Environ Geol (2008) 55:505–515

DOI 10.1007/s00254-007-0996-z

ORIGINAL ARTICLE

Applications of particle-tracking techniques to bank infiltration:

a case study from El Paso, Texas, USA
Ahmad Abdel-Fattah Æ Richard Langford Æ
Dirk Schulze-Makuch

Received: 12 March 2007 / Accepted: 9 August 2007 / Published online: 23 August 2007

Springer-Verlag 2007

Abstract This paper presents results of a small scale river. The examples presented in this paper indicate that
study that utilized particle-tracking techniques to evaluate operating the pumping well at a doubled distance from the
transport of river water through an alluvial aquifer in a river increased the time required for the water to travel to
bank infiltration testing site in El Paso, Texas, USA. The the well, but did not greatly change the capture zone.
particle-tracking survey was used to better define filtration
parameters. Several simulations were generated to allow Keywords Riverbank
Filtration
Particle tracking

visualization of the effects of well placement and pumping Modeling
MODPATH
rate on flow paths, travel time, the size of the pumping
influence zone, and proportion of river-derived water and
groundwater mixing in the pumping well. Simulations Introduction
indicate that migration of river water into the aquifer is
generally slow. Most water does not arrive at the well by In riverbank infiltration applications wells are located on
the end of an 18-day pumping period at 0.54 m3/min the banks to capture a portion of the river water through
pumping rate for a well located 18 m from the river. Forty- induced infiltration. The extraction of the groundwater near
four percent of the water pumped from the well was river the river, particularly for agricultural areas, such as the area
water. The models provided important information needed containing the bank infiltration site of this study, can lead
to design appropriate sampling schedules for bank filtration to river water infiltrating into the surrounding aquifer.
practices and ensured meeting adequate soil-retention Often, studies of river–aquifer interaction do not focus on
times. The pumping rate has more effect on river water mechanisms controlling movement of infiltrated river
travel time than the location of the pumping well from the water inside a nearby aquifer (for instance, Sontheimer
1980; Kim and Corapcioglu 2002; Ray 2002). However,
A. Abdel-Fattah (&)
more information on filtration parameters is required to
Hydrogeology Research Program, achieve a greater level of understanding of the movement
El Paso Agricultural Research and Extension Center, of river water inside nearby aquifers.
Texas A&M University, 1380 A&M Circle, According to Chen (2001), most river–aquifer interac-
El Paso, TX 79927, USA
e-mail: anabdel-fattah@ag.tamu.edu
tion studies have focused on the discharge losses in streams
due to extraction of the groundwater by a pumping well
R. Langford (e.g., Hantush 1965; Chen and Yin 1999, 2001). Chen
Department of Geosciences, (2001) emphasized that studies on river discharge depletion
The University of Texas at El Paso,
500 W. University Avenue,
must be expanded by including a determination of the
El Paso, TX 79968, USA following characteristics: the distance the infiltrated river
water can travel into the aquifer during a pumping period,
D. Schulze-Makuch the travel time from the river–aquifer boundary to the
School of Earth and Environmental Sciences,
Washington State University, P.O. Box 64281,
pumping well, and the area of aquifer influenced by river
Pullman, WA 99163, USA water. A few studies have used analytical solutions to deal

123
506 Environ Geol (2008) 55:505–515

with the movement of infiltrated river water inside a nearby

aquifer (for instance, Chen 2001; Chen and Yin 2001). In
our study we employed numerical simulations utilizing the
groundwater flow code MOFLOW (a 3-D, cell-centered,
finite difference, saturated groundwater flow model
developed by the USGS) to initiate particle-tracking sim-
ulations using MODPATH (a 3-D, particle-tracking code
developed by USGS which computes the paths for imagi-
nary particles of water moving through the simulated
groundwater system). The simulations provided useful
information on important parameters pertaining to infil-
tration such as pumping rate and optimal distance between
riverbank and the production well. In this study travel
times, pathlines, and influence zones of river water were
determined between a river and a nearby pumping well for
seasonal groundwater extractions. These flow/transport
parameters were determined to characterize the interactions
between water in the river and the alluvial aquifer.
Applications of such particle-tracking techniques are
important in transport studies during bank infiltration to
predict the attenuation of pathogens during transport and Fig. 1 A map of the study area. The study site is shown in the
artificial recharge. This is also important to ensure that enlarged view in the inset. Inset: map showing location of wells and
the geometry of the river bank filtration site
adequate soil-retention time requirements are met for
removal of human pathogens as a main goal of bank
infiltration operations. sampling campaigns. The stream is 1.5–2 m deep and 4 m
wide when full. The water table during the simulation
experiment was 2–2.1 m below the ground surface and
Study site 1.9 m below the stream surface, approximately at the same
level as the stream bed. Although the aquifer thickness is
The research site, about 200 m2 in area, is located in the 24 m, the bottom layer (model layer 4) was assumed to
Rio Bosque Wetlands Park in El Paso, Texas at the border have a 50-m thickness to minimize the effect of the lower
between Texas and the Mexican State of Chihuahua boundary condition in the simulation (Fig. 2).
(Fig. 1). The site utilizes surface water in an artificial As is common in arid settings, the stream is a losing
stream (stream and river in this manuscript are used reach that continually discharges into the groundwater. The
interchangeably) which meanders through the park. The water table fluctuates seasonally, being high during the late
stream flows approximately 4 months per year with treated summer and fall irrigation season when the nearby River-
water from an adjacent wastewater treatment plant. The side Canal (Fig. 1) is full and provides an additional
aquifer consists of medium- to fine-grained sand, with thin recharge source. Groundwater is the lowest in the winter
discontinuous beds of clay, silt, and gravel. The entire unit when flow in the Rio Grande and canal are reduced.
is generally less than 38 m thick and was deposited by the
Rio Grande during the late Pleistocene to Holocene time
period (Wilson et al. 1981). No wells in the study site Conceptual model
penetrate through the aquifer. However, the aquifer thick-
ness can be estimated from a well 75 m from the study site For the groundwater flow simulation, a 3-D conceptual
that penetrated 24 m of aquifer sands before encountering a model of the bank infiltration site (Fig. 2) was constructed
10.5-m thick layer of shale that confines the surficial based on integrated hydrogeological, radar penetration,
aquifer. The 10-cm diameter pumping well is located 18 m sediments, and aquifer test analyses of the site (Abdel-
from the stream. This relatively short distance to the Fattah 2005). The model is a four-layer system, consisting
pumping well was chosen so that little water withdrawal of an upper unconfined layer, a leaky confining unit, a
occurs from the surrounding aquifer and to guarantee the semi-confined unit, and low permeability shale at the bot-
larger amounts of pumped water is derived from the stream tom. The top and the third layer, both medium-grained sand
which will minimize the effects of dilution from ground aquifer layers, are separated by a leaky confining fine- and
water and maximize bank filtration. Also, this allows short very fine-grained sand layer. The assumptions behind our

123
Environ Geol (2008) 55:505–515 507

Fig. 2 A 3-D conceptual flow

model for the bank infiltration
site (dimensions not to scale)

model are that this semi-confined aquifer is homogeneous,

isotropic, and assumes an infinite extent. The pumping well
partially penetrates the semi-confined aquifer.
For the particle-tracking simulation, a 2-D conceptual
model of the study site was constructed and simplified
analogous to Chen’s (2001) conceptual model (Fig. 3).
Chen (2001) described a river–aquifer model to illustrate
the migration process of river water using an analytical
approach. He considered an unconfined alluvial aquifer
bordered on one side by a river (Fig. 3). The unstressed
gradient was nearly zero at the time of our tracer test,
similarly to Chen’s (2001) model. Under this circumstance,
exchange of groundwater and river water does not occur
(Fig. 3a). After withdrawal begins, the groundwater table
around the production well declines and a cone of
depression is produced (Fig. 3b). As withdrawal continues,
the diameter of the cone of depression expands, and the
cone eventually intercepts the river. With more pumping,
the segment of influenced river may expand in upriver and
downriver directions but becomes constant after steady-
state conditions are reached.
In our case, the already steep hydraulic gradient from
the stream is enhanced in the influenced segment. This
process may take days or even weeks (Chen 2001). River
water along the river–aquifer boundary (x = 0) begins to
leak into the aquifer, and the ultimate destination of the Fig. 3 Diagram showing hypothetical river-aquifer systems: a no
hydraulic gradient between river and aquifer; b a regional hydraulic
water is the pumping well. Initially, the movement of river
gradient toward the river, which gains water from the aquifer; ho is
water in the aquifer must be very slow because the the water table prior to pumping, h is the hydraulic head at location x,
groundwater velocity is generally small near the outer y, and Q is the pumping rate (figures modified from Chen 2001)
limits of the cone of depression. Velocity increases as
infiltrated water gets closer to the well because the gradient will eventually reach the pumping well. However, during a
is much larger near the pumping well. As long as a short-term pumping period, some river water may reach the
pumping period is sufficiently long, all infiltrated water well whereas other water may simply remain in the aquifer.

123
508 Environ Geol (2008) 55:505–515

The water at the location x = 0 and y = 0 (Fig. 3) requires The outflow of the system was due to stream output and
the least time to arrive at the pumping well. Hantush (1965) was represented by a constant head boundary, and due to a
termed this pathline the ‘‘meridian’’ line, which is the x- specified flow boundary represented by the discharging
axis in Fig. 3. pumping well.

Materials and methods Model parameter estimation

Setting the boundary conditions Hydraulic conductivities of model layers were estimated
from analysis of aquifer test data and laboratory measure-
Boundary conditions for the flow model consisted of: (a) ments. Vertical hydraulic conductivities of core samples
two specified head (or constant head cells), which were the extracted during well construction were measured using the
two water courses located northeast (Riverside Canal) and constant head method (Klute 1965). Horizontal and vertical
southwest (historic Rio Grande River channel) of the study hydraulic conductivities in addition to transmissivities,
area (Fig. 2). This extension of boundaries of the small storage coefficients, and leakage rates through the semi-
study area to these natural hydrological boundaries was confining unit in the aquifer were estimated from pump
done to minimize the effect of boundary conditions on the tests. These values were considered as initial estimates of
simulation results. The boundaries were specified as the hydraulic parameters applied to the model (Table 1).
stages of the water courses. The water table in the model Vertical hydraulic conductivity for the bed layer underly-
was referenced to mean sea level (WGS 84) and was held ing the stream (colmation layer) was also measured using
at 1114.25 m in the small study area. This depth was on the constant head method. The value was 3.0 · 10–6m/s
average 2 m below the ground surface and was measured after taking the geometric mean of ten samples from dif-
in pump and observation wells prior to the test. (b) Because ferent depths in the clogging layer underlying the stream.
no natural hydrogeologic boundaries occur north and south
of the study area, arbitrary no-flow boundaries were set
50 m (Fig. 2) north and south of a line drawn between the The steady-state (unstressed) groundwater flow model
pumping well and the stream. These were far enough from and calibration
the pumping well such that no effects of pumping should
reach them during the modeled time (Fig. 2). Also a no- The 3-D-finite difference model was developed using
flow boundary near the deep bottom boundary (shale layer) MODFLOW. The model was originally developed to
was set. Finally, the pumping well was specified as con- simulate the ground-water flow and transport of bromide
stant flux where the pumping rate is known. Recharge from and microspheres tracers to mimic transport of pathogens
the surface was considered a constant flux boundary in a testing bank infiltration system in El Paso, Texas, USA
(almost negligible), as only a single, short raining event (Abdel-Fattah 2005). The GIS coverage-based module (the
occurred concurrently with the test. Other recharge sources conceptual model approach in GMS) was used to construct
in the site were infiltration from the stream to the shallow the MODFLOW model by importing a digital image of the
aquifer due to the induced infiltration caused by pumping. site from the USGS Seamless Data Distribution System

Table 1 Summary of initial input parameters of aquifer layers for the groundwater flow modeling
Aquifer type Model layer 1 Model layer 2 Model layer 3 Model layer 4
unconfined confined confined confined

Top elevation (m) 1116.25 1113.85 1113.05 1092.25

Bottom elevation (m) 1113.85 1113.05 1092.25 1042.25
Thickness (m) 0.4 0.8 20.8 50
Horizontal conductivity (from aquifer test (m/s) 3.1 · 10–4 3.3 · 10–5 3.1 · 10–4 3.0 · 10–5
–4 –5 –4
Vertical conductivity (from aquifer test (m/s) 1.5 · 10 1.4 · 10 1.5 · 10 1.5 · 10–6
–6 –6 –5
Vertical conductivity (from laboratory measurements (m/s) 5.0 · 10 2.0 · 10 2.9 · 10 2.9 · 10–6
–7
Net recharge rate in (m/s) 7.0 · 10 n/a n/a n/a
Specific yield (dimensionless) 5.0 · 10–2 9.1 · 10–5 5.0 · 10–2 5.0 · 10–2
Specific storage (1/m) 1.0 · 10–2 5.0 · 10–3 2.0 · 10–4 8.0 · 10–6
2 –3 –5 –4
Transmissivity (m /s) 6.4 · 10 3.0 · 10 6.4 · 10 6.4 · 10–4

123
Environ Geol (2008) 55:505–515 509

(SDDS) as a Digital Orthophoto Quad (DOQ) created by the main area of interest to improve the model’s predictive
ArcGIS (inset in Fig. 1 and the background of Figs. 4, 5, power at the targeted small study site. New values for
6, 7, 8, 9, 10, 11, 12). The river in the simulation was hydraulic conductivity and recharge were entered, a new
represented by a group of nodes within the grid. The head solution was generated, and a new error estimate was
at the river nodes was specified to be equal to the river computed. The initial input parameters (Table 1) were
stage. Since the pumping well represents a point of con- adjusted during calibration until the degree of fit between
vergence in the groundwater flow and causes steep model simulations and field measurements was quantified
gradients in the head near the well, the grid was refined to by statistical means used in groundwater flow modeling:
accurately model the flow near the well to a 10-cm grid. The acceptable residual should be a small fraction of the
The GIS-built model was converted to 3-D finite difference difference between the highest and lowest heads across the
grid of the four-aquifer layers. The extended 3-D-grid site and be based on: (1) the ratio of the Root Mean Squared
containing the smaller study site covered a square region (RMS) of error to the total head loss should be small; (2)
measuring 1432.64 m in the x-direction by100 m in y- head differential of\5% for the residual mean and standard
direction. Length in z-direction was 74.0 m. The grid type deviation; and \10% for the ratio of the standard deviation
used was cell centered, consisted of 50 rows and 187 to total head change (Anderson and Woessner 1992). For
columns. this model, the estimated RMS of error was 0.049 which is
The governing equation for ground water flow through very small relative to the total head loss (0.480 m). More-
the porous medium, which was used in the model, can be over, residual mean and standard deviation were –0.034 and
written using the following partial differential equation; 0.037 m respectively, and the ratio of the residual standard
      deviation to the overall range in head across the model
o oh o oh o oh o gradient was estimated to be 0.077. Calibration statistics
Kxx þ Kyy þ Kzz  W ¼ Ss :
ox ox oy oy oz oz ot coupled with acceptable model parameter values indicate
ð1Þ that a very reliable calibration was achieved. It is important
to note that calibration was met with the residual mean
Where Kxx, Kyy, and Kzz are the hydraulic conductivities in being close to 0, and the ratio of residual standard deviation
the three orthogonal directions (m/s or ft/day), h is the head to the overall range in head is adequately less than 10%.
driving the flow or is the saturated thickness of the aquifer
(m or ft),W is the volumetric flux per unit volume and
represents the source/sink term for water or withdrawal Transient groundwater flow simulation and calibration
(m3/s or ft3/day).
Ss is the specific storage capacity of the porous medium Predicted water levels obtained by the calibrated steady-
(dimensionless), t is the time (s or day). state simulation were used as the initial heads in the tran-
When withdrawal of water occurs, –W = R, where R is a sient groundwater flow simulation of the actual 18-day
general sink/source term, which is defined to be intrinsi- pump test. A forced-gradient flow was created by operating
cally positive to represent recharge (defines the volume of the submersible pump inserted in the pumping well (Fig. 2)
inflow to the system per unit volume of aquifer per unit of for 6 h before running the test to reach steady-state flow in
time (ft3/ft3/day or m3/m3/s). the saturated zone. The pumping rate was set at 0.54 m3/
The model equations were solved numerically with a min. Changes in water level in the stream during the test
fully implicit finite difference method using the Block were negligible. Here, transient pumping conditions were
Successive Over-relaxation method (Wang and Anderson modeled to examine the effect of the pumping well dis-
1982). charging from layer 3 from December 16, 2003 (10:30
The first run of the simulation was to determine the a.m.) until January 3, 2003 (10:00 a.m.). Since flow was
unstressed steady-state head distribution. The calibration transient, the velocity field was calculated each time the
took the initial estimates of model parameters (mainly observed water level date was available. Therefore, 66
hydraulic conductivity) and adjusted them by trial and error stress periods were simulated with 1-time step of simula-
until the model successfully reproduced the observed tion assigned to each stress period. Results of transient flow
configuration of the water table prior to pumping. The simulation were presented in the form of head contour
model was repeatedly run until the computed solution maps at the end of each time step.
matched field-observed values within an acceptable level In the stressed-model calibration, a set of transient-field-
of accuracy. observed heads from the same observation wells at the site
A solution computed with this initial model was impor- was used. The model calibration was evaluated and
ted for the smaller bank filtration site and the errors were accepted based on hydrographs at each target location fol-
analyzed. The groundwater flow model was re-calibrated in lowing the same procedures described above in calibrating

123
510 Environ Geol (2008) 55:505–515

the unstressed steady-state model. For each transient model stream into the shallow aquifer and backward from the
run, an analysis of the observed versus computed water pumping well to locate the origin of the flow paths in the
levels was conducted to determine the accuracy of the stream for the 18-day test period.
simulation. The calibration was ultimately accepted through
a series of model runs using revised parameter values that
produced output that agreed reasonably with the real system Results
observations with an estimated error of ±30 cm with a 95%
confidence. For the stressed model, the estimated RMS of Forward tracking from the river: travel times
error was 0.13 m which is very small relative to the total and pathlines determination
transient head loss (total drawdown) which was 4.61 m.
Moreover, residual mean was 0.1 m, absolute mean of A forward tracking scheme was used in MODPATH to
residuals was 0.11 m, and the ratio of the residual standard determine which particles are captured by the pumping
deviation (taken her as the RMS) to the overall drawdown in well. Particles were placed along interface between stream
the model gradient (4.61 m) was estimated to be 0.03. and aquifer in a subdomain of the model, extending
97.46 m upriver and 113.84 m downriver from the x = 0
and y = 0 position. Figure 4a shows the pathlines of river
Particle-tracking simulations water converging at the pumping well in plan view after 18
days of pumping and Fig. 4b shows 3-D view of the same
To estimate travel times and construct pathlines of river event. The white squares in Fig. 4a (gray in the 3-D view in
water in the aquifer, the particle-tracking code MODPATH Fig. 4b) represent the starting locations of the particles at
was used. MODPATH (McDonald and Harbaugh 1988; the river–aquifer boundary, and the white arrowed lines
Pollock 1989) is 3-D USGS particle-tracking and post- represent the river water pathlines terminating at the
processing (display) program designed to work with pumping well whereas the black arrowed lines represent
MODFLOW to identify travel times and paths of particles the locations of river water pathlines that did not reach the
mimicking contaminants (or tracers). The results of this well during the 18-day pumping period. As shown in
program represent groundwater travel times and pathlines
for advective transport only. Using a flow field computed
by MODFLOW, MODPATH can track a set of fictitious
particles to simulate the movement of contaminants start-
ing from user-defined point-source locations. MODPATH
assumes the validity of Darcy’s law and the law of con-
servation of mass in the same fashion as MODFLOW.
MODFLOW models provide the velocities only at the
midpoints of the cell boundaries. The particle velocities are
calculated by MODPATH using linear interpolation. Since
the velocity component of the particle at any time are
known functions of the particle coordinates, the coordi-
nates of the particle at any future time can also be
computed (Pollock 1989).
The groundwater velocity was first estimated using the
calibrated groundwater flow model described above. In
order to estimate the tracking times, an effective porosity
value was defined for each of the cells in the grid. An
effective porosity value of 0.20 was used for the top and
lower two layers of the model based on results from lab-
oratory measurements. A lower effective porosity value
(0.15) was used for the less permeable model layer (layer
no. 2, Fig. 2). To show the migration process of infiltrated
river water in the unconfined aquifer, forward and back-
ward particle tracking were performed using MODPATH.
A set of particle starting locations were specified sur-
Fig. 4 Top portion (a) shows plan view of pathlines of induced river
rounding the cell containing the pumping well. Particles water that are converging and not converging at the pumping well.
were tracked forward to track infiltrated water from the Bottom portion (b) shows 3-D view of same event

123
Environ Geol (2008) 55:505–515 511

Fig. 4, not all the river water volumes reach the pumping (Scenario 2), none of the river water pathlines converged at
well after the 18-day pumping period for a 18-m river the pumping well (Fig. 6). However, when both the dis-
pumping well distance and pumping rate of 0.54 m3/min tance to the river and the pumping rate were doubled
(Scenario 1). But the river water in the reach close to the (1 m3/min) (Scenario 3) many pathlines now converge at
pumping well and extending 18 m upriver (pathline 9) and the pumping well (Fig. 7). At this combined effect of
17.7 m downriver (pathline 140 ) from x = 0 and y = 0 point doubled distance and pumping rate, the water particle at the
(white lines), reaches the well (Figs. 4, 5). Figure 5 shows location (x = 0, y = 0) takes 1.70 days (40.8 h) to reach the
the distances for the pathlines starting locations of the well. The new upriver and downriver starting locations that
pathlines that only converge at the pumping well. The have pathlines converge at the pumping well were located
times required for these water particles to get to the well at x = 0, y = 24.87 m with a travel time of 2.51 days
are different. Those particles farther from the well take (60.24 h) and x = 0, y = 29.5 m with a travel time of 2.79
much longer to reach the well. For example, the water days (66.96 h) respectively.
particle at the location (x = 0, y = 0) takes 1.46 days (35 h) These simulations indicate that the pumping rate has
to reach the well. This pathline is almost straight (pathline more effect on river water travel time and influence zone
1, Fig. 5). The water at the location (x = 0, y = 18 m; than does the location of the pumping well. Using Scenario
pathline 9) takes 2.71 days (65.04 h) to reach the well. 1, the pathlines 18 m upriver and 17.7 m downriver con-
From the downriver direction (the other side of the location verged at the pumping well. Whereas doubling both the
x = 0, y = 0), the water at the location x = 0, y = 17.7 m; distance (36 m) and the pumping rate (1 m3/min) (Scenario
pathline 140 , takes 2.55 days (61.2 h) to reach the well. 3) added more distant pathlines converging at the well
(24.78 m upriver and 29.5 m downriver). This conclusion
was also supported by the results obtained when using half
Simulations of varied pumping rates and river-pumping the distance (9 m) between the river and the pumping well
well distances and keeping the same original pumping rate (0.54 m3/min)
(Scenario 4). In this scenario, the water particles at the
When the river-pumping well distance was doubled (36 m) location (x = 0, y = 0) takes 0.7 days (16.8 h) to reach the
keeping the same original pumping rate of 0.54 m3/min well (Fig. 8). The upriver and downriver locations with

Fig. 6 Pathlines of induced river water after doubling the distance

Fig. 5 Pathlines of induced river water only converging at the between the river and the pumping well. None is converging at the
pumping well. Travel time along each pathline varies (Scenario 1) pumping well (Scenario 2)

123
512 Environ Geol (2008) 55:505–515

(58.56 h), respectively. However, when the pumping rate

for this simulation of half river-well distance was doubled
(Scenario 5), new upriver and downriver location with
pathlines converging at the pumping well were added
(Fig. 9) (29.11 m upstream and 29.67 m downstream). The
shortest pathline, x = 0, y = 0, needed only 0.43 days
(10.32 h) to reach the well.

Riverbank infiltration influence zone

Results from previous simulations indicated that the aqui-

fer fraction between the river and pumping well can be
replaced with infiltrated river water during a pumping
period of several days. The geometry of this zone is
another characteristic of interest in analyzing river–aquifer
interactions incorporated in bank infiltration schemes. The
river water extending 18 m upriver and 17.61 m downriver
from x = 0 and y = 0 location reaches the well (white lines
in Figs. 4a, b, 5) in the 18-day pumping period. The water
Fig. 7 Pathlines of induced river water after doubling both the
particles traveling along the meridian line reached the well
pumping rate and the distance between the river and the pumping well
(Scenario 3) after only 2.71 days. However, those particles farther from
the well take much longer to reach the well.
The river water particles moving along other pathlines
remain in the aquifer, some already near the well, and
others still far from the well. The area covered by the
pathlines in this portion of the aquifer is filled with water
from the river. Connecting the ends of these pathlines
forms a curve representing equal-pumping-time locations
for the induced river water (Fig. 10). The area is almost
uniform as modeled because the most important influence
is the 2-m head gradient between the river and the aquifer.

Fig. 8 Pathlines of induced river water converging at the pumping

well using half the original river-pumping well distance and the same
original pumping rate (Scenario 4)

pathlines converging at the pumping well were almost

identical to the original simulation (Scenario 1). The
upriver and downriver locations having pathlines con-
verging at the pumping well were located at x = 0,
Fig. 9 Pathline of induced river water converging at the pumping
y = 17.92 m with a travel time of 2.42 days (58.08 h) and well using half the original river-pumping well distance and doubling
x = 0, y = 17.60 m with a travel time of 2.44 days the pumping rate (Scenario 5)

123
Environ Geol (2008) 55:505–515 513

Fig. 10 Pathlines for 18-day pumping period with the area covered
by the pathlines (shaded area) has been filled with river water
infiltrated during the pumping period

This area was estimated and was found to be 2,890 m2,

which delineates the part of aquifer filled by infiltrated
river water or the bank infiltration influenced zone. Note
that this area represents the influenced aquifer after 18 days
of continuous pumping. Given the area of bank infiltration Fig. 11 Plane view (a) of delineation of backward flow paths of
influence and the aquifer parameters, the volume of river particles using MODPATH for 18-day pumping period. Capture zone
illustrated by the fan-shaped polygon. The lower part of the figure (b)
water discharged to the aquifer can be determined. The
shows cross-section view of the same event
area of bank infiltration influence zone (A) was 2,890 m2,
aquifer thickness (b) was 24 m, and specific yield (S) was
0.05; thus the river water leaked to the aquifer is approx- well is shown in Fig. 12. The area for 18-day-pumping
imately equal to A · b · S, which is 3,468 m3. period influence zone is 2,080 m2. The influenced area will
be smaller for a shorter pumping period as shown in
Fig. 12 that illustrates seven zones for different pumping
Backward tracking of particles: capture zone analysis periods.

The delineation of backward flow paths of particles using

MODPATH is shown in Fig. 11a which illustrates that Mixture of groundwater and river water
water contributed from the river is mixing with water at the pumping well
contributed from the surrounding aquifer. The 3-D nature
of delineation of backward flow paths of particles is best The particle-tracking models demonstrate that river water
seen in a cross sectional view (Fig. 11b). at the study site moves slowly to the well during pumping
The backward tracking option in MODPATH was used periods. For example, for a pumping well located only
to delineate capture zones for the pumping well for given 18 m from the river, it takes 1.46 days to reach the well for
time periods. It is clear from the pumping well capture the shortest river-pumping well pathline, which is the
zone after 18 days of pumping (Fig. 11a) that the pathlines straight pathline (pathline 1, Fig. 4). For wells located
intersect the area covered by the river in the study site farther from the river, it will take longer. The pumping well
which indicates that a connection between the river and the will produce a mixture of river water and groundwater, thus
production well occurred during the pumping period. The a water sample collected from the well may therefore be a
evolution of the capture zone with time for the pumping mixture of the river water and groundwater chemistry.

123
514 Environ Geol (2008) 55:505–515

water will increase in the well and the measured con-

verging pathlines’ angle will increase. Note that the
distance of the well to the river is relatively short for this
study. Most bank filtration wells, for example irrigation
wells, are located far from rivers. Although an irrigation
season generally lasts 3 months (Chen 2001), most often,
several days of continuous pumping are followed by a
number of days without pumping. Therefore, in reality,
infiltrated river water moves even more slowly to the
pumping well.

Discussion and conclusion

Particle tracking, pathlines, travel times, and influence

zones are important characteristics in understanding river–
aquifer interactions occurring during bank infiltration due
to groundwater extraction. These techniques can be
important in transport studies during bank infiltration to
predict the attenuation of pathogens during transport and
artificial recharge. This is especially important to ensure
that adequate soil-retention time requirements are met for
removal of human pathogens and other pollutants. The
Fig. 12 Evolution of the capture zone with time for the pumping well simulations of this study indicate that the pumping rate has
after 1, 3, 6, 9, 12, 15, and 18 days of pumping
more effect on river water travel time than the location of
the pumping well from the river. Operating the pumping
Chen (2001) described a method to determine the per- well at a doubled distance from the river increased the time
centage of river water at the pumping well using the required for the water to travel to the well, but did not
geometry of the river-water influenced zone around the greatly change the capture zone. The water along the
well. The area of the aquifer contained within the farther- meridian line (x = 0, y = 0) only traveled the pumping time
most upstream and downstream pathlines located above period allocated for the simulation (18 days) and more time
and below the meridian pathline (the one with particle’s was needed to reach the well. However, doubling the
location starts at x = 0, y = 0, Fig. 5) and both converging pumping rate greatly expanded the influence zone and
in the pumping well is expressed in terms of the angle reduced the travel time of particles. When the distance
created by the two pathlines. The streamwater migrated between the river and the pumping well was halved and the
along these two above- and below-pathlines has arrived at original pumping rate was used, the travel time for particles
the pumping well while the streamwater migrated in the was reduced, but the capture zone from the stream was not
area outside the two pathlines has not. This angle is mea- changed. When the pumping rate for this half river–well
sured and compared to a 360-angle (complete circle, distance scenario was doubled, more upriver and downriver
which represents the capture zone of the well). As an locations with pathlines converging at the pumping well
example, according to the area of aquifer contained within were added to the influence zone.
the two pathlines of 17.81 m upriver; (pathline 9 in Fig. 5, The examples considered in this study are for partially
of 2.71 days) and 17.65 m down-river; (pathline 140 of 2.55 penetrating streams. As it was shown in the earlier sections
days) both located above and below the meridian pathline of this study, when a river partially penetrates the aquifer,
and form a measured angle of 160 around the well, the the cone of depression created by the pumping well
percentage of river water is estimated to be 44% (or the expanded into the aquifer beneath the river and even to the
ratio of 160/360) of the total water flowing into the well. other side of the river (Figs. 4a, b, 6, 7, 8, 9). Also, the
Because the pathlines are curved around the well, the well was drawing from the aquifer on the other side of the
angles should be measured near the well (for detailed river as the river, which was represented by a specified
description of the method, readers are referred to Chen’s head in this simulation, worked as a water divide. Sopho-
2001 article). cleous et al. (1995) and Chen and Yin (1999) showed that a
This percentage was achieved after 18 days of pumping. river discharges less water when it partially penetrates the
However, as the pumping continues, the percentage of river aquifer. The discharge rate increases with greater river

123
Environ Geol (2008) 55:505–515 515

penetration. It is apparent that influence zones are smaller Acknowledgment The authors are grateful to US Environmental
for partially penetrating rivers when compared with results Protection Agency (EPA) for supporting the comprehensive project
‘‘Riverbank Filtration Effectiveness in an Arid Environment’’ (EPA
from fully penetrating river (for instance Chen 2001). Grant Number R829009) from which this paper was extracted.
The understanding of the relative importance of well
location and pumping rate can also be used to help design
appropriate sampling schedules for bank filtration appli- References
cations. The pathlines and travel times of the river water, as
well as the bank infiltration influence zone, can be used to Abdel-Fattah A (2005) GIS-based three-dimensional groundwater
predict the time when the induced river water gets into the flow and microbial transport modeling for an artificial bank
filtration site in El Paso, Texas. Ph.D. Thesis, University of
well and the ratio of river water to groundwater at the well. Texas at El Paso
The best dates and times to take water samples that better Anderson MP, Woessner WW (1992) Applied groundwater modeling:
represent the infiltrated river water at the pumping well or simulation of flow and advective transport. Academic Press, San
other observation point locations could be determined more Diego, p 381
Chen XH (2001) Migration of induced-infiltrated stream water into
accurately. Samples taken too early may contain no infil- nearby aquifer due to seasonal ground water withdrawal. Ground
trated river water at all because infiltrated river water has Water 39(5):721–728
not yet reached the well. It is recommended that pathlines Chen XH, Yin Y (1999) Evaluation of riverflow depletion for vertical
and travel times be taken into account in scheduling water anisotropic aquifers. J Environ Syst 27(1):55–69
Chen XH, Yin Y (2001) Riverflow depletion: modeling of reduced
samples from the pumping well. Moreover, the ratio of baseflow and induced river infiltration from seasonally pumped
river water to groundwater flowing into the well was wells. J Am Water Res Assoc 37(1):185–195
determined using particle tracking. This provides an idea Hantush MS (1965) Wells near stream with semipervious beds.
about groundwater dilution and mixing ratios, which is an J Geophys Res 70(12):2829–2838
Kim SB, Corapcioglu MY (2002) Contaminant transport in riverbank
important determining factor in assessing bank filtration filtration in the presence of dissolved organic matter and
removal capability. bacteria: a kinetic approach. J Hydrol 266(3):269–283
The results of these numerical simulations agreed well Klute A (1965) Laboratory measurements of saturated hydraulic
with findings of Chen’s (2001) analytical model. Discharge conductivity in saturated soil. In: Black CA (eds) Methods of soil
analysis. Monograph 9 of the American Society of Agronomy.
of river water into the aquifer creates an influence zone that American Society of Agronomy, Madison, pp 210–221
may have distinct water chemistry from other parts of the McDonald MG, Harbaugh AW (1988) A modular three-dimensional
aquifer. If the river water is contaminated and the bank finite-difference ground-water flow model. Techniques of Water
infiltration mechanisms were effective in removing certain Resources Investigations 06-A1, United States Geological Sur-
vey, Reston
contaminant, then the area of the aquifer that contains river Pollock DW (1989) Documentation of computer program to compute
water should have enhanced groundwater quality. Gener- and display pathlines using results from the US Geological
ally, a pumping episode in a bank filtration practice Survey modular three dimensional finite-difference groundwater
generates only a narrow influence zone in the direction flow model. US Geological Survey Open File-Report 89–381,
Denver
perpendicular to the river, and this zone is widest along the Ray C (2002) Riverbank filtration: understanding contaminant
meridian line and its width decreases rapidly upriver and biogeochemistry and pathogen removal. Kluwer Academic
downriver. The influence zone can grow toward the Publishers, Dordrech
pumping well and upriver and downriver with multiple Sontheimer H (1980) Experience with riverbank filtration along the
Rhine River. J Am Water Works Assoc 72(7):386–390
periods of pumping and/or increasing the rate of pumping. Sophocleous M, Koussis A, Martin JL, Perkins SP (1995) Evaluation
The results from the simulations in this study suggest of simplified river-aquifer depletion models for water rights
that migration of infiltrated river water into the nearby administration. Groundwater 33(4):579–588
aquifers is generally slow and most infiltrated river water Wang HF, Anderson MP (1982) Introduction to groundwater
modeling: finite difference and finite element methods. Freeman
does not arrive at the pumping well at the end of an 18-day and Co, San Francisco, p 237
pumping period at a pumping rate of 0.54 m3/min. Infil- Wilson CA, White RR, Orr BR, Roybal GR (1981) Water resources
trated river water may remain in the aquifer for several of the Rincon and Mesilla Valleys and adjacent areas, New
days or months before arriving at the pumping well. Mexico, USA. New Mexico State Engineer Technical Report 43

123