A Method for Predicting Drawdown at the Radius of

a Pumping Well for Large Complex Systems

by Soheil Afsharia, Richard Mandleb, Qun Liua, and Shu-Guang Lia

Abstract
One of the challenges in groundwater modeling is the prediction of hydraulic head in close proximity to a
pumping well using a regional-scale model. Typical applications of numerical models to field-scale problems
generally require large grids that can seldom accommodate cells as small as the actual well diameter [Anderson and
Woessner 1992]. Several methods have been used to simulate a more accurate head at the well scale. The two
primary methods for head prediction are: (1) an analytical method, and (2) grid refinement using a numerical
model. However, these two methods have limitations, particularly for applications that involve the development of
numerical models for large-scale hydrogeologic systems with multiple pumping wells. Traditional methods may
become computationally problematic and increase the time and resources needed for storing and processing
simulation results.
A method for predicting head at the well scale that is more general, computationally efficient, and accurate
was presented. Interactive Ground Water (IGW) [Li et al. 2002; Li and Liu 2003; Li and Liu in review], which
adopts a hierarchical modeling paradigm, was used to illustrate this approach and determine hydraulic head at the
well scale. The performance of the hierarchical-modeling approach against the analytical solution for a single well
and against a superposition of analytical solutions for a well field was presented. Our results demonstrate that the
hierarchical-modeling approach is capable of matching an exact solution of well drawdown and providing an
accurate representation of head and groundwater velocities in a well field in large-scale hydrogeologic systems.

Introduction groundwater-management and sustainability

strategies. In solute-transport modeling, the
estimated velocity field is derived from head values
In groundwater modeling, it is usually
at the model nodes; this is needed to predict the
impractical to employ grids that are comparable in
advection component of the transport of solutes. It
size and dimension to a pumping well. However,
is also necessary for the proper design of remedial
predicting head, or drawdown, in close proximity to
actions.
a pumping well (well scale) may be important for
In finite-difference models, numerous finite-
certain groundwater-flow and solute-transport
difference cells, that typically have relatively-large
modeling applications, especially in complex
spatial dimensions, are used to represent the aquifer
environments. In groundwater-flow modeling,
system. A point source or sink of water is injected,
long-term drawdown prediction at the well is
or extracted, over the volume of aquifer represented
critical for proper well design [Beljin 1987]. It is
by the cell that contains the point source or sink
also important to predict drawdown at the well
[Anderson and Woessner 1992]. However, the
scale where models are used for evaluation of
diameter of the well is typically much smaller than
the dimensions of the cell. Field-scale problems
a
Department of Civil and Environmental Engineering generally cover large geographic areas that require
Michigan State University, East Lansing, MI 48824
b grids having cells with large spatial dimensions and
Michigan Department of Environmental Quality
Constitution Hall, 3rd Floor South can seldom accommodate cells as small as the
525 W. Allegan Street, Lansing, MI 48909 actual well diameter. The resulting simulated heads
are generally not a good approximation of heads or examples to verify and show the capabilities of the
hydraulic gradients in close proximity to the hierarchical-modeling method.
pumping well, or any other source or sink;
numerous small cells are needed to simulate the
relatively-steep hydraulic gradients near a point
Traditional methodologies
source or sink accurately. However, regional-
model-derived heads may be correct at nodes Two traditional methodologies employed in
located away from the point source or sink finite-difference models for predicting the
[Anderson and Woessner 1992]. drawdown or hydraulic head at the radius of, or in
Several methods have been adopted in an close proximity to, a pumping well (well scale)
attempt to improve the accuracy of simulated heads using regional-scale models are (1) an analytical
at the well scale. Two methods used in finite- method (corrected drawdown), and (2) grid
difference models are: (1) an analytical method refinement using a numerical model (LGR or
(corrected drawdown) [Prickett 1967, Peaceman TMR).
1978, Pritchett and Garg 1980], and (2) grid
refinement using a numerical model local-grid- Analytical method: the corrected
refinement (LGR) [e.g. Mehl 2002], or telescopic- drawdown
mesh-refinement (TMR) [Ward et al. 1987].
However, these two methods may have severe
The model-calculated head can be thought
limitations, particularly for applications that
involve the development of numerical models for to represent the head at some distance ( re ) from
large-scale, complex hydrogeologic systems with the well node. An estimation of the head in the
multiple pumping wells, or other sources or sinks. well can be obtained from formulas based on the
In particular, the application of grid-refinement steady-state Thiem equation [Thiem 1906], which
techniques to these complex problems, may lead to can be applied to quasi-steady-state conditions
slow convergence, numerical oscillations, when the rate of removal of water from storage near
computational inefficiencies, and solution failure the pumping well is zero. The head in the well is
[Li et al. in review]. In addition, the time and calculated from the following equation [Anderson
computational resources needed to store and and Woessner 1992]:
process simulation results may be great.
The objectives of this paper are to address
QW T re
the limitations and drawbacks of the traditional hw = hi , j ln (1)
methods used to predict heads at the well-scale for 2T rw
large complex groundwater systems, and to present
an innovative methodology for determining well-
scale heads for such systems. Interactive Ground Where hw is the head in the well; hi , j is the head
Water (IGW) [Li and Liu 2003; Li and Liu in computed by the finite-difference model for the
review], which uses the hierarchical-modeling well node (i,j); QW T is the total pumping or
approach, overcomes the difficulties encountered
using traditional finite-difference methods in injection rate from the well; T is the
calculating well-scale heads. Furthermore, IGW is transmissivity; re is the radial distance measured
capable of modeling large complex groundwater from the node at which head is equal to hi , j ; and
systems in a flexible and computationally-efficient
framework on typical desktop computers. rw is the radius of the actual well. The radius re is
In the following sections, we review the two the effective well- block radius. Prickett [Prickett
traditional methods for calculating hydraulic heads 1967], Peaceman [Peaceman 1978, 1983] and
at the well scale, and point out the limitations with Trescott [Trescott et al. 1976] provided an equation,
the assumptions and implementation of these
which approximates re based on different model
methods. Next, we introduce the hierarchical-
modeling approach and explain the concept behind grid sizes. Their assumptions in applying this
this approach. Finally, we present illustrative approximation are: (1) flow to the well is within a

2
square finite-difference cell (well block) and can be estimation of hydraulic head or drawdown at the
described by a steady-state equation with no source well scale. In other words, there will be less
term except for the well discharge, (2) the aquifer is approximation error involved with averaging the
isotropic and homogenous in the well block, (3) predicted head for that particular well node.
only one well, located at the cell center, is in the Two different approaches are used to
well block, (4) the well fully-penetrates the aquifer, implement the grid-refinement method. In the first
(5) flow to the well is laminar, and (6) well loss is approach, LGR, the grid spacing is subdivided so
negligible. Trescott [Trescott et al. 1976] examined that it is finer in the area of interest (e.g.,
the performance of the corrected drawdown against surrounding of the well node). With LGR, there is
the analytical solution. Results revealed a one model with the area of grid refinement that is
reasonable improvement for predicting the part of the regional model. This often results in a
drawdown in the pumping well. Pritchett [Pritchett very large number of grid cells. In second
and Garg 1980], and Beljin [Beljin 1987] presented approach, TMR, grid-dependent information from
a more general approximation for the Thiem the regional model is used to construct a separate
equation. They investigated the cases where: (1) a model with finer grid spacing to obtain more
well is not positioned at the center of the well information around the area of interest. The
block, (2) the well block is rectangular having procedure is called telescopic-mesh-refinement
different aspect ratios, (3) the aquifer media is because it can be applied repeatedly to construct
anisotropic, and (4) there is more than one well multiple, successively-smaller, embedded models
within the block. [Ward et al. 1987].
The application of the analytical correction With TMR, the model with the finer grid is
method results in a drawdown prediction that is called a submodel or local model. In order to
comparable to the exact solution [Trescott et al. obtain detailed information for the local model, it is
1976]. However, this method suffers from several necessary to interpolate grid-dependent information
limiting factors; one of which is that it cannot easily from the regional model at the finer grid spacing of
be applied for multiple sources and sinks within the the local model. The transformation of information
well block. In addition, the corrected drawdown from regional model to local model and the
method is not valid for other general cases, such as selection of boundary conditions and starting
variable hydraulic conductivity, variable recharge, conditions for the local model are the most
anisotropic media with specified orientation (e.g., important issues using this numerical methodology
fractured rock), partial well penetration, and [Townley and Wilson 1980; Ward et al. 1987;
transient flow or pumping conditions. Buxton and Reilly 1986].
The second traditional methodology, grid
refinement, overcomes the limiting assumptions of Our objective here is to identify advantages and
the corrected-drawdown method. However, for limitations in using either approach to calculate
large grids and complex hydrogeologic systems, head or drawdown at the well scale. In using LGR
this method suffers from other conceptual, for simple or small-scale problems, the solution is
structure, and computational difficulties. obtained quickly, and consistency between the
regional and local model is maximized.
Numerical method: grid Furthermore, there will not be a significant increase
in effort for processing the results of the model
refinement simulations for the regional model or local model
[Leake et al. 1998]. However, for large-scale,
A model covering a relatively-large domain regional groundwater models where there may be a
and having grids with large spatial dimensions significant increase in the number of nodes (e.g.,
relative to the well scale is often referred to as the millions rather than thousands), the cost of
regional model. With the grid-refinement computation increases exponentially and the
method, the relatively-large finite-difference grid process becomes computationally intensive.
cell from a regional model is subdivided into Additionally, for large problems that involve
multiple cells with progressively-smaller spatial multiple sources and sinks, multiple scales of
dimensions. This results in a more accurate interest, transient flow conditions, complex aquifer

3
structure, and strong anisotropy and heterogeneity, number of submodels (e.g., 1 or 2), and are
the solution process using LGR can become implemented with little flexibility.
problematic. In these cases, the structure of the We present an innovative method that is
matrices may be highly ill-conditioned, which more computationally efficient than LGR or TMR
oftentimes leads to lack of convergence or while maintaining the capability of accurately
numerical oscillations [Li et al. in review]. predicting detailed head or drawdown at the well
Selecting a proper numerical algorithm or a matrix scale.
solver can improve convergence or dampen
oscillations, to some extent; however, the major
problem, which often results in solution failure, is
Hierarchical modeling
the large number of grids in a very large, non- approach
uniformly-structured matrix.
TMR avoids the potential difficulties Our approach for predicting head or
encountered with LGR by separating the regional drawdown at the well scale employs the concept of
model from the local model, and solving each hierarchical modeling in a unique, general purpose,
model individually. Therefore, instead of solving and object-oriented computational environment,
very large, complex matrices, a problem is solved Interactive Ground Water (IGW) [Li and Liu 2003;
using multiple, smaller-scale local models. This Li and Liu in review]. With this approach we are
procedure can be implemented (in principle) able to reduce a large, very-complex problem into a
successively in a continuous manner until the number of small, less complex problems that are
desired resolution for the well scale is obtained. solved individually in a dynamically-coupled and
The local model will derive its boundary and fully-integrated environment patch-dynamic
starting conditions from the parent model. Once framework [Li et al. in review]. The hierarchical
the local model is created, it performs as an modeling approach overcomes problems
independent model. The primary advantage of encountered when attempting to solve a very large
TMR is that it can reduce a large and complex set of complex, ill-conditioned matrices often
matrix system into multiple smaller and better- encountered with LGR. The approach is more
conditioned matrices. advantageous than the traditional, loosely coupled
However, the major drawback to TMR in that the interaction between the regional
implementing TMR is that the interaction between model and all submodels is seamless, dynamic, and
the parent and local models (of which there can be fully-integrated. In other words, regional-model-
several) depends on the offline analysis and simulation results (e.g., boundary conditions), and
processing of model modifications or simulation any changes to the regional model propagate
results from the parent model to obtain the automatically to all submodels, without the need for
boundary and starting conditions for the local offline post-processing of data or simulation results.
model. For example, once a new simulation is This transfer of information between the regional
completed using the regional model, new boundary model and submodels is accomplished for each
conditions and starting flow and solute-transport time step in near real-time. IGW hierarchical
conditions, if applicable, are determined for each modeling achieves these distinct capabilities
local model at the local-model grid spacing. through the adoption of the following new,
Making modifications to models or processing integrated computational paradigm [Li and Liu
simulation results for use in different scales of 2003; Li and Liu in review](showing in the
models can be very time consuming; especially following text box).
when the problem is a transient-flow or transport The dynamic paradigm eliminates the
condition, or there is uncertainty in selecting the traditional TMR disconnect and makes generalized
boundary conditions. The effort involved may hierarchical modeling practical. In particular, this
become impractical when the offline conceptual paradigm provides an efficient means of routing
changes must be made iteratively or in more than data between models and also provides
one model. Because of this, applications of the visualization controls at each time step for all
TMR are limited, in most cases, to very small models during the simulation. This gives the
modelers the perception of using a single model

4
that provides high resolution dynamics at the speed of submodels that are embedded in a parent model
that is very similar to that of a low-resolution in order to provide greater detail where it is
coarse-grid model. The object-oriented required. . IGW provides an intelligent and
implementation enables flexible, interactive integrated modeling environment in which to
creation of hierarchical models. This hierarchical conduct hierarchical model [Li and Liu 2003; Li
modeling process is an interactive and recursive and Liu in review].
process by which a modeler can create a hierarchy

t1 regional model, submodel1, submodel2, , on-line processing, analysis, integrated mapping & visualization
t2 regional model, submodel1, submodel2, , on-line processing, analysis, integrated mapping & visualization
t3 regional model, submodel1, submodel2, , on-line processing, analysis, integrated mapping & visualization
t4 regional model, submodel1, submodel2, , on-line processing, analysis, integrated mapping & visualization
.
tn regional model, submodel1, submodel2, , on-line processing, analysis, integrated mapping & visualization
end

Our experience with hierarchical modeling drawdown at the radius of a pumping well. The
and submodel development has shown that to analytical solution assumes that the confined
maintain computational efficiency and robustness aquifer is homogenous and isotropic, with
in modeling detailed flow dynamics around wells in impermeable boundaries bounding the top and
a very large complex system it is best to limit the bottom of the aquifer. It further assumes that the
number of grid cells and keep grid spacing aquifer has infinite lateral extent. The objectives of
relatively uniform. This may result in the this example are to (1) verify the use of hierarchical
development of multiple submodels with modeling in predicting the drawdown at the radius
successively-smaller grid spacing until it is possible of a pumping well, and (2) to illustrate the process
to estimate head accurately at the well scale. and concept of hierarchical modeling in assessing
However, with hierarchical modeling and IGW, this the solution, as it is implemented in IGW.
process of creating an appropriate number of In the finite-difference model developed to
submodels with successively-smaller grid spacing solve this problem, the aquifer is represented by a
is accomplished easily and in a way that is naturally rectangular layer whose extent is 10,000 meters by
intuitive [Li and Liu 2003; Li and Liu in review]. 10,000 meters. No-flow boundary conditions are
imposed along each face of the model. The aquifer
parameters are as follows: the transmissivity and
Illustrative examples storage coefficient of the aquifer are 17.3 m2/day
In this paper, we apply IGW to illustrate
and 0.001, respectively; the discharge rate of the
and verify the hierarchical-modeling-approach for
pumping well is 518.4 m3/day; the well radius is 0.1
predicting the drawdown at the radius of a pumping
meters (0.1m) and the duration of pumping is 50
well by performing a systematic comparison with
days.
an analytical solution (Theis equation). Following
Figure 1 presents the hierarchical-network
the first example, we extend our examination to a
and solutions to the single well example problem.
more general and complex well-field system and
The regional model and the six submodels have
the simulation of detailed flow dynamics around
grid resolutions of 500m, 100m, 30m, 9m, 2.7m,
wells in a large regional system.
and 0.81m respectively.
Each model has a grid that is 21 rows by
Single well example 21 columns in size. Figure 2 illustrates the
drawdown comparison between the numerical
The simulation results obtained using IGW model and analytical solution. Drawdown data
are compared to the analytical solution given by from every other node were used to prepare this
Theis [Theis 1935] to demonstrate the ability of the graph. It is apparent, in examining the results
hierarchical-modeling-approach to predict obtained from hierarchical-modeling-approach, that

5
the numerical model closely-approximates the hierarchical modeling, different drawdown values
analytical solution. Also, it is important to point may be recomputed, displayed, and analyzed very
out that the results from each submodel were quickly whenever the model stresses or other model
obtained and visualized instantaneously. And, with parameters are changed.

Figure 1 - Presents the hierarchical-network and simulated heads after 50 days pumping for the single well example
problem. Each model has a grid that is 21 rows by 21 columns in size.

6
 Figure 2 - Illustrates the drawdown comparison at the end of 50 days between the numerical model and analytical
solution. Each model has a grid that is 21 rows by 21 columns in size.

An alternative hierarchical model applied to model into several smaller submodels. The
the same single well problem is shown in Figure 3. decision on how to subdivide a regional model
In this case, each model has a larger grid that is 51 depends primarily on the computational ability of
rows by 51 columns in size, requiring only three the computer available to perform the analysis.
nested submodels to accurately predict the One may choose to divide the regional model into
drawdown at the well scale. Figure 4 illustrates the larger numbers of smaller submodels when using a
drawdown comparison between the analytical less powerful computer.
solution and numerical hierarchical models. On Likewise, an approach using a smaller
this graph, only the simulated drawdown results number of larger submodels would be taken when
from every other model node are plotted. The using a more powerful computer. This flexibility
alternative hierarchical solution show that there is eliminates the longstanding, infamous curse of
more than one way to subdivide a larger regional dimensionality in large-scale groundwater

7
modeling and allows the modeler to simulate wells, in a large, regional groundwater system.
detailed flow dynamics, even around low capacity

Figure 3 - Presents the hierarchical-network and simulated heads after 50 days pumping for the single well example
problem. Each model has a grid that is 51 rows by 51 columns in size.

Well field example pumping at different rates. This example applies

for groundwater-management problems, delineating
The objective of this example is to illustrate the the wellhead protection area (WHPA) for a well
capabilities of the hierarchical-modeling-approach field, or prediction of groundwater-flow directions
in calculating hydraulic heads for a large grid and velocities for solute-transport modeling.
containing a well field with multiple wells each

8
 Figure 4 - Illustrates the drawdown comparison at the end of 50 days between the numerical model and analytical
solution. Each model has a grid that is 51 rows by 51 columns in size.

Computational difficulties encountered in modeling model the system. Model grids of different extent
well fields center around calculating heads and and cell size are needed to model the system
groundwater flow rates where there are clusters of accurately so that the flow dynamics in a well field
wells that pump at different rates and with different with multiple well clusters and variable pumping
pumping schedules. In this case, the system may be rates are resolved.
very dynamic with spatially- and temporally- Another difficulty in modeling a well field
variable heads and groundwater flow. In areas far is representing the exact location of the individual
removed from the well field, predictions using a pumping wells, especially those in close proximity
regional model are usually accurate; however, to each other. With a regional model, or coarse
closer to the well node, a finer grid is needed to grid submodel, clusters of closely spaced wells are

9
grouped together in a single well node. A locally meters. As with the single-well example, the outer
refined grid is needed to simulate the impact of boundaries are represented by no-flow boundaries,.
individual wells on heads and flow rates. The no-flow boundaries in the regional model have
The model simulation results are compared been placed at this distance so they do not affect the
to an analytical solution for drawdown in a well simulated drawdown in the well field, satisfying the
field obtained by superimposing Theis solutions for assumption of an infinite aquifer. The aquifer
each pumping well to verify the hierarchical- parameters are as follows: the transmissivity and
modeling-approach for a well field in a regional storage coefficient of the aquifer are 200 m2/day
model. The distribution of the pumping wells and and 0.001, respectively; the well radius is 0.1 meter
their pumping rates for the well-field example are (0.1m); and the duration of pumping is 64 days.
shown in Figure 5. The regional-model domain is The discharge rates of the pumping wells are
100 km by 100 km with a uniform grid size of 4000 variable and are shown in Figure 5.

Figure 5 - The distribution of the pumping wells and their pumping rates for the well-field example.

10
Figure 6 - Hierarchical modeling layout which illustrates the relationship between submodels and their parent
a
models. [In label of Submodel M bc a represents the generation level (grandparent), b is the parent index, and c is
the kid index which is related to parent index].

11
Figure 7 - The modeling results are compared to the analytical solution on a cross-section profile (A-A)
for different submodel levels and grid resolutions (a)- large scale (b)- local scale.

12
 One possible example of the application of approaches the analytical solution. In addition, the
hierarchical modeling is shown in Figure 6. In this results from each submodel were obtained and
example, different levels of submodels are visualized instantaneously.
developed for each pumping well. Starting from a With IGW and hierarchical modeling,
coarse model we can successively approach the different drawdown values may be recomputed,
area of interest and obtain detailed information as displayed, and analyzed very quickly whenever the
close as the effective well radius. The diagram model stresses or other model parameters are
shown in the upper right quadrant of this figure changed.
illustrates the hierarchical relationship between
parent models and subsequent submodels. The
remainder of the figure shows the different model Summary
domains, pumping wells and simulated drawdowns.
Lines showing the relationship between parent With numerical models, the simulated head
model and dependent submodel domains were not at the well scale is the average of the head over the
included for figure clarity reasons. However, an well block and does not reflect the head value at the
examination of the hierarchical tree and the well or the hydraulic gradients in close proximity to
different model domains should be sufficient to the well. Since the applications of numerical
understand the relationship between models. The models to field-scale problems typically cover
grid resolution from the regional model to finest relatively-large geographic areas, a great number of
well-scale model varies between 4000 m to 0.49 m. cells having large size, relative to a pumping well,
The number of rows or columns in these models is are used. Analytical and numerical methods have
typically 30 to 100, depending on the size of the been adopted, to better evaluate the drawdown at
model area and the uniform grid spacing for that the radius of a pumping well. However, these
model. traditional methodologies have limitations. The
Figures 7a and 7b show the comparison analytical method has limiting assumptions that
between modeling results for different submodel cannot be applied for general groundwater
levels and grid resolutions and the analytical modeling. With numerical methods, LGR creates
solution obtained by superimposing Theis solutions computational difficulties because of the large
for the different pumping wells. The results are number of nodes, and TMR suffers from the
plotted along a cross-section profile (A-A) drawn discontinuity between parent and submodels and
through the regional model whose location is the time and effort required to process data between
shown in Figure 6. In the drawdown comparison parent and submodels.
1
we show the results from the regional model ( M 11 ) In this paper we presented an innovative
methodology to predict the head at the radius of a
through the finest-grid submodel ( M 118 ). The area pumping well for large-scale and complex
of interest in this problem is the pumping well groundwater-flow systems. Utilizing IGW, our
shown in the center of model M 118 (pumping well method employs the dynamically integrated, object-
oriented, hierarchical-modeling concept resulting in
of interest). Figure 7a shows the simulated heads
a highly efficient and flexible way to simulate
along a profile through the entire well field from
heads at the well scale. When modeling large-scale
model M 112 including the pumping well of interest. complex groundwater systems, the IGW
Figure 7b shows simulated heads from model M 118 hierarchical-modeling-approach allows the modeler
along the same profile through the pumping well of to obtain an accurate solution at the desired grid
interest. However, in this figure, the lateral extent resolution quickly and with little difficulty.
of the profile is limited to the extent of the domain Convergence to a solution using this method is
computationally efficient and can be obtained using
of model M 118 in the X-direction. As with the typical desktop computers, even for large complex
single well example, these results show that, as the field applications.
grid spacing is reduced, the numerical solution

13
References
1. Anderson, M. P. and W. W. Woessner. 1992. Applied groundwater modeling: simulation of flow and advection transport. San
Diego: Academic Press.
2. Beljin, M. S. 1987. Representation of individual wells in two dimensional groundwater modelin. Paper read at The
NWWA/IGMC Conference "Solving Groundwater Problems with Models", February 10-12, at Denver, Clorado.
3. Buxton, H. and T. E. Reilly. 1986. A technique for analysis of ground-water systems of regional and subregional scales applied
to Long Island.: U.S. Geological Survey,U.S. Geological Survey.
4. Harbaugh, A.W., and McDonald, M.G. 1966. Users Documentation for MODFLOW-96, an update to the U.S.
5. Geological Survey modular finite-difference ground-water flow model:: U.S. Geological Survey Open-File Report 96485.
6. Leake, S. A., P. W. Lawson, et al. 1998. Assignment of boundary conditions in embedded ground water flow models. Ground
Water 36 (4):621-625.
7. Li, S. G., et al. Interactive Ground Water (IGW). Li, S. G., et al. 2002 [cited. Available from http://www.egr.msu.edu/igw/.
8. Li, S.G. and Q. Liu. 2003. Interactive Ground Water (IGW): An Innovative Digital Laboratory For Groundwater Education
and Resear. Computer Applications in Engineering Education, 11 (4):179-202.
9. . (in review). A New Paradigm for Groundwater Modeling. Groundwater.
10. Li, S.G., Q. Liu, and S. Afshari. (in review). Modeling Complex Groundwater Systems Across Multiple Scales - A
Hierarchical Computational Steering Environment. Groundwater and Application.
11. McDonald, M.G., and Harbaugh, A.W. 1988. A modular three-dimensional finite-difference ground-water flow model: U.S.
Geological Survey Techniques of Water-Resources Investigations.
12. Mehl, S., and M. C. Hill. 2002. Development and evaluation of a local grid refinement method for block-centered finite-
difference groundwater models using shared nodes. Advances in Water Resources 25 (5):497-511.
13. Miller, R. T., and Voss, C.I. 1987. Finite-difference grid for a doublet well in an anisotropic aquifer. Ground Water, 24
(4):490-496.
14. Peaceman, D. W. 1978. Interpretation of Well-Block Pressures in Numerical Reservoir Simulation. Society of Petroleum
Engineers Journal 18 (3):183-194.
15. . 1983. Interpretation of Well-Block Pressures in Numerical Reservoir Simulation with Non-Square Grid Blocks and
Anisotropic Permeability. Society of Petroleum Engineers Journal 23 (3):531-543.
16. Prickett, T.A. 1967. Designing pumped well characterstics into electrical analog models. Ground Water 5 (4):38-46.
17. Pritchett, J. W., and S. K. Garg. 1980. Determination of Effective Well Block Radii for Numerical Reservoir Simulations.
Water Resources Research 16 (4):665-674.
18. Thiem, G. 1906. Hydrologische Methoden: Leipzig, Gebhardt.
19. Thies, C. V. 1935. The relation between the lowering of the pizometric surface and the rate and duration of discharge of a well
using groundwater storag. Trans. Amer. Geophys. Union. (2):519-524.
20. Townley, L. R. and J. L. Wilson. 1980. Description of and user's manual for a finite-element aquifer flow model AQUIFEM-1.
Cambridge Massachusetts.: Ralph M. Parson Laboratory Technology Adaption Program.
21. Trescott, P.C., G. F. Pinder, and S.P. Larson. 1976. Finite-difference model for aquifer simulation in two dimensional with
results of umerical experiment. U.S. Geological Survey, Chap. C1, Bk.7.
22. Ward, D. S., D. R. Buss, J. W. Mercer, and S. S. Hughes. 1987. Evaluation of a Groundwater Corrective Action at the Chem-
Dyne Hazardous-Waste Site Using a Telescopic Mesh Refinement Modeling Approach. Water Resources Research 23
(4):603-617.

14