only limited potential for dropping
NO3 concentrations.

251
Corresponding author: John Sigda
INTERA
[email protected]

Embracing Hydrogeologic Uncertainty: Rapid Robustness Assessment of the

Capture Well System for the EDB Plume from the Kirtland Air Force Base
Fuel Release Using Analytic Element Modeling

J. Sigda, INTERA
W. Linderfelts, INTERA
R. Shean, ABCWUA
E. Marcillo, INTERA

Abstract: Long-term fuel releases from Kirtland Air Force Base’s (KAFB) Bulk Fuels
Facility have contaminated the highly permeable sediments of the ancestral Rio Grande
River aquifer in Albuquerque, NM. An extensive dissolved-phase plume of carcinogenic
ethylene dibromide (EDB), originating from aviation gasoline, threatens a productive well
field operated by the Albuquerque Bernalillo County Water Utility Authority (ABCWUA).
This evaluation focused on determining the engineering robustness of a proposed 5-well
capture well network given the poorly constrained hydraulic properties and hydraulic gra-
dient dynamics of the contaminated aquifer. We developed a 3D analytic element modeling
tool using TIMML that calculated the EDB plume fraction not captured by the proposed
network for reasonable ranges of hydraulic conductivity, vertical anisotropy, and regional
gradient magnitude and direction. Results take minutes to hours and animations helped
inform ABCWUA’s staff and boards about the risks from the KAFB EDB plume. Our
results demonstrated that the proposed capture well network is not robust because it does
not contain the EDB plume for reasonable uncertainties in system characteristics.

Extended abstract: Methodology, findings, and implications

Long-term fuel releases from Kirtland Air Force Base’s (KAFB) Bulk Fuels Facility have
contaminated the highly permeable sediments of the ancestral Rio Grande River aquifer in
Albuquerque, NM. An extensive dissolved-phase plume of carcinogenic 1,2-dibromoethane
(ethylene dibromide or EDB), originating from aviation gasoline, threatens the productive
Ridgecrest well field operated by the Albuquerque Bernalillo County Water Utility Authority
(ABCWUA). Questions about whether KAFB’s proposed remediation system will protect
the wellfield led ABCWUA to have INTERA develop a rapid modeling tool to evaluate
different KAFB capture well networks, including the most recent KAFB network that had
been designed using a 3D MODFLOW-MT3DMS model.
Located in the unconfined part of the Albuquerque Basin aquifer, the Bulk Fuel Facility’s
EDB extends about 5,000 feet (ft) beyond the base (Figure 1). Ridgecrest well 5 is the
nearest public water supply well along the migration path, with about 3,700 ft between it
and the known distal plume edge. Groundwater heads in and around the EDB plume form
an elongate depression or trough created by long-term pumping of water supply wells in the
Ridgecrest well field and others.

252
 Comparison of USGS composite head contours for 2008 and 2012 (Falk et al., 2011;
Powell and McKean, 2014) show that the head trough appears to be narrowing, which could
indicate that the trough in the shallow unconfined aquifer with the plume is also narrowing
and perhaps changing direction. Examination of the shallow depth EDB plume outlines
(Figure 1) shows two bounding directions: 0 degrees east of north (N00E) and 22 degrees
east of north (N22E).
The magnitude of the regional hydraulic gradient along the plume migration path has
decreased since 2009 when groundwater pumping for water supply began to decrease. Data
from United States Geological Survey (USGS) monitoring wells located near the EDB plume
show that the annual mean magnitude of the hydraulic gradient was 0.0012 ft/ft in 2008 and
0.0006 ft/ft in 2015.
There is also uncertainty about the vertical and horizontal hydraulic conductivity for
this part of the aquifer. Analysis of a pump test for extraction well KAFB-106228 (CB&I,
2016) yielded a wide range of values: 40 to 240 feet per day (ft/d) for horizontal hydraulic
conductivity (K) and 0.001 to 0.06 for vertical anisotropy.
Our robustness evaluation of the capture well networks estimates the fraction of the EDB
plume not captured for selected ranges of aquifer and capture system characteristics. A given
capture system is considered robust if it captures all or nearly all of the EDB plume across
the ranges of values for the characteristics. The robustness modeling tool was conceived
to directly address the ranges of likely values for hydraulic properties and gradient as a
complement to the EPA MODFLOW-MT3DMS flow and transport model, which examines
plume capture for a single choice of hydraulic properties and gradient.
The following list summarizes the important features of the robustness modeling tool:

1. The modeling tool uses the three-dimensional, open-source analytic element code
TIMML (Bakker, 2015).

2. The contaminated groundwater aquifer is treated as an unconfined aquifer with ho-

mogenous but anisotropic hydraulic conductivity under quasi-steady flow conditions.
Hydraulic properties include the horizontal hydraulic conductivity, ratio of vertical to
horizontal hydraulic conductivity (vertical anisotropy), and porosity.

3. Regional groundwater flow direction and rate along the EDB plume caused by pumping
of the water supply wells are specified using the regional gradient element.

4. The 5 extraction wells (Figure 1) are simulated to start pumping according to a sched-
ule based on known completion dates.

5. Envelopes of the EDB plume were determined for three depths from the data for the
shallow, intermediate, and deep monitoring wells in the KAFB monitoring network.
The source areas were assumed to not contribute additional dissolved contamination
during the simulations.

6. Each simulation is created from values of the system characteristics selected from the
ranges of uncertainty and from the schedule and pumping rate for each of the extraction
and injection wells.

253
 7. The modeling tool records the number of EDB plume particles that are captured by
the extraction wells during the simulation and creates a table summarizing the results.
It also exports images suitable for creating an animation.

The combinations of hydraulic properties and gradient magnitudes yield 6 different seep-
age velocity values driving the EDB plume. Seepage velocity values for the 0.0003 ft/ft
gradient magnitude are 82, 96, and 110 feet/year (ft/yr) for the horizontal K values of 150,
175, and 200 ft/d, respectively. Seepage velocity values for the 0.0008 ft/ft gradient mag-
nitude are 219, 256, and 292 ft/yr for the horizontal K values of 150, 175, and 200 ft/d,
respectively.
Our robustness modeling tool allows rapid assessment of many different values for the
system characteristics because TIMML is computationally efficient compared to the KAFB
MODFLOW model. Our tool can distribute the different simulations to individual cores,
yielding run times across the full range of characteristics of only a few hours.
The results from the robustness simulations reveal that significant fractions of the EDB
could migrate beyond the 5-well extraction system for one or more of a limited set of system
characteristics. Table 1 presents the percentage EDB not captured given a vertical to hori-
zontal K ratio of 0.1 for the all-detections plumes after 10 years of simulation time. Showing
only a part of our results, Table 1 reveals that the system characteristics that lead to higher
percentages of EDB not captured include:

a ) If the regional groundwater head gradient is oriented near to south to north;

b ) If pumping rates at all but the southernmost extraction well are smaller; and

c ) If the aquifer’s effective horizontal hydraulic conductivity is 200 ft/d.

It may not be possible to sufficiently reduce the uncertainty about the hydraulic con-
ductivity of the plume aquifer with common analysis methods even if more pump tests are
carried out. The resulting uncertainty may still be too large to ensure adequate EDB capture
with the proposed 5-well extraction system.
This evaluation focused on determining the engineering robustness of a proposed 5-well
capture well network given uncertainty in hydraulic properties and hydraulic gradient dy-
namics of the contaminated aquifer. Results take minutes to hours and animations helped
inform ABCWUA’s staff and decision-makers about the risks from the KAFB EDB plume.
Our results demonstrated that the proposed capture well network is not robust because
it does not contain the EDB plume for reasonable uncertainties in aquifer characteristics.
Adding extraction wells may be more cost effective than trying to sufficiently decrease un-
certainties about aquifer characteristics.

254
 RI
DGECREST5
!
?

RI
DGECREST3
!
?

BURTON5
!
?
KAFB-106234

EXT-2

KAFB-106233
KAFB-106228

EXT-7

KAFB-7

±
1,
400 0 1,
400
Wat
erSupplyWells
Feet
!
?
Ext
ract
io n Well
I
nject
io n WellLo cat
i
on
CB&IShallo w EDBPlumeEn velo pe(
MCL)
I
NTERAShallo w EDBPlumeEn velo pe(
Det
ect
ions)
Figur
e1
Ext
ract
io n Wells ,Shallo w
EDBPlumeEn velo pes ,an d
So ur
ce(
s):St
reetmap– ESRI
,GI
So n li
ne WaterSupplyWells

Figure 1. Extraction Wells, Shallow EDB Plume Envelopes, and Water Supply Wells

255
 Table 1. Percentage EDB Not Captured After 10 Years for Anisotropy Ratio = 0.1: All Detections EDB Plume Envelopes

Plume Envelopes
Regional  Seepage  Shallow Plume  Intermediate Plume  Deep Plume 
Gradient  Velocity  Q (gpm)  Q (gpm)  Q (gpm) 
Direction  (ft/yr)  125  150  175  125  150  175  125  150  175 
   82  25  21  19  14  11  9  4  0  0 
   96  24  22  20  9  9  9  4  8  0 
N00E  110  26  22  20  10  8  8  0  0  4 
   219  32  29  24  14  11  7  0  0  0 
   256  38  31  26  24  15  10  0  0  0 
   292  42  34  30  28  20  15  0  0  0 
   82  22  20  20  11  9  10  4  0  0 
   96  20  18  18  9  9  8  4  8  0 
N11E  110  22  18  16  8  7  5  0  0  0 
   219  18  18  14  7  4  3  0  0  0 
   256  19  16  14  10  7  5  0  0  0 

256
   292  21  16  14  12  10  9  4  0  0 
   82  21  18  18  11  9  9  4  0  0 
   96  20  18  18  9  7  8  8  4  4 
N22E  110  21  18  17  9  9  8  8  0  8 
   219  14  12  12  8  5  4  4  4  0 
   256  16  11  9  8  6  5  15  4  4 
   292  16  14  13  11  7  7  19  15  4 

Table 1. Percentage EDB Not Captured After 10 Years for Anisotropy Ratio = 0.1: All Detections EDB
 References

Bakker, M. 2015. TimML A Multi-Aquifer Analytic Element Model Version 4.0 DRAFT. August 20, 2015.
https://github.com/mbakker7/timml
CB&I. 2016. Aquifer Test Report for Groundwater Extraction Well KAFB-106228 Bulk Fuels Facility, Solid
Waste Management Unit ST-106/ SS-111 Kirtland Air Force Base, Albuquerque, New Mexico. January
2016.
Falk, S.E., Bexfield, L.M., and Anderholm, S.K., 2011, Estimated 2008 groundwater potentiometric surface
and predevelopment to 2008 water-level change in the Santa Fe Group aquifer system in the Albuquerque
area, central New Mexico: U.S. Geological Survey Scientific Investigations Map 3162, 1 sheet.
Powell, R.I., and McKean, S.E., 2014, Estimated 2012 groundwater potentiometric surface and drawdown
from predevelopment to 2012 in the Santa Fe Group aquifer system in the Albuquerque metropolitan area,
central New Mexico: U.S. Geological Survey Scientific Investigations Map 3301, 1 sheet, http://dx.doi.org/10.3133/sim3301.

257
Corresponding author: Michael Takeda
Earthfx Incorporated
[email protected]

Characterization of Seasonal Thermal Plume Migration from a

Below-Water-Table Aggregate Extraction Operation
Michael G.S. Takeda, Earthfx Incorporated
E.J. Wexler
P.J. Thompson
J.D.C. Kassenaar

Abstract: Mining of sand and gravel from below the water table typically involves removal
of cover material and exposing the groundwater (in the form of a flooded pit or pond) directly
to the surface environment. The result can be a dramatic increase in water temperature
and creation of a thermal plume emanating from the exposed area. Earthfx developed a
groundwater flow and heat transport model to assess potential off-site thermal migration
at an aggregate operation in southern Ontario, Canada. Extraction of sand and gravel was
underway when a thermal plume emanating from the on-site pond was identified, potentially
impacting a nearby trout hatchery. Using available on-site and regional datasets, Earthfx
constructed a 3-D hydrostratigraphic model for the study area; mapping a buried drumlin
feature that strongly affected groundwater flow across the site. The hydrostratigraphic
surfaces formed the basis of a MODFLOW-NWT groundwater flow model, the outputs of
which (e.g., groundwater velocities, recharge rates, stream/lake seepage) were applied to a
transient thermal transport model using an early version of the updated MT3D-USGS code.
Recharge was estimated with a distributed hydrologic model of the area using the PRMS
code and calibrated to observed streamflows. The ability to simulate seasonally-fluctuating
temperature signals from recharging meteoric water and match observed patterns in on-
site monitors were key to model calibration. Thermal modeling confirmed that recharging
meteoric water and migration of the thermal plume from the on-site pond create two different
and interacting temperature signals, which are observed in the on-site boreholes. This study
illustrated how MODFLOW-NWT in conjunction with PRMS and the updated MT3D-
USGS code can be used to successfully characterize the thermal regime surrounding sand
and gravel operations, and evaluate potential impacts from thermal plume migration on
nearby ecological receptors.

258
Corresponding author: Albert Valocchi
University of Illinois at Urbana-Champaign
[email protected]

Computationally Efficient Methods for Modeling Diffusion and Reaction in

Low Permeability Zones in Field-Scale Simulations using
MODFLOW/MT3D/RT3D

Albert J. Valocchi, University of Illinois at Urbana-Champaign

Abstract: At many legacy waste sites contaminants have diffused over long times into low
permeability zones (LPZs). These LPZs could be thick aquitards or clay lenses of various
length scales. It is well-known that management of such waste sites is challenging since
active remediation can easily remove contaminants from the more permeable zones, but the
LPZs represent a reservoir of contaminant mass that can slowly diffuse back into the aquifer
over long time periods. Accurate modeling of mass transfer from LPZs requires very fine
spatial discretization to capture sharp gradients resulting from diffusion. This is not feasible
for practical field-scale applications. In addition, models should also include slow abiotic
or biotic transformation reactions that may have a significant effect on contaminant mass
over long time scales. Falta and Wang (2017) have recently presented a semi-analytical
approximate technique to circumvent these challenges. Rather than discretizing the LPZ,
an approximate analytical solution for one-dimensional transient diffusion is used to model
the diffusive flux at the LPZ interface. This can result in significant computational savings
under conditions when the approximation is valid. We have implemented the approach
of Falta and Wang (2017) using MODFLOW/MT3D/RT3D. We have made modifications
for finite size LPZs and other cases where the approximation breaks down. We have also
extended the framework to cases of multi-species nonlinear reactions within the LPZ.

259
Corresponding author: Assaf Wunsch
Leppert Associates
[email protected]

Use of Groundwater Modeling in Assessment of Corrective Actions for

Contaminated Groundwater Short-Circuit

Assaf Wunsch, Leppert Associates

S. Leppert, Leppert Associates

Abstract: An oil refinery in Arkansas has been exceeding National Pollutant Discharge
Elimination System (NPDES) water quality requirements at an outfall at the facility, even
though actual surface water discharges have proved to be compliant. A hydrological inves-
tigation of the facility concluded that contaminated groundwater was seeping into a natural
ditch upstream from the NPDES outfall, causing the compliance failure at the outfall. The
ditch is a local topographic low that deepened and incised over time due to erosion promoted
by large surface discharges from the facility and natural surface runoff events. The ditch
effectively created a groundwater / surface water short-circuit that shortened the natural
subsurface residence time and prevented implementation of monitoring natural attenuation
(MNA), which was already agreed upon by regulators to be the corrective action for ground-
water contamination at the facility. A three-dimensional FEFLOW model was constructed
and calibrated for the facility. The model utilized seepage-face capability to validate base-
line ditch flows due to groundwater seeps. Then, the model was used to evaluate several
corrective action scenarios, including partial-filling of the ditch and several de-watering de-
signs, featuring pumping wells and trenches. It was eventually decided to partially fill in the
ditch to prevent groundwater seeps. An iterative approach was required to establish a new
ditch bottom elevation that would not result in groundwater seeps, because the free surface
at the bottom of the incised ditch effectively lowered local groundwater potentiometric sur-
face. Lastly, the ditch was re-designed to accommodate typical 24-hour storm events and
facility discharges.

260
Corresponding author: Charlie Fitts
Fitts Geosolutions
[email protected]

Handling Dewatered Domains in Analytic Element Models

Charlie Fitts, Fitts Geosolutions, www.fittsgeosolutions.com

Abstract: For MODFLOW, many techniques have been proposed for dealing with layers
that become dewatered. This is important since good techniques can improve model sta-
bility, convergence, and accuracy where saturated thickness approaches zero. There has
been a lack of published work dealing with dewatered layers in analytic element models.
An approach to this issue has been implemented in AnAqSim. When saturated thickness
approaches zero in an unconfined or interface domain, the domain transitions to a very
thin confined domain with a minimum saturated thickness, which allows stability and con-
vergence. The minimum saturated thickness is an input parameter, so you can make the
horizontal flow in dewatered domains negligibly small. Vertical flows are allowed through a
dewatered domain, for example recharge passing to a deeper level. This approach and result-
ing discharges and heads are examined for example multi-level models of mine dewatering
and coastal interface flow.

Extended abstract: Methodology, findings, and implications

Background

In groundwater flow models, especially multi-level ones, it is beneficial to have layers than
can dewater without causing non-convergence or numerical instability. This is not an issue
for confined domains, but is for unconfined and fresh/salt interface domains, where saturated
thickness varies with head. For unconfined layers in MODFLOW, numerous approaches to
this issue have been proposed and implemented (e.g. Bedecker et al, 2012; Keating and
Zyvoloski, 2009; Painter et al, 2008; Doherty, 2001). There has been a lack of documented
solutions to this issue for analytic element method (AEM) programs.
In typical AEM fashion, the relationship between head (h) and discharge potential (Φ)
in an unconfined aquifer is
1
Φ = 𝐾ℎ2
2
where K is horizontal hydraulic conductivity and h is measured from the aquifer base.
For Φ < 0, a condition that could occur near a high discharge well or line element, there is
no corresponding h and this condition indicates a dewatered aquifer (zero saturated thick-
ness). This condition can cause non-convergence and numerical instability if other head-
based boundary conditions are present in the dewatered region.

How Dewatered Domains are Handled in AnAqSim

261
 An approach to this issue has been implemented in AnAqSim (Fitts, 2017). When
saturated thickness approaches zero in an unconfined or fresh/salt interface aquifer (domain),
the domain transitions to a very thin confined domain with a saturated thickness called the
minimum saturated thickness (M ) (Fitts, 2015). Φ(h) in AnAqSim for an unconfined domain
is given by
1 1
Φ = 𝐾ℎ2 (ℎ > 𝑀 ), Φ = 𝐾𝑀 ℎ − 𝐾𝑀 2 (ℎ ≤ 𝑀 )
2 2
This definition of Φ(h) is continuous and monotonic, which allows stability and conver-
gence, despite nearly dewatering the domain and allowing head to drop below the aquifer
base (h<0). M is an input parameter, and specifying a smaller M reduces the magnitude of
horizontal flow in domains where h<M. M is chosen small enough so that horizontal flows
in such dewatered regions are negligible.
Vertical flows are allowed through a dewatered domain with h<M, for example recharge
passing through a dewatered domain to an underlying domain. The vertical resistance to
flow offered by any layer, dewatered or not, is constant and based on the saturated thickness
at a specified average head in the domain.
The storage parameter for an unconfined domain transitions for h<M from the specific
yield to the storativity scaled down by the ratio M/B, where B is the saturated thickness at
the average head. With a very small M, the storativity of an unconfined domain with h<M
is negligible compared to its specific yield.
Example Application: Open Pit Mine Dewatering
The AnAqSim model shown in Figure 1 has a single layer in the far-field and 4 layers
near an open pit sand and gravel mine. At t=0, the pit doesn’t yet exist. Pit dewatering
is simulated as a constant head line boundary along the pit perimeter. As the simulation
proceeds, the perimeter dewatering system enlarges and deepens, starting in level 1 and later
switching to level 2 and then level 3. Pit dewatering causes heads to drop below the base of
level 1 and then later below the base of level 2. The “+” symbols indicate the location of
spatially-variable area sink (SVAS) basis points; SVAS simulate vertical leakage and storage
fluxes (Fitts, 2010). The recharge rate is 0.328 m/yr. The aquifer properties are Kh =0.25
m/d, Kv =0.025 m/d, Ss =4.9x10-5 m-1 , and Sy =0.13. The topmost layer is unconfined, and
in the 4-layer area, layers 2 and 3 are confined/unconfined (can transition), and layer 4 is
confined. The bases of layers 1 and 2 are at 225 and 220 m elevation.
The upper half of Figure 2 shows a profile of simulated heads in all levels along the profile
line shown in Figure 1, at the end of the simulation 6 years into dewatering. The gray lines
in the profiles are top and bottom domain boundaries. Near the dewatering system, heads
in levels 1 and 2 are below their base elevations. The perimeter dewatering system is in level
3 at this point in the simulation.
The lower half of Figure 2 shows horizontal domain discharge parallel to the profile line,
which is roughly parallel to flow, for each level at year 6. At this time, a large area of level
1 is dewatered (h < 225 m), and a smaller area of level 2 is dewatered (h < 220 m). The
areas with h<base are simulated with the specified minimum saturated thickness (0.03 m in
level 1 and 0.05 m in level 2). Horizontal discharges in the dewatered regions of level 1 and
2 are not significant compared to discharges in saturated regions of these and other levels.

262
Conclusions

The approach implemented in AnAqSim allows simulation of domains that get dewatered,
both for unconfined domains and fresh/salt interface domains that approach zero freshwater
saturated thickness. Transitioning to a very thin confined domain in such areas allows
numerical stability without inducing significant horizontal discharges in dewatered areas.
Vertical flow can pass through dewatered domains into overlying or underlying domains that
are not dewatered. Storage parameters transition to very small values, so dewatered domains
do not significantly contribute to storage fluxes in transient models.

Figure 1. Layout of elements in the example application.

263
Figure 2. Vertical profile of simulated heads along profile line (top) and profile of domain discharges parallel
to profile line (bottom), at the 6-year point in the transient simulation.

References

Bedekar, V. R. Niswonger, K. Kipp, S. Panday, and M. Tonkin, 2012, Approaches to the Simulation of
Unconfined Flow and Perched Groundwater Flow in MODFLOW. Groundwater 50(2), 187–198.
Doherty, J., 2001, Improved calculations for dewatered cells in MODFLOW. Groundwater 39(6), 863–869.

264
Fitts C., 2010, Modeling aquifer systems with analytic elements and subdomains. Water Resources Research
46, W07521, doi:10.1029/2009WR008331
Fitts, C.R., J. Godwin, K. Feiner, C. McLane, and S. Mullendore, 2015, Analytic Element Modeling of Steady
Interface Flow in Multilayer Aquifers Using AnAqSim, Groundwater, 53 (3), 432–439, doi: 10.1111/gwat.12225
Fitts, C., 2017, AnAqSim (analytic aquifer simulator) version 2017-1 and 121 page user guide. http://www.fittsgeosolutions.com
Keating, E. and G. Zyvoloski, 2009, A stable and efficient numerical algorithm for unconfined aquifer analysis.
Groundwater 47(4), 569–579.
Painter, S., H. Basagaoglu, and A. Liu, 2008, Robust representation of dry cells in single-layer MODFLOW
Models. Groundwater 46, 873–881.

265
Corresponding author: Charlie Fitts
Fitts Geosolutions
[email protected]

Modeling Vertical Leakage and Storage Changes with Spatially-Variable Area

Sinks

Charlie Fitts, Fitts Geosolutions, www.fittsgeosolutions.com

Abstract: Strack and Jankovic (1999) presented functions for spatially-variable area sinks
(SVAS) that simulate extraction [L/T] as a smooth surface that passes through constraining
basis points. They used these functions to model distributions of vertical leakage. Fitts
(2010) used SVAS in AnAqSim to simulate the combined extraction due to vertical leakage
and transient storage fluxes, where the latter is defined with finite difference time steps. The
SVAS approach creates a solution that obeys the governing flow equations at the basis points
and approximates it between them. It is important to understand how well the extraction
distribution approximates the ideal extraction distribution. AnAqSim has tools to examine
the modeled and ideal extraction distribution at points and along lines. There are also tools
that allow easy adjustment of basis point spacing and extraction accuracy in different areas
and near wells. These tools are explained and demonstrated. The pros and cons of the
approximate SVAS approach are compared to the exact extraction distributions that result
from the approach of Hemker (1984) and Bakker (1999). An example comparison of these
two approaches is presented.

Extended abstract: Methodology, findings, and implications

Methods

AnAqSim (analytic aquifer simulator) simulates multi-level and transient flow using these
key concepts:

• Backward finite difference approximation of the time derivative (as in MODFLOW)

• Spatially-variable area sinks (SVAS)

• Subdomains

The general equation is written in terms of the typical aquifer discharge potential Φ
employed in analytic element modeling:

Δℎ
∇2 Φ = 𝐿𝑡 + 𝐿𝑏 + 𝑆 (1)
Δ𝑡
The right side of equation 1 is a distributed sink term (extraction) for leakage out the
top of the layer (𝐿𝑡 ), leakage out the bottom (𝐿𝑏 ), and flux into storage within the layer
(𝑆Δℎ/Δ𝑡).
In addition to well and line elements where ∇2 Φ𝑊 = 0 and ∇2 Φ𝐿 = 0, AnAqSim adds
SVAS potential functions similar to those proposed by Strack and Jankovic (1999). SVAS
produce a smooth distribution 𝛾 that approximates the desired distribution of extraction.

266
 1 𝑛 1
Φ𝑆𝑉 𝐴𝑆 = ∑ 𝑏𝑖 𝑟𝑖3 + 𝐵0 𝑟02 (2)
9 𝑖=1 4
𝑛
2
∇ Φ𝑆𝑉 𝐴𝑆 = 𝛾 = ∑ 𝑏𝑖 𝑟𝑖 + 𝐵0 (3)
𝑖=1

where 𝐵0 and 𝑏𝑖 are constants and 𝑟𝑖 is the radial distance from one if the 𝑛 basis points.
The radial basis functions of Eq. 3 were developed by Hardy (1971). The right side of Eq.
3 is set equal to the right side of Eq. 1 at each basis point, which allows determining the
constants 𝐵0 and 𝑏𝑖 . The resulting SVAS potential function (2) meets equation 1 at each
basis point and approximates it between basis points.
AnAqSim, unlike most analytic element programs, simulates heterogeneity using subdo-
mains (Fitts, 2010) rather than line doublets internal to an infinite domain. Each subdomain
has its own bounded model consisting of elements inside or on its border, its own aquifer
properties, and it may have its own anisotropy direction and ratio. Each subdomain also has
its own SVAS, so the extraction is smooth within a subdomain but may jump at abutting
subdomain boundaries. Using subdomains allows flexible layering schemes to concentrate
more layers in key areas (e.g. multiple stacked subdomains abutting a single subdomain in
the far-field).

Example Application

An example application is the dewatering problem shown in Figure 1. The model consists
of 7 layers near the system, abutting a single layer that represents the far-field. There is a
cylindrical barrier (black) that penetrates the upper 3 layers which have 𝐾ℎ = 20 ft/d. The
lower 4 layers have 𝐾ℎ = 6 ft/d. All layers have 𝐾ℎ /𝐾𝑣 = 8. Inside the barrier are three
shallow dewatering wells (red), which discharge to a shallow trench (blue).
During the first 3.5 days of the simulation, the wells discharge at fixed rates. During the
next 7 days, the wells are head-specified to maintain dewatered conditions. Model-simulated
head change profiles that cut through the recharge trench and the barrier are shown in the
upper part of Figure 2.
The lower part of Figure 2 compares the modeled extraction (𝛾) with the extraction com-
puted from heads (𝐿𝑡 + 𝐿𝑏 + 𝑆Δℎ/Δ𝑡) in levels 1 and 5. With this model, the two curves for
each level are essentially identical. With sparser basis point spacing, one could see discrep-
ancies between basis points. Plots like the lower part of Figure 2 can be made in seconds
with AnAqSim, giving users direct visual evidence of the quality of this approximation.
Denser basis point spacing is desired where there are rapid lateral changes in leakage and
storage flux, such as near line boundaries and wells. AnAqSim allows automatic placement
of basis points on hexagonal networks of points within subdomains and within user-digitized
polygons, which can be nested and edited graphically (move, insert, or delete vertexes).
Special basis point spacing applies at wells; the density of basis points increases closer to the
well according to a logarithmic function. These tools allow you to quickly set up and alter
the basis point spacing. You can do most modeling with quick coarse versions, then easily
change to finer basis point spacing for final versions.

267
Comparison with Laplace Transform Analytic Element Method (LT-AEM)

The LT-AEM is an alternate way to simulate transient multi-level flow with analytic
elements (Hemker and Maas, 1987; Bakker and Strack, 2003; Bakker and Kuhlman, 2011;
Bakker, 2013). Current programs employing the LT-AEM include MLU and TTim.
The principal advantages of the LT-AEM compared to the SVAS approach in AnAqSim
are that it simulates essentially exact distributions of extraction without basis points, and
there is no finite difference approximation of the time derivative in the storage term. How-
ever, this elegance of LT-AEM comes with significant limitations compared to the SVAS/subdomain
approach in AnAqSim:

• Aquifers must be linear (confined), not nonlinear (e.g. unconfined, fresh/salt interface,
dewatered)

• Heterogeneity boundaries are limited to cylinders

• No layering transitions (e.g. 1 subdomain abutting several others) allowed

• No anisotropy in the plane of a layer.

• No high-order linesinks

• No rivers that dry up

• No drains with resistance

• Initial heads must be zero everywhere (models simulate head change only)

• Recharge distributions are less flexible (circular areas with constant rates, compared
to spatially variable rates defined by surfaces)

To check the two approaches, a pumping test in a 9-layer heterogeneous system with a
partially penetrating pumping well and partially-penetrating observation wells was simulated
by both AnAqSim and MLU using comparable inputs. The two approaches yielded identical
results.
The SVAS/subdomain approach involves approximation of the extraction distribution,
but doing that yields many additional capabilities. With the SVAS approach, it is important
to have good tools to place and adjust basis point spacing, and good tools to check the
accuracy of the extraction approximation; AnAqSim provides both.

268
Figure 1. Vertical cross section of example AnAqSim dewatering model.

269
Figure 2. Simulated transient head profiles in level 1 (top) and simulated profiles of extraction in the second
time step in levels 1 and 5 (bottom).

References

Bakker, M., 2013, Analytic modeling of transient multi-layer flow. In: Advances in Hydrogeology, P Mishra
and K Kuhlman eds., Springer, Heidelberg, 95-114.
Bakker M. and K. Kuhlman, 2011, Computational issues and applications of line-elements to model subsur-
face flow governed by the modified Helmholtz equation. Adv Water Resour, 34, 1186–1194
Bakker M. and O. Strack, 2003, Analytic elements for multiaquifer flow. J Hydrol 271(1–4), 119–129
Fitts, C., 2010, Modeling Aquifer Systems with Analytic Elements and Subdomains, Water Resources Re-
search, 46, W07521, doi:10.1029/2009WR008331.
Fitts, C., 2017, AnAqSim (analytic aquifer simulator) version 2017-1 and 121 page user guide. http://www.fittsgeosolutions.com
Hardy, R., 1971. Multiquadric equations of topography and other irregular surfaces. Journal of Geophysical
Research 76, 1905–1915.

270
Hemker C. and C. Maas, 1987, Unsteady flow to wells in layered and fissured aquifer systems. J Hydrol.,
90(3–4), 231–49.
Strack, O. and I. Jankovic, 1999, A multi‐quadric area‐sink for analytic element modeling of groundwater
flow, J. Hydrol., 226, 188–196.

271
Corresponding author: Jacob Fullerton
Brigham Young University
[email protected]

Analytic Element Modeling on the Cloud: Tethys and TimML

Jacob Fullerton, Brigham Young University

N. Jones
M. Bakker

Abstract: The interactive nature of web applications or ”web apps” makes it an excellent
medium for conveying complex scientific concepts to lay audiences and creating decision
support tools that harness cutting edge modeling techniques. However, the technical ex-
pertise required to develop web apps represents a barrier for would-be developers. Tethys
Platform has been developed to lower the technical barrier and minimize the initial develop-
ment investment that prohibits many scientists and engineers from making use of the web
app medium. It includes (1) a suite of tools and resources for spatial data management, (2)
a Python software development kit that streamlines development, and (3) a customizable
web portal that is used to deploy the completed web apps. Tethys synthesizes several soft-
ware projects including PostGIS, 52°North WPS, GeoServer, OpenLayers, and Highcharts.
Tethys Platform is particularly well-suited for developing groundwater modeling apps that
use the analytic element method. In this paper, we describe a set of Tethys apps that have
been developed using the TimML model as the computational engine. These apps can be
used for a variety of use cases including construction dewatering, wellhead protection stud-
ies, and well impact analysis. Users can either create new TimML models from scratch or
import, execute, and visualize existing models. Tethys Platform and the associated apps
are free and open source.

272
Corresponding author: Charles McLane
McLane Environmental, LLC
[email protected]

AEM Modeling to Support Site Conceptual Model Development and

Remediation
Charles McLane, McLane Environmental, LLC
Mark Kauffman, McLane Environmental, LLC
Timothy Pryshlak, McLane Environmental, LLC
Michael Moore, PARS Environmental, Inc.

Abstract: Analytic Element Method (AEM) modeling has advantages in being able to
efficiently lay out a regional hydraulic framework that sets the flow conditions for higher-
resolution small-scale flow system analysis near a particular site of interest. This concept
was put into practice using AnAqSim modeling software for a New Jersey industrial site
undergoing remedial investigation. A regional model was constructed for the the entire
portion of the State east of the Fall Line (Potomac-Raritan-Magothy (PRM) Aquifer, be-
cause decades of deep extraction at large municipal pumping centers had caused significant
drawdown over a 300-square mile area in the center of Coastal Plain aquifer that affected
hydraulic gradients north of the Site. Within the regional model, a fine-detail multi-layer
sub-model was developed for the Site area where Site hydrologic influences included shal-
low surface water bodies ranging in scale from the Delaware River, to nearby wetlands and
streams, to small on-site drainage ditch features, and extraction from industrial wells at and
in the vicinity of the Site. The complex distribution of hydraulic heads and vertical gradi-
ents in four hydrostratigraphic units beneath the site, caused by strong downward pumping
gradients, but mitigated by intervening aquitard layers, was accurately represented by the
site-scale AEM model as evidenced by on-site groundwater elevations and contaminant pat-
terns observed at the Site. The model continues to be updated with investigation-derived
data to support remedial activities.

Extended abstract: Methodology, findings, and implications

INTRODUCTION
The industrial site (the Site) that was the focus of this study is located in Salem County,
New Jersey, in a low-lying area along the Delaware River. As part of ongoing remedial
investigation activities, a groundwater flow model was developed to 1) better understand
the complex hydrogeologic conditions present at and in the vicinity of the Site that may
have an effect on groundwater flow conditions; 2) assist in the development of a Conceptual
Site Model (CSM); and 3) guide further investigation and remediation activities.
To better understand how both regional pumping influences and site-specific features may
affect groundwater flow conditions at the local scale, we utilized AnAqSim (Analytic Aquifer
Simulator, Fitts 2010), a groundwater flow modeling software based on the domain analytic
element approach. AnAqSim’s gridless solution approach allows for the computationally
efficient incorporation of large-area regional flow features, while still allowing for detailed
analysis of flow conditions at the local scale.

273
 Once constructed and calibrated to regional and site-specific flow conditions, as discussed
below, the model was able to meet established objectives, and continues to be updated with
site data as it becomes available.
STUDY AREA DESCRIPTION
The Site is located in the eastern outcrop area of the Potomac-Raritan-Magothy (PRM)
aquifer system which regionally includes the Upper, Middle, and Lower aquifers separated
by intervening confining units. Shallow aquifer units beneath the Site have been divided in
previous investigations into Hydrostratigraphic Units (HSUs) as follows: HSU-1 = Upper
PRM and overlying sands; HSU-2 = the upper portion of the Middle PRM aquifer; HSU-3
= the lower portion of the Middle PRM aquifer; and the HSU-4 = Lower PRM which is
connected to and extends beneath the Delaware River east of the Site.
Regional cones of depression (Figure 1) caused by large-volume municipal withdrawals
influence groundwater flow conditions in an area of approximately 300 square miles north of
the Site (Lewis et al. 1991, Martin 1998, Voronin et al. 2004, Charles et al. 2011). While
these hydrogeologic and modeling studies provide valuable insight into regional flow condi-
tions, the models are not adequately discretized and parameterized to accurately represent
Site-specific surface water features, hydrostratigraphy, and pumping stresses that may affect
groundwater flow conditions at the local scale.
At the Site, several active large capacity production wells are present in HSU-3 and HSU-
4, and multiple surface water features including the Delaware River, nearby wetlands, and
on-Site drainage features control boundary and flow conditions in the shallow zone.
MODELING APPROACH
To account for the effects of large municipal pumping centers on groundwater flow (Figure
1), and to properly link the model domain to well characterized hydraulic controls (Delaware
River and coastal ocean elevations), a regional scale groundwater flow model was constructed
using AnAqSim to represent the New Jersey Coastal Plain PRM aquifer system, extending
from the Delaware River on the west to the Atlantic Coast on the east. Regional AEM model
boundary locations and conditions, model stresses, and parameter values were similar to
those applied in previous USGS MODFLOW modeling studies for the PRM aquifer system.
Once the regional groundwater flow model was developed and calibrated to regional
flow conditions, the model was refined in a 15 by 5 mile area in the vicinity of the Site
better represent local hydrology and hydrogeology, and in turn, allow for a more detailed
evaluation of groundwater flow conditions, including the effects of localized pumping of site
production wells. The local-scale model consisted of seven layers (representative of HSUs 1
through 4 and intervening confining units), with model aquifer and aquitard layers stepped to
represent the dipping beds of the PRM system (Figure 2). The Site model also incorporated
several important hydrologic features, including wetlands, Delaware River tributaries, and
Site related drainage features. Local and regional AEM models were automatically linked
by the domain analytic element method embodied in the AnAqSim software. The local-scale
model was calibrated to site-specific flow conditions using water level elevation data and
potentiometric surface maps collected and prepared by site consultants.
DISCUSSION OF MODELING RESULTS
Review of simulated Site groundwater elevations indicated that modeled heads were
generally consistent with groundwater elevations and flow depictions in site consultants’
reports for HSU-1 and HSU-2, and provided a much more detailed and comprehensive picture

274
of the flow patterns in the Site area. Pumping from production wells completed in HSU-3
and HSU-4 exerts a limited influence on HSU-2 (due to intervening confining units), and
almost no effect on HSU-1 where groundwater elevations and flow directions are controlled
by precipitation recharge and surrounding surface water bodies.
CONCLUSIONS
The use of AnAqSim, with its domain analytic element method, allowed for efficient de-
velopment of a regional groundwater flow model to seamlessly create boundary conditions for
the higher-resolution, multi-layer, local-scale model required for Site-scale analyses. Model-
ing results were consistent with limited site data, and provided valuable information to “fill
in” the picture of Site flow conditions. Model results informed CSM development, and were
used to guide additional Site investigations, the data from which were incorporated into the
model to refine existing boundary conditions and improve local-scale model calibration. The
model will continue to be used to guide further investigation and remediation activities at
the Site.

Figure 1: Observed potentiometric contours - PRM Aquifer

275
 Figure 2: Representation of PRM Aquifer units in model

References

Barton, Cynthia, and Kozinski, Jane, 1991. Hydrogeology of the Region of Greenwich Township, Gloucester
County, New Jersey: U.S. Geological Survey Water-Resources Investigations Report 90-4198.
Charles, E.G., Nawyn, J.P., Voronin, L.M., and Gordon, A.D., 2011. Simulated effects of allocated and
projected 2025 withdrawals from the Potomac-Raritan-Magothy aquifer system, Gloucester and Northeastern
Salem Counties, New Jersey: U.S. Geological Survey Scientific Investigations Report 2011–5033, 145 p.
Fitts, C.R. 2010. Modeling aquifer systems with analytic elements and subdomains. Water Resources
Research, 46:W07521. DOI: 10.1029/2009WR008331.
Lewis, J.C., Hochreiter, J.J., Barton, G.J., Kozinski, J., Spitz, F.J., 1991. Hydrogeology of, and ground-
water quality in, the Potomac-Raritan-Magothy aquifer system in the Logan Township Region, Gloucester
and Salem counties, New Jersey: U.S. Geological Survey Water-Resources Investigations Report 90-4142.
Martin, M., 1998. Ground-water flow in the New Jersey Coastal Plain: U.S. Geological Survey Professional
Paper 1404-H
Voronin, L.M., 2004. Documentation of revisions to the regional aquifer system analysis model of the New
Jersey coastal plain: U.S. Geological Survey Water-Resources Investigations Report 2003-4268.

276
Corresponding author: Rishi Parashar
Desert Research Institute
[email protected]

Generalized Responses of Fractured Rock Aquifers to Pumping and Episodic

Recharge

Rishi Parashar, Division of Hydrologic Sciences, Desert Research Institute, Reno, NV

Donald M. Reeves, Department of Geosciences, Western Michigan University, Kalamazoo, MI

Abstract: Fractured rock aquifers are becoming an increasingly important source of potable
water worldwide due to depletion of current water resources and development in mountain-
ous regions. Sustainable development and management of fractured rock aquifers is ex-
amined in the context of drawdown responses and water table fluctuations to pumping and
episodic recharge events. A combination of saturated discrete fracture network (DFN) mod-
els and variably-saturated dual-permeability flow models are used to investigate generalized
responses in both unconsolidated porous media aquifers and fractured rock aquifers. Steady-
state DFN simulations in fractured rock produce irregularly-shaped patterns of drawdown
that greatly differ from the circular or elliptical cone of depressions generally found in porous
media aquifers. Drawdown responses, including the asymmetry of the cone of depression,
are highly sensitive to hydraulic and geometrical attributes of the fracture network. Contin-
uum simulations demonstrate that episodic recharge events are rapidly transmitted to the
water table within fractured rock aquifers owing to low storativity, high transmissivity, and
steep unsaturated drainage profiles. Porous media systems under similar external forcing
conditions are more effective in modulating recharge to the water table over longer time
scales and are, in general, less susceptible to climatic variations and resource exploitation.

Extended abstract: Methodology, findings, and implications

Introduction and Characteristics of Fractured Rock Aquifers

Fractured rock aquifers are pervasive worldwide and differ from typical porous media
aquifers by presence of fractures serving as primary pathways for transmission of water in
usable quantities. Ground water flow systems consisting of highly-connected, dense networks
of fractures can serve as prolific aquifers with high permeability values. Understanding the
dynamics of ground water movement in fractured rocks, especially when fractured aquifers
are subject to extraction of large quantities of water and perturbations in recharge are vital
to improving our ability to ensure sustainability of ground water resources (Parashar and
Reeves, 2017).
Fractured rock aquifers have high transmissivity and low storativity as water can be
easily transmitted to a pumping well but the fracture themselves can store very little water.
Interconnected networks of fractures exhibit strong multi-scale variability in geometric and
hydraulic properties (e.g., de Dreuzy et al., 2001; Painter et al., 2002; Neuman, 2005, 2008;
Parashar and Reeves, 2012; Reeves et al. 2012). The degree of fracture connectivity plays
a major role in how fracture networks transmit water through an aquifer. The contrast

277
in hydraulic properties of fractured and porous media aquifers, particularly with respect to
water storage, causes fractured aquifers to be more susceptible on average than porous media
aquifers to dramatic water level changes in response to pumping and recharge events. In the
following, responses of fractured rock aquifers to pumping and episodic recharge are explored
using a discrete fracture network (DFN) modeling approach and dual-continua equivalent
porous media (EPM) simulations.

Responses of Fractured Rock Aquifers to Pumping

In contrast to the un-fractured porous media, fractured rock aquifers often violate as-
sumptions of the Theis equation which predicts circular or elliptical cones of depression.
Inspection of steady-state drawdown maps for fracture networks show irregular patterns
(see Fig. 1) that are considerably different from the cone of depressions formed in un-
fractured porous media. The pattern and expanse of drawdown maps are highly sensitive to
network properties, such as the exponent of power-law length distribution, fracture density,
and distribution of fracture transmissivity (Parashar and Reeves, 2017).
A two-dimensional DFN code built using the minimum residual method (MINRES) of
solving large linear systems iteratively (Parashar and Reeves, 2012) is used here to conduct
simulations for networks with varying density and hydraulic properties. DFN approaches are
designed to solve for flow in each individual fracture of the network by computing values of
hydraulic head at all intersection points. The networks are stochastically generated according
to specified distributions of spacing, orientation, length, and transmissivity. The boundary
conditions (usually given in the form of varying hydraulic head) are interpolated to assign
value of hydraulic head to the boundary nodes (intersection points of fracture segments and
the domain boundary). Steady state flow solutions for the two-dimensional networks are
computed by employing conservation of mass at each fracture intersection point and using
Darcy’s law to model flow in individual fracture segments.
Qualitative examination of the drawdown maps show that the drawdown patterns in
fractured systems exhibit a significant departure from classical circular or elliptical cones of
depression. The extent of departure increases with increasing values of power-law exponent
� used to describe the distribution of fracture segment lengths, and is more pronounced for
less dense networks. The shapes of drawdown maps are more irregular for networks with
distributed values of transmissivity (i.e., heterogeneous conditions) as opposed to network
of fractures with uniform hydraulic properties. Examining drawdown behavior of fractured
rock aquifers in response to pumping events, and comparing these DFN based techniques
to traditional approaches help in better understanding of the system and development of
strategies for sustainable use of groundwater resources.

Responses of Fractured Rock Aquifers to Episodic Recharge

Studying variations of water levels in aquifers in response to perturbations in recharge is

an important metric in managing groundwater. Using dual-continua EPM simulations, it is
shown that responses to varying recharge events are very rapid and amplified in fractured
rock aquifers. This is a result of low storativity, high transmissivity, and steep drainage
profiles of fractured rocks.

278
 A recharge flux is applied to the top boundary of a dual-continua two-dimensional EPM
model built using NUFT (Nitao and Sun, 2007). All other boundaries except the vertical
boundaries below the water table are considered no-flow. The boundary conditions represent
a system with a hydraulic divide that allows flow in either longitudinal direction. Fractured
rock aquifers are seen to have dramatic water level variations caused by only minor pertur-
bations in recharge. Low porosity of fracture networks leads to lack of aquifer storage, and
negligible capillarity of fractures results in rapid movement of water to the saturated zone.
Thus climatic signals propagate very fast to the water table in fracture dominated aquifers.
On the contrary, porous media aquifers with much greater storage capacity, exhibits very
little variation in the water table because of perturbations in recharge.

Figure 1. Example of drawdown pattern of hydraulic head in a moderately dense fracture rock aquifer
obtained using two-dimensional DFN simulations (from Parashar and Reeves, 2017).

References

de Dreuzy, J.R., Davy, P., Bour, O. (2001), Hydraulic properties of two-dimensional random fracture net-
works following a power-law length distribution: 1. Effective connectivity, Water Resour. Res., 37(8),
2065-2078.
Neuman, S.P. (2008), Multiscale relationships between fracture length, density and permeability, Geophys.
Res. Lett., 35(22), doi:1029/2008GL035622.
Neuman, S.P. (2005), Trends, prospects and challenges in quantifying flow and transport through fractured
rocks, Hydrogeo. J., 13, 124-147.
Nitao, J.J. and Sun, Y. (2007), USNT reference manual of the NUFT code for Yucca Mountain Project,
Lawrence Livermore National Laboratory.
Painter, S., Cvetkovic, V., and Selroos, J.-O. (2002), Power-law velocity variations in fracture networks: Nu-
merical evidence and implications for tracer transport, Geophys. Res. Lett., 29(14), doi:10.1029/2002GL014960.

279
Parashar, R. and D.M. Reeves (2017), Ground water sustainability in fractured rock aquifers, Chapter 17
In: Sustainable Water Resources Management, Eds. C.S.P. Ojha, S. Rao, A. Bardossy, T.C. Zhang, and
C.M. Kao, American Society of Civil Engineers (ASCE), Reston, VA, in press.
Parashar, R. and D. M. Reeves (2012), On iterative techniques for solving flow in large two-dimensional
discrete fracture networks, Journal of Computational and Applied Mathematics, 236(18), 4712-4724.
Reeves, D.M., R. Parashar, and Y. Zhang (2012), Hydrogeologic characterization of fractured rock masses
intended for disposal of radioactive waste, In: Radioactive Waste, Ed. R.A. Rahman, InTech Publishing,
ISBN 978-953-51-0551-0.

280
Corresponding author: Nathan Young
Iowa State University
[email protected]

Linking FracMan and HydroGeoSphere to simulate watershed-scale nitrate

transport in fractured till

Nathan L. Young, Iowa State University

W. Simpkins
R. Horton
R. McLaren
R. Therrien

Abstract: The effect of fractures on nitrate transport in till at the watershed scale is largely
unknown. To address this problem, we developed an EPM approach that upscales fracture
flow using an equivalent 3-D K tensor of a fractured till block from the Des Moines Lobe in
central Iowa. This was accomplished by using a linkage of FracMan and HydroGeoSphere.
FracMan was used to create a 64-m3 block of fractured till with a fracture intensity of 24.4
m2 /m3 , an average fracture spacing of 0.043 m, and a fracture porosity of 0.038ύ. Five
smaller blocks, ranging in size from 0.00375 m3 (.06 x .06 x .06 m) to 8 m3 (2 x 2 x 2 m),
were subsampled from the center of the 64 m3 block. These blocks were converted to a
finite-element mesh and HydroGeoSphere was used to model flow and transport through
each block. We identified the representative elementary volume (REV) of fractured till as
the point when the fluid flux along each of the component directions (x, y, and z) remained
constant despite increasing block size. The 3-D K tensor of the REV was computed using
published methods. Breakthrough curves were simulated using both the discretely-fractured
grids and a single-continuum, equivalent porous media (EPM) model based on the computed
K tensor. Preliminary results showed an average fracture velocity of 7E-05 m/s (6 m/d)
and an average linear (porous media) velocity of 8E-08 m/s (0.007 m/d). These results are
consistent with laboratory column tests from till core of similar size.

281
Corresponding author: Steven Berg
Aquanty Inc.
[email protected]

Simulating Complex Surface Water/Groundwater Interactions during Flood

Events with a Fully-Integrated Physics Based Hydrologic Model
Steven J. Berg, Aquanty Inc.
H-T Hwang
Y-J Park
S. Frey
E. Sudicky
N. Grosso
M. Sherrier

Abstract: The optimal design of reactive covers for treating contaminants seeping from
a riverbank to a stream requires a comprehensive understanding of the discharge of water
and solute fluxes along the bank profile. This is particularly true following flood events
that result in riverbank discharge rates back into the river during the recession period that
are significantly higher than baseline conditions. In addition to the highly dynamic nature
of a flood event, the design can be complicated by many additional factors that control
the movement and mixing of surface water and groundwater, such as channel geometry,
flood stage, and bank soil heterogeneity. Traditional modeling approaches used to describe
surface water/groundwater interactions during flood events typically rely on simplified an-
alytical solutions, or highly abstracted numerical solutions. Because these approaches often
idealize a flood event, simplify the river geometry, and ignore bank soil heterogeneity, the
resulting model is highly abstracted. To overcome these limitations a rigorous physics-based
numerical model was developed using HydroGeoSphere. This numerical model incorporates
channel geometry, heterogeneous bank soil distribution, and observed river stage data to
drive the model. The model was used to simulate the interaction of surface water and
groundwater within the riverbank soils during flood events, including in-channel and flood-
plain events. The interaction between surface water and groundwater was computed by
applying a conservative tracer to the surface water in order to track its migration into the
bank soils on the rising limb of the flood hydrograph, and its return to the channel as the
flood recedes.

282
Corresponding author: Loring Crowley
Schnabel Engineering
[email protected]

Design of Infiltration Basins for Cooling Tower Water Bleed and Pool Water
Backwash

Loring W. Crowley, Schnabel Engineering

E. Rehwoldt

Abstract: Bleed water from cooling tower structures and backwash from recreational pools
at an overseas location were designed for infiltration into the subsurface via stormwater
management (SWM) infiltration basins. A three-dimensional groundwater flow and solute
transport model was used to design the geometry and select the location of the basins.
The groundwater model assessed the groundwater infiltration potential to accommodate
expected flow rates, the emergence of infiltrated water on steep slopes that could affect
slope stability, the potential environmental impacts to potable groundwater aquifers, and
the impacts of seepage into the existing offsite stream channel downhill from the site. In
addition, limited space available on the site, varying topography, and surface water piping
slopes and layout were considered. The model results showed that a series of injection (dry)
wells connected to concrete vaults at four locations, separate from SWM basins, could ap-
propriately accommodate the site layout constraints, mitigate slope stability concerns, meet
cooling tower bleed and pool backwash inflow requirements, and protect both groundwater
and surface water quality.

283
Corresponding author: Daniel Feinstein
U.S. Geological Survey
[email protected]

What makes a MODFLOW-derived metamodel useful in decision support?

Daniel Feinstein, U.S. Geological Survey

M. Fienen
H. Reeves

Abstract: Metamodels are one method for distilling the physics of a MODFLOW model
into correlation structures inherent in the input and output so that its findings can be ex-
tended outside the modeled domain. Metamodels are attractive for decision-support tools,
where the goal is rapid application to management problems over large areas. The first
step in developing a metamodel is to demonstrate its ability to reproduce - with a limited
number of explanatory variables derived from the MODFLOW model - a finding of interest
within the modeled domain. Metamodel examples discussed include those that reproduce
MODFLOW findings relative to: 1) the contribution of headwater streams to shallow pump-
ing, 2) the age distribution of shallow well discharge, and 3) the less tractable problem of
quantifying the baseflow depletion in headwater streams. The second step is to translate the
explanatory variables used in the metamodel to predictors that can be readily mapped across
the landscape, thereby allowing it to be applied widely. The reliability of the application
is subject to various tests. The transferability of the metamodel can be tested by applying
it to other hydrogeologically distinct domains where its results can be compared to other
MODFLOW models. The reproducibility of the metamodel can be tested by applying it
back to the original MODFLOW domain using only the mappable predictors. Meeting these
tests may require that the predictors themselves be modified (e.g., by scaling or censoring)
as illustrated with an application of a metamodel derived from the USGS Lake Michigan
Basin model.

284
Corresponding author: Stefan Finsterle
Finsterle GeoConsulting
[email protected]

Boundaries, Interfaces and Discontinuities: Where Things Happen!

Stefan Finsterle, Finsterle GeoConsulting

Abstract: Numerical models aim at solving governing equations representing processes that
are considered relevant within the physical domain of interest. Processes occurring outside
the model domain are either prescribed as boundary conditions, or their effects are supplied
through a link to another, specialized submodel. Moreover, simplifying assumptions are
made about the exchange of fluids, energy, and information across discontinuities within
the model domain. While it is recognized that boundaries, interfaces, and discontinuities
play a crucial role in the definition of groundwater flow problems (which are classified as
initial and boundary value problems), their impact on the solution is often underestimated.
The purpose of this presentation is to highlight the ubiquity, variety, and hidden complexity
of these elements, so we can pay due attention to their influence and avoid potential pitfalls
when crossing them in our daily modeling work.

285
Corresponding author: Mary Forrester
Colorado School of Mines
[email protected]

The role of hydrology in meteorological modeling: A case study on

insect-induced forest disturbance and its feedbacks from groundwater to
atmosphere
Mary M. Forrester, Colorado School of Mines
R. Maxwell
L. Bearup
D. Gochis

Abstract: Scientists frequently turn to physically-based hydrologic, land surface, or atmo-

spheric models to diagnose earth system interactions, interpret observations, and predict
extreme or hazardous events. Water, land, and atmosphere are often treated as isolated
systems, and rarely are all three coupled in a full representation of water and energy cycles.
Here, we present simulations that couple the Weather Research and Forecasting (WRF)
model, a mesoscale meteorological model, with ParFlow (PF), an integrated subsurface and
surface hydrologic model. PF-WRF was used to identify ecohydrologic and atmospheric
response to land disturbance by the mountain pine beetle (MPB) over the Colorado Head-
waters region. Our results support previous literature showing response to forest mortality,
including reduced transpiration, energy repartitioning, higher water tables, and higher plan-
etary boundary layer heights. We also find significant damping from the local scale to the
watershed scale, with the addition of unforested and unimpacted regions. Diurnal signals
play a large role in damping MPB, by lagging the response as disturbance moves from
surface to atmosphere. Most importantly, although infestation-induced perturbations to
convection and precipitation are insignificant when analyzed with ensemble methodologies,
we find that both energy and atmospheric processes are most sensitive to disturbance in re-
gions with higher water tables. When the PF component with dynamic subsurface hydrology
is removed, WRF’s simulated energy and atmospheric processes show greater impacts from
MPB. Groundwater representation in atmospheric modeling has been shown to affect the
response intensity of a major land disturbance event, a finding that has huge implications
to computational hydrology and meteorology.

286
Corresponding author: Pytrik Graafstra
Hydrology Department, Waternet, Vogelenzangseweg 21, 2114 BA Vogelenzang, The Netherlands
[email protected]

Incorporating insights from Time Series Analysis in groundwater modelling for

the urban area of the city of Amsterdam
Pytrik Graafstra, Hydrology Department, Waternet, The Netherlands
Frank Smits, Hydrology Department, Waternet, The Netherlands; Department of Water Management,
Water Resources Section, Faculty of Civil Engineering and Earth Sciences, TU Delft, The Netherlands
Theo Janse, Operations Research Department, Waternet, The Netherlands.

Abstract: As the public water authority of the city of Amsterdam and surrounding areas,
Waternet makes use of both steady-state and transient groundwater models for a variety
of purposes involving urban groundwater management. For instance when determining the
effect of planned measures on the occurrence of groundwater levels that are too low or too
high, potentially resulting in degradation of wooden pile foundations or mold in houses,
respectively. One of the challenges we face in that modelling, is the discrepancy between
the physical processes that play a part in the city, influencing groundwater measurements,
and the extent to which we are able to quantify those processes in groundwater models. We
addressed this discrepancy with automated Time Series Analysis (TSA). In the over 3000
time series of groundwater measurements that were analyzed, TSA often identified one or
more disturbances, such as groundwater extractions and measurement error. Along with
identifying disturbances and artifacts, TSA offers a way to structurally address these issues.
Incorporating (automated) TSA in generating model observations also results in a uniform,
reproducible approach, and the ability to evaluate enormous amounts of monitoring wells.
We argue that this approach is preferable to visual inspection and evaluation of measurement
series, and discuss ways to incorporate these insights into groundwater models.

Extended abstract: Methodology, findings, and implications

Introduction

As the public water authority of the city of Amsterdam and surrounding areas, Waternet
uses both steady-state and transient groundwater models for a variety of purposes involv-
ing urban groundwater management. For instance when determining the effect of planned
measures on the occurrence of groundwater levels that are too low or too high, potentially
resulting in degradation of wooden pile foundations or mold in houses, respectively. Other
uses include reviewing permit applications for groundwater extraction for building pits, and
determining the effectiveness of climate change mitigation measures. One of the challenges
we face in that modelling, is the discrepancy between the physical processes that play a
part in the city, influencing groundwater measurements, and the extent to which we are
able to quantify those processes in groundwater models. While this discrepancy between
groundwater models and measurements can be addressed manually, we opted for automated
Time Series Analysis (TSA) (Chatfield 1975; Box and Jenkins 1970). In the over 3000 time
series of groundwater measurements that were analyzed, TSA often identified one or more
disturbances.

287
Identifying measurement disturbances and artifacts through TSA

Both measurement disturbances (a physical process which influences measurements, but

is not represented in the groundwater model), and artifacts (disturbances created by hu-
man error or computational methods) can be a cause for discrepancy between groundwater
models and available measurements. In the case of steady-state models, these discrepancies
and their origin may be further obscured because measurements are aggregated into a single
observation. Of course, these issues can be identified manually, but visual inspection for
all monitoring wells is a time-consuming, tedious and subjective process. Waternet actively
measures over 3000 phreatic monitoring wells and 300 monitoring wells in deeper aquifers,
any of which can be subject to one or more of these disturbances. Additionally, repro-
ducibility of analyses and results is of increasing importance, as upcoming legislation in the
Netherlands will require argumentation when excluding available measurements from anal-
ysis (Ministerie van Infrastructuur en Milieu 2017). There are many different models that
can be used for TSA. In this case, we used a basic AR1-model with rainfall and evaporation
series as explanatory variables (Janse and Graafstra 2015). In order to deal with the effect
of unequally spaced measurements on the TSA model, innovations were incorporated (Von
Asmuth and Bierkens 2005).

Disturbances and artifacts that we were able to identify using TSA, are:

D1 Groundwater extractions (Figure 1).

Amsterdam has relatively high groundwater levels. Because of this, building sites that
include a subsurface component make use of extraction wells to temporarily lower the
groundwater table at and around the building site. At any given time, numerous
groundwater extractions of varying sizes are active in the city. The details necessary
to include these components in groundwater models are often not available, but their
effects may influence the groundwater measurements used as observations for ground-
water models.

D2 The presence of leaky sewers (Figure 2A).

The sewer system of Amsterdam is mostly located below the groundwater table. When
parts of the system become damaged, they can drain groundwater from their surround-
ing area. Oftentimes, this starts as a gradual process. As the damage progresses, or
other parts of the sewer in the vicinity become damaged as well, the lowering of the
groundwater table becomes more pronounced. When leaky sewers are repaired, an
increase of the groundwater table can often be seen.

D3 Measurement error (Figure 2B).

These errors include read-out or typing errors, and measuring the wrong monitoring
well.

D4 Varying measurement frequencies or gaps in a measurement series (Figure 2C).

When measurement frequencies are not constant throughout the year or the entire
modelling period, a single head based on the average of available measurements can

288
 lead to bias (Hill and Tiedeman 2007). Even weighted averaging may not always
resolve this problem, when measurements are unavailable over an extended period of
time. Especially when paired with other disturbances, such as leaky sewers (D2),
differences in the years in which measurements are carried out can lead to artifacts in
generated observations. Measurement data of monitoring well A may be carried out
before the sewer started leaking, while monitoring well B has data available only after
the leaking started.

Addressing measurement disturbances and artifacts

Along with identifying disturbances and artifacts, TSA offers a way to structurally ad-
dress these issues. Information obtained from TSA models can be included in the ground-
water model optimization process in one of the following ways, or a combination thereof:

S1 Adjusting observation weights based on residual analysis (Figure 1).

By adjusting the weight of an observation as a function of the TSA model residual,
less - or zero - emphasis can be put on sections of the series where, for instance, a
groundwater extraction was present. This method can also be used to subsequently
derive a (weighted) observation for a steady-state groundwater model.

S2 Decreasing the time-span of measurements used to calculate an average representative

head (Figure 2A).
When faced with measurement series displaying trends over time, using a subset of the
measurements when calculating a representative head will allow for a better (spatial)
interpretation of the presence of, for instance, leaking sewers.

S3 Incorporating TSA model outcome as input for generating observations (Figure 2C).
The TSA model can be used as addition to, or even replacement of, the measurements
used to generate groundwater model observations. Especially when dealing with vary-
ing measurement frequencies, this technique can be used to reduce bias.

Conclusion and discussion

Time Series Analysis (TSA) is not just a valuable tool when identifying measurement
disturbances (D1-D2) and artifacts (D3-D4), but also when addressing them (S1-S3). Aside
from qualitative advantages, incorporating (automated) TSA in generating model observa-
tions also results in a uniform, reproducible approach, and the ability to evaluate enormous
amounts of monitoring wells. However, care must be taken in choosing the means by which
to incorporate TSA when generating observations; leaning too heavily on TSA alone can
polish away the capricious nature of a groundwater system, especially when the amount of
measurements on which the TSA model is based, is limited.

289
Identification of groundwater extractions (D1). A: Measurements and TSA model for monitoring well
E05109A and E05744A. An extraction can be seen in the measurements of E05109A, while E05744 shows a
rise in its groundwater table. The rise is caused by extracted water being infiltrated back into the aquifer a
short distance from the site, reducing the risk of wooden foundation pile degradation. A distinct difference
in groundwater level can be observed in the period before and after the extraction. B: Residuals can be used
to adjust observation weights (S1). C: Five wells in an area where part of a sewer was replaced; three show
a temporary lowering, and two a rise of the groundwater table.

290
Figure 2. Identifying and addressing disturbances and artifacts. A: Identification of a leaky sewer (D2).
The mean of a subset was calculated (AMSL -1.27 m in 2010), instead of the overall mean (AMSL -1.00 m).
A difference of 27 cm seems minor, but the difference between too high and too low groundwater levels in
Amsterdam, on average, is only 90 cm. B: Assessing the plausibility of measurements (D3). Three lie outside
the 99.7 ύ confidence bandwidth, whose weights can be reduced or set to zero (S1). C: No measurements
were available for the summer of 2003 (D4). A corrected mean of the entire series (2000-2006) was calculate
(AMSL -2.07 m), instead of measurement mean (AMSL -1.74 m) (S3).

References

Box, G.E.P., and G.M. Jenkins. 1970. Time Series Analysis: Forecasting and Control. Holden-Day.
Chatfield, C. 1975. The analysis of time-series: Theory and practice. Springer US.
Hill, M.C., and C.R. Tiedeman. 2007. Effective groundwater model calibration. Wiley.
Janse, T., and P. Graafstra. How to get the most out of your historical groundwater data using a basic time
series approach. AIWW Conference Proceedings. 2015.
Ministerie van Infrastructuur en Milieu. FAQ’s BRO. https://bro.pleio.nl/file/download/44596202 (accessed
May 5, 2017).
Von Asmuth, J.R., and M.F.P. Bierkens. 2005. Modeling irregularly spaced residual series as a continuous
stochastic process. Water Resources Research 41, W12404. DOI:10.1029/2004WR003726.

291
Corresponding author: Maksym Gusyev
ICHARM, PWRI, Tsukuba, Japan
[email protected]

Coupling MODFLOW and distributed hydrologic model BTOP in the Fujikawa

River basin

Maksym A. Gusyev, ICHARM, PWRI, Tsukuba, Japan

Daniel Abrams, Illinois State Water Survey, Champaign, IL, USA
Jun Magome, University of Yamanashi, Kofu, Japan
Yoshio Tokunaga, ICHARM, PWRI, Tsukuba, Japan

Abstract: Distributed hydrologic models (DHMs) are widely utilized for simulating hy-
drological processes on a river basin scale while having greatly simplified groundwater flow
dynamics. This study is the first attempt to investigate a coupling of MODFLOW and
BTOP models and we used an existing BTOP model in the Fujikawa River basin, Japan.
The BTOP model, which is based on a topographic index approach of TOPMODEL with
river flow routing, has been applied globally with coarse grids of about 20-km as well as
at much smaller scales at several river basins. In our first investigation we attempted se-
quential linkage and full coupling between these two models. For the sequential linkage, the
BTOP model is simulated on a 30-arcsecond (about 1-km) grid in the Fujikawa basin and
provides recharge to the MODFLOW model for the groundwater simulations and baseflow
contribution to river discharge at each grid. Having iterative coupling between these two
models would allow us to further investigate surface-groundwater interactions, groundwater
abstraction as well as movement of environmental isotopes such as tritium in various river
basins in the future studies.

Extended abstract: Methodology, findings, and implications

Coupling MODFLOW and distributed hydrologic models (DHMs) is becoming a popular

approach to simulate impacts of water management scenarios on groundwater heads and
flows. MODFLOW allows representation of intricate details of subsurface features while
DHMs are commonly utilized for simulating surface hydrologic processes on a river basin
scale with a greatly simplified conceptualization of complex groundwater flow dynamics. For
example, a Block-wise TOPMODEL (BTOP) DHM has been applied in various river basins
with small grids (Gusyev et al., 2016a,b) as well as globally on 10-arcminute (about 20-km)
grid (Magome et al., 2015). In each BTOP grid, groundwater heads are incorporated using
a saturation deficit estimated with a topographic index, which is scaled with a cumulative
distribution function of groundwater travel distance from an upstream contributing area,
while lateral groundwater flows between grids are not considered (Takeuchi et al., 2008).
These BTOP model simplifications need to be evaluated by including a MODFLOW model
to represent subsurface groundwater movement in various river basins.
This study is the first attempt to develop a procedure of coupling of MODFLOW and
BTOP models. We used an existing BTOP model of the Fujikawa River basin, Japan, which
is provided in the BTOP user manual with supplementary tools used for the model coupling

292
(Gusyev et al., 2017). The MODFLOW model was setup in a steady-state with Groundwater
Vista 6 using the BTOP model files of the Fujikawa basin project. For the Fujikawa basin,
the BTOP model with 120 rows and 120 columns and 30-arcsecond (about 1-km) grids was
used to setup the 2-layer MODFLOW model, which had the same extent and grids with
variable length and width. The boundary of the BTOP and MODFLOW models is shown
in Figure 1. Figure 1a demonstrates the BTOP model digital elevation model (DEM) of the
Fujikawa basin that was used as a top layer elevation for MODFLOW model. The bottom
elevation of Layer 1 and 2 were assigned at 0 meters above sea level (m a.s.l.) and -100 m
a.s.l., respectively. In each active MODFLOW grid, we assigned a drain cell with a drain
bottom elevation estimated as 1 m below the DEM and a drain conductance, which was
estimated from the BTOP model river width and length at each grid. The water elevations
of these drains were imported from the surface water network of the BTOP model that is
equivalent to MODFLOW drains. In the Fujikawa River basin, the BTOP model simulated
river discharge on daily time steps using observed precipitation and meteorological gauging
station data between January 1st and December 31st 1999. Figure 1b shows the BTOP
simulated water infiltration in the root zone on January 18th. Other BTOP simulated fluxes
from root zone to unsaturated zone are not shown. A sequential linkage was implemented
by providing BTOP simulated root zone water infiltration fluxes as the MODFLOW gridded
recharge assigned to the top of Layer 1. The MODFLOW model was run in the steady-state
without calibration to obtain groundwater heads and velocities as well as drain inflows.
Figure 2 demonstrates the MODFLOW model results of sequential linkage with BTOP
inputs during winter baseflows on January 18th 1999. In Figure 2a, simulated groundwater
heads are below the DEM terrain surface shown in Figure 1a at the headwater catchments
and above the terrain along the main channel of the river valley. Where these simulated
heads are above the drain water elevation, drains receive groundwater inflows as demon-
strated by negative inflow values in Figure 2b. These MODFLOW simulated inflows occur
at the main river channel and have similar pattern to the BTOP model simulated ground-
water inflows without coupling (not shown). Despite the spatial similarity of these inflows,
the MODFLOW drain inflows are larger than the BTOP model inflows for the upstream
catchment of the Kitomatsuno River gauging station and can be explained by overestimated
recharge assigned to the MODFLOW model. Since the MODFLW model is bounded by
no-flow boundaries, the net recharge entering the model is equal to the flow out of the drain
cells. Having other water fluxes such as root zone or unsaturated zone drainages should be
used in the sequential version of the BTOP and MODFLOW coupling, but it will require
a transient MODFLOW simulation to follow the transient BTOP model simulation. For
example, BTOP simulated root zone and unsaturated zone drainage fluxes are negligible
for almost all grids on January 18th and the transient MODFLOW simulation allows to
accommodate zero recharge. The net inflow of water into the MODFLOW model is a fixed
value in the steady state simulation and the other groundwater sinks such as drain cells
can only remove water from the aquifer. For the next step, MODFLOW river cells may be
also considered to account for surface-groundwater interactions, where groundwater heads
are less that the drain water elevation. Such disconnected drain cells are shown in Figure
2b. These river cells along the main channel would recharge the groundwater and this extra
source of water, in turn, would influence the BTOP simulation. Consequently, an iterative
coupling of MODFLOW and BTOP models is needed allowing for the joint calibration of

293
river discharge and groundwater heads data of both models. Finally, having iterative cou-
pling between these two models would allow us to investigate groundwater abstraction as
well as movement of environmental isotopes such as tritium in various river basins in the
future studies.

294
Figure 1. The BTOP model of Fujikawa River basin digital terrain elevation (DEM) with 30-arcsec (about
1-km) grids a) and BTOP simulated water infiltration in the root zone on January 18th 1999.

295
Figure 2: MODFLOW simulated steady-state groundwater heads a) and drain inflows b) with the infiltration
in the root zone on January 18th 1999 from the BTOP model.

296
 References

Gusyev, M.A., Hasegawa, A., Magome, J., Sanchez, P., Sugiura, A., Sawano, H. and Tokunaga, Y. (2016a).
Evaluation of water cycle components with standardized indices under climate change in the Pampanga,
Solo and Chao Phraya basins. Journal of Disaster Research 11(6): 1091-1102, doi: 10.20965/jdr.2016.p1091
Gusyev, M.A., Gädeke, A., Cullmann, J., Magome, J., Sugiura, A., Sawano, H. and Takeuchi, K. (2016b).
Connecting global and local scale flood risk assessment: A case study of the Rhine River basin flood hazard.
Journal of Flood Risk Management 9(4): 343-354, doi: 10.1111/jfr3.12243
Gusyev, M., Magome, J., Kiem, A., and Takeuchi K. (2017). The BTOP Model with Supplementary Tools:
User Manual. Technical Note of PWRI No. 4357, Public Works Research Institute, Tsukuba, Japan, ISSN
0386-5878, 71 p.
Magome, J., Gusyev, M.A., Hasegawa, A. and Takeuchi, K. (2015). River discharge simulation of a dis-
tributed hydrological model on global scale for the hazard quantification. In Weber, T., McPhee, M.J. and
Anderssen, R.S. (eds) MODSIM2015, 21st International Congress on Modelling and Simulation. Modelling
and Simulation Society of Australia and New Zealand, December 2015: 1593-1599 pp. ISBN: 978-0-9872143-
5-5.
Takeuchi, K., Hapuarachchi, P., Zhou, M., Ishidaira, H. and Magome, J. (2008). A BTOP model to extend
TOPMODEL for distributed hydrological simulation of large basins. Hydrological Processes 22: 3236–3251.
doi: 10.1002/hyp.6910

297
Corresponding author: Rodrigo Herrera
Arcadis Chile
[email protected]

Using MODFLOW-USG to simulate the rise of an existing tailing storage

facility and estimate transient seepage

Rodrigo Herrera, Arcadis Chile

Jose Lagos, Arcadis Chile
Tomas Opazo, HMI Consulting

Abstract: Seepage estimation from tailings storage facilities (TSF) is critical for both op-
erational and environmental assessment purposes. In many cases, this is done indirectly by
developing a water budget for the TSF. In other cases, 1D or 2D numerical models are used
to simulate unsaturated flow through the tailings and directly estimating seepage. In this
work, we present a 3D MODFLOW-USG transient flow model that was constructed and
calibrated against data from an existing TSF to estimate seepage at monthly scale. The
numerical model consists of 37 layers with 20 m cells. The bottom 5 layers host the aquifer
and the upper 32 layers explicitly accommodate the tailings bathymetry and the impound-
ment’s pond water elevation. TSF layers were sequentially activated using the Transient
IBOUND package (TIB) and tailings permeability was reduced over time using the Time-
Variant Materials package (TVM2) to simulate consolidation. This allowed seepage to be
simulated implicitly. The model was calibrated using PEST against the impoundment’s
drain outflow and groundwater levels in downstream monitoring wells. Calibration used
a combination of pilot points to estimate the alluvial materials hydraulic properties, and
a zone-based approach for the tailing’s initial permeability. The 3D nature of the model
allowed capturing the transient response of measured drain outflows, suggesting the method-
ology was able to reproduce the close interrelationship between the 3D groundwater system
and the impoundment management that control seepage rates at this site.

Extended abstract: Methodology, findings, and implications

Introduction
Estimation of seepage from tailings storage facilities (TSF) is critical both for operational
and environmental assessment purposes. From the operational perspective, it is relevant for
water recovery at the TSF, especially in dry climates. From the environmental perspective,
seepage can create unwanted effects on solute concentrations downstream from the TSF,
requiring control and preventive measurements. Development of a water balance for the
impoundment is usually one of the first steps to estimate seepage. Its value is calculated
as the difference between inflows, outflows and water stored at the TSF. Other alternative
consists in analyzing the aquifer water level fluctuations and through inverse modeling cali-
brate the recharge rate that produces that effect. The use of 1D or 2D numerical models is
a common practice. Modeling can be performed using fully-saturated or variably-saturated
flow conditions. It may also consider the dependency between tailings hydraulic properties
and seepage rates, because of the consolidation process. In this work, we present a novel

298
approach to estimate seepage. It involves the use of a numerical model that encompasses
a 3D groundwater model and a 3D representation of the impoundment management, in
one integrated simulation. This approach was developed and implemented for an existing
TSF located in northern Chile in a mountainous area of a semi-arid regional basin that had
enough information to validate the approach.
Hydrogeologic Framework
Geological/hydrogeological units at the site include, from bottom to top, a Paleozoic
granitic basement, a Triassic andesite volcano-clastic sequence, and Jurassic sandstone and
conglomerate sedimentary sequence. Unconsolidated deposits, mainly coluvial and fluvial-
alluvial quaternary sediments are present along streams where the impoundment is located
(Figure 1a). The main stream, at the dam location, has at least 4 tributaries with vari-
ous orientations and different permeability values. Pre-TSF water table was located at the
overburden/rock interface, so that the alluvial sediments underneath the impoundment were
dry. After six months of TSF operation water level in a monitoring well located 1.5 km
downstream of the dam rose to its maximum value, suggesting the unsaturated zone under-
neath the tailings became fully saturated in a brief period and travel time within the alluvial
sediments was also short. Within four months of operation drain outflow started to occur
and after seven and twenty months it peaked for the first and second time, respectively. All
of this suggested a close interrelation between the morphological/hydraulic properties of the
site and the impoundment management so that a 3D modeling framework was required.
Numerical model setup
A 37-layer 3D MODFLOW-USG (Panday et al, 2013) numerical flow model was imple-
mented using Groundwater Vistas 6 (ESI). 5 layers host the hydrogeological units and, on
top of them, 32 accommodate the tailings bathymetry, permeability and pond elevation. A
20 m uniform grid size was considered (Figure 1b). A mixed steady-state/transient simula-
tion with 37 stress periods was set up, simulating 2.5 years of TSF operation at a monthly
scale. TSF layers gradually become active, one per stress period, from bottom to top, as
tailings increase in size and thickness, using the Transient IBOUND package (TIB). Upon
activation a constant head boundary condition (CHD BC) was applied to the TSF layer rep-
resenting pond elevation for that month. Once simulation advances to the next stress period
the CHD BC at the previous (lower) layer was released, making cells active, and a new CHD
BC takes in place in the activating layer. This procedure is repeated until the end of simula-
tion. Tailings properties were represented by an initial permeability, a void ratio (expressed
as porosity) and a sequence of consolidation. Tailings initial permeability was included in
the model using zones, that were defined based on a criterion of columns of similar final
thickness (Figure 1c). The consolidation process was represented only as a permeability re-
duction. The permeability reduction sequence was constructed from data collected at the site
and from specialized bibliography, and implemented in the model with the HydroAlgorith-
mics (2016) Time-Variant Materials (TVM2) package. Other important elements simulated
were finger drains of the impoundment dam and 13 pumping wells distributed along the main
stream and a cut-off trench located 1.6 Km downgradient from the dam. Calibration used a
combination of pilot points to estimate the alluvium hydraulic properties and a zone-based
approach for the tailing’s initial permeability. Because the TVM package uses permeabil-
ity value instead reduction factors, in order to calibrate the tailings initial permeability a
utility was written to update the TVM package between PEST iterations. Calibration was

299
constrained by drain outflow as the main target, supplemented by heads, head differences,
pumping rates, pumping rate differences and drain outflow differences.
Results and conclusions
The 3D nature of the model allowed capturing the transient response of measured drain
outflows, suggesting the methodology was able to reproduce the close interrelationship be-
tween the groundwater system and the impoundment management (Figure 2A). Estimated
seepage (Figure 2B) is mainly controlled by the site´s geomorphology and its relation with
the TSF pond volume and location, the hydraulic conductivity of the alluvial materials and
the tailings properties (initial permeability and consolidation sequence). The use of PEST
allowed to calibrate tailings initial permeability, something that is not common practice
for this type of seepage estimations. One key result of this exercise was the ‘prediction’ of
relatively higher permeability zone within the tailings materials that was not possible to
be identified using the standard manual trial-and-error calibration, providing one possible
explanation to the sharp increase in drain outflow. This implementation also facilitates the
use PEST-based uncertainty analysis methodologies to assess potential error of prediction of
interest such as optimizing pond location to minimize infiltration. Also has the potential to
allow the environmentally/operationally-constrained TSF management be based in a more
robust and comprehensive analysis.

Figure 1: Conceptual and numerical implementation

300
Figure 2: Model results. Figure 2A: Normalized drain outflow, comparing between simulated and measured
data. Comparison shows a reasonable agreement between both series. Data normalization was done using
the maximum value of the measured data series. Figure 2B: Normalized TSF simulated seepage. The shape
of seepage shows a strong correlation with the impound management. Sharp increases in seepage usually
relate to tailings encountering alluvial materials. Data normalization was done using the maximum simulated
value.

References

Doherty (2016). PEST Model-Independent Parameter Estimation. User Manual Part I and II. 6th Edition
published in 2016.
Environmental Simulations Inc (ESI) (2011). A Guide to Using Groundwater Vistas Version 6.
HydroAlgorithmics (2016). Time-Variant Materials Package (TVM2). Unpublished manuscript. February
2016.
Panday et al (2013). MODFLOW-USG Version 1: An Unstructured Grid Version of MODFLOW for
Simulating Groundwater Flow and Tightly Coupled Processes Using a Control Volume Finite-Difference
Formulation. Techniques and Methods 6-A45. United State Geological Survey.

301
Corresponding author: Randy Hunt
U.S. Geological Survey
[email protected]

Beyond hydrology: Application and uncertainty analysis of watershed heat

transport

Randy Hunt, U.S. Geological Survey

D. Feinstein
W. Selbig

Abstract: In many cases, the objective of hydrologic modeling is more than simulation of
current conditions, which are already well-known. Commonly, the goal is to make forecasts
of a range of outputs - some non-hydrologic - that are derived from hydrologic simula-
tions. However, realistic simulation, even if limited to only hydrologic components, can
result in: 1) a larger number of parameters, insensitivity, and correlation; 2) longer model
runtimes and increased model instability; and 3) large, transient datasets with redundant
information, which decreases the signal-to-noise ratio of the dataset. Multi-process simu-
lations extending beyond hydrologic components (for example transport-related or ecohy-
drological simulations) augment these challenges. How best to obtain useful model outputs
and representative uncertainty estimates with these constraints? A well-calibrated coupled
GSFLOW-SNTEMP model of a watershed in a humid temperature climate is used to fore-
cast watershed-scale heat transport employing a range of tools and levels of sophistication,
from historical regression to transport modeling. The work is used to evaluate levels of com-
plexity needed to successfully address a set of societally relevant questions. An important
extension of the work is identifying potential computationally efficient surrogate models
that can be leveraged to assess uncertainty in ways that are impractical or impossible using
sophisticated models alone. Given that the success of modeling depends on keeping what is
needed and omitting what is not, such insights will be important as coupled and integrated
models are increasingly applied to more complex questions.

302
Corresponding author: William Hutchings
GHD Services, Inc.
[email protected]

Perspectives on the Potential Migration of Fluids Associated with Hydraulic

Fracturing in Southwest Florida

William C. Hutchings, GHD Services, Inc., 2675 Winkler Avenue, Suite 180, Fort Myers, Florida 33901

Abstract: The variable-density flow model, SEAWAT Version 4, was used to evaluate the
hydrogeological conditions associated with hydraulic fracturing the limestone oil reservoir in
the Lower Cretaceous Sunniland Formation of Southwest Florida. Variations in hydraulic
head, hydraulic conductivity, total dissolved solids, temperature, heat flow, heat conduction,
and viscosity were used in the construction of a 52-layer flow and transport model extending
from land surface to approximately 12,000 feet below land surface. Both under-pressured
and over-pressured conditions were simulated. A hydraulic fracturing methodology used
recently in this formation was used as the base case simulation. Multiple horizontal stage
fracturing using typical stress periods, a design fracture zone radius, and various injection
rates were tested to evaluate the potential for horizontal and vertical fluid migration in
and from the reservoir under dynamic conditions, with TDS used as a tracer. Hypothetical
scenarios including preferential vertical pathways between the Sunniland Formation and
the Lower Floridan aquifer Boulder Zone were also simulated. Results indicate that in-
jected fluids do not migrate significantly in the lateral and vertical directions beyond the
design fractured zone, unless a preferential pathway exists within close proximity to the
fractured zone. In a worst-case scenario under the simulated conditions, vertical heads are
approximately 400 meters greater than static conditions and fracking fluids vertically mi-
grate approximately 500 meters; therefore, the quality of the deepest sources of drinking
water are not compromised.

Extended abstract: Methodology, findings, and implications

Introduction

Interest in the environmental effects of hydraulic fracturing has become an increasingly

important topic associated with the development of oil and natural gas because of the po-
tential risk to the quality of shallow and deep aquifers. Although hydraulic fracturing has
been in use since the 1960s, the ability of this technique to significantly enhance recovery
of oil and natural gas from low permeability reservoirs and shale, has currently led to its
widespread use in the US and other petroleum-producing countries. Hydraulic fracturing is
generally an environmentally safe practice because of the geologic conditions under which
it is employed; however, in some instances hydraulic fracturing has been associated with
inadvertent groundwater contamination and other health hazards. The USEPA has sum-
marized the development and potential adverse effects of hydraulic fracturing in a recent
publication that identifies the most likely practices that result in groundwater contamina-
tion. In addition, the USGS has identified cases of hydraulic fracturing that have resulted
in environmental impacts. The USEPA (2016) concluded that impacts to aquifers from the
injection of fluids into oil reservoirs and subsequent migration is not a known major source
of groundwater contamination. Several studies have also evaluated the hydrogeologic condi-
tions under which hydraulic fracturing takes place and have also concluded that in general

303
the vertical migration of injected fluids to groundwater aquifers is unlikely (Birdsell et al.,
2015; Flewelling and Sharma, 2014).
This study is based on the use of the numerical model SEAWAT Version 4 (Langevin et
al., 2008) to simulate the hydrogeologic conditions and hydraulic fracturing technology in
the Sunniland Formation of Southwest Florida. The Lower Cretaceous Sunniland Formation
is an oil-producing trend within the South Florida Basin, half of which occurs within onshore
Florida peninsula. The trend is approximately 145 miles long and 12 miles wide, extending
from Sarasota to Dade counties. Production began in 1943, with the eventual discovery of
11 fields through 1984. The Sunniland Formation occurs at a depth of approximately 12,000
feet below land surface, and is composed of 250 feet of limestone, dolomite, and anhydrite.
Exploration and production continues to date in the Sunniland Formation; however, hy-
draulic fracturing is not permitted, although adverse effects have never been encountered. A
single instance of hydraulic fracturing took place at the Collier-Hogan 20-3H oil well located
in Collier County in December 2013 that resulted in the immediate termination of further
development, due to non-permitted activity, i.e. hydraulic fracturing. This study is an as-
sessment of the potential migration of the injected fluids from the reservoir associated with
this event; however, this study could be used as an indication of the general behaviour of
hydraulic fracturing fluids under similar hydrogeologic conditions.
Research Methods
The hydraulic fracturing that was conducted was designed with the MFrac3D Simulator
that predicted fractures 4.378 m above the lateral and 17.941 m below the lateral for a total
of fracture height of 22.352 m. A total slurry volume of 691,068 gallons (16,454 barrels or
bbls) were used for the workover that was conducted in seven (7) stages. The total water
volume amounted to 662,298 gallons. On average, treatment pressures ranged from 8,287 to
8,397 psi gallons per minute and the injection rate per stage of 597 gpm. The injection time
per stage was equal to approximately 0.11 days, for a total treatment period of 2.09 days.
The finite difference grid for the three dimensional model consisted of a horizontal do-
main with dimensions of 1005 by 520 meters discretized into 256 columns and 74 rows. The
model consisted of 52 layers (variable thickness) with constant head boundaries set up along
the west and east sides of the model. The remaining two sides were designed as “no flow”
boundaries. An expanding grid was used with rows and columns increasing from 3.05 meters
in the vicinity of the horizontal injection well to 40 meters at the outer edges of the model.
The constant head boundaries ranged from 6.0 meters NGVD for the water table to -710
meters in the Sunniland Formation for under-pressured conditions. A westward hydraulic
gradient of 0.0001 was set up within each layer. For over-pressured conditions, a hydraulic
head of 6.0 meters was assigned to the Sunniland Formation. Total dissolved solids (TDS)
was assigned to the model, with 500 milligrams per Liter (mg/L) to the potable water of
the surficial aquifer, 35,000 mg/L to the Boulder Zone, and 270,000 m/L to the Sunniland
Formation (Figure 1, top). Representative values of TDS were assigned to the Floridan
aquifer and a linear gradient was used between the Boulder Zone and the Sunniland For-
mation. Temperature ranged from 24 to 100 degrees centigrade, with a gradient of 20.5
°C/Kilometer assigned to the model. Density ranged from 1000 to 1181 kg/m3 (𝛿𝜌/𝛿𝑇 -2.40
kg/(m3 °C). Viscosity varied throughout the model according to 𝛿𝜇/𝛿C 1.92 × 10-6 m2 /d.
Density and salinity varied according to the relationship, 𝛿𝜌/𝛿C 0.7.

304
 The simulated fracture was assigned to layer 52 with a length of 545 meters, width of
15.25 meters, and depth of 23 meters (Figure 1, bottom). Injection was simulated with
injection wells set up in 175 cells pumping 136 m3 /day and 275 m3 /day, in two simulations
to identify the potential difference in migration characteristics. A vertical fault (3.05x3.05 m)
extending from layer 51 to layer 37 was set above the center of the horizontal well. Hydraulic
conductivity of the preferential flow path was set at 280 m3 /day.

Results

A model simulation showing the initial flow system and TDS distribution (Figure 2, top)
depicts the upward flow in the Floridan aquifer, horizontal flow in the Boulder Zone, and
downward flow between the Boulder Zone and the Sunniland Formation. The hydraulic head
distribution after stage 7 of the injection period shows that the significant increase in head
(approximately 400 m) is restricted to the near vicinity of the horizontal well (Figure 2,
middle). The distribution of TDS after 360 days (Figure 2, bottom) exhibits the maximum
vertical extent of the injected plume to occur in layer 44 (for under and over-pressured
conditions), a vertical distance of approximately 460 meters. The plume continues to migrate
horizontally; however, vertical migration has essentially ceased.

Conclusions

This study demonstrates that vertical migration of the injected fluid from hydraulic
fracturing does not extend into the Underground Source of Drinking Water (USDW) that
occurs in the Floridan aquifer. The vertical migration of the injected solution is characterized
and controlled by the following:

• low permeability formations composed of clay, anhydrite,limestone, and dolomite;

• very high salinity and increasing density with depth;

• downward hydraulic gradient associated with under-pressured conditions;

• general absence of faults in the Sunniland Formation;

• without faulting, vertical migration from the reservoir does not occur;

• with presence of a fault, vertical migration is limited to approximately 460 m.;

• presence of Boulder Zone and associated Kohout convection, restricts potential migra-
tion into the USDW;and

• horizontal plume migration is restricted to near vicinity of the hydraulic fracture zone.

305
Figure 1. (top) Hydrogeologic column of Southwest Florida; (bottom) Horizontal well and hydraulic fracture
zone configuration.

306
Figure 2. (top) Initial TDS conditions and flow system; (middle) hydraulic head distribution in layer 52
after stage 7; (bottom) TDS distribution after 360 days showing the vertical extent of injected fluids plume
at layer 44.

References

307
Corresponding author: Shawn Leppert
Leppert Associates
[email protected]

Conceptual Model Simulation for the Establishment of Remedial System

Performance Criteria

Shawn C. Leppert, Leppert Associates

A. Wunsch
R. Farnsworth

Abstract: Environmental regulatory agencies require the establishment of performance

criteria for remediation systems. A direct performance criteria for groundwater extraction
systems often measure a system’s capture zone. Capture zones can be defined as the dif-
ference in the potentiometric surface between stressed (extraction) and ambient hydrologic
conditions. The capture zone is often estimated from the interpolation of drawdown mea-
sured at monitoring wells. The challenge in using drawdown as a performance measure
is that the ambient heads for every combination of hydrologic control state is not mea-
surable, it must be estimated. If a hydrologic environment is highly transient, empirical
estimation techniques may not be adequate. A methodology has been developed to estab-
lish quantitative performance criteria for groundwater extraction and control remediation
systems constructed in highly transient hydrologic environments. The foundation of the
methodology is the construction of a hydrologic conceptual flow model that can be simu-
lated to accurately represent the principal hydrologic processes. The model is thoroughly
calibrated to a relatively wide range of potential hydrologic conditions. The calibrated flow
model is used to simulate the ambient potentiometric surface for an adequate sampling
(matrix) of potential principal hydrologic control states. Estimates of hydraulic head for
each simulated monitoring well, for each matrix scenario is used to develop a mathemat-
ical relationship between monitoring well hydraulic head and principal hydrologic control
parameters. The hydraulic head at monitoring wells and primary hydrologic control param-
eters are measured in the field. The difference between the measured hydraulic heads and
the estimated ambient hydraulic heads for the measure hydrologic control parameters are
used to calculated a drawdown for each monitored well.

308
Corresponding author: Kara Marsac
Colorado School of Mines
[email protected]

Evaluating Non-potable Water Usage for Oil and Gas Purposes in the Permian
Basin

Kara Marsac, Colorado School of Mines

A. Navarre-Sitchler, Colorado School of Mines

Abstract: Oil and gas company water usage is currently an area of concern in the water
stressed western United States. 87ύ of recent wells in the Permian Basin are located in
areas of high or extreme water stress. Using recycled produced water or groundwater that
does not meet the USDW drinking water standards for oil and gas purposes could assist
in relieving both water stress and tension between oil and gas companies and the public.
However, non-USDW drinking water (TDS over 10,000 ppm) has the potential to react with
formation water causing mineral precipitation, reducing the permeability of the producing
formation. To evaluate the potential of non-potable water usage in the Permian Basin,
available groundwater chemistry data was compiled into a database. Data was collected
from the NETL-run NATCARB database, the USGS Produced Water and NWIS Databases,
and the Texas Water Development Board. The created database went through a system of
quality assurance and control for pH, TDS, depth and charge balance. Data was used to
make a set of waters representative of Permian Basin groundwater based on TDS, Ca/Mg
ratio and Cl/SO4 ratio. Low, medium and high of these three characteristics; representing
the 25th, 50th and 75th percentile respectively; was used to make a matrix of 27 waters. Low
TDS is 64,660 ppm, medium TDS is 98,486 ppm, and high TDS is 157,317 ppm. Ca/Mg
ratios range from 1.98 to 7.26, and Cl/SO4 ratios range from 32.96 to 62.34. Geochemist’s
Workbench was used to create mixing and titration models these 27 waters reacting with an
average water, specifically examining for possible precipitation. Initial results are positive,
with the highest total precipitation being 1.815 cm3 in 1 L of water with high TDS, high
Ca/Mg ratio and low Cl/SO4 ratio. This indicates a maximum of approximately 0.18ύ
of porosity would be filled with mineral precipitation during the mixing of chosen Permian
Basin waters.

309
Corresponding author: Scott Painter
Oak Ridge National Laboratory
[email protected]

Using fine-scale simulations to develop subgrid representations of

microtopography effects in integrated hydrology models
Scott L. Painter, Oak Ridge National Laboratory
A. Jan, Oak Ridge National Laboratory
E.T. Coon, Los Alamos National Laboratory

Abstract: Although it is well established that microtopography can attenuate and de-
lay runoff and alter the balance between infiltration and runoff, representing the effects
of microtopography in watershed-scale integrated hydrology models remains a challenge.
Submeter spatial resolution of microtopography is feasible in plot-scale models, but such
high-resolution simulations become intractable at watershed scales, which motivates the use
of subgrid representations. Using integrated hydrology of polygonal tundra implemented in
the Advanced Terrestrial Simulator as an example, we evaluate the use of fine-scale solu-
tions to the diffusion wave equation at submeter resolution as a strategy for constructing
subgrid representations. An uncalibrated subgrid model including depression storage and
flow obstruction effects on overland flow improves the match to fine-scale results relative
to a coarse-grid representation without the subgrid model. Additional improvement is ob-
tained by calibrating one parameter related to depression storage, resulting in a reasonably
accurate match to the fine-scale simulations. Catchment-scale simulations of the thermal
hydrology of polygonal tundra near Barrow, Alaska using the calibrated subgrid model show
that depression storage increases infiltration and reduces runoff in permafrost-affected re-
gions, thus confirming the importance of representing microtopography effects. This work
was supported by the Interoperable Design of Extreme-scale Application Software (IDEAS)
project and the Next-Generation Ecosystem Experiments-Arctic (NGEE Arctic) project.
NGEE-Arctic is supported by the Office of Biological and Environmental Research in the
DOE Office of Science. This manuscript has been authored by UT-Battelle, LLC under
Contract No. DE-AC05-00OR22725 with the U.S. Department of Energy. The United
States Government retains and the publisher, by accepting the article for publication, ac-
knowledges that the United States Government retains a non-exclusive, paid-up, irrevocable,
world-wide license to publish or reproduce the published form of this manuscript, or allow
others to do so, for United States Government purposes. The Department of Energy will
provide public access to these results of federally sponsored research in accordance with the
DOE Public Access Plan (http://energy.gov/downloads/doe-public-access-plan).

310
Corresponding author: Zachary Stanko
U.S. Geological Survey
[email protected]

Metamodel analysis of California’s coastal groundwater basins to identify

undesirable results of management actions

Zachary P. Stanko, U.S. Geological Survey

M. Fienen
C. Faunt

Abstract: California’s Sustainable Groundwater Management Act of 2014 provides a

framework for comprehensive, sustainable groundwater management, and empowers local
agencies to assess hydrologic conditions and address ”undesirable results” of management
actions. This study introduces a strategy to identify undesirable results in California’s
coastal groundwater basins. Groundwater-flow models can be used to predict the extent
of undesirable results under various conditions; however, existing models cover only a few
of the 131 coastal basins and may require substantial computation. Surrogates trained on
these models, called metamodels, can emulate the input-output relationships with trivial
runtime and, in many cases, be applied to unmodeled basins. Metamodels are developed
for existing models of representative basins to identify simulated outcomes of interest. The
viability of the metamodels is explored for other modeled and unmodeled basins. Appropri-
ateness of a metamodel is then determined for a given basin by considering model outcomes
and basin characteristics. The strategy is being tested on three California coastal ground-
water basins with existing MODFLOW-based models. The primary goal is to determine
whether there are undesirable results in the groundwater basins. Metamodeling techniques
are initially based on Gradient-Boosted Regression Trees to emulate model outputs that
represent undesirable results, such as excessive drawdowns. The metamodel performance is
evaluated using a range of unmodeled basins. Metrics for metamodel efficiency, including
cost of metamodel development and forecast accuracy, will indicate if alternative model-
ing strategies should be considered. The result of this analysis will contribute to the local
management goal of efficiently characterizing undesirable results.

311
Corresponding author: Jon Starn
U.S. Geological Survey
[email protected]

MODFLOW-based meta-modeling for estimating regional residence-time

distributions

Jon J. Starn, U.S. Geological Survey

K. Belitz
C. Carlson
L. Desimone
L. Kauffman

Abstract: Groundwater residence-time distributions (RTDs) are important for assessing

groundwater vulnerability; however, there is no approach currently available for mapping
RTDs at the multi-state or national scales. The meta-modeling approach used here gener-
ates particle RTDs with general MODFLOW/MODPATH models, summarizes the particle
RTDs by fitting parametric distributions to them, and then trains a statistical-learning
model to predict regional-scale parametric RTDs using mappable physical characteristics
such as recharge and aquifer volume. IPython notebooks are used to create general MOD-
FLOW/MODPATH models rapidly, consistently, and reproducibly. The notebooks use
FloPy and Python’s extensive GIS and scientific computing capabilities to create general
models based on nationally consistent data sets such as the NHDPlus. Typical work-flow
duration starting from a model-domain shapefile to a working model was about 5 minutes
for a model with 100,000s of cells, not including calibration. The models were created in
different hydrogeologic terranes in the northeastern glaciated U.S to represent aquifers with
differing geometries, extents, and thicknesses. The models were ”calibrated” by automat-
ically adjusting hydraulic conductivity in portions of the model domain to minimize the
fraction of cells where the water table was above land surface and below perennial streams.
Targets for statistical learning were created from particle RTDs by fitting unimodal as well
as mixtures of Weibull, gamma, and inverse-Gaussian distributions. Physical characteristics
of the modeled areas (including hydrogeologic terrane) were used to predict distribution pa-
rameters. A Multi-Task Lasso regression was trained on the general-model-based parametric
RTDs and used to extrapolate them throughout the northeastern U.S.

312
Corresponding author: Peter Vermeulen
iMod Software
[email protected]

Large scale high resolution modelling

Peter Vermeulen, iMod Software

B. Minnema
J. Verkaik
J. Hughes

Abstract: We have developed a high performance modeling approach that allows build-
ing, maintaining and running large scale high resolution models for very large areas (e.g.
the entire state of California) with reasonable simulation times. This can be an efficient
approach for large model areas that include a large number of groundwater pumping wells
with drawdown-interferences between individual wells. This drawdown-interference might
be essential for areas with low natural recharge. A large scale approach also minimizes the
risk that future sub-domain model applications are limited by nearby model boundaries.
For MODFLOW2005 models the sub-domain modelling technique includes the new Parallel
Krylov Solver package for tightly coupled sub-domains. The modeller is offered a flexible
environment to either make parallelized high resolution computations for the whole model
domain, or ad hoc ”related to political priorities” refine and compute only small parts of
the modelling domain, while maintaining consistency with the rest of the modelling area.
Consistency is realized through integrated up- and downscaling techniques. The techniques
are made accessible through the iMOD Graphical User Interface which has been an open
source product since 2014 (see: oss.deltares.nl/web/iMOD).

313