Benito Thesis

Investigation of Contaminant Transport over Multiple Length Scales:
From the Micro to the Macro
by
Pascual H. Benito
B.A. Geology (Amherst College, Amherst, MA) 1994,

M.Sc. (U.C. Berkeley Dept. Civil and Environmental Engineering) 2001
A dissertation submitted in partial satisfaction of the

requirements for the degree of
Doctor of Philosophy
in
Civil and Environmental Engineering
in the
GRADUATE DIVISION
of the
UNIVERSITY OF CALIFORNIA, BERKELEY
Committee in charge:
Prof. Tadeusz W. Patzek, Chair
Prof. Yoram N. Rubin
Dr. Dmitriy B. Silin
Prof. G. Mathias Kondolf
Fall 2008
The dissertation of Pascual H. Benito is approved:
Chair Date
Date
Date
Date
University of California, Berkeley

Fall 2008
Investigation of Contaminant Transport over Multiple Length Scales:
From the Micro to the Macro
c 2008

by
Pascual H. Benito
1
Abstract
Investigation of Contaminant Transport over Multiple Length Scales: From the

Micro to the Macro
by
Pascual H. Benito
Doctor of Philosophy in Civil and Environmental Engineering
University of California, Berkeley
Prof. Tadeusz W. Patzek, Chair
Theses have elements. Isn’t that nice?
Prof. Tadeusz W. Patzek, Chair Date

i
To myself,
Perry H. Disdainful,
the only person worthy of my company.

ii
iii
Contents
Acknowledgments xix
I First Part 1
1 Introduction 3
II Pore Scale 5
2 Pore network investigation of stability and scaling of two-phase immis-

cible displacement fronts with buoyancy forces 7
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 Research Question . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.3 Brief Introduction to Percolation Theory . . . . . . . . . . . . . . 11
2.3 Description of Invasion Percolation Model . . . . . . . . . . . . . . . . . . 13
2.3.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.2 Pore Filling Potential . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4 Model Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5.1 Phase Distribution at Breakthrough . . . . . . . . . . . . . . . . . 16
2.5.2 Invasion Front Geometry . . . . . . . . . . . . . . . . . . . . . . . 17
2.5.3 Capillary Pressure Curves . . . . . . . . . . . . . . . . . . . . . . . 19
2.6 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6.1 Capillary Pressure at Breakthrough . . . . . . . . . . . . . . . . . 21
2.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.8 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.8.1 Limitations and Caveats . . . . . . . . . . . . . . . . . . . . . . . . 26
2.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.10 Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.11 MATLAB code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
iv
2.12 Results from Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.13 Derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
III Multi-Scale 29
IV Regional Scale 31
3 Modeling the Impacts of the Food Processing Industry on Regional

Ground Water Quality in the Lower San Joaquin River Basin 33
3.1 Background and Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.1 Salinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1.2 Characteristics of Food Processing Waste Discharge . . . . . . . . 37
3.1.3 Attenuation of Salinity in the Vadose Zone . . . . . . . . . . . . . 37
3.2 Overview of Objectives and Methodology . . . . . . . . . . . . . . . . . . 39
3.3 Hydrogeological Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.3.1 Regional Hydrology . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 Water Recharge and Pumping . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.5 Numerical ground water Flow Model . . . . . . . . . . . . . . . . . . . . . 48
3.5.1 Ground Water Flow Equation . . . . . . . . . . . . . . . . . . . . . 52
3.5.2 Spatial Discretization . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.5.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.5.4 Hydrogeological Parameters . . . . . . . . . . . . . . . . . . . . . . 58
3.6 Results of Flow Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.7 Salinity Transport Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.7.1 Numerical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.7.2 Solute Transport Parameters . . . . . . . . . . . . . . . . . . . . . 71
3.7.3 Initial and Recharge Concentrations . . . . . . . . . . . . . . . . . 73
3.7.4 Land Application waste water Loading Rates . . . . . . . . . . . . 74
3.7.5 Salinity Transport Modeling Results . . . . . . . . . . . . . . . . . 75
3.7.6 The Effects of Spatial Variability . . . . . . . . . . . . . . . . . . . 82
3.8 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.9 Monte Carlo Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.9.1 Monte Carlo simulation Results . . . . . . . . . . . . . . . . . . . . 95
3.9.2 An alternative Monte Carlo analysis . . . . . . . . . . . . . . . . . 110
3.10 Comparison with field measurements . . . . . . . . . . . . . . . . . . . . . 116
3.11 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.11.1 Future Improvements . . . . . . . . . . . . . . . . . . . . . . . . . 126
V Closure 127
4 Summary and Conclusions 129

v
A Some Ancillary Stuff 137

vi
vii
List of Figures
2.1 Fluid phase distributions at breakthrough for different values of Bond

Number. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Phase distribution at breakthrough for four different bond numbers (Bond
Number increasing counterclockwise from upper left plot). . . . . . . . . 17
2.3 Non-wetting phase (red) displacement front at breakthrough with different
Bond Numbers (Bond Number increasing clockwise from upper left
plot). The dashed blue line represents the mean position of the front
above the inlet face. The width of the displacement front, σ is shown
through the arrows measuring 2σ. . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Capillary Pressure Curves shown for different Bond Numbers (Bond Num-
ber increasing clockwise from upper left plot). . . . . . . . . . . . . . . . 19
2.5 Breakthrough Capillary Pressure Curves shown for different Bond Numbers,NBo .
20
2.6 Breakthrough Capillary Pressure Curves for all simulations shown for
Bond Numbers increasing from NBo = −1 at left to NBo = 1 at right. . . 21
2.7 Breakthrough Capillary Pressure Curves for all simulations shown for
Bond Numbers increasing from NBo = −10 at left to NBo = 10 at right. 22
2.8 Non-wetting, connected-wetting, and trapped wetting phase saturations
at breakthrough versus Bond Number for all simulations. . . . . . . . . . 23
2.9 Non-wetting, connected-wetting, and trapped wetting phase saturations
at breakthrough versus Bond Number for all simulations, zoomed in view
(NBo from -0.1 to +0.1). . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.10 How do macroscopic properties of the two-phase fluid system scale with
the Bond Number? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.11 dimensionless capillary pressure at breakthrough as a function of Bond
Number. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.12 Non-wetting phase saturaiton at breakthrough as a function of Bond Num-
ber. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.13 dimensionless capillary pressure at breakthrough as a function of Bond
Number. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
viii
3.1 Map showing the location and industrial groups of over 200 food proces-
sors in Central Valley who discharge effluent to land and are currently
regulated by the Regional Water Quality Control Board. There are over
300 additional processors currently discharging without official regulation.
It is believed that land discharge practices have impacted ground water
quality at a majority of these sites [Central Valley Regional Water Quality
Control Board, 2006]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2 Digram showing conceptual model for the transport of waste water applied
to land as irrigation [Miller et al., 2008]. . . . . . . . . . . . . . . . . . . . 39
3.3 Aerial view of regional study area (red rectangle), and the food processing
facilities in the model area (red diamonds). The area of the numerical
model is shown by the black rectangle. Currently the model does not
include the region to the west of the San Joaquin River. . . . . . . . . . . 43
3.4 Generalized geologic section and view of the Central Valley as seen looking
Northwest along the axis of the valley (from Page, 1986). . . . . . . . . . 43
3.5 Map showing the structure of overlapping fan systems in the San Joaquin
Basin. The study area is shown by the red dashed rectangle and includes
the overlapping fans from the Stanislaus, Tuolumne, and Merced Rivers
(from Weissmann et al. [2005]). . . . . . . . . . . . . . . . . . . . . . . . . 45
3.6 The major sedimentary units in the Study Area (Source: Phillips et al.
2007). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.7 Map view with conceptual diagram, of regional ground water flow (from
Gronberg & Kratzer, 2006). . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.8 Cross-section with conceptual diagram of the regional ground water flow
showing regional flow and vertical components of flow due to agricultural
pumping and recharge(Source: Gronberg & Kratzer [2006]). . . . . . . . . 47
3.9 Well types and locations. Irrigation and municipal wells account for the
greatest volume of pumping. The median screened depth interval for
Irrigation wells is 51 – 74 m below land surface, for municipal wells it
is between 59-75m, and for domestic wells it is between 44-47 m [Burow
et al., 2004]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.10 Recharge map. Lighter colors correspond to greater recharge per unit
area, darker colors to less. (Source: Phillips et al. [2007]). . . . . . . . . . 50
3.11 Pumpage map. Lighter colors correspond to more greater pumping per
unit area, darker colors to less. (Source: Phillips et al. [2007]). . . . . . . 51
3.12 3-D view showing the active portion of the numerical model grid (vertical
exaggeration = 30×). The Corcoran Clay layer is shown in blue, un-
consolidated sediments in red and consolidated sediments in green. The
elevations of the top of the first layer are derived from the USGS digital
National Elevation Model data set [USGS, 1999]. . . . . . . . . . . . . . . 53
3.13 An East-West cross-section showing the five layers of the numerical grid.
The cells representing the Corcoran clay are shown in blue. The clay layer
pinches out roughly midway through the model area. Vertical exaggera-
tion is 30×. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
ix
3.14 Lateral boundary conditions and included surface water features. . . . . . 55

3.15 Map showing location of pumping wells (blue circles) in the numerical
model (base map from Phillips et al., 2007). . . . . . . . . . . . . . . . . . 56
3.16 Three dimensional view of the pumping well locations in the model. The
yellow circles are located in the center of each grid cell defined as a pump-
ing well. Vertical exaggeration is 30×. . . . . . . . . . . . . . . . . . . . . 57
3.17 Horizontal planar views through three-dimensional model of percentage
coarse-grained sediments in study area (source: Burow et al., 2004). The
red, elongated structures at the top maps suggest the presence of pale-
ochannels. These structures disappear at larger depths, suggesting the
presence of channels only at shallow formations. . . . . . . . . . . . . . . . 62
3.18 3-D view showing the four different lithologic regions in the numerical
model (green=Eastern Area, biege=Western Area above the Corcoran
Clay, blue= Corcoran Clay, yellow=Western Area below Corcoran Clay).
Vertical exaggeration is 30×. . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.19 Map view showing the Western and Eastern lithologic regions in the nu-
merical model. The lateral extent of the the underlying Corcoran Clay
layer is shown by the dashed lines. . . . . . . . . . . . . . . . . . . . . . . 64
3.20 Contour map of the water table for Spring of the Water Year 2000. The
contour interval is 2 meters; red corresponds to areas of higher water table
elevations and blue to lower water surface elevations. . . . . . . . . . . . . 65
3.21 Location of observed water levels for Water Year 2000 from DWR online
source [base map from Phillips et al. [2007]]. The dashed lines show the
lateral extent of the Corcoran clay layer. . . . . . . . . . . . . . . . . . . . 66
3.22 Contour map of computed water table elevations. The contour interval
is 2 meters; red corresponds to areas of higher water table elevations and
blue to lower water surface elevations. . . . . . . . . . . . . . . . . . . . . 67
3.23 Plot of Computed head values versus observed head values (meters). . . . 68
3.24 Difference map between computed water table elevations and interpolated
observed water table elevations. The observations correspond to data from
Spring 2000. Red corresponds to areas where the computed water table
elevations are higher than the observed; and blue to where the computed
values are lower than the observed. The color intensity scale is cut off
at ±5 meters. The location of the observations points are shown by the
small circles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.25 Map showing the land discharge areas (white polygons) for each facility
in the regional study area . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.26 Mass loading curves of Fixed Dissolved Solids (FDS) reaching the water
table over 30 year period due to land application of waste water from the
four primary food processing industries (wine & grape, meat, dairy, and
tomato processing), as estimated by 1-D, reactive vadose zone modeling
[Miller et al., 2008]. These loading curves serve as transfer functions for
salinity transport from the land surface through the vadose zone and to
the water table. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
x
3.27 Concentration of FDS (mg/l) at 30 years in model Layer 1. The colorbar is

cutoff at 1000 mg/l (so red indicated C ≥ 1000 mg/l FDS). The following
ratios were employed: αL /αT = 10 and αL /αV = 10, where αL , αT and
αV denote the longitudinal, transverse and vertical dispersivities. All
simulations assumed αV = αT . The small yellow circles indicate pumping
wells whose screened interval intersect Layer 1. . . . . . . . . . . . . . . . 76
3.28 Map of water table depths less than evapotranspiration extinction depth
(meters). Light blue regions show areas where the depth to water table is
less than or equal to extinction depth. Such areas are prime candidates for
increased FDS due to accumulations of salt via evapotranspiration. This
effect is unrelated to food-processing related discharge, unless if taking
place in such areas. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.29 A difference map of, ∆FDS = FDS(with land discharge)−FDS(without land discharge)
showing the net effect of simulated effluent land discharge at the water
table after 30 years. The color intensity scale is cutoff at 1,000 mg/l, so
concentrations greater than or equal to 1,000 mg/l appears as red. . . . . 79
3.30 A northward looking 3-D view of ∆FDS where the blue iso-surfaces con-
tain the subsurface volumes within which ∆FDS≥ 500 mg/l. The vertical
scale is exaggerated by 30 times and the blue contour lines indicate the
elevation of the water table. . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.31 3-D view of particle tracking flow-path lines emanating from the water
table underneath the land discharge areas. (vertical exaggeration = 30×) 81
3.32 Hypothetical hydraulic conductivity map for layer 1, which is used to an-
alyze the effects of paleo channels on transport of FDS in the saturated
zone. Values shown on the color bar are in meters per day. The red area
represents a hypothetical paleo-channel with a relatively high conductiv-
ity, passing through a land application area in the study area. The other
features on this map can be correlated with Figure 3.19. . . . . . . . . . . 83
3.33 30 year FDS concentration in Layer 1 with paleo-channel. Color bar in
mg/l and cutoff at 1000 mg/l. . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.34 Differences in FDS in Layer 1 with simulated paleo-channel, showing
the ∆FDS= FDS(with land Discharge) – FDS(without Land Discharge).
Color bar in mg/l and cutoff at 1000 mg/l The outline of the simulated
paleo-channel is visible in light blue. This map cannot be used to ana-
lyze the significance of the effects of the channel unless compared with a
similar one obtained assuming no channel, as shown below. . . . . . . . . 85
3.35 This map shows the difference between ∆FDS= FDS(with land Discharge)
– FDS(without Land Discharge), computed with the hypothetical paleo-
channel, and ∆FDS without the paleo-channel, in Layer 1. Color bar is in
mg/l. This shows the effect a coarse-grained paleo-channel would have as
compared to no channel, in the form of differences in concentrations. The
plume concentrations are higher at the leading edge of the plume, but are
smaller at the tail end. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
xi
3.36 This map shows a zoomed-in view of Figure 3.35 the difference between
∆FDS with the paleo-channel and ∆FDS without the hypothetical paleo-
channel, in Layer 1. Color bar in mg/l. This shows the effect a coarse-
grained paleo-channel would have as compared to no channel. The plume
concentrations are higher at the leading edge of the plume, but are smaller
at the tail end. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.37 This map shows the difference of ∆FDS with the paleo-channel and ∆FDS
with out the paleo-channel, in Layer 2. Color bar in mg/l. This shows
the effect a coarse-grained paleo-channel would have as compared to no
channel. The plume concentrations are higher at the leading edge of the
plume, but are smaller at the tail end. . . . . . . . . . . . . . . . . . . . . 88
3.38 This map shows the difference of ∆FDS with the paleo-channel and ∆FDS
with out the paleo-channel, in Layer 3 Color bar in mg/l. ItThis shows
the effect a coarse-grained paleo-channel would have as compared to no
channel. The effect of the channel is greatly diminished at greater depths. 89
3.39 This Figure shows a cross-section view with the color scale representing
the difference of ∆FDS with the paleo-channel and ∆FDS with out the
paleo-channel, in Layer 1. Color bar in mg/l. This Figure shows the effect
a coarse-grained paleo-channel would have as compared to no channel.
The differences in plume concentrations are positive higher at the leading
edge of the plume, and but are negative at the tail end. The vertical
exaggeration is 30x. In the view below we are seeing projections of the x-
and y-axes onto the East-West plane. . . . . . . . . . . . . . . . . . . . . . 90
3.40 Concentration maps showing the net effect of 30 yrs of land discharge for
the three different dispersivity ratio scenarios. For each dispersivity ratio
(columns) the maps shows the net concentration difference, ∆C between
30 yrs with and without land discharge activity for each of the five model
layers (L01-L05), starting from top layer (L01) . The color intensity scale
is in mg/l and is cutoff at 1000 mg/l (all values greater than 1000 are in
red). The white areas on east side the top 2 layers indicate dry cells above
water table. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.41 A close-up of the ∆C maps for the two end points of the dispersivity ratio
scenarios, for the top layer of the model. The maps show the concentration
difference between 30 yrs with and without land discharge activity for each
of the top model layer L01. The color intensity scale is in mg/l and is
cutoff at 1000 mg/l (all values greater than 1000 are in red). . . . . . . . . 94
3.42 This map shows the ensemble mean concentration, CM ean of FDS for each
grid block in layer 1 after 30 years. The color scale is in mg/l and is cutoff
at 1000 mg/l (red≥1000 mg/l, blue = 0 mg/l). All white areas on the left
side indicate dry cell blocks that are above the water table. . . . . . . . . 98
at 1000 mg/l (red≥1000 mg/l, blue = 0 mg/l). All white areas on the left
side indicate dry cell blocks that are above the water table. . . . . . . . . 99
xii
at 1000 mg/l (red≥1000 mg/l, blue = 0 mg/l). . . . . . . . . . . . . . . . 100
at 1000 mg/l (red≥1000 mg/l, blue = 0 mg/l). . . . . . . . . . . . . . . . 101
3.46 This map shows the concentration standard deviation, CStdev for the en-
semble of all realizations for each grid block in Layer 1 after 30 years. The
color scale is in mg/l and is cutoff at 1000 mg/l (red≥1000 mg/l, blue = 0
mg/l). All-white areas to on the left side indicate dry cell blocks that are
above the water table. The drying out or wetting of different cells as the
water table rises and falls with different parameter realizations creates a
fringe of artificially high standard deviations along the edge of the grid
subject to drying out. This is visible as the red to green colors that follow
the “shores” of the dried out cells. . . . . . . . . . . . . . . . . . . . . . . 102
mg/l). All-white areas to on the left side indicate dry cell blocks that are
above the water table. The drying out or wetting of different cells as the
water table rises and falls with different parameter realizations creates a
fringe of artificially high standard deviations along the edge of the grid
subject to drying out. This is visible as the red to green colors that follow
the “shores” of the dried out cells. . . . . . . . . . . . . . . . . . . . . . . 103
mg/l). The standard deviation of the concentration decrease sharply with
increasing depth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.49 This map shows the probability of FDS>500 mg/l in Layer 1 after 30
years. The color scale represents probability with 0 (blue) indicating 0%
probability of exceedence, and 1.0 (red) indicating 100% probability of
exceeding 500 mg/l. The location of the rivers and major streams are
shown as white lines. In addition to the ET hot spots described in the
previous section there are a few artifacts associated with grid blocks that
border the cells that dry out (i.e. cells that are entirely above the water
table). The large standard deviations associated with these grid cells
prone to drying out causes higher exceedence probabilities to be estimated
for these regions. One such artifact is clearly visible in the middle of the
eastern edge of the model. . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
xiii
border the cells that dry out in (i.e. cells that are entirely above the water
table). One such artifact is clearly visible in the eastern most corner of
the model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
border the cells that dry out in (i.e. cells that are entirely above the water
table). The large standard deviations associated with these cells prone to
drying out causes higher exceedence probabilities to be estimated for these
regions. One such artifact is clearly visible in near the eastern most corner
of the model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
probability of exceedence, and 1.0 (red) indicating 100% probability of ex-
ceeding 500 mg/l. The location of the rivers and major streams are shown
as white lines. The large “hot spots” that appear along the San Joaquin
river edge of the model are due to the initial background concentrations
assigned to the model, rather than to specific land application activity. . . 108
probability of exceedence, and 1.0 (red) indicating 100% probability of ex-
ceeding 500 mg/l. The location of the rivers and major streams are shown
as white lines. The large “hot spots” that appear along the San Joaquin
river edge of the model are due to the initial background concentrations
assigned to the model, rather than to specific land application activity. . . 109
3.54 This map shows the probability of ∆C >500 mg/l in layer 1 after 30
exceeding 500 mg/l. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
xiv
3.55 This map shows the probability of ∆C >500 mg/l in layer 2 after 30
exceeding 500 mg/l. In addition to the ET hot spots described in the
border the cells that dry out (i.e. cells that are entirely above the water
table,). One such artifact is clearly visible in the eastern most corner of
the model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.56 This map shows the probability of ∆C >500 mg/l in Layer 3 after 30
exceeding 500 mg/l. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
exceeding 500 mg/l. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
exceeding 500 mg/l. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
3.59 Plot of measured monthly FDS concentrations in shallow groundwater

monitoring wells (Figure 3.60 at the Hilmar Cheese plant for the period
from May 2004 to December 2005. . . . . . . . . . . . . . . . . . . . . . . 117
3.60 Aerial photo showing the outline (yellow) of the primary land application
area at the Hilmar Cheese plant, shallow ground water monitoring wells
(MW-1 – MW-20), and agricultural tile drain (cyan) and the associated
tile drain pump stations (red symbols). . . . . . . . . . . . . . . . . . . . . 118
3.61 Satellite photo showing aerial view of a large dairy processor (cheese &
whey) south of Modesto. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3.62 This contour map shows a salinity plume under an actual large dairy
processor in the study area. The groundwater measurements have been
averaged over from monthly monitoring data from May 2004 to December
2005. What is shown in actually a difference map of FDS above the back-
ground FDS concentration in a reference well (MW-20). The color scale
has been chosen to match that of the model results, with red representing
∆C ≥ 1000 mg/l FDS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
xv
3.63 This contour map shows the simulated salinity plume in the top model
layer under a large dairy processor in the study area, after 21 years of
continuous waste discharge. The color scale shows net difference FDS
above the background FDS concentration, ∆FDS. The upper bound of the
color scale has been set with dark red representing ∆C ≥ 1000 mg/l FDS.
The outer yellow polygon shows the “worst-case scenario” land discharge
area used in the model. The inner yellow polygon shows the actual land
application area used historically since 1985, and is about 1/8 the total
land application area on which the facility is permitted to discharge. . . . 121
3.64 Computed breakthrough curves for FDS in model layer 1 at the center of
plume for four processers from each of the main industry groups. . . . . . 122
xvi
xvii
List of Tables
3.1 Comparison of untreated winery and domestic waste water characteristics

[Central Valley Regional Water Quality Control Board, 2006] . . . . . . . 34
3.2 waste water characteristics (concentration ranges) of the four major food
processing industries. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3 Average values of the hydraulic conductivity calculated using the har-
monic (KH ), geometric (KG ) and arithmetic (KA ) means for the four
lithologic regions shown in Figures 3.18 and 3.19, as well as the ratio of
horizontal to vertical hydraulic conductivities (Khor /Kvert ). (Check new
wording.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.4 Table showing statistics of ∆FDS for each the three dispersivity ratio
scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.5 [Table showing the parameter values used in the Monte Carlos simula-
tions.] The salinity transport model is then run on each of these real-
izations to get a statistical distribution of salinity concentrations at each
point in the model grid. This distribution can be characterized by mean
concentrations and standard deviations of the salt concentrations in the
model. This probability density function can be used to generate a map
which shows the probability of the salt concentration exceeding a given
threshold, for example the MCL, at any given point in the model domain. 96
3.6 This table shows summary statistics for the values of the ensemble Min,
Mean, Max and standard Deviation of the concentration generated in the
Monte Carlo Simulations. . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.7 This table shows summary statistics for the values of the ensemble Min,
Mean, Max and standard Deviation of the concentration differences (∆C) 110
xviii
xix
Acknowledgments
Here I will thank everyone that has helped me along after so many years in grad school!
I would like to extend thanks to Steven Phillips of the U.S.G.S. for giving me access
to the original MODFLOW input files of their regional ground water model for the Lower
San Joaquin Basin.
1
Part I
First Part
3
Chapter 1
Introduction
Every dissertation should have an introduction. You might not realize it, but the intro-
duction should introduce the concepts, background, and goals of the dissertation.
4
5
Part II
Pore Scale
29
Part III
Multi-Scale
31
Part IV
Regional Scale
33
Chapter 3
Modeling the Impacts of the Food

Processing Industry on Regional
Ground Water Quality in the
Lower San Joaquin River Basin
3.1 Background and Motivation

Food processing is a major economic activity in California, serving the state’s multi-
billion dollar agriculture industry. California’s Central Valley contains over 640 food-
processing plants, which consume approximately 7.9×107 m3 of water per year [California
League of Food Processors, 2007]. Solid and liquid wastes from food processing indus-
tries (including canneries; meat, fruit, and vegetable packers; cheese manufacturers; and
wineries) contain significant quantities of organic matter, nitrogen, and salts. Approxi-
mately 80% (520) of these processors discharge their waste water to land [Central Valley
Regional Water Quality Control Board, 2005, 2006]. A large number of food processors
use land application of their waste water as a treatment method, with little or no pre-
treatment, in which ideally, the waste water serves to both irrigate and fertilize crops
grown in the discharge area, and natural soil and plant activity attenuate the waste com-
ponents before they reach the water table [Crites et al., 1974]. Postulated mechanisms
for attenuation include microbial processes such as denitrification, cation exchange, crop
nutrient uptake, and mineral precipitation [California League of Food Processors, 2007].
Ground water monitoring has shown, however, that significant ground water quality im-
pacts have occurred at many larger sites where these wastes have been applied [Central
Valley Regional Water Quality Control Board, 2005], casting doubt on the efficacy of
land application as a mechanism for attenuation. Initial investigations found elevated
salinity levels to be the most common form of ground water degradation near land ap-
plication sites, followed by elevated levels of nitrogen compounds, mainly ammonia and
nitrate.
Salt and nitrate pose significant threats to beneficial uses of ground water. The
34
palatability of domestic drinking water is affected by salinity, and California drinking

water standards include a recommended limit of 500 mg/l (ppm) total dissolved solids
(TDS) and an upper limit of 1,000 mg/l [Cal. Dept. of Public Health, 2006]. Drinking
water delivered to consumers must meet these requirements. Furthermore, individual
components of salt can have adverse health affects. Sulfate levels above 500 mg/l, for
example, can cause diarrhea in sensitive individuals [Central Valley Regional Water
Quality Control Board, 2005]. Salinity can also impact the use of water for irrigation
by reducing plants ability to uptake water from the soil and can cause desiccation of
leaves. Salt-sensitive crops such as strawberries, carrots, and beans can begin to suffer
reduced yields when sprinkler-irrigated by water containing over 450 mg/l TDS [Rhoades
et al., 1992; Ayers & Westcot, 1994]. Nitrate in drinking water can be toxic to humans,
especially infants, by limiting the ability of red blood cells to carry oxygen to the tissues of
the body. This condition in infants can be fatal, and is known as “Blue Baby Syndrome”.
High nitrate levels may also affect the oxygen-carrying ability of the blood in pregnant
women [Cal. Dept. of Public Health, 2008]. California’s primary drinking water standard
are set at 45 milligrams per liter (mg/l) for nitrate as NO3 (equivalent to 10 mg/L for
nitrate as nitrogen) [Cal. Dept. of Public Health, 2008].
As shown in the Table 3.1 food processing waste water is typically much higher
strength than domestic waste water, and therefore has a higher potential to adversely
impact water quality and to create nuisance conditions.
Table 3.1: Comparison of untreated winery and domestic waste water characteristics
[Central Valley Regional Water Quality Control Board, 2006]
Constituent Untreated Winery Untreated Domestic
waste water waste water
pH 2 – 11 6–8
Biochemical oxygen 300 – 30,000 100 – 400
demand (BOD, mg/l)
Total dissolved solids 80 – 7,000 250 – 1,000
(TDS, mg/l)
Total nitrogen (mg/l) 1 – 225 20 – 50
While California State regulations specify that the discharge of waste may not de-
grade ground water quality, they do not currently specify limits on waste application
rates or concentrations [Central Valley Regional Water Quality Control Board, 2005].
Enforcement actions have been taken against multiple food processors [California EPA,
2007], and the Central Valley Regional Water Quality Control Board, who regulates
waste discharge permits in California, has begun re-examining the land-disposal permit-
ting process [Central Valley Regional Water Quality Control Board, 2006]. The State
of California, has recently passed the Porter-Cologne Act [2006], which requires that
the economic impacts must be considered when stricter waste discharge regulations are
imposed [Sunding & Zilberman, 2005]. Understanding how to manage saline waste wa-
ter discharges from this industry can provide important lessons about alternatives for
management of point source discharges of salts from other sources, and can also aid in
35
the development of a comprehensive salt management policy for the entire San Joaquin
Valley, including salts from non-point sources. Because ground water is an important
source of agricultural and municipal water supplies, salts emanating from food processing
facilities must be analyzed in the context of regional water use and treatment systems.
This circumstance calls for a multidisciplinary and comprehensive approach to salt man-
agement. This study was carried out as part of a larger investigation on the water quality
and economic impacts of the food processing industry in the Central Valley, documented
in Sunding et al. [2007a]; Rubin et al. [2007]; Sunding et al. [2007b].
The portion of the study presented here aims to develop and demonstrate a frame-
work for modeling the fate of chemicals of concern from land discharge of waste water
from multiple facilities from different food processing industries, once they have traveled
through the vadose zone and reached the water table. The emphasis is on characterizing
the effects at a regional scale. To develop this framework, the study will focus on the fate
of the salinity component of the waste water, with the intent of expanding this analysis
to nitrogen and organic compounds at a later stage.
Irrigated agriculture causes ground water impacts, even without considering the reuse
of waste water. All irrigation water, with the exception of direct rainfall, contains dis-
solved salts. Salts become concentrated within the soil by evaporation and by transpira-
tion from crops. The addition of more irrigation water results in a build up of salts in the
soil that can harm crops if left unchecked. The arid and semi-arid climates of the Central
Valley make this process more severe as rainfall rates are insufficient to naturally leach
the salts from the soil [Central Valley Regional Water Quality Control Board, 2005]. Salt
accumulation resulting from intensive irrigated agriculture is already a major problem
in the the Central Valley, and the impacts from waste discharge from food processors
take place against this background. The focus of this study is limited to the net impact
of the food processing waste discharge practices, and is not meant to directly address
the broader impacts of irrigated agriculture on water quality in the Central Valley.
3.1.1 Salinity
The term “salinity” encompasses multiple individual ion species and is commonly rep-
resented as either electrical conductivity (EC) or fixed dissolved solids (FDS). FDS is
the direct measurement of the concentration of ionic species, whereas EC represents
an indirect measure of their charge. The major ions comprising salinity are: chloride
(CL− ), calcium (Ca2 +), magnesium (Mg2 +), potassium (K+ ), sodium (Na+ ), ammo-
nium (NH+ 2− 3−
4 ), nitrate (NO3 ), sulfate (SO4 ), and phosphate (PO4 ). Two carbonate
−
3−
species (CO4 and HCO3 ) are also significant contributors. The trace elements alu-
−
minum, manganese, zinc, copper, and iron can also contribute to FDS [Hillel, 2000;
Appelo & Postma, 2005]. It is useful to briefly relate FDS to the more commonly re-
ported, Total Dissolved Solids (TDS). The fixed dissolved solids represents the portion
of the total dissolved Solids that are not volatile under ignition, and often can be taken
to represent the non-organic component of the Total Dissolved Solids (although some
mineral salts can also be volatilized).
36
Figure 3.1: Map showing the location and industrial groups of over 200 food proces-
sors in Central Valley who discharge effluent to land and are currently regulated by the
Regional Water Quality Control Board. There are over 300 additional processors cur-
rently discharging without official regulation. It is believed that land discharge practices
have impacted ground water quality at a majority of these sites [Central Valley Regional
Water Quality Control Board, 2006].
37
Total Dissolved Solids = Volatile Dissolved Solids + Fixed Dissolved Solids (3.1)
When the organic component is low, then TDS and FDS are approximately equal.
For reference, standard seawater has a salinity of 35,000 mg/l TDS. From here on out we
will refer to salinity in terms of FDS, and for the purpose of this study will not consider
the fractions of the individual component species.
3.1.2 Characteristics of Food Processing Waste Discharge

Rubin et al. [2007] characterized the waste streams and land discharge practices from
over 200 food processors with waste discharge permits, based on their monthly data
on effluent concentrations and volumes reported to the Central Valley Regional Water
Quality Control Board (CVRWQCB) from 2003 to 2005. Four food processing indus-
tries: tomato canners, dairy processors (cheese and whey), wineries (and grape juice),
and meat packers, were identified as having the largest impact both in terms of the
volume of waste, and in terms of overall economic output. Representative waste stream
characteristics were developed for each of the these four industries. The characteristics
developed assumed that the processors would discharge their waste water at the maxi-
mum rates and applied over the largest areas allowed by their waste discharge permits.
That is to say the “characteristic” waste streams represent “worst-case” scenarios for a
processor operating in compliance with their permits. The range of values presentative
of these scenarios for each industry is given in Table 3.2.
3.1.3 Attenuation of Salinity in the Vadose Zone

Several key biogeochemical processes in the vadose zone alter the composition of the
waste water before it reaches the water table, including microbial degradation, crop
nutrient uptake, evapotranspiration, mineral precipitation, and ion exchange [California
League of Food Processors, 2007]. These mechanisms are shown schematically in Figure
3.2. The attenuation capacity of the vadose zone above the water table and salinity
mass loadings at the water table were estimated based on the results from a series
of numerical simulations performed using the process-based, multi-component reactive
flow and transport code MIN3P [Mayer et al., 2002]. Each simulation calculated the
flow rate and concentration at the water table over time using a one-dimensional, vertical
vadose zone column model with detailed biogeochemical interactions. To address the high
variability in site conditions and in the waste-stream characteristics, several scenarios
were evaluated by Miller et al. [2008] for each of the four major food industry groups. The
loading rates used in this study correspond to their “Case 2”. This scenario represents
low soil moisture saturation conditions (40-50 percent saturation), a very high ratio of soil
saturated hydraulic conductivity to waste water application rate (Ksat /qwaste ≈ 10, 000),
and a waste water composition high in FDS, total ammonia, and nitrate, while low in
organic carbon. This scenario was created to show the effects of aerobic conditions
throughout the unsaturated zone, and provides a worst case scenario for nitrate loading
Table 3.2: waste water characteristics (concentration ranges) of the four major food processing industries.
Component FDS† Organic C, Ammonium, Nitrate Land
as BOD as TKN Application
Rates qwaste
(mg/l) (mg O2 /l) (mg N/l) (cm/day)
Wineries and grape 82–4300 42–6200 8.4–200 0.1–53 0.03–2.23
processors
Tomato canner 430–4300 170–4400 3.1–520 nd–2.4 1.42–8.48
Dairy Processors‡ nd–4800 570–5700 1.4–140 6.8–80 0.95–2.39
Meat packers 530–3300 18–1700 22–800 nd–2.3 1.57–2.25
†FDS, fixed dissolved solids; BOD, biochemical oxygen demand; TKN, total
Kjehldahl nitrogen; nd, non-detect.
‡Cheese producer data from literature [Danalewich et al., 1998]. All other data
obtained from self-reporting by individual food processors.
38
39
Figure 3.2: Digram showing conceptual model for the transport of waste water applied
to land as irrigation [Miller et al., 2008].
to the aquifer. This scenario was also found to be most representative of conditions
found in monitoring data in ground water wells down-gradient of waste discharge sites
in the study area [Miller et al., 2008]. The details of these vadose zone simulations are
documented fully in Rubin et al. [2007] and Miller et al. [2008]. The results from the
vadose zone model represent the “transfer functions” from waste loadings at the land
surface to the saturated zone ground water model.
3.2 Overview of Objectives and Methodology

The studies of Rubin et al. [2007] and Miller et al. [2008] analyzed and characterized
the flow and transport of salts in soils underlying land application sites. An important
outcome was the computation of the fluxes of salts at the water table under various
scenarios. This study extends their analysis to the saturated zone. Here we will analyze
the flow and transport of salts in the ground water aquifer underlying the land application
sites. Specifically, we will analyze here the temporal and spatial evolution of the ground
water salinity plumes associated with land discharge operations.
We do not intend to explore the evolution of plumes next to each and every land
application sites in the Central Valley. This is unattainable given the scarcity of data,
on the one hand, and the large number of land application sites, on the other. Instead,
we plan to assess the extent of ground water contamination under conditions typical
to the Central Valley. The analysis does not intend to replace the role of site-specific
analysis when evaluating specific sites. Rather, it intends to provide a realistic ensemble
of scenarios that will then be combined to create a realistic set of expectations in terms
of the extent of ground water contamination. Our scenarios intend to cover extreme con-
40
ditions. They were designed as a combination of hydrologic conditions and the effluents’
chemical profiles that are expected to yield concentration values close to the bounds of
the concentrations. As such, the conditions and discharges constituting these worst case
scenarios are not meant to be representative of actual conditions at the specific sites we
consider.
There is no guarantee that the absolute extreme values in concentrations or travel
distances from the discharge sites are captured. The range of concentrations and travel
distances expected in the field depend on local hydrologic conditions and on the scale of
the measurement. Smaller measurement devices will yield increasingly larger concentra-
tion values. The concentration values reported in our study should be viewed as averages
over the scale of the numerical grid elements (which is of the order of hundreds of meters),
and considerable deviations to higher values at smaller scales. The large-scale averages
are in line with the flux-averaged concentrations, which represent the concentration of
the water pumped through wells.
Although they are applied here under assumed scenarios, the methodologies devel-
oped in this study could be applied using site-specific waste discharge data for individual
processors and more detailed, site-specific hydrogeological data. The methodology could
also be used to study other operations which impact ground water quality, such as the
disposal of biosolids, leaching from septic systems, and waste discharges from dairies and
other large scale livestock operations.
To create a meaningful ensemble of scenarios, we chose to focus on a limited section of
the Central Valley, namely the Lower San Joaquin River Basin (see 3.3). We refer to this
area as the Representative Area (RA). This study area is relatively well-characterized in
terms of hydrological and hydrogeological conditions as it has been studied extensively
by researchers from the United States Geological Survey (USGS) [Burow et al., 2004;
Phillips et al., 2007]. This figure also identifies the locations of the food processors
modeled in this study. An extended discussion of the food processors is provided in
Rubin et al. [2007]. Each of these facilities was studied through its periodic waste
discharge and ground water monitoring reports and the relevant documentation in its
permit for operating a land discharge site. The wide range of food-processing facilities
in the study area (Figure 3.3) reflect the diversity of food processors operating in the
Central Valley.
Focusing on an RA was necessary since analyzing the entire Central Valley was unfea-
sible given the limited budget and time available for this project. To be representative of
the Central Valley and to allow for subsequent findings to be generalized, the RA needed
to be large enough to cover a sufficient number of processors, processes and geologi-
cal conditions. It also needed to be large enough geographically to cover a wide range
of conditions encountered in the Central Valley. A benefit of working with an RA is
that it allowed us to focus on an area whose hydrogeological conditions are much better
documented than in other sections of the Central Valley, thus eliminating the need to
speculate about conditions in poorly documented regions. As will be discussed in more
detail below, we developed a numerical model for analyzing flow and solute transport
processes at the RA considered in this study. The model is subjected to loading of solute
fluxes from a large number of food processing facilities located in the RA. These fluxes
41
were computed in previous work [Rubin et al., 2007; Miller et al., 2008]. The study
area was subjected to salinity loading at various locations (marked on Figure 3.3), from
facilities representing both small and large operations in terms of land discharge areas
and flow rates. The salt loads used to simulate the land discharge are representative of
a variety of food processing industrial groups,namely, the four large-impact industrial
groups of wineries, tomato processors, meat processors and cheese makers. An extended
discussion of the rationale leading to salinity loads used is provided in Rubin et al. [2007];
Miller et al. [2008], and is not repeated here for brevity. However, it is worth emphasiz-
ing that the “design loads” used to simulate the salinity loading at the land discharge
sites represent extreme conditions. We refer to these loads as our best and worst case
scenarios, as was discussed in Rubin et al. [2007]; Miller et al. [2008]. An extreme “design
load”would entail an extreme environmental impact to the ground water and would thus
be useful to identify the applicable limits and constraints. Conversely, when an extreme
“design load” fails to make a notable impact, this provides an indication that we should
not expect significant impacts for loads that are not at the extremes.
A simulation period of 30 years was chosen for modeling the transport of salt reach-
ing the ground water table. A thirty year period was deemed sufficiently long to allow
for the environmental impacts to ground water to become noticeable, and at the same
time for land use changes to take place, such that the environmental impacts could pos-
sibly translate into economic impacts, addressed in Sunding et al. [2007b]. Numerical
simulations over a longer period of time would require the consideration of effects such
as climate change, as well as consumer and technological trends, and all this could intro-
duce significant uncertainty as well as a diversion of the discussion away from the issues
being addressed in this project. The 30 year simulation period is also consistent with
the planning horizon of the California State Water Plan, which extends to Year 2030.
An important aspect to consider in this study is model uncertainty. Spatial viability of
hydrological conditions and hydraulic parameters, coupled with scarcity of data, makes
our predictions subject to a large uncertainty. Although the study area is relatively
well documented (compared to other regions of the Central Valley), there is still large
uncertainty associated with the various model parameters that are needed. To capture
this uncertainty, we formulated our study in a probabilistic framework. We opted to
do that by modeling dependent variables such as the ground water concentrations at
various times and locations as random variables. This implies that they are modeled
through their statistical moments such as means, variances and probability distribution
functions. We will provide our results using, for example, quantitative assessment of
the probability for the concentration to exceed a certain threshold value at a certain
time. An extended discussion on probabilistic methods is provided in Rubin [2003].
Note that a statement such as “the probability for the concentration to exceed a value
C0 is equal to 80% implies that 8 out of ten sites with similar conditions are expected
to show concentrations exceeding C0 . As a way to model the effects of uncertainty in
the groundwater and contaminant transport parameters, a Monte Carlo was employed
method, as will be described in 3.9. It is worth noting that the Monte Carlo method
was only used for analysis of the saturated zone but not for the unsaturated zone. Rel-
ative to the unsaturated zone, flow and transport processes in the saturated zone are
42
less complex to model, and the needed parameters and their statistical distributions are
much better documented in the literature; in this case a Monte Carlo analysis can be
performed relatively easily to yield meaningful results. This is unlike the unsaturated
zone, however, where knowledge of the many parameters needed for complex transport
models is painfully scarce; in this case a Monte Carlo analysis is less useful. Another
issue we wish to address in this introduction section is that of subsurface variability of
parameters such as the hydraulic conductivity. This subject has been explored exten-
sively in recent years (see for example Kitanidis [1990] and Rubin [2003], which includes
many additional references). In general, one would like to be able to characterize the
flow domain at the highest resolution possible, which is especially important for the case
of modeling point sources of contamination. Such an undertaking is very demanding
in terms of field data acquisition and in terms of numerical modeling. Fortunately, the
land application sites cover large areas that are much larger than the typical length
scales of heterogeneity of the relevant properties. In such situations, it is appropriate
to use effective properties rather then resorting to detailed subsurface characterization
data. Effective properties are useful for modeling the mean behavior, that is, the large-
scale trends of the concentration field. While effective properties are not expected to
produce the small scale variability of the concentration–perhaps more importantly–they
are successful in their ability to estimate the flux-averaged concentration, which is the
concentration that is measurable at a pumping well.
3.3 Hydrogeological Setting

The California Central Valley is the largest ground water basin in the state, in terms of
both total storage capacity and high utilization rate. Annual extraction rates are on the
order of 11 km3 per year ( 9,000,000 acre-feet), largely to support California’s agriculture
industry. The basin is recharged by direct precipitation and infiltration along the beds
of the San Joaquin and Sacramento river systems, whose discharge is primarily from
rainfall and snowmelt in the Sierra Nevada [Ferriz, 2001].
The San Joaquin Valley is more than 400 km long, ranging in width from 40 to 90
km, and it makes up the southern two-thirds of the Central Valley. The San Joaquin
Valley is bordered on the east by the Sierra Nevada, which rises up to elevations of more
than 4,200 meters, and on the west by the Coast Range, which is of more moderate
elevation. The San Joaquin Valley is made up of two sub-regions: the San Joaquin Basin
in the north, which drains to the Sacramento-San Joaquin Delta; and at the southern
end an interior drainage called the Tulare Basin, named after a Pleistocene Lake that
occupied most of the area [Page, 1986; Bertoldi, 1991]. The valley floor is underlain
by an asymmetric structural trough filled with marine and continental sediments up to
10 km thick, underlain by granitic bedrock of the Sierra Nevada batholith [Page, 1986;
Bertoldi, 1991]. Figure 3.4 shows a generalized geologic section and view of the Central
Valley as seen looking Northward along the axis of the valley.
Several large overlapping fluvial fans have formed along the eastern margin of the
San Joaquin Basin where their rivers exit the Sierra Nevada. Seven of these “Eastern”
43
Figure 3.3: Aerial view of regional study area (red rectangle), and the food processing
facilities in the model area (red diamonds). The area of the numerical model is shown
by the black rectangle. Currently the model does not include the region to the west of
the San Joaquin River.
Figure 3.4: Generalized geologic section and view of the Central Valley as seen looking
Northwest along the axis of the valley (from Page, 1986).
44
fans–the Mokelumne, Stanislaus, Tuolumne, Merced, San Joaquin, Kings and Kaweah
River fans–have drainage basins that connect to glaciated portions of the Sierra Nevada
(Figure 3.5). The fans in the northern portion are smaller due to the narrower width of
the valley compared to the southern area [Weissmann et al., 2005]. Smaller fans have
formed along the western edge of the valley along the base of the Coast Range, referred
to as the “Western” fans. Although no two fluvial fans are identical, they display similar
sedimentary structures. They can generally be divided into three physiographic regions:
the western fans, the eastern fans, and the basin deposits as illustrated in Figure 3.6
[Burow et al., 2004]. The sediments on the western side are unconsolidated and have
relatively higher hydraulic conductivities. The eastern edge is more consolidated and is
characterized by lower hydraulic conductivity. The Corcoran Clay layer underlies the
western half of the regional study area and is present at depths of about 30 to 40 meters
below ground surface. The Corcoran Clay layer is a member of the regionally extensive
lacustrine “E”-clay layer underlying a large portion of San Joaquin and Tulare basins.
This pattern of repeating fans suggests that studying the transport of salinity over
a single or a couple of such fans can provide insight about the transport of salinity at
other locations, and that studying salinity transport over a limited section of the Central
Valley seems like a reasonable means for obtaining general results.
3.3.1 Regional Hydrology

Under natural conditions prior to intensive agricultural development, ground water in
the Central Valley was primarily recharged in the upper parts of alluvial fans where the
streams enter the valley (Figure 3.7 and Figure 3.8). Ground water followed the dip
of the underlying basement rock and flowed southwest towards the San Joaquin River,
around which Artesian conditions existed [Page, 1986; Bertoldi, 1991].
Prior to agricultural development the Corcoran Clay is believed to have acted as
an effective confining layer. Most of the ground water flowed laterally through the
aquifer, with a small of amount of upward leakage through the confining layer. The
consequent drilling of large diameter wells through the Corcoran Clay with perforations
above and below the clay unit (see Figure 3.9 showing well types, locations, and screened
interval depths), however, has diminished the effectiveness of the layer as a confined unit
[Bertoldi, 1991], allowing the passage of water between the upper aquifers and the deeper
units. Reductions in the hydraulic head of the confined aquifer systems due to large scale
pumping has resulted in a net downward movement of water through well casings. In
some places, the amount of water that flows downward through one large-diameter well
has been estimated to be equivalent to the natural leakage through the “E-clay” over
an area of approximately 7 square miles. During the peak of the withdrawal season, the
net downward flow may be, on average, as much as 0.3 cubic foot per second per well
[Planert et al., 1995].
Extensive ground water development in the valley (Figure 3.9), coupled with the
increase in effective vertical conductivity of the Corcoran Clay layer have greatly altered
the regional flow patterns. The current flow regime is strongly affected by agricultural
pumping and irrigation, and these local effects lead to strong vertical gradients. The
45
Figure 3.5: Map showing the structure of overlapping fan systems in the San Joaquin
Basin. The study area is shown by the red dashed rectangle and includes the overlapping
fans from the Stanislaus, Tuolumne, and Merced Rivers (from Weissmann et al. [2005]).
46
Figure 3.6: The major sedimentary units in the Study Area (Source: Phillips et al.
2007).
47
Figure 3.7: Map view with conceptual diagram, of regional ground water flow (from
Gronberg & Kratzer, 2006).
Figure 3.8: Cross-section with conceptual diagram of the regional ground water flow
showing regional flow and vertical components of flow due to agricultural pumping and
recharge(Source: Gronberg & Kratzer [2006]).
48
induced gradients in the relatively high pumping zones (discussed in Section 3.4) may
overwhelm the natural gradient by introducing a strong vertical gradient.
3.4 Water Recharge and Pumping

The surface recharge and ground water pumping rates were derived from data from
Burow et al. [2004]. Recharge includes water infiltration from irrigation and precipita-
tion. It also includes leakage from reservoirs based on estimates provided in Phillips
et al. [2007]. Irrigation rates were estimated based on crop type and land use data.
Sixty percent of recharge is from irrigation and forty percent is from precipitation. Of
the total irrigation, sixty percent is from surface water deliveries and the rest from pump-
ing. This information is important for calculations of the chemical composition of the
recharge water, which will be discussed in the next section.
Pumping includes municipal and agricultural wells. Pumping rates were determined
in Burow et al. [2004], based on data provided by water districts in the study area and by
estimates determined from irrigation demands. Domestic wells make a relatively small
contribution to the total pumping rates and were ignored, following the recommendations
of Burow et al. [2004].
Recharge and pumpage estimates were developed for irrigation districts, which were
further broken down into 64 subregions, shown by the various regions in Figures 3.10
and 3.11. Figure 3.10 shows a map of the spatially variable recharge applied at the water
table in the model. Lighter colors correspond to greater recharge, darker colors to less.
Figure 3.11 shows a map of the estimated pumping per unit area,or pumpage. Lighter
colors correspond to more greater pumping per area, darker colors to less. While this
map does not identify depths of the well screens interval, but this data is implemented
in the model.
3.5 Numerical ground water Flow Model

This section provides an overview of the numerical model constructed for the regional
study area shown in Figure 3.3. The numerical model follows closely the principles and
conceptual models employed in Burow et al. [2004] and Phillips et al. [2007], and uses
information obtained directly from the USGS numerical model’s input files. This section
discusses the numerical model’s spatial discretization, boundary conditions, sources and
sinks, and the hydrogeologic parameters.
The model is three-dimensional, to allow for modeling the effects of vertical gradients.
It is a steady-state model based on information from water-year 2000 (October 1, 1999 –
September 30, 2000) when the ground-water system was in a quasi-steady-state condition
[Burow et al., 2004; Phillips et al., 2007]. Variables such as irrigation, precipitation and
pumping rates were computed on a monthly basis, and were then averaged to yield annual
rates which were assumed to be constant over time. We are using MODFLOW-2000
[Harbaugh et al., 2000], a program that solves the three-dimensional ground water flow
equation using a finite-difference numerical method, in combination with GMS (ground
49
Figure 3.9: Well types and locations. Irrigation and municipal wells account for the
greatest volume of pumping. The median screened depth interval for Irrigation wells is
51 – 74 m below land surface, for municipal wells it is between 59-75m, and for domestic
wells it is between 44-47 m [Burow et al., 2004].
50
Figure 3.10: Recharge map. Lighter colors correspond to greater recharge per unit area,
darker colors to less. (Source: Phillips et al. [2007]).
51
Figure 3.11: Pumpage map. Lighter colors correspond to more greater pumping per unit
area, darker colors to less. (Source: Phillips et al. [2007]).
52
water Modeling System), which provides pre/post-processing capabilities and integration

with GIS layers.
The steady-state strategy was chosen mainly for computational efficiency, and due to
the difficulty in obtaining sufficiently detailed data. The steady-state modeling strategy
is an approximation to a flow regime that is in fact transient in nature. The potential
consequences of making this approximation are as follows.
Increased recharge due to temporarily reduced pumping rates (or increased infiltra-
tion rates) could raise the water table and reduce ground water contaminant concentra-
tions (and vice versa). Our model provides an annually averaged concentration. Another
important effect to consider is the potential for transient reversals of gradients at the
local scale. The induced changes in the flow direction due to transients can, under some
circumstances, enhance mixing of contaminants. These reversals can lead to increases
in the area affected by contaminants, but they also lead to dilution and hence reduc-
tion in the concentration due to the larger volume of ground water that mixes with the
contaminants.
3.5.1 Ground Water Flow Equation

MODFLOW [Harbaugh et al., 2000] solves the following partial-differential equation
describing transient ground water flow in a three-dimensional, heterogeneous, anisotropic
medium, in which the axes of the hydraulic conductivity tensor are aligned with the
model grid coordinate system:

∂ ∂h ∂ ∂h ∂ ∂h ∂h
Kxx + Kyy + Kzz + W = Ss (3.2)
∂x ∂x ∂y ∂y ∂z ∂z ∂t
where, x,y, and z are principal coordinates of the grid system, Kxx ,Kyy , and Kzz are
the diagonal components of the hydraulic conductivity tensor [LT−1 ], h is the hydraulic
head [L], W is a volumetric flux per unit volume of aquifer, representing sources and
sinks of water [T− 1], Ss is the specific storage coefficient of the porous material [L− 1],
and t is time [T]. Additional assumptions include constant and uniform fluid (water)
density, viscosity, and temperature.
MODFLOW solves Equation 3.2 using the finite-difference method in which the
groundwater flow system is divided into a grid of cells. For each cell, there is a sin-
gle point, called a node, at which they hydraulic head is calculated. The finite-difference
equation method used is documented in McDonald & Harbaugh [1988].
3.5.2 Spatial Discretization

The spatial discretization of the model was chosen to align with the USGS numerical
ground water model being developed for the study area (Phillips et al., 2007). Figure
3.12 and 3.13 provide an areal view and a vertical cross-section view of the numerical
grid. The lateral dimensions of the grid area are 61.20×54.8 kilometers. In plan view, the
model area was discretized into a grid of 153×137 blocks with the individual grid block
dimensions of 400 meters by 400 meters. This block size is approximately the size of the
53
Figure 3.12: 3-D view showing the active portion of the numerical model grid (vertical
exaggeration = 30×). The Corcoran Clay layer is shown in blue, unconsolidated sedi-
ments in red and consolidated sediments in green. The elevations of the top of the first
layer are derived from the USGS digital National Elevation Model data set [USGS, 1999].
smallest land discharge area, which is important because it allows us to model accurately
the loading of the discharge. In the vertical direction the model area was discretized into
five layers, generally of increasing thickness with depth, thus allowing greater resolution
at the depths near the water table. The layer thickeness were defined by combining layers
from the 16-layer model of Phillips et al. [2007], resulting in a coarser vertical resolution.
The Corcoran clay layer (see Figure 3.12 and Figure 3.13) is modeled as a distinct layer
because of the large conductivity contrasts with the adjacent layers. The region of the
numerical grid to the west of the San Joaquin River was not modeled, so these grid
blocks were set as inactive. The elevation of the top layer is taken by mapping from the
USGS digital elevation model [USGS, 1999] to the grid. The total thickness of the model
ranges between 220 and 430 meters, as shown in a representative cross-section in Figure
3.13. With the exception of the Corcoran Clay layer whose thickness is constrained by
well logs [Page, 1986], the thickness of the layers increases from top to bottom.
The orientation of the model layers follows the dip angle of the sediments. The model
axes are generally aligned with the geologic structures of the valley to allow modeling of
the hydraulic conductivity and dispersivity as diagonal tensors.
54
Figure 3.13: An East-West cross-section showing the five layers of the numerical grid.
The cells representing the Corcoran clay are shown in blue. The clay layer pinches out
roughly midway through the model area. Vertical exaggeration is 30×.
3.5.3 Boundary Conditions

Boundary conditions are assigned for the lateral boundaries of our study area as well as at
the water table, as shown in Figure 3.14. The easternmost edge of the model at the base
of the Sierra foothills was defined as a no-flow boundary. This is based on the assumption
that ground water flow from the foothills is negligible when compared to the flow through
the other boundaries. The northern and southern edges were modeled as general-head
boundaries using the MODFLOW General Head Boundary package [Harbaugh et al.,
2000]. This boundary condition allows for flux into or out of the model by defining a
constant head at a known distance from the model boundary. The western edge of the
model along the San Joaquin River was also modeled using the general-head boundary
package. This allows for pumping induced cross-flow from under the river. Figure 3.14
shows a map of the lateral boundary conditions.
The bottom of the aquifer is set at depths determined by the USGS that represent
topographic variability and the general dip of the Corcoran Clay and which is assumed
to be deep enough to avoid any undesirable modeling effects on the flow regime [Phillips
et al., 2007].
The water table was modeled as a flux boundary, based on the spatial distribution
of recharge shown in Figure 3.10. Following the findings of Phillips et al. [2007], the
recharge rate for all sub-regions shown Figure 3.10 had to be reduced by 10 percent in
order to achieve numerical convergence in the flow model.
Pumping Wells
Pumping wells were modeled using the MODFLOW Well Package [Harbaugh et al., 2000].
The locations and pumping rates of the wells are based on the USGS model [Phillips
et al., 2007], as shown in Figure 3.15 and Figure 3.16. Due to the large number of wells
considered in this study, no grid refinement was carried out around pumping wells. If
well locations were not known, then wells were evenly distributed within a sub-region
and pumping rates defined to add up to total estimated pumpage for the sub-region, as
described in Phillips et al. [2007].
55
Figure 3.14: Lateral boundary conditions and included surface water features.
56
Figure 3.15: Map showing location of pumping wells (blue circles) in the numerical model
(base map from Phillips et al., 2007).
To approximate the depths of the screened intervals the pumping rates were only
assigned to the layers of the model that most closely matched the screened intervals.
The current model has only 5 layers (a decision made in order to facilitate the Monte
Carlo analysis) as compared to the 16 layers in the USGS model, so the depth resolution
of the screened intervals is limited by the layer thicknesses.
Evapotranspiration
Evapotranspiration (ET ) was simulated using the MODFLOW Evapotranspiration Pack-
age Harbaugh et al. [2000]. The evapotranspiration package models ET with a maxi-
mum rate of ET0 at the land surface, decreasing linearly to zero at extinction depth.
The USGS value of ET0 = 0.043 m/day was used and is in line with the annual average
determined from data from the California Irrigation Management Information System
(CIMIS) website for the water-year 2000 [California Dept. of Water Resources, 2007],
57
Figure 3.16: Three dimensional view of the pumping well locations in the model. The
yellow circles are located in the center of each grid cell defined as a pumping well. Vertical
exaggeration is 30×.
ET0 = 0.0039 m/day. The extinction depth of 2.1 meters was also used based on the
value used in the USGS model [Phillips et al., 2007].
Ground Water - Surface Water Interactions

Leakage from the three reservoirs along the eastern edge of the model was simulated by
applying additional recharge based on estimated leakage rates from the reservoirs. The
river/ground water interactions were modeled in the same fashion as in the USGS model.
The San Joaquin River and the lower reaches of it’s tributaries are generally taken as
being gaining reaches, where the ground water table is above the river bottom, whereas
the upper reaches of the tributaries have been found to be under losing conditions,
with the water table below the river bottom [Jones and Stokes Associates & Mussetter
Engineering Inc, 2002; Corporation, 2002; Phillips et al., 2007], although this can vary
locally and seasonally in some areas. Gaining reaches were modeled with General Head
Boundary conditions. The prescribed heads were based on river stage measurements,
linearly interpolated between gaging stations, and the river bottom conductances were
taken from the USGS model [Phillips et al., 2007]. The losing reaches of the rivers (in
the upslope regions) were modeled by applying recharge over the losing reaches (0.005
m/day). The locations of the different surface water zones are shown in Figure 3.14.
58
3.5.4 Hydrogeological Parameters

For a steady-state model, the most important parameter to consider is the hydraulic
conductivity. Two strategies are commonly pursued in this regard. The first calls for
high-resolution modeling of the spatial distribution of the hydraulic conductivity, and the
second allows for modeling of the conductivity field through the effective conductivity.
Each of these two strategies is suited for different situations, due to their advantages and
limitations. An extended discussion of this topic is provided in Rubin [2003, Chap.5],
which includes numerous references to previous related work.
We modeled the hydraulic conductivity field through the effective conductivity con-
cept. Such an approach is suitable for computing average fluxes in large-scale applica-
tions, and when modeling contaminant transport from large, distributed sources, such
as is the case for land application sites. The most general brackets on the values of the
effective hydraulic conductivity, Keff are given by the Batchelor bounds, in which the
lower limit is equal to the harmonic average (KH ) and the upper limit to the arithmetic
average (KA ) of the hydraulic conductivity [Batchelor, 1974]. The geometric mean of
the conductivity, KG , can be viewed as a convenient choice for an effective conductivity
somewhere in the Batchelor bounds range, although it is not rigorous because strictly
speaking the geometric mean is the effective conductivity for two-dimensional flow and
for a symmetrically distributed distribution of the natural logarithm of the conductivity,
ln K)[Rubin, 2003]. Rather than limiting ourselves to a single value, we chose to work
with a range of values within those bounds.
The grid block hydraulic conductivities in the USGS model of Phillips et al. [2007]
were distributed spatially based on geostatistically interpolated sediment texture maps
(percentage coarse grained) from thousands of driller’s well logs as shown in Figure 3.17.
Phillips et al. [2007] then used a mixing law to calculate the horizontal and vertical
conductivities based on the percentage of two textural end members representing 100%
coarse-grained and 100% fine-grained material. The hydraulic conductivity of these two
end members are Kcoarse and Kf ine respectively, and these values were varied as part of
their model calibration.
Phillips et al. [2007] divided the alluvial deposits into three lithologic subregions.
Together with the Corcoran clay, these three subregions where used to divide the hy-
draulic properties of our model into four broad lithologic subregions illustrated in Figures
3.18-3.19 and listed below:
1. Corcoran Clay Layer
2. Western Area above Corcoran Clay (downslope region, representing younger, less
consolidated deposits)
3. Western Area below Corcoran Clay (downslope region, representing younger, less
consolidated deposits)
4. Eastern Area (upslope region, representing older, more consolidated fan deposits)
59
Lithologic Subregion KH KG KA Khor /Kvert

[m/day] [m/day] [m/day]
Western Area (above 19 40 84 300
Corcoran Clay)
Western Area (below 12 25 54 300
Corcoran Clay)
Eastern Area (entire 111 24 50 300
thickness)
Corcoran Clay 1.3 × 10−3 1.3 × 10−3 1.3 × 10−3 1
Table 3.3: Average values of the hydraulic conductivity calculated using the harmonic
(KH ), geometric (KG ) and arithmetic (KA ) means for the four lithologic regions shown in
Figures 3.18 and 3.19, as well as the ratio of horizontal to vertical hydraulic conductivities
(Khor /Kvert ). (Check new wording.)
Phillips et al. [2007] provide histograms showing the distribution of their calibrated
hydraulic conductivities values for all the lithologic sub-regions in the model except the
Corcoran clay, which they assumed to be of uniform and isotropic conductivity. The
calibrated value from Phillips et al. [2007] of 1.3×10−3 m/day) was used. This is about
three orders of magnitude larger than the conductivity values determined from core-scale
samples of the Corcoran Clay [Page, 1977], most likely due to enhanced conductivity
because of a large number of wells drilled through and screened across the clay layer,
and also possibly due to fracturing [Williamson et al., 1989; Phillips et al., 2007].
These data were used to compute effective conductivity tensors for each of the four
lithological subregions. As stated earlier, the Batchelor bounds for the effective conduc-
tivity are defined by the harmonic and arithmetic averages of the conductivities, with
the geometric mean lying somewhere next to the middle of this range. Log-normal dis-
tributions were fit to each of the conductivity histograms, and the harmonic, geometric,
and arithmetic mean of the log of the conductivity were computed for each zone and
used to assign a mean value to each subregion in the current model, using equation 5.8
from Rubin [2003]. A variance was also determined for each zone. These statistics will
be used to develop different conductivity realizations for Monte Carlo simulations. Table
3.3 below provides the arithmetic, geometric, and harmonic means of the four lithologic
regions, as well as the ratio of horizontal to vertical hydraulic conductivities.
3.6 Results of Flow Model

Figure 3.22 shows a contour map of the water table elevations computed by the model for
the case where for each lithologic region Kef f = KG . Figure 3.20 shows a contour map
generated from water level observations from over 500 wells for Spring 2000 obtained
from the Department of Water Resources online data repository [DWR, 2006]. Figure
3.15 identifies the locations of these observed water surface elevations. In general, the
model captures the general features and trends of the observed water elevations, although
60
not perfectly. The local ground water depression under the city of Modesto, for example,
is not well captured, and while the depression in the southwestern corner of the model is
captured, the computed heads are slighly higher than the observed heads. Figure 3.23 is
a scatter plot of the observed and computed water table elevations. Perfect agreement
between the observed and computed heads would result in the points falling along a
straight line of unit slope. The match between observed and computed values is not
perfect, but the general trend is captured. It appears that in general the computed
heads are higher than the the Spring 2000 observations.
The model error can be quantified by the average error, ē, and the root mean square
error (RM SE):
n
1X
ē = (hobs,i − hcomp,i ) (3.3)
n
i=1
v
u n
u1 X
RM SE = t (hobs,i − hcomp,i )2 (3.4)
n
i=1
where hobs,i and hcomp,i are the ith observed and computed hydraulic head pair,
respectively, and n is the total number of head observations.
The RM SE is a measure of the error magnitude, whereas the average error indicates
if the computed heads are overall higher (ē > 0) or lower (ē < 0) than the observed
values. The RM SE was 3.8 meters, and the average error was +1.3 meters, with n=526,
indicating that in general the model is predicting water levels that are slightly higher
than the observed values.
Figure 3.24 shows a difference map between computed water table elevations and in-
terpolated observed water table elevations. Red corresponds to areas where the computed
water table elevations are higher than the observed; and blue to where the computed
values are lower than the observed. The color intensity scale is cut off at ±5 meters. The
location of the observations points are shown by the small circles. The map shows that
the largest model over-predictions of head generally occur in the upland eastern area
of the model and in the area below the city of Modesto. The Monte Carlo simulations
(that will be discussed in more detail in Section 3.9) in which different values of Ke f f
were assigned to the different regions, showed that that the standard deviation of the
head (over all generated random realizations) was largest in the up-slope Eastern Region.
This indicated that the hydraulic heads in this region of the model are the most sensitive
to variations in the value of Kef f .
It is not unexpected that the computed heads should not perfectly match the ob-
served heads for several reasons. First, the resolution of the numerical model, basically
the numerical grid block size, is different from the actual measurement scale. Second,
water table elevations are affected by local variations in hydrologic input and aquifer
properties, which cannot be captured by a numerical model due to scarcity of the data
needed for fine-scale characterization. Next, the model assumed steady-state conditions,
whereas the measured water table elevations reflect the actual conditions at the day of
61
the measurement. Furthermore, the water level observations represent hundreds of wells
all screened at different depths, and with different lengths of screened intervals, this
will also make harder to compare directly the observed water levels to the computed
(the depths of the screened intervals for these observation wells were not publicly avail-
able due to homeland security concerns). Given vertical gradients due to recharge and
deeper pumping, it is expected that wells screened at deeper intervals would have lower
hydraulic heads, and this is an additional source of mismatch. However, we note the
general existence of a strong correlation between the computed and observed water table
elevations. This is a solid indication with regard to the suitability of our model for the
general purpose of this analysis.
62
Figure 3.17: Horizontal planar views through three-dimensional model of percentage

coarse-grained sediments in study area (source: Burow et al., 2004). The red, elongated
structures at the top maps suggest the presence of paleochannels. These structures dis-
appear at larger depths, suggesting the presence of channels only at shallow formations.
63
Figure 3.18: 3-D view showing the four different lithologic regions in the numerical model
(green=Eastern Area, biege=Western Area above the Corcoran Clay, blue= Corcoran
Clay, yellow=Western Area below Corcoran Clay). Vertical exaggeration is 30×.
64
Figure 3.19: Map view showing the Western and Eastern lithologic regions in the nu-
merical model. The lateral extent of the the underlying Corcoran Clay layer is shown
by the dashed lines.
65
Observed Heads
Figure 3.20: Contour map of the water table for Spring of the Water Year 2000. The
contour interval is 2 meters; red corresponds to areas of higher water table elevations
and blue to lower water surface elevations.
66
Figure 3.21: Location of observed water levels for Water Year 2000 from DWR online
source [base map from Phillips et al. [2007]]. The dashed lines show the lateral extent
of the Corcoran clay layer.
67
Computed Heads
Figure 3.22: Contour map of computed water table elevations. The contour interval is 2
meters; red corresponds to areas of higher water table elevations and blue to lower water
surface elevations.
68
K =K : Computed vs. Observed Heads

eff G
50
45 Head
1:1
40
35
Computed Head [m]
30
25
20
15
10
0
0 5 10 15 20 25 30 35 40 45 50
Spring 2000: Observed Head [m]
Figure 3.23: Plot of Computed head values versus observed head values (meters).
69
Computed Heads - Observed Heads
Figure 3.24: Difference map between computed water table elevations and interpolated
observed water table elevations. The observations correspond to data from Spring 2000.
Red corresponds to areas where the computed water table elevations are higher than the
observed; and blue to where the computed values are lower than the observed. The color
intensity scale is cut off at ±5 meters. The location of the observations points are shown
by the small circles.
70
3.7 Salinity Transport Modeling
3.7.1 Numerical Model
MT3DMS [Zheng & Wang, 1999] was used as the numerical engine for the salinity trans-
port modeling. MT3DMS is a modular, three-dimensional transport model that can
simulate transient advection, dispersion, and chemical reactions of multiple dissolved
species, by solving the following 3-dimensional, transient advection-dispersion equation
(ADE) for each chemical species:
∂(θC k ) ∂ ∂C k ∂ X
= θDij − θvi C k + qs Csk + Rn (3.5)
∂t ∂xi ∂xj ∂xi
where
θ = the effective porosity of the subsurface medium, [-]
C k = dissolved concentration of kth species, [ML−3 ]
t = time, [T]
xi,j = distance along the respective Cartesian coordinate axis, [L]
Dij = hydrodynamic dispersion coefficient tensor, [L2 T−1 ]
vi = the linear pore water velocity, [LT−1 ], related to the darcy velocity, ui , by the
relationship vi = ui /θ
qs = volumetric flow rate per unit volume of aquifer representing fluid sources (positive)
and sinks (negative), [T−1 ]
Csk = concentration of sources or sink flux for species, k, [ML−3 ]
P
Rn =chemical reaction term, [ML−3 T−1 ]
MT3DMS uses the following approximation [Burnett & Frind, 1987] for the hydro-
dynamic dispersion tensor:
71
vx2 vy2 v2
Dxx = αL + αT + αV z + D⋆ (3.6)
|v| |v| |v|
vy2 v 2 v2
Dyy = αL + αT x + αV z + D⋆ (3.7)
|v| |v| |v|
v 2 v 2 v2
Dzz = αL z + αV x + αV z + D⋆ (3.8)
|v| |v| |v|
vx vy
Dxy = Dyx = (αL − αT ) (3.9)
|v|
vx vz
Dxz = Dzx = (αL − αV ) (3.10)
|v|
vy vz
Dyz = Dzy = (αL − αV ) (3.11)
|v|
(3.12)
where αL is the longitudinal dispersivity [L], in the direction of flow, αT is the

transverse horizontal dispersivity [L], αV is the transverse vertical dispersivity [L], v is
the pore velocity vector [LT−1 ], and D⋆ is the effective molecular dispersion coefficient
for the porous medium [L2 T−1 ].
The transport equation (Eq. 3.5) is related to the flow equation (Eq. 3.2 through
Darcy’s Law:
ui Ki ∂h
vi = =− (3.13)
θ θ ∂xi
MT3DMS uses the output hydraulic head and cell-by-cell flow data computed by
MODFLOW to establish the advective flow field. The ADE is then solved by either a
finite-difference method, particle tracking, or a combination of both.
Fixed Dissolved Solids (FDS) was modeled as an ideal, non-reactive, non-sorbing
tracer, and molecular diffusion was assumed negligible. The model was run using
the MT3DMS Advection, Dispersion, and Source/Sink mixing packages. The solution
scheme chosen was the Hybrid Method of Characteristics (HMOC) with upstream weight-
ing. The HMOC method belongs to the family of the mixed Eularian-Lagrangian type
whereby particle tracking is used to model the advective transport while the disper-
sive transport, reactions, sources/sinks are computed by the standard finite-difference
method. The HMOC package was chosen because it is considered best suited for mod-
eling situations where large time steps may be required (as is the case for simulations
over several decades), and where the grid blocks are large compared to the dispersivities
[Schäfer, 2003].
3.7.2 Solute Transport Parameters

There are several issues that need to be addressed in this context. The first issue is the
adequacy of modeling contaminant transport using large numerical grid blocks. Large
72
numerical grid blocks are not suitable for modeling contaminant transport from point
sources. However, the land application sites are of the order of hundred of meters,
much larger than the length scale of heterogeneity of the conductivity field. Under these
conditions, large grid blocks are suitable for computing space-averaged concentrations
and fluxes.
Large numerical grid blocks are not suitable if high resolution modeling of the con-
centration field is needed. Such an analysis is needed if one is interested in point values
of the concentrations. Such values are of interest from a theoretical point of view, but for
applications needed in this study, which involve computing large scale economic impacts,
such high resolution modeling is not necessary. Point values of the concentration are of
greater concern when dealing with health risks, but again, health risks are determined
based on long-term exposures, which are correlated with spatial and temporal averages
of the concentration and not with point values.
The second issue is concerned with determining an adequate strategy for modeling
the effects of spatial variability. In our approach, the large-scale effects of the spatial
variability of the conductivity field on contaminant transport are modeled directly on
the numerical grid, by assigning different conductivity values to different regions of the
numerical model. The effects of the small scale variability are modeled using hydrody-
namic, or macro-dispersion coefficients (Dij ). The macrodispersion coefficients need to
account for the effects of sub-grid block scale spatial variability on transport. This is the
spatial variability that is characterized by scales of the order of meters in the horizontal
directions, and dozens of centimeters in the vertical direction, and it the one that is
wiped out due to the uniformity of conductivity over the numerical grid blocks.
An extensive discussion of this topic is provided in Rubin et al. [2000, 2003] and in
Bellin et al. [2004] and the approach in this study follows the strategy outlined in these
research papers. To determine the value of the longitudinal dispersivity, values of the
variance and integral scale of the log-conductivity were taken from the literature [Rubin,
2003, p.35] corresponding to three-dimensional variability of alluvial fan deposits. To
account for uncertainty, a range of values for the dispersivity ratio were used, as given
below.
1. Alluvial Sediments (Western Area above Corcoran Clay, Western Area Below Cor-
coran Clay, and Eastern Area)
(a) Longitudinal Dispersivity, αL = 19.2 m

(b) Ratio of Longitudinal to Transverse Dispersivity, αL /αT = 1 to 20
(c) Porosity, θ = 0.30
2. Corcoran Clay Layer
(a) Longitudinal Dispersivity, αL = 4 m

(b) Ratio of Longitudinal toTransverse Dispersivity, αL /αT = 1 to 20
(c) Porosity = θ=0.30
73
3.7.3 Initial and Recharge Concentrations

It is necessary to specify the initial concentration for the ground water in order to solve
Eq. 3.5. A map for the initial concentration map was generated by interpolating ground
water measurements of total dissolved solids (TDS) from the food processor water well
information, and from a USGS water quality data set for the Central Valley compiled in
a report by Purkey et al. [2001].
For ground water (as opposed to the direct waste effluent or shallow vadose zone
water) it is not completely unreasonable to treat FDS and TDS as almost equivalent,
with the assumption that organic compounds do not represent a significant contribution,
having been removed or attenuated in the vadose zone. In our study there is a notable
exception: the Dairy processing effluent. The unsaturated zone modeling for the fate
of the effluent discharge for this industrial group shows that a large organic component
reaches the water table. Thus for plumes from this group it is not clear that FDS can be
directly compared with TDS. But the assumption that TDS≈FDS is a fairly reasonable
approximation for the winery, meat and tomato sites.
Because FDS values were not widely available, a reasonable assumption was made
that in the TDS and FDS in ground water are nearly equivalent. A large number of
measurements are also available from required quarterly reporting by the food processors
to the Water Quality Control Board. To avoid biasing the initial concentrations, we did
not include values from source wells immediately down-gradient of the land discharge
zones, because these wells represent localized conditions. It is expected that over the
large planning horizon, the selection of initial conditions will be of minor consequence.
Furthermore, as described later we will adopt a line of investigation that filters out the
effect of initial conditions altogether.
The recharge concentrations of FDS at the areas outside of the land application
sites were estimated by calculating a flux-averaged recharge concentration based on the
fluxes and compositions of the different sources of recharge water. This is in line with
the procedure employed by Schoups [2004]. For each water budget sub-region, a flux
averaged recharge concentration, Crecharge, was calculated using the following equation
([Schoups, 2004], Eq.4.7):
CprecipQprecip + Cgw Qgw + Csw Qsw

Crecharge = (3.14)
Qtot
where Cprecip, Cgw , and Csw , are the source water concentrations for precipitation,
ground water, and surface water, respectively. Qtot is the total recharge flux applied over
the area, and is given by
Qtot = Qprecip + Qgw + Qsw (3.15)

Recharge due to leakage from the three reservoirs shown in Figure 3.14 was assigned
the same concentration as surface water (65 mg/l), as was recharge over the losing reaches
of the three tributaries of the San Joaquin River. To account for the salinity added
due to mineral dissolution in the water water as it infiltrates through the unsaturated
zone [Schoups, 2004, see] and to account for the displacement of higher FDS porewater,
74
Figure 3.25: Map showing the land discharge areas (white polygons) for each facility in
the regional study area
Crecharge was increased by about 90 mg/l, to a value of 220 mg/l. This value was
found to result in a concentration that more closely matched the initial background
concentrations.
3.7.4 Land Application waste water Loading Rates

The time-varying salt loads at the water table calculated by the vadose zone modelling
[Miller et al., 2008] were seasonally averaged over 91.25 day periods and were applied
over the total land discharge area for each processor in the model (see Figure 3.25),
by specifying the concentration of the recharge water of the grid blocks over the land
discharge areas. These salt loading curves, shown in Figure 3.26, represent the “transfer
functions” from the vadose zone model to the saturated zone ground water model, and
are the fluxes and loads predicted at the water table by the vadose zone modeling.
Three of the food processing facilities in the model area (Figure 3.25) do not belong
to one of the four main industry groups (Meat, Wine, Tomato, Dairy) analyzed in the
vadose zone modeling, and they were assigned salt loading rates from the most similar
industrial groups.
75
Figure 3.26: Mass loading curves of Fixed Dissolved Solids (FDS) reaching the water ta-
ble over 30 year period due to land application of waste water from the four primary food
processing industries (wine & grape, meat, dairy, and tomato processing), as estimated
by 1-D, reactive vadose zone modeling [Miller et al., 2008]. These loading curves serve as
transfer functions for salinity transport from the land surface through the vadose zone
and to the water table.
Land application of waste water was assumed to occur over the entire permitted land
application area. Monthly data reported to the Regional Water Quality Control Board
indicate that it is common practice to apply the effluent over subsections of the total
permitted area, rotating areas from season to season. This rotation of discharge areas is
expected to reduce concentrations in the ground water, because it enhances mixing at
the edges of the solute body. Given the lack of precise information about the rotation
regimes, however, a conservative strategy was adopted that assumes the entire permitted
land area was used for application of the waste water.
3.7.5 Salinity Transport Modeling Results

A preliminary perspective on the 30 year salinity impact is shown in the concentration
map in Figure 3.27. The parameters used to obtain this perspective are as follows. For
the effective conductivity we used the geometric mean conductivity, KG , of each lithologic
subregion. For longitudinal macrodispersivity we used αL =19.2 m and a ratio of 1:10
between both transverse to longitudinal dispersivity (αT /αL ) and vertical to horizontal
dispersivity (αV /αL ). This figure shows the evolution of the concentration field due to
salinity loading over a period of 30 years, and it suggests that the areas affected are
mostly in the close vicinity of the land application sites, rather than regionally spread
out mixed plumes.
It is interesting to note in Figure 3.27 the development of FDS hot spots along the
76
Figure 3.27: Concentration of FDS (mg/l) at 30 years in model Layer 1. The colorbar
is cutoff at 1000 mg/l (so red indicated C ≥ 1000 mg/l FDS). The following ratios were
employed: αL /αT = 10 and αL /αV = 10, where αL , αT and αV denote the longitudinal,
transverse and vertical dispersivities. All simulations assumed αV = αT . The small
yellow circles indicate pumping wells whose screened interval intersect Layer 1.
77
Figure 3.28: Map of water table depths less than evapotranspiration extinction depth
(meters). Light blue regions show areas where the depth to water table is less than or
equal to extinction depth. Such areas are prime candidates for increased FDS due to
accumulations of salt via evapotranspiration. This effect is unrelated to food-processing
related discharge, unless if taking place in such areas.
78
upper reaches of the Stanislaus and Tuolumne Rivers, and the lower reaches of the San
Joaquin River in the Northern half of the model. These hot spots are unrelated to land
discharge operations, but rather are attributed to accumulation and concentration of
salts due to evapotranspiration in regions of shallow ground water. This is confirmed in
Figure 3.28, which identifies (in light blue) the areas where the water table is less than
2.5 meters below the ground surface, which in the model is the assigned extinction depth
for evapotranspiration. We note that the light blue areas along the rivers correlate well
with the appearance of concentration hot spots, and are unrelated to locations of effluent
land application areas.
In order to more clearly highlight the contribution of land application activities to the
FDS distribution and remove the effects of spatial variations in the natural background
concentrations and to minimize the effects of the poorly documented initial conditions
we can define the following concentration difference:
∆FDS(t) = FDS(t, with land discharge) − FDS(t,without land discharge) (3.16)
where FDS(t, with land discharge) represents the predicted concentration of FDS at
time, t, when the model is run with the land application of effluent, and FDS(t,without land discharge)
represents the predicted concentrations for the model at time, t, without land application
of effluent. Thus ∆FDS serves to effectively provide a view of the net impact of the land
application practices.
Figure 3.29 shows an areal view of ∆FDS at the water table at t=30 years. Figure 3.30
provides a three-dimensional perspective of ∆FDS where the blue iso-surfaces contain
the subsurface volumes within which ∆FDS is greater than or equal to 500 milligrams
per liter. The vertical scale is exaggerated by 30 times. The blue contour lines indicate
the elevation of the water table.
Figure 3.31 shows a semi-transparent, Northward-looking 3-Dimensional view of the
model, showing the pathlines that depict the 30–year trajectories of particles that started
at the water table underneath each of food processing land discharge areas. The particle
paths were calculated using MODPATH, a particle tracking post-processing package
that was developed to compute three-dimensional flow paths using output from steady-
state or transient ground-water-flow simulations by MODFLOW [Pollock, 1994]. The
color map represents the 3-dimensional hydraulic head distribution. The vertical scale
is exaggerated 30 times, which allows one to see the effects of the vertical flow gradients
caused by the combination of recharge and deeper pumping.
These three figures illustrate that the impacts on ground water salinity due to land
application of the effluent are expected to be limited in spatial extent when considered
in a regional perspective, but may be very pronounced locally near the application area.
79
Figure 3.29: A difference map of, ∆FDS = FDS(with land discharge) −

FDS(without land discharge) showing the net effect of simulated effluent land discharge
at the water table after 30 years. The color intensity scale is cutoff at 1,000 mg/l, so
concentrations greater than or equal to 1,000 mg/l appears as red.
Figure 3.30: A northward looking 3-D view of ∆FDS where the blue iso-surfaces contain the subsurface volumes within which
∆FDS≥ 500 mg/l. The vertical scale is exaggerated by 30 times and the blue contour lines indicate the elevation of the water
table.
80
30-year particle tracking flowpath lines under land discharge sites
Figure 3.31: 3-D view of particle tracking flow-path lines emanating from the water table underneath the land discharge areas.
(vertical exaggeration = 30×)
81
82
3.7.6 The Effects of Spatial Variability
As stated earlier, the hydraulic conductivity field was homogenized over large portions
of the study area. In effect, the spatial variability is modeled only in an average sense
using effective parameters. How to model the spatial variability of the conductivity is
a decision that depends on the length scales that characterize the flow and transport
problem. Modeling the effects of spatial variability depends on the relationship between
the length scales of heterogeneity and the dimension of the solute body, travel time
and distance. Numerous studies [add reference or material]have shown that when
the length scale of heterogeneity is small compared to the other length scales (e.g., the
grid scale), the use of effective parameters are justified except for evaluating small scale
effects. On the other hand, larger length scales may have pronounced effects on spatial
variability in the concentrations. Such effects may be significant at early travel times
(short travel distances), but lessen with time and travel distance due to the effects of
dispersivity [add reference or material]. [Consider rewording this paragraph.]
While general modeling results for this study will be presented in a subsequent sec-
tion (Section 3.6), the purpose of the current section is to evaluate the effects of the
large-scale heterogeneity on concentration. Motivation was provided by Burow et al.
(2004) in their investigation of extensive well-logs. Their study indicated that buried
paleo-channels of coarse-grain sediments may exist parallel to the course of the current
river channels, as seen for example to the south of the Tuolumne River in Figure 3.17B.
These paleo-channels could potentially act as fast pathways for salinity transport. The
channeling effect created by contrast between the higher hydraulic conductivity of the
channel deposits could lead to sharper concentration fronts, and thus should be investi-
gated.
To study the possible effect that such a channel (or channels) would have on the rate
and extent of spread of salt away from the land discharge areas, a synthetic paleo-channel
was introduced into the model domain as a narrow band of coarse-grained material with
hydraulic conductivity of 80 m/day, running parallel to the Merced river, as shown in
Figure 3.32, and located directly below the land discharge area for a food processor.
This location and the very high value of hydraulic conductivity were chosen to represent
what can be considered the “worst-case” scenario, providing conservative estimates of
potential impacts. The rest of the hydraulic conductivities were kept at the geometric
mean for each lithologic zone. The dispersivities were kept the same as in the other runs
(αL =19.2 m, αL /αT = 10).
Figures 3.33 to 3.39) present the simulated spatial distributions of FDS for different
cases and layers or cross-sections of the model. Figure 3.33 shows the FDS concentration
in Layer 1 for the case in which the model contains the hypothetical paleo-channel. Note
that there are regions with high concentrations (FDS due to land discharge) in addition
to gradual background variations (FDS not due to land discharge). In the subsequent
figures (Figures 3.34 to 3.39) the background concentrations are removed by subtracting
the FDS for the cases with no land discharge (that is, ∆FDS= FDS(with land Discharge)
– FDS(without Land Discharge)). For example, 3.34 shows ∆FDS in Layer 1 for the
case in which the channel is present. It is evident that the red-colored regions represent
83
Figure 3.32: Hypothetical hydraulic conductivity map for layer 1, which is used to
analyze the effects of paleo channels on transport of FDS in the saturated zone. Values
shown on the color bar are in meters per day. The red area represents a hypothetical
paleo-channel with a relatively high conductivity, passing through a land application
area in the study area. The other features on this map can be correlated with Figure
3.19.
84
Figure 3.33: 30 year FDS concentration in Layer 1 with paleo-channel. Color bar in
mg/l and cutoff at 1000 mg/l.
salinity from land applications at different food processing facilities. To further examine
the impact of the paleo-channel on the FDS distributions, relative to the case in which no
channel is present, we take the difference of the ∆FDS distributions with and without
the channel (a difference of differences!), as is shown in Figure 3.35 for layer 1. The
difference between these two cases highlights the effect of the channel while masking
other effects. It is interesting to note that there are some minor differences, with the
plume concentrations higher at the leading edge of the plume, but smaller at the tail
end (see also Figure 3.36).
Figures 3.37) and 3.37) show distribution similar to that in Figure 3.34), but for
layers 2 and 3, respectively, of the model. These figures show that the effect of the
channel is greatly diminished at depths below Layer 1.
What we note from these figures is that the effects of the channel appear to be local.
The channel in our simulation enhances the downstream migration of the solutes by a
distances of a few hundred meters (the dimension of the yellow spot corresponding to
X=8 km in Figure 3.39)), but this further downstream migration leads only to a small
increase in concentration, on the order of 300 mg/l. It is limited to Layers 1 and 2, and
it appears to be of the order of 100 to 200 mg/l. This change is pronounced when the
85
Figure 3.34: Differences in FDS in Layer 1 with simulated paleo-channel, showing the
∆FDS= FDS(with land Discharge) – FDS(without Land Discharge). Color bar in mg/l
and cutoff at 1000 mg/l The outline of the simulated paleo-channel is visible in light
blue. This map cannot be used to analyze the significance of the effects of the channel
unless compared with a similar one obtained assuming no channel, as shown below.
channel passes directly underneath the land discharge area, and more so, if its axis is
aligned with the flow direction. The potential exists, in general, for a significant effect
on the migration of solutes. Our testing, however, shows this effect to be somewhat
limited, primarily due to mitigating action of the transverse dispersivity and the vertical
gradient, which act together to divert mass out of the channel, and in this way diminish
significantly its capacity to conduct mass over large distances. This conclusion should
be viewed in perspective: if a pumping well is located a short distance downstream
from a land discharge area, and it is pumping its water from the channel, concentrations
measured in the well may not be predicted accurately using a homogenized conductivity
field. For the purpose of understanding the impact on regional water quality in this
study, it is justified to adopt a uniform conductivity field. For local effects, however, a
more detailed characterization of the conductivity field would be needed.
86
Figure 3.35: This map shows the difference between ∆FDS= FDS(with land Discharge)
– FDS(without Land Discharge), computed with the hypothetical paleo-channel, and
∆FDS without the paleo-channel, in Layer 1. Color bar is in mg/l. This shows the effect
a coarse-grained paleo-channel would have as compared to no channel, in the form of
differences in concentrations. The plume concentrations are higher at the leading edge
of the plume, but are smaller at the tail end.
87
Figure 3.36: This map shows a zoomed-in view of Figure 3.35 the difference between
∆FDS with the paleo-channel and ∆FDS without the hypothetical paleo-channel, in
Layer 1. Color bar in mg/l. This shows the effect a coarse-grained paleo-channel would
have as compared to no channel. The plume concentrations are higher at the leading
edge of the plume, but are smaller at the tail end.
88
Figure 3.37: This map shows the difference of ∆FDS with the paleo-channel and ∆FDS
with out the paleo-channel, in Layer 2. Color bar in mg/l. This shows the effect a coarse-
grained paleo-channel would have as compared to no channel. The plume concentrations
are higher at the leading edge of the plume, but are smaller at the tail end.
89
Figure 3.38: This map shows the difference of ∆FDS with the paleo-channel and ∆FDS
with out the paleo-channel, in Layer 3 Color bar in mg/l. ItThis shows the effect a
coarse-grained paleo-channel would have as compared to no channel. The effect of the
channel is greatly diminished at greater depths.
90
Figure 3.39: This Figure shows a cross-section view with the color scale representing the
difference of ∆FDS with the paleo-channel and ∆FDS with out the paleo-channel, in
Layer 1. Color bar in mg/l. This Figure shows the effect a coarse-grained paleo-channel
would have as compared to no channel. The differences in plume concentrations are
positive higher at the leading edge of the plume, and but are negative at the tail end.
The vertical exaggeration is 30x. In the view below we are seeing projections of the x-
and y-axes onto the East-West plane.
91
3.8 Sensitivity Analysis

This section evaluates the sensitivity of the transport modeling results to the assumption
concerning the ratio between the transverse (lateral) and longitudinal dispersivities, αL
and αT , respectively. The longitudinal dispersivity αL is used to model the effects of
small-scale heterogeneity on the longitudinal spreading of solutes in ground water, which
implies spreading in the direction of flow. The transvere, or lateral, dispersivity αT serves
a similar purpose only in the directions orthogonal to the mean flow direction, both in
the horizontal and vertical directions. The ratio between αL and αT can be determined
from field experiments. In the absence of experimental data, it is common to take the
ratio αL /αT of about 10:1. Our sensitivity analysis explores the effects of this ratio on
the concentration field.
In this section we explore the effects of this ratio on the spatial extent of the areas
affected by the land discharge. Toward this goal, we will evaluate the evolution of the
FDS concentration field in the study area for scenarios both with and without land
discharge. Again we will define ∆C(αL /αT ) as the difference at a given point in time
between the two concentrations fields (obtained by subtracting the “no land discharge”
from the “land discharge” case). For a complete picture, we will look at ∆C(αL /αT ) for
the ratios αL /αT equal to 1, 10, and 20. Towards this goal, we will look at the spatial
distributions of ∆CI (αL /αT = 1), ∆CB (αL /αT = 10), and ∆CII (αL /αT = 20), all
computed at 30 years. For clarity, note that a positive ∆C(αL /αT ) implies an increase
in the concentration due to land discharge.
Figure 3.40 shows the spatial distributions of the differences for the various ratios
and at different depths, from a regional perspective. The maps shown cover the entire
study area, and are difficult to interpret locally, but they provide the general impression
that there is little sensitivity of the spatial extent of contamination to this ratio. In all
cases, the effects appear to be localized.
Differences between the various ratios are expected at the local scale. A larger ra-
tio αL /αT will enhance longitudinal spreading at the expense of lateral spreading. In
essence, a larger ratio will lead to cigar-shaped plumes, and smaller ratios will lead to
more spherical looking plumes. We will take a closer look at these effects, but from the
regional perspective, again, it appears that the effects of changing the ratio of disper-
sivities are limited to the close vicinity of the discharge areas, rather than the regional
concentrations.
Figure 3.41 provides a closer look at the difference maps for the two ratios αL /αT
of 1 and 20, which are the two end points of the investigated range of ratios. It can
be noticed by looking at the dairy facility at the southern part of the study area that
larger αL /αT ratios enhance the longitudinal spread in a westerly direction. We also
note that the αL /αT =20 case leads to larger differences in concentrations compared to
the αL /αT =1 case, as is indicated by the prevalence of red hot spots over green, that is
particularly visible in the winery located as the southern part of the study area. This is
because the solute mass spreads over a smaller volume when moving in geological flow
domains of higher αL /αT ratios which tend to channelize the solute mass. It is expected
that higher ratios will have lower vertical spread as well.
92
Table 3.4: Table showing statistics of ∆FDS for each the three dispersivity ratio scenarios
Scenario ∆CI ∆CB ∆CII
αL /αT 1 10 20
Min -147 -356 -472
Max 7019 10296 10926
∆FDS Range 7166 10652 11398
(mg/l) Mean 11 12 12
Median 0 0 0
Stdv 113 152 166
N 104805 104805 104805
The statistics of the differences in concentrations between the αL /αT =20 and αL /αT =1
cases are shown in, Table 3.4 obtained by analyzing close to 104,000 grid nodes. The
αL /αT =20 leads to the largest differences in absolute value, both positive and nega-
tive, with both being the outcome of the channeling effect. The negative differences
are somewhat surprising, because land discharge adds solute mass to the aquifer, and
one should not expect a reduction in concentration anywhere, but only increases or no
changes. However, the negative differences, which are of the order of magnitude of the
background concentration, indicate channeling of the initial solute mass in the aquifer
away form the high initial concentration areas, which leads to reduction in concentration.
The negative values are not real physical effects, but only an artifact of working with a
uniform set of initial conditions for all numerical simulations.
In order to filter out this artifact, it is beneficial to evaluate the difference in the
differences ∆C(αL /αT ), namely ∆CI − ∆CB and ∆CII − ∆CB . This allows evaluating
the effects of the land discharge while avoiding any numerical artifacts. Figures 37 and
38 evaluate these differences at the vicinity of the Hilmar Facility, at the southern part
of our study area. With blue indicating a decrease in concentration, green indicating
near-zero changes and red indicating regions with an increase in concentration, we note
that the ratio αL /αT =1 leads to larger concentrations going deeper into the aquifer,
and to lower concentrations in the shallower portions of the aquifer, closer to the ground
surface. This is due to the enhanced lateral spread that characterizes the αL /αT =1 case,
that enhances mass transfer in a direction orthogonal to streamlines.
3.9 Monte Carlo Simulations

The set of sensitivity analyses of the previous section is useful for exploring extreme
situations, but they cannot provide probabilities for observing any event (e.g., concen-
tration exceeding certain threshold values). This can be done, however, using a Monte
Carlo[Metropolis & Ulam, 1949] simulation approach . The process of Monte Carlo sim-
ulations [Rubin, 2003, see] includes generating alternative, physically and statistically
plausible realizations of the flow domain, and analyzing flow and transport in each of
them. The variation between the realizations represent the uncertainty in the subsur-
93
∆CI (αL /αT = 1) ∆CB (αL /αT = 10) ∆CII (αL /αT = 20)
L01
L02
L03
L04
L05
Figure 3.40: Concentration maps showing the net effect of 30 yrs of land discharge for
the three different dispersivity ratio scenarios. For each dispersivity ratio (columns) the
maps shows the net concentration difference, ∆C between 30 yrs with and without land
discharge activity for each of the five model layers (L01-L05), starting from top layer
(L01) . The color intensity scale is in mg/l and is cutoff at 1000 mg/l (all values greater
than 1000 are in red). The white areas on east side the top 2 layers indicate dry cells
above water table.
94
Layer 01, t = 30 years
∆CI (αL /αT = 1)
∆CII (αL /αT = 20)
Figure 3.41: A close-up of the ∆C maps for the two end points of the dispersivity ratio
scenarios, for the top layer of the model. The maps show the concentration difference
between 30 yrs with and without land discharge activity for each of the top model layer
L01. The color intensity scale is in mg/l and is cutoff at 1000 mg/l (all values greater
than 1000 are in red).
95
face characterization. The ensemble of realizations thus analyzed provide an ensemble of

plausible values of the dependent variables, such as the concentration of various chemi-
cals of concern at many locations and at many time steps, which can then be analyzed
statistically. Such an approach is particularly suitable for situations such as we have
here, where little information is available in terms of direct measurements, yet there
is a reasonable understanding, from previous modeling studies, of the conditions and
parameter values ranges that can be implemented in the Monte Carlo Scheme. [add
reference or material]
The sensitivity analysis indicated that the anisotropy ratio of the dispersivity plays a
minor role in affecting impacts at a regional scale. The effective conductivity is expected
to potentially hold a more significant role. In the Monte Carlo analysis, the effective hy-
draulic conductivity of each lithologic region is treated as a stochastic random variable,
defined by a probability density function (pdf). The hydraulic conductivity for each
lithologic region is kept uniform spatially, but different realizations are randomly gener-
ated from the statistical distribution. The pdf’s for each zone were defined as uniform
distributions with a minimum value equal to the harmonic mean of the conductivity of
the ith lithologic region , KH i , and the maximum value equal to the arithmetic mean
i
for that region, KA . Five random realizations were generated for each lithologic unit,
except for the Corcoran Clay layer which was not randomized. Random realizations
were generated using the Latin Hypercube method in order to optimize the coverage of
the parameter space sampled [Iman & Conover, 1980]. Thus a total of 125 (5×5×5)
different combinations of effective conductivities for the model domain were generated,
each comprised of a different set of effective conductivities. The different realizations
were then used to run the ground water flow model and the salinity transport model for
the thirty year period. The summary of the parameters used is given below in Table 3.5.
3.9.1 Monte Carlo simulation Results

The Monte Carlo simulations generate an ensemble of realizations for the modeled con-
centration field, from which we can compute statistics for the concentration at any posi-
tion, x, and time, t. We can define C(xi , t) as the concentration at each cell centered at
position, xi , at time, t. For each grid block we can thus calculate an ensemble mean con-
2 ),
centration, hC(xi , t)i and a concentration variance, σC (xi , t) (or standard deviation,σC
which measures the variability of the concentration around the ensemble mean. The en-
semble of realizations of the concentration field
Table 3.6 summarizes the statistics of the FDS concentration at 30 years, includ-
ing minimum, maximum and average concentrations as well as the standard deviation.
Note, this table shows the “statistics of the ensemble statistics” based on all grid cells,
for example, the mean of hC(xi , t)i when taken over all all grid cells. It shows that
the FDS values are characterized by relatively low averages and very high maximum
concentration values. The standard deviations are also low, which suggests that most of
the model domain is characterized by the low FDS values, except for hot spots of very
high concentrations.
The probability, P , that the concentration at a given location and time is less than or
Table 3.5: [Table showing the parameter values used in the Monte Carlos simulations.] The salinity transport model is then
run on each of these realizations to get a statistical distribution of salinity concentrations at each point in the model grid. This
distribution can be characterized by mean concentrations and standard deviations of the salt concentrations in the model. This
probability density function can be used to generate a map which shows the probability of the salt concentration exceeding a
given threshold, for example the MCL, at any given point in the model domain.
Monte Carlo Simulation Parameters
Lithologic Kmin = Kmax = K hor /Kvert # K- Distrib- αL [m] αL /αT αL /αV
Subregion KH KA Realiz- ution
[m/day] [m/day] ations
Western 19 84 300 5 uniform 19.2 10 10
Area above
Corcoran
Clay
Western 1 53.51 300 5 uniform 19.2 10 10
Area below
Corcoran
Clay
Eastern 10.98 49.23 300 5 uniform 19.2 10 10
Area (entire
thickness)
Corcoran 1.3 × 1.3 × 1 1 uniform 4 10 10
Clay 10−3 10−3
Total Num- 125

ber of Real-
izations
96
97
Table 3.6: This table shows summary statistics for the values of the ensemble Min,
Mean, Max and standard Deviation of the concentration generated in the Monte Carlo
Simulations.
FDS Summary Statistics [mg/l] over all realizations
CM in CM ean CM ax Cstdev
Min 18 31 49 0
Max 9979 13045 15108 12847
Range 9962 13014 15059 12847
Mean 247 260 274 18
Median 223 227 230 1
Stdev 170 209 251 186
#Cells 104805 104805 104805 104805
equal to some concentration, C ⋆ , is defined by the distribution function, or, cumulative

density function (CDF):
P (C(xi , t) ≤ C ⋆ ) = CDF(C ⋆ , xi , t) (3.17)
The probability that C(xi , t) is greater than C ⋆ is called the exceedence the proba-
bility and can be calculated as:
P (C(xi , t) > C ⋆ ) = 1 − CDF(C ⋆ , xi , t) (3.18)
In order to calculate the exceedence probability at each cell, we adopted the approach
described in Rubin [2003], suggested by the investigation of Bellin et al. [1994], for
approximating the CDF of the concentration at each cell. This was done by computing
the ensemble mean and variance at each cell and assuming a lognormal model for the
probability distribution. This approach provides a reasonable approximation for large
sampling volumes (as we have in the case of large grid blocks), and is less computationally
demanding than the generating the large number of realizations needed for computing
the entire CDF.
The next set of figures summarize graphically the spatial distribution of the FDS
concentration after 30 years in in each of the model layers. Figures 3.42–3.48 illustrate the
spatial distribution of the mean and standard deviation of the concentration. Figures 3.49
to 3.53 provide the spatial distributions of the probabilities for the FDS concentration
to exceed 500 mg/l, which is the non-enforceable guideline based on the cosmetic and
aesthetic qualities of water. These figures show consistently that the spatial extent of
the impact on ground water by land application is limited. These figures include also
the impact of water infiltration along the river banks coupled with evapotranspiration
described earlier.
98
Figure 3.42: This map shows the ensemble mean concentration, CM ean of FDS for each
grid block in layer 1 after 30 years. The color scale is in mg/l and is cutoff at 1000 mg/l
(red≥1000 mg/l, blue = 0 mg/l). All white areas on the left side indicate dry cell blocks
that are above the water table.
99
(red≥1000 mg/l, blue = 0 mg/l). All white areas on the left side indicate dry cell blocks
that are above the water table.
100
(red≥1000 mg/l, blue = 0 mg/l).
101
(red≥1000 mg/l, blue = 0 mg/l).
102
Figure 3.46: This map shows the concentration standard deviation, CStdev for the en-
semble of all realizations for each grid block in Layer 1 after 30 years. The color scale is
in mg/l and is cutoff at 1000 mg/l (red≥1000 mg/l, blue = 0 mg/l). All-white areas to
on the left side indicate dry cell blocks that are above the water table. The drying out
or wetting of different cells as the water table rises and falls with different parameter
realizations creates a fringe of artificially high standard deviations along the edge of the
grid subject to drying out. This is visible as the red to green colors that follow the
“shores” of the dried out cells.
103
semble of all realizations for each grid block in Layer 2 after 30 years. The color scale is
in mg/l and is cutoff at 1000 mg/l (red≥1000 mg/l, blue = 0 mg/l). All-white areas to
on the left side indicate dry cell blocks that are above the water table. The drying out
or wetting of different cells as the water table rises and falls with different parameter
realizations creates a fringe of artificially high standard deviations along the edge of the
grid subject to drying out. This is visible as the red to green colors that follow the
“shores” of the dried out cells.
104
semble of all realizations for each grid block in Layer 3 after 30 years. The color scale
is in mg/l and is cutoff at 1000 mg/l (red≥1000 mg/l, blue = 0 mg/l). The standard
deviation of the concentration decrease sharply with increasing depth.
105
Figure 3.49: This map shows the probability of FDS>500 mg/l in Layer 1 after 30
years. The color scale represents probability with 0 (blue) indicating 0% probability
of exceedence, and 1.0 (red) indicating 100% probability of exceeding 500 mg/l. The
location of the rivers and major streams are shown as white lines. In addition to the
ET hot spots described in the previous section there are a few artifacts associated with
grid blocks that border the cells that dry out (i.e. cells that are entirely above the water
table). The large standard deviations associated with these grid cells prone to drying
out causes higher exceedence probabilities to be estimated for these regions. One such
artifact is clearly visible in the middle of the eastern edge of the model.
106
location of the rivers and major streams are shown as white lines. In addition to the ET
hot spots described in the previous section there are a few artifacts associated with grid
blocks that border the cells that dry out in (i.e. cells that are entirely above the water
table). One such artifact is clearly visible in the eastern most corner of the model.
107
location of the rivers and major streams are shown as white lines. In addition to the
ET hot spots described in the previous section there are a few artifacts associated with
grid blocks that border the cells that dry out in (i.e. cells that are entirely above the
water table). The large standard deviations associated with these cells prone to drying
out causes higher exceedence probabilities to be estimated for these regions. One such
artifact is clearly visible in near the eastern most corner of the model.
108
location of the rivers and major streams are shown as white lines. The large “hot
spots” that appear along the San Joaquin river edge of the model are due to the initial
background concentrations assigned to the model, rather than to specific land application
activity.
109
location of the rivers and major streams are shown as white lines. The large “hot
spots” that appear along the San Joaquin river edge of the model are due to the initial
background concentrations assigned to the model, rather than to specific land application
activity.
110
Table 3.7: This table shows summary statistics for the values of the ensemble Min, Mean,
Max and standard Deviation of the concentration differences (∆C)
∆C Summary Statistics [mg/l] over all realizations
∆CM in ∆CM ean ∆CM ax ∆CStdev
Max 7698 10087 11583 914
Range 8260 10384 11653 915
Mean 8 11 14 1
Median 0 0 0 0
Stdev 117 145 167 13
#Cells 104805 104805 104805 104805
3.9.2 An alternative Monte Carlo analysis

In this Section we develop an alternative analysis. Rather than assessing the statistics
of FDS concentrations, we shall analyze the statistics of the net differences in concentra-
tions caused by land discharge. This will allow us to assess directly the impact of land
discharge, and to eliminate from the picture important yet unrelated effects such as the
hot spots of FDS concentrations along the rivers.
Let us define ∆C as the difference between the FDS concentration with land dis-
charge and the FDS concentration assuming no land discharge. The difference ∆C was
calculated using pairs of realizations of the concentration field, each pair including a real-
ization of the concentration field computed with land discharge, CLD , and the realization
obtained using the same set of parameters, but without land discharge, CN LD .
∆C(xi , t) = C(xi , t)LD − C(xi , t)N LD (3.19)
In the “no land discharge” scenario the land discharge areas were assigned the same
recharge fluxes and concentration equal to the corresponding background values.
A large number of pairs were generated, yielding a large ensemble of ∆C. The ensem-
ble of ∆C was then used to calculate statistics, concentration maps, and concentration
exceedence probability maps. The FDS concentration at grid blocks that dried out was
assumed to be zero. The exceedence probability was estimated by calculating the mean
and standard deviation of the ∆C at each grid block and then assuming a log-normal
distribution in order to calculate the cumulative density function (CDF). The probability
of ∆C > 500 mg/l, is then calculated as 1-CDF(∆C =500 mg/l).
Table 3.7 summarizes the statistics of the ∆C concentration at 30 years, including
minimum, maximum and average concentrations as well as the standard deviation. It
shows that the ∆C values are characterized by relatively low averages and very high
maximum concentration values. The standard deviations are also low, which suggests
that most of the domain is characterized by the low ∆C values, except for hot spots of
very high concentrations changes.
Figures 3.54 – 3.58 depict the spatial distributions of the exceedence probabilities
for the five different model layers. The probability to have ∆C larger than 500 mg/l is
111
very high at most of the land application areas in the 4 top model layers, and reduces
significantly at larger depths.
Figure 3.54: This map shows the probability of ∆C >500 mg/l in layer 1 after 30
years. The color scale represents probability with 0 (blue) indicating 0% probability of
exceedence, and 1.0 (red) indicating 100% probability of exceeding 500 mg/l.
112
Figure 3.55: This map shows the probability of ∆C >500 mg/l in layer 2 after 30
exceedence, and 1.0 (red) indicating 100% probability of exceeding 500 mg/l. In addition
to the ET hot spots described in the previous section there are a few artifacts associated
with grid blocks that border the cells that dry out (i.e. cells that are entirely above the
water table,). One such artifact is clearly visible in the eastern most corner of the model
113
Figure 3.56: This map shows the probability of ∆C >500 mg/l in Layer 3 after 30
114
115
116
3.10 Comparison with field measurements

An obvious question to ask now is“Given all the simplifying assumptions made, does
this model predict reasonable concentrations of FDS consistent with groundwater mon-
itoring beneath land application sites at actual food processors in the region? Do our
conservative assumptions result in concentrations that provide realistic upper bounds
on expected spatial extent and peak concentrations of the salinity plumes?” To answer
these questions we will take a closer look at monitoring data from the large dairy proces-
sor south of Modesto (in Hilmar, CA). This dairy processor has been in operation since
1985 and annually produces more cheese from a single site than any other manufacturer
in the world. The facility processes on the order of 1.5 million gallons of milk each day
received from hundreds of California dairies producing more than 1.4 million pounds of
cheese, whey, and lactose products each day, seven days a week, 365 days a year. Once
the marketable components of the milk are extracted, approximately 85% of the original
volume remains (referred to as “cow water” at the plant). A mixture of “cow water”
and locally produced groundwater is also used for cleaning operations, and this adds
additional solids and salts to the final waste stream, resulting in a waste stream that
can be as high as 1.5 million gallons per day [Brown and Caldwell, 2004b,a; Brown &
Caldwell, 2006].
Although recently the plant has added reverse osmosis as a pretreatment method,
as well as separation of biosolids for off-site disposal, historically the waste water was
applied directly to land application with no treatment over a region of 147 acres [Brown
and Caldwell, 2004b,a; Brown & Caldwell, 2006].
Figure 3.60 shows the an aerial photo of the processing facility and the surrounding
dairies and fields with the traced outline of the primary historical land application area
for the facility shown in yellow. Shallow ground water monitoring wells (MW-1 – MW-
20), agricultural tile drains (cyan) underlaying the fields and the associated tile drain
pump stations (red “P” symbols). The facility is visible just to the left of the discharge
area, in the region with the largest cluster of wells.
Figure 3.61)
Monthly monitoring data of FDS levels in the wells for the period between May 2004
to December 2005 are shown in Figure 3.59. The monitoring wells are screened in the
upper aquifer above the Corcoran Clay layer.
The water table in the local area is found at a depth of around 3 meters ( 10 ft). A
significant ground water mound has formed under the land application area, raising the
the water table directly under the site to within 2 meters of the ground surface. It is
unclear if the ponding is due to the land application, or leakage from a the two holding
ponds. It should be noted that this a much shallower than the water table depth of 14
meters assumed in the unsaturated zone modeling used to compute the loading curves,
an assumption which reflected the average conditions for the entire regional model area.
On-site investigations reported that ground water levels showed TDS above the 1,000
mg/l threshold as deep as 30 meters below ground surface (the maximum vertical extent
of investigation).
117
Figure 3.59: Plot of measured monthly FDS concentrations in shallow groundwater

monitoring wells (Figure 3.60 at the Hilmar Cheese plant for the period from May 2004
to December 2005.
118
Figure 3.60: Aerial photo showing the outline (yellow) of the primary land application
area at the Hilmar Cheese plant, shallow ground water monitoring wells (MW-1 – MW-
20), and agricultural tile drain (cyan) and the associated tile drain pump stations (red
symbols).
119
Figure 3.61: Satellite photo showing aerial view of a large dairy processor (cheese &
whey) south of Modesto.
120
Figure 3.62: This contour map shows a salinity plume under an actual large dairy
processor in the study area. The groundwater measurements have been averaged over
from monthly monitoring data from May 2004 to December 2005. What is shown in
actually a difference map of FDS above the background FDS concentration in a reference
well (MW-20). The color scale has been chosen to match that of the model results, with
red representing ∆C ≥ 1000 mg/l FDS.
121
Figure 3.63: This contour map shows the simulated salinity plume in the top model
layer under a large dairy processor in the study area, after 21 years of continuous waste
discharge. The color scale shows net difference FDS above the background FDS concen-
tration, ∆FDS. The upper bound of the color scale has been set with dark red represent-
ing ∆C ≥ 1000 mg/l FDS. The outer yellow polygon shows the “worst-case scenario”
land discharge area used in the model. The inner yellow polygon shows the actual land
application area used historically since 1985, and is about 1/8 the total land application
area on which the facility is permitted to discharge.
122
Figure 3.64: Computed breakthrough curves for FDS in model layer 1 at the center of
plume for four processers from each of the main industry groups.
123
3.11 Conclusions
This really needs to be redone to add the context missing by not having the
whole report. Also need to address the research questions to give more of the
research emphasis,etc.. Chapter 3 discusses the development of a ground water model
for the Lower San Joaquin River (LSJR) Basin, and the implementation of the model to
investigate the transport of waste water discharged form food processing facilities. This
model is part of an integrated saturated-unsaturated system for modeling the fate of the
waste water discharge. This system includes an unsaturated zone modeling component
that is described in detail in Rubin et al. [2007] and Miller et al. [2008]. The unsaturated
zone modeling component computes the mass (water and solutes) that are discharged
into the saturated zone and modeled with the ground water model.
Our numerical models can cover a wide range of processes affecting flow and trans-
port in the subsurface, including multi-component chemical reactions. The models are
suitable and ready for additional large- and small-scale investigations that may be re-
quested as a follow-up to this study, or those that can hopefully become part of the
routine monitoring and licensing of industrial and agricultural activities in the Central
Valley. Large scale investigation implies a modeling effort that covers a large region
with multiple processors, whereas small-scale investigations can focus, for example, on a
single food processor.
Our study of the LSJR basin intends to provide conclusions that are applicable for the
entire Central Valley. There are always limitations to such projections due to variations
in local hydrological conditions, soil types, geology, as well as in pumping rates. However,
two observations can be made to support the generality of the study’s conclusions. First,
the geology of the Central Valley is stationary, as is evident from Figure 3, which shows
the Central Valley as comprising of a series of repeating geological units. (This is too
strong a statement that can not be fully supported) In this regard, the findings
from studying one geological unit can be generalized to other units. Additionally, we
modeled the LSJR Basin’s hydrologic response (i.e., degradation in ground water quality)
to an ensemble of food processors representing various industrial groups covering the
highest impact industrial groups including tomato canners, wineries, meat packers and
others. This allows investigating the impacts of various chemical profiles of the waste
water discharge.
There are possibly significant differences between the ground water degradation at
specific sites compared to those modeled here. This is primarily because there is no
data base currently available to support detailed, site-specific analysis of ground wa-
ter degradation, and our model was developed under many assumptions. While these
assumptions are based on engineering judgment, they are subject to uncertainty until
verified through additional testing.
Another reason why we expect differences between actual impacts and modeled im-
pacts is that our investigation is carried out under assumed worst case scenarios. We
thus expect that actual impacts overall (meaning, in most land discharge sites) will be
more moderate than those anticipated by the worst case scenario analysis. Worst case
scenarios were created by considering several aspects of the modeling effort intended to
124
maximize the ground water degradation. Worst-case scenarios were constructed by max-
imize the mass loading from the unsaturated to the saturated zone. For modeling ground
water flow, we consider a broad range of parameters, including the presence of highly-
conductive channels. The ground water is assumed to be at steady state. Transients in
the ground water are known to enhance dispersion [Dagan et al., 1996], which in turn
leads to reduction in concentrations. This steady state leads to higher concentrations.
The uncertainty regarding actual site conditions is the motivation behind the model-
ing decision to work under assumed worst case scenarios, and the rationale is that worst
case scenarios may put brackets on the range of expected impacts to ground water qual-
ity. There can always be surprises, such as the presence of highly conductive channels
that connect the land discharge site with a particular water supply well. Surprises can
be prevented through site investigation. We tried to minimize surprises through investi-
gating a wide range of conditions, including a hypothetical case of a highly-conductive
channel as well as a wide range of hydrologic parameter values, intended to cover the
range of parameters expected at the sites.
It should be emphasized again that the scenarios modeled do not represent actual
conditions imparted by food processors at the specific locations. Rather the model
identifies a “worst case scenario profile” for a particular industry, and then attached it
to the various facilities based on their industrial affiliation. The only modification to
local conditions was in matching the total discharge at the site to the one reported by
the facility. But again, there is no direct correlation between the chemical composition
modeled and the actual one reported at the site. However, prelimiary comparisons with
actual monitoring data, such as from the Hilmar Cheese processing facility, indicate
that the assumptions used provide reasonable estimates and brackets on the extent and
magnitude of impact from land application of waste water.
Our investigation indicates that the degradation to ground water quality is likely to
occur within the close vicinity of the discharge sites. Whereas the solutes can migrate
downstream of the discharge sites over distances of thousands of meters, increases in
FDS concentrations larger than 500 mg/l compared to background are limited to the
ground water underneath the discharge sites and over distances of hundreds of meters
downstream. The probability for increase in concentrations of such magnitude over
larger distances were computed and reported above, and were found to be close to zero.
These are the predicted outcomes from our worst-case scenario analysis. Most land
discharge sites do not operate under the “worst case scenario”, and their impacts on
ground water quality are expected to be milder compared to those predicted by the
worst-case scenarios.
The impacts of land discharge on ground water quality are given in terms of excee-
dence probabilities. These probabilities represent the probability for the concentration
to exceed a threshold at a given location and time. The statement of impact in terms
of probability intends to address the uncertainty in estimating such impact. This un-
certainty stems from limited knowledge of actual site conditions, including hydrologic
conditions and hydrogeochemical parameters such as the permeability and chemical reac-
tion parameters. To address this uncertainty, we opted to work with a range of values for
the uncertain parameters instead of a single value. Each range of values is summarized
125
in the form of a histogram, and values are drawn at random from this range to produce
a realization. Many realizations are thus produced, each being a physically plausible
representation of reality. The underlying assumption is that an ensemble of realizations
will create an ensemble of performance that will encapsulate the actual one. This ap-
proach follows a well-known procedure in statistics called Monte Carlo simulation. The
way to interpret the exceedence probability is as follows. If an exceedence probability at
a certain location and time is equal to 90%, this means that 9 out of ten sites operating
with similar waste characteristics and hydrology are expected to produce a concentration
above the threshold at that location and time. As additional data is collected at the
discharge site, and as additional information becomes available, the exceedence probabil-
ities will change. Instead of working with large ranges for the unknown parameters, we
will be able then to work with narrower ranges. A narrower range can lead to higher or
lower number of realizations with concentrations above the threshold values. Additional
measurements do not necessarily reduce exceedence probabilities, they only make predic-
tions more accurate and closer to reality. The procedure for incorporating measurements
into prediction is called conditioning [Rubin, 2003].
What can explain the limited spatial extent of the spreading of solutes underneath
the discharge sites? It is a combination of effects, including the reduction in concentra-
tions due to dispersion, and in a few cases due to the buffering effects of the vadose zone.
There is a strong vertical hydraulic gradient in the shallow aquifers caused by the com-
bined effects of recharge and deeper pumping, leading to downward vertical migration
of solutes, deeper into the aquifer. This downward migration limits the lateral extent
of the migration of the pollutants, and at the same time leads to development of hot
spots of high concentrations in the volume of aquifer just underneath and downstream
of the land discharge sites. Such hot spots will persist as as land discharge occurs and
as long as vertical gradients of sufficient magnitude persist. These gradients are created
by pumping at the deeper formation, and possibly by other effects such as mountain
font recharge. A reversal in the direction of the gradient is possible if the water table
in the hydrostratigraphic units surrounding the central Valley increase, for example in
high precipitation years or reduction on the deep aquifer pumping.
As the concentration in the hot spots increases, pollutants will continue to migrate
into the deeper formations, eventually leading to termination of the deep-formation
pumping, as the need arise to drill wells deeper into the aquifer. Once the pumping
in the deeper formation ceases or in fact is performed deeper, or the gradient reverses
from any other reason, the containment effect will reduce or even vanish, and the pol-
lutants accumulated in the hot spots will migrate and spread over a much larger area.
However, this will happen primarily in the upper aquifer, and the FDS hot spots will
migrate horizontally. The impact to ground water quality will become more widespread,
but since the pumping wells will be located deeper, the impact on drinking water will
not be immediate. If one is willing to write off the water stored in the upper aquifer
above the Corcoran Clay layer, then this impact can be considered minor. However,
that would imply losing the storage capacity of the upper aquifer. And with time, due
to vertical migration of solutes, the contaminants will spread even deeper. Our analysis
assumes that the deep formation pumping will continue in its current form into the fore-
126
seeable future, which means the 30 years’ design horizon for this study. This is a working
hypothesis and there is no guarantee that this will be the case. An alternative scenario
would be to investigate the effects of pumping termination and gradient reversal, and to
evaluate this effect on the FDS hot spots. We have not investigated this scenario due to
budgetary constraints. But it needs further investigation.
Hydrological conditions, soil parameters, chemical profiles as well as the magnitude of
the waste water discharge vary across land discharge sites. There is no one formula that
can be used to predict what the actual impacts to ground water will be at each of the
sites because of the wide range of conditions and parameters that need to be recognized.
It is thus imperative that specific site investigation and modeling analysis are conducted
at each active and potential land discharge site. Strategies need to be developed to
make such studies economic and rapid. The tools we developed and presented here are
applicable immediately for such purposes.
3.11.1 Future Improvements

Instead of using just vadose zone transfer functions with just one assumed water table
depth, have three (deep, medium, shallow water table).
high/low ratios of Ksat to Qs .
Look at the evolution of Nitrate.
Incorporate annually averaged, industry-specific recharge rates over the land appli-
cation areas.
More sophisticated handling of the rivers
higher vertical spatial resolution of grid
local scale model to evaluate at higher resolution and with
127
Part V
Closure
129
Chapter 4
Summary and Conclusions

130
131
Bibliography
Appelo, C. A. J., & Postma, D. 2005. Geochemistry, groundwater and pollution. 2nd
edn. Leiden, Netherlands: A. A. Balkema.
Ayers, R.S., & Westcot, D.W. 1994. Water quality for agriculture. Food and Agriculture
Organization of The United Nations.
Batchelor, G K. 1974. Transport Properties of Two-Phase Materials with Random Struc-

ture. Annual Review of Fluid Mechanics, 6(1), 227–255.
Bellin, A., Rubin, Y., & Rinaldo, A. 1994. Eulerian-Lagrangian approach for modeling of
flow and transport in heterogeneous geological formations. Water Resources Research,
30, 2913–2924.
Bellin, A., Lawrence, A. E., & Rubin, Y. 2004. Models of sub-grid variability in numerical
simulations of solute transport in heterogeneous porous formations: three-dimensional
flow and effect of pore-scale dispersion. Stochastic Environmental Research and Risk
Assessment, 18(1), 31–38.
Bertoldi, G. L.; Johnston, R. H.; Evenson K. D. 1991. Ground water in the Central
Valley, California; a summary report. Professional Paper 1401-A. U.S.G.S.
Broadbent, S.R., & Hammersley, J.M. 1957. Percolation processes I. crystals and mazes.
Proceedings of the Cambridge Philosophical Society, 53, 629–641.
Brown & Caldwell. 2006. Hilmar Cheese Company, Monthly Water Quality Monitoring
Report, December 2005, Permit No. 97-206. Tech. rept. California Regional Water
Quality Control Board.
Brown and Caldwell. 2004a. Hilmar Cheese Company, Merced Country, CA, Report of
Waste Discharge. Tech. rept.
Brown and Caldwell. 2004b. Hilmar Cheese Company, Monthly Water Quality Monitor-
ing Report, December 2004, Permit No. 97-206. Tech. rept. California Regional Water
Quality Control Board.
Burnett, R. D., & Frind, E. O. 1987. Simulation of Contaminant Transport in Three

Dimensions 2. Dimensionality Effects. Water Resources Research, 23, 695–705.
132
Burow, Karen R., Shelton, Jennifer L., Hevesi, Joseph A., & Weissmann, Gary S. 2004.
Hydrogeologic characterization of the Modesto Area, San Joaquin Valley, California.
SIR 2004-5232. USGS, Reston, VA.
Cal. Dept. of Public Health. 2006 (May 2). California Code of Regulation,
Title 22. Division 4. Environmental Health Chapter 15. Domestic Water
Quality and Monitoring Regulations Article 16. Secondary Water Standards.
http://www.cdph.ca.gov/certlic/drinkingwater/Documents/Recentlyadoptedregulations/R-
21-03-finalregtext.pdf. (verified 3 Dec 2008).
Cal. Dept. of Public Health. 2008 (June). Nitrates and Nitrites in Drinking Water.
Web-Page: http://www.cdph.ca.gov/certlic/drinkingwater/Pages/Nitrate.aspx.
California Dept. of Water Resources. 2007. California Irrigation Management Informa-

tion System (CIMIS), Monthly ET Records for Oct 1999- Oct-2000. Web-Database:
http://wwwcimis.water.ca.gov/cimis/monthlyReport.do.
California EPA. 2007. California Central Valley Regional Water Quality Control Board
adopted orders. http://www.swrcb.ca.gov/rwqcb5/adopted_orders/index.html.
(verified 11 May 2007).
California League of Food Processors. 2007. Manual of good practice for land application
of food processing/rinse water. Davis, CA: CLFP.
Central Valley Regional Water Quality Control Board. 2005. Staff report: Regulation of
food processing waste discharges to land. Cal. Dept. Water Resources, Rancho Cordova,
CA.
Central Valley Regional Water Quality Control Board. 2006. Staff report: Update re-
garding the regulation of food processing waste discharges to land. Cal. Dept. Water
Resources, Rancho Cordova, CA.
Chaouche, M., Rakotomalala, N., Salin, D., Xu, B., & Yortsos, Y. C. 1994. Invasion
Percolation in a Hydrostatic or Permeability Gradient - Experiments and Simulations.
Physical Review E, 49(5), 4133–4139.
Clement, E, Baudet, C, Guyon, E, & Hulin, J P. 1987. Invasion front structure in a 3-D
model porous medium under a hydrostatic pressure gradient. Journal of Physics D:
Applied Physics, 20(5), 608.
Corporation, URS. 2002 (October). Development of water supply alternatives for use in
habitat restoration for the San Joaquin River. Tech. rept. Prepared for Friant Water
Users Authority, Lindsay, California and Natural Resources Defense Council Coalition
San Francisco, California.
Crites, R., Pound, C., & Smith, R. 1974 (17-19 April). Experience with land treatment
of food processing wastewater. p. 4 - 30. USEPA.
133
Dagan, G., Bellin, A., & Rubin, Y. 1996. Lagrangian analysis of transport in Het-
erogeneous formations under transient flow conditions. Water Resour. Research, 32,
891–900.
Danalewich, J. R., Papagiannis, T. G., Belyea, R. L., Tumbleson, M. E., & Raskin,
L. 1998. Characterization of dairy waste streams, current treatment practices, and
potential for biological nutrient removal. Water Research, 32(12), 3555 – 3568.
DWR, CA. 2006 (December). California Water Data Library. Electronic Database.
Ferriz, Horacio. 2001. Groundwater Resources of Northern California: An Overview.

Pages 19–48 of: Ferriz, H., & Anderson, R. (eds), Engineering Geology Practice In
Northern California. California Geological Survey (and the Association of Engineering
Geologists).
Frisch, H. L., & Hammersley, J. M. 1963. Percolation Processes and Related Topics.
Journal of the Society for Industrial and Applied Mathematics, 11(4), 894–918.
Glass, R. J., & Yarrington, L. 1996. Simulation of gravity fingering in porous media
using a modified invasion percolation model. Geoderma, 70(2-4), 231–252.
Gronberg, J.M., & Kratzer, C.R. 2006. Environmental setting of the lower Merced River
Basin, California. Scientific Investigations Report 20065152. U.S. Geological Survey.
Harbaugh, A.W., Banta, E.R., Hill, M.C., & McDonald, M.G. 2000. MODFLOW-2000,
the U.S. Geological Survey modular ground-water model–User guide to modularization
concepts and the ground-water flow process: U.S. Geological Survey Open-File Report
0092, 121 p. Tech. rept.
Hillel, Daniel. 2000. Salinity management for sustainable irrigation : integrating science,
environment, and economics. Washington, DC.: World Bank.
Huling, S.G., & J.S., Weaver. 1991. Ground Water Issue: DENSE NONAQUEOUS
PHASE LIQUIDS. Tech. rept. EPA/540/4-91-002. US, Environmental Protection
Agency.
Iman, Ronald L., & Conover, W.J. 1980. Small sample sensitivity analysis techniques for
computer models.with an application to risk assessment. Communications in Statistics
- Theory and Methods, 9(17), 1749–1842.
Jones and Stokes Associates, & Mussetter Engineering Inc. 2002. San Joaquin River Pilot
Project 2000 Final baseline vegetation and physical variable data collection summary,
Sacramento, CA. Tech. rept.
Kitanidis, PK. 1990. Effective hydraulic conductivity for gradually varying flow. Water
Resources Research, 26(6), 1197–1208.
Lenormand, R., & Zarcone, C. 1985. Invasion Percolation in an Etched Network -

Measurement of a Fractal Dimension. Physical Review Letters, 54(20), 2226–2229.
134
Mayer, K.U., Frind, E. O., & Blowes, D. W. 2002. Multicomponent reactive trans-
port modeling in variably saturated porous media using a generalized formulation for
kinetically controlled reactions. Water Resourc. Res., 38(9), 1174–1195.
McDonald, M.G., & Harbaugh, A.W. 1988. A modular three-dimensional finite-difference

ground-water flow model: U.S. Geological Survey Techniques of Water-Resources In-
vestigations, book 6, chap. A1, 586 p. Tech. rept. U.S. Geological Survery.
Metropolis, Nicholas, & Ulam, S. 1949. The Monte Carlo Method. Journal of the
American Statistical Association, 44(247), 335–341.
Miller, Gretchen R., Rubin, Yoram, Mayer, K. Ulrich, & Benito, Pascual H. 2008. Mod-
eling Vadose Zone Processes during Land Application of Food-Processing Waste Water
in California’s Central Valley. J Environ Qual, 37(5 Supplement), 43–57.
Newell, Charles J, Bowers, Richard L., & Rifai, Hanadi S. 1994. IMPACT OF NON-
AQUEOUS PHASE LIQUIDS (NAPLS) ON GROUNDWATER REMEDIATION. In:
Summer National AICHE Meeting;, August 16, 1994 Symposium 23, ”Multimedia
Pollutant Transport Models”. Unpublished Presentation.
Newell, Charles J., Acree, Steven D., Ross, Randall R., & Huling, Scott G. 1995. Ground
Water Issue: LIGHT NONAQUEOUS PHASE LIQUIDS. Tech. rept. EPA/540/S-
95/500. US, Environmental Protection Agency.
Page, R.W. 1977. Guide for data collection to calibrate a predictive digital ground-
water model of the unconfined aquifer in and near the city of Modesto, California:
U.S. Geological Survey Water-Resources Investigations Report 76-41. Tech. rept. U.S.
Geological Survery.
Page, R.W. 1986. Geology of the fresh ground-water basin of the Central Valley, Cali-
fornia, with texture maps and sections. Professional Paper 1401-C. USGS.
Phillips, Steven P., Green, Christopher T., Burow, Karen R., Shelton, Jennifer L., &
Rewis, Diane L. 2007. Simulation of multiscale ground-water flow in part of the North-
eastern San Joaquin Valley, California. SIR 20075009. USGS.
Planert, Michael, Williams, John S., & Geological, Survey. 1995. Ground water atlas of
the United States. Segment 1, California, Nevada. U.S. Geological Survey.
Pollock, David W. 1994 (September). Users Guide for MODPATH/MODPATH-PLOT,

Version 3: A particle tracking post-processing package for MODFLOW, the U. S.
Geological Survey finite-difference ground-water flow model. Tech. rept. Open-File
Report 94-464. U. S. Geological Survey, Reston, Virginia.
Porter-Cologne Act. 2006. Porter-Cologne Water Quality Control Act. Cal. State Water
Resources Control Board.
135
Purkey, D.R., Thomas, G.A., Byrne, S., Cheng, A., & Harrison., N.E. 2001 (Septem-
ber). System-wide Conjunctive Water Management: The hydrogeologic suitability of
potential groundwater banking sites in the Central Valley of California. Tech. rept.
The Natural Heritage Institute, (www.n-h-i.org).
Rhoades, J.D., Kandiah, A., & Mashali, A.M. 1992. The use of saline waters for crop
production. Food and Agriculture Organization of The United Nations.
Rubin, Y, Sun, A., Maxwell, R., & Bellin, A. 2000. The concept of block-effective
macrodispersivity and a unified approach for grid-scale- and plume-scale-dependent
transport doi:null. Journal of Fluid Mechanics, 395(-1), 161–180.
Rubin, Y., Bellin, A., & Lawrence, A. E. 2003. On the use of block-effective macrodis-
persion for numerical simulations of transport in heterogeneous formations. Water
Resources Research, 39(9).
Rubin, Yoram. 2003. Applied stochastic hydrogeology. New York, NY: Oxford Univ.
Press.
Rubin, Yoram, Benito, P., Miller, G.R., McLauglin, J., Hou, Z., Hermanowicz, S., Mayer,
K.U., & Sillin, D. 2007. Hilmar Supplemental Environmental Project (SEP) draft
report: Volume II. Tech. rept. Hilmar SEP, Oakland, CA.
Schäfer, Wolfgang. 2003. Applied 3-D Groundwater Modeling with MODFLOW / MT3D
/ MT3DMS – Transport Modeling. Odenwaldstrae 6, 69168 Wiesloch, Germany:
Steinbeis-Transferzentrum Grundwassermodellierung.
Schoups, G. 2004. Regional-scale hydrologic modeling of subsurface water flow and reac-
tive salt transport in the western San Joaquin Valley. Ph.D. thesis, U.C. Davis.
Sunding, David, & Zilberman, David. 2005. A guide to consideration of economics under
the California Porter-Cologne Act. Sacramento, CA: California Resource Management
Institute.
Sunding, David, Berkman, M., Hamilton, S., Anderson, T., Kavanaugh, M., Mannap-
peruma, J., & Zilberman, David. 2007a. Hilmar Supplemental Environmental Project
(SEP) draft report: Volume I. Oakland, CA: Hilmar SEP.
Sunding, David, Berkman, M., Hamilton, S., Anderson, T., Kavanaugh, M., Mannap-
peruma, J., & Zilberman, David. 2007b. Hilmar Supplemental Environmental Project
(SEP) draft report: Volume III. Oakland, CA: Hilmar SEP.
USGS. 1999. National Elevation Dataset: U.S. Geological Survey (USGS), EROS Data
Center. raster digital data.
Weissmann, G.S., Bennett, G.L., & Lansdale, A.L. 2005. Factors controlling sequence
development on Quaternary fluvial fans, San Joaquin Basin, California, USA. Page
136
169186 of: Harvey, A.M., Mather, A.E., & Stokes, M. (eds), Alluvial Fans: Geomor-
phology, Sedimentology, Dynamics. Geological Society, London, Special Publications,
251. 03058719/05/15. The Geological Society of London.
Wilkinson, D. 1984. Percolation Model of Immiscible Displacement in the Presence of

Buoyancy Forces. Physical Review A, 30(1), 520–531.
Wilkinson, D., & Willemsen, J. F. 1983. Invasion Percolation - a New Form of Percolation
Theory. Journal of Physics a-Mathematical and General, 16(14), 3365–3376.
Williamson, A.K., Prudic, D.E., & Swain, L.A. 1989. Ground-Water Flow in the Central
Valley, California: US Geological Survey Professional Paper 1401-D. Tech. rept. U.S.
Geological Survery.
Xu, B., Yortsos, Y. C., & Salin, D. 1998. Invasion percolation with viscous forces.
Physical Review E, 57(1), 739–751.
Zheng, Chunmiao, & Wang, P. Patrick. 1999 (December). MT3DMS: a modular three-
dimensional multispecies transport model for simulation of advection, dispersion, and
chemical reactions of contaminants in groundwater systems : documentation and users
guide. Tech. rept. Contract Report SERDP-99-1. prepared for U.S. Army Corps of
Engineers ; monitored by U.S. Army Engineer Research and Development Center.
137
Appendix A
Some Ancillary Stuff
Ancillary material should be put in appendices, which appear after the bibliography.

Benito Thesis

Uploaded by

Document Information

Original Description:

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

Benito Thesis

Uploaded by

Copyright:

Available Formats

Investigation of Contaminant Transport over Multiple Length Scales:

From the Micro to the Macro

B.A. Geology (Amherst College, Amherst, MA) 1994,

A dissertation submitted in partial satisfaction of the

Civil and Environmental Engineering

University of California, Berkeley

Investigation of Contaminant Transport over Multiple Length Scales: From the

Theses have elements. Isn’t that nice?

Prof. Tadeusz W. Patzek, Chair Date

the only person worthy of my company.

2 Pore network investigation of stability and scaling of two-phase immis-

2.12 Results from Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3 Modeling the Impacts of the Food Processing Industry on Regional

4 Summary and Conclusions 129

A Some Ancillary Stuff 137

2.1 Fluid phase distributions at breakthrough for different values of Bond

3.14 Lateral boundary conditions and included surface water features. . . . . . 55

3.27 Concentration of FDS (mg/l) at 30 years in model Layer 1. The colorbar is

3.59 Plot of measured monthly FDS concentrations in shallow groundwater

3.1 Comparison of untreated winery and domestic waste water characteristics

Modeling the Impacts of the Food

3.1 Background and Motivation

palatability of domestic drinking water is affected by salinity, and California drinking

3.1.2 Characteristics of Food Processing Waste Discharge

3.1.3 Attenuation of Salinity in the Vadose Zone

3.2 Overview of Objectives and Methodology

3.3 Hydrogeological Setting

3.3.1 Regional Hydrology

3.4 Water Recharge and Pumping

3.5 Numerical ground water Flow Model

water Modeling System), which provides pre/post-processing capabilities and integration

3.5.1 Ground Water Flow Equation

3.5.2 Spatial Discretization

3.5.3 Boundary Conditions

Ground Water - Surface Water Interactions

3.5.4 Hydrogeological Parameters

1. Corcoran Clay Layer

Lithologic Subregion KH KG KA Khor /Kvert

3.6 Results of Flow Model

Figure 3.17: Horizontal planar views through three-dimensional model of percentage

K =K : Computed vs. Observed Heads

Computed Heads - Observed Heads

3.7 Salinity Transport Modeling

3.7.1 Numerical Model

θ = the effective porosity of the subsurface medium, [-]

C k = dissolved concentration of kth species, [ML−3 ]

xi,j = distance along the respective Cartesian coordinate axis, [L]

Dij = hydrodynamic dispersion coefficient tensor, [L2 T−1 ]

Csk = concentration of sources or sink flux for species, k, [ML−3 ]

where αL is the longitudinal dispersivity [L], in the direction of flow, αT is the

3.7.2 Solute Transport Parameters

(a) Longitudinal Dispersivity, αL = 19.2 m

2. Corcoran Clay Layer

(a) Longitudinal Dispersivity, αL = 4 m

3.7.3 Initial and Recharge Concentrations

CprecipQprecip + Cgw Qgw + Csw Qsw

Qtot = Qprecip + Qgw + Qsw (3.15)

3.7.4 Land Application waste water Loading Rates

3.7.5 Salinity Transport Modeling Results

∆FDS(t) = FDS(t, with land discharge) − FDS(t,without land discharge) (3.16)

Figure 3.29: A difference map of, ∆FDS = FDS(with land discharge) −

3.7.6 The Effects of Spatial Variability

3.8 Sensitivity Analysis