Artigo

Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

Journal of Hydrology 556 (2018) 539–556

Contents lists available at ScienceDirect

Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol

Research papers

Comparison of new generation low-complexity flood inundation


mapping tools with a hydrodynamic model
Shahab Afshari a,⇑, Ahmad A. Tavakoly b, Mohammad Adnan Rajib c, Xing Zheng d, Michael L. Follum b,
Ehsan Omranian e, Balázs M. Fekete f
a
Department of Civil Engineering, The City College of New York, CUNY Environmental CrossRoads Initiative, CUNY-CREST Institute, City University of New York,
160 Convent Ave, Marshak Rm #826, New York, NY 10031, USA
b
Coastal and Hydraulics Laboratory, U.S. Army Engineer Research and Development Center, 3909 Halls Ferry Road, Vicksburg, MS 39180, USA
c
Lyles School of Civil Engineering, Purdue University, 550 W Stadium Ave, West Lafayette, IN 47907, USA
d
Department of Civil, Architectural & Environmental Engineering, University of Texas at Austin, Austin, TX 78712, USA
e
Department of Civil and Environmental Engineering, University of Texas at San Antonio, San Antonio, TX 78249, USA
f
Department of Civil Engineering, The City College of New York, CUNY Environmental CrossRoads Initiative, CUNY-CREST Institute, City University of New York,
Steinman Hall Room #188, 160 Convent Avenue, New York, NY 10031, USA

a r t i c l e i n f o a b s t r a c t

Article history: The objective of this study is to compare two new generation low-complexity tools, AutoRoute and
Received 11 May 2017 Height Above the Nearest Drainage (HAND), with a two-dimensional hydrodynamic model (Hydrologic
Received in revised form 19 November 2017 Engineering Center-River Analysis System, HEC-RAS 2D). The assessment was conducted on two hydro-
Accepted 22 November 2017
logically different and geographically distant test-cases in the United States, including the 16,900 km2
Available online 23 November 2017
This manuscript was handled by G. Syme,
Cedar River (CR) watershed in Iowa and a 62 km2 domain along the Black Warrior River (BWR) in
Editor-in-Chief, with the assistance of Ashok Alabama. For BWR, twelve different configurations were set up for each of the models, including four dif-
Mishra, Associate Editor ferent terrain setups (e.g. with and without channel bathymetry and a levee), and three flooding condi-
tions representing moderate to extreme hazards at 10-, 100-, and 500-year return periods. For the CR
Keywords: watershed, models were compared with a simplistic terrain setup (without bathymetry and any form
Hyper-resolution modeling of hydraulic controls) and one flooding condition (100-year return period). Input streamflow forcing data
AutoRoute representing these hypothetical events were constructed by applying a new fusion approach on National
HAND Water Model outputs. Simulated inundation extent and depth from AutoRoute, HAND, and HEC-RAS 2D
HEC-RAS 2D were compared with one another and with the corresponding FEMA reference estimates. Irrespective of
Multi-model comparison the configurations, the low-complexity models were able to produce inundation extents similar to HEC-
National Water Model
RAS 2D, with AutoRoute showing slightly higher accuracy than the HAND model. Among four terrain set-
ups, the one including both levee and channel bathymetry showed lowest fitness score on the spatial
agreement of inundation extent, due to the weak physical representation of low-complexity models com-
pared to a hydrodynamic model. For inundation depth, the low-complexity models showed an overesti-
mating tendency, especially in the deeper segments of the channel. Based on such reasonably good
prediction skills, low-complexity flood models can be considered as a suitable alternative for fast predic-
tions in large-scale hyper-resolution operational frameworks, without completely overriding hydrody-
namic models’ efficacy.
Ó 2017 Elsevier B.V. All rights reserved.

1. Introduction perhaps more disastrous (Hirabayashi et al., 2013). In the past


30 years, estimated costs of average annual flood damage is
With an increasing stress of climate and land use changes in approximately $8 billion within the United States (US) (National
recent times, flood events are becoming more frequent and Weather Service – Hydrologic Information Center, 2016). Accord-
ingly, there is a growing interest in regional to continental scale
high/hyper resolution flood forecasting and risk assessment across
⇑ Corresponding author. various parts of the globe (e.g. Alfieri et al., 2013; Bierkens et al.,
E-mail addresses: safshar00@citymail.cuny.edu (S. Afshari), Ahmad.A.Tavakoly@ 2015; Paiva et al., 2011; Pappenberger et al., 2012; Winsemius
erdc.dren.mil (A.A. Tavakoly), adnanrajib@purdue.edu (M.A. Rajib), zhengxing@ et al., 2013; Wood et al., 2011). Maidment (2015) proposed a mod-
utexas.edu (X. Zheng), michael.l.follum@usace.army.mil (M.L. Follum), SeyedEhsan. eling architecture to forecast streamflow in 2.7 million river
Omranian@utsa.edu (E. Omranian), bfekete@ccny.cuny.edu (B.M. Fekete).

https://doi.org/10.1016/j.jhydrol.2017.11.036
0022-1694/Ó 2017 Elsevier B.V. All rights reserved.
540 S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556

reaches across the continental US, which became operational in (e.g. channel and flood-plain cross-sections, optimum parameter
2016 under the National Water Model (NWM) framework values), which are often not readily available. Accordingly, the
(http://water.noaa.gov/about/nwm). Despite these advancements, majority of these modeling packages come with a ‘‘black-box” con-
translating streamflow forecasts into time-varying flood inunda- figuration that can be executed only for research purposes in a
tion maps with reasonable accuracy and speed remains an out- stand-alone desktop environment (Kauffeldt et al., 2016; Néelz,
standing concern. 2009). These models also require considerable setup and computa-
Hydrologic models contain a rainfall-runoff estimator and a tion time, especially with high resolution river networks. Accord-
channel routing scheme, therefore, another model component is ingly, using a model that is as realistic as possible is not the
required to simulate the over-bank conditions (i.e. flood inunda- panacea (Hunter et al., 2007); the choice should be balanced
tion). Many model applications for inundation mapping exist in lit- against several other considerations when it comes to the question
erature (Table 1). Out of these alternatives, Hydrologic Engineering of integration into a continental scale operational system such as
Center-River Analysis System (HEC-RAS), with 1D flow simulation the NWM.
functionality, has been the principal model used in US Federal Choice of a hydraulic/hydrodynamic model as component of a
Emergency Management Agency (FEMA)’s National Flood Insur- large scale framework is determined less by the superior model
ance Program (FEMA, 2015) and National Oceanic and Atmospheric physics and more by its suitability to be executed in cyber infras-
Administration (NOAA)’s Advanced Hydrologic Prediction Service tructures, computational overhead, interoperability with the driver
(NOAA, 2011). The ability of performing coupled 1D/2D analysis hydrologic model, output retrieval, and visualization capabilities
has been recently added to HEC-RAS (hereafter, HEC-RAS 2D; (Rajib et al., 2016). Being driven by such constraints, Follum
Table 1) which is still being tested under different geophysical set- (2012) introduced AutoRoute (Table 1) as a rapid tool to create
tings. With a few exceptions of the LISFLOOD-FP model (e.g. Alfieri flood inundation mapping over large scales. Using the simulated
et al., 2014; Rajib et al., 2016; Schumann et al., 2013), most of the streamflow outputs from Tavakoly et al. (2017) as an input forcing
model applications listed in Table 1 are limited to small spatial to the AutoRoute, Follum et al. (2017) generated high resolution
scales over either a single river reach or a low-density river (10 m) flood maps for the Midwest US (230,000 km2) and the
network. Mississippi Delta (109,500 km2). Despite such intensive applica-
Executing most of the hydraulic/hydrodynamic models requires tion, computational overhead for executing AutoRoute was
modelers’ intervention to provide substantial spatial details remarkably small. Liu et al. (2016) adopted the concept of Height
Above the Nearest Drainage (HAND; Nobre et al., 2011; Rennó
et al., 2008) and transformed 10 m National Elevation Dataset
Table 1 (NED) for the continental US into a HAND raster. This HAND raster
Existing models being applied by researchers and flood modeling communities along shows the relative height of a given location above the nearest
with those applied in current study. reach in the nationally mapped river network (National Hydrogra-
Model Reference(s) Developer phy Dataset Plus). Maidment et al. (2016) featured several case
(s) studies based upon the loose coupling of NWM streamflow outputs
1 FESWMS-2DH (Finite Froehlich (1989) and US with this HAND raster to generate near real-time flood inundation
Element Surface Water Musser and Dyar (2007) Geological maps. Considering these recent advancements, it is timely to
Modeling System for 2D flow Survey examine whether fast-computing, ‘‘low-complexity” inundation
in the Horizontal plane) mapping tools with simplified input requirements and process-
2 FaSTMECH (Flow and Kim et al. (2011) and
representations can be preferred from an operational standpoint,
Sediment Transport with Nelson et al. (2003)
Morphological Evolution of particularly in time-limited emergency response scenarios, over
Channels) computationally exhaustive, input intensive, physics based and
3 MIKE 11 1D, MIKE 21 2D and Ballesteros et al. (2011), The Danish presumably accurate hydraulic/hydrodynamic models.
MIKE FLOOD 1D/2D coupled Patro et al. (2009), Hydraulic
Ability to capture natural floodplain processes and the influence
hydrodynamic suit of models Wright et al. (2008) Institute
4 SOBEK 1D/2D Vanderkimpen et al. Deltares- of man-made control structures is different in each model. No
(2009) Delft model has the perfect realization of flooding; hence, simplification
Hydraulics of the model physics may further undermine its already-limited
5 BreZo/HiResFlood Begnudelli and Sanders University ability. In this regard, a multi-model comparison can help measure
(2007), Nguyen et al. of
relative accuracy of each model. Previous studies are heavily
(2015a), Nguyen et al. California,
(2015b), Sanders (2007) Irvine, US skewed towards the comparison of 1D versus 2D hydraulic/hydro-
6 FLDWAV (Flood Wave Fread (1998) US National dynamic models (Cook and Merwade, 2009; Alho and Aaltonen,
Dynamic Model) Weather 2008; Benjankar et al., 2014; Horritt and Bates, 2002; Leandro
Service
et al., 2009; Tayefi et al., 2007; Vojinovic and Tutulic, 2009). Sev-
7 HEC-RAS (Hydrologic USACE (2014) US Army
Engineering Center-River Corps of
eral studies have compared different 2D models (Horritt and
Analysis System) 1D Engineers Bates, 2001; Vanderkimpen et al., 2009) or the same model under
8 HEC-RAS (Hydrologic Brunner (2014) different configurations of topographic resolution and/or surface
Engineering Center-River roughness (Bates et al., 2003; Cook and Merwade, 2009; Horritt
Analysis System) 2D*
and Bates, 2001; Mason et al., 2003; Pappenberger et al., 2005).
9 LISFLOOD-FP Alfieri et al. (2014), Bates University
and De Roo (2000), Bates of Bristol, Effects of other geophysical and man-made attributes including
et al. (2010), Rajib et al. UK channel bathymetry, levees, and bridges on model-simulated flood
(2016), Schumann et al. inundation has remained relatively unexplored (e.g. Cook and
(2013)
Merwade, 2009; Pappenberger et al., 2006).
10 AutoRoute* Follum (2012) and US Army
Follum et al. (2017) Corps of
The new-generation low-complexity inundation mapping tools,
Engineers such as AutoRoute and HAND, have not been compared with each
11 HAND (Height Above the Maidment et al. (2016) Liu et al. other, or with advanced hydrodynamic models (e.g. HEC-RAS 2D).
Nearest Drainage) for and Zheng et al. (2016) (2016) Although AutoRoute was compared with reference inundation
continental US*
extents (Follum et al., 2017), HAND’s efficacy is yet to be tested.
*
Models being applied and tested in current study. This study, developed upon the preliminary work of Afshari et al.
S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556 541

Fig. 1. Workflow of multi-model comparison scheme adopted in this study including inputs, models, and outputs.

(2016), attempts to fill this gap with a view to provide a ‘‘practical, iod) and four different terrain setups (e.g. with and without
yet reliable” flood inundation modeling alternative to be coupled channel bathymetry and levee). Concerning the much larger test-
with continental scale hydrologic forecasting models. The main case (i.e. CR), only a single flood event (100-year return period)
objective of this study is to evaluate the relative accuracy between and terrain setup (without channel bathymetry and levee) were
AutoRoute, HAND, and HEC-RAS 2D for different magnitudes of considered. The models were compared with one another for inun-
flooding events, in terms of both inundation extents and depths. dation extent and depth, separately in each of the terrain-flood
The Cedar River (CR) in Iowa and Black Warrior River (BWR) in Ala- configurations. In one of the configurations of BWR, FEMA esti-
bama in the US were considered as test-cases to represent two dif- mated flood extents and depths were also used as a reference to
ferent spatial scales, terrain, land use and hydro-climatic compare with model simulations. All these configurations are sum-
conditions. For one of the test-cases (BWR), models are compared marized in Table 2. To keep the terminology obvious and self-
after incorporating geophysical and man-made attributes (e.g. explanatory, configurations were named in terms of the attributes
channel bathymetry and levee) such that the resultant difference in their respective terrain setups (e.g. NED, NED + Bathymetry,
in the outcomes invoke avenues of future refinement in their cur- NED + Levee, NED + Bathymetry + Levee).
rent structures.
2.1. Study domains and hydraulic control structures
2. Methodology
Fig. 2 illustrates the test-cases – a 16,900 km2 Cedar River (CR)
The assessment presented in this study is based upon 39 model watershed in Iowa and a much smaller 62 km2 area along the Black
configurations involving three models, three flood events, four ter- Warrior River (BWR) in Alabama in the US. The CR watershed
rain setups, and two test beds. Fig. 1 summarizes the general stretches about 380 km in a mild/moderate terrain (0.0001 m/
design of this study, showing that each model (i.e. HEC-RAS 2D, m), from Mower County in Minnesota to a US Geological Survey
AutoRoute, and HAND) was executed over the BWR test-case sep- (USGS) outlet in Cedar Rapids, Iowa (USGS 05464500). Along its
arately for three flood events (10-, 100-, and 500-year return per- way downstream, the main channel is fed by 11 major tributaries
some of which are also divided into lower order upstream headwa-
ter reaches. The BWR domain stretches approximately 15 km
Table 2 through a moderate/slightly terrain (0.0003 m/m) in Tuscaloosa
Configurations for multi-model comparison. Each of the three models used in this
county, Alabama with 15 adjoining tributaries. The river networks
study had 13 configurations as listed below.
used for flood simulation in both cases were obtained from
Configuration Flood event Terrain setup Test-case NHDPlus (McKay et al., 2012). The dominant land use in CR is agri-
(return
NED Bathymetry Levee cultural, whereas the BWR domain used in this study is mostly an
period)
urban landscape with some forested areas.
1 10-year 10 m   Black Warrior
p CR is relatively a natural landscape with least obstruction by
2  River (BWR),
p man-made hydraulic control structures. In contrast, the BWR
3  Alabama
p p
4 domain has one levee and two lock/dams in the main channel
5 100-year 10 m   (associated with two USGS gauge stations; Fig. 2). The levee has
p
6  been in operation since August 1999 (construction date) to reduce
p
7 
p p flooding damage in the city of Northport. According to the US Army
8
9 500-year 10 m   Corps of Engineers National Levee Database, the approximate
p
10  length and average crest elevation of the levee are 3.3 km and
p
11  47.5 m, respectively. The Oliver lock and dam is located at the out-
p p
12
let of the study domain (USGS 02465000), whereas the Holt lock
13 100-year 10 m   Cedar River (CR), Iowa
and dam (USGS 02462961) defines the upstream boundary loca-
542 S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556

identical when the models were compared under the same config-
uration. The following sub-sections describe the model setups,
including the sources of these input data and details on how they
were processed for this study.

2.2.1. Construction of input streamflow data


The specific flood magnitudes used for model simulation for
this study are not actual events. The representative streamflow
data to force the inundation models were constructed while main-
taining the hydrologic ‘‘connectivity” of the river network and
therefore accounting for the contribution of tributaries. For
instance, difference between monthly mean downstream and
upstream streamflow in the BWR domain implies higher values
in the downstream station 87% of the time (Fig. 3a). Fig. 3b, com-
paring daily streamflow data at the upstream and downstream
gauge stations for 19 peak flow events during a past 40-year period
(1976–2014), also validates this notion. These findings help to real-
ize two critical factors regarding the ‘‘connectivity” aspect men-
tioned above. First, amplification of flood magnitude while
proceeding towards the outlet was due to the lateral flows from
the tributaries; attenuation of flood peak by some diffusion effects
can be considered negligible (discussed further in a later section).
Secondly, the flood peaks in the tributaries occurred at the same
time as in the main channel. These data-driven assessments justi-
fied the approach adopted here to determine the input streamflow
forcing data for the inundation models.
In this study, input streamflow data for the three hypothetical
flood events (10-, 100-, and 500-year return periods) were con-
structed by fusing NWM simulated outputs with USGS and FEMA
estimates. Simulated hourly streamflow in all the associated
NHDPlus reaches were obtained from a pre-operational offline
repository of the NWM (personal communications with the NOAA
National Water Center, Alabama) for a recent flooding event in
respective study domains (September 25–October 1, 2016 in CR
(Buchmiller and Eash, 2010 also referenced as USGS, 2010 in
Fig. 5) and December 24–31, 2015 in BWR (Hedgecock and
Feaster, 2007 also referenced as USGS, 2007 in Fig. 5)). Like other
hydrologic models, NWM outputs have some discrepancies relative
to the observed data (Fig. 4). Since the use of NWM in this study
was kept limited only to ensure hydrologic connectivity of river
network, bias in its streamflow simulations did not affect the ‘‘rel-
ative accuracy” of the inundation models. Hence, further diagnostic
evaluation of NWM’s performance was not included here.
Based on the statistical analyses of long-term observations at
the outlet of BWR, peak flow values for different return periods
were estimated by Hedgecock and Feaster, (2007). Accordingly,
Fig. 2. Test beds: (a) Cedar Rapid (CR) watershed including 15 USGS streamflow an event-specific scaling factor was calculated for the outlet of
monitoring stations (USGS 05464500 signifies the outlet); (b) segment of the Black
Warrior River (BWR) between Oliver and Holt lock and dam (respectively located at
BWR, being defined as the ratio of corresponding USGS estimated
USGS 02465000 and 02462951 stations). Northport levee located at the right bank magnitude and the NWM peak flow (USGS 02462951; Fig. 4). For
of the BWR is protecting the city of Northport from extreme flood events. The river all other NHDPlus reaches including the main channel’s upstream
networks in both test-cases are obtained from nationally mapped NHDPlus boundary location in BWR, respective NWM streamflow hydro-
database. Line thickness of the river network indicates ranges of National Water
graphs were multiplied by the outlet’s event-specific scaling factor
Model (NWM) streamflow outputs during selected peak flow events.
to obtain three different sets of input forcing data. Similar
tion for the main channel. Flood inundation is influenced by these approach was followed for CR except it was kept limited only for
dams/levees due to their flow regulatory role and to possible a 100-year flood event. CR being a much larger domain with den-
backwater effects during extreme events. Incorporation of these sified gauge network, calculation of scaling factor for this case
man-made control structures, even in their simplest forms, can involved using USGS estimated peak flow at 14 upstream gauge
supplement the limited hydrodynamic simulation capacity of an stations (addressed in Buchmiller and Eash, 2010) in addition to
inundation mapping tool. that at the outlet. For those reaches in CR with no USGS estimate,
the nearest downstream station was selected for scaling purposes.
FEMA (2008, 2010, 2013, 2014) also estimates peak flows for dif-
2.2. Model inputs ferent return periods leveraging some field-studies and local
expertise; however, such studies were available for five reaches
To enable an even assessment, the forcing data (i.e. input in BWR and 16 reaches in CR (dashed lines in Fig. 5). The FEMA sug-
streamflow), topographic resolution, land cover classification, and gested values were used in scaling factor calculation instead of
the values of channel/surface roughness parameters were kept USGS estimates, wherever seemed appropriate (i.e. specific reaches
S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556 543

Fig. 3. Flow regime variations and differences between upstream (USGS 02462951) and downstream (USGS 02465000) stations at the BWR test-cases: (a) difference of 450
monthly means of observed streamflow data during 1976–2014 period; (b) 19 daily peak streamflow at the upstream and downstream stations, both observed in the same
day during 1976–2014. In 18 out of 19 events, peak streamflow at downstream stations were significantly larger than upstream records.

2.2.2. Inclusion of floodplain features


As summarized in Table 2, the terrain setups used for flood
modeling in BWR include: (1) NED without channel bathymetry
or levee (NED), (2) NED with channel bathymetry (NED + Bathyme-
try), (3) NED with levee (NED + Levee), and (4) NED with both
channel bathymetry and levee (NED + Bathymetry + Levee). How-
ever, flood modeling for CR is conducted only with a single terrain
setup based on NED (without channel bathymetry or levee), con-
sidering its large spatial extent (16,900 km2 compared to BWR’s
62 km2), lack of continuous bathymetry data and least hydraulic
controls. The spatial resolution of NED was kept the same (10 m)
in every case. Hence, plausible differences in simulated inundation
depth and extent while using these different terrain setups evolve
solely from the respective ability of the models to capture flood-
plain hydrodynamics. Although the fourth terrain setup (i.e. NED
+ Bathymetry + Levee) is the best case to closely represent river
corridor/floodplain, others help to create insights on the sensitivity
of a model to particular floodplain feature(s).
The 10 m resolution of NED is not fine enough to identify abrupt
topographical variability (e.g. height, width, and location of a levee,
shape and thalweg of a channel), let alone the persistent inaccu-
racy that might have induced from the acquisition of elevation data
and associated interpolation techniques. Use of Light Detection and
Ranging (LiDAR) topography data significantly improves detection
of land surface features such as levee, but its inability to ‘‘see
through” water surface and capture channel bathymetry is not
unknown (Cook and Merwade, 2009). In case of BWR, limited
information on the longitudinal/cross-sectional dimensions and
changes in elevation along the levee was obtained from a design-
inventory of the US Army Corps of Engineers (USACE, 2014), part
of which were tentatively validated by the authors in a non-
exhaustive field survey using ground global positioning system
(GPS) equipment. Furthermore, the main channel’s water surface
and bed elevation data were obtained from another field-
Fig. 4. Comparison of simulated (NWM, blue curves) and observed daily stream-
flows (USGS stations, red curves) during selected peak flow events at the outlet of campaign (obtained from the USACE Tuscaloosa Field Office), at a
the two test-cases: (a) CR, and (b) BWR. (For interpretation of the references to spatial resolution of 2–10 m along and across the channel. These
colour in this figure legend, the reader is referred to the web version of this article.) are point elevation data, which were transformed into raster for-
mat to mosaic with the 10 m NED, producing the aforementioned
terrain setups. Fig. 6 shows the difference between the NED and
and flood return periods, depending on availability). Effect of these NED + Bathymetry + Levee setups along a cross-section and a lon-
scaling factors on the constructed input streamflow are presented gitudinal section of the main channel. The effect of the dam/locks
in Fig. 5 in terms of relative channel thickness. was implicitly incorporated in the streamflow data (Section 2.2.1).
544 S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556

values were selected based upon the suggestions from Moore


(2011); however, the ‘‘low roughness” category was used consider-
ing the recommendation by Follum et al., (2017). All the 30 m
NLCD grid-cells classified as ‘open water’ basically represent the
river network, hence, they were assigned the roughness value of
a natural channel (n = 0.03). Unlike the HEC-RAS 2D and AutoRoute
models, HAND does not require assigning n for the land surface.
The current version of HAND only uses channel n, which was kept
consistent with the other two models to enable an even
comparison.

2.3. Flood inundation modeling

2.3.1. Hydrodynamic model: HEC-RAS 2D


Hydrologic input to HEC-RAS 2D refers to the streamflow
hydrographs (time-series) for each of the NHDPlus reaches
involved and a streamflow-stage rating curve at the outlet location.
Geospatial inputs include: (1) terrain (topography data) and (2)
spatially distributed (gridded) surface/channel Manning’s n. The
model solves a 2D unsteady flow equation at hourly resolution
using a diffusive wave approach. Although HEC-RAS 2D can also
employ a ‘‘full momentum” approach (the Saint Venant equation),
it was avoided in this study as it does not produce substantial dif-
ferences in simulated inundation in a fairly uniform terrain like
BWR or CR. Despite the relatively intensive computational
demand, application of the Saint Venant equation in HEC-RAS 2D
would be more useful in simulating critical scenarios such as levee
breach and design of hydraulic structures. The simulation was per-
formed on a heterogeneous mesh, simultaneously having struc-
tured and un-structured cells (e.g. the mesh highlighted in
Fig. 7). Abrupt changes in the terrain (e.g. river bank, levee) are
delineated by ‘‘breaklines” and un-structured cells (up to nine
faces and with different sizes), while square cells (25 m) are nested
on the other parts of the landscape. Weighted average of elevation
and roughness values respectively from all intersecting/encom-
passing cells of the 10 m terrain and 30 m roughness grid (NLCD,
being linked with the lookup table) were ‘‘poured” on to a model
cell regardless of its size/shape. In addition to capturing topo-
graphic details as precisely (un-structured cells) and parsimo-
niously (square cells) as possible, faces of these model cells work
as ‘‘virtual cross-sections” that regulate the propagation of flood
wave. Although such a detailed model setup would enable
enhanced simulation of floodplain/channel’s response in flooding
conditions, it requires substantial intervention from the modelers.
Outputs from HEC-RAS 2D are the time-varying flood inundation
extents and depths out of which only those at the time-stamp of
peak flow were extracted for comparison purposes. Running time
is highly dependent on the amount of physical details imparted
into the model (e.g. mesh resolution, number of boundary
Fig. 5. The schematic representation of how National Water Model (NWM) daily
streamflow outputs were synthesized to create (a) a 100-year flood event in CR, and conditions).
(b) 10-, 100-, and 500-year flood events in BWR. Thickness of the river network
indicates the relative contribution of streamflow from respective tributaries in 2.3.2. Low-complexity model: AutoRoute
developing a particular peak flow at the outlet. Detail methodology are provided in
Unlike HEC-RAS 2D, AutoRoute uses peak flow (not the time-
Section 2.2.1.
series/hydrograph) at each NHDPlus reach. Other inputs to Auto-
Route, including topography data and distributed surface/channel
2.2.3. Land cover and surface/channel roughness roughness values, were the same as those used in HEC-RAS 2D.
A common feature in HEC-RAS 2D and AutoRoute (also HAND, Assuming 1D steady-state flow, AutoRoute uses Manning’s equa-
with some exceptions) is the provision of spatially distributed val- tion to calculate the normal flow depth over a high-density num-
ues of surface and/or channel roughness parameters. In this study, ber of cross-sections, while the cross-section geometry and
the same set of roughness values was used in each of the inunda- channel slope were automatically generated from topography data.
tion models. The 2011 National Land Cover Database (Homer et al., Flood depth and inundation maps were simulated using a volume-
2015) at a 30 m spatial resolution is selected as the input land use fill numerical method at each cross-section. Follum et al. (2017)
data for both test-cases (e.g. highlighted in Fig. 7a as an example provided more details on the setup rubrics and computational
from BWR). Depending on the respective land use class, a separate techniques of AutoRoute. Output from AutoRoute was a static set
lookup table linked each grid cell of NLCD 2011 with a representa- of inundation extent and depth, corresponding to the peak flow
tive value of Manning’s roughness coefficient (n). Manning’s n used to force the model (for each of the specific flood events;
S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556 545

Fig. 6. Bathymetry integration in the BWR test-case: (a) plan view of topographical and geospatial variation, (b) profile view of a cross-section at the channel highlighting four
terrain setups, (c) longitudinal profile view of the main channel from upstream to downstream direction. For clarity, this figure features a smaller portion of the BWR test-case
which includes a levee.

Fig. 7. Application of spatially distributed channel roughness parameters and terrain setups by HEC-RAS 2D, AutoRoute, and HAND models: (a) a magnified view of 30 by 30
m NLCD 2011 layer including 16 distinct land cover classifications, (b) a magnified view of 10 by 10 m terrain setups. In both (a) and (b), structured and non-structured mesh
grids represent HEC-RAS 2D’s computational configuration. While this figure features only BWR, the same protocol was applied in the case of CR test-case as well.
546 S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556

Table 2). Although streamflow and other input data for AutoRoute
were manually processed in this study, supplementary tools have
been developed for automatic pre-processing to enable its execu-
tion in an operational setting (Snow, 2016; AutoRoutePy: 2.1.0).
It should be noted that AutoRoute is currently closed-source at
the request of one of our Military sponsors but may soon be open
source. Due to this uncertainty, the paper does not state whether it
is open source. The AutoRoute executable is publically available by
sending an access request to Michael Follum (Michael.L.Follum@
erdc.dren.mil).

2.3.3. Low-complexity model: HAND


HAND is a hydrological terrain analysis approach, which has
been tested for reasonable functionality in producing flood inunda-
tion maps (Rodda, 2005; Rennó et al., 2008; Nobre et al., 2011). In
Fig. 8. The error matrix applied for quantifying the conformity of two flood
this approach, vertical distance between a grid-cell in topography
inundation maps. Each squared block of the error matrix indicates number of
data and the nearest cell along a stream that it drains into defines incidents at which model 1 and model 2 are either resulting in similar (i.e. both
the ‘‘HAND value”. All cells on the landscape that have a HAND models simulate wet or dry condition, denoted as nw1,w2 and nd1,d2) or dissimilar
value smaller than the specified stage (water level) are treated as (i.e. model 1 simulates wet while the model 2 simulates dry or vice versa, denoted
inundated. HAND is entirely raster-based and defines the inun- as nw1,d2 and nd1,w2) predictions. Margins of the error matrix, indicate the summation
of cells in a particular row or column (e.g. nw1,w2+d2 imply sum of nw1,w2 and nw1,d2 as
dated zone by a corresponding river segment. Therefore, it does the first row elements of the error matrix).
not require the creation of cross sections. A user-friendly, seamless
workflow for the HAND model is currently under development,
however, an executable prototype framework for US watersheds
can be supported by Xing Zheng (zhengxing@utexas.edu). where n is total number of cells; nw1;w2 is number of cells predicted
In this study, 10 m HAND rasters were created with different wet by both inundation models; nd1;d2 is number of cells predicted
terrain setups (four rasters for BWR and one for CR; Table 1), each dry by both inundation models; nd1;w2 is number of cells predicted
with respect to the NHDPlus river network (e.g. Liu et al., 2016). dry by model 1 but as wet by model 2; nw1;d2 is number of cells pre-
These HAND rasters were then used to estimate stage height – dicted wet by model 1 but as dry by model 2; nw1;w2þd2 is number of
channel hydraulic geometry relationships for each of the reaches cells where model 1 predicted them as wet while model 2 predicted
(e.g. Zheng et al., 2016). Taking these relationships, estimated either wet or dry; nw1þd1;d2 is number of cells where model 2 pre-
channel length and average slope from the NHDPlus database, dicted them as dry while model 1 predicted either wet or
and predefined channel roughness value (Section 2.2.3), Manning’s dry;nd1;w2þd2 and nw1þd1;w2 are being read in same fashion to
equation was applied to generate streamflow-stage rating curves nw1;w2þd2 and nw1þd1;d2 respectively. Both of these inundation metrics
for all the reaches. Using these rating curves, input peak flow cor- range from 0 to 1 denoting lowest and highest conformity,
responding to a given return period was converted into a stage respectively.
height. Finally, the HAND raster was used to create the inundation Flood inundation depths were compared by calculating Mean
extents at these particular stage heights. Difference (MD) and Root Mean Squared Difference (RMSD)
between the simulated outputs from two models or between a
2.4. Model comparison metrics model and a reference. Both MD and RMSD were computed based
upon an average of cell-by-cell difference and squared difference of
Quantifying the differences in inundation extents and depths two flood inundation depth layers:
between two flood models, and between flood models and the ref-
erence (e.g. FEMA in BWR case study), needs a mathematical 1X N
scheme. For the comparison of inundation extents, an error matrix MD ¼ ðZ1;i  Z2;i Þ ð3Þ
N i¼1
(e.g. Congalton and Green, 2008) was developed (Fig. 8) using
which Kappa-statistic and Fitness-statistic (being denoted as j
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
and F , respectively) (Yu and Lane, 2006) were calculated to mea- u N
1u X
sure the degree of agreement or disagreement between two flood RMSD ¼ t ðZ1;i  Z2;i Þ2 ð4Þ
maps. The j statistic is a ratio between the actual agreement (indi- N i¼1
cated by major diagonal of the error matrix) of the two models and
the chance of agreement (expressed through marginal rows and where N is total number of raster cells; Z1,i and Z2,i are depth
columns of the error matrix) (Eq. (1)). Hence, an impressive j value values simulated respectively by flood model 1 and 2 at the ith
is possible even with fewer matching wet cells. For instance, in the cell. MD and RMSD need careful interpretation, if used together.
case of flood events, where there are few number of conforming Lower MD may not always come with lower RMSD. For a con-
wet cells (i.e. nw1,w2) relative to the large number of conforming stant MD, RMSD can increase as the variance associated with
dry cells (i.e. nd1,d2), j might be close to 1. the frequency distribution of error magnitudes also increases.
Accordingly, the sole purpose of MD was kept limited in this
n:ðnw1;w2 þ nd1;d2 Þ  ðnw1þd1;w2 nw1;w2þd2 þ nw1þd1;d2 nd1;w2þd2 Þ
j¼ study only to evaluate a model’s general overestimating/underes
n2  ðnw1þd1;w2 nw1;w2þd2 þ nw1þd1;d2 nd1;w2þd2 Þ
timating tendency with respect to the other model or the FEMA
ð1Þ reference, while RMSD should be seen as a metric of models’ rel-
F reduces bias into results since it only considers the number of ative accuracy.
conforming wet cells predicted by both flood models (Eq. (2)): An R code (RStudio, 2017) was developed to perform one-to-
one comparison of model products (i.e. flood inundation extent
nw1;w2
F¼ ð2Þ and depth) with the option of calculating a suite of conformity
nw1;w2þd2 þ nw1þd1;w2  nw1;w2 statistics as described above (Afshari, 2016, 2017).
S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556 547

3. Results and discussion case, with a view to have a closer look on some critical aspects
where flood models usually struggle over large spatial scales.
This section presents the outcome of the study from three Specifically, the model comparisons on CR solicited a general
aspects: (1) comparison among models for flood extents, (2) com- assessment whether low-complexity models ‘‘behave” in the same
parison among models for flood depths, and (3) comparison of the manner as the hydrodynamic model regardless of meandering
models with FEMA flood estimates for a specific flood magnitude main channel segments, confluence, and lower order headwater
(only for BWR test-case). reaches.
In general, HEC-RAS-2D resulted in notably larger inundated
3.1. Comparison of flood extents area compared to AutoRoute and HAND. Across the entire test-
case, differences in inundated area between HEC-RAS 2D and Auto-
3.1.1. Inter-comparison of models for inundation extent Route and between HEC-RAS 2D and HAND were respectively 382
For the CR test-case, HEC-RAS 2D, AutoRoute, and HAND were and 229 km2, which made HAND simulations closer to HEC-RAS
compared for flood extent and depth only for a 100-year flood. 2D. Although HAND inundated a slightly larger area than Auto-
Only the most simplistic terrain setup, without channel bathyme- Route, spatial patterns of their respective inundation were nearly
try or other possible floodplain features, was considered in this identical in each of the four cases (A1–A4) highlighted in Fig. 9.

Fig. 9. Comparison of HEC-RAS 2D, AutoRoute and HAND for 100-year flood inundation extent in CR test-case. Only a simplistic terrain setup is considered here (10 m NED
without bathymetry or levee). A1–A4 blocks in all columns are highlighting magnified portions of selected flooded area at different regions of the test bed from upstream
through downstream. Blue colored areas denote inundation simulated by both models. Yellow and red colored areas denote inundation solely by a particular model. In all
figures, j and F denote Kappa-statistic and measure of fit respectively, to measure the degree of agreement or disagreement between two flood maps. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)
548 S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556

This was also evident from j and F as both of the low-complexity disagreement between the two models; F values were found sub-
models basically showed the same fitness scores against the inun- stantially higher when looked specifically into A3 or A4 portions of
dation extent of HEC-RAS 2D. the test-case. This could also be supported by visually assessing the
As indicated in Section 2.4, concurrent occurrence of high j and highlighted portions in CR (Fig. 9). For example, with respect to
low F is quite possible. j shows conformity in the number and HEC-RAS 2D, low-complexity models showed large differences
location of dry cells between the models, not the conformity of along the main channel and its confluences with the tributaries
their actual inundated extents. Still, a j value as high as one in (A1 and A2 in Fig. 9); however, such difference was found minimal
CR strongly suggests that the low-complexity models function very in the lower order less-meandering reaches in the upstream head-
reasonably. Concerning the actual match of wet cells (i.e. inunda- water catchments (A3 and A4 in Fig. 9).
tion), obtaining a lower F value could be potentially misinter- A small-scale yet more comprehensive comparison was
preted. It is likely for an uncalibrated low-complexity model to deduced from the BWR test-case focusing more on the models’
show lower F values against a much more detailed hydrodynamic response to floodplain features such as bathymetry and levee.
model, especially when executed over a large area such as the CR. HEC-RAS 2D, AutoRoute, and HAND were compared for flood
Nevertheless, a relatively low fitness of inundated boundaries extent using 10-, 100-, and 500-year events, and for four different
(F values 0.5 in Fig. 9) cannot be undermined as the typical terrain setups (Figs. 10–12). The results suggest expansion of

Fig. 10. Comparison of HEC-RAS 2D, AutoRoute and HAND for 10-year flood inundation extent over the four different terrain setups in BWR test-case. From Figs. 10–12: rows
compare between different set of models for a particular terrain setup, while columns compare effect of different terrain setups for a particular set of models. Blue colored
areas denote inundation simulated by both models. Yellow and red colored areas denote inundation solely by a particular model. In all figures, j and F denote Kappa-statistic
and measure of fit respectively, to measure the degree of agreement or disagreement between two flood maps. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556 549

Fig. 11. Comparison of HEC-RAS 2D, AutoRoute and HAND for 100-year flood inundation extent over the four different terrain setups in BWR test-case. For interpretation of
this figure, Fig. 10 provides required details.

inundation extents when the return period increased from 10-year representation of the wave propagation (e.g. backwater effect) is
to 500-year for all the models; however, the percent change of a limitation in HAND. Moreover, HAND showed flooding in physi-
extent between return periods was not identical when models cally implausible locations which have similar elevation difference
were compared with one another. For four terrain setups in HEC- with respect to the nearest reach (the main channel in this case).
RAS 2D, the average expansion of the flood extent was 25% and As a result, HAND produced larger inundated area than AutoRoute
11% from 10-year to 100-year and from 100-year to 500-year flood when both were compared against HEC-RAS 2D. This is similar to
events, respectively. Although AutoRoute is not a hydrodynamic what was observed in the case of CR test-case. For BWR, HAND
model, it closely mimics HEC-RAS 2D in most parts of the study showed an average expansion of flood extent by 31% between
region. AutoRoute showed an average expansion of flood extent 10-year and 100-year events and 5% between 100-year and 500-
by 20% between 10-year and 100-year and 10% between 100- year flood events. Among all the configurations, regardless of ter-
year and 500-year event simulations. Compared to HEC-RAS 2D, rain setup or flood event, AutoRoute invariably showed better per-
a limitation in AutoRoute is the absence of any downstream formance than HAND in capturing the inundation extent of BWR in
boundary condition (dam/lock in BWR main channel). Accordingly, terms of j and F .
difference in inundation extents between HEC-RAS 2D and Auto-
Route, as observed in the first column of Figs. 10–12, might be 3.1.2. Effect of terrain setups on inundation extent
due to AutoRoute’s inability to capture the possible ‘‘backwater Depending on the model, effects of floodplain attributes (e.g.
effect” during the higher magnitude events. Similar to AutoRoute, levee and bathymetry) on the simulated inundation extents can
550 S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556

Fig. 12. Comparison of HEC-RAS 2D, AutoRoute and HAND for 500-year flood inundation extent over the four different terrain setups in BWR test-case. For interpretation of
this figure, Fig. 10 provides required details.

be noticeably different. In the BWR test-case, HEC-RAS 2D and ‘‘within-bank” water because of the deeper and wider channel in
AutoRoute showed reasonably good agreement irrespective of the NED + Bathymetry and NED + Bathymetry + Levee setups. This
return periods, even without the incorporation of channel bathy- phenomenon was vivid for 10-year and 100-year events (Figs. 10
metry or levee (e.g. F ¼ 0:71  0:75). Incorporation of the levee and 11, respectively). Accordingly, the best model conformity in
(NED + Levee setup) could not mitigate HAND’s overestimating the NED + Levee setup (i.e. highest j and F ) does not necessarily
tendency. As the relative elevation with respect to the nearest mean that model simulated inundation extents were more accu-
channel is a key determinant for HAND, the model essentially rate in this particular terrain setup. Considering NED + Bathyme-
over-laid the same ‘‘over-bank” stage from the particular channel try + Levee to be a relatively better realization of the floodplain
segment on both sides of the levee. In reality, spatial orientation compared to other terrain setups, raises questions to why Auto-
of the levee acts as a confinement and thus controls the movement Route and HAND are unable to leverage from this terrain setup
of flood wave, which was relatively well-captured by HEC-RAS 2D and remains as an outstanding question needing further investiga-
and AutoRoute. tion looking into possible scopes of model re-conceptualization.
Both AutoRoute and HAND showed prominent overestimation Regardless, the NED + Bathymetry + Levee setup exemplified a cru-
of inundation extents relative to HEC-RAS 2D throughout the entire cial aspect regarding the effect of bathymetry in case of an extreme
length of the main channel once bathymetry was incorporated in flood event (e.g. 500-year; Fig. 12). In that case, flood water was
the terrain. More specifically, HEC-RAS 2D tended to retain more found to have over-topped the levee in the NED + Levee setup in
S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556 551

all the models, but HEC-RAS 2D and AutoRoute did not allow over- 3.2.1. Inter-comparison of models for inundation depth
topping when bathymetry was included in the model configuration Fig. 13 demonstrates the comparison among HEC-RAS 2D, Auto-
(NED + Bathymetry + Levee). This is a clear example of bathymetry Route, and HAND for inundation depth in CR, only for a 100-year
being a controlling factor that determines the shape of flood inun- flood event. Over the entire stream network of CR, HEC-RAS 2D
dation map, except in HAND. Here as well, AutoRoute has relatively produced 11–16 m (MD) deeper floods than AutoRoute and HAND,
better capability than HAND, at least with current versions. which is in line with its overestimation of inundation extent
(Fig. 9). Similarly, both low-complexity models had minimal differ-
ence (<2 m) in their respective simulated depths. However, HAND
3.2. Comparison of flood depth produced a slightly deeper flood than AutoRoute as evident from
MD, making the models’ behavior coherent with what was
Inter-comparison of simulated inundation depths (as raster depicted in Fig. 9 for their flood extent simulation. Furthermore,
data layers) was carried out only for the 100-year flood event in HAND simulated depth had an RMSD of 9 m against HEC-RAS
both CR and BWR test-cases. The common (intersecting) area of 2D, contrary to AutoRoute’s 17 m. This led to the notion that HAND
the inundation extents generated by two particular models was simulations were relatively more consistent with HEC-RAS 2D, at
taken as the boundary within which flood depth comparison was least in CR. The meandering main channel (a near-outlet location)
carried out. and confluences (A1 and A2) again appeared to be the hot-spots

Fig. 13. Comparison of flood depths simulated by HEC-RAS 2D, AutoRoute and HAND in CR test-case, for a 100-year flood event with a particular terrain setup. MD and RMSD
denote Mean Difference and Root Mean Squared Difference respectively to evaluate the similarity of two flood models in predicting flood depth. Significance of A1–A4 is
explained in Fig. 9.
552 S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556

where both AutoRoute and HAND struggled, but their performance in most parts of the river network. On the other hand, variability of
to simulate flood depth is equivalent to HEC-RAS 2D in the case of depth simulated by HAND was relatively more erratic (hence, may
less-meandering upstream headwater catchments (A3 and A4). not be realistic).
For the BWR test-case, inundation depth of a 100-year flood In the 16,900 km2, 380 km long CR test-case, the low-
event for all terrain setups showed an average MD of 0.8 m complexity models generally showed an underestimating ten-
between HEC-RAS 2D and AutoRoute, with AutoRoute resulting dency to simulate inundation extent and depth, while HAND pro-
in deeper inundation (Fig. 14). HAND produced even deeper inun- duced closer results to the HEC-RAS 2D model. The opposite was
dation as its average MD from HEC-RAS 2D was about 4 m. In gen- found in the 62 km2, 15 km long BWR test-case with AutoRoute
eral, both AutoRoute and HAND overestimated depth with respect being relatively accurate with respect to HEC-RAS 2D. Here, the
to HEC-RAS 2D which is opposite to what was detected from the CR scale is important. For example, there are specific locations with
test-case. Analogous to the results shown for inundation extent short reach segments within CR, where the low-complexity models
(Section 3.1.1), bias in depth simulation was nearly 50% less in behaved similarly as in the case of BWR. Hence, the outcome from
AutoRoute than that in HAND in terms of their RMSD against the BWR test-case, showing AutoRoute and HAND overestimating
HEC-RAS 2D. Variability of depth along the main channel was inundation extent and depth is not an anomaly, rather it is a very
found to be more consistent between HEC-RAS 2D and AutoRoute likely subset of possibilities that might have happened if a much

Fig. 14. Comparison of flood depths simulated by HEC-RAS 2D, AutoRoute and HAND in BWR test-case, for a 100-year flood event with four different terrain setups. MD and
RMSD denote Mean Difference and Root Mean Squared Difference respectively to evaluate the similarity of two flood models in predicting flood depth. Rows compare between
different set of models for a particular terrain setup, while columns compare effect of different terrain setups for a particular set of models.
S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556 553

larger domain along the Black Warrior River system was modeled interesting here is the remarkable similarity between AutoRoute
in this study. and HAND pertaining to their behavior in depth simulation in cer-
It could seem ambiguous why AutoRoute differs from HAND tain segments of the main channel. Contrary to their general depth
even though both are based on Manning’s equation. Despite using overestimating tendency relative to HEC-RAS 2D, both AutoRoute
Manning’s equation, the procedure of generating flood maps is not and HAND distinctly underestimated flood inundation depth in
similar in these models. AutoRoute generates flood inundation the middle and near-outlet locations in the ‘‘bathymetry-informe
map (extent) and depth ‘‘vertically”, whereas HAND does it ‘‘later- d” terrain setups. Even when comparing HAND and AutoRoute,
ally”. More specifically, AutoRoute automatically generates cross- NED + Bathymetry + Levee showed the highest RMSD (7.64 m)
sections, calculates flood extent for user-defined input streamflow out of all terrain setups being modeled here. Overall, this analysis
values using iterative calculation of flow for every cross section resonates the insensitivity of these low-complexity flood models to
(Follum et al., 2017). The depth is incrementally increased from cope with hydrodynamics, especially in meandering portions of the
the lowest point in the reach until the calculated streamflow from channel (middle section), the deeper portions in the near-outlet
Manning’s equation matches the input streamflow. In the HAND location with a navigational dam/lock, and nearby levee, not to
method, first a synthetic stage-discharge rating curve is generated mention their inability to capture back water effects. However, in
for each NHDPlus reach using Manning’s equation. For a user- such cases, AutoRoute seemed to be better-equipped than HAND,
defined input streamflow, a corresponding depth value is extracted similar to the findings highlighted in Section 3.1.2 for simulation
from such a rating curve (Liu et al., 2016). All cells on the landscape of inundation extent.
with a HAND value smaller than this depth will be considered as
inundated. Hence, this method does not require cross sections
and it also does not have the direct incorporation of streamflow 3.3. Evaluation of model performance with FEMA reference estimates
as in AutoRoute. In this way, use of Manning’s equation in HAND
does not produce similar results as in the AutoRoute model. Fig. 15 compares model simulated inundation extent and depth
for a 100-year flood event with the corresponding FEMA estimates.
This assessment was kept limited only on BWR test-case as it has
3.2.2. Effect of terrain setups on inundation depth nearly all real-life examples of floodplain features that often make
Similar to the multi-model multi-terrain assessment of inunda- flood models under-perform. The FEMA flood map (i.e. inundation
tion extent (Section 3.1.2), simulated flood depth in the BWR test- extent) was obtained via Flood Map Service Center (https://msc.
case was also evaluated under four different terrain setups. When fema.gov/portal), which is the official public source for sharing
the models were configured with the simplest terrain setups, NED flood hazard information in terms of flood maps and other related
and NED + Levee, AutoRoute showed the least bias for flood depth products. The technical information from FEMA-Coordinated
simulation with respect to HEC-RAS 2D (relevant RMSD values for Needs Management Strategy (FEMA-CNMS) platform as well as
a 100-year event; Fig. 14). Once bathymetry was incorporated, FEMA’s R4 Regional Service Center assured that channel bathyme-
AutoRoute showed signs of discrepancy as the RMSD value relative try was not considered for developing the 100-year flood hazard
to HEC-RAS 2D increased by more than 1 m. Exactly similar phe- map in BWR. Another considerable factor was that FEMA estimates
nomenon was observed in the case of HAND when it was compared on flood extent and depth were available only for the downstream
with HEC-RAS 2D (middle column of Fig. 14). What is more portion of BWR encompassing the levee. Hence, comparing the

Fig. 15. Comparison of HEC-RAS 2D, AutoRoute and HAND with FEMA estimated reference for a 100-year flood event. The comparison was conducted only for downstream
region of the study domain nearby the Northport levee, due to the limited availability of FEMA studies. Top Row: Flood inundation maps, Bottom Row: flood depth.
554 S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556

models with FEMA estimates seemed justified only for the ‘‘NED + model. The HAND model results were relatively closer to HEC-RAS
Levee” terrain setup. Clearly, HEC-RAS 2D had the most conformity 2D compared to the AutoRoute model for depth and flood extent in
with FEMA estimates for flood extent (Fig. 15). The j and F scores the CR study domain. However, in the relatively less meandering
were 0.94 and 0.74, respectively, for the HEC-RAS 2D versus FEMA upstream headwater rivers, both AutoRoute and HAND behaved
case. These scores were 0.93 and 0.69 when the flood extent gen- the same as HEC-RAS 2D. This can be attributed to the nature of
erated by AutoRoute was compared with the FEMA flood map. low complexity models, their inherent capabilities, and limitation,
Comparison of the HAND and FEMA flood extents resulted in j which might perform well in simple landscape. Results also
and F equal to 0.88 and 0.53, respectively. In case of flood depths, showed that in complex conditions such as meandering main
HEC-RAS 2D quite expectedly produced the lowest RMSD (2.23 m), channels and confluences, low complexity models struggle to gen-
while AutoRoute and HAND had an RMSD of 3.33 m and 5.36 m, erate results that are comparable to the HEC-RAS 2D model. In the
respectively. The average of differences (MD) for FEMA flood depth BWR test-case, AutoRoute generally outperformed HAND when
against HEC-RAS 2D, AutoRoute, and HAND were respectively cap- compared to the HEC-RAS 2D model when hydraulic controls such
tured as 1.55, 0.37, and 1.27 m. The MD values show that Auto- as dams and levee were incorporated. One should note that both
Route and FEMA are relatively similar whereas HAND and HEC-RAS low complexity models demonstrated identical spatial variation
2D underestimate the flood depth derived by FEMA. AutoRoute and of flood extents, despite the difference between magnitudes of
FEMA both use similar methods (Manning equation) which can flood extent or flood depth. Considering model performance with
explain similarity of MD values. Since HEC-RAS 2D applies the different terrain setups, HEC-RAS 2D and AutoRoute models
unsteady hydrodynamic (2D Diffusion Wave) equation, it might showed similar results for the NED + Levee terrain setup with
produce lower depth compared to the steady state approach. This HAND showing some inconsistency in capturing the effect of any
is due to the fact that the unsteady state equation factors in addi- abrupt geophysical variation (e.g. enforced by the levee). Contrar-
tional physically-based terms may lower the simulated water ily, terrain setups with bathymetry (e.g. the NED + Bathymetry +
stages compared to ones estimated by steady state approach Levee terrain setup) showed prominent discrepancy between the
(being applied in FEMA and AutoRoute). HAND model shows over- low-complexity models in comparison with the hydrodynamic
estimation of depth at upstream and abruptly underestimation of model. This behavior is speculated to be driven by assimilation of
depth when a levee exists (Fig. 15) which resulted in overall under- different river bed-slope at each NHDPlus reach (i.e. channels with
estimation of depth with this model compared to the FEMA model. bathymetry versus channels having a ‘‘flat-bed” as given by NED),
and thus, resultant change in the vertical distance between a given
cell on the floodplain and the nearest channel cell that it drains
4. Conclusion into. In general, the low-complexity models should be set up with
caution in flat and densely urbanized zones (e.g. in the down-
This paper inaugurates a new line of research to compare the stream regions of the BWR study domain) since they do not cap-
sensitivity and suitability of new-generation low-complexity flood ture the backwater effects created by existing hydraulic
models. With increasing flood hazards across the world, it has been structures. Further study in different climate and land use condi-
a burning question whether it is sustainable to employ computa- tions would be helpful to validate these findings.
tionally intensive yet supposedly better hydrodynamic models in Despite the generally favorable results obtained in this study, it
large-scale hyper-resolution operational flood simulation. Against is up to the users discretion whether a low-complexity flood inun-
such concern, a few models were developed with the fast- dation mapping tool should be preferred or complex hydrody-
computing capability due to simplified input requirements and namic models. As a near future application, low complexity
process-representations yet reasonably good in terms of prediction models can provide a rapid, first order estimate of flood inundation
accuracy. This paper compares two such genres of recently- to prioritize evacuation areas during severe flood events. Further-
developed inundation models: a hydrodynamic model (i.e. HEC- more, the combination of both types of modeling approaches can
RAS 2D) and two low-complexity models (i.e. AutoRoute and be considered. For instance, the low complexity models can be
HAND). To the best of authors’ knowledge, this is also the first flood used for flood mapping at the regional scale with a nested high-
inundation assessment using United States National Water Model fidelity model at the local scale where anthropogenic effects and
(NWM) streamflow data. topographic conditions need to be considered. Results of this study
The assessment presented in this study is based upon 39 model showed that low-complexity tools can have ‘‘nearly equal applica-
configurations involving three models, three flood events, four ter- bility” while retaining the value of complex hydrodynamic models.
rain setups, and two test beds. Models were compared for two This notion is based upon a trade-off between highest possible
hydrologically different and geographically distant test-cases in accuracy and computational efficiency, which is permissible for
the United States, including the 16,900 km2 Cedar River (CR) operational needs. Moreover, to provide operational hydrologic
watershed in state of Iowa and a 62 km2 domain along the Black support in geographic areas where hydrologic data is sparse, and
Warrior River (BWR) in state of Alabama. Model comparison in because the assessments provided are time critical, alternative
CR was conducted using only one terrain setup (i.e. National Eleva- approaches can be employed to develop terrain and bathymetry
tion Dataset, NED) and a 100-year flood event, with a view to focus data. For instance, idealized power-law hydraulic geometries
on issues such as meandering channel segments and confluences which are derived based on bank-full hydraulics of the channel
where flood models usually struggle. A much more detailed analy- can be established given measured basic hydraulics (i.e. discharge,
sis was conducted over the BWR test-case including moderate to water surface width, average depth, and average velocity) and
extreme flood events at 10-, 100-, and 500-year return periods applied to generate asymptotic forms of the channel bed geometry.
and various terrain setups with levee and/or bathymetry. The Application and deployment of this methodology in the low com-
streamflow time-series from the pre-operational offline repository plexity models can be considered as a far future work.
of the NWM were used and processed to define inflow boundary
conditions of the flood events for all NHDPlus reaches in the test- Acknowledgments
cases.
Results showed HEC-RAS 2D, AutoRoute, and HAND can be This work was conducted in a continuing effort to complete the
ranked according to their model complexity and computational multi-model comparison study which was initiated at the 2016
capability. HEC-RAS 2D is the most and HAND is the least complex National Water Center Innovators Program supported by the
S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556 555

Consortium of Universities for the Advancement of Hydrologic Follum, M.L., Tavakoly, A.A., Niemann, J.D., Snow, A.D., 2017. AutoRAPID: a model
for prompt streamflow estimation and flood inundation mapping over regional
Science, Inc. (CUAHSI) and the National Oceanic and Atmospheric
to continental extents. J. Am. Water. Resour. Assoc. 53 (2), 280–299.
Administration. The authors would like to appreciate the support Fread, D.L., 1998. NWS FLDWAV Model: Theoretical Description Hydrologic
of those involved in the coordination and execution of this project Research Laboratory, Office of Hydrology. National Weather Service, NOAA.
including Dr. David Maidment, Dr. Sagy Cohen, Dr. Sarah Prask- Froehlich, David C., 1989. Local scour at bridge abutments. Proceedings of the 1989
National Conference on Hydraulic Engineering. Publ by ASCE, New Orleans, LA,
ievicz, Dr. Edward P. Clark, Mr. Alan Snow, Ms. Elissa Yeates, and USA.
Ms. Kayla (Hudson) Cotterman. This project was partially funded Hedgecock, T.S., Feaster, T.D., 2007. Magnitude and frequency of floods in Alabama,
by Deputy Assistant Secretary of the Army for Research and Tech- 2003. U.S. Geological Survey Scientific Investigations Report 2007–5204, 28p., +
app. (available online at http://pubs.water.usgs.gov/sir2007-5204).
nology through the Engineer Research and Development Center’s Hirabayashi, Y., Mahendran, R., Koirala, S., Konoshima, L., Yamazaki, D., Watanabe,
Military Engineering applied research work package title Austere S., Kim, H., Kanae, S., 2013. Global flood risk under climate change. Nat. Clim.
Entry and was supported exclusively by internal funding from Change 3, 816–821 https://doi.org/doi:10.1038/nclimate1911.
Homer, C.G., Dewitz, J.A., Yang, L., Jin, S., Danielson, P., Xian, G., Coulston, J., Herold,
The City College of New York including the Andrew S. Grove N.D., Wickham, J.D., Megown, K., 2015. Completion of the 2011 National Land
Endowment. cover database for the conterminous United states-representing a decade of
land cover change information. Photogramm. Eng. Rem. Sens. 81 (5), 345–354.
Horritt, M.S., Bates, P.D., 2001. Effects of spatial resolution on a raster based model
References of flood flow. J. Hydrol. 253 (1), 239–249.
Horritt, M.S., Bates, P., 2002. Evaluation of 1D and 2D numerical models for
Afshari, S., Omranian, E., Feng, D., 2016. Relative sensitivity of flood inundation predicting river flood inundation. J. Hydrol. 268, 87–99.
extent by different physical and semi-empirical models. In: Maidment (Ed.), Hunter, N.M., Bates, P.D., Horritt, M.S., Wilsond, M.D., 2007. Simple spatially-
Technical Report 13, National Water Center Innovators Program Summer distributed models for predicting flood inundation: a review. Geomorphology
Institute Report 2016. Consortium of Universities for the Advancement of 90 (3–4), 208–225. https://doi.org/10.1016/j.geomorph.2006.10.021.
Hydrologic Science Inc. (CUAHSI), USA, pp. 19–24. Kauffeldt, A., Wetterhall, F., Pappenberger, F., Salamon, P., Thielen, J., 2016.
Afshari, S., 2016. R, Repository: . License: GNU General Public License. http://doi. Technical review of large-scale hydrological models for implementation in
org/10.5281/zenodo.221189. operational flood forecasting schemes on continental level. Environ. Model.
Afshari, S., 2017. R, Repository: https://github.com/sha17hab/Working-With- Softw. 75, 68–76. https://doi.org/10.1016/j.envsoft.2015.09.009.
NWM-NetCDF-Products/tree/V.1.1. License: GNU General Public License. Kim, M.H., Morlock, S.E., Arihood, L.D., Kiesler, J.L., 2011. Observed and forecast
http://doi.org/10.5281/zenodo.893563. flood-inundation mapping application—a pilot study of an eleven-mile reach of
Alfieri, L., Burek, P., Dutra, E., Krzeminski, B., Muraro, D., Thielen, J., Pappenberger, F., the White River, Indianapolis, Indiana. U.S. Geological Survey Scientific
2013. GloFAS – global ensemble streamflow forecasting and flood early Investigations Report 2011–5138, 63p. 3 Appendixes.
warning. Earth Syst. Sci. 17, 1161–1175 https://doi.org/doi:10.5194/hess-17- Leandro, J., Chen, A.S., Djordjević, S., Savić, D.A., 2009. Comparison of 1D/1D and 1D/
1161-2013. 2D coupled (sewer/surface) hydraulic models for urban flood simulation. J.
Alfieri, L., Salamon, P., Bianchi, A., Neal, J., Bates, P., Feyen, L., 2014. Advances in pan- Hydraul. Eng. 135 (6), 495–504.
European flood hazard mapping. Hydrol. Process. 28 (13), 4067–4077. Liu, Y.Y., Maidment, D.R., Tarboton, D.G., Zheng, X., Yildirim, A., Sazib, N.S., Wang, S.,
Alho, P., Aaltonen, J., 2008. Comparing a 1D hydraulic model with a 2D hydraulic 2016. A CyberGIS Approach to Generating High-resolution Height Above
model for the simulation of extreme glacial outburst floods. Hydrol. Process. 22 Nearest Drainage (HAND) Raster for National Flood Mapping. CyberGIS 16,
(10), 1537–1547. The Third International Conference on CyberGIS and Geospatial Data Science,
Ballesteros Cánovas, J.A., Eguíbar, M., Bodoque, J.M., Díez-Herrero, A., Stoffel, M., Urbana, Illinois, July 24–26. https://doi.org.10.13140/RG.2.2.24234.41925/1.
Gutiérrez-Pérez, I., 2011. Estimating flash flood discharge in an ungauged CyberGIS Center Technical Report, CYBERGIS-TR-2016-005.
mountain catchment with 2D hydraulic models and dendrogeomorphic Maidment, D.R., Rajib, M.A., Lin, P., Clark, E.P., (Eds.). 2016. National Water Center
paleostage indicators. Hydrol. Process. 25 (6), 970–979. Innovators Program Summer Institute Report 2016. Technical Report 13,
Bates, P.D., De Roo, A.P.J., 2000. A simple raster-based model for flood inundation Consortium of Universities for the Advancement of Hydrologic Science Inc.
simulation. J. Hydrol. 236 (1–2), 54–77. (CUAHSI), USA, 122p. https://doi.org/10.4211/technical.20161019.
Bates, P.D., Marks, K., Horritt, M.S., 2003. Optimal use of high-resolution Maidment, D.R., 2015. A Conceptual Framework for the National Flood
topographic data in flood inundation models. Hydrol. Process. 17 (3), 537–557. Interoperability Experiment. cuahsi 22p. Available online at: https://www.
Bates, P.D., Horritt, M.S., Fewtrell, Timothy J., 2010. A simple inertial formulation of cuahsi.org/Files/Pages/documents/13623/nfieconceptualframework_revised_
the shallow water equations for efficient two-dimensional flood inundation feb_9.pdf (last cited on February 14, 2016).
modelling. J. Hydrol. 387 (1–2), 33–45. https://doi.org/10.1016/j. Mason, D.C., Cobby, D.M., Horritt, M.S., Bates, P.D., 2003. Floodplain friction
jhydrol.2010.03.027. parameterization in two-dimensional river flood models using vegetation
Begnudelli, L., Sanders, B.F., 2007. Simulation of the St. Francis Dam-Break Flood. J. heights derived from airborne scanning laser altimetry. Hydrol. Process. 17
Eng. Mech. (ASCE) 133 (11). (9), 1711–1732.
Benjankar, R., Tonina, D., McKean, J., 2014. One-dimensional and two-dimensional McKay, L., Bondelid, T., Dewald, T., Johnston, J., Moore, R., Rea, A., 2012. NHDPlus
hydrodynamic modeling derived flow properties: impacts on aquatic habitat Version 2: User Guide. Accessed Dec 2016. https://nctc.fws.gov/courses/
quality predictions. Earth. Surf. Process. Land. https://doi.org/10.1002/esp.3637. references/tutorials/geospatial/CSP7306/Readings/NHDPlusV2_User_Guide.pdf.
Bierkens, M.F.P., Bell, V.A., Burek, P., Chaney, N., Condon, L.E., David, C.H., de Roo, A., Moore, M.R. 2011. Development of a High-Resolution 1D/2D Coupled Flood
et al., 2015. Hyper-resolution global hydrological modelling: what is next? Simulation of Charles City, Iowa. Master of Science Thesis, University of Iowa.
Hydrol. Process. 29, 310–320 https://doi.org/doi:10.1002/hyp.10391. http://ir.uiowa.edu/etd/1032.
Brunner, G.W., 2014. Combine 1D and 2D Modeling with HEC-RAS. USACE-HEC. Musser, Jonathan W., and Thomas R. Dyar. 2007. Two-dimensional flood-inundation
Buchmiller, R.C., Eash, D.A., 2010. In: Floods of May and June 2008 in Iowa: U.S. model of the Flint River at Albany, Georgia: Atlanta, Georgia. http://pubs.
Geological Survey Open-File Report 2010–1096, p. 10. usgs.gov/sir/2007/5107/, U.S. Geological Survey Scientific Investigations Report
Congalton, R.G., Green, K., 2008. Assessing the accuracy of remotely sensed data: 2007-510.
principles and practices. CRC Press – Technology & Engineering. Néelz, S., 2009. Desktop review of 2D hydraulic modelling. Environment Agency,
Cook, A., Merwade, V., 2009. Effect of topographic data, geometric configuration and Bristol.
modeling approach on flood inundation mapping. J. Hydrol. 377, 131–142 Nelson, J.M., Bennett, J.P., Wiele, S.M., 2003. Flow and sediment transport modeling.
https://doi.org/doi:10.1016/j.jhydrol.2009.08.015. In: Tools in Geomorphology. John Wiley & Sons, Chichester, pp. 539–576.
FEMA, 2008. Flood Insurance Study, Benton County, Iowa and Incorporated Areas. Nguyen, P., Thorstensen, A., Sorooshian, S., Hsu, K., Aghakouchak, A., 2015a. Flood
Federal Emergency Management Agency, Flood Insurance. forecasting and inundation mapping using HiResFlood-UCI and near-real-time
FEMA, 2010. Flood Insurance Study, Linn County, Iowa and Incorporated Areas. satellite precipitation data: the 2008 Iowa Flood. Am. Meteorol. Soc. 1171–
Federal Emergency Management Agency. Flood Insurance. 1183. https://doi.org/10.1175/JHM-D-14-0212.1.
FEMA, 2013. Flood Insurance Study, Mower County, Minnesota and Incorporated Nguyen, P., Thorstensena, A., Sorooshian, S., Hsu, K., AghaKouchak, A., Sanders, B.,
Areas. Federal Emergency Management Agency, Flood Insurance. Koren, V., Cuib, Z., Smith, M., 2015b. A high resolution coupled hydrologic–
FEMA, 2014a. Flood Insurance Study, Floyd County, Iowa and Incorporated Areas. hydraulic model (HiResFlood-UCI) for flash flood modeling. J. Hydrol. 541 (Part
Federal Emergency Management Agency, Flood Insurance. A), 401–420.
FEMA, 2014. Flood Insurance Study, Volume 1 of 2, Tuscaloosa County, Alabama and NOAA, 2011. AHPS Guidline Final v3. Debbery, under contract EA133C05CQ1048 –
Incorporated Areas. Task Order 35.
FEMA, 2015. Policy for Use of Hydrologic Engineering Center-River Analysis System Nobre, A., Cuartas, L., Hodnett, M., Rennó, C., Rodrigues, G., Silveira, A., Waterloo, M.,
in the National Flood Insurance Program Available online at https://www. Saleska, S., 2011. Height Above the Nearest Drainage – a hydrologically relevant
fema.gov/policy-use-hydrologic-engineering-center-river-analysis-system- new terrain model. J. Hydrol. 404 (1–2), 13–29.
national-flood-insurance-program. Last accessed on May 8, 2017. Paiva, R.C.D., Collischonn, W., Tucci, C.E.M., 2011. Large scale hydrologic and
Follum, M.L., 2012. AutoRoute Rapid Flood Inundation Model. Coastal and hydrodynamic modeling using limited data and a GIS based approach. J. Hydrol.
Hydraulics Engineering Technical Note ERDC/CHL CHETN-IV-88. U.S. Army 406, 170–181. https://doi.org/10.1016/j.jhydrol.2011.06.007.
Engineer Research and Development Center, Coastal and Hydraulics Laboratory, Pappenberger, F., Beven, K.J., Hunter, N.M., Bates, P.D., Gouweleeuw, B.T., Thielen, J.,
Vicksburg, Mississippi. de Roo, A.P.J., 2005. Cascading model uncertainty from medium range weather
556 S. Afshari et al. / Journal of Hydrology 556 (2018) 539–556

forecasts (10 days) through a rainfall-runoff model to flood inundation J. Am. Water. Resour. Assoc. 53 (2), 258–279. https://doi.org/10.1111/1752-
predictions within the European Flood Forecasting System (EFFS). Hydrol. 1688.12456.
Earth Syst. Sci. 9 (4), 381–393. Tayefi, V., Lane, S., Hardy, R., Yu, D., 2007. A comparison of one- and two-
Pappenberger, F., Matgen, P., Beven, K.J., Henry, J., Pfister, L., de Paul, F., 2006. dimensional approaches to modelling flood inundation over complex upland
Influence of uncertain boundary conditions and model structure on flood floodplains. Hydrol. Process. 21, 3190–3202.
inundation predictions. Adv. Water Resour. 29 (10), 1430–1449. U.S. Army Corps of Engineers, 2014. Design Deficiency Report With Environmental
Pappenberger, F., Dutra, E., Wetterhall, F., Cloke, H.L., 2012. Deriving global flood Assessment, Northport Levee Repair Project. Northport, Alabama.
hazard maps of fluvial floods through a physical model cascade. Hydrol. Earth Vanderkimpen, P., Peeters, E., Melger, P., 2009. Chapter 9. Flood modeling for risk
Syst. Sci. 16, 4143–4156. https://doi.org/10.5194/hess-16-4143-2012. evaluation—a MIKE FLOOD vs SOBEK 1D2D benchmark study. Flood Risk
Patro, S., Chatterjee, C., Mohanty, S., 2009. J. Indian Soc. Remote Sens. 37 (107). Management: Research and Practice (CRC Press 2008), 77–84. https://doi.org/
https://doi.org/10.1007/s12524-009-0002-1. 10.1201/9780203883020.ch9.
Rajib, M.A., Merwade, V., Liu, Z., 2016. Large scale high resolution flood inundation Vojinovic, Z., Tutulic, D., 2009. On the use of 1D and coupled 1D–2D modelling
mapping in near real-time. In: 40th Anniversary of the Association of State approaches for assessment of flood damage in urban areas. Urban Water J. 6 (3),
Flood Plain Managers National Conference, Grand Rapids, Michigan, USA, 19–24 183–199.
June. Available online at: http://www.asfpmfoundation.org/ace-images/Adnan_ Winsemius, H.C., Van Beek, L.P.H., Jongman, B., Ward, P.J., Bouwman, A., 2013. A
Full2016.pdf (last cited on December 21, 2016). framework for global river flood risk assessments. Hydrol. Earth Syst. Sci. 17,
Rennó, C.D., Nobre, A.D., Cuartas, L.A., Soares, J.V., Hodnett, M.G., Tomasella, J., 1871–1892. https://doi.org/10.5194/hess-17-1871-2013.
Waterloo, M.J., 2008. HAND, a new terrain descriptor using SRTM-DEM: Wood, E.F., Roundy, J.K., Troy, T.J., Beek, L.P.H.V., Bierkens, M.F.P., Blyth, E., Roo, A.D.,
mapping terra-firme rainforest environments in Amazonia. Remote Sens. et al., 2011. Hyperresolution global land surface modeling: Meeting a grand
Environ. 112, 3469–3481. challenge for monitoring Earth’s terrestrial water. Water Resour. Res. 47 (12).
Rodda, H.J.E., 2005. The Development and Application of a Flood Risk Model for the https://doi.org/10.1029/2010WR010090.
Czech Republic. Nat. Hazards 36 (1–2), 207–220. https://doi.org/10.1007/ Wright, N.G., M.ASCE, Villanueva, I., Bates, P.D., Mason, D.C., Wilson, M.D., Pender,
s11069-004-4549. G., Neelz, S., 2008. Downloaded 59 times TECHNICAL PAPERS Case Study of the
RStudio: Integrated Development for R. RStudio, Inc., n.d. Boston, MA, 2017. http:// Use of Remotely Sensed Data for Modeling Flood Inundation on the River
www.rstudio.com/. Severn, U.K. J. Hydraul. Eng. (ASCE) 134 (5).
Sanders, B.F., 2007. Evaluation of on-line DEMs for flood inundation modeling. Adv. Yu, D., Lane, S.N., 2006. Urban fluvial flood modelling using a two-dimensional
Water Resour. 30 (8), 1831–1843. https://doi.org/10.1016/j.advwatres.2007. diffusion-wave treatment, part 2: development of a sub-grid-scale treatment.
02.005. Hydrol. Process. 20, 1567–1583.
Schumann, G.J.-P., Neal, J.C., Voisin, N., Andreadis, K.M., Pappenberger, F., Zheng, X., Lin, P., Keane, S., Kesler, C., Rajib, M.A., 2016. NHDPlus-HAND evaluation.
Phanthuwongpakdee, N., Hall, A.C., Bates, P.D., 2013. A first large-scale flood In: Maidment (Ed.), National Water Center Innovators Program Summer
inundation forecasting model. Water Resour. Res. 49, 6248–6257. https://doi. Institute Report 2016. Technical Report 13. Consortium of Universities for the
org/10.1002/wrcr.20521. Advancement of Hydrologic Science Inc. (CUAHSI), USA, pp. 26–36. https://doi.
Snow, A.D., 2016. AutoRoutePy: 2.1.0. Zenodo. http://doi.org/10.5281/zenodo.54974. org/10.4211/technical.20161019.
Tavakoly, A.A., Follum, M.L., Wahl, M., Snow, A., 2017. Continental-scale river flow
modeling of the mississippi river basin using high-resolution NHDPlus dataset.

You might also like