Monflow Manual

Electronic Manual for
MODFLOW
A Modular 3D Finite-Difference Ground-Water Flow Model
Original MODFLOW code and manual by Michael G. McDonald and Arlen W. Harbaugh, USGS
MODFLOW
win32
code developed by Doug Rumbaugh, ESI
MODFLOW
win32
manual written by Doug and Jim Rumbaugh
Electronic MODFLOW manual reformatted by Jim Rumbaugh
This manual was produced using Microsoft Word Version 6.0 and Doc-To-Help
, by WexTech Systems, Inc.

/7
Environmental
Simulations
Inc.
Environmental Simulations, Inc.
2997 Emerald Chase Drive
Suite 100
Herndon, Virginia 22071
Tel: (703) 834-3054
Fax: (703) 689-2827
Email: [email protected]
[email protected]
WWW: http://www.us.net/envisim/welcome.html
ftp://ftp.us.net/pub/envisim
Electronic Manual for MODFLOW Contents i
Contents
1 Introduction 1
About this Help File...................................................................................................................1
Abstract ......................................................................................................................................1
Purpose ......................................................................................................................................2
Organization of This Report ......................................................................................................3
Acknowledgement......................................................................................................................5
2 Model Derivation 7
Mathematical Model ..................................................................................................................7
Discretization Convention..........................................................................................................8
Finite-Difference Equation.......................................................................................................10
Iteration....................................................................................................................................20
Formulation of Equations for Solution.....................................................................................23
Types of Model Cell and Simulation of Boundaries ................................................................24
Conceptual Aspects of Vertical Discretization.........................................................................25
3 Program Design 31
Overall Structure......................................................................................................................31
Array Boundaries and Aquifer Boundaries ..............................................................................40
Volumetric Budget ...................................................................................................................42
Space Allocation......................................................................................................................44
Three-Dimensional Subscripts for Model Arrays ....................................................................45
Input Structure .........................................................................................................................45
Output Structure.......................................................................................................................50
4 Basic Package 51
Conceptualization and Implementation....................................................................................51
Selection of Major Options and Designation of Input Files.......................................51
The IBOUND Array ..................................................................................................51
Initial Conditions .......................................................................................................52
Discretization of Time ...............................................................................................54
Output ........................................................................................................................54
Budget Calculations in the Basic Package.................................................................57
Basic Package Input .................................................................................................................58
Explanation of Fields.................................................................................................59
Output Control Input ................................................................................................................62
Explanation of Fields.................................................................................................62
5 Block-Centered Flow Package 65
Basic Conductance Equations....................................................................................66
ii Contents Electronic Manual for MODFLOW
Horizontal Conductance Under Confined Conditions................................................70
Horizontal Conductance Under Water Table Conditions ..........................................72
Vertical Conductance Formulation ............................................................................73
Vertical Flow Calculation Under Dewatered Conditions...........................................80
Storage Formulation...................................................................................................84
Storage Term Conversion ..........................................................................................85
Applicability and Limitations of Optional Formulations ...........................................87
Data Requirements.....................................................................................................87
Block-Centered Flow Package Input........................................................................................90
Explanation of Fields Used in Input Instructions.......................................................91
6 River Package 95
River Package Input ...............................................................................................................105
Explanation of Fields Used in Input Instructions.....................................................105
7 Recharge Package 107
Conceptualization and Implementation..................................................................................107
Recharge Package Input .........................................................................................................111
8 Well Package 113
Well Package Input ................................................................................................................115
9 Drain Package 117
Drain Package Input ...............................................................................................................122
10 Evapotranspiration Package 125
Evapotranspiration Package Input..........................................................................................130
11 General-Head Boundary Package 133
General-Head Boundary Package Input .................................................................................136
12 SIP Solver Package 139
General Theory ........................................................................................................139
Transfer of Arrays....................................................................................................155
Order of Calculation ................................................................................................156
Iteration Parameters .................................................................................................157
Strongly Implicit Procedure Package Input............................................................................161
Electronic Manual for MODFLOW Contents iii
13 SOR Solver Package 163
Slice-Successive Overrelaxation Package Input.....................................................................169
14 Utility Modules 171
Introduction............................................................................................................................171
Input Instructions For Array Reading utility Modules ...........................................................173
References 175
References..............................................................................................................................175
Appendix A - Program Portability 177
Introduction............................................................................................................................177
The Impact of the Programming Language on Portability .....................................................177
The Impact of Computational Precision on Portability..........................................................178
The Impact of Allocating Array Storage in a Single Array on Portability .............................180
The Impact on Portability of Preconnected File Units ...........................................................181
Conversion to FORTRAN 66.................................................................................................181
Appendix B - Space Requirements 183
X Array Memory Requirements.............................................................................................183
Appendix C - Continuing a Run 185
Procedure ...............................................................................................................................185
Appendix D - Sample Problem 187
Introduction............................................................................................................................187
Appendix E - Input Instructions 191
Introduction............................................................................................................................191
Array Input.............................................................................................................................191
Basic Package Input ...............................................................................................................193
Output Control Input ..............................................................................................................195
Block-Centered Flow Package Input......................................................................................197
River Package Input ...............................................................................................................199
Recharge Package Input.........................................................................................................199
Well Package Input ................................................................................................................200
Drain Package Input...............................................................................................................200
Evapotranspiration Package Input..........................................................................................201
General-Head Boundary Package Input .................................................................................202
Strongly Implicit Procedure Package Input............................................................................203
Slice-Successive Overrelaxation Package Input.....................................................................203
Glossary of Terms 201
Index 203
iv Contents Electronic Manual for MODFLOW
Electronic Manual for MODFLOW 1 Introduction 1
1 Introduction
About this Help File
The MODFLOW Electronic Manual is based upon the original documentation
written by Michael McDonald and Arlen Harbaugh of the USGS. The original
document has been changed in the following ways:
The sections of each chapter describing the source code have been removed,
The figures have been scanned at a high resolution and are included in the
help file, and
Manuals for the newer MODFLOW Packages (BCF2, BCF3, Stream-
Routing, Horizontal Flow Barrier, Aquifer Compaction, PCG2 Solver) are
provided in a separate help file.
Many of the figures are difficult to see due to the difference in resolution between the
scanned image and your screen. We recommend that if you want to get a better look
at a figure simply print the topic. Select File and Print Topic from the the Help
main menu (when viewing the help file). This sends the currently selected topic (part
of a chapter) to the printer.
The Microsoft Help Compiler does not support the use of subscripts and superscripts,
so the equations will not look like the original document. In addition, some of the
original mathematical symbols and greek symbols have been altered from the original
to be compatible with the Microsoft help format. For this reason, if you have a
question about an equation, we suggest going to the original manual.
2 1 Introduction Electronic Manual for MODFLOW
Abstract
This report presents a finite-difference model and its associated modular computer
program. The model simulates flow in three dimensions. The report includes
detailed explanations of physical and mathematical concepts on which the model is
based and an explanation of how those concepts are incorporated in the modular
structure of the computer program. The modular structure consists of a Main
Program and a series of highly independent subroutines called "modules." The
modules are grouped into "packages." Each package deals with a specific feature of
the hydrologic system which is to be simulated, such as flow from rivers or flow into
drains, or with a specific method of solving linear equations which describe the flow
system, such as the Strongly Implicit Procedure or Slice-Successive Overrelaxation.
The division of the program into modules permits the user to examine specific
hydrologic features of the model independently. This also facilitates development of
additional capabilities because new packages can be added to the program without
modifying the existing packages. The input and output systems of the computer
program are also designed to permit maximum flexibility.
Ground-water flow within the aquifer is simulated using a block-centered finite-
difference approach. Layers can be simulated as confined, unconfined, or a
combination of confined and unconfined. Flow associated with external stresses,
such as wells, areal recharge, evapotranspiration, drains, and streams, can also be
simulated. The finite-difference equations can be solved using either the Strongly
Implicit-Procedure or Slice-Successive Overrelaxation.
The program is written in FORTRAN 77 and will run without modification of most
computers that have a FORTRAN 77 compiler. For each program module, this
report includes a narrative description, a flow chart, a list of variables, and a module
listing.
Purpose
Since their inception, the two- and three-dimensional finite-difference models
described by Trescott (1975), Trescott and Larson (1976), and Trescott, Pinder, and
Larson (1976) have been used extensively by the U.S. Geological Survey and others
for the computer simulation of ground-water flow. The basic concepts embodied in
those models have been incorporated in the model presented here. The primary
objectives in designing a new ground-water flow model were to produce a program
that could be readily modified, was simple to use and maintain, could be executed on
a variety of computers with minimal changes, and was relatively efficient with
respect to computer memory and execution time.
The model program documented in this report uses a modular structure wherein
similar program functions are grouped together, and specific computational and
hydrologic options are constructed in such a manner that each option is independent
of other options. Because of this structure, new options can be added without the
necessity of changing existing subroutines. In addition, subroutines pertaining to
options that are not being used can be deleted, thereby reducing the size of the
program. The model may be used for either two- or three-dimensional applications.
Input procedures have been generalized so that each type of model input data may be
stored and read from separate external files. Variable formatting allows input data
arrays to be read in any format without modification to the program. The type of
output that is available has also been generalized so that the user may select various
model output options to suit a particular need. The program was originally written
using FORTRAN 66 (McDonald and Harbaugh, 1984). It has subsequently been
modified to use FORTRAN 77. This report documents the FORTRAN 77 version.
The program is highly portable; it will run, without modification, on most computers.
On some computers, minor modification may be necessary or desirable. A discussion
about program portability is contained in Appendix A.
The major options that are presently available include procedures to simulate the
effects of wells, recharge, rivers, drains, evapotranspiration, and "general-head
boundaries". The solution-algorithms available include two iteration techniques, the
Strongly Implicit Procedure (SIP) and the SliceSuccessive Overrelaxation method
(SSOR).
Organization of This Report
The purpose of this report is to describe the mathematical concepts used in this
program, the design of the program, and the input needed to use the program. The
program has been divided into a main program and a series of highly independent
subroutines called modules. The modules, in turn, have been grouped into
"packages." A package is a group of modules that deals with a single aspect of the
simulation. For example, the Well Package simulates the effect of wells, the River
Package simulates the effect of rivers, and the SIP Package solves a system of
equations using the Strongly Implicit Procedure. Many of the packages represent
options which the user may or may not have occasion to use. Each of the packages is
described in a separate chapter of this report. Two preliminary chapters describe
topics relating to the overall program; Chapter 2 derives the finite-difference
equation that is used in the model and Chapter 3 describes the overall design of the
program. Chapter 14 describes utility modules that are used by various packages to
perform special tasks. Appendices A-E cover topics relating to the operation of the
model.
Chapters 4 through 13 describe individual packages. The description of each
package consists of (1) a section entitled "Conceptualization and Implementation,"
(2) input instructions for the package, and (3) documentation of the individual
modules contained in the package. The Conceptualization and Implemementation
section describes the physical and mathematical concepts used to build the package.
For example, in the chapter describing the River Package, an equation is derived
which approximates flow through a riverbed, and a discussion is provided to show
how that equation can be incorporated into the finite-difference equation. Chapters
12 and 13 describe the solution procedures currently available in the model.
4 1 Introduction Electronic Manual for MODFLOW
The input instructions in Chapters 4 through 13 are presented in terms of input
"items." An item of input may be a single record or a collection of similar records, or
it may be an array or a collection of arrays.(In the model described herein, three-
dimensional arrays are always read as a collection of two-dimensional arrays, one
associated with each model layer.) The input section in each chapter presents a list of
the input items associated with the package described in that chapter; the entries in
this list are numbered, and generally consist of two lines (sometimes followed by a
note or comment). For items which consist of a single record or a group of similar
records, the first line in the entry gives the names of the fields comprising the
records, while the second line shows the format of those fields, in standard
FORTRAN notation. For an input item which consists of an array, the first line of
the entry gives the name of the array, while the second line gives the name of the
utility module which reads the array. Further details concerning utility modules are
provided in Chapter 14.
For most of the packages, the list of input items is subdivided into two major
sections. One of these falls under the heading "FOR EACH SIMULATION" and
includes all items for which only one entry is needed in each simulation; the other
falls under the heading "FOR EACH STRESS PERIOD", and includes those items
for which several entries may be needed in each simulation (for example, pumping
rate, which may change with time during the period represented in a simulation).
These major sections of the input list are further subdivided by headings which
indicate the modules (subroutines) which read the item, or, in the case of an array,
which call a utility subroutine to read the array. Input items that are printed entirely
in capital letters are used as FORTRAN variables or arrays in the model program;
input items which appear in mixed upper and lower case print are terms used in the
instructions to describe the input fields or procedures, and do not appear in the model
itself as FORTRAN variables. Chapter 4, which describes the Basic Package,
includes two lists of input items; one of these describes input which is always
required, while the other describes input associated with the optional "output control"
section of the Basic Package.
An explanation of input fields is presented following the list of input items in
Chapters 4 through 13. This explanation is followed in most cases by a sample input
for the package under consideration. In Chapter 4, again, the input items associated
with the output control option are treated separately; thus an independent explanation
of fields and sample input are provided for output control.
In each simulation, the user must designate which of the options of the program are to
be utilized, and must indicate the file from which the input for each option is to be
read. This is done through a one-dimensional array, IUNIT; the entries in this array
are the unit numbers associated with the required files by the computer operating
system. A location in the IUNIT array is given at the beginning of the input sections
in Chapters 5 through 13, and at the beginning of the input discussion for "output
control" in Chapter 4. If the option is to be utilized, the user must enter, in the
designated IUNIT array location, the unit number of the file or channel through
which input for the option is to be read; if the option is not required a zero is entered
in this location. Further discussion of the IUNIT array is provided in Chapters 3 and
4.
Following the input section in Chapters 4 through 13, each chapter provides a
documentation of the modules making up the associated package. This
documentation consists of a list of the modules in the package, followed by detailed
descriptions of each of the modules. The detailed description of a module generally
contains four documents: (1) a narrative description of the module, (2) a flow chart of
the module, (3) a FORTRAN listing of the module, and (4) a list of the variable
names which are used in the module. For very simple modules, the flow chart is
omitted. The narrative description is a numbered list of the functions performed by
the module showing the order in which they are performed. The flow chart is a
graphic equivalent of the narrative. The blocks in the flow chart are numbered with
the same numbers used in the narrative so that the two documents can be cross
referenced. An explanation of terms used in the flow chart is contained on the sheet
with the flow chart. The program listing contains comments with numbers
corresponding to those used in the flow charts and the narratives. The fourth record
of the listing contains a comment showing the time and day that the module was last
modified. The list of variables shows the name, range, and definition of every
variable used in the module. If the variable is used only in that module, its range is
given as "Module"; if it is used in other modules of the package, but not outside the
package, its range is given as "Package"; if it is used in the modules of more than one
package, its range is given as "Global."
To summarize the organization of this report, Chapters 2 and 3, and the
"Conceptualization and Implementation" section of Chapter 4, provide discussions
relevant to the overall design and functioning of the program; the formulation of
coefficients representing flow within the aquifer is discussed under
"Conceptualizationand Implementation" in Chapter 5; Chapters 6 through 11 provide
discussions of particular external sources or sinks and their representation in the
model; and Chapters 12 and 13 discuss the operation of particular solvers for the
systems of finite difference equations generated in the model. Input instructions for
each package are provided in the relevant chapter; a discussion of input for utility
modules is provided in Chapter 14. The appendices provide a sample problem,
abbreviated input instructions, and discussions of certain computer-related topics.
Acknowledgement
The authors wish to extend special thanks to Gordon Bennett. In addition to
providing the administrative support for the model development, he provided
encouragement and guidance along the way. His critical review of the report greatly
improved its clarity.
Electronic Manual for MODFLOW 2 Model Derivation 7
2 Model Derivation
Mathematical Model
The three-dimensional movement of ground water of constant density through porous
earth material may be described by the partial-differential equation
x
(
K
h
x
)+
y
(
K
h
y
)+
z
(
K
h
z
) - W =
S
h
t
xx yy zz s
(1)
where
K
xx
, K
yy
, and K
zz
are values of hydraulic conductivity along the x, y,and z coordinate
axes, which are assumed to be parallel to the major axes of hydraulic conductivity
(Lt
-1
);
h is the potentiometric head (L);
W is a volumetric flux per unit volume and represents sources and/or sinks of
water (t
-1
)
S
s
is the specific storage of the porous material (L
-1
); and
t is time (t).
For a derivation of equation (1) see for example Rushton and Redshaw (1979). In
general, S
s
, K
xx
, K
yy
, and K
zz
may be functions of space (S
s
= S
s
(X,Y,Z), K
xx
=
K
xx
(x,y,z), etc.) and W may be a function of space and time (W = W(x,y,z,t));
equation (1) describes ground-water flow under nonequilibrium conditions in a
8 2 Model Derivation Electronic Manual for MODFLOW
heterogeneous and anisotropic medium, provided the principal axes of hydraulic
conductivity are aligned with the coordinate directions.
Equation (1), together with specification of flow and/or head conditions at the
boundaries of an aquifer system and specification of initial-head conditions,
constitutes a mathematical representation of a ground-water flow system. A solution
of equation (1), in an analytical sense, is an algebraic expression giving h(x,y,z,t)
such that, when the derivatives of h with respect to space and time are substituted into
equation (1), the equation and its initial and boundary conditions are satisfied. A
time-varying head distribution of this nature characterizes the flow system, in that it
measures both the energy of flow and the volume of water in storage, and can be used
to calculate directions and rates of movement.
Except for very simple systems, analytical solutions of equation (1) are rarely
possible, so various numerical methods must be employed to obtain approximate
solutions. One such approach is the finite-difference method, wherein the continuous
system described by equation (1) is replaced by a finite set of discrete points in space
and time, and the partial derivatives are replaced by terms calculated from the
differences in head values at these points. The process leads to systems of
simultaneous linear algebraic difference equations; their solution yields values of
head at specific points and times. These values constitute an approximation to the
time-varying head distribution that would be given by an analytical solution of the
partial-differential equation of flow.
The finite-difference analog of equation (1) may be derived by applying the rules of
difference calculus; however, in the discussion presented here, an alternative
approach is used with the aim of simplifying the mathematical treatment and
explaining the computational procedure in terms of familiar physical concepts
regarding the flow system.
Discretization Convention
Figure 1 shows a spatial discretization of an aquifer system with a mesh of blocks
called cells, the locations of which are described in terms of rows, columns, and
layers. An i,j,k indexing system is used. For a system consisting of "nrow" rows,
"ncol" columns, and "nlay" layers, i is the row index, i = 1,2,. . . nrow; j is the
column index, j = 1,2,. . . ncol; and k is the layer index, k = 1,2,. . . nlay. For
example, figure 1 shows a system with nrow = 5, ncol = 9, and nlay = 5. In
formulating the equations of the model, an assumption was made that layers would
generally correspond to horizontal geohydrologic units or intervals. Thus in terms of
Cartesian coordinates, the k index denotes changes along the vertical, z; because the
convention followed in this model is to number layers from the top down, an
increment in the k index corresponds to a decrease in elevation. Similarly rows
would be considered parallel to the x axis, so that increments in the row index, i,
would correspond to decreases in y; and columns would be considered parallel to the
y axis, so that increments in the column index, j, would correspond to increases in x.
These conventions were followed in constructing figure 1; however, applications of
the model requires only that rows and columns fall along consistent orthogonal
directions within the layers, and does not require the designation of x, y, or z
coordinate axes.
Figure l.-A discretized hypothetical aquifer system.
Following the conventions used in figure 1, the width of cells in the row direction, at
a given column, j, is designated r
j
; the width of cells in the column direction at a
given row, i, is designated c
i
; and the thickness of cells in a given layer, k, is
designated v
k
. Thus a cell with coordinates (i,j,k) = (4,8,3) has a volume of
r
8
c
4
v
3
. Within each cell there is a point called a "node" at which head is to be
calculated. Figure 2 illustrates, in two dimensions, two conventions for defining the
configuration of cells with respect to the location of nodes--the block-centered
formulation and the point-centered formulation. Both systems start by dividing the
aquifer with two sets of parallel lines which are orthogonal. In the block-centered
formulation, the blocks formed by the sets of parallel lines are the cells; the nodes are
at the center of the cells. In the point-centered formulation, the nodes are at the
intersection points of the sets of parallel lines, and cells are drawn around the nodes
with faces halfway between nodes. In either case, spacing of nodes should be chosen
so that the hydraulic properties of the system are, in fact, generally uniform over the
extent of a cell. The finite-difference equation developed in the following section
holds for either formulation; however, only the block-centered formulation is
presently used in the model.
In equation (1), the head, h, is a function of time as well as space so that, in the finite-
difference formulation, discretization of the continuous time domain is also required.
Finite-Difference Equation
Development of the ground-water flow equation in finite-difference form follows
from the application of the continuity equation: the sum of all flows into and out of
the cell must be equal to the rate of change in storage within the cell. Under the
assumption that the density of ground water is constant, the continuity equation
expressing the balance of flow for a cell is
i
Q = SS
h
t
V

where
Q
i
is a flow rate into the cell (L
3
t
-1
);
SS has been introduced as the notation for specific storage in the finite-difference
formulation; its definition is equivalent to that of SS in equation (1)--i.e., it is the
volume of water which can be injected per unit volume of aquifer material per unit
change in head (L
-1
);
V is the volume of the cell (L
3
); and
h is the change in head over a time interval of length t.
Figure 2.-Grids showing the difference between block-centered and point-centered
grids.
The term on the right hand side is equivalent to the volume of water taken into
storage over a time interval t given a change in head of h. Equation (2) is stated in
terms of inflow and storage gain. Outflow and loss are represented by defining
outflow as negative inflow and loss as negative gain.
Figure 3 depicts a cell i,j,k and six adjacent aquifer cells i-1,j,k; i+1,j,k; i,j-1,k;
i,j+1,k; i,j,k-1; and i,j,k+1. To simplify the following development, flows are
considered positive if they are entering cell i,j,k; and the negative sign usually
incorporated in Darcy's law has been dropped from all terms. Following these
conventions, flow into cell i,j,k in the row direction from cell i,j-1,k (figure 4), is
given by Darcy's law as
i, j - 1 / 2, k
i, j - 1 / 2, k i k
i, j - 1, k i, j, k
j - 1 / 2
q =
KR c v
h
-
h
r

where
h
i,j,k
is the head at node i,j,k, and h
i,j-1,k
that at node i,j-l,k; q
i,j-1/2,k
is the volumetric
fluid discharge through the face between cells i,j,k and i,j-1,k (L
3
t
-1
);
KR
i,j-1/2,k
is the hydraulic conductivity along the row between nodes i,j,k and i,j-1,k
(Lt
-1
);
c
i
v
k
is the area of the cell faces normal to the row direction; and r
j-1/2
is the
distance between nodes i,j,k and i,j-1,k (L).
Figure 3.-Cell i,j,k and indices for the six adjacent cells.
Figure 4.-Flow into cell i,j,k from cell i,j-1,k.
Although the discussion is phrased in terms of flow into the central cell, it can be
misleading to associate the subscript j-1/2 of equation (3) with a specific point
between the nodes. Rather, the term KR
i,j-1/2,k
of equation (3) is the effective
hydraulic conductivity for the entire region between the nodes, normally calculated as
a harmonic mean in the sense described by, for example, Collins (1961). If this is
done, equation (3) gives the exact flow, for a one-dimensional steady-state case,
through a block of aquifer extending from node i,j-1,k to node i,j,k and having a
cross sectional area c
i
v
k
.
Similar expressions can be written approximating the flow into the cell through the
remaining five faces, i.e., for flow in the row direction through the face between cells
i,j,k and i,j+l,k,
i, j +1 / 2, k
i, j +1 / 2, k i k
i, j +1, k i, j, k
j +1 / 2
q =
KR c v
h
-
h
r

while for the column direction, flow into the block through the forward face is
i +1 / 2, j, k
i +1 / 2, j, k j k
i +1, j, k i, j, k
i +1 / 2
q =
KC r v
h
-
h
c


and flow into the block through the rear face is
i - 1 / 2, j, k
i - 1 / 2, j, k j k
i - 1, j, k i, j, k
i - 1 / 2
q =
KC r v
h
-
h
c

For the vertical direction, inflow through the bottom face is

i, j, k +1 / 2
i, j, k +1 / 2 j i
i, j, k +1 i, j, k
k +1 / 2
q =
KV r c
h
-
h
v

while inflow through the upper face is given by

i, j, k - 1 / 2
i, j, k - 1 / 2 j i
i, j, k - 1 i, j, k
k - 1 / 2
q =
KV r c
h
-
h
v

Each of equations (3)-(8) expresses inflow through a face of cell i,j,k in terms of
heads, grid dimensions, and hydraulic conductivity. The notation can be simplified
by combining grid dimensions and hydraulic conductivity into a single constant, the
"hydraulic conductance" or, more simply, the "conductance," For example,
i, j -1 / 2, k
i, j -1 / 2, k i k
j - 1 / 2
CR
=
KR c v
r

where
CR
i,j-1/2,k
is the conductance in row i and layer k between nodes i,j-1,k and
i,j,k (L
2
t
-1
).
Conductance is thus the product of hydraulic conductivity and cross-sectional area of
flow divided by the length of the flow path (in this case, the distance between the
nodes.) Substituting conductance from equation (9) into equation (3) yields
q
i,j-1/2,k
= CR
i,j-1/2,k
(h
i,j-1,k
-h
i,j,k
). (10)
Similarly, equations (4)-(8) can be rewritten to yield
q
i,j+1/2,k
= CR
i,j+1/2,k
(h
i,j+1,k
- h
i,j,k
) (11)
q
i-1/2,j,k
= CC
i-1/2,j,k
(h
i-1,j,k
- h
i,j,k
) (12)
q
i+1/2,j,k
= CC
i+1/2,j,k
(h
i+1,j,k
- h
i,j,k
) (13)
q
i,j,k-1/2
= CV
i,j,k-1/2
(h
i,j,k-1
- h
i,j,k
) (14)
q
i,j,k+1/2
= CV
i,j,k+1/2
(h
i,j,k+1
- h
i,j,k
) (15)
where the conductances are defined analogously to CR
i,j-1/2,k
in equation (9).
Equations (10)-(15) account for the flow into cell i,j,k from the six adjacent cells. To
account for flows into the cell from features or processes external to the aquifer, such
as streams, drains, areal recharge, evapotranspiration or wells, additional terms are
required. These flows may be dependent on the head in the receiving cell but
independent of all other heads in the aquifer, or they may be entirely independent of
head in the receiving cell. Flow from outside the aquifer may be represented by the
expression
a
i,j,k,n
= P
i,j,k,n
h
i,j,k
+ q
i,j,k,n
(16)
where
a
i,j,k,n
represents flow from the nth external source into cell i,j,k (L
3
t
-1
), and P
i,j,k,n
and
q
i,j,k,n
are constants ((L
2
t
-1
) and (L
3
t
-1
), respectively).
For example, suppose a cell is receiving flow from two sources, recharge from a well
and seepage through a riverbed. For the first source (n=1), since the flow from the
well is assumed to be independent of head, P
i,j,k,1
is zero and q
i,j,k,1
is the recharge rate
for the well. In this case,
a
i,j,k,1
= q
i,j,k,1
. (17)
For the second source (n=2), the assumption is made that the streamaquifer
interconnection can be treated as a simple conductance, so that the seepage is
proportional to the head difference between the river stage and the head in cell i,j,k
(figure 5); thus we have
a
i,j,k,2
= CRIV
i,j,k,2
(R
i,j,k
- h
i,j,k
) (18)
where R
i,j,k
is the head in the river (L) and CRIV
i,j,k,2
is a conductance (L
2
t
-1
)
controlling flow from the river into cell i,j,k. For example, in the situation shown
schematically in figure 5, CRIV would be given as the product of the vertical
hydraulic conductivity of the riverbed material and the area of the streambed as it
crosses the cell, divided by the thickness of the streambed material. Equation (18)
can be rewritten as
a
i,j,k,2
= - CRIV
i,j,k,2
h
i,j,k
+ CRIV
i,j,k,2
R
i,j,k
. (19)
The negative conductance term, -CRIV
i,j,k,2
corresponds to P
i,j,k,2
of equation 16,
while the term CRIV
i,j,k,2
R
i,j,k
corresponds to q
i,j,k,2
. Similarly, all other external
sources or stresses can be represented by an expression of the form of equation 16.
In general, if there are N external sources or stresses affecting a single cell, the
combined flow is expressed by
i, j,k
n=1
N
i, j,k,n
n=1
N
i, j,k,n i, j,k
n=1
N
i, j,k,n
QS =
a
p
h
+ q =

(20)
Figure 5.-Conceptual representation of leakage through a riverbed into a cell.
Defining P
i,j,k
and Q
i,j,k
by the expressions
i, j,k
n=1
N
i, j,k,n
P
= p
i, j,k
n=1
N
i, j,k,n
Q = q
the general external flow term for cell i,j,k is

QS
i,j,k
= P
i,j,k
h
i,j,k
+ Q
i,j,k
. (21)
Applying the continuity equation (2) to cell i,j,k, taking into account the flows from
the six adjacent cells, as well as the external flow rate, QS, yields
i, j -1/ 2,k i, j+1/ 2,k i -1/ 2, j,k i+1/ 2, j,k i, j,k -1/ 2 i, j,k+1/ 2 i, j,k i, j,k
i, j,k
j i k
q +q +q +q +q +q +QS =
SS
h
t
r c v

(22)
where
h
i,j,k
/ t is a finite-difference approximation for the derivative of head with respect
to time (Lt
-1
);
SS
i,j,k
represents the specific storage of cell i,j,k (L
-1
);
and r
j
c
i
v
k
is the volume of cell i,j,k (L
3
).
Equations (10) through (15) and (21) may be substituted into equation (22) to give
the finite-difference approximation for cell i,j,k as
CR
i,j-1/2,k
(h
i,j-1,k
- h
i,j,k
) + CR
i,j+1/2,k
(hi
,j+1,k
- h
i,j,k
)
+ CC
i-1/2,j,k
(h
i-1,j,k
- h
i,j,k
) + CC
i+1/2,j,k
(h
i+1,j,k
-h
i,j,k
)
+ CV
i,j,k-1/2
(h
i,j,k-1
- h
i,j,k
) + CV
i,j,k+1/2
(h
i,j,k+1
- h
i,j,k
)
+ P
i,j,k
h
i,j,k
+ Q
i,j,k
= SS
i,j,k
(r
j
c
i
v
k
) hi
,j,k
/ t. (23)
The finite-difference approximation for the time derivative of head, h
i,j,k
/t must
next be expressed in terms of specific heads and times. Figure 6 shows a hydrograph
of head values at node i.j,k. Two values of time are shown on the horizontal axis: t
m
,
which is the time at which the flow terms of equation (23) are evaluated; and t
m-1
, a
time which precedes t
m
. The head values at node i,j,k associated with these times are
designated by superscript as h
m
i,j,k
and h
m
i-1,j,k
, respectively. An approximation to the
time derivative of head at time t
m
is obtained by dividing the head difference h
m
i,j,k
-
h
m
i-1,j,k
by the time interval t
m
-t
m-1
; that is,
i, j,k i, j,k
m
i, j,k
m-1
m m-1
h
t
=
h
-
h
t
-
t
Thus the hydrograph slope, or time derivative, is approximated using the change in
head at the node over a time interval which precedes, and ends with, the time at
which flow is evaluated. This is termed a backward-difference approach, in that
h/t is approximated over a time interval which extends backward in time from tm,
the time at which the flow terms are calculated. There are other ways in which h/t
could be approximated; for example, we could approximate it over a time interval
which begins at the time of flow evaluation and extends to some later time; or over a
time interval which is centered at the time of flow evaluation, extending both forward
and backward from it. These alternatives, however, may cause numerical instability--
that is, the growth or propagation of error during the calculation of heads at
successive times in a simulation.
Figure 6.-Hydrograph for cell i,j,k.
In an unstable situation, errors which enter the calculation for any reason at a
particular time will increase at each succeeding time as the calculation progresses,
until finally they completely dominate the result. By contrast, the backward-
difference approach is always numerically stable--that is, errors introduced at any
time diminish progressively at succeeding times. For this reason, the backward-
difference approach is preferred even though it leads to large systems of equations
which must be solved simultaneously for each time at which heads are to be
computed.
Equation (23) can be rewritten in backward-difference form by specifying flow terms
at tm, the end of the time interval, and approximating the time derivative of head over
the interval t
m-1
to t
m
; that is:
i, j-1/ 2,k
i, j -1,k
m
i, j,k
m
i, j+1/ 2,k
i, j+1,k
m
i, j,k
m
CR
(
h
-
h
)+
CR
(
h
-
h
)
+
CC
(
h
-
h
)+
CC
(
h
-
h
)
i -1/ 2, j,k
i -1, j,k
m
i, j,k
m
i+1/ 2, j,k
i+1, j,k
m
i, j,k
m
+
CV
(
h
-
h
)+
CV
(
h
-
h
)
i, j,k-1/ 2
i, j,k-1
m
i, j,k
m
i, j,k+1/ 2
i, j,k+1
m
i, j,k
m
+
P
h
+Q =
SS
(
r c v
)
h
-
h
t
-
t
i, j,k
i, j,k
m
i, j,k
i, j,k j i k
i, j,k
m
i, j,k
m-1
m m-1

Equation (24) is a backward-difference equation which can be used as the basis for a
simulation of the partial differential equation of ground water flow, equation (1).
Like the term Q
i,j,k
, the coefficients of the various head terms in equation (24) are all
known, as is the head at the beginning of the time step, h
m-1
i,j,k
. The seven heads at
time t
m
, the end of the time step, are unknown; that is, they are part of the head
distribution to be predicted. Thus equation (24) cannot be solved independently,
since it represents a single equation in seven unknowns. However, an equation of
this type can be written for each active cell in the mesh; and, since there is only one
unknown head for each cell, we are left with a system of "n" equations in "n"
unknowns. Such a system can be solved simultaneously.
The objective of transient simulation is generally to predict head distributions at
successive times, given the initial head distribution, the boundary conditions, the
hydraulic parameters and the external stresses. The initial-head distribution provides
a value of h
1
i,j,k
at each point in the mesh --- that is, it provides the values of head at
the beginning of the first of the discrete time steps into which the time axis is divided
in the finite-difference process. The first step in the solution process is to calculate
values of h
2
i,j,k
--that is, heads at time t
2
, which marks the end of the first time step. In
equation (25), therefore, the head superscript m is taken as 2, while the superscript
m-1, which appears in only one head term, is taken as 1. The equation therefore
becomes
i, j-1/ 2,k
i, j -1,k
2
i, j,k
2
i, j+1/ 2,k
i, j+1,k
2
i, j,k
2
CR
(
h
-
h
)+
CR
(
h
-
h
)
+
CC
(
h
-
h
)+
CC
(
h
-
h
)
i -1/ 2, j,k
i -1, j,k
2
i, j,k
2
i+1/ 2, j,k
i+1, j,k
2
i, j,k
2
+
CV
(
h
-
h
)+
CV
(
h
-
h
)
i, j,k-1/ 2
i, j,k-1
2
i, j,k
2
i, j,k+1/ 2
i, j,k+1
2
i, j,k
2
+
P
h
+Q =
SS
(
r c v
)
h
-
h
t
-
t
i, j,k
i, j,k
2
i, j,k
i, j,k j i k
i, j,k
2
i, j,k
1
2 1

where again the superscripts 1 and 2 refer to the time at which the heads are taken
and should not be interpreted as exponents.
An equation of this form is written for every cell in the mesh in which head is free to
vary with time (variable-head cells), and the system of equations is solved
simultaneously for the heads at time t
2
. When these have been obtained, the process
is repeated to obtain heads at time t
3
, the end of the second time step. To do this,
equation (25) is reapplied, now using 2 as time subscript m-1 and 3 as time subscript
m. Again, a system of equations is formulated, where the unknowns are now the
heads at time t
3
; and this set of equations is solved simultaneously to obtain the head
distribution at time t3. This process is continued for as many time steps as necessary
to cover the time range of interest.
It is important to note that the set of finite-difference equations is reformulated at
each time step; that is, at each step there is a new system of simultaneous equations to
be solved. The heads at the end of the time step make up the unknowns for which
this system must be solved; the heads at the beginning of the step are among the
known terms in the equations. The solution process is repeated at each time step
yielding a new array of heads for the end of the step.
Iteration
The model described in this report utilizes iterative methods to obtain the solution to
the system of finite-difference equations for each time step. In these methods, the
calculation of head values for the end of a given time step is started by arbitrarily
assigning a trial value, or estimate, for the head at each node at the end of that step.
A procedure of calculation is then initiated which alters these estimated values,
producing a new set of head values which are in closer agreement with the system of
equations. These new, or interim, head values then take the place of the initially
assumed heads, and the procedure of calculation is repeated, producing a third set of
head values. This procedure is repeated successively, at each stage producing a new
set of interim heads which more nearly satisfies the system of equations. Each
repetition of the calculation is termed an "iteration." Ultimately, as the interim heads
approach values which would exactly satisfy the set of equations, the changes
produced by succeeding stages of calculation become very small. This behaviour is
utilized in determining when to stop iteration, as discussed in a subsequent
paragraph.
Thus, during the calculations for a time step, arrays of interim head values are
generated in succession, each array containing one interim head value for each active
node in the mesh. In figure 7, these arrays are represented as three-dimensional
lattices, each identified by an array symbol, h, bearing two superscripts. The first
superscript indicates the time step for which the heads in the array are calculated,
while the second indicates the number, or level, of the iteration which produced the
head array. Thus h
m,2
represents the array of values computed in the first iteration for
the end of step M; h
m,2
would represent the array of values computed in the second
iteration; and so on. The head values which were initially assumed for the end of
time step m, to begin the process of iteration, appear in the array designated h
m,o
. In
the example of figure 7, a total of n iterations is required to achieve closure for the
heads at the end of time step m; thus the array of final head values for the time step is
designated h
m,n
. Figure 7 also shows the array of final head values for the end of the
preceding time step h
m-1,n
(where again it is assumed that n iterations were required
for closure). The head values in this array appear in the storage term of equation
(24)--i.e., they are used in the term h
m-1
i,j,k
on the right side of equation (24)--in the
calculation of heads for time step m. Because they represent heads for the preceding
time step, for which computations have already been completed, they appear as
predetermined constants in the equation for time step m; thus they retain the same
value in each iteration of the time step. Similarly, the final values of head for time
step m are used as constants in the storage term during calculations for time step
m+1.
Figure 7.-Iterative calculation of a head distribution.
Ideally, one would like to specify that iteration stop when the calculated heads are
suitably close to the exact solution. However, because the actual solution is
unknown, an indirect method of specifying when to stop iterating must be used. The
method most commonly employed is to specify that the changes in computed heads
occuring from one iteration level to the next must be less than a certain quantity,
termed the "closure criterion" or "convergence criterion," which is specified by the
user. After each iteration, absolute values of computed head change in that iteration
are examined for all nodes in the mesh. The largest of these absolute head change
values is compared with the closure criterion. If this largest value exceeds the
closure criterion, iteration continues; if it is less than the closure criterion, iteration is
said to have "closed" or "converged," and the process is terminated for that time step.
Normally, this method of determining when to stop iteration is adequate. Note that
the closure criterion refers to change in computed head, and that values of head are
not themselves necessarily calculated to a level of accuracy comparable to the closure
criterion. As a rule of thumb, it is wise to use a value of closure criterion that is an
order of magnitude smaller than the level of accuracy desired in the head results.
The program described herein also incorporates a maximum permissible number of
iterations per time step. If closure is not achieved within this maximum number of
iterations, the iterative process will be terminated and a corresponding message
printed in the output. The closure criterion is designated HCLOSE in the model
input, while the maximum number of iterations per time step is designated MXITER.
The initial estimates of head for the end of time step m, in array h
m,o
of figure 7,
could be assigned arbitrarily, or they could be chosen according to a number of
different conventions. Theoretically, the iterative process would eventually converge
to the same result regardless of the choice of initial head values, although the work
required would be much greater for some choices than for others. In the model
described in this report, the heads computed for the end of each time step are used as
the initial trial values of head for the end of the succeeding time step. Thus in figure
7, the array h
m-1,n
contains the final estimates of head, obtained after n iterations, for
the end of time step m-1. When the calculations for step m-1 are complete, these
same values of head are transferred to the array h
m,o
, and used as the initial estimates,
or trial values, for the heads at the end of time step m. Head values for the end of the
first time step in the simulation are assumed initially to be equal to the heads
specified by the user for the beginning of the simulation.
Discussions of the mathematical basis of various iterative methods may be found in
many standard references, including Peaceman (1977), Crichlow (1977) and Remson,
Hornberger and Molz (1971). It is suggested that the reader review one of these
discussions, both to clarify general concepts and to provide an introduction to such
topics as the use of matrix notation, the role of iteration parameters, and the influence
of various factors on rate of convergence. In particular, such a review is
recommended prior to reading Chapters 12 and 13 of this report.
An iterative procedure yields only an approximation to the solution of the system of
finite-difference equations for each time step; the accuracy of this approximation
depends upon several factors, including the closure criterion which is employed.
However, it is important to note that even if exact solutions to the set of finite-
difference equations were obtained at each step, these exact solutions would
themselves be only an approximation to the solution of the differential equation of
flow (equation (1)). The discrepancy between the head, h
m
i,j,k
, given by the solution
to the system of difference equations for a given node and time, and the head
h(x
i
,y
j
,z
k
,t
m
) which would be given by the formal solution of the differential equation
for the corresponding point and time, is termed the truncation error. In general, this
error tends to become greater as the mesh spacing and time-step length are increased.
Finally, it must be recognized that even if a formal solution of the differential
equation could be obtained, it would normally be only an approximation to
conditions in the field, in that hydraulic conductivity and specific storage are seldom
known with accuracy, and uncertainties with regard to hydrologic boundaries are
generally present.
Formulation of Equations for Solution
The model described in this report presently incorporates two different options for
iterative solution of the set of finite-difference equations, and is organized so that
alternative schemes of solution may be added without disruption of the program
structure. Whatever scheme of solution is employed, it is convenient to rearrange
equation (24) so that all terms containing heads at the end of the current time step are
grouped on the left-hand side of the equation, and all terms that are independent of
head at the end of the current time step are on the right-hand side. The resulting
equation is
i, j,k-1/ 2
i, j,k-1
m
i -1/ 2, j,k
i, j -1,k
m
i, j -1/ 2,k
i, j-1,k
m
CV
h
+
CC
h
+
CR
h
+(-
CV
-
CC
-
CR
-
CR i, j,k -1/ 2 i -1/ 2, j,k i, j -1/ 2,k i, j+1/ 2,k
-
CC
-
CV
+
HCOF
)
h
+
CR
h i+1/ 2, j,k i, j,k+1/ 2 i, j,k
i, j,k
m
i, j+1/ 2,k
i, j+1,k
m
+
CC
h
+
CV
h
=
RHS i+1/ 2, j,k
i+1, j,k
m
i, j,k+1/ 2
i, j,k+1
m
i, j,k
(26)
where
i, j,k i, j,k
i, j,k
m m-1
HCOF
=
P
- SC
1
t
-
t
(L
2
t
-1
)
i, j,k
i, j,k
i, j,k
i, j,k
m
m m-1
RHS
= - Q -
SC
1
h
t
-
t
(L
3
t
-1
)
SC
1
=
SS r c v i, j,k i, j,k j i k

(L
2
)
The entire system of equations of the form of (26), which includes one equation for
each variable-head cell in the mesh, may be written in matrix form as
[A] {h} = {q} (27)
where [A] is a matrix of the coefficients of head, from the left side of equation (26),
for all active nodes in the mesh; {h} is a vector of head values at the end of time step
m for all nodes in the mesh; and {q} is a vector of the constant terms, RHS, for all
nodes of the mesh. The model described in this report assembles the vector {q} and
the terms that comprise [A] through a series of subroutines, or "modules". The
vector {q} and the terms comprising [A] are then transferred to modules which
actually solve the matrix equations for the vector {h} .
Types of Model Cell and Simulation of Boundaries
In practice, it is generally unnecessary to formulate an equation of the form of (24)
for every cell in a model mesh, as the status of certain cells is specified in advance in
order to simulate the boundary conditions of the problem. In the model described in
this report, cells of this type are grouped into two categories--"constant-head" cells
and "inactive" (or "no-flow") cells. Constant-head cells are those for which the head
is specified in advance, and is held at this specified value through all time steps of the
simulation. Inactive or no-flow cells are those for which no flow into or out of the
cell is permitted, in any time step of the simulation. The remaining cells of the mesh,
termed "variable-head" cells in this report, are characterized by heads which are
unspecified and free to vary with time. An equation of the form of (24) must be
formulated for each variable-head cell in the mesh, and the resulting system of
equations must be solved simultaneously for each time step in the simulation.
Constant-head and no flow cells are used in the model described herein to represent
conditions along various hydrologic boundaries. For example, figure 8 shows the
map of an aquifer boundary superimposed on an array of cells generated for the
model. The aquifer is of irregular shape, whereas the model array is always
rectangular in outline; no-flow cells have therefore been used to delete the portion of
the array beyond the aquifer boundary. The figure also shows constant-head cells
along one section of the boundary; these may be used, for example, where the aquifer
is in direct contact with major surface water features. Other boundary conditions,
such as areas of constant inflow or areas where inflow varies with head, can be
simulated through the use of external source terms or through a combination of no-
flow cells and external source terms.
Figure 8.-Discretized aquifer showing boundaries and cell designations.
Conceptual Aspects of Vertical Discretization
The model described in this document handles discretization of space in the
horizontal direction by reading the number of rows, the number of columns and the
width of each row and column (that is, the width of the cells in the direction
transverse to the row or column). Discretization of space in the vertical direction is
handled in the model by specifying the number of layers to be used, and by
specifying hydraulic parameters which contain or embody the layer thickness. This
approach is followed in preference to explicit reading of layer thickness in order to
accomodate two different ways of viewing vertical discretization.
At one extreme, vertical discretization can be visualized simply as an extension of
areal discretization--a more or less arbitrary process of dividing the flow system into
segments along the vertical, governed in part by the vertical resolution desired in the
results. At the opposite extreme, vertical discretization can be viewed as an effort to
represent individual aquifers or permeable zones by individual layers of the model.
Figure 9-a shows a typical geohydrologic sequence which has been discretized
according to both interpretations--in 9-b according to the first viewpoint, and in 9-c
according to the second. The first viewpoint leads to rigid superposition of an
orthogonal three-dimensional mesh on the geohydrologic system; while there may be
a general correspondence between geohydrologic layers and model layers, no attempt
is made to make the mesh conform to stratigraphic irregulaties. Under the second
viewpoint, model layer thickness is considered variable, to simulate the varying
thickness of geohydrologic units; this leads, in effect, to a deformed mesh.
Figure 9.-Schemes of vertical discretization.
Each of these methods of viewing the vertical discretization process has advantages,
and each presents difficulties. The model equations are based on the assumption that
hydraulic properties are uniform within individual cells, or at least that meaningful
average or integrated parameters can be specified for each cell; these conditions are
more likely to be met when model layers conform to geohydrologic units as in figure
9-c. Moreover, greater accuracy can be expected if model layers correspond to inter-
vals within which vertical head loss is negligible, and this is also more likely under
the configuration of 9-c. On the other hand, the deformed mesh of 9-c fails to
conform to many of the assumptions upon which the model equations are based; for
example, individual cells may no longer have rectangular faces, and the major axes of
hydraulic conductivity may not be aligned with the model axis. Some error is always
introduced by these departures from assumed conditions.
In practice many vertical discretization schemes turn out to be a combination of the
viewpoints illustrated in figures 9-b and 9-c. For example, even where layer
boundaries conform to geohydrologic contacts, it may be necessary to use more than
one layer to simulate a single geohydrologic unit, simply to achieve the resolution
required in the results. Figure 10 shows a system consisting of two sand units
separated by a clay; the units are of uniform thickness, and each could be represented
by a single layer without deformation of the mesh. However, flow is neither fully
horizontal nor fully vertical in any of the layers; if information on the direction of
flow within each unit is required, several layers must be used to represent each unit.
Similarly, figure 11 shows a sand-clay system in which pumpage from the sands is
sustained partially by vertical flow of water released from storage in the clay. If the
objective of analysis is to determine the pattern of storage release in the clay, several
model layers would be required to represent that unit, as shown in the figure. On the
other hand, figure 12 shows a sand-clay system in which storage release occurs only
in the sands, flow in the sand is essentially horizontal, and flow in the clay is
essentially vertical. In this case a single model layer may be used to represent each
sand, while the clay may be represented simply by the vertical conductance between
layers. This approach to vertical discretization has sometimes been termed the "quasi
three-dimensional" approach.
Figure 10.-Possible pattern of flow in a cross section consisting of two high
conductivity units separated by a low conductivity unit.
Figure 11.-A cross section in which a low conductivity unit is represented by six
model layers.
The approaches to vertical discretization described above all lead to a set of
equations of the form of (26), which must be solved simutaneously at each time step.
The differences among these approaches arise in the way the various conductances
and storage terms are formulated and, in general, in the number of equations to be
solved, the resolution of the results, and the accuracy of the results. The model
described in this document is capable of implementing any of these approaches to
vertical discretization in that, as noted above, the thickness of individual layers ( v
k
of figure 1 and equation (24)) is never read explicitly by the program; rather, this
thickness is embedded in various hydraulic coefficients specified by the user. For
example, in confined layers transmissivity, which is the product of hydraulic
conductivity and layer thickness, is specified; and storage coefficient, the product of
specific storage and layer thickness, is also used,. For an unconfined layer, aquifer
bottom elevation and hydraulic conductivity are input for each cell. Saturated
thickness is calculated as head minus bottom elevation, and transmissivity is then
calculated as hydraulic conductivity times saturated thickness. Thus, layer thickness
can vary from cell to cell depending on bottom elevation and head. Chapter 5, which
describes the Block Centered Flow Package, contains a discussion of the formulation
of conductance and storage terms corresponding to the various ways of
conceptualizing the vertical discretization.
Figure 12.-A cross section in which a low conductivity unit is represented by the
conductance between model layers.
Electronic Manual for MODFLOW 3 Program Design 31
3 Program Design
Overall Structure
This chapter describes the overall design of the model program. The program
consists of a main program (MAIN) and a large number of highly independent
subroutines called modules. This chapter will explain the functions of MAIN and
explain how the modules can be grouped into "packages" and "procedures".
The functions which must be performed for a typical simulation are shown in figure
13. The period of simulation is divided into a series of "stress periods" within which
specified stress parameters are constant. Each stress period, in turn, is divided into a
series of time steps. The system of finite-difference equations of the form of
equation (27) is formulated and solved to yield the head at each node at the end of
each time step. Iterative solution methods are used to solve for the heads for each
time step. Thus within a simulation, there are three nested loops: a stress-period
loop, within which there is a time-step loop, which in turn contains an iteration loop.
Each rectangle in figure 13 is termed a "procedure". For example, prior to entering
the stress loop, the program executes three procedures which pertain to the
simulation as a whole. In the Define Procedure, the problem to be simulated is
defined: the size of the model, the type of simulation (transient or steady-state), the
number of stress periods, the hydrologic options, and the solution scheme to be used
are specified. In the Allocate Procedure, memory space required by the program is
allocated. In the Read and Prepare Procedure, all data that are not functions of time
are read. These data may include all or some of the following: boundary conditions,
initial heads (starting heads), transmissivity, hydraulic conductivity, specific yield,
storage coefficients, elevations of layer tops and bottoms, and parameters required by
the specified solution scheme. Certain preliminary calculations are also made in this
procedure to prepare data for further processing.
32 3 Program Design Electronic Manual for MODFLOW
Figure 13.-Overall program structure.
Within the stress period loop the first procedure is termed the Stress Procedure. In
this procedure the number of time steps (NSTP) in the stress period and certain
information to calculate the length of each time step are read. In a second Read and
Prepare Procedure, all data that pertain to a stress period, such as pumping rates and
areal recharge, are read and processed. The time-step loop is then entered (figure
13); in the Advance Procedure, the length of the time step is calculated and the heads
for the start of the time step are initialized. The iteration loop contains the Formulate
Procedure which determines the conductances and coefficients for each node as
required by equation (27), and the Approximate Procedure which approximates a
solution to the system of linear equations for head. Iteration proceeds until closure is
achieved or until a specified maximum number of allowable iterations is reached. At
the end of the iteration loop, the Output Control Procedure determines the disposition
of the computed heads, budget terms, and cell-by-cell flow terms. In the Budget
Procedure, budget entries are calculated and cell-by-cell flow terms are printed or
recorded, as explained in a subsequent section. In the Output Procedure, heads,
drawdown, and the volumetric budget are printed or recorded.
As shown in the preceding discussion, figure 13 provides a flow chart for the overall
program structure, a list of the various procedures, and an indication of the sequence
in which those procedures are implemented; it also provides a flow chart for the main
program of the model. The work within the procedures--i.e., within the rectangles of
figure 13--is performed by individual subroutines, or modules, called by the main
program. The main program itself is simply an organized sequence of call
statements, most of which are coupled to "IF" tests which determine whether a
module is required. Accordingly, the main program does not itself do the work of
simulation; it merely calls the various modules in the proper sequence to do that
work. Modules which are called directly by the main program are termed "primary"
modules; those that are called by other modules are termed "secondary" modules.
Thus the various procedures indicated in figure 13 are implemented through
individual modules; and the modules can accordingly be grouped according to the
procedure which they help to perform. As noted in Chapter 1, modules can also be
grouped by "packages", where a package (for example, the River Package, the Well
Package, or the SIP Package) includes those modules required to incorporate a
particular hydrologic process or solution algorithm into the simulation. In terms of
understanding the operation of the model, these two methods of grouping modules
are both useful. The package classification, for example, indicates which modules
will be active in a given simulation. (Modules are called by the main program only if
they are part of a package which is required in the simulation; and while some
packages are required in all simulations, most are needed only when the hydrologic
process or solution method embodied in the package is specified by the user.) The
procedure classification, on the other hand, defines the specific function of the
module in relation to the functions of other modules of the package. For example,
several modules whose function is to allocate space are grouped under the Allocate
Procedure; each of these modules allocates the space required for the arrays used in a
single package. If few options or features are specified, relatively few packages are
involved in the simulation, and the Allocate Procedure is handled by a relatively
small number of modules. As the options specified by the user increase, more
packages enter the simulation, and more modules are called to complete the space
allocation task.
Figure 14 illustrates the classification of modules by procedure and by package in
terms of a matrix of primary modules (i.e., modules called by the main program).
The horizontal rows in figure 14 correspond to procedures while the vertical columns
correspond to packages. An "X" is entered in each block of the matrix for which a
module exists; absence of an "X" indicates that the procedure in question is not
required in the indicated package. Entries marked with a subscript "S" indicate
primary modules which utilize submodules in accomplishing their function;
submodules are secondary modules which are utilized only in a single package.
Entries marked with the subscript "U" indicate primary modules which utilize utility
modules to accomplish their tasks; utility modules are secondary modules which are
available to many packages.
Figure 14.-Organization of modules by procedures and packages.
The primary modules are named according to a convention which indicates both the
package and the procedure to which they belong. The first three characters designate
the package, the fourth is a package version number, and the last two indicate the
procedure. For example, in figure 14, a module is indicated for the Well Package
and Allocate Procedure. This module is designated as WEL1AL; the first three
letters, WEL, indicate that the module is part of the Well Package; the last two
letters, AL, indicate that it performs the Allocate Procedure in that package. Thus
this module is one of those that deals with the simulation of specified withdrawal or
input, as through wells, and its particular function is to allocate the space in computer
memory used to store well data. The number one appearing in the fourth place of the
six-character module designation is a package version number. If the package is
modified to effect improvements, a different integer would be used in this place to
distinguish the modified package from the original or from other modified versions.
Figure 15 shows the names of the primary modules arranged in the same matrix
format that was used in figure 14. As in figure 14, a subscript "S" indicates that
submodules are utilized and "U" indicates that utility modules are utilized.
Submodules are designated by a six-character name in which the first character is
always the letter "S". This is followed by three characters designating the package
name, a numeral indicating the package version number, and a one-character
mnemonic to distinguish the module from other submodules of the same package; for
example, the secondary module "SBCF1C" is a submodule in version one of the
Block-Centered Flow Package. Utility modules are designated by the letter "U"
followed by a five-character mnemonic. For example, the secondary module
"U2DREL" is a utility module which reads two-dimensional real arrays.
Figure 15.-Primary modules organized by procedure and package.
Table 1 lists the various packages documented in this publication, gives the three-
character abbreviation used in the module designation scheme, and provides a brief
description of the package operation. Two major categories of package may be
recognized--the flow component packages and the solver packages; within the
category of flow component packages, a stress package subcategory may be
recognized. The flow component packages are those which calculate the coefficients
of the finite-difference equation for each cell. This category includes the Block-
Centered Flow Package, which formulates the internal flow terms (describing flow
between cells and flow to or from storage); and the subcategory of stress packages.
Each of the stress packages formulates the coefficients describing a particular
external or boundary flow; for example, the River Package calculates the coefficients
describing flow between a cell and a surface stream. The solver packages are those
which implement algorithms for solution of the systems of finitedifference equations.
This documentation describes two packages in this category, one incorporating the
Strongly Implicit Procedure of solution, and the other utilizing Slice-Successive
Overrelaxation. The only package which does not fit into any of these categories is
the Basic Package, which addresses a variety of tasks in support of the entire
simulation.
The Block-Centered Flow Package is the only option described in this documentation
for the formulation of internal flow terms in the equations. However, alternative
packages, for example, utilizing a point centered approach, could certainly be
developed and used in place of the BlockCentered Flow Package.
Table 1.--List of packages.
Package Name Abbreviation Package Description
Basic BAS Handles those tasks that are part of the model as a whole. Among
those tasks are specification of boundaries, determination of time-
step length, establishment of initial conditions, and printing of
results.
Block-Centered Flow BCF Calculates terms of finite difference equations which represent flow
with-in porous medium; specifically, flow from cell to cell and
flow into storage. (NOTE: There are now 3 versions of the BCF
package; the two new versions are BCF2 and BCF3. BCF2 allows
dry cells to resaturate and BCF3 provides additional inter-cell
conductance calculations for water-table aquifers. These are
documented in separate manuals)
Stress Component Packages
Well WEL Adds terms representing flow to wells to the finite-difference
equations.
Recharge RCH Adds terms representing areally distributed recharge to the finite-
difference equations.
River RIV Adds terms representing flow to rivers to the finite-difference
equations.
Drain DRN Adds terms representing flow to drains to the finite-difference
equations.
Evapotranspiration EVT Adds terms representing ET to the finite-difference equations.
General-Head Bnd. GHB Adds terms representing general-head boundaries to the finite-
difference equations.
Solver Packages
Strongly Implicit Proc. SIP Iteratively solves the system of finite-difference equations using the
Strongly Implicit Procedure.
SOR Solver SOR Iteratively solves the system of finite-difference equations using
Slice-Successive Overrelaxation.
Similarly, additional solver packages, incorporating different solution algorithms,
could be added, as could additional stress packages. Every simulation must include
the Basic Package, the Block-Centered Flow Package (or a suitable replacement) and
a solver package. Beyond this, the packages to be included in a simulation are at the
option of the user, and will depend on the hydrologic processes influencing the
problem. The individual modules in the program have been designed in such a way
that the packages are totally independent; with the exception of the three required
packages noted above, addition or removal of an individual package has no effect on
other packages. If an entirely new package is desired, modules can be developed for
each of the procedures involved (and the main program modified to call those
modules in proper sequence) without affecting other packages of the program.
Figure 16 shows a detailed flow chart of the main program, indicating all of the
primary modules together with the tests which determine whether or not each module
is to be called. Figure 16 may be studied in conjunction with figures 13 and 15, and
table 1, for an appreciation of the overall structure and operation of the model.
The overall design of the model is such that the conductance terms for cell-to-cell
flow (CC, CR, and CV of equation (26)) are formulated at the beginning of the
simulation, and are reformulated if necessary at each iteration during solution.
Reformulation takes place only in unconfined situations, where the conductances
depend upon saturated thickness, which may change at each iteration. At present,
formulation of conductances is done only by the Block-Centered Flow Package,
although again, a replacement package could readily be developed. The lateral
conductance terms (CC and CR in equation (26)) are computed as harmonic means
for the intervals between nodes, using parameters specified by the user for individual
cells; the vertical conductance (CV in equation (26)) is calculated using information
which is specified directly for the vertical interval between nodes. The various
conductance terms are stored in arrays which are ultimately passed to the solver
package, where the matrix equations (equation (27)) are solved.
Figure 16.-Overall program structure showing all primary modules.
The coefficient HCOFi,j.k and the term RHSij,k of equation (26) are formulated
anew at each iteration, for all active nodes in the mesh. This formulation is done
progressively, as each package calculates and adds terms for the particular process
associated with that package. At the beginning of each iteration, the values of
HCOFi,j,k and RHSi,j,k are set to zero throughout the mesh. The Block-Centered
Flow Package then adds the term -SS
i,j,k
~r
j
~c
i
~v
k
/(t
m
- t
m-1
) to HCOF
i,j,k
at each node,
and adds the term
-
SS r c v
t
-
t
h
i, j,k j i k
m m-1
i, j,k
m-1

to RHS
i,j,k
at each node. For cells that are affected by flow from a stream, given by
an expression of the form P
si,j,k
(h
s
-h
i,j,k
), where h
s
is the (constant) stream head, the
River package adds the term -P
si,j,k
to HCOF
i,j,k
and adds the constant term -P
si,j,k
h
s
to
RHS
i,j,k
. This process continues until each package specified by the user has added
its contribution to HCOF and RHS at each indicated node of the mesh. The HCOF
and RHS arrays are then transferred to the solver package, together with the three
conductance arrays (CC,CR and CV), an array containing heads at the beginning of
the time step, and the IBOUND array, which identifies constand head, no flow and
active nodes. The solver package sums the six conductance terms and the value of
HCOF at each node to create a single coefficient of h
i,j,k
(corresponding to the term in
brackets in equation (26)), and carries out one iteration of the solution procedure.
The various arrays used in the solution procedure are actually stored as segments of a
single one-dimensional array, the "X" array.
As noted in Chapter 1, Chapters 4 through 13 of this document discuss the program
in terms of individual packages. Each of these chapters contains a detailed
description of a particular package, including a listing and discussion of each module
included in the package. The remainder of this chapter describes the way boundaries,
water budget calculations, space allocation and input-output are handled in the
model, and provides a brief description and listing of the main program.
Array Boundaries and Aquifer Boundaries
As noted in Chapter 2, the model may be visualized in terms of a threedimensional
assemblage of cells, each cell associated with a node of the model array. The size of
the model array is specified by the user in terms of the number of rows (NROW),
number of columns (NCOL) and number of layers (NLAY); these terms define a
three-dimensional array of cells in the form of a rectangular box. In formulating the
finite-difference equations, cell-tocell conductance terms are omitted for the exterior
of cells on the outer surface of this rectangular array. Thus considering flow along a
row, a cell-to-cell conductance term is developed for the interval between column 1
and column 2, but not for the interval to the opposite side of column 1; similarly, a
conductance term is developed for the interval between column (NCOL-1) and
column (NCOL), but not for the interval beyond column (NCOL). Similar
conventions are established in the other two directions, so that in effect the array is
bounded externally by planes across which no cell-tocell flow occurs. If these
boundaries of the model array, which are actually embedded in the program, coincide
with impermeable boundaries in the aquifer, they can be relied upon to simulate the
no-flow condition along those aquifer boundaries without further intervention by the
user. In general, however, the aquifer boundaries will be irregular in form, or will
not be of a simple impermeable character. In these cases, the aquifer boundary must
be simulated by specifying certain cells within the array as no-flow or constanthead,
by using external stress terms, or by using a combination of no-flow cells and
external stress terms. This was discussed in Chapter 2, and is further discussed
below. It should also be noted that while no cell-tocell conductance terms are
formulated for the interval above the uppermost layer of the model array, flow into
this layer from above is frequently represented in the model through external stress
terms--for example, terms representing evapotranspiration or stream seepage.
A finite-difference equation of the form of (26) is formulated for each variable-head
cell in the mesh. For constant-head cells, no equation is formulated; however, the
equation for each variable-head cell adjacent to a constant-head cell contains a term
describing flow to and from the constant-head cell. For inactive no-flow cells, no
equation is formulated, and no term appears in the equation of any adjacent cell for
flow to or from the inactive cell; thus no flow is simulated across the interval between
an inactive cell and any adjacent cell.
As pointed out above, the model array as initially generated always has the form of a
rectangular box. Where the limits of an aquifer do not coincide with this rectangular
shape, inactive cells may be used to delete portions of the array which fall outside the
aquifer boundaries; this was discussed through an example in Chapter 2. As noted in
the same example, constant-head cells may be used to represent such features as
surface water bodies of constant level which are in full contact with the aquifer.
Boundaries which are characterized by a constant rate of flow into or out of the
aquifer may be simulated using a no-flow boundary in conjunction with the Well
Package, by assigning appropriate withdrawal or recharge rates to nodes just inside
the boundary. Boundaries characterized by inflow which varies in proportion to head
can be simulated using the General Head Boundary Package or the River Package,
where these again are applied to nodes just interior to a no-flow boundary. Use of
the River Package would involve specifying artificial streambed conductance and
stream-head values at each cell along the boundary, where these values are
deliberately chosen in such a way as to duplicate the required head-flow
relationships.
Constant-head cells, inactive cells and variable-head cells are distinguished from one
another in the model through the IBOUND array, which contains one element for
each cell in the mesh. The entry in the IBOUND array for a given cell indicates the
type of cell according to the following convention:
IBOUND (I,J,K) < 0..Cell I,J,K is constant head
IBOUND (I,J,K) = 0..Cell I,J,K is inactive
IBOUND (I,J,K) > 0..Cell I,J,K is variable head
The IBOUND codes are initially specified by the user. If necessary, the codes are
adjusted so that they are consistent with other data specified by the user and with
intermediate results. For example, cells which are specified as active but are given
transmissivity and vertical-leakance values equal to zero are changed to inactive cells
by the program.
Volumetric Budget
A summary of all inflows and outflows to a region is generally called a water budget.
In this report, the water budget is termed a volumetric budget because it deals with
volumes of water and volumetric flow rates; thus strictly speaking it is not a mass
balance, although this term has been used in reference to volumetric budgets in other
model reports. The model program calculates a water budget for the overall model as
a check on the acceptability of the solution, and in order to provide summarized
information on the flow system.
Numerical solution techniques for simultaneous equations do not always result in a
correct answer; in particular, iterative solvers may stop iterating before a sufficiently
close approximation to the solution is attained. A water budget provides an
indication of the overall acceptability of the solution. The system of equations solved
by the model actually consists of a flow continuity statement for each model cell.
Continuity should also exist for the total flows into and out of the model--that is, the
difference between total inflow and total outflow should equal the total change in
storage. In the model program, the water budget is calculated independently of the
equation solution process, and in this sense may provide independent evidence of a
valid solution.
Each flow component package calculates its own contribution to the budget. The
total budget as printed in the output does not include internal flows between model
cells--only flows into or out of the model as a whole. For example, flow to or from
rivers, flow to or from constant head cells, and flow to wells are all included in the
overall budget terms. Flow into and out of storage is also considered part of the
overall budget inasmuch as accumulation in storage effectively removes water from
the flow system, and storage release effectively adds water to the flow--even though
neither process, in itself, involves the transfer of water into or out of the ground water
regime.
For every time step, the budget module of each flow component package calculates
the rate of flow into and out of the system due to the process simulated by the
package. The inflows and outflows for each component of flow are stored separately
in the VBVL array. Most packages deal with only one such component of flow, but
the Block-Centered Flow Package deals with two--flow to constant head cells and
flow to storage. In addition to flow, the volumes of water entering and leaving the
model during the time step are calculated as the product of flow rate and time step
length. Cumulative volumes, from the beginning of the simulation, are then
calculated and stored in array VBVL.
Module SBAS1V in the BAS Package uses the inflows, outflows and cumulative
volumes in the VBVL array to print the budget at the times requested by the model
user. When a budget is printed, the flow rates for the last time step and cumulative
volumes from the beginning of simulation are printed for each component of flow.
Inflows are printed separately from ouflows; following the convention indicated
above, water entering storage is treated as an outflow while water released form
storage is treated as an inflow. In addition, total inflow and total outflow are printed,
as well as the difference between total inflow and outflow. The difference is then
printed as a percent error, calculated using the formula:
D=
100(INFL - OUT)
(INFL+OUT)0.5
where INFL is the total inflow to the system, OUT is the total outflow and D is the
percent error term. If the model equations are correctly solved, the percent error
should be small. In general, flow rates may be taken as an indication of solution
validity for the time step to which they apply, while cumulative volumes are an
indication of validity for the entire simulation up to the time of the printout. The
budget is printed at the end of each stress period whether requested or not.
There are situations in which it is useful to calculate flow terms for various
subregions of the model. To facilitate such calculations, provision has been made to
save flow terms for individual cells on disk so they can be used in computations
external to the model itself. These individual cell flows are referred to here as "cell-
by-cell" flow terms, and are of four general types: (1) cell-by-cell stress flows, or
flows into or from an individual cell due to one of the external stresses represented in
the model, such as evapotranspiration or recharge; (2) cell-by-cell storage terms,
which give the rate of accummulation or depletion of storage in an individual cell; (3)
cell-by-cell constant-head flow terms, which give the net flow to or from individual
constant-head cells; and (4) internal cell-bycell flows, which are actually the flows
across individual cell faces--that is, between adjacent model cells. These four kinds
of cell-by-cell term are further discussed in subsequent paragraphs. To save any of
these cell-by-cell terms, two flags in the model input must be set. The input to the
Output Control section of the Basic Package includes a flag, ICBCFL, which must be
set for each time step for which any cell-by-cell terms are to be saved. In addition,
each flow component package includes a flag which is set if the cell-by-cell terms
computed by that package are to be saved. Thus if the appropriate flag in the
Evapotranspiration Package input is set, cell-by-cell evapotranspiration terms will be
saved for each time step for which the ICBCFL flag in the Basic Package input is
also set. Three of the four types of cell-by-cell flow terms listed above--storage,
constant-head cell and internal flows--are computed in the Block-Centered Flow
Package, and thus fall under the control of a single flag, IBCFCB, in the input to that
package. Thus in general all three types are saved on disk if this flag is set, and
ICBCFL is also set for the time step. Only flow values are saved in the cell-by-cell
disk files; neither water volumes nor cumulative water volumes are included. The
flow dimensions are volume per unit time, where volume and time are in the same
units used for all model input data. The cell-by-cell flow values are stored in
unformatted form to make the most efficient use of disk space; see the narrative for
the UBUDSV module for information on how the data are written to disk.
Cell-by-cell stress flows are flow rates into or out of the model, at a particular cell,
due to one particular external stress. For example, the cell-by-cell evapotranspiration
term for cell i,j,k would give the flow out of the model by evapotranspiration from
cell i,j,k. Cell-by-cell stress flows are considered positive if flow is into the cell, and
negative if it is out of the cell. A cell-by-cell stress flow value is saved for every
model cell, for each stress component for which the cell-by-cell flow is requested.
That is, an array the size of the model grid is saved on disk for each requested
component of flow. For many of the stress components, flow will be zero at most
model cells. For example, when using the River Package, there will be nonzero cell-
by-cell budget values only at those cells that are traversed by rivers. Thus the
amount of disk space required for cell-by-cell flow terms can be large; a flow value is
stored for each model cell even when that value is zero, and terms may be saved at
many time steps.
The cell-by-cell storage term gives the net flow to or from storage in a variable-head
cell. An array of these terms, one for each cell in the mesh is saved in transient
simulations if the appropriate flags are set. Withdrawal from storage in the cell is
considered positive, whereas accumulation in storage is considered negative.
The cell-by-cell constant-head flow term gives the flow into or out of an individual
constant-head cell. This term is always associated with the constant-head cell itself,
rather than with the surrounding cells which contribute or receive the flow. A
constant-head cell may be surrounded by as many as six adjacent variable-head cells.
The cell-by-cell calculation provides a single flow value for each constant-head cell,
representing the algebraic sum of the flows between that cell and all of the adjacent
variable-head cells. A positive value indicates that the net flow is away from the
constant-head cell (into the variable-head portion of the mesh); a negative value
indicates that the net flow is into the constant-head cell.
The internal cell-by-cell flow values represent flows across the individual faces of a
model cell. Three such terms are saved by the Block-Centered Flow Package for
each variable-head cell and constant-head cell in the mesh, whenever the appropriate
cell-by-cell flags are set. These three terms are flow across the front cell face
(between cell i,j,k and i+l,j,k), flow across the right face (between cell i,j,k and
i,j+l,k), and flow across the lower face (between cell i,j,k and i,j,k+l). Each of these
represents flow between a given cell and a neighboring cell. (Although each cell has
six neighbors, only three flow terms are required; flow across the other three sides is
accounted for in the calculations of flow for cells adjacent to those sides.) Flows are
considered positive if they are in the direction of increasing row number, increasing
column number or increasing layer number, and are considered negative if in the
opposite directions. These internal cell-by-cell flow values are useful in calculations
of the ground-water flow into various subregions of the model, or in constructing
flow vectors.
In theory one could calculate a budget identical to the overall budget by using the
cell-by-cell flow terms. This is not always true in practice because in some situations
the budgets may be summed differently. The cell-by-cell value at a cell for a given
stress or flow component is the net flow for that component, which could possibly
include two or more flows of the same type, some negative and some positive. Only
the net flow for the cell is saved in the cell-by-cell disk file. In the overall budget
calculations as performed in the model, on the other hand, positive and negative
flows are assembled separately, so that a negative flow at an individual cell would be
added to the outflow term and a positive flow at the same cell would be added to the
inflow term. Thus if inflow and outflow terms for the entire model are calculated by
summing individual cell-by-cell values, they may differ from the corresponding terms
as calculated by the model program in the overall budget. However, the difference
between inflow and outflow should be the same for either calculation.
Space Allocation
Space in the central memory of the computer used by data arrays and lists is allocated
at execution time in a one-dimensional array called the "X" array. The Allocate
Procedure contains a module for each package of the model which allocates space
needed by that package. The total number of words needed in the X array depends
on the type and number of packages required in a simulation and generally will range
from 10 to 20 times the number of cells in the grid. The main program contains two
statements referring to the length of the X array, both of which appear in the first part
of the program listing. In the listing reproduced with this documentation these
statements are COMMON X(30000) and LENX = 30000. The number 30000 in the
statements refers to the length of the X array; this number must be increased if the
storage requirements of the problem exceed 30000 elements.
NOTE: MODFLOW
win32
dynamically allocates the memory (space) needed for
each problem. Thus, there is no need to make changes to the X array or be
concerned about the dimensions of the program. If there is not enough memory
available, the program will display an error message when you run the model.
Three-Dimensional Subscripts for Model Arrays
The conceptualization and implementation sections of this report designate cell
locations by row, column, and layer indices in that order (usually designated as i,j,k),
as is customary in scientific literature; however, this order of indices is not the most
efficient order for array subscripts in the model program. Many model parameters
are declared to be three dimensional arrays and accordingly have row, column and
layer subscripts. The order of array subscripts in the FORTRAN language
determines how data are stored in computer memory. The design of the program is
such that the model array subscripts should be in column, row, and layer order for the
most efficient memory access on most computers; this order has been used
throughout the program. Typically in the program, J is used for the column subscript,
I is used for row, and K for layer, but the order is J,I,K rather than I,J,K. It is
important to bear in mind this difference in the subscript ordering when comparing
the model program to the conceptualization and implementation sections of the
report.
Input Structure
The input structure of the program is designed to permit input to be gathered, as it is
needed, from many different files. It is based on an element of the FORTRAN
language called the unit number, which identifies the file from which the input is to
be read (or to which the output is to be written). The user must provide a link
between the name of each input or output file and the corresponding unit number; this
is generally done externally to the program, through operating system statements.
For input purposes, the program may be discussed in terms of "major options"; these
are major segments of the program which are utilized only at the user's request. They
correspond generally to the individual packages; in fact, all of the existing packages
except the Basic Package are considered major options. Output Control, which is not
an individual package but rather an optional segment of the Basic Package providing
flexibility in program output, is also considered a major option. The balance of the
Basic Package is not considered an option since it is always utilized and input for it
must always be read. Block-Centered Flow has been treated here as an option, even
though it is presently required in all simulations. This has been done to allow for the
addition of replacement packages for Block-Centered Flow in the future.
One of the first steps in organizing input data is to specify which of the major options
are to be used. This is done using the "IUNIT" array (figure 17) which is read in the
Define Procedure by the Basic Package. An option is invoked by inserting an input
unit number in the appropriate element of the IUNIT array; if an option is not
desired, the value of the element is set to zero. Thus the IUNIT array serves as a flag
to indicate whether an option is active, and also serves to specify the unit number
containing input data required by the option. For example, if the Drain Package is
not to be used, the third element of the IUNIT array (figure 18) is set to zero; if it is
to be used, the third element of the array is set to the unit number of the file
containing the input data for the package. In the main program, the value of IUNIT
(3) is tested in several of the program procedures. If it is zero, the Drain module
associated with the procedure is not called. If IUNIT (3) is greater than zero, the
subroutine is called and input data is read from the file associated with the unit
number.
As noted above, the Basic (BAS) Package, exclusive of the Output Control option, is
used for every simulation; and input data for the Basic Package are always required.
Basic Package data (figure 18) are read from unit number 1 as specified in the main
program. If necessary, the unit number for BAS input can be changed to meet the
requirements of a particular computer.
The first element of the IUNIT array must contain the unit number from which data
for the Block-Centered Flow (BCF) Package are to be read. At present, because
BCF is the only package available for the formulation of cell-to-cell flow terms, a
non-zero entry in the first element of IUNIT is always required.
Figure 17.-Specification of major options using the IUNIT array.
Most of the data submitted by the user will consist of one-dimensional and two-
dimensional arrays. Those arrays are submitted as an "array control record" plus,
optionally, a series of records containing the array elements. The array control
record is read from the unit number specified for the major option which calls for the
array. If all the elements of an array have the same value, the value is specified on
the control record and it is not necessary to read the associated array. If the elements
of the array vary, records containing the array values are read from the unit specified
on the array control record according to a format which is also specified in the
control record. The unit number may be the same as that from which the control
record is read, or it may be different. Thus there is a great deal of flexibility
regarding the organization of the input data for a simulation.
Any consistent length and time units may be used for model data. This gives a
certain amount of freedom to the user, but care must be exercised to avoid any
mixing of units. There is no way for the program to detect the use of inconsistent
units. For example, if transmissivity is entered in units of ft
2
/day and pumpage as
m
3
/s, the program will run, but the results will be meaningless.
Figure 18.-Sample input data showing role of the IUNIT array.
Output Structure
The output structure is designed to control the amount, type, and frequency of
information to be printed or written on disk. It controls the printing of head and
drawdown by layer and time step, and the printing of the overall volumetric budget.
It also controls disk output of head, drawdown, and cell-by-cell flow terms for use in
calculations external to the model, or in user-supplied printing and plotting programs.
Output Control, which is a major option contained within the Basic Package, receives
instructions from the user to control the amount and frequency of output. To utilize
this option, the user must specify the unit number of the file or channel from which
the input data for the Output Control option are to be read. This unit number must be
entered as the twelfth element of the IUNIT array (IUNIT 12); the input information
is then read, at each time step, from the file identified by this unit number. If a zero
is specified as the twelfth element of the IUNIT array, a default output convention is
invoked. This default output consists of head values and budget terms printed for the
end of each stress period. Every simulation generates some printer output. All
printer output goes to unit number 6 as specified in the main program. This unit
number can be changed to meet the requirements of a particular computer.
Electronic Manual for MODFLOW 4 Basic Package 51
4 Basic Package
Conceptualization and Implementation
The Basic Package handles a number of administrative tasks for the model. It reads
data on the number of rows, columns, layers, and stress periods, on the major options
to be used, and on the location of input data for those options. It allocates space in
computer memory for model arrays; it reads data specifying initial and boundary
conditions; it reads and implements data establishing the discretization of time; it sets
up the starting head arrays for each time step; it calculates an overall water budget;
and it controls model output according to user specification.
Selection of Major Options and Designation of
Input Files
The selection of major options and the designation of their input unit numbers were
discussed in the preceding chapter. The primary role of the Basic Package in these
operations is to read the IUNIT array; as noted in Chapter 3, the entries in this array
determine (a) whether or not a major option is to be used and (b) the unit number
from which data for the option is to be read. Whenever a new major option is added
to the program, an element corresponding to that option must be added to the IUNIT
array.
The IBOUND Array
Recall that the finite-difference equation for a cell has the form
i, j-1/ 2,k
i, j -1,k
m
i, j,k
m
i, j+1/ 2,k
i, j+1,k
m
i, j,k
m
CR
(
h
-
h
)+
CR
(
h
-
h
)
+
CC
(
h
-
h
)+
CC
(
h
-
h
)
i -1/ 2, j,k
i -1, j,k
m
i, j,k
m
i+1/ 2, j,k
i+1, j,k
m
i, j,k
m
+
CV
(
h
-
h
)+
CV
(
h
-
h
)
i, j,k-1/ 2
i, j,k-1
m
i, j,k
m
i, j,k+1/ 2
i, j,k+1
m
i, j,k
m
52 4 Basic Package Electronic Manual for MODFLOW
+
P
h
+Q =
SC
1
(
h
-
h
)
t
i, j,k
i, j,k
m
i, j,k
i, j,k
i, j,k
m
i, j,k
m-1
m
(28)
One equation of this form is written for each variable-head cell in the grid. The
IBOUND array, which is specified by the user and read by the Basic Package,
contains a code for each cell which indicates whether (1) the head varies with time
(variable-head cell), (2) the head is constant (constant-head cell), or (3) no flow takes
place within the cell (no-flow or inactive cell). The IBOUND array can be modified
by other packages if the state of a cell changes. Figure 19 illustrates the distribution
of IBOUND code entries for a typical model layer.
Initial Conditions
Because equation (28) is in backward-difference form, a head distribution at the
beginning of each time step is required to calculate the head distribution at the end of
that time step (figure 20). For each time step after the first, the head distribution at
the start of one time step is set equal to the head distribution at the end of the
previous time step. For the first time step, "starting heads" are specified by the user.
These specified initial heads are used for head calculation only in the first time step;
however, they may also be saved, in the array STRT, and used to calculate
drawdown, the difference between the starting head distribution and some later head
distribution.
Figure 19.-Example of the boundary array (IBOUND) for a single layer.
Figure 20.-Flow of head distributions during a simulation.
Discretization of Time
Simulation time is divided into stress periods--time intervals during which all
external stresses are constant--which are, in turn, divided into time steps as shown in
figure 21. Within each stress period, the time steps form a geometric progression.
The user specifies the length of the stress period, the number of time steps into which
it is to be divided, and the time step multiplier, or ratio of the length of each time step
to that of the preceding time step. Using these terms, the program calculates the
length of each time step in the stress period.
Output
The primary output of the program is head distribution. The user may control the
frequency at which heads are printed or saved on disk through the "Output Control"
option, a major option contained in the Basic Package. Other output items include
drawdowns and volumetric budget terms; the Output Control option also provides for
storage or printing of these terms. If Output Control is not utilized, a default output
option is invoked--the head distribution and the overall volumetric budget are printed
at the end of each stress period, and drawdowns are also printed if starting heads
were saved. Figure 22 shows an example of a volumetric budget printout for the end
of a stress period.
Figure 21.-Division of simulation time into stress periods and time steps.
5
6

4

B
a
s
i
c

P
a
c
k
a
g
e
E
l
e
c
t
r
o
n
i
c

M
a
n
u
a
l

f
o
r

M
O
D
F
L
O
W
V
O
L
U
M
E
T
R
I
C

B
U
D
G
E
T

F
O
R

E
N
T
I
R
E

M
O
D
E
L

A
T

E
N
D

O
F

T
I
M
E

S
T
E
P

1

I
N

S
T
R
E
S
S

P
E
R
I
O
D

1
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
C
U
M
U
L
A
T
I
V
E

V
O
L
U
M
E
S
L
*
*
3

R
A
T
E
S

F
O
R

T
H
I
S

T
I
M
E

S
T
E
P

L
*
*
3
/
T
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
I
N
:
I
N
:
S
T
O
R
A
G
E

=
.
0
S
T
O
R
A
G
E

=
.
0
C
O
N
S
T
A
N
T

H
E
A
D

=
.
0
C
O
N
S
T
A
N
T

H
E
A
D

=
.
0
W
E
L
L
S

=
.
0
W
E
L
L
S

=
.
0
D
R
A
I
N
S

=
.
0
D
R
A
I
N
S

=
.
0
R
E
C
H
A
R
G
E
=
.
1
3
6
0
8
E
+
0
8
R
E
C
H
A
R
G
E
=
1
5
7
.
5
0
T
O
T
A
L

I
N
=
.
1
3
6
0
8
E
+
0
8
T
O
T
A
L

I
N
=
1
5
7
.
5
0
O
U
T
:
O
U
T
:
S
T
O
R
A
G
E
=
.
0
S
T
O
R
A
G
E
=
.
0
C
O
N
S
T
A
N
T

H
E
A
D
=
.
4
3
2
6
5
E
+
0
7
C
O
N
S
T
A
N
T

H
E
A
D
=
5
0
.
0
7
5
W
E
L
L
S
=
.
6
4
8
0
0
E
+
0
7
W
E
L
L
S
=
7
5
.
0
0
0
D
R
A
I
N
S
=
.
2
8
0
1
0
E
+
0
7
D
R
A
I
N
S
=
3
2
.
4
1
9
R
E
C
H
A
R
G
E
=
.
0
R
E
C
H
A
R
G
E
=
.
0
T
O
T
A
L

O
U
T
=
.
1
3
6
0
7
E
+
0
8
T
O
T
A
L

O
U
T
=
1
5
7
.
4
9
I
N

-

O
U
T
=
3
0
3
.
0
0
I
N

-

O
U
T
=
.
3
4
9
4
3
E
-
0
2
P
E
R
C
E
N
T

D
I
S
C
R
E
P
A
N
C
Y

=

0
.
0
0

P
E
R
C
E
N
T

D
I
S
C
R
E
P
A
N
C
Y

=

0
.
0
0
F
i
g
u
r
e

2
2
.
-
-
S
a
m
p
l
e

o
v
e
r
a
l
l

v
o
l
u
m
e
t
r
i
c

w
a
t
e
r

b
u
d
g
e
t
.
Budget Calculations in the Basic Package
The calculation of the volumetric budget is carried out in two parts, the calculation of
budget entries and the summation of those entries. As explained in Chapter 3 the
entries, which correspond to individual components of flow, are calculated in the
flow component packages and stored in the I one-dimenisonal array VBVL. The
array VBVL is passed to the Basic Package which sums and prints the budget entries.
Basic Package Input
Input for the Basic (BAS) Package except for output control is read from unit 1 as
specified in the main program. If necessary, the unit number for BAS input can be
changed to meet the requirements of a particular computer. Input for the output
control option is read from the unit number specified in IUNIT(12).
Information for the Basic Package must be submitted in the following order:
FOR EACH SIMULATION
BAS1DF
1. Data: HEADNG(32)
Format: 2OA4
2. Data: HEADNG (continued)
Format: 12A4
3. Data: NLAY NROW NCOL NPER ITMUNI
Format: I10 I10 I10 I10 I10
4. Data: IUNIT(24)
Format: 24I3
(BCF WEL DRN RIV EVT XXX GHB RCH SIP XXX SOR OC)
BAS1AL
5. Data: IAPART ISTRT
Format: I10 I10
BAS1RP
6. Data: IBOUND(NCOL,NROW)
Module: U2DINT (One array for each layer in the grid)
7. Data: HNOFLO
Format: F10.0
8. Data: Shead(NCOL,NROW)
Module: U2DREL (One array for each layer in the grid)
NOTE: IBOUND and Shead are treated as three-dimensional arrays in the
program. However, the input to each of these arrays is handled as a series of two-
dimensional arrays, one for each layer in the grid.
FOR EACH STRESS PERIOD
BAS1ST
9. Data: PERLEN NSTP TSMULT
Format: F10.0 I10 F10.0
Explanation of Fields
HEADNG--is the simulation title that is printed on the printout. It may be up to 132
characters long; 80 in the first record and 52 in the second. Both records must be
included even if they are blank.
NLAY--is the number of model layers.
NROW--is the number of model rows.
NCOL--is the number of model columns.
NPER--is the number of stress periods in the simulation.
ITMUNI--indicates the time unit of model data. (It is used only for printout of
elapsed simulation time. It does not affect model calculations.)
0 - undefined 3 - hours
1 - seconds 4 - days
2 - minutes 5 - years
The unit of time must be consistent for all data values that involve time. For
example, if years is the chosen time unit, stress-period length, timestep length,
transmissivity, etc., must all be expressed using years for their time units. Likewise,
the length unit must also be consistent.
IUNIT is a 24-element table of input units for use by all major options. Only 10
elements (1-5, 7-9, 11, and 12) are being used. Element 6 has been reserved for a
transient leakage package, while element 10 has been reserved for an additional
solver, both on the assumption that such packages will be added to the model in the
future. Elements 13-24 are reserved for future major options.
IUNIT MAJOR
LOCATION OPTION
1 Block-Centered Flow Package
2 Well Package
3 Drain Package
4 River Package
5 Evapotranspiration Package
6 Reserved for Transient Leakage Package
7 General-Head Boundary Package
8 Recharge Package
9 SIP Package
10 Reserved for additional solver
11 SSOR Package
12 Output Control Option
If IUNIT(n) 0,the corresponding major option is not being used.
If IUNIT(n) > 0,the corresponding major option is being used and data for
that option will be read from the unit number contained in IUNIT(n). The unit
numbers in IUNIT should be integers from 1 to 99. Although the same number may
be used for all or some of the major options, it is recommended that a different
number be used for each major option. Printer output is assigned to unit 6 (unless it
is changed to meet computer requirements). That unit number should not be used for
any other input or output. The user is also permitted to assign unit numbers for
output. Those numbers should be different from those assigned to input. The Basic
Package reads from unit 1 (unless it is changed to meet computer requirements). It is
permissible but unwise to use that unit for other major options.
IAPART--indicates whether array BUFF is separate from array RHS.
If IAPART =0, the arrays BUFF and RHS occupy the same space. This
option conserves space. This option should be used unless some other package
explicitly says otherwise.
If IAPART 0, the arrays BUFF and RHS occupy different space. This
option is not needed in the program as documented in this publication. It may be
needed for packages yet to be written.
ISTRT--indicates whether starting heads are to be saved. If they are saved, they will
be stored in array STRT. They must be saved if drawdown is calculated.
If ISTRT = 0, starting heads are not saved.
If ISTRT 0, starting heads are saved.
IBOUND--is the boundary array.
If IBOUND(I,J,K) < 0, cell I,J,K has a constant head.
If IBOUND(I,J,K) = 0, cell I,J,K is inactive (no-flow).
If IBOUND(I,J,K) > 0, cell I,J,K is variable-head.
HNOFLO is the value of head to be assigned to all inactive cells (IBOUND = 0)
throughout the simulation. Since heads at inactive cells are unused, this does not
affect model results but serves to identify inactive cells when head is printed. This
value is also used as drawdown at inactive cells if the drawdown option is used.
Even if the user does not anticipate having inactive cells, a value for HNOFLO must
be submitted.
Shead is head at the start of the simulation. Regardless of whether starting head is
saved, these values must be input to initialize the solution.
PERLEN is the length of a stress period. It is specified for each stress period.
NSTPis the number of time steps in a stress period.
TSMULT is the multiplier for the length of successive time steps. The length of the
first time step DELT(1) is related to PERLEN, NSTP and TSMULT by the relation
DELT(1) = PERLEN(1-TSMULT)/(1-TSMULT**NSTP).
Output Control Input
Output Control is a major option separate from the rest of the Basic Package. Input
to Output Control is read from the unit specified in IUNIT(12). If IUNIT(12) is zero,
no output control data are read, and default output control is used. Under the default,
head and total budget are printed at the end of every stress period. Additionally, if
starting heads are saved (ISTRT is not 0), drawdown is printed at the end of every
stress period. The default printout format for head and drawdown is 1OG11.4. All
printer output goes to unit 6 as specified in the main program. If necessary, the unit
number for printer output can be changed to meet the requirements of a particular
computer.
FOR EACH SIMULATION
BAS1RP
1. Data: IHEDFM IDDNFM IHEDUN IDDNUN
Format: I10 I10 I10 I10
FOR EACH TIME STEP
BAS10C
2. Data: INCODE IHDDFL IBUDFL ICBCFL
3. Data: Hdpr Ddpr Hdsv Ddsv
(Record 3 is read 0, 1, or NLAY times, depending on the value of
INCODE.)
Explanation of Fields
IHEDFM--is a code for the format in which heads will be printed.
IDDNFM--is a code for the format in which drawdowns will be printed. Format
codes have the same meaning for both head and drawdown. A positive format code
indicates that each row of data is printed completely before starting the next row.
This means that when there are more columns in a row than will fit on one line,
additional lines are used as required to complete-the row. This format is called the
wrap format. A negative format code indicates that the printout is broken into strips
where only that number of columns that will fit across one line are printed in a strip.
As many strips are used as are required to print the entire model width. This format
is called the strip format. The absolute value of the format code specifies the printout
format as follows.
0 (10G11.4) 7 (20F5.0)
1 (11G10.3) 8 (20F5.1)
2 (9G13.6) 9 (20F5.2)
3 (15F7.1) 10 (20F5.3)
4 (15F7.2) 11 (20F5.4)
5 (15F7.3) 12 (l0G11.4)
6 (15F7.4)
IHEDUN--is the unit number to which heads will be written if they are saved on
disk.
IDDNUN--is the unit number to which drawdowns will be written if they are saved
on disk.
INCODE--is the head/drawdown ouput code. It determines the number of records in
input item 3.
If INCODE < 0, layer-by-layer specifications from the last time steps are
used. Input item 3 is not read.
If INCODE = 0, all layers are treated the same way. Input item 3 will
consist of one record.
If INCODE > 0, input item 3 will consist of one record for each layer.
IHDDFL--is a head and drawdown output flag.
If IHDDFL= 0, neither heads nor drawdowns will be printed or saved on
disk.
If IHDDFL 0, heads and drawdowns will be printed or saved according to
the flags for each layer specified in input item 3.
IBUDFL--is a budget print flag.
If IBUDFL= 0, overall volumetric budget will not be printed.
If IBUDFL does not equal 0, overall volumetric budget will be printed.
(Note that the overall volumetric budget will always be printed at the end of
a stress period, even if the value of IBUDFL is zero.)
ICBCFL--is a cell-by-cell flow-term flag.
If ICBCFL = 0, cell-by-cell flow terms are not saved or printed.
if ICBCFL does not equal 0, cell-by-cell flow terms are printed or recorded
on disk depending on flags set in the component of flow packages, i.e., IWELCB,
IRCHCB, etc.
Hdpr--is the output flag for head printout.
If Hdpr = 0, head is not printed for the corresponding layer.
If Hdpr does not equal 0, head is printed for the corresponding layer.
Ddpr--is the output flag for drawdown printout.
If Ddpr = 0, drawdown is not printed for the corresponding layer.
If Ddpr does not equal 0, drawdown is printed for the corresponding layer.
Hdsv--is the output flag for head save.
If Hdsv = 0, head is not saved for the corresponding layer.
If Hdsv does not equal 0, head is saved for the corresponding layer.
Ddsv--is the output flag for drawdown save.
If Ddsv = 0, drawdown is not saved for the corresponding layer.
If Ddsv does not equal 0, drawdown is saved for the corresponding layer.
Electronic Manual for MODFLOW 5 Block-Centered Flow Package 65
5 Block-Centered Flow Package
The Block-Centered Flow (BCF) Package computes the conductance components of
the finite-difference equation which determine flow between adjacent cells. It also
computes the terms that determine the rate of movement of water to and from storage.
To make the required calculations, it is assumed that a node is located at the center of
each model cell; thus the name Block-Centered Flow is given to the package.
In Chapter 2, the equation of flow for each cell in the model was developed as
i, j,k-1/ 2 i, j,k -1 i-1/ 2, j,k i -1, j,k i, j -1/ 2,k i, j-1,k CV h
+
CC h
+
CR h
+(-
CV
-
CC
-
CR
-
CR i, j,k -1/ 2 i -1/ 2, j,k i, j -1/ 2,k i, j+1/ 2,k
-
CC
-
CV
+
HCOF
)
h
+
CR h i+1/ 2, j,k i, j,k+1/ 2 i, j,k i, j,k i, j+1/ 2,k i, j+1,k
+
CC h
+
CV h
=
RHS i+1/ 2, j,k i+1, j,k i, j,k+1/ 2 i, j,k+1 i, j,k
(29)
The CV, CR, and CC coefficients are conductances between nodes--sometimes called
"branch conductances." The HCOF and RHS coefficients are composed of external
source terms and storage terms. Besides calculating the conductances and storage
terms, the BCF Package calculates flow-correction terms that are added to HCOF and
66 5 Block-Centered Flow Package Electronic Manual for MODFLOW
RHS when an underlying aquifer becomes partially unsaturated. Under these
conditions the flow to the underlying aquifer no longer increases in proportion to the
head difference between aquifers, but rather reaches a constant limiting value. The
additional terms correct the flow equations, in effect reducing the expressions for
downward flow to correspond to this limiting value.
The following discussion of the conceptualization and implementation of the BCF
package is divided into nine sections: Basic Conductance Equations, Horizontal
Conductance Under Confined Conditions, Horizontal Conductance Under Water
Table Conditions, Vertical Conductance Formulation, Vertical Flow Calculation
Under Desaturating Conditions, Storage Formulation, Storage Term Conversion,
Applicability and Limitations of Optional Formulations and Data Requirements.
Basic Conductance Equations
The concept of hydraulic conductance was introduced in Chapter 2 (equation (9)). It
is reviewed here and extended to cover the calculation of equivalent conductance for
elements arranged in series.
Conductance is a combination of several parameters used in Darcy's law. Darcy's law
defines one-dimensional flow in a prism of porous material (figure 23) as
Q=KA(h
2
-h
1
)/L (30)
where
Q is the flow (L
3
t
-1
);
K is the hydraulic conductivity in the direction of flow (Lt
-1
);
A is the cross-sectional area perpendicular to the flow (L
2
);
h
2
-h
1
is the head differences across the prism parallel to flow (L); and
L is the length of the flow path (L).
Conductance, C, is defined as
C = KA/L. (31)
Explanation
K Is Hydraulic Conductivity
h
2
Is the Head at the Left End of the Prism
h
1
Is the Head at the Right End of the Prism
Q Is the Flow Rate from the Left End to the Right End
L Is the Length of the Flow Path
A Is the Cross Sectional Area Perpendicular to the Direction of Flow
Figure 23.-Prism of porous material illustrating Darcy's law.
Therefore, Darcy's law can be written as
Q = C(h
2
-h
1
). (32)
Another form of the conductance definition for horizontal flow in a prism is
C = TW/L (33)
where
T is transmissivity (K times thickness of the prism) in the direction of flow
(L2t-1); and
W is the width of the prism (L).
Conductance is defined for a particular prism of material and for a particular
direction. In an anisotropic medium characterized by three principal directions of
hydraulic conductivity, the conductances of a prism in these three principal directions
will generally differ.
If a prism of porous material consists of two or more subprisms in series--that is,
aligned sequentially in the direction of flow, as shown in figure 24--and the
conductance of each subprism is known, a conductance representing the entire prism
can be calculated. The equivalent conductance for the entire prism is the rate of flow
in the prism divided by the head change across the prism.
C = Q/(h
A
-h
B
) (34)
Assuming continuity ofhead across each section in series gives the identity
i=1
n
i A B h
=
h
-
h
(35)
Substituting for head change across each section using Darcy's law (equation (32))
gives
i=1
n
i
i
A B
q
C
=
h
-
h
(36)
Figure 24.-Calculation of conductance through several prisms in series.
Since flow is one-dimensional and we are assuming no accumulation or depletion in
storage, all qi are equal to the total flow Q; therefore,
Q
1
C
=
h
-
h
i=1
n
i
A B
A B
i=1
n
i
h
-
h
Q
=
1
C
(37)
By comparison with equation (34), it can be seen that
1
C
=
1
C
i=1
n
i
(38)
Thus for a set of conductances arranged in series, the inverse of the equivalent
conductance equals the sum of the inverses of the individual conductances. When
there are only two sections, the equivalent conductance reduces to
C=
C C
C
+
C
1 2
1 2
(39)
(39)
Horizontal Conductance Under Confined
Conditions
The finite-difference equations presented in this report use equivalent conductances
between nodes of adjacent cells--i.e., "branch conductances,"--rather than
conductances defined within individual cells. The horizontal conductance terms, CR
and CC of equation (29), are calculated between adjacent horizontal nodes. CR
terms are oriented along rows and thus specify conductance between two nodes in the
same row. Similarly, CC terms specify conductance between two nodes in the same
column. To designate conductance between nodes, as opposed to conductance within
a cell, the subscript notation "1/2" is used. For example, CR
i,j+1/2,k
represents the
conductance between nodes i,j,k and i,j+1,k.
Figure 25 illustrates two cells along a row, and the parameters used to calculate the
conductance between nodes. Two assumptions are made: (1) the nodes are in the
center of the cells and (2) the transmissivity is uniform over each cell. Thus the
conductance between the nodes is the equivalent conductance of two half cells in
series (C1 and C2). Applying equation (39) gives
i, j+1/ 2,k
1 2
1 2
CR
=
C C
C
+
C
(40)
Figure 25.-Calculation of conductance between nodes using transmissivity and dimensions of cells.
Substituting the conductance for each half cell from equation (33) gives
i, j+1/ 2,k
i, j,k i
j
i, j+1,k i
j+1
i, j,k i
j
i, j+1,k i
j+1
CR
=
TR DELC
1 / 2
DELR
TR DELC
1 / 2
DELR
TR DELC
1 / 2
DELR
+
TR DELC
1 / 2
DELR
where
TR is transmissivity in the row direction (L
2
t
-1
);
DELR is the grid width along a row (L); and
DELC is the grid width along a column (L).
DELR and DELC are identical to the terms ~r and ~c, respectively, which
were introduced in figure 4 and equation (3), Chapter 2. The new notation is
introduced here to conform to the input of the Block-Centered Flow Package.
Simplification of the above expression gives the final equation
i, j+1/ 2,k i
i, j,k i, j+1,k
i, j,k j+1 i, j+1,k j
CR
= 2
DELC
TR TR
TR DELR
+
TR DELR
(41)
The same process can be applied to the calculation- of CC
i+1/2,j,k
giving
i+1/ 2, j,k j
i, j,k i+1, j,k
i, j,k i+1 i+1, j,k i
CC
= 2
DELR
TC TC
TC DELC
+
TC DELC
(42)
where
TC is the transmissivity in the column direction (L
2
t
-1
). Equations (41) and (42) are
used in the BCF Package to calculate the horizontal conductances between nodes
within each layer of the model. However, where the transmissivity of both cells is
zero, the conductance between the nodes in the cells is set equal to zero without
invoking the equations.
Horizontal Conductance Under Water Table
Conditions
In a model layer which is confined, horizontal conductance will be constant for the
simulation. If a layer is unconfined or potentially unconfined, new values of
horizontal conductance must be calculated as the head fluctuates. This is done at the
start of each iteration. First, transmissivity is calculated as the product of hydraulic
conductivity and saturated thickness; then conductance is calculated from
transmissivity and cell dimensions using equations (41) and (42).
Transmissivity within a cell in the row direction is calculated using one of the
following three equations
if HNEW
i,j,k
> TOP
i,j,k
,
then TR
i,j,k
= (TOP
i,j,k
- BOT
i,j,k
) HYR
i,j,k
; (43)
if TOP
i,j,k
> HNEW
i,j,k
> BOT
i,j,k
,
then TR
i,j,k
= (HNEW
i,j,k
- BOT
i,j,k
) HYR
ij,k
; (44)
if HNEW
i,j,k
< BOT
i,j,k
,
then TR
i,j,k
= 0 (45)
where
HYR
i,j,k
is the hydraulic conductivity of cell i,j,k in the row direction (Lt
-1
);
(this notation is introduced here to conform to the input of the Block-Centered Flow
Package);
TOP
i,j,k
is the elevation of the top of cell i,j,k (L); and
BOT
i,j,k
is the elevation of the bottom of cell i,j,k (L).
Transmissivity in the column direction is the product of transmissivity in the row
direction and a horizontal anisotropy factor specified by the user; the horizontal
anisotropy factor is a constant for each layer. Conductances in each direction are
calculated from transmissivity and cell dimensions. When head drops below the
aquifer bottom (equation (45)), the cell is considered to be dewatered, and is
permanently set to no flow; the model has no provision for the resaturation of a
dewatered cell. Thus errors may arise in attempts to simulate situations in which
actual reversals in water-level occur. Errors can also arise if oscillations of computed
heads occur during iteration; if such computational oscillations cause head to drop
erroneously below the botton of the cell, the cell will change to no flow for all
succeeding iterations and time steps. As a means of controlling this problem, the
iterative solvers contain provisions for slowing the rate of convergence.
In the program described herein a layer-type flag, LAYCON, is used to specify
whether or not the simulation of water table conditions through equations (43)-(45) is
to be invoked. This is discussed more fully in the section on data requirements.
Vertical Conductance Formulation
Vertical conductance terms are calculated within the model using data from an input
array which incorporates both thickness and vertical hydraulic conductivity in a
single term, and using horizontal (or map) areas calculated from cell dimensions. In
general, the vertical interval between two nodes, i,j,k and and i,j,k+1, may be
considered to contain n geohydrologic layers or units, having vertical hydraulic
conductivities K1,K2 . . . . Kn and thicknesses z
1
, z
2
. . . . z
n
. The map area of
the cells around nodes i,j,k and i,j,k+1 is DELRj*DELCi; the vertical conductance of
an individual geohydrologic layer, g, in this area is given by
g
g j i
g
C
=
K DELR
*
DELC
z
(46)
The equivalent vertical conductance, Ci
,j,k+1/2
, for the full vertical interval between
nodes i,j,k and i,j,k+1 is found by treating the n individual geohydrologic layers as
conductances in series; this yields
1
C
=
1
C
=
1
K DELR DELC
z
=
1
DELR DELC
z
K i, j,k+1/ 2 g=1
n
g g=1
n
g j i
g
j i g=1
n
g
g

(47)
rearranging equation (47)
i, j,k+1/ 2
j i
g=1
n
g
g
C
DELR DELC
=
1
z
K
(48)
The quantity C
i,j,k+1/2
/(DELR
j
*DELC
i
) has been termed the "vertical leakance" and is
designated Vcont
i,j,k+1/2
in this report; thus we have
i, j,k+1/ 2
g=1
n
g
g
Vcont
=
1
z
K
(49)
Vcont is the term actually used as input in the model described herein. That is, rather
than specifying a total thickness and an equivalent (or harmonic mean) vertical
hydraulic conductivity for the interval between node i,j,k and node i,j,k+1, the user
specifies the term Vcont
i,j,k+1/2
, which is actually the conductance of the interval
divided by the cell area, and as such incorporates both hydraulic conductivity and
thickness. The program multiples Vcont by cell area to obtain vertical conductance.
The values of Vcont must be calculated or determined externally to the program; this
is generally done through an application of equation (49). The Vcont values are
actually read as the elements of a two-dimensional input array, Vcont
i,j
, for each
layer. Each value of Vconti,j is the vertical leakance for the interval between cell
i,j,k and cell i,j,k+1--that is, for the interval between the layer for which the array is
read, and the layer below it. It follows that the Vcont array is not read for the
lowermost layer in the model. Although values of Vcont are thus read into the model
through a series of two-dimensional input arrays, the discussion in this section will
continue to be given in terms of three-dimensional array notation, Vcont
i,j,k+1/2
, to
emphasize the fact that the Vcont values refer to the intervals between layers.
Figure 26 shows a situation in which nodes i,j,k and i,j,k+1 both fall within a single
hydrogeologic unit, having a vertical hydraulic conductivity Kz i,j which is uniform
at least within the cell area. For this case, application of equation (49) yields
i, j,k+1/ 2
i, j
k+1/ 2
Vcont
=
Kz
z
(50)
where z
k+1/2
, the vertical distance between nodes, is the sum of v
k
/2 and v
k+1
/2
in which v represents layer thickness as in figure 1. This situation might be found,
for example, where several model layers are used to represent a single geohydrologic
unit in order to provide greater vertical resolution.
Figure 27 shows a case in which two adjacent model layers are used to represent two
vertically adjacent hydrogeologic units, so that nodes i,j,k and i,j,k+1 fall at the
midpoints of these geohydrologic layers. Each layer is characterized by its own value
of vertical hydraulic conductivity, which is again assumed to be uniform at least over
the cell area. The expression for Vcont in this case becomes
i, j,k+1/ 2
k
i, j,k
k+1
i, j,k+1
Vcont
=
1
v
/ 2
Kz
+
v
/ 2
Kz

(51)
where
v
k
is the thickness of model layer k
v
k+1
is the thickness of model layer k+1
Kz i,j,k is the vertical hydraulic conductivity of the upper layer in cell i,j,k
Kz i,j,k+1 is the vertical hydraulic conductivity of the lower layer in cell
i,j,k+1
Figure 26.-Diagram for calculation of vertical leakance, Vcont, between two nodes which fall within a single
geohydrologic unit.
Figure 27.-Diagram for calculation of vertical leakance, Vcont between two nodes located at the midpoints of vertically
adjacent geohydrologic units.
If one value of K
z
is much smaller than the other, the term containing the larger
Kz value will be negligible in equation (51). Thus for this condition, only the term
involving the smaller K
z
value need be retained in the denominator of (51).
Figure 28 shows a third situation, in which node i,j,k and node i,j,k+1 are taken
within (i.e., at the median depths of) two aquifers which are separated by a
semiconfining unit. In this case, three intervals must be represented in the summation
of equation (49)--the lower half of the upper aquifer, the semiconfining unit, and the
upper half of the lower aquifer. The resulting expression for Vcont is
i, j,k+1/ 2
u/ 2
zu
c
zc
L/ 2
zL
Vcont
=
1
z
K
+
z
K
+
z
K

(52)
where
z
u
is the thickness of the upper aquifer
z
c
is the thickness of the confining bed
z
L
is the thickness of the lower aquifer
K
zu
is the vertical hydraulic conductivity of the upper aquifer
K
zc
is the vertical hydraulic conductivity of the semiconfining unit
K
zL
is the vertical hydraulic conductivity of the lower aquifer; and
each of these terms must in general be considered to vary with the map location (i,j)
of the nodes. In many applications it turns out that K
zc
is much smaller than either
K
zu
or K
zL
; in these situations the terms involving K
zu
and K
zL
are negligible in
equation (52) so that the expression for Vcont becomes
i, j,k+1/ 2
zc
c
Vcont
=
K
z
(53)
Figure 28.-Diagram for calculation of vertical leakance, Vcont, between two nodes located at the midpoints of aquifers
which are separated by a semiconfining unit.
If the formulation of equation (53) is applied to the situation shown in figure 28, and
if the further assumptions are made that the confining bed makes no measureable
contribution to the horizontal conductance or the storage capacity of either model
layer, then in effect model layer k represents the upper aquifer, model layer k+1
represents the lower aquifer, and the confining bed is treated simply as the vertical
conductance between the two model layers. This formulation is equivalent to that of
figure 12, and is frequently referred to as the "quasi-three-dimensional" approach.
In summary, the model described herein utilizes a single input array, Vcont, which
incorporates both vertical hydraulic conductivity and thickness, rather than
independent inputs for thickness and conductivity. The program multiplies Vcont by
cell area to obtain vertical conductance. This requires the user to calculate Vcont
values externally to the program, using equation (49) in the general case (where n
hydrogeologic layers occur in the vertical interval between nodes) or equations (50),
(51), (52) or (53) in the situations shown in figures 26-28. While this approach
involves some preprocessing of input data, it actually increases the flexibility of
model application. Because layer transmissivity (or hydraulic conductivity and
bottom elevation if unconfined) and layer storage coefficient are also used as input
terms, the model never actually reads vertical gird spacing data. Thus the model can
implement either the orthogonal mesh of figure 9-b or a deformed mesh such as that
of figure 9-c, and can similarly be adapted to either a direct three-dimensional
simulation or to the quasithree-dimensional formulation, without modification of the
program.
Vertical Flow Calculation Under Dewatered
Conditions
The basic finite difference equation for cell i,j,k (equation (24)) was given as
i, j-1/ 2,k
i, j -1,k
m
i, j,k
m
i, j+1/ 2,k
i, j+1,k
m
i, j,k
m
CR
(
h
-
h
)+
CR
(
h
-
h
)
+
CC
(
h
-
h
)+
CC
(
h
-
h
)
i -1/ 2, j,k
i -1, j,k
m
i, j,k
m
i+1/ 2, j,k
i+1, j,k
m
i, j,k
m
+
CV
(
h
-
h
)+
CV
(
h
-
h
)
i, j,k-1/ 2
i, j,k-1
m
i, j,k
m
i, j,k+1/ 2
i, j,k+1
m
i, j,k
m
+
P
h
+Q =
SS
(
r c v
)
h
-
h
t
-
t
i, j,k
i, j,k
m
i, j,k
i, j,k j i k
i, j,k
m
i, j,k
m-1
m m-1

(54)
In this equation the term CV
i,j,k+1/2
(h
m
i,j,k+1
- h
m
i,j,k
) gives the flow into cell i,j,k
through its lower face, i.e.
q
i,j,k+1/2
= CV
i,j,k+1/2
(h
m
i,j,k+1
- h
m
i,j,k
) (55)
where following the convention of equation (24), a positive value of q
i,j,k+1/2
indicates
flow into cell i,j,k and a negative value indicates flow out of the cell. Equations (54)
and (55) are based on the assumption that cells i,j,k and i,j,k+1 are fully saturated -
i.e., that the water level in each cell stands higher than the elevation of the top of the
cell. There are, however, situations in which a portion of a confined aquifer may
become unsaturated--for example, when drawdown due to pumpage causes water
levels to fall, at least locally, below the top of the aquifer. In terms of simulation, this
condition is shown in figure 29. Two aquifers separated by a confining bed are
simulated using the quasi-three-dimensional approach, in which the upper aquifer is
represented by cell i,j,k, the underlying aquifer by cell i,j,k+1, and the confining bed
by the vertical conductance between the two layers, CV
i,j,k+1/2
. Pumping from the
lower layer has lowered the water level in cell i,j,k+1 below the elevation of the top
of the cell, so that the aquifer is effectively unconfined within the cell area. An
assumption is made that the confining layer remains fully saturated from top to
bottom, and we consider the head difference across this confining unit. At the upper
surface of the confining unit the head is simply that in the upper aquifer in cell i,j,k--
hi,j,k. Just below the lower surface of the confining unit, however, unsaturated
conditions prevail, so that the pressure sensed on the lower surface of the confining
unit is atmospheric-taken as zero in the model formulation. Thus the head at the base
of the confining unit is simply the elevation at that point-i.e., the elevation of the top
of the lower cell. If this elevation is designated TOPi,j,k+1, the flow through the
confining bed is obtained by substituting TOPi,j,k+1 for hi,j,k+1 in equation (55),
q
i,j,k+1/2
= CV
i,j,k+1/2(
TOP
i,j,k+1
- h
m
i,j,k
) (56)
Figure 29.-Situation in which a correction is required to limit the downward flow into cell i,j,k+1, as a result of partial
desaturation of the cell.
Thus the flow will be downward, from cell i,j,k to cell i,j,k+1 (i.e., following the
convention of equation (26), q
i,j,k+1/2
will be negative); but under this condition the
flow will no longer be dependent on the water level, h
i,j,k+1
, in the lower cell. The
simplest approach to this problem in formulating the equation for cell i,j,k would be
to substitute the flow expression of equation (56) into equation (54), in place of the
expression given in (55). However, if we consider the matrix of coefficients of the
entire system of finite difference equations (matrix [A] of equation (27)), direct
substitution of the expression in (56) into the equation for node i,j,k would render
this matrix unsymmetric, generating problems in the solution process. To avoid this
condition, an alternative approach is used. The flow term of equation (55) is allowed
to remain on the left side of equation (54). The flow into cell i,j,k as computed by
this term, is
CV
i,j,k+1/2
(h
m
i,j,k+1
- h
m
i,j,k
)
(where in this case, since h
i,j,k
> h
i,j,k+1
, the computed flow is negative, indicating
movement out of cell i,j,k.) The "actual" flow into cell i,j,k is given by equation (56)
as CV
i,j,k+1
/2(TOP
i,j,k+1
- h
m
i,j,k
) (where again h
m
i,j,k
> TOP
i,j,k+1
indicating movement
out of the cell). A correction term, qc, can be obtained by subtracting equation (56)
from equation (55), i.e.
qc = (computed flow into cell i,j,k)
- ("actual" flow into all i,j,k)=
CV
i,j,k+1/2
(h
m
i,j,k+1
- TOP
i,j,k+1
)
To compensate for allowing the computed flow to remain on the left side of equation
(54), the term qc is added to the right side of equation (54). In the operation of the
model, equation (54), which is identical to equation (24), is rearranged to the form of
equation (26); and in practice, the term is added to the right side, RHS, of equation
(26). This immediately introduces a difficulty, since qc contains the term hmi,j,k+1,
and all terms involving unknown heads must be kept on the left side of equation (26).
to circumvent this difficulty, qc is actually computed using the value of hmi,j,k+l
from the preceding interation, rather than that from the current iteration, i.e.
q
c,n
= CV
i,j,k+1/2
(h
m,n-1
i,j,k+1
- TOP
i,j,k+1
) (58)
where q
c,n
is the value of q
c
to be added in the nth iteration, and h
m,n-1
i,j,k+1
is the value
of h
m
i,j,k+1
from the preceding iteration, n-1. As convergence is approached the
difference between h
m,n-1
i,j,k+l
and h
m,n
i,j,k+l
becomes progressively smaller, and the
approximation involved in (58) thus becomes more accurate. In the first iteration of
each time step, the initial trial value of hi,j,k+1 is used in computing qc.
The process described above is used in formulating the equations for cell i,j,k when
the underlying cell, i,j,k+1, has "dewatered"-i.e., when the water level in i,j,k+1 has
fallen below the top of the cell. A correction must also be applied in formulating the
equations for the dewatered cell itself. To examine this correction, we now take cell
i,j,k to be the dewatered cell, and we consider flow into i,j,k from the overlying cell,
i,j,k-1. For this case, the computed flow into cell i,j,k from above is CV
i,j,k-1/2
(h
m
i,j,k-1
- h
m
i,j,k
) whereas the "actual" flow into the cell is CV
i,j,k-1/2
(h
m
i,j,k-1
- T0P
i,j,k
). The
difference, computed minus "actual" flow, is thus q
c
' = CV
i,j,k-1/2
(TOP
i,j,k
- h
m
i,j,k
)
where q
c
' should be added to the right hand side of equation (54) or (26). From a
programming point of view, the most efficient way to handle this correction is to add
the term CV
i,j,k-1/2
to HCOF on the left side of equation (26), while adding the term
(CV
i,j,k-1/2
TOP
i,j,k
) to the RHS term. Because HCOF forms part of the coefficient of
hmi,j,k, which falls on the main diagonal of the coefficient matrix, this correction
does not affect the symmetry of the coefficient matrix; at the same time, the problems
entailed in placing an unknown head value on the right side of the equation are
avoided.
In summary, whenever dewatering of a cell occurs, two corrections must be made--
one in formulating equation (26) as it applies to the overlying cell, and one in
formulating equation (26) as it applies to the dewatered cell itself. These two
corrections are discussed separately above, in each case using the designation i,j,k to
represent the cell for which equation (26) is formulated. It is important to keep in
mind, however, that both corrections are applied in any dewatering event, and that
the form of the corrections has been developed to preserve the symmetry of the
coefficient matrix [A] of equation (27), and to maximize program efficiency.
In the program described herein, the user specifies whether or not the procedure for
limiting vertical flow under dewatered conditions is to be implemented. This is done
through the layer type-flag, LAYCON, as discussed in the section on data
requirements.
Storage Formulation
In the formulation of storage terms, the program described herein distinguishes
between layers in which storage coefficient values remain constant throughout the
simulation, and those in which the storage coefficient may "convert" from a confined
value to a water table value, or vice-versa, as the water level in a cell falls below or
rises above the top of the cell. This distinction is made through the use of the layer
flag, LAYCON, as described in the section on data requirements.
For a layer in which storage coefficient is to remain constant during the simulation,
the storage formulation is based upon a direct application of the storage expression in
equation (24) or (54). This expression, which applies to an individual cell, i,j,k, has
the form

V
t
=
SS
(
r c v
)
h
-
h
t
-
t
i, j,k j i k
i, j,k
m
i, j,k
m-1
m m-1
(60)
where V/t is the rate of accumulation of water in the cell, and as such must appear
on the right side of equation (24) or (54); SSi,j,k is the specific storage of the
material in cell i,j,k; r
j
, c
i
and v
k
are the cell dimensions; h
m
i,j,k
is the head in
cell i,j,k at the end of time step m; h
m-1
i,j,k
is the head in cell i,j,k at the end of time
step m-1; t
m
is the time at the end of time step m; and t
m-1
is the time at the end of
time step m-1. In equation (26) the notation SC1i,j,k was introduced, where SC1
i,j,k
=
Ss
i,j,k
r
j
c
i
v
k
. In this report the term SC1
i,j,k
is termed the "storage capacity" or
the "primary storage capacity" of cell i,j,k; the "primary" designation is used to
distinguish SC1
i,j,k
from a secondary storage capacity which is used when storage
term conversion is invoked, as explained in the following section. Using the concept
of storage capacity, the expression for rate of accumulation in storage in cell i,j,k can
be written
SC1
i,j,k
(h
m
i,j,k
- h
m-1
i,j,k
)/(t
m
- t
m-1
).
This expression is separated into two terms in equation (26), SC1
i,j,k
h
m
i,j,k
/(t
m
-t
m-1
),
which is incorporated in the left side of (26) through the term HCOF
i,j,k
, and SC1
i,j,k
h
m-1
i,j,k
/(t
m
-t
m-1
), which is included in the term RHS
i,j,k
on the right side of (26).
The input to the Block-Centered Flow Package requires specification of
dimensionless storage coefficient values in each layer of the model; for a confined
layer these storage coefficient values are given by the specific storage of the cell
material multiplied by layer thickness in the cell, SS
i,j,k
v
k
; for an unconfined layer
they are equal to the specific yield of the material in the cell. The incorporation of
layer thickness into the confined storage term maintains the flexibility of the program
to represent layers of varying thickness, and to implement either the direct three-
dimensional or "quasi-three-dimensional" conceptualizations of vertical
discretization. The storage coefficient values are read layer by layer; they are
designated as array sfl in the input instructions. These values are then multiplied by
the cell areas, r
j
c
i
, to create storage capacity values, and they are stored in the
SC1 array.
Storage Term Conversion
The primary storage capacity described above, SC1
i,j,k
is adequate for simulations in
which the water level in each individual cell remains either above the top of the cell
or below the top of the cell throughout the course of the simulation. If the water level
crosses the top of a cell during a simulation--i.e., if the water level in a confined
(fully saturated) cell falls below the top of the cell as a result of simulated pumpage,
or if the water level in an unconfined cell rises above the top of the cell--then in
effect the system "converts" from confined to water table conditions, or vice versa,
during the simulation. Where these conditions appear to be possible, the user may
invoke storage term conversion for the entire layer through use of the layer-type flag.
When this is done, the primary storage capacity, SC1
i,j,k
for any cell in the layer will
represent the confined storage coefficient multiplied by cell area; a secondary storage
capacity, SC2
i,j,k
is used to represent specific yield multiplied by cell area. Values of
confined storage coefficient for each cell in the layer are read through the two-
dimensional input array sf1. These confined storage coefficient values are multiplied
by cell areas to obtain confined storage capacities, which are stored in the array SC1.
Values of specific yield for each cell in the layer are read through the two-
dimensional input array sf2. These specific yield values are multiplied by cell areas
to obtain unconfined storage capacities, which are stored in array SC2.
In a layer which has been designated for storage term conversion, the expression for
rate of accumulation in storage in cell i,j,k is formulated as follows
V
t
=
SCB(
h
-
LTOP
)+SCA(
LTOP
-
h
t
-
t
i, j,k
m
i, j,k i, j,k
i, j,k
m-1
m m-1
(61)
where again V/ t is rate of accumulation of water in storage in cell i,j,k and as
such must appear on the right side of equation (24) or (54); SCA is the storage
capacity in effect in cell i,j,k at the start of the time step; and SCB is the "current"
storage capacity--that is, the storage capacity in effect during the iteration in process.
Consider a case in which the head in cell i,j,k at the beginning of time step m (h
m-1
i,j,k
)
is above the top of the cell. Since there is no free surface in the cell at the start of the
time step, the storage capacity at that time is taken as the confined storage capacity--
that is, SCA is set equal to SC1
i,j,k
. If, during a given iteration for time step m, the
computed value of head for the end of the time step (h
m
i,j,k
) is found to be above the
top of the cell, SCB for the following iteration is also set equal to SC1
i,j,k
; equation
(61) for that iteration then reverts to the form of equation (60). However, if the
computed value of hmi,j,k in a given iteration turns out to be below the top of the
cell, as shown in figure 30, the value of SCB for the following iteration is set equal to
SC2, the unconfined storage capacity. In this case the computed rate of release of
water from storage in the time step has two components:
SC1
i,j,k
(LTOP
i,j,k
- h
m-1
i,j,k
)/(t
m
- t
m-1
) ,
the rate of release from confined or compressive storage; and
SC2
i,j,k
(h
m
i,j,k
- TOP
i,j,k
)/(t
m
- t
m-1
) ,
the rate of release from water table storage.
If the head at the beginning of the time step, h
m-1
i,j,k
, is below the top of cell i,j,k, so
that a free surface exists within the cell, SCA in equation (61) is set equal to SC2
i,j,k
.
If, during an iteration for time step m, the computed value of head for the end of the
time step turns out to be below the top of the cell, SCB in the subsequent iteration is
also set equal to SC2
i,j,k
and equation (61) again reverts to the form of equation (60).
However, if the computed head for the end of the time step turns out to be above the
top of the cell, SCB in the subsequent iteration is set equal to SC1
i,j,k
, the confined
storage capacity. This situation occurs during intervals of rising water level, and
again two components are computed for the rate of accumulation of water in storage-
-one corresponding to unconfined or water table storage and one corresponding to
confined or compressive storage.
Equation (61) can be rearranged as follows
V
t
=
SCB
t
-
t
h
+
SCA(
LTOP
-
h
) - SCB*
LTOP
t
-
t m m-1
i, j,k
m
i, j,k
i, j,k
m-1
i, j,k
m m-1
(62)
Figure 30.-A model cell which uses two storage factors during one iteration.
Again, V/ t represents rate of accumulation in storage and as such would appear
on the right in equation (24) or (54). In the formulation of equation (26), therefore,
the term SCB/(t
m
- t
m-1
) is subtracted from HCOF
i,j,k
on the left hand side, while the
term
SCA(
LTOP
-
h
) - SCB*
LTOP
t
-
t
i, j,k
i, j,k
m-1
i, j,k
m m-1
is added to RHS
i,j,k
on the right.
Applicability and Limitations of Optional
Formulations
The options for calculation of horizontal conductance under water table conditions,
limitation of vertical flow under desaturating conditions, and storage term conversion
were all developed on the assumption that each model layer corresponds to a distinct
aquifer or permeable horizon, and that these horizons are separated by distinct units
of low permeability. Use of these options where these conditions are not satisfied
may lead to a variety of problems and inaccuracies in simulation. For example, if the
option for horizontal conductance calculation under water table conditions is used
where a water table aquifer is represented by several model-layers, and the water
table is expected to traverse more than one layer during simulation, incorrect (and
irreversible) conversion of cells to a no-flow condition may occur as iterations are
carried out. Thus care should be exercised in the decision to use any of the three
options noted above.
Data Requirements
The fundamental variables controlling cell-to-cell flow and storage in the model are
entered through the Block-Centered-Flow Package input. These variables, depending
on the options which are invoked, may include transmissivity, hydraulic conductivity,
specific yield, confined storage coefficient, vertical leakance, aquifer bottom
elevation and aquifer top elevation. Each of these variables is entered using the
utility arrayreader module U2DREL, which is described in Chapter 14. This module
either reads a two-dimensional array of data for a single layer, or accepts a single
value provided by the user and applies that value throughout the array, for all cells in
the layer.
The model utilizes a layer-type code to classify layers according to the simulation
options that are used. In particular, the layer-type code indicates whether specified
transmissivity values are to be used, or transmissivities are to be calculated at each
iteration as the product of hydraulic conductivity and saturated thickness; whether
storage term conversion is to be used; and whether limitation of vertical flow from
above is to be invoked under dewatered conditions. Because the layer-type code
identifies the options to be employed in a given layer, it indicates the kinds of data
required for the layer, and thus identifies the data arrays to be read. The data are
entered layer by layer; for each layer a set of two-dimensional arrays, one array for
each required parameter, is read in turn. That is, all of the required arrays for layer 1
are read initially, in sequence, then all of the arrays for layer 2, and so on until all
layers have been covered. This method of data organization provides a simpler input
process than would be possible using the alternative of a series of three-dimensional
arrays corresponding to the various parameters.
Within each layer the required parameters should be specified for every cell,
including constant-head and no-flow cells. For no-flow cells, the entered values are
never used in calculation, and thus any values may be specified; for constant head
cells, the storage terms are not used but the other parameters are, and realistic values
for those parameters must be entered.
Two parameters, transmissivity and hydraulic conductivity, each require the-
designation of two values at each cell--one in the row direction and one in the column
direction. To reduce input effort, only a single array is read for each of these
parameters, giving only the values in the row direction; these row-direction values
are subsequently multiplied by an anisotropy factor to obtain the corresponding
column-direction values. A single value of the anisotropy factor is specified by the
user for each layer, through the one-dimensional array TRPY (NLAY).
Vertical leakance terms (Vcont, or Kz/ z) are associated with each layer except the
lowermost; the values associated with a given layer actually apply to the interval
between that layer and the next lower layer. For example, the array of Vcont values
entered during the input sequence for layer 1 actually applies to the interval between
the midpoint of layer 1 and the midpoint of layer 2.
In addition to the terms mentioned above, the Block-Centered Flow Package input
includes cell dimensions (DELR and DELC), a flag to indicate whether the
simulation is transient or steady state (ISS), and a flag to indicate whether cell-by-cell
flow terms are to be saved (IBCFCB). If the ISS flag is set for steady-state
conditions (ISS = 1), no space is allocated for storage coefficient or specific yield
and storage calculations are skipped. Thus for steady-state runs, arrays of storage
coefficients or specific yields must not be included in the input data; if they are
included, the data sequence will be misread. Note that erroneous specification of ISS
or of a LAYCON value will also cause misreading of the data array sequence.
Four types of layer are recognized by the model, incoporating various combinations
of the options provided by the Block-Centered-Flow Package. These four layer types
are identified by their layer-type codes, which are stored in the one-dimensional array
LAYCON (NLAY). The code values and the corresponding layer characteristics are
given below.
Layer-type 0--In this category there is no provision for modification of transmissivity
as water level varies, for storage term conversion, or for limitation of vertical flow
from above if water level falls below the top of the cell. This layer type is normally
used to simulate confined conditions, but could also be used to simulate a layer in
which unconfined conditions will always prevail, provided drawdowns are expected
to be a small fraction of layer thickness and flow from the overlying layer (if present)
is expected to be negligible. If the simulation is transient, storage coefficient or
specific yield values are entered in the input array sfl(NCOL, NROW); then row-
direction transmissivities are entered in the input array Tran (NCOL, NROW); and
following the transmissivities, unless the layer is the lowermost in the model, vertical
leakance values are entered in the input array Vcont (NCOL, NROW). Again,
parameter values may be specified by providing the entire array, or by providing a
single default value which is applied to all cells of the layer. The parameter values
assigned at the beginning of a simulation in this type of layer are retained without
change throughout the simulation.
Layer-type 1--This layer type is utilized only in a single-layer model or in the
uppermost layer of a model, and only where unconfined conditions are expected to
persist in the layer throughout the entire period of simulation. No provision is made
for storage term conversion, by virtue of the assumption that water table conditions
will always prevail; and no provision is made for limiting flow from above under
dewatered conditions, since layer-type 1 is used only for the uppermost layer of a
model. However, transmissivities are computed at each iteration as the products of
hydraulic conductivity and saturated thickness values within the layer. Thus the input
data includes hydraulic conductivity and cell bottom elevation, rather than
transmissivity. If the simulation is indicated as transient, specific yield values are
entered in the input array sfl(NCOL, NROW). Row direction hydraulic conductivity
values are then entered in the input array HY(NCOL, NROW) and cell bottom
elevations are entered in the array BOT(NCOL, NROW). If the model contains more
than one layer, vertical leakance values are entered in the input array Vcont(NCOL,
NROW). Because use of this layer type would be inappropriate except in the
uppermost layer, a check of the layer number is made whenever LAYCON is given a
value of one; if the layer number is not also equal to one, indicating the uppermost
model layer, an error message is printed.
Layer-type 2--This layer type is used where the situation may alternate between
confined and unconfined conditions, so that storage term conversion and limitation of
flow from above under dewatered conditions are both desirable; but where the
saturated thickness is expected to remain everywhere a high fraction of the layer
thickness throughout the period of simulation, so that recalculation of transmissivity
as the product of hydraulic conductivity and saturated thickness is not necessary.
The storage term conversion option requires that both a confined storage coefficient
and a specific yield value be specified for each cell, and that the top elevation be
specified for each cell; the top elevation is also used in the option to limit flow from
above under dewatered conditions. If the simulation is transient, confined storage
coefficient values are entered in the input array sf1(NCOL, NROW). Transmissivity
values are then entered in the array Tran(NCOL, NROW). Unless the layer is the
lowermost in the model, vertical leakance values are next entered in the array
Vcont(NCOL, NROW). Specific yield values are then entered in the array
sf2(NCOL, NROW) if the simulation is transient; and finally layer top elevations are
entered in the array TOP (NCOL, NROW).
Layer-Type 3--This layer type incorporates all of the Block-CenteredFlow options
associated with water table conditions. Transmissivities are recalculated at each
iteration using hydraulic conductivities and layer bottom elevations, and both storage
term conversion and limitation of flow from above under dewatered conditions are
implemented. The required data thus includes hydraulic conductivities, layer bottom
elevations, confined storage coefficients (if transient), specific yields (if transient),
vertical leakances and layer top elevations. Confined storage coefficients are entered
in the input array sfl (NCOL, NROW); hydraulic conductivity values are then entered
in the array HY(NCOL, NROW), and aquifer bottom elevations in BOT(NCOL,
NROW). Unless the layer is the lowermost in the model, vertical leakance values are
next entered in the array Vcont(NCOL, NROW). Specific yield values are then
entered in the array sf2(NCOL,NROW); and finally aquifer top elevations are entered
in the array TOP(NCOL, NROW).
The input sequence is outlined in the following section. Both of the utility modules
which are used are described in Chapter 14, and the required formats are illustrated
in the "Sample Input to the BCF Package" and in appendix D.
Block-Centered Flow Package Input
Input for the Block-Centered Flow (BCF) Package is read from the unit specified in
IUNIT(1).
FOR EACH SIMULATION
BCF1AL
1. Data: ISS IBCFCB
Format: I10 I10
2. Data: LAYCON(NLAY) (Maximum of 80 layers)
Format: 4012
(If there are 40 or fewer layers, use one record; otherwise, use two records.)
BCF1RP
3. Data: TRPY(NLAY)
Module: U1DREL
4. Data: DELR(NCOL)
Module: U1DREL
5. Data: DELC(NROW)
Module: U1DREL
A subset of the following two-dimensional arrays are used to describe each layer.
The arrays needed for each layer depend on the layer type code (LAYCON) and
whether the simulation is transient (ISS = 0) or steady state (ISS = 1). If an array is
not needed, it must be omitted. All of the arrays (items 6-12) for layer 1 are read
first; then all of the arrays for layer 2, etc.
IF THE SIMULATION IS TRANSIENT
6. Data: sf1(NCOL,NROW)
Module: U2DREL
IF THE LAYER TYPE CODE (LAYCON) IS ZERO OR TWO
7. Data: Tran(NCOL,NROW)
Module: U2DREL
IF THE LAYER TYPE CODE (LAYCON) IS ONE OR-THREE
8. Data: HY(NCOL,NROW)
Module: U2DREL
9. Data: BOT(NCOL,NROW)
Module: U2DREL
IF THIS IS NOT THE BOTTOM LAYER
10. Data: Vcont(NCOL,NROW)
Module: U2DREL
IF THE SIMULATION IS TRANSIENT AND THE LAYER TYPE CODE
(LAYCON) IS TWO OR THREE
11. Data: sf2(NCOL,NROW)
Module: U2DREL
IF THE LAYER TYPE CODE IS TWO OR THREE
12. Data: TOP(NCOL,NROW)
Module: U2DREL
Explanation of Fields Used in Input Instructions
ISS--is the steady-state flag.
If ISS = 0, the simulation is steady state.
If ISS not equal to 0, the simulation is transient.
IBCFCB--is a flag and a unit number.
If IBCFCB > 0, it is the unit number on which cell-by-cell flow terms will
be recorded whenever ICBCFL (see Output Control) is set; the terms which are saved
will include cell-by-cell storage terms, cell-by-cell constant head flows, and internal
cell-by-cell flows.
If IBCFCB = 0, cell-by-cell flow terms will not be printed or recorded.
If IBCFCB < 0, flow for each constant-head cell will be printed, rather
than saved on disk, whenever ICBCFL is set; cell-by-cell storage terms and internal
cell-by-cell flows will neither be saved nor printed.
LAYCON--is the layer type table. Each element holds the code for the respective
layer. Read one value for each layer. There is a limit of 80 layers. Leave unused
elements blank.
0 - confined--Transmissivity and storage coefficient of the layer are constant
for the entire simulation.
1 - unconfined--Transmissivity of the layer varies. It is calculated from the
saturated thickness and hydraulic conductivity. The storage coefficient is constant;
valid only for layer 1.
2 - confined/unconfined--Transmissivity of the layer is constant. The storage
coefficient may alternate between confined and unconfined values. Vertical leakage
from above is limited if the layer desaturates.
3 - confined/unconfined--Transmissivity of the layer varies. It is calculated
from the saturated thickness and hydraulic conductivity. The storage coefficient may
alternate between confined and unconfined values. Vertical leakage from above is
limited if the aquifer desaturates.
TRPY--is a one-dimensional array containing an anisotropy factor for each layer. It
is the ratio of transmissivity or hydraulic conductivity (whichever is being used)
along a column to transmissivity or hydraulic conductivity along a row. Read one
value per layer. Set to 1.0 for isotropic conditions. NOTE: This is one array with
one value for each layer.
DELR--is the cell width along rows. Read one value for each of the NCOL columns.
DELC--is the cell width along columns. Read one value for each of the NROW
rows.
sf1--is the primary storage coefficient. Read only for a transient simulation (steady-
state flag, ISS, is 0). Note that for Laycon=1, sfl will always be specific yield, while
for Laycon=2 or 3, sf1 will always be confined storage coefficient. For Laycon=0,
sf1 would normally be confined storage coefficient; however, layer-type 0 can also
be used for simulation of water table conditions where drawdowns are expected to
remain everywhere a small fraction of the saturated thickness, and where there is no
layer above, or flow from the layer above is negligible; and in this case specific yield
values would be entered in sf1.
Tran--is the transmissivity along rows. Tran is multiplied by TRPY to obtain
transmissivity along columns. Read only for layers where LAYCON is zero or two.
HY--is the hydraulic conductivity along rows. HY is multiplied by TRPY to obtain
the hydraulic conductivity along columns. Read only for layers where LAYCON is
one or three.
BOT--is the elevation of the aquifer bottom. Read only for layers where LAYCON
is one or three.
Vcont--is the vertical hydraulic conductivity divided by the thickness from a layer to
the layer beneath it. Since there is not a layer beneath the bottom layer, Vcont cannot
be specified for the bottom layer.
sf2--is the secondary storage coefficient. Read it only for layers where LAYCON is
two or three and only if a transient simulation (steady-state flag, ISS, is zero). The
secondary storage coefficient is always specific yield.
TOP--is the elevation of the aquifer top. Read only for layers where LAYCON is
two or three.
Electronic Manual for MODFLOW 6 River Package 93
6 River Package
Rivers and streams contribute water to the ground-water system or drain water from it
depending on the head gradient between the stream and the ground-water regime.
The purpose of the River Package is to simulate the effects of flow between surface-
water features and ground-water systems. To accomplish this, terms representing
seepage to or from the surface features must be added to the ground-water flow
equation (equation (26)) for each cell affected by the seepage.
Figure 32 shows a stream divided into reaches so that each reach is completely
contained in a single cell. Stream aquifer seepage is simulated between each reach
and the model cell that contains that reach.
The cross-section of figure 33-a shows a situation in which the open water of a
stream is separated from the ground-water system by a layer of low permeability
streambed material. Figure 33-b shows an idealization of this system in which the
stream-aquifer interconnection is represented as a simple conductance through which
one-dimensional flow occurs. The system of figure 33 is helpful in conceptualizing
and describing the simulation of stream-aquifer interaction; however, it must be
recognized that, in many instances, no discrete low-permeability streambed layer is
present. The techniques of simulation developed through the conceptualization of
figure 33 can still be applied to represent these situations, provided the proper
interpretation is placed on the various terms and parameters that are used.
Figure 34 shows an isolated view of the idealized streambed conductance of figure
33-b, as it crosses an individual cell. The length of the conductance block is taken as
the length of the stream, L, as it crosses the node; the width is taken as the stream
width, W; the distance of flow is taken as the thickness, M, of the streambed layer;
and the hydraulic conductivity of the streambed material is designated K. The
assumption is made that measurable head losses between the stream and the aquifer
are limited to those across the streambed layer itself--that is, that there is no signif-
icant head loss between the bottom of the streambed layer and the point represented
by the underlying model node. It is further assumed that the underlying model cell
remains fully saturated--that is, that its water level does not drop below the bottom of
the streambed layer. Under these assumptions, flow between the stream and the
ground-water system is given by
94 6 River Package Electronic Manual for MODFLOW
QRIV =
KLW
M
(HRIV -
h
)
i, j,k
(63-a)
QRIV = CRIV(HRIV -
h
)
i, j,k
(63-b)
Figure 32.-Discretization of a stream into reaches. Some small reaches are ignored.
Figure 33.-(a) Cross section of an aquifer containing a stream and (b) Conceptual representation of stream-aquifer
interconnection in simulation.
Figure 34.-Idealization of streambed conductance in an individual cell.
where QRIV is the flow between the stream and the aquifer, taken as positive if it is
directed into the aquifer; HRIV is the head in the stream; CRIV is the hydraulic
conductance of the stream-aquifer interconnection (KLW/M), and h
i,j,k
is the head at
the node in the cell underlying the stream reach.
If the assumption is satisfied that all significant head loss occurs across a discrete
streambed layer, the application of equations (63) is straightforward. More
frequently, however, equations (63) must be applied to situations in which no discrete
streambed layer can be identified, or in which head losses are not restricted to those
across such a layer. In these cases, the task is to formulate a single conductance term,
CRIV, which can be used in (63-b) to relate flow between the stream and the depth
represented by node i,j,k to the corresponding head difference. This flow is in
general a three-dimensional process, and its representation through a single
conductance term can never be more than approximate. If reliable field
measurements of stream seepage and associated head difference are available, they
may be used to calculate an effective conductance. Otherwise, a conductance value
must be chosen more or less arbitrarily and adjusted during model calibration.
Certain rules can be formulated to guide the initial choice of conductance. For
example, the assumed crosssectional area of flow should normally be of the same
order of magnitude as the product of channel width and stream reach length within
the cell; the assumed distance of flow should not exceed the vertical interval between
the streambed and node i,j,k; and, if distinct layers can be recognized within this
interval, these should normally be treated as conductances in series in formulating an
equivalent conductance. In general, however, it should be recognized that
formulation of a single conductance term to account for a three-dimensional flow
process is inherently an empirical exercise, and that adjustment during calibration is
almost always required.
Equations (63) normally provide an acceptable approximation of streamaquifer
interaction over a certain range of aquifer head values. In most cases, however, if
water levels in the aquifer fall below a certain point, seepage from the stream ceases
to depend on head in the aquifer. This can be visualized by returning to the concept
of a discrete streambed layer. Figure 35-a shows the situation described by equations
(63); water level in the aquifer is above the bottom of the streambed layer, and flow
through that layer is proportional to the head difference between the stream and the
aquifer. In figure 35-b, water level in the aquifer has fallen below the bottom of the
streambed layer, leaving an unsaturated interval beneath that
Figure 35.-Cross sections showing the relation between head at the bottom of the streambed layer and head in the cell.
Head in the cell is equal to the water-table elevation.
layer; if it is assumed that the streambed layer itself remains saturated, the head at its
base will simply be the elevation at that point. If this elevation is designated RBOT,
the flow through the streambed layer is given by
QRIV = CRIV (HRIV - RBOT) (64)
where QRIV, CRIV, and HRIV are as defined for equation (63-b). Obviously,
further declines in head below RBOT produce no increase in flow through the
streambed layer; the flow simply retains the constant value given by equation (64), as
long as head remains below RBOT. The model described herein utilizes these
concepts in simulating stream-aquifer interaction--that is, flow between a stream and
a node i,j,k is simulated according to the equation set
QRIV = CRIV (HRIV - h
i,j,k
), h
i,j,k
> RBOT
QRIV = CRIV (HRIV - RBOT), h
i,j,k
< RBOT (65)
Figure 36 shows a graph of flow between the stream and cell i,j,k as a function of the
head, h
i,j,k
, as calculated using equations 65. Flow is 'zero when h
i,j,k
, is equal to the
water level in the stream, HRIV.. For higher values Of h
i,j,k
, flow is negative, that is,
into the stream; for lower values of h
i,j,k
, flow is positive, that is, into the aquifer.
This positive flow increases linearly as h
i,j,k
decreases, until h
i,j,k
reaches RBOT;
thereafter the flow remains constant.
A relationship similar to that of equations (65) and figure (36) generally prevails in
stream-aquifer interaction whether or not a discrete streambed layer is present. For
example, once a break in saturation occurs between the stream and the aquifer,
seepage from the stream to the aquifer must become independent of head in the
aquifer. In most cases, moreover, this independence is established even before a
break in saturation occurs.
Figure 36.-Plot of flow, QRIV, from a stream into a cell as a function of head, h, in the cell where RBOT is the elevation
of the bottom of the streambed and HRIV is the head in the stream.
Figure 37 shows a situation in which levels in the aquifer have fallen far enough
below a stream so that only a narrow saturated connection exists between the
streambed and the regional water table. Examination of figure 37 will show that the
head gradient in the saturated connection must be approximately unity, and that
further lowering of the regional water table will not increase this gradient. Thus,
once a condition similar to that in figure 37 is established, seepage from the stream is
independent of further head decline in the aquifer. The situation shown in figure 37
is itself an overimplification of field conditions, which may often involve complex
patters of saturated and unsaturated material beneath the stream. In all situations,
however, seepage from the stream must at some point become independent of head in
the aquifer, as that head continues to decline.
If hydrologic conditions indicate that seepage from a stream will increase as the local
water table elevation declines, but will reach the limiting condition illustrated in
figure 37 when the water table reaches an elevation h
l
, RBOT should be taken equal
to h
l
. Because the vertical hydraulic gradient beneath the stream is approximately
one under the conditions of figure 37, seepage from the stream into cell i,j,k is given
approximately by the product KLW, where K is the vertical hydraulic conductivity of
the material in the saturated column beneath the stream, and again L is the length of
the stream as it crosses cell i,j,k, and W is stream width. A value of the stream-
aquifer conductance term, CRIV, consistent with this estimated value of limiting
seepage and with the selected value of RBOT can be obtained by substituting KLW
for QRIV in equation (64) and solving for CRIV. This yields
CRIV = (KLW)/(HRIV - RBOT) (66)
Figure 37.-Limiting seepage from a stream at unit hydraulic gradient
In summary, if the limiting condition of stream seepage is expected to follow the
model of figure 37, RBOT should be chosen as the water table elevation at which the
transition to this limiting seepage can be expected, and CRIV may be calculated from
equation (66). The model simulation technique based on equations (65) should then
provide a reasonable approximation to the stream-aquifer interaction.
The simplified model of stream-aquifer interaction utilized here assumes that this
interaction is independent of the location of the stream reach within the cell, and that
the level of water in the stream is uniform over the reach, and constant over each
stress period. The latter assumption implies that conditions of flow in the stream do
not vary significantly during the stress period--for example, that the stream does not
go dry or overflow its banks, or that such events are of such short duration as to have
no effect on stream-aquifer interaction.
Data describing each river are specified by the user for each stress period. Input
consists of six entries for each river reach, specifying the layer, row, and column of
the cell containing the reach, and the three parameters needed to calculate seepage--
stream level or stage (HRIV), the conductance of the stream-aquifer interconnection
(CRIV), and the "bottom elevation," or level at which the limiting value of stream
seepage is attained (RBOT).
At the start of each iteration, terms representing river seepage are added to the flow
equation for each cell containing a river reach. The choice of which river seepage
equation to use, equation 63 or equation 64, is made by comparing the most recent
value of head at the cell to the value of RBOT for the reach. Since this process is
done at the start of each iteration, the most current value of head (HNEW) is the
value from the previous iteration. Thus, the check for which river seepage equation
to use lags behind the seepage calculations by one iteration. If equation 63 is
selected, the term -CRIV is added to the term HCOF and the term -CRIV*HRIV is
added to RHS. If equation 64 is selected, the term -CRIV (HRIV - RBOT) is added
to the term RHS.
River Package Input
Input to the River (RIV) Package is read from the unit specified in IUNIT(4).
FOR EACH SIMULATION
RIV1AL
1. Data: MXRIVR IRIVCB
Format: I10 I10
RIV1RP
2. Data: ITMP
Format: I10
3.Data: Layer Row Column Stage Cond Rbot
Format: I10 I10 I10 F10.0 F10.0 F10.0
(Input item 3 normally consists of one record for each river reach. If ITMP is
negative or zero, item 3 is not read.)
MXRIVR--is the maximum number of river reaches active at one time.
IRIVCB--is a flag and a unit number.
If IRIVCB > 0, it is the unit number on which cell-by-cell flow terms will be
recorded whenever ICBCFL (see Output Control) is set.
If IRIVCB = 0, cell-by-cell flow terms will not be printed or recorded.
If IRIVCB < 0, river leakage for each reach will be printed whenever
ICBCFL is set.
ITMP--is a flag and a counter.
If ITMP < 0, river data from the last stress period will be reused.
If ITMP > 0, ITMP will be the number of reaches active during the current
stress period.
Layer--is the layer number of the cell containing the river reach.
Row--is the row number of the cell containing the river reach.
Column--is the column number of the cell containing the river reach.
Stage--is the head in the river.
Cond--is the riverbed hydraulic conductance.
Rbot--is the elevation of the bottom of the riverbed.
Electronic Manual for MODFLOW 7 Recharge Package 105
7 Recharge Package
The Recharge (RCH) Package is designed to simulate areally distributed recharge to
the ground-water system. Most commonly, areal recharge occurs as a result of
precipitation that percolates to the ground-water system. Recharge applied to the
model is defined as
QR
i,j
= I
i,j
*DELR
j
*DELC
i
(67)
where QR
i,j
is the recharge flow rate applied to the model at horizontal cell location
(i,j) expressed as a fluid volume per unit time; and I
i,j
is the recharge flux (in units of
length per unit time) applicable to the map area, DELR
j
*DELC
i
, of the cell. The
recharge, QR
i,j
, is applied to a single cell within the vertical column of cells located at
(i j). There is no need to allow for recharge to occur simultaneously at multiple
depths in the same vertical column because natural recharge enters the ground-water
system at its top. In the simplest situation, the top of the ground-water system will
occur in model layer 1; however, the vertical position of the top of the system may
vary with horizontal location and with time as the water-table rises and falls. Three
options for specifying the cell in each vertical column of cells that receives the
recharge have been implemented as described below. The RCH Package can
potentially be used to simulate recharge from sources other than precipitation -- for
example, artificial recharge. If the ability to apply recharge to more than one cell in a
vertical column of cells is required, then the Well Package, which allows recharge or
discharge to be specified at any model cell, can be used.
In the package described herein, values of recharge flux, I
i,j
, are read into a two
dimensional array, RECH
i,j
, at each stress period (unless an option is exercised to use
recharge fluxes from the previous stress period). These values of recharge flux are
immediately multiplied by horizontal cell areas, DELR
j
DELC
i
, to obtain values Of
QR
i,j
, which are maintained in the RECH array. The cell within each vertical column
to which the recharge is applied is specified through the recharge option code,
NRCHOP, and optional array IRCH. The options include: (1) application of the
recharge to model layer 1; (2) application of the recharge to any cell in the vertical
column as specified by layer numbers contained in two dimensional array IRCHi,j;
and (3) application of the recharge to the uppermost active cell in the vertical
column, provided there is no constant head cell above it in the column. Under
options 1 and 2, if a cell designated to receive recharge is no-flow, then no recharge
106 7 Recharge Package Electronic Manual for MODFLOW
is added. Under the third option, if there is a constant head cell in a vertical column
of cells and there is no active cell above, then no recharge is applied to this column
because it is assumed that any recharge would be intercepted by the constant head
source. Recharge flux values that are read into the model must be expressed in units
that are consistent with the length and time units used to represent all other model
parameters.
In the formation of the matrix equations, the recharge flow rate, QR
i,j
, associated with
a given horizontal cell location (i,j) and vertical location, k, that is determind by the
recharge option is subtracted from the value of RHS
i,j,k
(equation (26) or (29)). This
is done at each iteration for all cells that receive recharge. Because recharge as
defined is independent of aquifer head, nothing is added to the coefficient of head,
HCOFi,j,k.
Figure 38.-Hypothetical problem showing which cells receive recharge under the three options available in the Recharge
Package.
Careful consideration should be given to the problem under study and to the other
options employed in its simulation before deciding which of the three recharge
options listed above to utilize in a given situation. For example, figure 38 shows a
situation in which a cross sectional model has been used to simulate a hypothetical
problem involving recharge, seepage from a stream, and seepage into a stream (figure
38-a). Using the provision described in Chapter 5 for horizontal conductance
formulation under water table conditions, the model mesh has been progressively
truncated during simulation so that the uppermost active cells in each vertical column
fall approximately at the water table. This process yields the final distribution of
active, constant head and no-flow cells shown in figure 38-b.
Figure 38-C illustrates the recharge distribution to the model if option 1 above is
utilized. Under this option recharge is permitted only to the top layer of the model.
Thus once, the water table shape has been simulated by the use of no flow cells in the
top layer, recharge to the vertical columns beneath those cells is shut off. This
clearly fails to simulate the given system.
Figure 38-d illustrates the recharge distribution if option 2 is utilized, assuming that
the user specifies recharge cells prior to the simulation on the basis of an estimated
water table position, which differs slightly from that finally obtained in the simulation
process. Four of the cells which the user had designated as recharge cells have
converted to an inactive condition and thus receive no recharge.
Figure 38-e illustrates the Simulation under the third option, which turns out to be the
one best suited for this particular situation. Under this option, recharge enters the
uppermost active cell in each vertical column, except where constant head cells have
been used to represent the streams. Thus, a continuous distribution of recharge to the
water table is simulated.
For the typical situation of recharge from precipitation, option 3 is the easiest to use.
The model user does not have to be concerned about determining which is the highest
active cell in a vertical column because the program automatically determines this
throughout the simulation. Option 1, however, can be useful in situations where
recharge should not pass through the no-flow cells in layer 1. For example, some
cells may be designated no-flow because they are impermeable. Any recharge
specified for those cells should not pass into layer 2. Of course option 3 could still be
used in this situation by specifying that the recharge rate is zero at the impermeable
cells. The user should select the option that will result in the least effort for
specification of input data. Similarly, option 2 may be useful when layers other than
layer 1 have outcrop areas and when recharge to the specified layers should not
penetrate through no-flow cells to a lower layer. Other factors to consider when
choosing the best option are that option 2 uses more computer memory than options 1
and 3, and option 3 uses slightly more computer time than options 1 and 2.
Recharge Package Input
Input to the Recharge (RCH) Package is read from the unit specified in IUNIT(8).
FOR EACH SIMULATION
RCH1AL
1.Data: NRCHOP IRCHCB
Format: I10 I10
RCH1RP
2 .Data: INRECH INIRCH
Format: I10 I10
3. Data: RECH(NCOL,NROW)
Module: U2DREL
IF THE RECHARGE OPTION IS EQUAL TO 2
4. Data: IRCH(NCOL, NROW)
Module: U2DINT
NRCHOP--is the recharge option code. Recharge fluxes are defined in a two-
dimensional array, RECH, with one value for each vertical column. Accordingly,
recharge is applied to one cell in each vertical column, and the option code
determines which cell in the column is selected for recharge.
1 - Recharge is only to the top grid layer.
2 - Vertical distribution of recharge is specified in array IRCH.
3 - Recharge is applied to the highest active cell in each vertical column. A constant-
head node interceptsrecharge and prevents deeper infiltration.
IRCHCB--is a flag and a unit number.
If IRCHCB >0,it is the unit number on which cell-by-cell flow terms will be
recorded whenever ICBCFL (see Output Control) is set.
If IRCHCB < 0, cell-by-cell flow terms will not be printed or recorded.
INRECH--is the RECH read flag.
If INRECH > 0, an array of recharge fluxes, RECH, is read.
If INRECH < 0, recharge rates from the preceding stress period are used.
INIRCH--is the IRCH read flag. When NRCHOP is two,
If INIRCH > 0, an array of layer numbers (IRCH) is read.
If INIRCH < 0, the array (IRCH) used in the preceding stress period is
reused.
Note: When NRCHOP is one or three, INIRCH is ignored.
RECH--is the recharge flux (Lt-1). Read only if INRECH is greater than or equal to
zero.
IRCH--is the layer number array that defines the layer in each vertical column where
recharge is applied. Read only if NRCHOP is two and if INIRCH is greater than or
equal to zero.
Electronic Manual for MODFLOW 8 Well Package 111
8 Well Package
The Well Package is designed to simulate features such as wells which withdraw
water from the aquifer (or add water to it) at a specified rate during a given stress
period, where the rate is independent of both the cell area and the head in the cell.
The discussion in this section is developed on the assumption that the features to be
simulated are actually wells, either discharging or recharging.
Well discharge is handled in the Well Package by specifying the rate, Q, at which
each individual well adds water to the aquifer or removes water from it, during each
stress period of the simulation. Negative values of Q are used to indicate well
discharge, while positive values of Q indicate a recharging well.
At the beginning of each stress period, the WEL1RP module reads four values for
each well--the row, column and layer number of the cell in which the well is located,
and the discharge or recharge rate, Q, of the well during that stress period. At each
iteration, as the matrix equations are formulated, the value of Q for each well is
subtracted from the RHS value (equation (26) or (29)) for the cell containing that
well. Where more than one well falls within a single cell, the calculation is repeated
for each well as the RHS term for that cell is assembled. Thus the user specifies the
discharge associated with each individual well, and these are in effect summed within
the program to obtain the total discharge from the cell.
The Well Package, as it is presently formulated, does not accommodate wells which
are open to more than one layer of the model. However, a well of this type can be
represented as a group of single-layer wells, each open to one of the layers tapped by
the multi-layer well, and each having an individual Q term specified for each stress
period. If this approach is used, the discharge of the multi-layer well must be divided
or apportioned in some way among the individual layers, externally to the model
program. A common method of doing this is to divide the well discharge in
proportion to the layer transmissivities i.e.
1
w
1
Q
Q
=
T
T
(68)
112 8 Well Package Electronic Manual for MODFLOW
where Q
1
is the discharge from layer 1 to a particular well in a given stress period, Q
w
is the well discharge in that stress period, T
1
is the transmissivity of layer 1 and ET
represents the sum of the transmissivities of all layers penetrated by the well. Again,
it's important to note that equation (68), or some other method of apportioning the
discharge, must be implemented by the user externally to the program for each multi-
layer well, and for each stress period.
This approach, in which a multi-layer well is represented as a group of single layer
wells, fails to take into account the interconnection between various layers provided
by the well itself, and is thus an incomplete representation of the problem. A
package which will provide an improved approximation of multi-layer well effects is
under development.
Electronic Manual for MODFLOW 8 Well Package 113
Well Package Input
Input for the Well (WEL) Package is read from the unit specified in IUNIT(2).
FOR EACH SIMULATION
WEL1AL
1. Data: MXWELL IWELCB
Format: I10 I10
FOR EACH STRESSPERIOD
WEL1RP
2.Data: ITMP
Format: I10
3.Data: Layer Row Column Q
Format: I10 I10 I10 F10.0
(Input item 3 normally consists of one record for each well.
If ITMP is negative or zero, item 3 is not read.)
MXWELL--is the maximum number of wells used at any time.
IWELCB--is a flag and a unit number.
If IWELCB > 0, it is the unit number on which cell-by-cell flow terms will
be recorded whenever ICBCFL (see Output Control) is set.
If IWELCB = 0, cell-by-cell flow terms will not be printed or recorded.
If IWELCB < 0, well recharge will be printed whenever ICBCFL is set.
If ITMP < 0, well data from the last stress period will be reused.
If ITMP > 0, ITMP will be the number of wells active during the current
stress period.
114 8 Well Package Electronic Manual for MODFLOW
Layer--is the layer number of the model cell that contains the well.
Row--is the row number of the model cell that contains the well.
Column--is the column number of the model cell that contains the well.
Q--is the volumetric recharge rate. A positive value indicates recharge and a
negative value indicates discharge.
Electronic Manual for MODFLOW 9 Drain Package 115
9 Drain Package
The Drain Package is designed to simulate the effects of features such as agricultural
drains, which remove water from the aquifer at a rate proportional to the difference
between the head in the aquifer and some fixed head or elevation, so long as the head
in the aquifer is above that elevation, but which have no effect if head falls below that
level. The discussion in this section is phrased on the assumption that the features to
be simulated are actually agricultural drains.
Figure 39 shows a cross section through a cell, illustrating concepts underlying the
simulation of drains in the model. The drain is assumed to run only partially full, so
that head within the drain is approximately equal to the median drain elevation, di,j,k.
The head computed by the model for cell i,j,k (hi,j,k) is actually an average value for
the cell, and is normally assumed to prevail at some distance from the drain itself.
The drain head, di,j,k prevails only locally, within the drain--it does not characterize
the cell as a whole. Between the drain and the area in which head hi,j,k prevails there
exists a radial or semiradial flow pattern in the vertical plane, normally characterized
by progressively steeper head gradients as the drain is approached. The head loss
within this converging flow pattern forms one part of the head difference hi,j,k-di,j,k.
An additional component of head loss may occur in the immediate vicinity of the
drain if the hydraulic conductivity in that region differs from the average value used
for cell i,j,k--because of the presence of foreign material around the drain pipe, or
channel-bed material in the case--of an open drain (figure 40). Finally, head losses
occur through the wall of a drain pipe, depending upon the number and size of the
openings in the pipe, and the degree to which those openings may be blocked by
chemical precipitates, plant roots, etc.
116 9 Drain Package Electronic Manual for MODFLOW
Figure 39.-Cross section through cell i,j,k illustrating head loss in convergent flow into drain.
The three processes discussed above--convergent flow toward the drain, flow through
material of different conductivity immediately around the drain, and flow through the
wall of the drain--each generate head losses which may be assumed proportional to
the discharge, QD, through the system-that is, the discharge from cell i,j,k into the
drain. Because these head losses occur in series, the total head loss hi,j,k-di,j,k may
also be taken as proportional to QD. This has been done in the method of simulation
embodied in the Drain Package. That is, it has been assumed that the drain function
is described by the equation pair
QD
i,j,k
= CD
i,j,k
(h
i,j,k
- d
i,j,k
) for h
i,j,k
> d
i,j,k
(69)
QD
i,j,k
= 0 for h
i,j,k
< d
i,j,k
(70)
The coefficient CD
i,j,k
of equation (69) is a lumped (or equivalent) conductance
describing all of the head loss between the drain and the region of cell i,j,k in which
the head h
i,j,k
can be the assumed to prevail. It depends on the characteristics of the
convergent flow pattern toward the drain, as well as on the characteristics of the drain
itself and its immediate environment.
One could attempt to calculate values for CD by developing approximate equations
for conductance for the three flow processes, and then calculate the equivalent series
conductance. The conductance for each process would be based on the formulation
of a one-dimensional flow equation. The formulations vary significantly depending
on the specific drain system being simulated, so no general formulation for
calculating CD is presented here. Also, in most situations a specific formulation
would require detailed information that is not usually available, such as detailed head
distribution around the drain, aquifer hydraulic conductivity near the drain,
distribution of the fill material, hydraulic conductivity of fill material, number and
size of the drain pipe openings, the amount of clogging materials, and the hydraulic
conductivity of the clogging materials. In practice, it is more common to calculate
CD from measured values of QD and h-d using equation (69). If h-d is not accurately
known, CD is usually adjusted during model calibration in order to match measured
values of QD to model calculated values.
Figure 40.-Factors affecting head loss immediately around a drain: (a) buried drain pipe in backfilled ditch and (b) open
drain.
Figure 41 shows a graph of QD vs. h
i,j,k
as defined by equations (69) and (70); the
function is similar to that for flow between a surface stream and the aquifer (figure
36) except that flow into the aquifer is excluded, and positive values of QD have
been taken as corresponding to flow into the drain. With proper selection of
coefficients, the River Package could in fact be utilized to perform the functions of
the Drain Package.
Because QD
i,j,k
in equation (69) has been taken as a flow out of cell i,j,k, it must be
subtracted from the left side of equation (24) for each cell affected by a drain,
provided the head h
i,j,k
is above the drain elevation. This is accomplished in the
Drain Package by testing to determine whether head exceeds drain elevation, and if
so, by adding the term -CD
i,j,k
to HCOF
i,j,k
(equation (26)) and adding the term -
CD
i,j,k
d
i,j,k
to RHS
i,j,k
, as the matrix equations are assembled.
Figure 41.-Plot of flow, QD, into a drain as a function of head, h, in a cell where the elevation of the drain is d and the
conductance is CD.
Drain Package Input
Input to the Drain (DRN) Package is read from the unit specified in IUNIT(3).
FOR EACH SIMULATION
DRN1AL
1. Data: MXDRN IDRNCB
Format: I10 I10
DRN1RP
2. Data: ITMP
Format: I10
3. Data: Layer Row Col Elevation Cond
Format: I10 I10 I10 F10.0 F10.0
(Input item 3 normally consists of one record for each drain. If ITMP is
negative or zero, item 3 will not be read.)
MXDRN--is the maximum number of drain cells active at one time.
IDRNCB--is a flag and a unit number.
If IDRNCB > 0, it is the unit number on which cell-by-cell flow terms will
If IDRNCB = 0, cell-by-cell flow terms will not be printed or recorded.
If IDRNCB < 0, drain leakage for each cell will be printed whenever
ICBCFL is set.
If ITMP < 0, drain data from the last stress period will be reused.
If ITMP > 0, ITMP will be the number of drains active during the current
stress period.
Layer--is the layer number of the cell containing the drain.
Row--is the row number of the cell containing the drain.
Column--is the column number of the cell containing the drain.
Elevation--is elevation of the drain.
Cond--is the hydraulic conductance of the interface between the aquifer and the
drain.
Electronic Manual for MODFLOW 10 Evapotranspiration Package 121
10 Evapotranspiration Package
The Evapotranspiration (ET) Package simulates the effects of plant transpiration and
direct evaporation in removing water from the saturated ground water regime. The
approach is based on the following assumptions: (1) when the water table is at or
above a specified elevation, termed the "ET surface" in this report,
evapotranspiration loss from the water table occurs at a maximum rate specified by
the user; (2) when the depth of the water table below the ET surface elevation
exceeds a specified interval, termed the "extinction depth" or "cutoff depth" in this
report, evapotranspiration from the water table ceases; and (3) between these limits,
evapotranspiration from the water table varies linearly with water table elevation.
This can be expressed in equation form as
RET
i,j
= RETM
i,j
h
i,j,k
> h
si,j
(71)
RET
i,j
= 0 h
i,j,k
< h
si,j
- d
i,j
(72)
RET
i,j
= RETM
i,j
{h
i,j,k
- (h
si,j
- d
i,j
)}/d
i,j
(h
si,j
-d
i,j
) <= h
i,j,k
<= h
si,j
(73)
where RET
i,j
is the rate of loss per unit surface area of water table due to
evapotranspiration, in volume of water per unit area per unit time, within the map
area DELR
j
DELC
i
; h
i,j,k
is the head, or water table elevation in the cell from which
the evapotranspiration occurs; RETM
i,j
is the maximum possible value of RET
i,j
; h
si,j
is the ET surface elevation, or the water table elevation at which this maximum value
of evapotranspiration loss occurs; and d
i,j
is the cutoff or extinction depth, such that
when the distance between h
si,j
and h
i,j,k
exceeds d
i,j
evapotranspiration ceases.
In implementing the finite difference approach the volumetric rate of
evapotranspiration loss from a given cell is required. This is given as the product of
the loss rate per unit area, and the horizontal surface area, DELR
j
DELC
i
, of the cell
from which the loss occurs, i.e.
QET
i,j
= RET
i,j
*DELR
j
*DELC
i
(74)
122 10 Evapotranspiration Package Electronic Manual for MODFLOW
where QET
i,j
is the evapotranspiration, in volume of water per unit time, through the
area DELRjDELCi. If the maximum value of QETi,j (corresponding to RETMi,j) is
designated QETMi,j, equations (7l)-(73) can be expressed in terms of volumetric
discharge as
QET
i,j
= QETM
i,j
h
i,j,k
> h
si,j
(75)
QET
i,j
= 0 h
i,j,k
< h
si,j
-d
i,j
(76)
QET
i,j
= QETM
i,j
{h
i,j,k
- (h
si,j
- d
i,j
)}/d
i,j
(h
si,j
-d
i,j
) <= h
i,j,k
<= h
si,j
(77)
Figure 42 shows a graph of evapotranspiration loss, QET
i,j
, vs head in cell i,j,k based
on equations (75)-(77). Comparison of the ET function with- the river or drain
functions shows that the three are mathematically similar, except that the linear
portion of the ET function is bounded at both ends by constant values, rather than
only at the lower end.
Evapotranspiration is drawn from only one cell in the vertical column beneath the
map area DELR
j
*DELC
i
; the user designates the cell (i.e. the layer, k) using one of
two options. Under the first option, evapotranspiration is always drawn from the
uppermost layer of the model; under the second, the user specifies the cell, within the
vertical column at i,j, from which the evapotranspiration is to be taken. In either case
the computed evapotranspiration has no influence on the simulation if the designated
cell is either a no-flow cell or a constant head cell.
Figure 42.-Plot of volumetric evapotranspiration, QET, as a function of head, h, in a cell where d is the cutoff depth and
hs is the ET surface elevation.
For each cell location, (i, j), in the horizontal plane, and for each stress period (unless
an option is exercised to use prior values) the ET package reads values of RETM
(maximum evapotranspiration loss per unit area per unit time) into an array labelled
EVTR. These rates are immediately multiplied by cell areas, DELR
j
*DELC
i
, to
obtain the maximum volumetric rate of evapotranspiration from each cell, QETM;
these maximum volumetric rates then replace the values of RETM
i,j
in the EVTR
array. Thus, the input to the EVTR array consists of maximum evapotranspiration
rates per unit 1 area, and as such must have dimensions Lt
-1
. In the calculation
carried out within the model, however, the entries in the EVTR array appear as
maximum volumetric rates, having dimensions L
3
t
-1
.
Values of h
si,j
, the ET surface elevation (or water table elevation at which
evapotranspiration is maximum), are read into the two dimensional array SURF by
the ET package; values of the cutoff depth or extinction depth are read into the two-
dimensonal array EXDP. Because the term QET
i,j
of equations (75)-(77) has been
defined as an outflow from the aquifer it must be subtracted from the left side of
equation (24). In terms of the expressions HCOF and RHS of equation (26), this is
accomplished in the ET package as follows:
1) if h
i,j,k
< (h
si,j,k
- d
i,j
) no changes are made in the terms HCOF or RHS for cell i,j,k;
2) if h
i,j,k
> h
si,j
, QETM
i,j
is added to RHS
i,j
; and
3) if (h
si,j,k
- d
i,j
) < h
i,j,k
< h
si,j,k
, -QETM
i,j
/d
i,j
is added to HCOF
i,j,k
and -QETM
i,j
{h
si,j
-
d
i,j
}/d
i,j
is added to RHS
i,j,k
.
The value of h
s,i,j
, the water table elevation at which evapotranspiration is maximum,
should normally be taken as the average land surface elevation in the map area
DELR
j
DELC
i
; the cutoff or extinction depth, d
i,j
, is then frequently assumed to be on
the order of six to eight feet below land surface (although considerable variation can
be introduced by climatic factors, the presence of deep-rooted phreatophytes, or so
on). Where the distance from land surface to the water table varies extensively
within the area of a cell, care must be exercised in implementing the ET package and
in choosing the various parameters of equation (70), or misleading results may be
obtained.
The options for selection of the layer from which ET is to be drawn provide some
flexibility in adapting the package to special situations, but also require some care in
implementation. Figure 43 shows a situation similar to that discussed for the
recharge package, in which a cross sectional model has been progressively truncated
to follow the water table, using the provision for horizontal conductance formulation
under water table conditions (Chapter 5). Figure 43-a shows the hydrologic situation
under study, and figure 43-b the final distribution of variable head and inactive (no
flow) cells obtained in the simulation.
Under option 1 (figure 43-c), evapotranspiration is drawn only from the uppermost
layer of the model; in the problem shown, the presence of no flow cells in this layer
deletes evapotranspiration from the right half of the model, so that the simulation
fails to represent field conditions.
Figure 43-d shows the situation which could be achieved through the use of option 2,
assuming that the simulation was carried out in stages and that the user interacted
with the simulation process, designating the cells from which evapotranspiration was
to be drawn as the truncation of the mesh developed.
Figure 43.-Hypothetical problem showing cells from which ET will be abstracted under the two options available in the
ET Package.
Evapotranspiration Package Input
Input to the Evapotranspiration (EVT) Package is read from the unit specified in
IUNIT (5).
FOR EACH SIMULATION
EVT1AL
1. Data: NEVTOP IEVTCB
Format: I10 I10
EVT1RP
2.Data: INSURF INEVTR INEXDP INIEVT
3.Data: SURF
Module: U2DREL
4.Data: EVTR
Module: U2DREL
5.Data: EXDP
Module: U2DREL
IF THE ET OPTION IS EQUAL TO TWO
6.Data: IEVT
Module: U2DINT
NEVTOP--is the evapotranspiration (ET) option code. ET parameters (ET surface,
maximum ET rate, and extinction depth) are specified in two-dimensional arrays,
SURF, EVTR, and EXDP, with one value for each vertical column. Accordingly, ET
is calculated for one cell in each vertical column. The option codes determine for
which cell in the column ET will be calculated.
1 - ET is calculated only for cells in the top grid layer.
2 - The cell for each vertical column is specified by the user in array IEVT.
IEVTCB--is a flag and a unit number.
If IEVTCB > 0, it is the unit number on which cell-by-cell flow terms will
be recorded whenever ICBCFL (see Output Control ) is set.
If IEVTCB < 0, cell-by-cell flow terms will not be printed or recorded.
INSURF--is the ET surface (SURF) read flag.
If INSURF > 0, an array containing the ET surface elevation will be read.
If INSURF < 0, the ET surface from the preceding stress period will be
reused.
INEVTR--is the maximum ET rate (EVTR) read flag.
If INEVTR > 0, an array containing the maximum ET rate will be read.
If INEVTR < 0, the maximum ET rate from the preceding stress period will
be reused.
INEXDP--is the extinction depth (EXDP) read flag.
If INEXDP > 0, an array containing the extinction depth (EXDP) will be
read.
If INEXDP < 0, the extinction depth from the preceding stress period will
be reused.
INIEVT--is the layer indicator (IEVT) read flag. It is used only if the ET option
(NEVTOP) is equal to two.
If INIEVT > 0, an array containing the layer indicators (IEVT) will be read.
If INIEVT < 0, layer indicators used during the preceding stress period will
be reused.
SURF--is the elevation of the ET surface.
EVTR--is the maximum ET rate (volume of water per unit area (Lt-1)).
EXDP--is the ET extinction depth.
IEVT--is the layer indicator array. For each horizontal location, it indicates the layer
from which ET is removed. It is needed only if the ET option is equal to two.
Electronic Manual for MODFLOW 11 General-Head Boundary Package 129
11 General-Head Boundary
Package
The function of the General-Head Boundary (GHB) Package is mathematically
similar to that of the River, Drain and ET Packages, in that flow into or out of a cell
i,j,k, from an external source is provided in proportion to the difference between the
head in the cell, h
i,j,k
, and the head assigned to the external source, h
bi,j,k
. Thus a
linear relationship between flow into the cell and head in the cell is established, i.e.
Q
bi,j,k
= C
bi,j,k
(h
bi,j,k
- h
i,j,k
) (78)
where Q
bi,j,k
is the flow into cell i,j,k from the source; C
bi,j,k
is the conductance
between the external source and cell i,j,k; h
bi,j,k
is the head assigned to the external
source; and h
i,j,k
is the head in cell i,j,k. The relationship between cell i,j,k and the
external source is shown schematically in figure 44. The constant-head source is
represented by the apparatus on the right in figure 44, which holds the source head at
the level hb regardless of other factors; the link between the source and cell i,j,k is
represented by the block of porous material C
bi,j,k
. Note that figure 44 shows no
mechanism to limit flow in either direction as h
i,j,k
rises or falls.
A graph of Q
bi,j,k
versus h
i,j,k
as given by equation (78) is shown in figure 45. In
contrast to the River, Drain and ET Packages, the GHB Package provides no limiting
value of flow to bound the linear function in either direction; and as the head
difference between cell i,j,k and the source increases, flow into or out of the cell
continues to increase without limit. Care must accordingly be used in utilizing the
GHB Package to insure that unrealistic flows into or out of the system do not develop
during the course of simulation.
Because Q
bi,j,k
of equation (78) is defined as an inflow to the aquifer it must be added
to the left side of equation (24). In terms of the expressions HCOF and RHS of
equation (26), this is accomplished in the model by subtracting the term C
bi,j,k
from
HCOF
i,j,k
and subtracting the term C
bi,j,k
h
bi,j,k
from RHS
i,j,k
as the matrix equations are
assembled.
130 11 General-Head Boundary Package Electronic Manual for MODFLOW
Figure44.-Schematic diagram illustrating principle of general-head boundary package.
Figure 45.-Plot of flow, Q
b
, from a general-head boundary source into a cell as a function of head, h, in the cell where h
b
is the source head.
132 11 General-Head Boundary Package Electronic Manual for MODFLOW
General-Head Boundary Package Input
Input for the General-Head Boundary (GHB) Package is read from the unit specified
in IUNIT(7).
FOR EACH SIMULATION
GHB1AL
1.Data: MXBND IGHBCB
Format: I10 I10
GHB1RP
2.Data: ITMP
Format: I10
Boundary
3.Data: Layer Row Column Head Cond
Format: I10 I10 I10 F10.0 F10.0
(Input item 3 normally consists of one record for each GHB. If ITMP is
negative or zero, item 3 is not read.)
MXBND--is the maximum number of general-head boundary cells at one time.
IGHBCB--is a flag and a unit number.
If IGHBCB > 0, it is the unit number on which cell-by-cell flow terms will
If IGHBCB = 0, cell-by-cell flow terms will not be printed or recorded.
If IGHBCB < 0, boundary leakage for each cell will be printed whenever
ICBCFL is set.
If ITMP < 0, GHB data from the preceding stress period will be reused.
If ITMP > 0, ITMP is the number of general-head boundaries during the
current stress period.
Layer--is the layer number of the cell affected by the head-dependent boundary.
Row--is the row number of the cell affected by the head-dependent boundary.
Column--is the column number of the cell affected by the head-dependent boundary.
Boundary head--is the head on the boundary.
Cond--is the hydraulic conductance of the interface between the aquifer cell and the
boundary.
Electronic Manual for MODFLOW 12 SIP Solver Package 135
12 SIP Solver Package
General Theory
The discussion of the Strongly Implicit Procedure (SIP) presented here utilizes
certain general concepts of matrix algebra and numerical analysis which may be
reviewed in any standard reference, including those noted earlier by Peaceman
(1977), Crichlow (1977) or Remson, Hornberger and Molz (1971). In addition to
general background material, these three references provide descriptions of the
Strongly Implicit Procedure itself which may be consulted to supplement the
discussion presented here.
SIP is a method for solving a large system of simultaneous linear equations by
iteration. The finite difference equation for a single cell, i,j,k, was shown in Chapter
2 to be of the form
CV
i,j,k-1/2
h
i,j,k-1
+ CC
i-1/2,j,k
h
i-1,j,k
+ CR
i,j-1/2,k
h
i,j-1,k
+ (-CV
i,j,k-1/2
- CC
i-1/2,j,k
- CR
i,j-1/2,k
- CR
i,j+1/2,k
- CC
i+1/2,j,k
- CV
i,j,k+1/2
+ HCOF
i,j,k
)h
i,j,k
+ CR
i,j+1/2,k
h
i,j+l,k
+ CC
i+1/2,j,k
h
i+l,j,k
+ CV
i,j,k+1/2
h
i,j,k+1
= RHS
i,j,k
. (79)
One equation of this form is written for each cell in the finite-difference grid,
expressing the relationship among the heads at node i,j,k, and at each of the six
adjacent nodes at the end of a time step. Because each equation may involve up to
seven unknown values of head, and because the set of unknown head values changes
from one equation to the next through the grid, the equations for the entire grid must
be solved simultaneously at each time step. The solution consists of one value of
head for each node, for the end of the step.
136 12 SIP Solver Package Electronic Manual for MODFLOW
The discussion of the SIP procedure presented here is based on the notation of
Weinstein, Stone and Kwan (1969), the developers of SIP. Using their notation,
equation (79) may be written.
Z
i,j,k
h
i,j,k-1
+ B
i,j,k
h
i-1,j,k
+ D
i,j,k
h
i,j-1,k
+ E
i,j,k
h
i,j,k
+ F
i.j,k
h
i,j+1,k
+ H
i,j,k
h
i+1,j,k
+ S
i,j,k
h
i,j,k+1
= Q
i,j,k
(80)
The coefficients in equation (80) all are labelled with the index i,j,k to show that they
are associated with the equation for node i,j,k. Thus Z
i,j,k
of equation (80) is
equivalent to CL
i,j-1/2
of equation (79); E
i,j,k
of equation (80) is equivalent to the
expression (-CV
i,j,k-1/2
- CC
i-1/2,j,k
- CR
i,j-1/2,k
- CR
i,j+1/2,k
- CC
i+1/2,j,k
- CV
i,j,k+1/2
+
HCOF
i,j,k
) of equation (79); and so on.
As pointed out in Chapter 2, the entire set of equations of the form of (80) can be
summarized in matrix form as
[A] {h} = {q} (81)
where [A] is the matrix of coefficients of head, {h} is a vector of head values, and
{q} is a vector of the right-hand terms of equation (80). Figure 46 shows the
elements of the coefficient matrix and of the two vectors for a mesh of three rows,
four columns and two layers. Notice that the matrix [A] is sparse--i.e., that there are
very few nonzero elements--and that these are all located on just seven diagonals, as
indicated in figure 47.
Examination of equations (79) and (80) will show that the term CL
i,j,k-1/2
of equation
(79) appears both as the coefficient Z in equation (80) for node i,j,k, and as the
coefficient S in the corresponding equation for node i,j,k-1, that is
Z
i,j,k
= S
i,j,k-1
(82)
Figure 46.-Correspondence between the finite-difference equations and the matrix equation for a grid of three rows, four
columns, and two layers.
Figure 47.-Structure of coefficient matrix showing nonzero diagonals.
Similarly,
B
i,j,k
= H
i-1,j,k
(83)
and
D
i,j,k
= F
i,j-1,k
(84)
Replacing each Z, B, and D coefficient in the matrix of Figure 46 with the equivalent
S, H, or F element, as defined by equations (82) - (84), yields the matrix of Figure
48, which is readily seen to be symmetric. Thus the coefficient matrix [A] of
equation (81) is symmetric as well as sparse.
A system of equations of the form of (81) can be solved by direct methods if [A] can
be factored into two matrices [L*] and [U*], such that [L*]is in lower triangular form
(all nonzero elements are on or below the main diagonal), while [U*] is in upper
triagular form (all nonzero elements are on or above the main diagonal), and all
elements on the main diagonal of [U*] are equal to one. Figure (49) illustrates the
characteristics of [L*] and [U*] relative to [A] for a 3 x 3 matrix [A]. Once this
factoring has been accomplished, a technique known as "backward and forward
substitution" can be used to complete the solution. However, a difficulty arises in
that, even though [A] is a sparse matrix, [L*] and [U*] are generally not sparse, and a
great deal of computer memory and time may be needed to calculate all of their
nonzero elements. In addition, roundoff errors may become unacceptably large.
The Strongly Implicit Procedure seeks to find a matrix [B] such that the sum matrix
[A + B] can be factored easily into two matrices [L] and [U], where [A + B], [L], and
[U] meet the following conditions:
(1) [A + B] is "close" to [A];
(2) [L] is in lower triangular form while [U] is in upper triangular form, and all
entries along the main diagonal of [U] are equal to unity;
(3) [L] and [U] are both sparse; and
(4) both [L] and [U] have just four nonzero diagonals.
Suppose a matrix [B] is constructed in an attempt to satisfy these conditions; the term
[B] (h) can be added to each side of equation (81) to give
[A + B]{h} = {q} + [B]{h} (85)
A solution vector {h} for equation (85) must also be a solution vector for equation
(81). The presence of the vector {h} on both sides of equation (85) presents an
immediate difficulty; however, if an iterative approach to the solution is utilized
(chapter 2), values of h from the preceding iteration may be used in the head vector
on the right. That is, equation (85) may be expressed in the form
[A + B] {h
l
} = {q} + [B] {h
l-1
} (86)
where {h
l
} is a vector of head values from iteration l, and {h
l-1
} a vector of head
values from iteration l-1. In equation (86), {h
l-1
} is actually used as an
approximation to {h
l
}. If the matrix [B] were known, solution of (86) would be
straightforward; for according to the properties postulated above, [A + B] could be
factored easily into the sparse matrices [L] and [U], allowing the use of forward and
backward substitution. Thus the problem of solving equation (86) is equivalent to
that of finding an appropriate matrix [B]. In practice, however, the solution is
pursued in terms of the matrices [A], [A + B], [L] and [U]. The term [A + B] {h
l-1
}
is subtracted from each side of (86) to yield
[A + B] {h
l
} - [A + B] {h
l-1
} = {q} - [A] {h
l-1
} (87)
or
[A + B] {h
l
-h
l-1
} = {q} - [A] {h
l-1
} (88)
Figure 48.-Symmetric coefficient matrix for a grid containing two rows, three columns, and two layers.
Figure 49.-Decomposition of a coefficient matrix into lowerand upper triangular matrices.
In order that the conditions specified above for [L], [U], and [A + B] may be
satisified, [A + B] must contain six nonzero diagonals which were not present in [A],
as shown in figure 50; the effect of these additional nonzero diagonals is to introduce
new terms into the equation for node i,j,k, involving heads at nodes not adjacent to
i,j,k. The relationship between the elements of [A + B] and the elements of [L] and
[U] is as given in the following equations, where as indicated in figures 50 and 51, a,
b, c, and d, refer to elements of [L], e, f, and g, refer to elements of [U] above the
main diagonal, and capital letters refer to elements of [A + B].
Z
'
i,j,k
= a
i,j,k
(89-a)
A
'
i,j,k
= a
i,j,k
e
i,j,k-1
(89-b)
T
'
i,j,k
= a
i,j,k
f
i,j,k-1
(89-c)
B
'
i,j,k
= b
i,j,k
(89-d)
C
'
i,j,k
= e
i-1,j,k
b
i,j,k
(89-e)
D
'
i,j,k
= c
i,j,k
(89-f)
E
'
i,j,k
= a
i,j,k
g
i,j,k-1
+ b
i,j,k
f
i-1,j,k
+ e
i,j-l,k
c
i,j,k
+d
i,j,k
(89-g)
F
'
i,j,k
= d
i,j,k
e
i,j,k
(89-h)
G
'
i,j,k
= f
i,j-1,k
c
i,j,k
(89-i)
H
'
i,j,k
= f
i,j,k
d
i,j,k
(89-j)
U
'
i,j,k
= b
i,j,k
g
i-1,j,k
(89-k)
R
'
i,j,k
= g
i,j-1,k
c
i j,k
(89-i)
S
'
i,j,k
= g
i,j,k
d
i,j,k
(89-m)
If the subscript of an element in equations (89-a...m) places the element outside of the
grid boundary, the element is assumed to be equal to zero. The 13 equations contain
20 unknown values, the elements of [L], [U], and [A + B]. This indicates that there
are many matrices [B] which can be added to [A] so that the sum can be factored into
upper and lower triangular matrices of the form of [L] and [U]. However, the
requirement that [A + B] must be "close" to [A], or equivalently that
[A + B] {h} ~ [A] {h} (90)
has not yet been used. In terms of the elements of [A + B] (h) and [A] (h) associated
with an individual node, i,j,k, equation (90) implies that
Z
'
i,j,k
h
i,j,k-1
+ A
'
i,j,k
h
i,j+1,k-1
+ T
'
i,j,k
h
i+1,j,k-1
+ B
'
i,j,k
h
i-1,j,k
+ C
'
i,j,k
h
i-1,j+1,k
+ D
'
i,j,k
h
i,j-1,k
+ E
'
i,j,k
h
i,j,k
+ F
'
i,j,k
h
i,j+1,k
+ G
'
i,j,k
h
i+1,j-1,k
+ H
'
i,j,k
h
i+1,j,k
+ U
'
i,j,k
h
i-1,j,k+1
+ R
'
i,j,k
h
i,j-1,k+1
+ S
'
i,j,k
h
i,j,k+1
~ Z
i,j,k
h
i,j,k-1
+ B
i,j,k
h
i-1,j,k
+ D
i,j,k
h
i,j-1,k
+ E
i,j,k
h
i,j,k
+ F
i,j,k
h
i,j+1,k
+ H
i,j,k
h
i+1,j,k
+ S
i,j,k
h
i,j,k+1
(91)
Equation (91) can be rearranged so that the terms from the six nonzero diagonals not
present in [A] are all on the right side, while the left side is made up of differences
between elements of matrix [A] and corresponding elements of matrix [A + B], i.e.
(Z
i,j,k
-Z
'
i,j,k
)h
i,j,k-1
+ (B
i,j,k
-B
'
i,j,k
)h
i-1,j,k
+ (D
i,j,k
-D
'
i,j,k
)h
i,j-1,k
+ (E
i,j,k
-E
'
i,j,k
)h
i,j,k
+ (F
i,j k
-F
'
i,j,k
)h
i,j+1,k
+ (H
i,j,k
-H
'
i,j,k
)h
i+1,j,k
+ (S
i,j,k
-S
'
i,j,k
)h
i,j,k+1
~ A
'
i,j,k
h
i,j+1,k-1
+ T
'
i,j,k
h
i+1,j,k-1
+ C
'
i,j,k
h
i-1,j+1,k
+ G
'
i,j,k
h
i+1,j-1,k
+ U
'
i,j,k
h
i-1,j,k+1
+ R
'
i,j,k
h
i,j-1,k+1
(92)
The terms on the right side of (92), corresponding to the six nonzero diagonals of [A
+ B] not appearing in [A], are all derived from the matrix B, and all involve the
heads at nodes not adjacent to node, i,j,k; by contrast, the terms on the left side of
(92) are derived from both [A] and [B], and involve the heads at i,j,k and the six
adjacent nodes.
Figure 50.-Structure of matrix [A+B] showing nonzero diagonals.
Figure 51.-Structure, showing nonzero diagonals, of the lower triangular factor [L] and the upper triangular farctor [U] of
matrix [A+B].
To reduce the effect of the terms corresponding to nodes not adjacent to i,j,k, three
parameters, here termed , and , and all chosen between zero and one, are
introduced as multipliers of the terms on the right side of equation (92). Ultimately,
as the solution of the matrix equations ((85) or (86)) is implemented, these
multipliers take on the role of iteration parameters. They are brought into equation
(92) as follows:
(Z
i,j,k
-Z'
i,j,k
)h
i,j,k-1
+ (B
i,j,k
-B
'
i,j,k
)h
i-1,j,k
+ (D
i,j,k
-D
'
i,j,k
)h
i,j-1,k
+ (E
i,j,k
-E
'
i,j,k
)h
i,j,k
+ (F
i,j,k
-F
'
i,j,k
)h
i,j+1,k
+ (H
i,j,k
-H
'
i,j,k
)h
i+1,j,k
+ (S
i,j,k
-S
'
i,j,k
)h
i,j,k+1
A
'
i,j,k
h
i,j+1,k-1
+ T
'
i,j,k
h
i+1,j,k-1
+ C
'
i,j,k
h
i-1,j+1,k
+ G
'
i,j,k
h
i+1,j-1,k
+ U
'
i,j,k
h
i-1,j,k+1
+ R
'
i,j,k
h
i-1,j,k+1
(93)
Next the heads on right side of (93), corresponding to nodes not adjacent to i,j,k, are
expressed in terms of heads at nodes which are adjacent to i,j,k. This is done by
noting that, for example, node i, j+l, k-1 lies at the corner of a rectangle, the other
three corners of which are: i,j,k-1; i,j+l,k; and i,j,k. Thus using the rules in
interpolation illustrated in figure 52, h
i,j+1,k-1
is given approximately by
h
i,j+1,k-1
= h
i,j+1,k
+ h
i,j,k-1
- h
i,j,k
. (94-a)
Similarly,
h
i+1,j,k-1
= h
i,j,k-1
+ h
i+1,j,k
- h
i,j,k
(94-b)
h
i-1,j+1,k
= h
i-1,j,k
+ h
i,j+1,k
- h
i,j,k
(94-c)
Figure 52.-Estimation of a function at one corner of a rectangle in terms of the values of the function at the other three
corners.
h
i+1,j-1,k
= h
i+1,j,k
+ h
i,j-1,k
- h
i,j,k
(94-d)
h
i-1,j,k+1
= h
i,j,k+1
+ h
i-1,j,k
- h
i,j,k
(94-e)
h
i,j-1,k+1
= h
i,j,k+1
+ h
i,j-1,k
- h
i,j,k
(94-f)
Substituting equations (94-a... f) into equation (93) and reorganizing gives
(Z
'
i,j,k
- Z
i,j,k
+ A
'
i,j,k
+ T
'
i,j,k
) h
i,j,k-1
+ (B
'
i,j,k
- B
i,j,k
+ C
'
i,j,k
+ U
'
i,j,k
) h
i-l,j,k
+ (D
'
i,j,k
- D
i,j,k
+ G
'
i,j,k
+ R
'
i,j,k
) h
i,j-l,k
+ (E
'
i,j,k
- E
i,j,k
- A
'
i,j,k
- T
'
i,j,k
- C
'
i,j,k
- G
'
i,j,k
- U
'
i,j,k
- R
'
i,j,k
) h
i,j,k
+ (F
'
i,j,k
- F
i,j,k
+ A
'
i,j,k
+ C
'
i,j,k
) h
i,j+l,k
+ (H
'
i,j,k
- H
i,j,k
+ T
'
i,j,k
+ G
'
i,j,k
)h
i+1,j,k
+ (S
'
i,j,k
- S
i,j,k
+ U
'
i,j,k
+ R
'
i,j,k
)h
i,j,k+l
~ 0 (95)
The relation expressed in equation (95) can be satisfied ifeach coefficient is
approximately equal to zero. Setting these coefficients equal to zero yields the
equations
Z
'
i,j,k
- Z
i,j,k
+ A
'
i,j,k
+ T
'
i,j,k
= 0 (96-a)
B
'
i,j,k
- B
i,j,k
+ C
'
i,j,k
+ U
'
i,j,k
= 0 (96-b)
D
'
i,j,k
- D
i,j,k
+ G
'
i,j,k
+ R
'
i,j,k
= 0 (96-c)
E
'
i,j,k
- E
i,j,k
- A
'
i,j,k
- T
'
i,j,k
- C
'
i,j,k
- G
'
i,j,k
- U
'
i,j,k
- R
'
i,j,k
= 0 (96-d)
F
'
i,j,k
- F
i,j,k
+ A
'
i,j,k
+ C
'
i,j,k
= 0 (96-e)
H
'
i,j,k
- H
i,j,k
+ T
'
i,j,k
+ G
'
i,j,k
= 0 (96-f)
S
'
i,j,k
- S
i,j,k
+ U
'
i,j,k
+ R
'
i,j,k
= 0 (96-g)
Equations (96-a ... g) and (89-a ... m) form a system of 20 equations in 20 unknowns
which when solved, will yield the entries of [A + B], [L] and [U] such that EA + B] is
"close" to [A], and can be readily factored into [L] and [U], where [L] and [U] are
both sparse and have the required lower triangular and upper triangular forms. For
example, substituting equations (89-a, -b, and -c) into equation (96-a) and
rearranging yields
a
i,j,k
= Z
i,j,k
/(1 + e
i,j,k-1
+ f
i,j,k-1
). (97-a)
Similarly,
b
i,j,k
= B
i,j,k
/(1 + e
i-1,j,k
+ g
i-1,j,k
) (97-b)
c
i,j,k
= D
i,j,k
/(1 + f
i,j-1,k
+ g
i,j-1,k
) (97-c)
A
'
i,j,k
= a
i,j,k
e
i,j,k-1
(97-d)
C
'
i,j,k
= e
i-1,j,k
b
i,j,k
(97-e)
G
'
i,j,k
= f
i,j-l,k
c
i,j,k
(97-f)
R
'
i,j,k
= g
i,j-1,k
c
i,j,k
(97-g)
T
'
i,j,k
= a
i,j,k
f
i,j,k-1
(97-h)
U
'
i,j,k
= b
i,j,k
g
i-1,j,k
(97-i)
d
i,j,k
= E
i,j,k
+ A
'
i,j,k
+ T
'
i,j,k
+ C
'
i,j,k
+ G
'
i,j,k
+ U
'
i,j,k
+ R
'
i,j,k
-a
i,j,k
g
i,j,k-1
-b
i,j,k
f
i-1,j,k
- e
i,j-1,k
c
i,j,k
(97-j)
e
i,j,k
= (F
i,j,k
-A
'
i,j,k
-C
'
i,j,k
)/d
i,j,k
(97-k)
f
i,j,k
= (H
i,j,k
-T
'
i,j,k
-G
'
i,j,k
)/d
i,j,k
(97-l)
g
i,j,k
= (S
i,j,k
-R
'
i,j,k
-U
'
i,j,k
)/d
i,j,k
(97-m)
Using these relations to provide the elements of [L] and [U], [A + B] may be
replaced with the product [E][U] in (88) to yield
[L][U]{h
l
-h
l-1
} = {q} - [Al {h
l-1
} (98)
where again the superscript l refers to the current iteration level, and l-l to the
preceding iteration level. We next define the vector (RES
l
) by
{RES
l
} = {q} - [A] {h
l-1
} (99)
Using this notation equation (98) can be written
[L][U]{h
l
-h
l-1
} = {RES
l
} (100)
Equation (100) can now be solved by a process of forward and backward
substitution. The first step involves forward substitution to solve for the vector (v) in
the equation
[L] {v} = {RES
l
} (101)
where {v} = [U] {h
l
-h
l-1
}. The vector (v) determined in this way is then ultilized in a
process of back substitution to solve for the vector (h
l
-h
l-1
) in the equation
[U] {h
l
-h
l-1
} = {v} (102)
In earlier discussions, the coefficients of the equations and hence the elements of the
matrices were identified by the indices of the cells, as shown in figure 53-a. To
illustrate the process of forward substitution, used to calculate the elements of the
vector (v), it is convenient to renumber the equations sequentially using a single
index, as shown in figure 53-b. Because all elements in [L] above the main diagonal
are zero, the first linear equation represented by matrix equation (101) is
d
1
v
1
= RES
1
l
(103)
In equation (103), the term dl has been determined through equation (97-j), and
RES1l has been calculated through equation (99) as an element of the vector (RES
l
);
thus (103) can be solved immediately for the value of v
1
. The second equation
represented by matrix equation (101) is
c
2
v
1
+ d
2
v
2
= RES
2
l
(104)
Figure 53.-Cell numbering schemes for a grid using three indices and using one index.
Again, c
2
and d
2
are known from equations (97), and RES
2
l
is known from equation
(99); using the value of vl from the solution of equation (103), (104) can be solved
for v
2
.
The general equation for an element of {v} has the form
v
n
= (RES
n
l
- a
n
v
n-NRC
- b
n
v
n-NCOL
- c
n
v
n-1
)/d
n
(105)
where NRC is the number of cells in the layer, NCOL is the number of columns in
the model, the coefficeints a
n
,b
n
,c
n
and d
n
are all determined through equations (97)
and RES
n
l
- is determined through equation (99). The terms an and bn are zero for the
first and second eqations ((103) and (104)); and each equation involves elements of
(v) determined earlier in the sequence. This procedure of forward substitution, in
which the elements of (v) are determined in sequence, is possible because of the
lower triangular form of the matrix [L]--i.e., because [L] has only zeros to the right of
the main diagonal.
Back substitution is next used to calculate the elements of the vector (h
l
-h
l-1
) from the
elements of (v), thus solving equation (102). The process of back substitution is
similar to that of forward substitution except that, because the matrix [U] is upper
triangular, the order of calculation is reversed. When the vector {h
l
-h
l-1
} has been
calculated, it is added to the vector (h
l-1
) to obtain (h
l
), the vector of head values
corresponding to iteration l.
In summary, the problem of solving the equation
[A]{h} = {q} (106)
has thus been converted into an iterative process in which: (1) the matrices [L] and
[U] are determined using equations (97); (2) the vector (RESl) is calculated using the
vector (q), the matrix [A] and heads from the preceding iteration; (3) equation (100)
is then solved using forward and backward substitution to obtain the vector {h
l
-h
l-1
}
and (4) the vector (h
l-1
) is added to the vector (h
l
-h
l-l
) to obtain the vector (h
l
).
However, while these are the essential steps of the SIP procedure, several aspects of
the method remain to be discussed.
Transfer of Arrays
As noted previously, the coefficient matrix [A] is sparse, with only seven nonzero
diagonals. Rather than passing the entire matrix to the SIP Package, only the nonzero
diagonals are needed; and because of symmetry of the matrix, only the main diagonal
and the three lower diagonals are needed. The three lower diagonals correspond to
the conductance arrays CC, CR, and CV. The main diagonal is formed from the
three conductance arrays and the array HCOF described in Chapter 2. The right hand
side of the matrix equation, (q), corresponds to the array RHS described in Chapter 2.
The latest estimate of the head distribution (h
l-1
), corresponds to the array HNEW.
As new estimates of head are calculated by SIP, they are stored in HNEW replacing
the previous estimates. Thus input to SIP consists of the following arrays: CC, CR,
CV, RHS, HCOF, and HNEW. Output from SIP consists of a new HNEW array. As
explained in Chapter 3, the Formulate Procedure is inside the iteration loop;
therefore, the input arrays may be modified at each iteration.
Order of Calculation
Experience has shown that if the finite-difference equations are solved in two
different orders on alternate iterations, the number of iterations needed to converge to
a solution is reduced. The order assumed in the discussion, to this point, has been to
begin at the first column, the first row, and the first layer, and to proceed in ascending
column order, ascending row order, and ascending layer order. An alternative is to
start at the first column, the last row, and the last layer, and to proceed in ascending
column order, descending row order, and descending layer order. Using the same
ordering of diagonal names used in figure 51, equations similar to equations (97-a ...
m) can be developed. They are
a
i,j,k
= Z
i,j,k
/(1+ e
i,j,k+1
+ f
i,j,k+1
) (107-a)
b
i,j,k
= B
i,j,k
/(1+e
i+1,j,k
+ g
i+1,j,k
) (107-b)
c
i,j,k
= D
i,j,k
/(1+f
i,j-1,k
+ g
i,j-1,k
) (107-c)
A
'
i,j,k
= a
i,j,k
e
i,j,k+1
(107-d)
C
'
i,j,k
= e
i+1,j,k
b
i,j,k
(107-e)
G
'
i,j,k
= f
i,j-1,k
c
i,j,k
(107-f)
R
'
i,j,k
= g
i,j-1,k
c
i,j,k
(107-g)
T
'
i,j,k
= a
i,j,k
f
i,j,k+1
(107-h)
U
'
i,j,k
= b
i,j,k
g
i+1,j,k
(107-i)
d
i,j,k
= E
i,j,k
+ A
'
i,j,k
+ T
'
i,j,k
+ C
'
i,j,k
+ G
'
i,j,k
+ U
'
i,j,k
+ R
'
i,j,k
- a
i,j,k
g
i,j,k+1
- b
i,j,k
f
i+1,j,k
- e
i,j-1,k
c
i,j,k

(107-j)
e
i,j,k
= (F
i,j,k
- A
'
i,j,k
-C
'
i,j,k
)/d
i,j,k
(107-k)
f
i,j,k
= (H
i,j,k
- T
'
i,j,k
-G
'
i,j,k
)/d
i,j,k
(107-1)
g
i,j,k
= (S
i,j,k
- R
'
i,j,k
- U
'
i,j,k
)/d
i,j,k
(107-m)
In the model described herein, equations (107-a ... m) and equations (97-a ... m) are
in effect invoked alternately in successive iterations. The model program actually
uses one general set of equations in which the variables are identified by single
indices. The ordering of (97) or of (107) is then achieved through the sequence of
values assigned to the indices. In the following list of these general equations, the
index nll refers to the cell in the previous layer calculated, but in the same row and
column as cell n; the indices nrl and ncl are defined analogously. Also, in these
equations, the iteration parameters , , and have each been replaced by a single
parameter as explained in the following section. Note that one additional equation
has been added to the list-the equation for vn, the element of the vector (v)
corresponding to cell n. This equation can be added inasmuch as vn can be calculated
as soon as the nth rows of the matrices [L] and [U] have been calculated. The
equations are
a
n
= Z
n
/(1 + (e
nll
+ f
nll
)) (108-a)
b
n
= B
n
/(1 + (e
nrl
+ g
nrl
)) (108-b)
c
n
= D
n
/(1 + (f
ncl
+ g
ncl
)) (108-c)
A
'
n
= a
n
e
nll
(108-d)
C
'
n
= b
n
e
nrl
(108-e)
G
'
n
= c
n
f
ncl
(108-f)
R
'
n
= c
n
g
ncl
(108-g)
T
'
n
= a
n
f
nll
(108-h)
U
'
n
= b
n
g
nrl
(108-i)
d
n
= E
n
+ (A
'
n
+ T
'
n
+ C
'
n
+ G
'
n
+ U
'
n
+ R
'
n
) -
a
n
g
nll
- b
n
f
nrl
- c
n
e
ncl
(108-j)
e
n
= (F
n
- (A
'
n
+ C
'
n
))/d
n
(108-k)
f
n
= (H
n
- (T
'
n
+ G
'
n
))/d
n
(108-l)
g
n
= (S
n
- (R
'
n
+ U
'
n
))/d
n
(108-m)
v
n
= (RES
n
- a
n
v
nll
- b
n
v
nrl
- c
n
v
ncl
)/d
n
(108-n)
Since the backward substitution requires all values of e
n
, f
n
, g
n
, and v
n
, space is
allocated in the SIP Package for four arrays to store those values. Each of these
arrays has as many elements as there are cells in the grid.
Iteration Parameters
While Weinstein, Stone and Kwan (1969) define three iteration parameters in their
theoretical development, they utilize a single value in practice. Thus the terms , ,
and of equation (93) are replaced by a single parameter, , which multiplies each
term on the right side of the equation; however, must be cycled through a series of
values in successive iterations to achieve satisfactory rates of convergence. In the
model described herein, values of are calculated from the expression
() = 1-(WSEED)^
(-1)
/(NPARM-1) =1,2.... NPARM (109)
where NPARM is the total number of w values to be used; is an index taking on
integral values from 1 to NPARM; () is the corresponding iteration parameter
value; and WSEED is the iteration parameter "seed", calculated according to rules
outlined below, and used as a basis for determining the sequence of w values.
The value of WSEED is in turn developed as follows. The terms pl, P2, and P3 are
calculated for each cell in the mesh using the conductances between that cell and its
neighbors, as follows
P1 = (CC
max
+ CV
max
)/CR
min
(110)
P2 = (CR
max
+ CV
max
)/CC
min
(111)
P3 = (CR
max
+ CC
max
)/CV
min
(112)
where CCmax for a given cell, i,j,k, is the larger of CC
i-1/2,j,k
and CC
i+1/2,j,k
, while
CCmin is the smaller of these values; and similarly CR
max
is the larger of CR
i,j-1/2,k
and CR
i,j+1/2,k
, while CR
min
is the smaller, and CV
max
is the larger of CV
i,j,k-1/2
and
CV
i,j,k+1/2
, while CV
min
is the smaller. Using these values, the terms
2
2
1
2
NCOL
(1+ )
2
2
2
2
NROW
(1+ )
2
2
3
2
NLAY
(1+ )
are computed for each cell in the grid, where again NCOL is the number of columns
in the model, NROW is the number of rows and NLAY is the number of layers; and
the minimun value for each of these three terms is taken to be the cell seed. The seed
term, WSEED, is then taken as average of all the cell seeds. The iteration
parameters, () , generated from the WSEED value are then used sequentially in
successive iterations, recycling each time the entire set has been used (i.e., each
NPARM iterations).
The process described above for calculating the sequence of values differs slightly
from that used by Weinstein, Stone and Kwan (1969), but produces values that are in
the same range and that appear to function well in many problems. However, several
points should be made regarding iteration parameter selection. First, the process is
essentially empirical, and there is little understanding of why one sequence of
parameters performs better than another. Second, the parameters chosen affect the
rate of convergence but (assuming that convergence is achieved) should not influence
the final solution. Third, the influence of the parameters on the rate of convergence
is extremely significant.
The model described herein provides for an additional iteration parameter, referred to
here as the acceleration parameter to distinguish it from . This parameter, which is
designated ACCL in the program, functions as a multiplier of (RES
l
); thus where a
parameter of this type is not to be used, ACCL is simply assigned a value of one.
Parameters similar to ACCL have been used in various versions of SIP (Peaceman,
1977, page 130) although Weinstein, Stone and Kwan (1969) do not employ a
parameter of this type. ACCL is not cycled, but rather is assigned a single value by
the user. As a general rule, it should initially be given a value of one, and
improvement in the rate of convergence should be pursued through adjustment of the
seed term, as explained below. If problems with convergence persist, values of
ACCL other than one can be tried.
Experience has shown that setting the acceleration/relaxation parameter (ACCL) to 1
and using the seed value calculated by the program does not always produce
optimum convergence--that is, the number of iterations required to achieve
convergence is not minimum. Convergence rates will deviate from the optimum if
the absolute value of head change in each iteration is consistently either too small or
too large. When the head change is too large, the computed head overshoots the
correct value, and oscillations occur as the head change repeatedly reverses to
compensate for the overshoot. Severe overshoot causes divergence, while moderate
overshoot simply slows down convergence. When head change is too small, the
opposite problem occurs; head tends to approach the correct value monotonically, but
very slowly. In severe situations, the head changes at each iteration may be so small
that the criterion for convergence is satisfied at all points, even though the computed
heads are still far from the correct values. In such situations, a significant volumetric
budget imbalance will occur.
Weinstein, Stone, and Kwan (1969) suggest that a trial and error method can be used
to improve the choice of seed. This can be done by making an initial run using the
seed calculated by the program or chosen from experience, and using ACCL=1. The
trend of head change per iteration, with increasing iteration number, is observed for
the iterations of a single time step. There is normally some variation in head change
from one iteration to the next due to the cycling of iteration parameters, but this
variation is often superimposed on an overall trend in which head change tends either
to increase or decrease as iterations continue; it is this overall trend (which is often
most evident in the later iterations of the test) that is of interest here. Some
oscillations (reversals in sign) of the computed head change are normal during
convergence; however, repeated oscillation is a sign of overshoot, indicating that
computed head changes are too great for optimal convergence. Head changes which
are too small, on the other hand, are indicated by a very flat overall trend. For proper
evaluation of the trend, the trial should generally be run for a number of iterations
equal to 4 or 5 times the number of iteration parameters, unless convergence occurs
before this.
Following the initial trial, the seed is multiplied by a number between two and ten if
head changes in the initial trial appear to be too great, and divided by a number
between two and ten if those head changes appear to be too small. If the trend in the
initial trial is unclear either multiplication or division of the seed may be tried. In any
case, a second run is made using the new seed value, and the trend of head change vs.
iteration level is again examined. The results are compared with those of the initial
trial to see if the rate of convergence has improved. If both runs have converged, the
comparison is based on the number of iterations required for convergence; if they
have failed to converge, the comparison is based on the head changes observed in the
final iterations.
The trial runs can be continued to further refine the choice of seed; in general the
seed value will be multiplied or divided by progressively smaller numbers at each
step of the procedure. However, it is usually not worthwhile to carry the process too
far; multiplication or division of the seed by factors less than 2 is seldom warranted.
In most cases, a satisfactory seed value developed by this procedure will remain
satisfactory even though changes in the model are introduced-for example, additional
stresses, modifications in boundary conditions or changes in the model mesh.
However, if convergence problems arise after such changes, the trial and error
procedure can be repeated. It should be noted that the more strongly diagonal the
coefficient matrix, the less important the choice of seed will be. Thus, source terms
such as evapotranspiration or stream seepage, which affect only coefficients on the
main diagonal, normally tend to make the choice of seed less critical; and the
addition of such terms to a model seldom necessitates modification of the seed.
For each iteration, the program stores the value of |h
l
|
max
, where |h
l
|

refers to the
absolute value of the head change computed during iteration l at a given node, and
|h
l
|
max
is the maximum such absolute value for the entire mesh in that iteration. For
the last time step of each stress period, a table of |h
l
|
max
values for each iteration of
the time step is printed in the program output; at the user's option, a table of these
values may be printed for every time step, or for time steps which fall at specified
intervals. (See Narrative for Module SIP1AP and Sample Problem Output in
Appendix D.) In addition to the |h
l
|
max
value, each entry in the table shows the
indices of the node at which the maximum change was recorded, and a sign
indicating whether the change was positive or negative. This table can be used in the
head change trend evaluations described above, under the assumption that the
behavior of the |h
l
|
max
values is representative of the behavior of head change
throughout the mesh.
Improvements in the rate of convergence can also be obtained by adjusting the
acceleration parameter, ACCL. Increases in ACCL will cause increases in the head
change at each iteration, while decreases in ACCL will cause decreases in head
change. The trial procedure described above can be used for this case as well;
however changes in the seed and in ACCL should never be attempted in the same set
of trial runs.
It is sometimes necessary to slow the process of convergence in order to prevent cells
from converting to the no-flow condition as a result of head overshoot during an
iteration. In these situations, optimal convergence cannot be considered convergence
in the minimum number of iterations, but rather convergence in the smallest number
of iterations that does not involve head overshoot. The procedure of examining head
change per iteration and adjusting iteration parameters can again be used to
determine when this condition is being met, and to develop the required seed or
ACCL terms.
Strongly Implicit Procedure Package Input
Input to the Strongly Implicit Procedure (SIP) Package is read from the unit specified
in IUNIT(9).
FOR EACH SIMULATION
SIP1AL
1. Data: MXITER NPARM
Format: I10 I10
SlPlRP
2.Data: ACCL HCLOSE IPCALCWSEEDIPRSIP
Format: F10.0 F10.0 I10 F10.0 I10
MXITER--is the maximum number of times through the iteration loop in one time
step in an attempt to solve the system of finite-difference equations. Fifty iterations
are generally sufficient.
NPARM--is the number of iteration parameters to be used. Five parameters are
generally sufficient.
ACCL--is the acceleration parameter. It must be greater than zero and is generally
equal to one. If a zero is entered, it is changed to one.
HCLOSE--is the head change criterion for convergence. When the maximum
absolute value of head change from all nodes during an iteration is less than or equal
to HCLOSE, iteration stops.
IPCALC--is a flag indicating where the iteration parameter seed will come from.
0 - the seed will be entered by the user.
1 - the seed will be calculated at the start of the simulation from problem
parameters.
WSEED--is the seed for calculating iteration parameters. It is only specified if
IPCALC is equal to zero.
IPRSIP--is the printout interval for SIP. If IPRSIP is equal to zero, it is changed to
999. The maximum head change (positive or negative) is printed for each iteration of
a time step whenever the time step is an even multiple of IPRSIP. This printout also
occurs at the end of each stress period regardless of the value of IPRSIP.
Electronic Manual for MODFLOW 13 SOR Solver Package 159
13 SOR Solver Package
Successive overrelaxation is another method for solving large systems of linear
equations by means of iteration. It is implemented in the model discussed herein
through the Slice Successive Overrelaxation (SSOR) Package. Background material
on the successive overrelaxation approach can be found in many standard references,
including those already noted by Peaceman (1977), Crichlow (1977) and Remson,
Hornberger and Molz (1971).
The successive overrelaxation technique is implemented in the SSOR Package by
dividing the finite difference grid into vertical "slices," as shown in figure 54, and
grouping the node equations into discrete sets, each set corresponding to a slice. In
every iteration, these sets of equations are processed in turn, resulting in a new set of
estimated head values for each slice. As the equations for each slice are processed,
they are first expressed in terms of the change in computed head between successive
iterations. The set of equations corresponding to the slice is then solved directly by
Gaussian elimination, treating the terms for adjacent slices as known quantities (that
is, inserting the most recently computed values of head for the adjacent slices as
"known" values in the equations for the slice being processed). The values of head
change computed for the slice in this Gaussian elimination process are then each
multiplied by an acceleration parameter, , generally taken between 1 and 2; the
results are taken as the final values of head change in that iteration for the slice.
They are added to the respective head values from the preceding iteration to obtain
the final estimates of head for the iteration, for that slice. This procedure is repeated
for each slice in sequence until all of the slices in the three-dimensional array have
been processed, thus completing a single iteration. The entire sequence is then
repeated, in successive passes through the series of slices, until the differences
between the head values computed in successive iterations is less than the closure
criterion at all nodes in the mesh.
160 13 SOR Solver Package Electronic Manual for MODFLOW
Figure 54.-Division of the three-dimensional model array into vertical slices for processing in the SSOR package.
It should be noted that even though a direct method of solution (Gaussian
elimination) is used within each iteration to process the equations for each individual
slice, the overall solution procedure is not direct but iterative. Each direct solution
produces only interim values or estimates of head change based on the most recently
computed heads in adjacent slices; as successive slices are processed, the computed
values continue to change until closure is achieved.
The process of solution described above can be illustrated in more detail through
consideration of the node equations. The equation of flow for an individual cell, as
developed in chapter 2, is reproduced below with the addition of a second superscript
to indicate iteration level
CV
i,j,k-1/2
h
m,l
i,j,k-1
+ CC
i-1/2,j,k
h
m,l
i-1,j,k
+ CR
i,j-1/2,k
h
m,l
i,j-1,k
+ (- CV
i,j,k-1/2
- CC
i-1/2,j,k
- CR
i,j-1/2,k
- CR
i,j+1/2,k
- CC
i+1/2,j,k
- CV
i,j,k+1/2
+ HCOF
i,j,k
) h
m,l
i,j,k
+ CR
i,j+1/2,k
h
m,l
i,j+1,k
+ CC
i+1/2,j,k
h
m,l
i+1,j,k
+ CV
i,j,k+1/2
h
m,l
i,j,k+1
= RHS
i,j,k
(113)
In equation (113), the superscript m refers to the time step, while the superscript l
refers to the iteration level. If an equation of the form of (113) is written for the
following iteration level, l+1, and the left side of equation (113) is then subtracted
from each side of the new equation, the result can be written as
CV
i,j,k-1/2
(h
m,l+1
i,j,k-1
- h
m,l
i,j,k-1
) + CC
i-1/2,j,k
(h
m,l+1
i-1,j,k
- h
m,l
i-1,j,k
)
+ CR
i,j-1/2,k
(h
m,l+1
i,j-1,k
- h
m,l
i,j-1,k
) + (- CV
i,j,k-1/2
- CC
i-1/2,j,k
- CR
i,j-1/2,k
- CR
i,j+1/2,k
- CC
i+1/2,j,k
- CV
i,j,k+1/2
+ HCOF
i,j,k
) (h
m,l+1
i,j,k
- h
m,l
i,j,k
) + CR
i,j+1/2,k
(h
m,l+1
i,j+1,k
- h
m,l
i,j+1,k
)
+ CC
i+1/2,j,k
(h
m,l+1
i+1,j,k
- h
m,l
i+1,j,k
) + CV
i,j,k+1/2
(h
m,l+1
i,j,k+1
- h
m,l
i,j,k+1
) =
RHS
i,j,k
- CV
i,j,k-1/2
h
m,l
i,j,k-1
- CC
i-1/2,j,k
h
m,l
i-1,j,k
- CR
i,j-1/2,k
h
m,l
i,j-1,k
- (- CV
i,j,k-1/2
- CC
i-1/2,j,k
- CR
i,j-1/2,k
- CR
i,j+1/2,k
- CC
i+1/2,j,k
- CV
i,j,k+1/2
+ HCOF
i,j,k
) h
m,l
i,j,k
- CR
i,j+1/2,k
h
m,l
i,j+1,k
- CC
i+1/2,j,k
h
m,l
i+1,j,k
- CV
i,j,k+1/2
h
m,l
i,j,k+1
(114)
In equation (114) the unknown terms are taken as the changes in computed head
between iteration l and iteration l+1--for example, (h
m,l+1
i,j,k
- h
m,l
i,j,k
). Note that when
the l-th iteration has been completed, the right hand side of (114) consists entirely of
known terms--it includes the RHS and conductance terms assembled in the
formulation process, and estimates of head already obtained during iteration l.
Now suppose that we divide the mesh into vertical slices taken along rows, as shown
in figure 54, and isolate the equations associated with the nodes of an individual
slice--for example, slice 4 of figure 54, which is taken along row 4 of the three
dimensional array. In terms of equation (114), if we are processing slice i,
corresponding to row i, we retain the head changes at nodes within this slice as
unknown terms, but consider the head changes at nodes in the two adjacent slices to
be known values. Thus the terms CC
i-1/2,j,k
(h
m,l+1
i-1,j,k
- h
m,l
i-1,j,k
) and CC
i+1/2,j,k
(h
m,l+1
i+1,j,k
- h
m,l
i+1,j,k
), on the left side of equation (114), are treated as known
quantities. If we move these two expressions to the right side of the equation and
rearrange, we find that the terms in h
m,l
i-1,j,k
and h
m,l
i+1,j,k
drop out, leaving
CV
i,j,k-1/2
(h
m,l+1
i,j,k-1
- h
m,l
i,j,k-1
) + CR
i,j-1/2,k
(h
m,l+1
i,j-1,k
- h
m,l
i,j-1,k
)
+ (- CV
i,j,k-1/2
- CC
i-1/2,j,k
- CR
i,j-1/2,k
- CR
i,j+1/2,k
- CC
i+1/2,j,k
- CV
i,j,k+1/2
+ HCOF
i,j,k
) (h
m,l+1
i,j,k
- h
m,l
i,j,k
)
+ CR
i,j+1/2,k
(h
m,l+1
i,j+1,k
- h
m,l
i,j+1,k
) + CV
i,j,k+1/2
(h
m,l+1
i,j,k+1
- h
m,l
i,j,k+1
) =
RHS
i,j,k
- CV
i,j,k-1/2
h
m,l
i,j,k-1
- CC
i-1/2,j,k
h
m,l
i-1,j,k
- CR
i,j-1/2,k
h
m,l
i,j-1,k
- (- CV
i,j,k-1/2
- CC
i-1/2,j,k
- CR
i,j-1/2,k
- CR
i,j+1/2,k
- CC
i+1/2,j,k
- CV
i,j,k+1/2
+ HCOF
i,j,k
) h
m,l
i,j,k
- CR
i,j+1/2,k
h
m,l
i,j+1,k
- CC
i+1/2,j,k
h
m,l
i+1,j,k
- CV
i,j,k+1/2
h
m,l
i,j,k+1
(115)
Now suppose the slices are processed in the order of increasing row number, i; then
calculations for slice i-1 will be completed in each iteration before calculations for
slice i are initiated. It follows that a value of h
m,l+1
i-1,j,k
will be available when the
processing of slice i is initiated in iteration l+1, whereas a value of h
m,l+1
i+1,j,k
will not
be available. Thus the term CC
i-1/2,j,k
h
m,l+1
i-1,j,k
can be incorporated directly as a
known term in the processing of slice i, but the term CC
i+1/2,j,k
h
m,l+1
i+1,j,k
cannot. To
circumvent this difficulty, the value of h
m
i+1,j,k
from the preceding iteration, h
m,l
i+1,j,k
,
is substituted for h
m,l+1
i+1,j,k
on the right side of (115). (Thus in effect we are using the
most recently calculated value of head for each adjacent slice.) The resulting
equation is
CV
i,j,k-1/2
(h
~m,l+1
i,j,k-1
- h
m,l
i,j,k-1
) + CR
i,j-1/2,k
(h
~m,l+1
i,j-1,k
- h
m,l
i,j-1,k
)
+ (- CV
i,j,k-1/2
- CC
i-1/2,j,k
- CR
i,j-1/2,k
- CR
i,j+1/2,k
- CC
i+1/2,j,k
- CV
i,j,k+1/2
+ HCOF
i,j,k
) (h
~m,l+1
i,j,k
- h
m,l
i,j,k
)
+ CR
i,j+1/2,k
(h
~m,l+1
i,j+1,k
- h
m,l
i,j+1,k
) + CV
i,j,k+1/2
(h
~m,l+1
i,j,k+1
- h
m,l
i,j,k+1
) =
RHS
i,j,k
- CV
i,j,k-1/2
h
m,l
i,j,k-1
- CC
i-1/2,j,k
h
m,l
i-1,j,k
- CR
i,j-1/2,k
h
m,l
i,j-1,k
- (- CV
i,j,k-1/2
- CC
i-1/2,j,k
- CR
i,j-1/2,k
- CR
i,j+1/2,k
- CC
i+1/2,j,k
- CV
i,j,k+1/2
+ HCOF
i,j,k
) h
m,l
i,j,k
- CR
i,j+1/2,k
h
m,l
i,j+1,k
- CC
i+1/2,j,k
h
m,l
i+1,j,k
- CV
i,j,k+1/2
h
m,l
i,j,k+1
(116)
In equation (116), the notation h
~
has been introduced for the head terms in slice i at
iteration l+1. The purpose of this notation will become clear as the solution process
is described. The number of nodes in the slice is NC*NL, where NC is the number of
columns in the model and NL the number of layers; and an equation of the form of
(116) is formulated at each node. Thus a system of NC*NL equations in NC*NL
unknowns is established. Because the number of layers is usually small, the total
number of equations is generally small enough so that direct solution by Gaussian
elimination is an efficient approach (note that such a procedure would generally not
be feasible for the larger set of equations associated with the entire three-dimensional
model array.)
The set of equations associated with an individual slice, i, can be written in matrix
form as
[A]
i
{h
~
}
i
= {R}
i
(117)
where [A]
i
is the coefficient matrix for slice i; {h
~
}
i
is a vector of estimates,
h
~m,l+1
i,j,k
- h
m,l
i,j,k
for the change in computed head at each node in the slice between
iteration l and iteration l+1; and {R}
i
is the vector of "constant" terms, representing
the right side of equation (116), for slice i.
The Gaussian elimination procedure applied to the matrix equations (117) yields one
value of the term (h
~m,l+1
i,j,k
- h
m,l
i,j,k
) for each node in the slice. These terms are taken
as first estimates for the change in computed head from iteration l to iteration l+1.
Each is multiplied by the acceleration parameter, , and each result is added to the
corresponding head from the preceding iteration to obtain the final estimate of head
for iteration l+1; that is,
h
m,l+1
i,j,k
= h
m,l
i,j,k
+ (h
~m,l+1
i,j,k
- h
m,l
i,j,k
) (118)
When values of h
m,l+1
i,j,k
have been computed for each node (j,k) in slice i, the
procedure of calculation is initiated for the succeeding slice, i+1. When all slices
have been processed the iteration is complete, and calculations are initiated for the
next iteration unless closure has been achieved.
As illustrated in figure 55-a, the matrix of coefficients [A]
i
of equation (117) is
symmetric and banded, with a maximum half-bandwidth equal to the number of
layers. Because of the symmetry of the matrix, only the lower triangular portion has
to be stored; this storage is provided in the program in a two-dimensional array, as
illustrated in figure 55-b, with dimensions NL*NC and NL+1. In this example,
NL=NC=3.
Adjustment of the acceleration parameter is frequently necessary in SSOR to achieve
optimal rates of convergence. For this purpose, methods similar to the trial and error
procedure described in Chapter 12, for adjustment of the SIP "seed" value can be
applied.
Figure 55.-Coefficient matrix for slice equations and corresponding computer storage array.
Slice-Successive Overrelaxation Package Input
Input to the Slice-Successive Overrelaxation (SOR) Package is read from the unit
specified in IUNIT(11).
FOR EACH SIMULATION
SOR1AL
1. Data: MXITER
Format: I10
SOR1RP
2. Data: ACCL HCLOSE IPRSOR
Format: F10.0 F10.0 I10
MXITER--is the maximum number of iterations allowed in a time step.
ACCL--is the acceleration parameter, usually between 1.0 and 2.0.
HCLOSE--is the head change criterion for convergence. When the maximum
absolute value of head change from all nodes during an iteration is less than or equal
to HCLOSE, iteration stops.
IPRSOR--is the printout interval for SOR. IF IPRSOR is equal to zero, it is changed
to 999. The maximum head change (positive or negative) is printed for each iteration
of a time step whenever the time step is an even multiple of IPRSOR. This printout
also occurs at the end of each stress period regardless of the value of IPRSOR.
Electronic Manual for MODFLOW 14 Utility Modules 165
14 Utility Modules
Introduction
Utility modules are those submodules which perform general tasks common to
several different packages. The name of a utility module always consists of a "U"
followed by a five-character mnemonic. There are eight utility modules:
UBUDSV - Writes an unformatted record consisting of an array with one real
number for each cell in the grid.
ULASAV - Writes an unformatted record consisting of an array with one real
number for each cell in a layer.
ULAPRS and ULAPRW - Prints one two-dimensional array which contains and one
real number for each cell in a layer. ULAPRS prints, in strip form, the first N
columns (where N is the number of values that can fit on one print line) of each row
and then the next N columns, etc., until all columns of each row are printed.
ULAPRW prints, in wrap form, all of row 1, all of row 2, and all of row 3, etc. The
format for printing arrays is shown in table 2.
UCOLNO - Prints column numbers at the top of each page of data printed by
ULAPRS and ULAPRW.
U2DREL - Reads a two-dimensional array of real numbers.
U2DINT - Reads a two-dimensional array of integers.
U1DREL - Reads a one-dimensional array of real numbers.
166 14 Utility Modules Electronic Manual for MODFLOW
Table 2.--Print-format codes for utility modules ULAPRS and ULAPRW.
IPRN FORMAT
1 11G10.3
2 9G13.6
3 15F7.1
4 15F7.2
5 15F7.3
6 15F7.4
7 20F5.0
8 20F5.1
9 20F5.2
10 20F5.3
11 20F5.4
12 10G11.4
Electronic Manual for MODFLOW 14 Utility Modules 167
Input Instructions For Array Reading utility Modules
The real two-dimensional array reader (U2DREL), the integer two-dimensional array
reader (U2DINT), and the real one-dimensional array reader (U1DREL) read one
array-control record and, optionally, a data array in a format specified on the array-
control record. The control record is read from the input unit number specified for
the major option that is requesting the array. For example, the Recharge Package
uses U2DREL to read the RECH array. The input unit for the recharge option is
contained in IUNIT (8), and accordingly, the RECH array control record is read on
this input unit.
FOR REAL ARRAY READER (U2DREL or U1DREL)
Data: LOCAT CNSTNT FMTIN IPRN
Format: I10 F10.0 5A4 I10
FOR INTEGER ARRAY READER (U2DINT)
Data: LOCAT ICONST FMTIN IPRN
Format: I10 I10 5A4 I10
LOCAT--indicates the location of the data which will be put in the array.
If LOCAT < 0, the sign is reversed to give the unit number from which an
unformatted record will be read.
If LOCAT = 0, every element in the array will be set equal to the value
CNSTNT/ICONST.
If LOCAT > 0, it is the unit number from which data values will be read
in the format specified in the third field of the array-control record
(FMTIN).
CNSTNT/ICONST--is a constant. Its use depends on the value of LOCAT.
If LOCAT = 0, every element in the array is set equal to
CNSTNT/ICONST.
If LOCAT 0, and if CNSTNT/ICONST ~ 0, every element in the array is
multiplied by CNSTNT/ICONST.
FMTIN--is the format of records containing the array values. It is used only if the
first field in the array-control record (LOCAT) contains a positive number. The
format must be enclosed in parentheses; for example, (15F5.0) for real data and
(15I5) for integer data.
IPRN--is a flag indicating that the array being read should be printed and a code for
indicating the format that should be used. It is used only if LOCAT is not equal to
zero. The format codes are different for each of the three modules. IPRN is set to
zero when the specified value exceeds those defined in the chart below. If IPRN is
less than zero, the array will not be printed.
168 14 Utility Modules Electronic Manual for MODFLOW
IPRN U2DREL U2DINT U1DREL
0 10G11.4 10I11 10G12.5
1 11G10.3 60I1
2 9G13.6 40I2
3 15F7.1 30I3
4 15F7.2 25I4
5 15F7.3 20I5
6 15F7.4
7 20F5.0
8 20F5.1
9 20F5.2
10 20F5.3
11 20F5.4
12 10G11.4
Electronic Manual for MODFLOW References 169
References
References
Collins, R. E., 1961, Flow of fluids through porous materials; New York, Reinhold
Publishing Corp., 270 p.
Crichlow, Henry B., 1977, Modern reservoir engineering - A simulation approach;
Englewood Cliffs, N.J., Prentice Hall Inc., 354 p.
McDonald, M.G., and Harbaugh, A.W., 1984, A modular three-dimensional finite-
difference ground-water flow model: U.S. Geological Survey Open-File Report 83-
875 , 528 p.
Peaceman, Donald W., 1977, Fundamentals of numerical reservoir simulation; New
York, Elsevier Scientific Publishing Company, 176 p.
Remson, Irwin, Hornberger, George M. and Molz, Fred J., 1971, Numerical methods
in subsurface hydrology; New York, Wiley-Interscience, 389 p.
Rushton, K. R., Redshaw S. C., 1979, Seepage and groundwater flow-numerical
analysis by analog and digital methods; New York, John Wiley and Sons
Trescott, Peter C., 1975, Documentation of finite-difference model for simulation of
three-dimensional ground-water flow: U.S. Geological Survey Open-File Report 75-
438, 32 p.
Trescott, Peter C. and Larson, S. P., 1976, Supplement to Open-File Report 75-438,
Documentation of finite-difference model of three-dimensional ground-water flow;
U.S. Geological Survey Open-File Report 76-591, 21 p
170 References Electronic Manual for MODFLOW
Trescott, P. C., Pinder, G. F., and Larson, S. P., 1976, Finitedifference model for
aquifer simulation in two dimensions with results of numerical experiments: U.S.
Geological Survey Techniques of Water-Resources Investigations, Book 7, Chapter
C1, 116 p.
Weinstein, H. C., Stone, H. L., and Kwan, T. V., 1969, Iterative procedure for
solution of systems of parabolic and elliptic equations in three dimensions: Indus.
Engineering Chemistry Fundamentals, v. 8, no. 2, p. 281-287.
Electronic Manual for MODFLOW Appendix A - Program Portability 171
Appendix A - Program Portability
Introduction
One of the major design requirements for the model program was that it should be
portable. A portable program is one that can be run with a minimum of modification
on most computers that are physically capable of running a program of its type. The
goal of portability for the model program has been attained as evidenced by the fact
that it has run successfully in either its present form or the earlier form (McDonald
and Harbaugh, 1984) on computers manufactured by many companies including
mainframe computers, minicomputers, and microcomputers. The following
discussion explains in more detail the concept of portability, what was done to
maximize portability of the model program, and circumstances that might require that
the program be changed in order to run successfully on a particular computer.
The Impact of the Programming Language on Portability
The programming language is the most important factor that determines program
portability. There are a variety of programming languages available, and for each
language, there are numerous versions which have resulted from the desire of
vendors to improve the power of the language and to take advantage of the hardware
features of their particular computers. The most commonly available language
suitable to use for the model program is FORTRAN. There are two versions defined
by the American National Standards Institute (ANSI) on which most commercial
versions are based, ANSI X3.9-1966 and ANSI X3.9-1978.
1
These versions are
commonly referred to as FORTRAN 66 and FORTRAN 77, respectively.
FORTRAN 66 was selected for the original version of the model because, at the
time# it was far more widely supported than was FORTRAN 77. Now that the
FORTRAN 77 standard is more widely supported the program has been converted to
this standard.
The program in this report is nearly identical to the original program except for the
changes required to make it comply with the FORTRAN 77 standard. These changes

1
American National Standards Institute, 1966, FORTRAN: American National Standards Institute, X3.9-1966, 36 p.
American National Standards Institute, 1978, Programming Language FORTRAN: ANSI, X3.9-1978, chs. 1-18.
172 Appendix A - Program Portability Electronic Manual for MODFLOW
are minor and are described by McDonald and Harbaugh (1984, p. 505). The
conversion was done solely for the purpose of making the program comply with the
FORTRAN 77 standard; many of the features that make FORTRAN 77 more
powerful than FORTRAN 66 were not used. An effort to make more extensive
FORTRAN 77 revisions to the program is judged to be uneconomical. Such an effort
would not result in significant improvements in program clarity or efficiency.
Most commercial versions of FORTRAN 77 compilers include some extended
features not defined as part of the FORTRAN 77 standard. Such features vary widely
among computer vendors and were not used in the model program. The program
contains only one exception to complete compliance to the FORTRAN 77 standard
that the authors are aware of. This exception# which is explained in a following
section (see The Impact of Allocating Array Storage in a Single Array on
Portability), is commonly allowed on computers and should not be a major restriction
to portability. Because the program closely follows the FORTRAN 77 standard, the
program should work on any computer supporting this language provided that the
computer has adequate computational power.
It is recognized that some users may want the program converted back into the
FORTRAN 66 language because they have access only to older compilers. Only a
few changes are required to convert the program back to FORTRAN 66. Information
about how to convert is provided in a following section (see Conversion to
FORTRAN 66).
The Impact of Computational Precision on Portability
Variation of precision among computers causes some problems with program
portability. Computational precision refers to the accuracy at which numbers are
calculated and stored in the computer. To prevent the imposition of constraints on
the computers on which FORTRAN is implemented the computational precision was
not defined as part of the standards. The accuracy of model results are dependent on
the computational precision, so precision must be considered when moving the model
program among computers.
The model program was developed on computers using 32 binary bits to represent
single precision real numbers. This gives from 6 to 7 decimal digits of precision and
includes values that range in magnitude from approximately 10**-39 to 10**38.
Double precision real numbers are represented by 64 binary bits and range in
magnitude from approximately 10**-10000 to 10**10000 with 14 to 15 decimal
digits of precision. The head array, HNEW, and some variables in the solvers are
stored as double precision, and accordingly many calculations in the solvers are
double precision. This was necessary for accuracy under some conditions. The
model program should perform adequately for most problems on computers that use
32 bits for single precision and 64 bits for double precision. However, the required
precision depends on the problem being simulated. Thus.. the user must ultimately
determine if adequate precision is being used for solving a particular problem.
There are some situations for which there is a need to modify the program to make all
real number calculations in double precision. If using a computer that represents
single precision real numbers with less than 32 bits, then double precision is probably
necessary for all real numbers and calculations. Even on computers that represent
single precision numbers with 32 bits, certain problems are difficult to solve without
making all real numbers and calculations double precision. Unfortunately, it is
difficult to predict if a specific problem requires all double precision. Simulations
with very large numbers of cells, for example more than 50000, are more likely to
have precision problems than are smaller simulations. Simulations in which there are
areas having significant ground-water flow and yet the heads in the adjacent model
cells in these areas are equal within .01 percent or less are also more likely to have
precision problems. In addition# precision problems depend on the preciseness of
the attempted solution. In general, the symptoms of inadequate precision are either a
poor volumetric flow balance or lack of convergence by the solver. However, these
same symptoms are more commonly caused by bad Input data or improper
adjustment of parameters that control Iteration. Because precision problems are in
general fairly unlikely and use of all double precision results in increased memory
and computer time usage, conversion to double precision should probably be done as
a last resort in order to solve convergence problems. If the program is converted to
all double precision# almost twice the computer memory is required for model data.
To convert all real numbers and calculations to double precision,, do the following:
1. Declare all single precision real arrays and variables as DOUBLE PRECISION in
the main program and in every subroutine. This can be done by adding the statement
IMPLICIT DOUBLE PRECISION (A-H,O-Z)
to the beginning of the specifications section of every subroutine and the main
program. Alternately* some compilers have a special command that will accomplish
this without the need to modify the program itself.
2. In the Basic Package Allocate module (subroutine BAS1AL) change the statement
ISUM=ISUM+2*NRCL
to
ISUM=ISUM+NRCL
3. Change the real intrinsic functions in subroutine SSIP1I to their double precision
equivalents. Specifically, change all occurrences of AMAX1 to DMAXl, and
AMIN1 to DMIN1.
If using a computer on which significantly more than 32 bits are used to represent
single precision real numbers, then the partial use of double precision data and
calculations as included In the present model program may be unnecessary. The
program should still work on such computers without modification, but
computational time and memory would be saved if double precision were eliminated.
The program was designed so that changing to all single precision would be easy.
For this reason# the use of double precision intrinsic functions and constants is
avoided, and all conversions between real and double precision are done by implied
type changes in assignment statements. If it is determined that the use of all single
precision is acceptable,, make the following changes to the program:
1. Delete all DOUBLE PRECISION specification statements.
2. In the Basic Package Allocate module (subroutine BAS1AL) change the statement
ISUM=ISUM+2*NRCL
to
ISUM=ISUM+NRCL
The model program was developed on computers that use 32 bits for integer numbers
and calculations. This is more than adequate to represent the range of integer
numbers that are used in the program under any conditions. The largest integer that
the computer must be able to represent is the dimensioned size of the X array in the
main program. Many FORTRAN compilers allow one to specify that integers be
represented by 16 bits, which does not provide enough precision for larger model
simulations. That is, a 16 bit integer in a typical computer can represent numbers in
the range -32768 to 32767, and an X array dimension of 32767 is adequate only for
simulations that have 2200 to 3000 cells, depending on what options are used.
The Impact of Allocating Array Storage in a Single Array
on Portability
Because almost all model data are stored in the X array, which is dimensioned in the
main program, this array can be quite large. It is generally 10 to 15 times the size of
the number of model cells. Some computers require that special compiler options be
used when an array exceeds a specified size. Users are cautioned to be aware of this
and use the appropriate options as needed. The result of using the incorrect option
can sometimes be that the program will execute without producing error messages,
but answers will be incorrect.
All computers limit the amount of array space that a program can use. If a user's
simulation exceeds this limit,, the only options may be to use a different computer or
make the simulation smaller. However, some computers make a distinction between
total amount of memory used for all arrays and the total used by a single array. That
is, some computers might allow 4 arrays each having 32000 elements, but not allow a
single array consisting of 128000 elements. On such a computer, it is quite possible
that one could exceed the size limit for a single array without exceeding the total
array size limit because nearly all model data are stored in the single X array. If this
situation occurs, it is possible to break the X array into smaller pieces. Although this
can be done by modifying only the main program# the modification is fairly complex.
Such modification requires a knowledgeable programmer who has a clear
understanding of how model data are stored within the X array. Before making such
a modification, it would be prudent to assess how long the desired simulation might
take to execute. Generally, computer execution speed is more of a constraint on
maximum problem size than is array size.
The only known exception to the use of the FORTRAN 77 standard in the model
program is that the data type of actual subroutine arguments does not always match
the type of the corresponding dummy arguments. The standard requires the data type
of actual and dummy arguments to match. This only happens in the main program
where it calls subroutines using actual arguments that are elements from the X array.
The X array itself is data type real, and several model arrays stored withi X are either
integer (arrays IOFLG, IBOUND, IRCH, IEVT, and LRCH) or double precision
(array HNEW). For example, actual argument X(LCIRCH) in the main program is
passed to dummy argument IRCH in subroutine RCHIRP. X(LCIRCH) is of type
real, and IRCH is of type integer. This practice hasnot been a problem in the past.
Should this become a problem on future computers, required changes will be fairly
minor, affecting only the main program and some of the Allocate modules.
The Impact on Portability of Preconnected File Units
FORTRAN 77 provides 2 ways for a file unit to become connected to (associated
with) a file -- preconnection and the OPEN statement. The model program use
preconnection. This means that the computer's operating system provides the
connection between files and file units prior to program execution. Often the user
must issue commands to the system, which connect the necessary files and file units,
prior to running the program. The specific method for doing this varies among
computers. Generally, preconnection is adequate# but it may be inconvenient on
some computers. Modification of the main program to use OPEN statements is a
simple task for a programmer.
Conversion to FORTRAN 66
Because the model program uses only one feature of FORTRAN 77 that is not part of
FORTRAN 66, few changes are needed in order to make the program comply with
the FORTRAN 66 standard. All of the required changes are a result of differences in
the way character (alphanumeric) data are handled by the two versions of
FORTRAN. Any variables or arrays holding character data must be declared to be
the character data type in a FORTRAN 77 program; in a FORTRAN 66 program,
character data are stored in numeric variables or arrays. The specific changes that
are required to make the model program comply with the FORTRAN 66 standard are
shown below. It is assumed that at least four characters can be stored In a single
precision real variable or array element.
1. Delete all CHARACTER*4 statements throughout all subroutines and the main
program.
2. In each of the array reading utility modules (U1DREL, U2DREL, and U2DINT),
change the statement
CHARACTER*20 FMTIN
to
DIMENSION FMTIN(5)
3. In each of the array reading utility modules (U1DREL, U2DREL, and U2DINT),
change all occurrences of "A20" to "5AM". For example# change
1 ' USING FORMAT: ',A20/30X,79('-')/)
in subroutine U1DREL to
1 ' USING FORMAT: ',5M/30X,79('-')/)
4. To make the program strictly comply with the FORTRAN 66 standard, it is
necessary to change character constants in DATA statements to Hollerith constants.
However,, most FORTRAN 66 compilers accept character constants specified using
FORTRAN 77 notation (using apostrophes). Thus, it is unlikely that this change will
be required.
Electronic Manual for MODFLOW Appendix B - Space Requirements 177
Appendix B - Space
Requirements
X Array Memory Requirements
The outline below gives the X-array space requirements for each package. The
formulas can be used to calculate the exact size of the X array for a given problem.
However, it is generally easier to simply run the model program and let it calculate
the size. The total space required is printed as part of the model printout even if the
X array is dimensioned too small. As a rough estimate, the X array is approximately
10 to 15 times the number of nodes in the model depending on the options selected.
In the outline below, NODES is defined as NCOL*NROW*NLAY, the number of
nodes in the model.
I. BAS Package
A. 8*NODES + (NLAY - 1)*NCOL*NROW + NROW + NCOL + 4*NLAY
B. Additionally, add NODES if start head is saved (ISTRT is not 0), and add
NODES if BUFF is separate from RHS (IAPART is not 0)
II. BCF Package
A. NLAY
B. If a transient simulation (ISS is 0), add NODES
C. For each layer where LAYCON is one or three, add 2*NCOL*NROW
D. For each layer where LAYCON is two or three, add NCOL*NROW
E. If a transient simulation (ISS is 0), for each layer where LAYCON is two or three,
add NCOL*NROW
III. WEL Package 4*MXWELL
IV. DRN Package 5*MXDRAN
178 Appendix B - Space Requirements Electronic Manual for MODFLOW
V. RIV Package 6*MXRIVR
VI. EVT Package
A. Option 1 3*NCOL*NROW
B. Option 2 4*NCOL*NROW
VII. GHB Package 5*MXBND
VIII. RCH Package
A. Options 1 and 3 -- NCOL*NROW
B. Option 2 2*NCOL*NROW
IX. SIP Package 4*NODES + 4*MXITER + NPARM
X. SOR Package (NLAY + 4)*NCOL*NLAY + 4*MXITER
Generally, it is advisable to have the X-array dimension relatively close to the
amount of space needed for a specific problem. On the other hand, it is inconvenient
to redimension the X array every time a new option is selected. Several load
modules with the size of the X array differing by a factor of two should be adequate.
Electronic Manual for MODFLOW Appendix C - Continuing a Run 179
Appendix C - Continuing a Run
Procedure
There is often value in breaking long simulations into several short model runs. This
allows one to decide, between runs, whether or not to continue the simulation.
Although the model program in this report does not have a special option for making
continuation runs, it is quite simple to continue a simulation by using the output of
one run as input of the next. Simply save the heads from the run that is to be
continued on a disk file, and specify that file as starting heads for the next run. The
subroutine that reads starting heads (U2DREL) is capable of reading modelgenerated
disk files of saved heads without the need for reformatting. Because volumetric
budget terms are always set to zero at the start of a model run, the printed budget on a
model run represents only that one run, not the total of all runs in a series of
continuation runs. If a total budget for a series of continuation runs is desired, the
totals from each run can be added externally. Similarly, the model program keeps
track of simulation time only for single model runs, but total simulation time for a
series of continuation runs can be calculated externally by adding the simulation
times of each run.
Electronic Manual for MODFLOW Appendix D - Sample Problem 181
Appendix D - Sample Problem
Introduction
This sample problem is intended to illustrate input and output from the program.
There are three simulated layers, as shown in the accompanying illustration, which
are separated from each other by confining layers. Each layer is a square 75,000 feet
on a side and is divided by a grid into 15 rows and 15 columns which form squares
5,000 feet on a side. Flow within the confining layers is not simulated, but the effects
of the confining layers on flow between the active layers are incorporated in the
vertical leakance (Vcont) terms. Flow into the system is infiltration from
precipitation; flow out of the system is to buried drain tubes, discharging wells, and a
lake which is represented by a constant-head boundary.
182 Appendix D - Sample Problem Electronic Manual for MODFLOW
Setting starting heads equal to 0.0, the program was run to get a steady-state solution.
The Strongly Implicit Procedure was used to solve the system of difference
equations: the error criterion was set at 0.001 feet, the acceleration parameter was set
to 1.0, and the maximum number of iterations was set equal to 50. A seed of 0.001
was specified for use in calculating the iteration parameters; 31 iterations were
needed to close.
List of Wells (Q = 5 ft
3
/s for each well)
Layer Row Column
3 5 11
2 4 6
2 6 12
1 9 8
1 9 10
1 9 12
Electronic Manual for MODFLOW Appendix D - Sample Problem 183
1 9 14
1 11 8
1 11 10
1 11 12
1 11 14
1 13 8
1 13 10
1 13 12
1 13 14
List of Drains (Conductance = 1 ft
2
/s)
Layer Row Column Elevation
1 8 2 0.0
1 8 3 0.0
1 8 4 10.0
1 8 5 20.0
1 8 6 30.0
1 8 7 50.0
1 8 8 70.0
1 8 9 90.0
1 8 10 100.0
Electronic Manual for MODFLOW Appendix E - Input Instructions 185
Appendix E - Input Instructions
Introduction
These input instructions are intended as a quick reference for the experienced user.
Most explanations that are contained in the complete input instructions given in
package documentation have been omitted. The format of input fields is given only
for those records that contain fields that are not 10 characters wide. Each input item,
for which format is not given, is identified as either a record or an array. For records,
the fields contained in the record are named. For arrays, only the array name is
given. Input fields which contain codes or flags are described. All other field and
array descriptions have been dropped.
Array Input
The real two-dimensional array reader (U2DREL), the integer two-dimensional array
reader (U2DINT), and the real one-dimensional array reader (U1DREL) read one
array-control record and, optionally, a data array in a format specified on the array-
control record.
FOR REAL ARRAY READER (U2DREL or U1DREL)
Data: LOCAT CNSTNT FMTIN IPRN
Format: I10 F10.0 5A4 I10
FOR INTEGER ARRAY READER (U2DINT)
Data: LOCAT ICONST FMTIN IPRN
Format: I10 I10 5A4 I10
IPRN--is a flag indicating that the array being read should be printed and a code for
indicating the format that should be used. It is used only if LOCAT is not equal to
zero. The format codes are different for each of the three modules. IPRN is set to
186 Appendix E - Input Instructions Electronic Manual for MODFLOW
zero when the specified value exceeds those defined in the chart below. If IPRN is
less than zero, the array will not be printed.
IPRN U2DREL U2DINT U1DREL
0 10G11.4 10I11 10G12.5
1 11G10.3 60I1
2 9Gl3.6 40I2
3 15F7.1 30I3
4 15F7.2 25I4
5 15F7.3 2015
6 15F7.4
7 20F5.0
8 20F5.1
9 20F5.2
10 20F5.3
11 20F5.4
12 10G11.4
LOCAT--indicates the location of the data which will be put in the array.
If LOCAT < 0, unit number for unformatted records.
If LOCAT = 0, all elements are set equal to CNSTNT or ICONST.
If LOCAT > 0, unit number for formatted records.
Basic Package Input
Input for the Basic (BAS) Package except for output control is read from unit 1 as
specified in the main program. If necessary, the unit number for BAS input can be
changed to meet the requirements of a particular computer. Input for the output
control option is read from the unit number specified in IUNIT(12).
FOR EACH SIMULATION
1. Record: HEADNG(32)
2. Record: HEADNG (continued)
3. Record: NLAY NROW NCOL NPER ITMUNI
4. Data: IUNIT(24)
Format: 24I3
(BCF WEL DRN RIV EVT XXX GHB RCH SIP XXX SOR OC)
1 2 3 4 5 6 7 8 9 10 11 12
5. Record: IAPART, ISTRT
6. Array: IBOUND(NCOL,NROW)
(One array for each layer in the grid)
7. Record: HNOFLO
8. Array: Shead(NCOL,NROW)
(One array for each layer in the grid)
9. Data: PERLEN NSTP TSMULT
ITMUNI--is the time unit of model data.
0 - undefined 3 - hours
1 - seconds 4 - days
2 - minutes 5 - years
Consistent length and time units must be used for all model data. The user
may choose one length unit and one time unit to be used to specify all input data.
IUNIT--is a 24-element table of input units for use by all major options.
IAPART--indicates whether array BUFF is separate from array RHS.
If IAPART= 0, the arrays BUFF and RHS occupy the same space. This
option conserves space. This option should be used unless some other package
explicitly says otherwise.
If IAPART 0, the arrays BUFF and RHS occupy different space.
ISTRT--indicates whether starting heads are to be saved.
If ISTRT = 0, starting heads are not saved.
If ISTRT 0, starting heads are saved.
IBOUND--is the boundary array.
If IBOUND(I,J,K) < 0, cell I,J,K has a constant head.
If IBOUND(I,J,K) = 0, cell I,J,K is inactive.
If IBOUND(I,J,K) > 0, cell 1,J,K is active.
HNOFLO--is the value of head to be assigned to all inactive cells.
Shead--is head at the start of the simulation.
PERLEN--is the length of a stress period.
NSTP--is the number of time steps in a stress period.
TSMULT--is the multiplier for the length of successive time steps.
Output Control Input
Input to Output Control is read from the unit specified in IUNIT(12). All printer
output goes to unit 6 as specified in the main program. If necessary, the unit number
for printer output can be changed to meet the requirements of a particular computer.
FOR EACH SIMULATION
1. Record: IHEDFM IDDNFM IHEDUN IDDNUN
FOR EACH TIME STEP
2. Record: INCODE IHDDFL IBUDFL ICBCFL
3. Record: Hdpr Ddpr Hdsv Ddsv
(Record 3 is read 0, 1, or NLAY times, depending on the value of INCODE.)
IHEDFM--is a code for the format in which heads will be printed.
IDDNF --is a code for the format in which drawdowns will be printed.
0 (10G11.4) 7 (20F5.0)
1 (11G10.3) 8 (20F5.1)
2 (9G13.6) 9 (20F5.2)
3 (15F7.1) 10 (20F5.3)
4 (15F7.2) 11 (20F5.4)
5 (15F7.3) 12 (10G11.4)
6 (15F7.4)
IHEDUN--is the unit number on which heads will be saved.
DDNUN--is the unit number on which drawdowns will be saved.
INCODE--is the head/drawdown ouput code.
If INCODE < 0, layer-by-layer specifications from the last time steps are
used. Input item 3 is not read.
= 0, all layers are treated the same way. Input item 3 will consist of one
record. IOFLG array will be read.
> 0, input item 3 will consist of one record for each layer.
IHDDFL-- is a head and drawdown output flag.
IF IHDDFL = 0, neither heads nor drawdowns will be printed or saved.
IF IHDDDL not equal to 0, heads and drawdowns will be printed or saved.
IBUDFL--is a budget print flag.
If IBUDFL = 0, overall volumetric budget will not be printed.
If IBUDFL not equal to 0, overall volumetric budget will be printed.
ICBCFL--is a cell-by-cell flow-term flag.
Hdpr--is the output flag for head printout.
If Hdpr = 0, head is not printed for the corresponding layer.
If Hdpr not equal to 0, head is printed for the corresponding layer.
Ddpr--is the output flag for drawdown printout.
If Ddpr = 0, drawdown is not printed for the corresponding layer.
If Ddpr not equal to 0, drawdown is printed for the corresponding layer.
Hdsv--is the output flag for head save.
If Hdsv = 0, head is not saved for the corresponding layer.
If Hdsv not equal to 0, head is saved for the corresponding layer.
Ddsv--is the output flag for drawdown save.
If Ddsv = 0, drawdown is not saved for the corresponding layer.
If Ddsv not equal to 0, drawdown is saved for the corresponding layer.
Block-Centered Flow Package Input
Input for the BCF Package is read from the unit specified in IUNIT(l).
FOR EACH SIMULATION
1. Record: ISS IBCFCB
2. Data: LAYCON(NLAY) (maximum of 80 layers)
Format: 4012
(If there are 40 or fewer layers, use one record.)
3. Array: TRPY(NLAY)
4. Array: DELR(NCOL)
5. Array: DELC(NROW)
All of the arrays (items 6-12) for layer 1 are read first; then all of the arrays for layer
2, etc.
IF THE SIMULATION IS TRANSIENT
6. Array: sfl(NCOL,NROW)
IF THE LAYERTYPE CODE (LAYCON) IS ZERO OR TWO
7. Array: Tran(NCOL,NROW)
IF THE LAYERTYPE CODE (LAYCON) IS ONE OR THREE
8. Array: HY(NCOL,NROW)
9. Array: BOT(NCOL,NROW)
IF THIS IS NOT THE BOTTOM LAYER
10. Array: Vcont(NCOL,NROW)
IF THE SIMULATION IS TRANSIENT AND THE LAYER TYPE CODE
(LAYCON) is TWO OR THREE
11. Array: sf2(NCOL,NROW)
IF THE LAYER TYPE CODE IS TWO OR THREE
12. Array: TOP(NCOL,NROW)
ISS--is the steady-state flag.
If ISS 0, the simulation is steady state.
If ISS = 0, the simulation is transient.
IBCFCB--is a flag and a unit number.
If IBCFCB > 0, cell-by-cell flow terms will be recorded if ICBCFL (see
Output Control) is set.
If IBCFCB = 0, cell-by-cell flow terms will not be printed or recorded.
If IBCFCB < 0, print flow for constant-head cells if ICBCFL is set.
LAYCON--is the layer type table:
0 - confined, I - unconfined,
2 - confined/unconfined (T constant), and 3 - confined/unconfined.
TRPY--is an anisotropy factor for each layer: T or K along a column to T or K along
a row.
DELR--is the cell width along rows.
DELC--is the cell width along columns.
SF1--is the primary storage factor.
Tran --is the transmissivity along rows.
HY--is the hydraulic conductivity along rows.
BOT--is the elevation of the aquifer bottom.
Vcont--is the vertical hydraulic conductivity divided by the thickness from a layer to
the layer beneath it.
sf2--is the secondary storage factor.
MT--is the elevation of the aquifer top.
River Package Input
Input to the River (RIV) Package is read from the unit specified in
IUNIT(4).
FOR EACH SIMULATION
1. Record: MXRIVR IRIVCB
2. Record: ITMP
3.Record: Layer Row Column Stage Cond Rbot (Input item 3 normally consists of
one record for each river reach. If ITMP is negative or zero, item 3 is not read.)
IRIVCB--is a flag and a unit number.
If IRIVCB > 0, cell-by-cell flow terms will be recorded. If IRIVCB = 0,
cell-by-cell flow terms will not be printed or recorded.
If IRIVCB < 0, river leakage will be printed if ICBCFL is set.
If ITMP < 0, river data from the last stress period will be reused.
If ITMP > 0, ITMP will be the number of reaches active during the current
stress period.
Recharge Package Input
Input to the Recharge (RCH) Package is read from the unit specified in IUNIT(8).
FOR EACH SIMULATION
1. Record: NRCHOP IRCHCB FOR EACH STRESS PERIOD
2. Record: INRECH INIRCH 3. Array: RECH(NCOL,NROW)
IF THE RECHARGE OPTION IS EQUAL TO 2
4. Array: IRCH(NCOL,NROW)
NRCHOP--is the recharge option code.
1 - Recharge is only to the top grid layer.
2 - Vertical distribution of recharge is specified in array IRCH.
3 - Recharge is applied to the highest active cell in each vertical
column.
IRCHCB--is a flag and a unit number.
If IRCHCB > 0, unit number for cell-by-cell flow terms.
If IRCHCB < 0, cell-by-cell flow terms will not be printed or recorded.
INRECH--is the RECH read flag.
If INRECH < 0, recharge fluxes from the preceding stress period are used.
If INRECH > 0, an array of recharge fluxes, RECH (Lt-1), is read.
INIRCH--is similar to INRECH.
Well Package Input
Input for the Well (WEL) Package is read from the unit specified in
IUNIT(2).
FOR EACH SIMULATION
1. Record: MXWELL IWELCB
2. Record: ITMP
3. Record: Layer Row Column
(Input item 3 normally consists of one record for each well. If ITMP is negative or
zero, item 3 is not read.)
MXWELL--is the maximum number of wells used at any time.
IWELC --is a flag and a unit number.
If IWELCB > 0, unit number for cell-by-cell flow terms.
If IWELCB = 0, cell-by-cell flow terms will not be printed or recorded.
If IWELCB < 0, well recharge will be printed whenever ICBCFL is set.
If ITMP < 0, well data from the last stress period will be reused. If ITMP >
0, ITMP will be the number of wells active during the current stress period.
Drain Package Input
Input to the Drain (DRN) Package is read from the unit specified in
IUNIT(3).
FOR EACH SIMULATION
1. Record: MXDRN IDRNCB
2. Record: ITMP
3. Record: Layer Row Col Elevation Cond
(Input item 3 normally consists of one record for each drain.
If ITMP is negative or zero, item 3 will not be read.)
MXDRN--is the maximum number of drain cells active at one time. IDRNCB--is a
flag and a unit number.
If IDRNCB > 0, unit number for cell-by-cell flow terms. If IDRNCB = 0,
cell-by-cell flow terms will not be printed or recorded.
If IDRNCB < 0, drain leakage for each cell will be printed whenever
ICBCFL is set.
If ITMP < 0, drain data from the last stress period will be reused.
If ITMP > 0, ITMP will be the number of drains active during the current
stress period.
Evapotranspiration Package Input
Input to the Evapotranspiration (EVT) Package is read from the unit specified in
IUNIT (5).
FOR EACH SIMULATION
1. Record: NEVTOP IEVTCB
2. Record: INSURF INEVTR INEXDP INIEVT
3. Array: SURF
4. Array: EVTR
5. Array: EXDP
IF THE ET OPTION IS EQUAL TO TWO
6. Array: IEVT
NEVTOP--is the evapotranspiration (ET) option code.
1 - ET is calculated only for cells in the top grid layer.
2 - The cell for each vertical column is specified by the user in array IEVT.
IEVTCB--is a flag and a unit number.
If IEVTCB > 0, unit number for cell-by-cell flow terms.
If IEVTCB < 0, cell-by-cell flow terms will not be printed or recorded.
INSURF--is the ET surface (SURF) read flag.
If INSURF > 0, an array containing the ET surface elevation will be read.
If INSURF < 0, the ET surface from the preceding stress period will be
reused.
INEVTR--is similar to INSURF.
INEXDP--is similar to INSURF.
INIEVT--is similar to INSURF.
General-Head Boundary Package Input
Input for the General-Head Boundary (GHB) Package is read from the unit specified
in IUNIT(7).
FOR EACH SIMULATION
1. Record: MXBND IGHBCB
2. Record: ITMP
3. Record: Layer Row Column Head Cond
(Input item 3 normally consists of one record for each GHB. If ITMP is negative or
zero, item 3 is not read.)
MXBND--is the maximum number of general-head boundary cells at one time. I
IGHBCB--is a flag and a unit number.
If IGHBCB > 0, unit number for cell-by-cell flow terms.
If IGHBCB = 0, cell-by-cell flow terms will not be printed or recorded.
If IGHBCB < 0, boundary leakage-for each cell will be printed whenever
ICBCFL is set.
If ITMP < 0, GHB data from the preceding stress period will be reused.
If ITMP > 0, ITMP is the number of general-head boundaries during the
current stress period.
Strongly Implicit Procedure Package Input
Input to the Strongly Implicit Procedure (SIP) Package is read from the unit specified
in IUNIT(9).
FOR EACH SIMULATION
1. Record: MXITER NPARM
2. Record: ACCL HCLOSE IPCALC WSEED IPRSIP
IPCALC--is a flag indicating where the iteration parameter seed will come from.
0 - the seed will be entered by the user.
1 - the seed will be calculated at the start of the simulation from problem
parameters.
IPRSIP--is the printout interval for SIP.
Slice-Successive Overrelaxation Package Input
Input to the Slice-Successive Overrelaxation (SOR) Package is read from the unit
specified in IUNIT(Il).
FOR EACH SIMULATION
1. Record: MXITER
2. Record: ACCL HCLOSE IPRSOR
IPRSOR--is the printout interval for SOR.

Monflow Manual

Uploaded by

Copyright:

Available Formats

Monflow Manual

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Monflow Manual

Uploaded by

Copyright:

Available Formats

Electronic Manual for

, by WexTech Systems, Inc.

14 2 Model Derivation Electronic Manual for MODFLOW

For the vertical direction, inflow through the bottom face is

while inflow through the upper face is given by

the general external flow term for cell i,j,k is

You might also like