hysteretic, variably saturated, transient flow and transport models … · 2020. 4. 2. ·...
TRANSCRIPT
Hysteretic, variably saturated, transient flowand transport models with numerical inversion
techniques to characterize a field soil in central Arizona
Item Type text; Dissertation-Reproduction (electronic)
Authors Thomasson, Mark John
Publisher The University of Arizona.
Rights Copyright © is held by the author. Digital access to this materialis made possible by the University Libraries, University of Arizona.Further transmission, reproduction or presentation (such aspublic display or performance) of protected items is prohibitedexcept with permission of the author.
Download date 02/12/2020 08:28:02
Link to Item http://hdl.handle.net/10150/284257
INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI films the
text directly from the original or copy submitted. Thus, some thesis and
dissertation copies are in typewriter face, while others may be from any type of
computer printer.
The quality of this reproduction is dependent upon the quality of the copy
submitted. Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleedthrough, substandard margins, and improper alignment
can adversely affect reprxluction.
In the unlikely event that the author did not send UMI a complete manuscript arxJ
there are missing pages, these will t>e noted. Also, if urtauthorized copyright
material had to be removed, a note will indicate the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning
the original, beginning at the upper left-hand comer and continuing from left to
right in equal sections with small overtaps.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Higher quality 6' x 9' black and white photographic
prints are available for any photographs or illustrations appearing in this copy for
an additional charge. Contact UMI directly to order.
Bell & Howell lnformatk>n and Learning 300 North Zeeb Road, Ann Arbor. Ml 48106-1346 USA
UIVQ ̂800-521-0600
HYSTERETIC, VARIABLY SATURATED, TRANSIENT FLOW AND TRANSPORT MODELS WITH NUMERICAL INVERSION TECHNIQUES TO CHARACTERIZE A
FIELD SOIL IN CENTRAL ARIZONA
by
Mark J. Thomasson
Copyright © Mark J. Thomasson 2000
A Dissertation Submitted to the Faculty of the
DEPARTMENT OF HYDROLOGY AND WATER RESOURCES
In Partial Fulfillment of the Requirements For the Degree of
DOCTOR OF PHILOSOPHY WITH A MAJOR IN HYDROLOGY
In the Graduate College
THE UNIVERSITY OF ARIZONA
2000
UMI Number: 9992070
Copyright 2000 by
Thomasson, Mark John
All rights reserved.
UMI UMI Microform 9992070
Copyright 2001 by Bell & Howell Information and Learning Company. All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
Bell & Howell Information and Leaming Company 300 North Zeeb Road
P.O. Box 1346 Ann Arbor. Ml 48106-1346
2
THE UNIVERSITY OF ARIZONA ® GRADUATE COLLEGE
As members of the Final Examination Committee, we certify that we have
read the dissertation prepared by J- Thomasson
entitled Hysteretic, Variably Saturated. Transient Finw anH
Transport Models with Numerical Inversion Techniques to
Characterize a Field Soil in Central Arizona
and recommend that it be accepted as fulfilling the dissertation
requirement for the Degree of Doctor of Philosophy
/ J Zdoo Date ' Dr. Pe J. wiere
dy L. Bassett Date
/0/4 Date
0 ?) "E?
Sr. Arthur W. Warrick
Date
Date
Dr. Tian-Chyi J. \ ) Date Final approval and acceptance of this dissertation is contingent upon the candidate's submission of the final copy of the dissertation to the Graduate College.
I hereby certify that I have read this dissertation prepared under my direction and recommend that it be accepted as fulfilling the dissertation requirement.
)issertation^irect
-7/ ,7 Qrr -̂ Di ssertaticfn Di
-2. Dr. Peter J. Wierenga Date
»-o
rector Dr. Randy L. Bassett li /J Date
3
STATEMENT BY AUTHOR
This dissertation has been submitted in partial fulfillment of requirements for an advanced degree at The University of Arizona and is deposited in the University Library to be made available to borrowers under rules of the Library.
Brief quotations from this dissertation are allowable without special permission, provided that accurate acknowledgment of source is made. Requests for permission for extended quotation from or reproduction of this manuscript in whole or in part may be granted by the copyright holder.
SIGNED:
4
ACKNOWLEDGEMENTS
Many people were involved in the successful completion of this dissertation. I
would first like to thank my committee for their valuable time and input: Dr. Peter
Wierenga, Dr. Randy Bassett, Dr. Paul Ferre, Dr. Arthur Warrick, Dr. Shlomo Neuman,
and Dr. Jim Yeh.
I thank Dr. Peter Wierenga and the College of Agriculture for funding this
research. Dr. Wierenga has provided me with thoughtful guidance throughout this
project and his keen observations and questions have challenged me to grow as a
scientist. Dr. Wierenga also dedicated a tremendous amount of his time to the reading
and editing of this dissertation.
Dr. Randy Bassett has supplied me with a great deal of invaluable mentoring
throughout my graduate academic career. He has consistently contributed
encouragement and expert advice. Dr. Bassett's enthusiastic approach to teaching and
research has always inspired me to match his enthusiasm. In addition. Dr. Bassett
provided numerous opportunities to grow as an instructor, encouraging me to teach a
wide range of topics in his classes.
I thank Dr. Paul Ferre for the copious amount of time he has spent with me
discussing both my research ideas and the challenges I have faced during the course of
this project. His perceptiveness and impressive scientific knowledge have not only
offered me many ways to overcome obstacles, but also have stimulated a wealth of new
directions to explore.
Dr. Arthur Warrick, Dr. Shlomo Neuman, and Dr. Jim Yeh, have all given their
time to explore specific issues and have challenged me, through their courses, to
continue to develop my scientific knowledge.
Finally, there are a number of people that I have encountered throughout my
graduate academic career that have enhanced my experience at Arizona: Rob, Ty,
Leslie, Dave, Lori, Ingrid, Emory, Eric, and Evan. Most importantly, a big thank you
and much love to Alissa and Bo.
DEDICATION
To my friends and family with love. Thank you.
6
TABLE OF CONTENTS
LIST OF FIGURES 8
LIST OF TABLES 13
ABSTRACT 14
1. INTRODUCTION AND OBJECTIVES 16 l . L O v e r v i e w 1 6 1.2. Objectives 16
2. INFILTRATION EXPERIMENTS 19 2.1. Site Description 19 2.2. Infiltration and Transport Experiments 20
3. FLOW IN VAIUABLY SATLFRATED HETEROGENEOUS MEDL\ INCORPORATING HYSTERESIS 29 3.1. Introduction 29 3.2. Materials, Methods, and Theory 30
3.2.1. Water Budget 30 3.2.2. Modeling Water Flow 31
3.3. Results and Discussion 46 3.3.1. Textural Characterization 46 3.3.2. Water Budget 46 3.3.3. Simulation of Cases I and II 59 3.3.4. Comparison of Cases II and III: Incorporating Hysteresis 64 3.3.5. Simulation of Cases IV and V 71 3.3.6. Water Contents with Depth, Experiment 1 72 3.3.7. Simulation of Cases VI and VII, Experiment II 78
3.4. Conclusions 83
4. FIELD SCALE SOLUTE TRANSPORT THROUGH VARIABLY SATURATED MEDIA INCORPORATING HYSTERESIS 86 4.1. Introduction 86 4.2. Transport Models 89
4.2.1. Parameter Estimation and Analytical Solutions to Solute Transport 89 4.2.2. Numerical Modeling of Solute Transport 92
4.3. Measured Bromide Concentrations 95 4.4. Methods 101
4.4.1. Mass Balance 101 4.4.2. Approach 102
4.4.2.1. Parameter Estimation 103 4.4.2.2. Predictions Using an Analytical Model 105
7
TABLE OF CONTENTS - Continued
4.4.2.3. Numerical Solute Transport Simulations 107 4.5. Results and Discussion 113
4.5.1. Mass Balance 113 4.5.2. Modeling 117
4.5.2.1. Parameter Estimation, Cases I and II 117 4.5.2.2. Evaluation of the Local and Global Approaches Using Analytical
Models, Cases in - VI 119 4.5.2.3. Numerical Solute Transport Simulations, Cases VII - X 122
4.5.2.3.1 Effect of Hysteresis on Bromide Transport 122 4.5.2.3.2 Preferential Transport 130
4.6. Conclusions 145
5. CONCLUSIONS 149
APPENDIX: CHARACTERIZATION 157
REFERENCES 167
8
LIST OF FIGURES
Figure 2.1. Location of the Maricopa Agricultural Center (MAC) 21
Figure 2.2. Experimental plot showing locations where data were collected 22
Figure 2.3. Monitoring island schematic 23
Figure 2.4. Input concentrations of bromide during Experiment 1 28
Figure 3.1. Locations where soil cores were collected 36
Figure 3.2. Representative soil tension profile for the monitoring island locations 40
Figure 3.3. Hydraulic conductivity versus water content. Solid circles are the field-calculated values fi'om O.S to 3 m in 0.5 m intervals; the solid line is based on a laboratory-derived value of 45.8 cm d"' for Ks, the solid line with open circles is based on a Ks of450 cm d ' 41
Figure 3.4. Model initial conditions and observed initial conditions. The circles represent the mean observed water contents. The horizontal lines represent ± one standard deviation and the solid line represents the interpolated water contents used in the model (HYDRUS) 44
Figure 3.5. Mean percent sand, silt, and clay composition at three locations 47
Figure 3.6. Mean percent gravel (circles) and mean measured water content (solid line) versus depth 48
Figure 3.7. Spatial distribution of the water budget data for irrigation day 4, 1.0 m depth. Diamonds represent vertical neutron probe access tube locations. The numbers represent under- or over-estimation of water applied (in percent) relative to the water meter data. On average the neutron probe data over- predicted how much was actually applied by 28.4 percent 50
Figure 3.8. Spatial distribution of the water budget data for irrigation day 5, 1.0 m depth. Diamonds represent vertical neutron probe access tube locations. The numbers represent under- or over-estimation of water applied (in percent) relative to the water meter data. On average the neutron probe data under- predicted how much was actually applied by 17.6 percent 51
Figure 3.9. Spatial distribution of the water budget data for irrigation day 18, 9.0 m depth. Diamonds represent vertical neutron probe access tube locations. The numbers represent under- or over-estimation of water applied (in percent) relative to the water meter data. On average the neutron probe data over- predicted how much was actually applied by 0.2 percent 53
9
LIST OF FIGURES - Continued
Figure 3.10. Computed water depths applied versus the water depths applied according to the water meters for 34 access tube locations on irrigation days when data are available. The dashed line represents +/- 32 percent error from the true amount applied 55
Figure 3.11. Computed water depths applied versus the water depths applied according to the water meters for the perimeter access tube locations (8 locations). The dashed line represents +/- 32 percent error from the true amount applied 56
Figure 3.12. Computed water depths applied versus the water depths applied according to the water meters for the access tube locations located at least 5 m within the irrigated perimeter (26 locations). The dashed line represents +/- 32 percent error from the true amount applied 58
Figure 3.13. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 7.5 m depth. The vertical lines denote ± one standard deviation in the mean water content 62
Figure 3.14. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 7.5 m depth. The vertical lines denote ± one standard deviation in the mean water content 65
Figure 3.15. Comparison of the simulated water content time series to the mean measured water contents for the 1.5 m depth. Triangles depict tension 66
Figure 3.16. Comparison of the simulated water content time series to the mean measured water contents for the 3.0 m depth. Triangles depict tension 67
Figure 3.17. Comparison of the simulated water content time series to the mean measured water contents for the 5.0 m depth. Triangles depict tension 68
Figure 3.18. Retention curves for hysteretic model and models based on parameters derived from wetting experiments and drying experiments 69
Figure 3.19. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 7.5 m depth. The vertical lines denote ± one standard deviation in the mean water content. A, Case III with Ks = 250.0 cm d"', B, Case II with Ks = 450.0 cm d"', C, Case V with Ks = 1250.0 cm d ', D, Case IV with Ks = 450.0 cm d"' 73
Figure 3.20. Water content measurements and model predictions for day 22. The circles represent measured water contents 74
Figure 3.21. Water content measurements and model predictions for day 24. The circles represent measured water contents 75
Figure 3.22. Water content measurements and model predictions for day 93. The circles represent measured water contents 76
10
LIST OF FIGURES - Continued
Figure 3.23. Model initial conditions and observed initial conditions for Experiment II. The circles represent the mean observed water contents. The horizontal lines represent ± one standard deviation and the solid line represents the interpolated water contents used in the model (HYDRUS) 79
Figure 3.24. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 1.5 m depth. The vertical lines denote ± one standard deviation in the mean water content 80
Figure 3.25. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 3.0 m depth. The vertical lines denote ± one standard deviation in the mean water content 81
Figure 3.26. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 7.5 m depth. The vertical lines denote ± one standard deviation in the mean water content 82
Figure 4.1. Model initial conditions and observed initial conditions. The circles represent the mean observed water contents. The horizontal lines represent ± one standard deviation and the solid line represents the interpolated water contents used in the model (HYDRUS). The dashed line represents the initial conditions for the bromide used in the model (HYDRUS) 94
Figure 4.2. Relative bromide concentrations from the NMI, east side 96
Figure 4.3. Relative bromide concentrations from the NMI, west side 97
Figure 4.4. Relative bromide concentrations from the SMI, east side 98
Figure 4.5. Relative bromide concentrations from the SMI, west side 99
Figure 4.6. Amounts of bromide recovered at the monitoring islands from day 5 to day 15, the solid line represents 100 percent recovery 114
Figure 4.7. Comparison of measured bromide data and the predicted bromide concentrations at the SMI, west location. The solid line represents model predictions from Case III based on transport properties estimated locally (Case I), the dashed line represents model predictions from Case IV based on transport properties estimated globally (Case II) 120
Figure 4.8. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations. The dashed line represents model predictions from Case V based on average transport properties estimated locally (Case I), the solid line represents model predictions from Case VI based on transport properties estimated globally (Case II) 123
II
LIST OF FIGURES - Continued
Figure 4.9. Comparison of all measured bromide data from around the monitoring islands and the predicted bromide concentrations. The solid line represents model predictions with hysteresis in the hydraulic properties (Case VIII); the solid line with solid diamonds represents model predictions with no hysteresis in the hydraulic properties (Case VII) 124
Figure 4.10. Measured bromide concentrations and model predictions for day 16, irrigation day 16 127
Figure 4.11. Measured bromide concentrations and model predictions for day 24, the last day of irrigation 128
Figure 4.12. Measured bromide concentrations and model predictions for day 93, the last measured day of redistribution 129
Figure 4.13. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 50 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VIII {R ~ 0.87), the dashed line represents results from Case IX (/? ~ 1.0), the solid line with solid circles represents results from Case VIII (/? ~ 0.8) 131
Figure 4.14. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 100 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VIII {R ~ 0.87), the dashed line represents results from Case IX (/? ~ 1.0), the solid line with solid circles represents results from Case VIII (/?-0.8) 132
Figure 4.15. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 150 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VIII {R ~ 0.87), the dashed line represents results from Case IX (/? ~ 1.0), the solid line with solid circles represents results from Case VIII (/?-0.8) 133
Figure 4.16. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 200 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VIII {R ~ 0.87), the dashed line represents results from Case IX (/? ~ 1.0), the solid line with solid circles represents results from Case VIII (/?~0.8) 134
12
LIST OF FIGURES - Continued
Figure 4.17. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 250 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VIII {R ~ 0.87), the dashed line represents results from Case IX (/? ~ 1.0), the solid line with solid circles represents results from Case VIII (/? ~ 0.8) 135
Figure 4.18. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 300 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VTII {R ~ 0.87), the dashed line represents results from Case IX (/? ~ 1.0), the solid line with solid circles represents results from Case Vin(/?~0.8) 137
Figure 4.19. Comparison of measured bromide data and the predicted bromide concentrations at the NMI, west location. The solid line represents model predictions from Case VTU {R ~ 0.87), the dashed line represents model predictions from Case IX (/? = 1.0, Young et al., 1999), and the solid line with solid circles represents model predictions from Case VIII {R ~ 0.80) 138
Figure 4.20. Comparison of measured bromide data and the predicted bromide concentrations at the NMI, east location. The solid line represents model predictions from Case VIII {R ~ 0.87), the dashed line represents model predictions from Case IX (/? = I.O, Young et al., 1999), and the solid line with solid circles represents model predictions from Case VTII (/? ~ 0.80) 139
Figure 4.21. Comparison of measured bromide data and the predicted bromide concentrations at the SMI, west location. The solid line represents model predictions from Case VIII (/? ~ 0.87), the dashed line represents model predictions from Case IX (^ = 1.0, Young et al., 1999), and the solid line with solid circles represents model predictions from Case VIII (/? ~ 0.80) 140
Figure 4.22. Comparison of measured bromide data and the predicted bromide concentrations at the SMI, east location. The solid line represents model predictions from Case VTII (/? ~ 0.87), the dashed line represents model predictions from Case IX (/? = 1.0, Young et al., 1999), and the solid line with solid circles represents model predictions from Case VIII {R - 0.80) 143
Figure 4.23. Comparison of all measured bromide data from around the monitoring islands and the predicted bromide concentrations. The solid line represents model predictions using a uniform input concentration (Co = 31.6 mg /1, Case VIII); the dashed line represents model predictions using variable input concentrations (Figure 2.4, Case X) 144
13
LIST OF TABLES
Table 2.1. Tabulated values of applied water for Experiment 1 26
Table 2.2. KBr applied during Experiment I* 27
Table 3.1. Mean Hydraulic Properties and the statistical parameters for the core samples obtained by the various laboratory methods 37
Table 3.2. Calculated unsaturated hydraulic conductivity values for the NMI and SMI locations 42
Table 3.3. Summary of mean hydraulic properties used in models 45
Table 3.4. Computed depths of water added on irrigation day 5 at 1.0 m 52
Table 3.5. Mean square error* for Cases I - VTI 60
Table 3.6. Saturated hydraulic conductivity values used for Cases I - VII 63
Table 4.1. Breakthrough times for the monitoring island locations 100
Table 4.2. Locations where bromide breakthrough occurred within the irrigation timeframe 104
Table 4.3. Input parameters used in the inverse models. Cases I and II 106
Table 4.4. Transport parameters for the analytical model, Cases III - VI 106
Table 4.5. Mean transport properties used in the numerical models, Cases VII - IX.. 109
Table 4.6. Summary of test cases for field scale solute transport 112
Table 4.7. Estimated transport properties from Cases I and II 118
Table 4.8. MSE values for Cases III - VI 121
Table 4.9. Mean transport properties from Cases I and II used in Cases V and VI 121
Table 4.10. Mean square error for Cases VI - IX using average measured bromide concentrations 125
14
ABSTRACT
Field-measured data collected from infiltration experiments conducted at the Maricopa
Agricultural Center, Arizona, and one-dimensional numerical models were used in an
iterative approach to develop conceptual models of unsaturated flow and transport.
Conceptual models were developed using field data, estimates of hydraulic and
transport properties, and assimiptions regarding the types of processes occurring in the
field plot. Using the field data and numerical models in an iterative approach revealed
that adjustments to the parameterization of the conceptual models were necessary to
allow for acceptable predictions of flow and transport. In addition, the numerical
models delineated where the conceptual models needed adjustment and indicated where
further analysis of the field data was necessary. Specifically, in regards to the
unsaturated water flow, it was found that: 1) without increasing the laboratory
determined value of Ks tenfold, it was impossible to make acceptable predictions for the
infiltration experiments; 2) the laboratory-derived hydraulic properties provided a
superior first approximation to texturally based hydraulic properties taken from an
internal database; 3) incorporation of hysteresis into the soil hydraulic functions
resulted in predictions of the water contents through time and soil water content profiles
that were marginally better than not incorporating hysteresis; and 4) based on the mean
square errors of the predictions and a post-audit based on the adjusted hydraulic
properties, the use of a homogeneous soil profile was supported. In regard to the
unsaturated solute transport, it was found that: I) inversion techniques to estimate the
15
unsaturated transport properties at each location (local approach) yielded the best results
with respect to the fit of the data compared to the techniques which fit data from all
depths simultaneously (global approach); 2) predictions using unsaturated transport
properties averaged fi-om the global approach yielded the best results compared to the
average local approach properties; 3) incorporating hysteresis had little effect on the
predicted bromide breakthrough curves and concentration profiles, yielding only
marginally better results than predictions fi-om models with no hysteresis; 4) it was
impossible to accurately predict the breakthrough of bromide at depths deeper than 200
cm without using a retardation factor less then one.
16
1. INTRODUCTION AND OBJECTIVES
1.1. Overview
Topical introductions are offered at the beginning of chapters specific to the
subject being examined. Chapter One outlines the objectives of this dissertation.
Chapter Two gives a description of the experimental plot where the infiltration studies
were conducted and outlines the specific details of the infiltration experiments. Chapter
Three examines the unsaturated water flow. Chapter Four investigates the unsaturated
solute transport. Chapter Five summarizes the conclusions of the unsaturated flow and
transport modeling.
1.2. Objectives
The objective of this dissertation was to use field measured data and numerical
models in an iterative approach to assess and quantify the conceptual model of flow and
transport. Specifically, the objective was to determine how changes to the
parameterization of the conceptual model affect the flow and transport predictions.
This objective was attained by separate treatments of the water flow and the transport of
a solute through a variably saturated heterogeneous field soil. These objectives are
discussed in detail in two chapters: Chapter 3 focuses on simulating the unsaturated
water flow; Chapter 4 investigates the transport of a solute.
17
The simulation of the unsaturated water flow was compared to data obtained from
infiltration experiments conducted at a field plot. The infiltration and redistribution of
water into a large (50 m by 50 m), controlled and monitored field plot was modeled and
the results compared with measured data. Effects of the inclusion of hysteresis in the
retention and hydraulic conductivity functions were determined, and the ability of
laboratory derived hydraulic parameters from core samples to accurately predict the
propagation of the wetting firont under a field setting was examined. Both the wetting
front arrival and the changes in water content using laboratory derived hydraulic
properties incorporating hysteresis were modeled. Sub-objectives of the study were:
1. assess the ability of hydraulic properties measured on soil cores in the laboratory to
represent field properties,
2. compare the results of models using laboratory derived hydraulic properties to
models using hydraulic properties taken from a database based on textural
considerations,
3. examine the effects of incorporating hysteresis into the soil hydraulic functions on
unsaturated flow predictions, and
4. evaluate the validity of using a homogeneous representation of the profile to capture
the average field behavior.
The transport of a solute through a variably saturated field soil was simulated
using analytical and numerical models. This was achieved through integrating the
18
estimation of the transport properties, the development of a conceptual model, and the
impact of the model on transport predictions. The techniques for estimating transport
properties were evaluated and the effects of spatial averaging of point measurements (0,
R, D, a) on the construction of an equivalent porous media were investigated. Location
specific and spatially integrated inversion methods were compared to assess their ability
to characterize the flux and transport properties. Additionally, the impact of averaging
these properties on the prediction of flow and transport and the effects of hysteresis on
solute transport were investigated. Sub-objectives of the study were:
1. use inversion techniques to estimate the solute transport properties,
2. evaluate two schemes for averaging the estimated transport properties and assess
which scheme provides the best results using analytical models,
3. examine the effects of incorporating hysteresis into the soil hydraulic functions on
the solute transport predictions, and
4. assess and examine the effect of the retardation coefficient on the solute transport
predictions.
19
2. INFILTRATION EXPERIMENTS
2.1. Site Description
The field study was conducted by Young et al. [1999] at the Maricopa
Agricultural Center (MAC) of the University of Arizona starting in the spring 1997 and
ending in the summer 1998. The MAC site is located 42 kilometers (km) southwest of
Phoenix [Young et al., 1999] (Figure 2.1). The study was designed and implemented to
evaluate unsaturated zone monitoring techniques and strategies applicable to low-level
radioactive waste (LLW) disposal facilities [Young et al., 1999]. The study consisted of
two successive infiltration experiments.
The MAC site is situated within a northwest trending alluvial basin, situated
within the Basin and Range region. The Basin and Range region is characterized by
northwest to southeast trending mountain ranges separated by broad alluvial valleys
[Resources, 1993]. The alluvium in the basin is anywhere fi-om 60 to 150 meters (m) in
thickness, and is comprised primarily of unconsolidated to slightly consolidated sands
and gravels. Some finer grained lenses are distributed throughout the unit. The
regional groundwater table is at approximately 200 m depth. The site lies in the
alluvium with localized perched zones of groundwater at approximately 11 - 12 m
depth. The region is in a semi-arid environment and receives approximately 20
centimeters (cm) of rainfall annually. Precipitation is characterized by intense, short
duration summer storms and longer duration low intensity winter rains. The region
20
experiences hot siunmers, and mild winters with an average annual temperature of 21°
Celsius.
2.2. Infiltration and Transport Experiments
Solute transport tracer experiments were conducted on a 50 m by 50 m plot
located within field plot F-115 at the MAC site (Figure 2.2). The 50 m by 50 m plot
was covered by a 60 m by 60 m, 0.081 cm thick Hypalon® pond liner to minimize
surface evaporation. The two monitoring islands shown as circles on Figure 2.2 were
constructed using 4 m long, 3.65 m diameter highway culverts placed vertically in 3.7
m diameter auger holes. Through ports in the culvert walls, the undisturbed soil around
the culvert could be accessed (Figure 2.3). Single-chamber solution samplers were
installed through the walls of the north and south monitoring islands (NMI and SMI)
from 0.5 m to a depth of 3.0 m in 0.5 m intervals [Young et al., 1999]. Dual-chamber
solution samplers were installed in the nested boreholes at depths of 3, 5, and 10 m at
the locations shown by solid triangles in Figure 2.2.
Water with potassium bromide (KBr) was applied to the experimental plot using
164 drip lines. The parallel drip lines were spaced 0.3 m apart with self-cleaning
emitters every 0.3 m along the line. A total of 27,000 emitters were so distributed over
the 50 m X 50 m area.
21
M7
Phoenix area
Maricopa site 1-8
1-10
Tucson 100
1-1!
MEXICO
Figure 2.1. Location of the Maricopa Agricultural Center (MAC).
22
Covered area Field Plot F-115
^ Irrigated area
h O Monitoring island
• Neutron probe access tube
Tensiometer
10 meters
Figure 2.2. Experimental plot showing locations where data were collected.
23
Annular backfill
Tensiometers, Single chamber solution sampler
Neutron probe access tube
Plan view Soil Surface
0 m
0.5 m
3.0 m
Side view
Figure 2.3. Monitoring island schematic.
24
Experiment I lasted 93 days, starting April 28, 1997 and ending on July 30,
1997. A KBr solution with a mean concentration of 31.6 milligrams per liter (mg /1)
was applied for 15 days immediately followed by a 9 day application of irrigation water
that contained less than 1.0 mg / 1 of bromide (April 28, 1997 to May 21, 1997) at an
average flux rate of 1.85 centimeters per day (cm d ') (Table 2.1) for an application of
44.4 cm of water and 2.19 x 10^ mg of KBr (Table 2.2). Redistribution of the solute in
the profile was measured for 70 days following cessation of irrigation. Although a
mean value of 31.6 mg / 1 of bromide was applied during Experiment I it should be
noted that the input concentration was quite variable during application with a standard
deviation of 6.1 mg /1 (Figure 2.4). The variability of the input bromide concentrations
cannot be explained by analytical error alone. Other sources of error are incomplete
mixing of the KBr in the mixing tank and a non-uniform distribution of the bromide by
the application system. Young et al. [1999] noted that it was difficult to control the
fluid volume in the mixing tanks thereby creating a variable input bromide
concentration during the irrigation period. Day 12 represents an extreme case of the
mixing problems where the fluid overflowed the top of the mixing tank resulting in a
very low input bromide concentration for that day.
Experiment II lasted 210 days starting on December 3, 1997 and ending July 1,
1998. Water was applied at an average flux rate of 1.97 cm d"' for 33 days (total of 64.8
cm applied) with redistribution measured for the following 177 days. Although no KBr
added to the irrigation water during the infiltration phase of Experiment II, samples
were collected from the solution samplers to quantify the displacement process.
During the 93 days of Experiment I, 898 samples were collected and analyzed for
bromide from the monitoring island locations, 426 of these samples were collected
during the irrigation period. For the duration of Experiment II, 258 samples were
collected and analyzed for bromide from the monitoring island locations. All of these
were collected during the irrigation phase of the experiment.
26
Table 2.1. Tabulated values of applied water for Experiment I
Time Experiment I Experiment II (days) Applied (cm) Sum applied (cm) Applied (cm) Sum applied (cm)
0 1.81 1.81 0.72 0.72 1 1.84 3.64 1.33 2.04 2 1.85 5.49 1.96 4.01 3 1.82 7.31 1.97 5.97 4 1.95 9.26 1.97 7.94 5 1.87 11.12 1.99 9.93 6 1.87 12.99 1.96 11.89 7 1.86 14.86 1.97 13.85 8 1.86 16.71 1.96 15.82 9 1.85 18.56 1.97 17.79 10 1.86 20.42 1.91 19.70 11 1.82 22.23 1.98 21.68 12 1.82 24.05 1.98 23.66 13 1.84 25.89 1.98 25.64 14 1.83 27.72 1.98 27.63 15 1.83 29.55 1.98 29.60 16 1.83 31.38 1.97 31.58 17 1.85 33.23 1.96 33.54 18 1.83 35.06 1.99 35.53 19 1.84 36.90 1.97 37.50 20 1.85 38.75 1.97 39.48 21 1.86 40.61 1.97 41.45 22 1.84 42.45 1.93 43.38 23 1.96 44.41 1.93 45.31
Table 2.1 (Continued). Applied water for Experiment I
Time Experiment I Experiment II (days) Applied (cm) Sum applied (cm) Applied (cm) Sum applied (cm)
24 1.92 47.23 25 1.92 49.15 26 1.97 51.13 27 1.97 53.10 28 1.97 55.07 29 1.96 57.03 30 1.96 58.98 31 1.95 60.93 32 1.95 62.87 33 1.95 64.82
Mean 1.85 1.96 SD 0.04 0.02
CV (%) 1.90 0.99
Table 2.2. KBr applied during Experiment I*
Time (days) Concentration (mg cm'^) 0 0.0345 1 0.0337 2 0.0312 3 0.0300 4 0.0365 5 0.0405 6 0.0276 7 0.0311 8 0.0350 9 0.0400 10 0.0344 11 0.0290 12 0.0158 13 0.0266 14 0.0278
Mean 0.0316 * from Young et al. [1999]
28
45
CO
40
S 35
30
31.6 mg/1
t
c o u c o U
25
20
15
0 6 8 Time (day)
10 12 14
Figure 2.4. Input concentrations of bromide during Experiment I.
29
3. FLOW IN VARIABLY SATURATED HETEROGENEOUS MEDIA INCORPORATING HYSTERESIS
3.1. Introduction
Water flow in unsaturated soils is often difficult to quantify because it is a
function of moisture content, matric potential and hydraulic conductivity. Water flow
problems are usually solved numerically although analytical solutions are available for
simplified systems. However, these simplified systems rarely exist in the real world.
Understanding and quantifying the mechanisms of a wetting front advance in an
unsaturated soil profile is an important step in studying the complex nature of
unsaturated flow. The advance of a wetting front is also important in understanding the
transport of contaminants under unsatiu^ted conditions.
A field study was conducted at the Maricopa Agricultural Center (MAC) (Figure
2.1) of the University of Arizona starting in the spring 1997 and ending in the summer
1998. The MAC s i te i s loca ted 42 ki lometers (km) southwest o f Phoenix [Young e t a i ,
1999]. The study was designed and implemented to evaluate unsaturated zone
monitoring techniques and strategies applicable to low-level radioactive waste (LLW)
disposal facilities [Young et al., 1999]. The study consisted of two successive
infiltration experiments.
The objective of the present study is to analyze the data obtained fi*om these two
experiments, to simulate the advance of the wetting firont and compare the results with
measured data, to determine if inclusion of hysteresis in the retention and hydraulic
30
conductivity functions affects the above outcome, and to evaluate the ability of
laboratory derived hydraulic parameters from core samples to accurately predict the
propagation of the wetting front.
3.2. Materials, Methods, and Theory
3.2.1. Water Budget
A water budget was calculated for the first experiment using the measured water
content data from the neutron probe access tubes. The water budget compares the
amount of water applied to the upper surface of the plot through the trickle system, to
the amount of water contained within the soil profile at specific times during the
experiment. The measured amount of water in the soil profile was determined from the
field measured water contents by integrating the soil water profile at specific times
during the experiment. The amount of water applied to the plot through the trickle
system was calculated from water meters located at the inlet water pipes of the irrigated
plot. A water budget was calculated according to the following equations;
Water budget = measured - fv .
known (3.1)
fV{z , t ) = Water measured = (3.2) 0
^0=^,-.One-^..O (3.3)
31
where Wmeasured IS thc depth (cm) of water applied, Az is the vertical depth interval along
the length of the neutron probe, and ^is the water content. The water budget was
calculated for various depths and times to determine the difference between water
applied and water measured in the soil profile. This also provided us with an
independent check of the neutron probe calibration and allowed us to quantify the
effects of two-dimensional flow at the perimeter of the irrigated plot. It should be noted
that the depth interval and times over which the mass balance was conducted were
carefully chosen to make certain that the termination point was in front of the wetting
front and that the water content profile was not truncated.
3.2.2. Modeling Water Flow
HYDRUS 1-D [Simuneket al., 1998] was utilized to simulate the flow of water
through the soil profile. HYDRUS 1-D solves a mixed form of Richards' equation.
The governing equation, without a source or sink term, is given by:
= (3-4) dt dz ^ dz y 3z
where ^is the volumetric water content, t is time, K is hydraulic conductivity, h is the
water pressure head, and z is elevation.
The initial and boundary conditions with respect to water content for this study
are given as:
Kz, t ) = h ,{z ) t=t„ (3.5)
32
-K
-K
dz
dz
= -1.85 cm/d r= l l J
= 0
r=113
0 < r < 23 days
/> 23 days
t > 0 days
(3.6)
(3.7)
(3.8)
where Equation (3.5) is the initial distribution of pressure head, hi is a prescribed
function of z, to is the time the simulation begins, and z represents the position in the
soil profile (i.e. z = 11.5 m at the soil surface, z = 0 m at the bottom of the profile); a
water table condition (perched water table) is imposed at the lower boundary. The 11.5
m profile was discretized using a grid spacing of 0.10 m, resulting in a total of 116
vertical nodes.
The soil-hydraulic functions of van Genuchten [1980] were used for the cases
not employing hysteresis and are given by:
/ ,<0 \m (3.9)
h > 0
0 = ( '+M")
0
K{h) = K^S°^ l - ( l -5 j "" ) ' " J (3.10)
(3.11)
where 6r and 6s are the residual and saturated water content, h is the pressure head, K is
the unsaturated hydraulic conductivity, and Se is the effective saturation. Equations
33
(3.9) and (3.10) are subject to the /w=l—(!/«) condition (« >1) [van Genuchten,
1980].
HYDRUS incorporates hysteresis into the retention function using an empirical
model developed by Scott et al. [1983] and modified by Kool and Parker [1987] to
account for air entrapment. The air entrapment effects incorporated into the model were
developed by Land [1968] and given by Aziz and Settari [1979]. Hysteresis is
incorporated into the hydraulic conductivity function using a model analogous to that of
Kool and Parker developed by Vogel et al. [1996] in which they apply an approach
similar to that used in the hysteretic retention function. Specifying hysteresis in the
retention and hydraulic conductivity functions requires that the user define both the
main wetting and drying curve. In addition, the initial conditions must be specified as
falling on either the main wetting or drying curve. Hysteresis is incorporated by scaling
scanning curves from the main wetting or drying curve. The wetting scanning curves
are scaled from the main wetting curve and drying scanning curves are scaled from the
main drying curve. The following discussion outlines the methodology of the hysteretic
model as outlined by Kool and Parker [1987]. The approach used to implement
hysteresis in HYDRUS is slightly different, but inspection reveals that both approaches
yield identical results.
Assumptions of the empirical model include the restrictions:
n ' '=n '^=n (3 .12)
where 0r is the residual water content, tar and n are empirical fitting parameters, the
superscripts w and d refer to wetting and drying, respectively. The main wetting and
34
drying curves are defined in the HYDRUS code by sending the parameter vectors
, 9^, d", /i) and to the following modified version of the van
Genuchten expression given by:
/? —f) 0,+ " ^ h<h^
f f (h ) = \ ( l + M") (3 .13)
_0. h>h.
Wetting scaiming curves are obtained by scaling the wetting curve to pass through a
reversal point. The scaled wetting curve is calculated by using the parameter vector
, in (3.13). The parameter ^ replaces 0^ in the parameter vector and
in effect forces the scanning wetting curve to pass through the reversal point. This
scaling parameter, ^, is defined as:
. 0. -O,S:(hA
where 0^ is the water content at the reversal point and is the effective
saturation on the main wetting curve at the reversal pressure head, and is given by
(3.11). By analogy, drying scanning curves are calculated using the parameter vector
{^0^, d],a'', ri^) where is calculated to adjust the main drying curve to pass through
the reversal point and is given by:
. 0^-0[ \ -SUhA\
35
Air entrapment effects are accounted for in the model by allowing the saturated water
content to be defined in terms of a wetting or drying saturated water content, and
ffl; this results in a hysteresis loop that is not closed at saturation. In general, studies
have found that < 0^ [Hopmans et al., 1987]. Air entrapment during the re wetting
phase can be approximated by the following empirical equation:
and 0^ . Equations (3.16) - (3.17) are used with (3.14) to predict the scanning
wetting curves. When the regime shifts from drying to wetting equations (3.16) - (3.17)
are used to calculate 9^ which is then used in place of 6^ in solving equations (3.13)
and (3.14). This substitution can also be used in equation (3.15) if another reversal
occurs.
HYDRUS uses an analogous procedure for applying hysteresis to the hydraulic
conductivity function given by (3.10). The same parameter vectors are used as in the
retention functions in addition to the requirement for and K'l to be specified.
Hydraulic properties were estimated using 11 core samples. The 11 core
samples were collected from an north south transect through the 50 m by 50 m plot at a
depth of approximately 1.5 m (Figure 3.1). Table 3.1 lists the mean, standard deviation.
(3.16)
where
(3.17)
36
Covered area Field PIotF-II5
iiii^aicu cuca •— —Depth • 1.49 m
• Depth = .49 m
• Depth = .45 m
• Depth = .38 m
• Depth = .50 m
• Depth = .61 m
NMI • Depth = .50 m
O o
• Depth = .49 m
SMI • Depth = .74 m
• Depth = .41 m
—•— Depth " :45-nt
-L J- X
O Monitoring island
• Location of soil core
h 10 meters
Figure 3.1. Locations where soil cores were collected.
37
Table 3.1. Mean Hydraulic Properties and the statistical parameters for the core samples obtained by the various laboratory methods
dr (cm^ cm "^) 4(cm^cm'^) ar(cm'') n Ks (cm d ') Y Saturated Hydraulic Conductivity Experiment
Mean 0.301 45.8 S.D. 0.034 23.3 Max 0.353 80.8 Min 0.262 16.6
CV (%) I I .4 50.9 Multi-step Outflow Experiment
Mean 0.00035 0.0425 1.312 S.D. 0.00041 0.0217 0.062 Max 0.00099 0.0788 1.145 Min 0.00003 0.0136 1.223
CV (%) 116.8 51.0 4.7 Wetting Experiment (upward infiltration)
Mean 0.0610 0.0984 1.33 -0.489 S.D. 0.0513 0.0821 0.20 1.253 Max 0.1420 0.2920 1.74 0.916 Min 0.0000 0.0408 1.11 -3.44
CV (%) 84.1 83.4 14.9 -256.2 Drying Experiment (evaporation)
Mean 0.0942 0.0534 1.746 -1.35 S.D. 0.0599 0.0665 0.376 1.58 Max 0.2030 0.2360 2.420 0.002 Min 0.0000 0.0097 1.220 -4.57
CV (%) 63.5 124.4 21.500 -116.8
38
maximum and minimum, and coefficient of variation for the hydraulic parameters for
the 11 core samples. The following outline of experimental work conducted on the core
samples and the estimation of the hydraulic parameters is summarized from
unpublished work conducted by John Fleming (1999). The intact soil cores collected
were 7.62 cm in length and 7.62 cm in diameter. The following experiments were used
to estimate the hydraulic properties: saturated hydraulic conductivity (Ks); multi-step
outflow (MSO); upward infiltration (UI); and evaporation (EV). After the completion
of the experiments, the bulk density and particle density (pipette method ) of the cores
were determined. The experiments were conducted in such a way that the ending
conditions for one experiment coincided with the required initial conditions for the
following experiment. The saturated water content was also experimentally determined.
The remaining parameters, a, n, and 0r, and y were optimized by using the inverse
procedure of the numerical code HYDRUS 1-D {Simunek et al., 1998] utilizing data
from aforementioned experiments. The optimizations in HYDRUS 1-D were also
conducted by Mr. Fleming. Efforts were made to characterize these hydraulic
properties using field-measured data with mixed results. Some of these approaches are
outlined in Appendix A.
The hydraulic conductivities calculated using Equation (3.10) and the data in
Table 3.1, were compared to field measured hydraulic conductivity values. This
comparison allows the predictive hydraulic conductivity relationship to be scaled to the
measured values. The concept is a variation of the scaled-predictive method introduced
by Mace et al. [1998]. In the scaled-predictive method the predictive K(0) relationship
39
is scaled using a single measurement of unsaturated hydraulic conductivity that replaces
the saturated hydraulic conductivity as the matching point. The approach utilized here
makes use of several unsaturated hydraulic conductivity measurements to scale the
predictive unsaturated hydraulic conductivity relationship. The methodology used to
calculate the field measured unsaturated hydraulic conductivities involved computing
an average pore-water velocity determined from the observed tensiometer data and
combining this with the observed mean water content at some depth to derive a flux
value. The tensiometer data measured at the monitoring islands (Figure 2.2) at depth
intervals of 0.5 m to a depth of 3 m was used to calculate the fleld measured unsaturated
hydraulic conductivities. Making a unit gradient assumption, supported by the observed
tensiometer data , the subsequent unsaturated hydraulic conductivity values can be
calculated and compared to those generated using Mualem's [1976] model based on the
laboratory derived hydraulic properties. Figure 3.2 shows a representative soil tension
profile from the monitoring island locations. After approximately 5 days of irrigation, a
steady-state unit gradient profile is attained. Figure 3.3 shows the hydraulic
conductivity as a function of water content based on the soil hydraulic properties model
of Mualem [1976] as well as calculated hydraulic conductivities from the field-
measured data from 0.5 to 3 m in 0.5 m increments. Table 3.2 shows the calculated
unsaturated hydraulic conductivity values as well as the values used to arrive at the
calculated values. As seen, when the saturated hydraulic conductivity is raised to 450
cm d ', the hydraulic conductivity curve passes through the field calculated values,
while the curve based on a laboratory derived hydraulic conductivity value of
40
Tension (cm H,0)
0 200 400 600 800
Initial conditions (early time)| Steady-state (later time) - m O O®
T^^nJ ! I ~ 5 days
100
E
a. u Q
200 r
300
Figure 3.2. Representative soil tension profile for the monitoring island locations.
41
>
o •o o u
3 CO L. -a
50
40
30 -a E u
S ^20 T3 O
« U3 C
10
0
0.32 0.28 0.24 0.2 0.16 0.12 0.08
Water content (cm^ cm"^)
Figure 3.3. Hydraulic conductivity versus water content. Solid circles are the field-calculated values from 0.5 to 3 m in 0.5 m intervals; the solid line is based on a laboratory-derived value of 45.8 cm d ' for Ks, the solid line with open circles is based on a Ks of 450 cm d"'.
42
Table 3.2. Calculated unsaturated hydraulic conductivity values for the NMI and SMI locations.
Depth (cm)
Location WF arrival'
(day) Velocity
(cm / day) Water
content^ K(0)^
(cm / day) 50 SMI West 4.0 12.50 0.26 3.25
East 4.0 12.50 0.26 3.25 NMI West 4.0 12.50 0.26 3.25
East 3.0 16.67 0.26 4.33 100 SMI West 7.5 13.33 0.25 3.33
East 6.5 15.38 0.25 3.85 NMI West 5.5 18.18 0.25 4.55
East 4.5 22.22 0.25 5.56 150 SMI West 9.5 15.79 0.24 3.71
East 9.0 16.67 0.24 3.92 NMI West 6.0 25.00 0.24 5.88
East 7.5 20.00 0.24 4.70 200 SMI West 11.0 18.18 0.23 4.18
East 11.5 17.39 0.23 4.00 NMI West 8.0 25.00 0.23 5.75
East 9.0 22.22 0.23 5.11 250 SMI West 13.0 19.23 0.21 4.04
East 13.0 19.23 0.21 4.04 NMI West lO.O 25.00 0.21 5.25
East 11.5 21.74 0.21 4.57 300 SMI West 14.0 21.43 0.21 4.46
East 14.0 21.43 0.21 4.46 NMI West 11.0 27.27 0.21 5.67
East 12.0 25.00 0.21 5.20 ' Determined from measure tensiometer data ^ Average water content at that depth behind the wetting front ^ Computed using a unit gradient assumption
43
45.8 cm d ' under-predicts the field calculated values by 72 to 98 percent (treating the
field calculated values as the known values). The same procedure was followed when
applicable for the other cases where different soil hydraulic properties were used (fi-om
the laboratory derived hydraulic properties).
The initial distribution of pressure head for the model was obtained from the
observed initial averaged water contents (Figure 3.4) given by the neutron probe data,
and the water retention model Equation (3.9).
Seven simulation tests, labeled Cases I through VII (Table 3.3), were conducted
using different soil hydraulic properties. The same initial and boundary conditions were
used for Experiment I (Cases I - V); different initial conditions were used for
Experiment II (Cases VI - VII). Hysteresis was incorporated into the models that use
laboratory derived hydraulic properties by using a combination of the properties
estimated from the wetting and drying experiments.
The interpolated initial conditions (water contents) as calculated by the model
and the observed water contents are shown in Figure 3.4.
44
0 0.1 Water Content (cm^ cm'^)
0.2 0.3 0.4 0.5
Figure 3.4. Model initial conditions and observed initial conditions. The circles represent the mean observed water contents. The horizontal lines represent ± one standard deviation and the solid line represents the interpolated water contents used in the model (HYDRUS).
45
Table 3.3. Summary of mean hydraulic properties used in models
Case dr (cm^ cm'^) 0s (cm^ cm'^) ar(cm"') n Ks (cm d"') y
I Experiment I, drying properties 0.06 0.301 0.053 1.75 45.8 -1.35
II Experiment I, wetting properties 0.06 0.301 0.098 1.33 45.8 -0.489
III Experiment I, drying and wetting properties 0.06 0.301 0^ = 0.098 1.33
ct = 0.053 45.8 -0.489
IV Experiment I, HYDRUS database properties with no hysteresis 0.065 0.410 0.075 1.89 106.1 0.5
V Experiment I, HYDRUS database properties with hysteresis 0.065 0.410 q>^ = 0.15 1.89 106.1
£/ = 0.075 0.5
VI Experiment II, wetting properties 0.06 0.301 = 0.098 1.33 45.8 -0.489
VII Experiment II, drying and wetting properties 0.06 0.301 a^ = 0.098 1.33
= 0.053 45.8 -0.489
46
3.3. Results and Discussion
3.3.1. Textural Characterization
The mean percent (from three locations) sand, silt, and clay with depth is shown
in Figure 3.5. Based on the data the soil is classified as a sandy loam. Figure 3.6 shows
the mean percent (from the same three locations) gravel, and the mean (from nine
locations) observed water content versus depth. Collectively, the information was used
to delineate six approximate boundaries at depths of 3 m, 6 m, 7.5 m, 9 m, 10 to 11 m,
and 12 m.
3.3.2. Water Budget
The computed depths of water added for each of the neutron probe access tube
locations were calculated for the 0 to 1.0 m depth interval on irrigation days 4 and 5,
and for the 0 to 9.0 m depth interval on irrigation day 19. The computed depths of
water application at each access tube were then compared to the depth of water applied
through the drip system. The magnitude of the error associated with the water budget
revealed a problem with the neutron probe calibration. A small change in the slope of
the neutron probe calibration equation, consistent with data from a small plot adjacent
to the large plot, improved results for the water budget. Plotting the computed depths of
the water added against the actual amount of water applied revealed a strong two-
dimensional flow component at the perimeter access tubes locations.
47
400
c u 800
o. u Q
1200
1600 20 10 60 70 80 90 0 10 20 0
Sand Silt Clay
Figure 3.5. Mean percent sand, silt, and clay composition at three locations.
48
0 0
20
Percent gravel
40 60 80 100
400
800 a. u Q
1200
1600
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Water content (cm^ cm'^)
Figure 3.6. Mean percent gravel (circles) and mean measured water content (solid line) versus depth.
49
Figure 3.7 shows the spatial distribution for the water budget at the 1.0 m depth
on irrigation day 4, and is based on data from 26 neutron probe access tubes. On
average, the neutron probe data over-estimated how much water was actually applied,
as measured by the water flow meters, by 28.4 percent. While the mass balance
calculation yielded acceptable results with respect to the mean computed depths of
water added, the standard deviation was quite large, 33.7 percent. Note also that the
neutron probe access tubes are located at least 5 m within the irrigated area.
Figure 3.8 shows the spatial distribution for the water budget at the 1.0 m depth
on irrigation day 5, and is based on data from 26 neutron probe access tubes. On
average, the neutron probe data under-estimated how much water was actually applied,
as measured by the water flow meters, by 17.6 percent. While the mass balance
calculation yielded acceptable results with respect to the mean computed depths of
water added, the standard deviation was quite large, 32.9 percent. Table 3.4 more
clearly shows the spatial variability of the computed depths of water added for each of
the neutron probe access tube locations.
Figure 3.9 is a plot of the spatial distribution for the water budget to the 9.0 m
depth on irrigation day 18, based on data from 9 neutron probe access tubes. On
average, the neutron probe data over-estimated how much water was actually applied,
as compared to that measured by the flow meters, by 0.2 percent. The standard
deviation of the error was 34.0 percent.
50
Over-Estimates (%)
Under-Estimates (%)
Figure 3.7. Spatial distribution of the water budget data for irrigation day 4, 1.0 m depth. Diamonds represent vertical neutron probe access tube locations. The numbers represent under- or over-estimation of water applied (in percent) relative to the water meter data. On average the neutron probe data over- predicted how much was actually applied by 28.4 percent.
51
Over-Estimates (%)
Under-Estimates (%)
Figure 3.8. Spatial distribution of the water budget data for irrigation day 5, 1.0 m depth. Diamonds represent vertical neutron probe access tube locations. The numbers represent under- or over-estimation of water applied (in percent) relative to the water meter data. On average the neutron probe data under- predicted how much was actually applied by 17.6 percent.
Table 3.4. Computed depths of water added on irrigation day 5 at 1.0 m
Location x (m) y (m) Total W (cm) % Difference*
402 10 5 7.61 -17.83 403 23 5 9.22 -0.45 404 30 5 6.17 -33.31 405 40 5 3.67 -60.42 412 5 15 8.96 -3.23 413 15 15 10.57 14.17 414 22.5 15 10.98 18.62 415 35 15 4.06 -56.16 416 45 15 6.63 -28.35 422 10 25 1.30 -85.94 423 23 25 8.86 -4.26 424 30 25 10.16 9.74 425 40 25 4.84 -47.78
432 5 35 7.93 -14.36 433 15 35 8.34 -9.92 434 22.5 35 5.26 -43.21 435 35 35 6.54 -29.32 436 45 35 5.46 -41.06 442 10 45 8.79 -5.11 443 22.5 45 9.71 4.90 444 30 45 8.30 -10.31 445 40 45 5.95 -35.74 454 22.5 10 8.76 -5.40
455 22.5 20 8.48 -8.42 456 22.5 30 4.80 -48.14 457 22.5 40 17.11 84.75
Mean 7.63 -17.6
Std. Dev. 3.05 32.9 Compared to 7.31 cm, as measured by the water flow meters
53
Over-Estimates (%)
Under-Estimates (%)
Figure 3.9. Spatial distribution of the water budget data for irrigation day 18, 9.0 m depth. Diamonds represent vertical neutron probe access tube locations. The numbers represent under- or over-estimation of water applied (in percent) relative to the water meter data. On average the neutron probe data over- predicted how much was actually applied by 0.2 percent.
54
The data in Figure 3.7 and in Figure 3.9 show considerable spatial variation and
there are no clear trends in the two figures. This would indicate that the water is applied
in a uniform manner. The computed depths of water added for the neutron probe access
tube location 422 (x = 10 m, y = 25 m) under-estimates how much water was actually
applied, as measured by the water flow meters, on both irrigation days. However, this
is not interpreted as a bias in the water application system as this under-estimation
would be expected to show up on Figure 3.9. This under-estimation is more likely
attributable to a disturbed soil profile in the upper 0.5 m due to plot preparation. There
does appear to be a spatial trend in Figure 3.9. The east side shows an under-estimate
of the amount of water applied, while the west side displays an over-estimate of the
amount of water applied. Also, for a field experiment of this magnitude, a difference of
28.4 percent for irrigation day 4, the 0 to 1.0 m depth, 17.6 percent for irrigation day 5,
the 0 to I.O m depth data (26 access tubes both days) and 0.23 percent (9 access tubes)
for the 0 to 9 m depth data are quite acceptable.
Because of lateral flow from the wet plot to the drier surrounding soil, the access
tubes at the perimeter under-estimated the amount of water applied by 50 to 60 percent,
a much larger error than predicted using the interior access tubes. Plotting the
computed depths of the water added versus the actual amounts of water applied
revealed a strong two-dimensional flow component at the perimeter access tube
locations (Figure 3.10 and Figure 3.11). Figure 3.10 shows all the neutron probe access
tube locations (34 locations, including the locations at the perimeter) through the
irrigation phase of the experiment. The solid circles represent locations at least 5 m
55
CJ o o Urn c. o C/3
-T3 'S &0 o
1 1 ' i ! i 1 1
•
' H ^ V ^ Over-predicts « \m ^ ^
•
0
Under-predicts
0 10 20 30 40
Water applied (cm)
50 60
Figure 3.10. Computed water depths applied versus the water depths applied according to the water meters for 34 access tube locations on irrigation days when data are available. The dashed line represents +/- 32 percent error from the true amount applied.
56
Over-predicts
Under-predicts
Water applied (cm)
Figure 3.11. Computed water depths applied versus the water depths applied according to the water meters for the perimeter access tube locations (8 locations). The dashed line represents +/- 32 percent error from the true amount applied.
57
within the irrigated perimeter; the open squares represent the perimeter locations. As
can be seen, the access tubes located on the perimeter under-estimate the computed
depths of the water applied to a much larger degree than the access tubes locations 5 m
within the plot. Figure 3.11 shows only the access tubes located on the perimeter (8
locations) and more clearly shows the trend of under-estimation. Figure 3.12 shows
only the neutron probe access tube locations at least 5 m within the irrigated perimeter;
the majority of these locations fall within the error bars of one standard deviation. The
solid line on both figures represents the values where the amount applied as measured
by the water meters equals the computed depth of water applied (0 percent error), the
dashed lines represent +/- 32 percent of this value. The range of +/- 32 percent
represents the mean value of the standard deviation from all the irrigation days. The 32
percent was calculated by first, determining the average depth of water applied (based
on data available from 26 access tubes) for each irrigation day, then calculating the
associated standard deviation of the access tubes for each irrigation day. The
percentage the standard deviation represented of the mean depth of water applied was
calculated for each irrigation day, and finally, a mean of all the computed percentages
was calculated. This approach was used because of the large degree of variation in the
standard deviation. It should be noted that for both figures not all locations possess data
for each irrigation day.
58
Over-predicts
Under-predicts
0 10 20 30 40 50 60
Water applied (cm)
Figure 3.12. Computed water depths applied versus the water depths applied according to the water meters for the access tube locations located at least 5 m within the irrigated perimeter (26 locations). The dashed line represents +/- 32 percent error from the true amount applied.
59
3.3.3. Simulation of Cases I and II
The "goodness of fit" for the models (Cases I - VII, Table 3.3), were evaluated
using three criteria; the ability of the model to predict the arrival time of the wetting
front, and the ability of the model to accurately predict the water contents during the
irrigation and redistribution phases of the experiment. The ability of the model to
predict the arrival time of the wetting front was evaluated by comparing the simulated
arrival time to the observed arrival time using the tensiometer data. The tensiometers
are very responsive to the changes in water content. In addition, tensiometer
measurements were collected at six-hour time intervals, thus the tensiometer data
provided a more accurate measure of the wetting front arrival time as compared to the
neutron probe data and was one of the criteria used to evaluate the ability of the model
to predict the arrival time of the wetting front. The mean square errors (MSE) for the
observed versus predicted water contents for a particular depth was the method used to
evaluate the ability of the model to accurately predict the water contents during the
irrigation and redistribution phases of the experiment. The MSE for a particular depth
was computed according to:
2
MSE = -"^(0,-0) (3.18)
w h e r e ^ ^ , a r e t h e m e a s u r e d a n d p r e d i c t e d v o l u m e t r i c w a t e r c o n t e n t s r e s p e c t i v e l y , f o r
a particular depth, and / represents the time interval. Table 3.5 lists the MSE values
(multiplied by 1000 to better illustrate the differences) for the particular cases at
specified depths.
60
Table 3.5. Mean square error* for Cases I - VII
Depth (m) Ossc
1.5 3.0 5.0 7.5 I 3.82 1.62 2.62 1.58 II 0.68 0.22 0.40 0.14 III 0.51 0.19 0.21 0.10 IV 1.61 0.50 0.88 0.40 V 1.96 0.72 1.48 0.67 VI 0.55 0.36 0.47 0.18 VII 0.32 0.34 0.40 0.30
Note that the MSE was multiplied by 1000
61
Figure 3.13 presents the measured (circles with vertical bars) and the simulated
water contents versus time for the 7.5 m depth (Cases I and II). The simulated wetting
front arrived at approximately 24 days at the 7.5 m depth. The hydraulic properties
obtained from the drying cores clearly result in underestimation of the soil water
content at this depth (Figure 3.13, A). However, the arrival of the wetting front is
reasonably well predicted using the properties from the wetting experiments. An
increase in the saturated hydraulic conductivity to 75 cm day"' makes the prediction of
the water content during the redistribution phase worse (Figure 3.13, B).
Using the hydraulic properties from the wetting cores (Case II, Table 3.3) with a
measured saturated hydraulic conductivity of 45.8 cm day"' (Table 3.1), caused the
simulated wetting front to lag approximately 33 days behind the measured wetting front
(Figure 3.13, C). A 10-fold increase (450 cm day"') in the saturated hydraulic
conductivity in accordance with the data in Figure 3.3, greatly improved the wetting
front arrival time prediction, and also resulted in a good fit during the drainage phase
(Figure 3.13, D). These same results could not be obtained with the properties from the
drying experiments unless major adjustments were made to a, n, and y. Thus the best
results for the 7.5 m depth were obtained with wetting properties and a Ks of450 cm
day"'. Table 3.6 summarizes the adjustments made for Cases I and II as well as
subsequent cases (Cases III - VII).
62
c u c o U L. o "S
c c o u a>
- A, Drying, = 46 cm day
_ I -
-1 B, Drying, K^ = 15 cm day 1
0.16 -
0.12 r
r C, Wetting, AT- = 46 cm day"' D, Wetting, = 450 cm day"' -! 0.24
0.2 r
0.16
0.12
0 20 40 60 80 0 20 40 60 80 100
Time (Days) Time (Days)
Figure 3.13. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 7.5 m depth. The vertical lines denote ± one standard deviation in the mean water content.
63
Table 3.6. Saturated hydraulic conductivity values used for Cases I - VII
Case Description Ks (cm d'^) Adjusted Ks (cm d"')
I Experiment I, drying properties 46.0 75.0
II Experiment I, wetting properties 46.0 450.0
III Experiment I, drying and wetting properties 46.0 250.0
IV Experiment I, HYDRUS database properties with 106.1
no hysteresis 450.0
V Experiment I, HYDRUS database properties with hysteresis 106.1 1250.0
VI Experiment II, drying and wetting properties 46.0 250.0
VII Experiment II, drying and wetting properties 46.0 250.0
64
3.3.4. Comparison of Cases II and III: Incorporating Hysteresis
Hysteresis was incorporated into the soil hydraulic model incrementally (first in
the water retention function, and then in the water retention and hydraulic conductivity
functions). Incorporating hysteresis into just the water retention function decreased the
accuracy of the predicted water contents (results not shown). Figure 3.14 compares the
simulated water content time series with hysteresis included (Case III, Table 3.3), and
the simulated water content time series with no hysteresis (using hydraulic parameters
from the wetting experiment. Case II) with the mean measured water content time
series. Adjustment of the saturated hydraulic conductivity from 46.0 cm day"' to 250.0
cm day ' for the model incorporating hysteresis (Case III) yielded much better
predictions with respect to the wetting fi-ont arrival time as well as a better fit during the
drainage phase, in comparison with using a Ks of 46.0 cm d"'.
The predictions based on the model with no hysteresis (Case II) match closely to
the predictions of the model incorporating hysteresis (Case III) although, as indicated
by the MSE (Table 3.5) the model incorporating hysteresis (Case III) yielded better
results at every depth (Figure 3.15 to Figure 3.17), including the shallower depths. This
may be attributed to ±e fact that the retention curve from the model with no hysteresis
(Case II) falls on the main wetting curve of the model incorporating hysteresis (Case
III). Figure 3.18 illustrates this idea; the main wetting and drying curves based on the
parameters used in the hysteretic model (Case III) are plotted along with the retention
curves based on the drying and wetting parameters with no hysteresis (Cases I and II,
respectively).
u « 0.14 ^ "S I
^ 0 1 2 - " " ' C a s e D , K s = 4 5 0 . 0 c m / d j Case in, Ks = 250.0 cm / d
0 . 1 ^ ^ ^ ^
-20 0 20 40 60 80 100
Time (day)
Figure 3.14. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 7.5 m depth. The vertical lines denote ± one standard deviation in the mean water content.
66
0.3
E O n E CJ 0.2 t.
C o o
CJ
"S
i } f
0.1 -
0
0
- 400
- 300 ^ wJm - E
Case n, Ks = 450.0 cm / d Case in, Ks = 250.0 cm / d
200
T 100
0
20 40 60 80 100
c _o c ti H
Time (day)
Figure 3.15. Comparison of the simulated water content time series to the mean measured water contents for the 1.5 m depth. Triangles depict tension.
67
0.3 -
0.2
c CJ o
U ba o
0.1 T
0
0
- 300
- 200
^ 100
- - Case n, Ks = 450.0 cm / d —Case in, Ks = 250.0 cm / d
0
20 40 60
Time (day)
80 100
O r4 X E u c o c u
Figure 3.16. Comparison of the simulated water content time series to the mean measured water contents for the 3.0 m depth. Triangles depict tension.
68
0.3 - ^ 400
0
- 300
- 200
^ 100
- - Case n, 450.0 cm / d Case in, 250.0 cm / d
0
O <N
s u e ,o 'cn C u
20 40 60 Time (day)
80 100
Figure 3.17. Comparison of the simulated water content time series to the mean measured water contents for the 5.0 m depth. Triangles depict tension.
69
0.4
i O . 3
I 0-2 O
O "S
Main drying curve. Case HI ™ Main wetting curve. Case III, main wetting and drying cu^eTCase Main wetting and drying curve. Case I
0.1 10 100 1000 10000 100000 1
Suction (cm of H2O)
Figure 3.18. Retention curves for hysteretic model and models based on parameters derived from wetting experiments and drying experiments.
70
The retention curve for the model using parameters taken from the wetting
experiment falls on the main wetting curve of the hysteretic model (as expected since
the parameters are the same) while the retention curve based on the parameters from the
drying experiments differs significantly from the others. This is because of the use of a
different shape factor, n, as well as a and y. With models developed using the drying
parameters (Case I), I was unable to achieve a satisfactory calibration without large
adjustments to the parameters. This would indicate that for these field conditions,
where the profile stays fairly wet during most of the experiment, the wetting parameters
yield better results. In addition, the similarity of the results between Cases II and III
would indicate that hysteresis does not play a large role in determining the wetting front
arrival and the shape of the water content curves. However, it should be noted that
incorporation of hysteresis did result in the use of a more realistic value for the
saturated hydraulic conductivity as compared to the case without hysteresis (Case II).
The suitability of using a homogeneous representation for this site was evaluated
by comparing the simulated results with data from shallower depths (1.5, 3.0, and 5.0
m). Figure 3.15 through Figure 3.17 compare the simulated water content time series
for the 1.5, 3.0 and 5.0 m depths to the observed water content and tension time series.
The open triangles are the mean measured tension time series from a representative
location within the experimental plot. Because of noise inherent in the water content
data, the observed tensiometer response was also used to delineate the arrival of the
wetting front and the beginning of the drainage phase for the shallower depths.
Tensiometers are very responsive to changes in water content and provide a more
71
distinct delineation of the wetting front arrival time as well as of the time that the
drainage phase begins. The plots (Figure 3.15 - Figure 3.17) clearly illustrate how the
measured tensiometer response agrees with the predicted water contents with respect to
the arrival of the wetting front and the start of the drainage phase. In addition, the plots,
as well as the MSE, support the use of a homogeneous representation for this field site.
3.3.5. Simulation of Cases IV and V
The uniqueness of the laboratory-derived parameters was also evaluated by
comparing the results from Cases II and III, against a model using hydraulic parameters
taken from an internal database in HYDRUS 1-D [Simuneket al., 1998]. The hydraulic
parameters chosen from the internal database are based on textural considerations, as
stated previously; a sandy loam is the dominant texture throughout the profile at the
site. Table 3.3 lists the hydraulic parameters used as a first approximation from the
internal HYDRUS database; Case IV corresponds to the model with no hysteresis. Case
V incorporates hysteresis into the retention and hydraulic conductivity functions.
Calibration of the models using hydraulic parameters taken from the internal database
followed the same methodology used in calibrating the models based on the laboratory
derived hydraulic properties: saturated hydraulic conductivity was used as a fitting
parameter to adjust the arrival time of the wetting front and minor adjustments were
made to alpha (a) and n to adjust the shape of the curve. Using this methodology it was
not possible to fine-tune either of the two models using hydraulic properties taken from
72
the internal database (Cases IV and V) as a first approximation to the same degree of
quality as the models using the laboratory derived properties (Cases II and III). This is
supported by the MSE values (Table 3.5). Figure 3.19 shows the final results for the
four cases for the 7.5 m depth. Using the laboratory derived hydraulic parameters
clearly resulted in a superior fit as compared to using the internal database parameters
from HYDRUS (Figure 3.19). This implies that under these circumstances the
laboratory derived parameters provide a better first approximation for predicting the
wetting front arrival and the shape of the curves than parameters taken from the internal
database solely based on textural considerations.
3.3.6. Water Contents with Depth, Experiment I
The mean measured vertical water content profiles for the plot are shown for
days 22 (Figure 3.20), 24 (Figure 3.21), and 93 (Figure 3.22). Also shown are the
predicted water contents for the two models based on the laboratory derived hydraulic
parameters, with and without hysteresis. The plots for days 22 and 24 represent the day
before and the day after the end of the irrigation phase, respectively. Day 93 is the last
day neutron probe measurements were taken during the drainage phase of the first
experiment.
73
3 c o U
Urn CJ
c o c o U
WH o rt
0.28
0.24
0.2
0.16
_ Wetting & drying properties J Wetting properties _ 1 y - - K^ = 250.0 cm d' _J -^-Ar^ j= 450.0 cm d'
~ T d i i l i f i i : ^ : o ;—, ** 1 " . ~ j p o ^ ^ ^ ^
- ' i ' i J " L ^
r 1
— i
i i 1 i i i ; 1 ! 1 1 , I , ; , 1 : : n
Database properties, hysteresis J Database properties, no hysteresisj , r K = 1 2 5 0 . 0 c m d - ' -Kj = 450.0 cm d '
0 20 40 60 80
Time (Days)
0 20 40 60 80 100
Time (Days)
Figure 3.19. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 7.5 m depth. The vertical lines denote ± one standard deviation in the mean water content. A, Case III with Ks = 250.0 cm d ', B, Case II with Ks = 450.0 cm d ', C, Case V with Ks = 1250.0 cm d ', D, Case IV with Ks = 450.0 cm d-'.
74
Water Content (cm^ cm"^)
0 0.05 0.1 0.2 0.25 0.35 0.15 0.3 0
4
8
Case II, Ks = 450.0 cm / d i Case III, Ks = 250.0 cm / d |
12
Figure 3.20. Water content measurements and model predictions for day 22. The circles represent measured water contents.
75
o. o Q
0 0
8 —
12
0.05
Water Content (cm^ cm*^)
O.i 0.15 0.2 0.25 0.3 0.35
Case II, Ks = 450.0 cm / d"
Case III, Ks = 250.0 cm / d i
Figure 3.21. Water content measurements and model predictions for day 24. The circles represent measured water contents.
76
Water Content (cm^ cnr^)
0.15 0.2 0.25 0.35 0 0.05 0.1 0.3 0
4
8
Case n, Ks = 450.0 cm / d~j Case III, Ks = 250.0 cm / d I
Figure 3.22. Water content measurements and model predictions for day 93. The circles represent measured water contents.
77
The simulated profile based on the model incorporating full hysteresis (Case III)
provides the best overall results for this situation as compared to the model with no
hysteresis. These results are supported by similar findings by Pickens and Gillham
[1980] who found that including or neglecting hysteresis had the most pronounced
effect on the pressure head and water content profiles. With regard to the simplification
of using a homogenous profile model to describe a heterogeneous field profile, the
model appears to do an adequate job during the first 24 days of infiltration (flux
controlled irrigation) with the quality of the predictions decreasing as the water moves
deeper in the soil profile and drainage is dominant. During the drainage phase (and
without layering in the model) the simulated profile moves toward a more homogeneous
profile. The predictions from the hysteretic model (Case III) provide visually better
results to the observed profile during the infiltration phase than the model with no
hysteresis (Case II). The quality of these results may be due to the variability of the
initial conditions and the different retention characteristics between the two models, and
are therefore somewhat artificial. Given a different set of initial conditions the non-
hysteretic model may have produced superior results compared to the model
incorporating hysteresis. Clearly both models fail to accurately predict the shape of the
soil water profile as the late time drainage phase progresses. It should also be pointed
out that the model was adjusted with respect to the wetting front arrival and drainage
time series. Had the model been adjusted using soil water profiles instead of the
drainage time series better predictions for the soil water profile could have been
78
obtained. However, taking this approach would have necessitated the use of a multi-
layered model.
3.3.7. Simulation of Cases VI and VII, Experiment II
A post audit of the model was conducted to verify the quality of the predictions.
The post audit was carried out using data from the second infiltration experiment and
the hydraulic parameters from the two models (Cases II and III), one without hysteresis
(wetting parameters. Case VI) and the other employing hysteresis (Case VTI). Figure
3.23 shows the initial conditions for the second experiment; they are the final water
contents observed at the end of Experiment I. Figure 3.24 through Figure 3.26 show the
mean measured water content time series for the 1.5 m, 3.0 m and 7.5 m depths. As for
the Experiment I cases, the case with no hysteresis specified (using hydraulic
parameters based on the wetting experiment. Case VI) provided comparable results to
the case incorporating hysteresis in both the retention and hydraulic conductivity
functions (Case VII). The predicted water content time series for Experiment II at the
1.5 m depth (Figure 3.24) fit the observed data with a higher degree of accuracy than
the predictions from Experiment I (Figure 3.15) and provided comparable results to
Experiment I simulations at other depths (Table 3.5). The change in the tensiometer
readings during passage of the wetting front again clearly delineates the wetting front
arrival and the start of the drainage phase and coincides almost exactly with the
simulated results.
79
0.05 Water content (cm^ cm'^)
0.1 0.15 0.2 0.25 0.3
Figure 3.23. Model initial conditions and observed initial conditions for Experiment II. The circles represent the mean observed water contents. The horizontal lines represent ± one standard deviation and the solid line represents the interpolated water contents used in the model (HYDRUS).
80
-50
1 }
0 50
" Case VI, Ks = 450.0 cm / d
—Case Vn, Ks = 250.0 cm / d
160
140
+ 120 ^ O
100 = E
80 ^ §
60 g o
+ 40 ^
20
0
100 Time (day)
150 200 250
Figure 3.24. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 1.5 m depth. The vertical lines denote ± one standard deviation in the mean water content.
81
-50 0 50
120
100
80
- 60
x E o
40 I E-
Case VI, Ks = 450.0 cm / d -|- 20
Case Vn, Ks = 250.0 cm / d
100
Time (day)
150 200 250
Figure 3.25. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 3.0 m depth. The vertical lines denote ± one standard deviation in the mean water content.
82
0.2 <>
- - Case VI, Ks = 450.0 cm / d
Case Vn, Ks = 250.0 cm / d
0 50 100 150 Time (day)
200 250
Figure 3.26. Comparison of the mean measured and predicted water contents and wetting front arrivals at the 7.5 m depth. The vertical lines denote ± one standard deviation in the mean water content.
83
3.4. Conclusions
The arrival of the wetting front was simulated and compared to observed data
from two successive field infiltration events. Several models were developed using
hydraulic properties derived from drying and wetting experiments. These models were
then compared and fine-tuned if necessary, against data taken from the first infiltration
experiment. A post-audit of the models from Experiment I was conducted using data
from a second experiment. Using the hydraulic properties from Experiment I,
simulations were performed and the results compared to the measured data from
Experiment II to assess the quality of the predictions as well as the assumption
regarding using a homogeneous representation of the soil profile.
From this study it was determined that the saturated hydraulic conductivity, Ks,
had to be adjusted to predict the time of the wetting front arrival and the changes in
water content observed in the field during and after infiltration. A modified version of
the scaled-predictive method introduced by Mace et al. [1998] offered a valuable tool
for determining how much to adjust Ks to adjust the predictive hydraulic conductivity
relationship. Without increasing Ks tenfold from the value of 46 cm d"', determined in
the laboratory, to 450 cm d"', determined in the field from tensiometer observations at a
range of depths from 0.5 to 3 m from around the monitoring islands, acceptable
predictions for Experiments 1 and II could not be made. Thus Ks became an effective
saturated hydraulic conductivity, Ks.eff, which incorporates uncertainties not included in
the laboratory determined Ks. Simulations solely based on hydraulic properties
estimated in the laboratory with core samples as first approximations yielded poor
84
results with regards to estimating the arrival of the wetting front at various depths as
well as predicting the drainage phase of the experiment. Adjustments made to the
parameter, lead to reasonably good results for estimating the arrival of the wetting front
and changes in water content. The results also support the use of a homogeneous
vertical soil profile for this particular field site. The simulations utilizing properties
taken from wetting experiments (Case II) and the case incorporating hysteresis into the
retention and hydraulic conductivity functions (Case III) yielded the best results as
compared to the results from utilizing hydraulic properties taken from drying
experiments. Comparison of the results from Cases II and III indicate that hysteresis
was not a dominant mechanism during these experiments; this may be because the soil
profile remained in the "wet" region of the retention curve during most of the
experiment. Incorporation of hysteresis into the soil hydraulic functions, however,
yielded superior predictions of the soil water content profile as compared to predictions
without hysteresis and is supported by similar findings of Pickens and Gillham [1980].
Another factor affecting these results is that while Case III follows a separate main
drying curve from Case 11 (Figure 3.18), the maximum difference between the drying
curves for Cases II and III is approximately 7 % at 100 cm of suction; this would
indicate that a large difference should not be seen between the two cases. However,
incorporation of hysteresis did result in the use of a more realistic value for the
saturated hydraulic conductivity as compared to the case without hysteresis (Case II).
Utilizing hydraulic properties (based on soil texture) taken from an internal,
general database in HYDRUS (Cases FV and V) the question of whether or not the
85
laboratory derived hydraulic properties (Cases II and III) offer an advantage over the
database properties was addressed. Using saturated hydraulic conductivity as a fitting
parameter it was not possible to adjust either of the two models using hydraulic
properties taken from the internal database (Case IV and V) as a first approximation to
the same degree of quality as the models using the laboratory derived properties (Case
II and III). This is supported by the MSE values (Table 3.5). Based on these results,
the laboratory-derived parameters provide a superior first approximation for predicting
the wetting front arrival and the shape of the curves.
A post-audit of the fine-tuned models, using fine-tuned laboratory derived
hydraulic parameters, conducted by simulating a second irrigation event in the same
plot, yielded good results, with the MSE (Table 3.5) of the same order as the results
obtained from Cases II and III. In addition, it would appear that the assumption
regarding the use of a homogeneous soil profile is supported.
In conclusion, for this field site, and under these conditions, the use of a relatively
simple one-dimensional model is suitable for predicting processes such as wetting front
advance at the scale of tens of meters. The results also show that, although laboratory
derived hydraulic properties may be limited in their applicability to field settings, for
this field setting they clearly offered a superior first approximation as compared to
hydraulic properties based on textural considerations taken from a database.
86
4. FIELD SCALE SOLUTE TRANSPORT THROUGH VARIABLY SATURATED MEDL\ INCORPORATING HYSTERESIS
4.1. Introduction
Prediction of flow and transport through variably saturated porous media requires
the estimation of hydraulic and transport properties. Solution of the unsaturated flow
component, approximated by Richards' equation, requires that the unsaturated hydraulic
properties be estimated. In most cases the van Genuchten model [van Genuchten, 1980]
requires a minimum of five parameters: reduced water content (fir), saturated water
content (6^), the empirical parameters affecting the shape of the hydraulic functions (a,
m and n) and the saturated hydraulic conductivity (/Tj). The equilibriimi CDE, used to
predict the transport of solutes, requires that at least three parameters be defined: the
retardation factor (/?), the average pore-water velocity (v), and the dispersion coefficient
(Z)), a parameter that combines the effects of mechanical mixing and molecular or
fickian diffusion. These properties are often difficult to define with any confidence,
even for homogeneous laboratory columns. Further complications arise when
estimating these properties in field settings because of heterogeneous conditions and
experimental uncertainties associated with field scale experiments. This study
addresses the issues affecting the estimation of flow and transport at the field scale. The
effects of spatial averaging of point measurements to infer these properties are also
evaluated here by their subsequent use in analytical and numerical models. The effects
of incorporation of hysteretic relationships into the unsaturated flow model are
87
investigated through simulation of solute transport and comparison with field
measurements.
There is no clear consensus among investigators regarding the dependence of
transport properties on the flow conditions. Many investigations have observed a linear
increase of the dispersion coefficient with increasing depth for higher flow rates in
saturated media in field scale transport studies [Ellsworth et al., 1996; Jacques et al.,
1998; Vanderborght et al., 2000]. In contrast, Porro et al. [1993] found no scale
dependence of the dispersion coefficient in unsaturated layered soil proflles up to 6 m
deep where the layer thickness is small relative to the scale of observation. Jacques et
al. [1998] found deviations from the expected linear increase in dispersivity with depth
for steady-state chloride transport through two heterogeneous field soils. Finally,
Padilla et al. [1999] show a dependence of dispersivity on both the properties of the
media as well as the water content.
As with scale dependence of flow and transport properties, there is no generally
accepted approach to spatial averaging of point measurements for large-scale analyses.
For example, Porro et al. [1993] found significant differences in the estimation of
transport properties using data from individual locations (local approach) as opposed to
using the data to simultaneously fit data fi-om all depths to estimate the transport
properties (global approach). In this study, the local and global approaches for both the
flow and transport properties are compared and forward models generated using these
averaged properties are compared with field measurements. The effects of averaging
88
these properties are evaluated by using them in a simple analytical solution to the CDE
and a numerical model, HYDRUS 1-D [Simunek et al., 1998],
Hysteresis presents a further complication when estimating flow parameters for
variably saturated media. Pickens and Gillham [1980] examined the effects of
incorporating hysteretic relationships into the unsaturated hydraulic functions on the
pressure head, water content and concentration profiles for a hypothetical one-
dimensional case where a slug of water containing a non-reactive tracer is introduced
into a column of sand and allowed to infiltrate and redistribute. They concluded that the
effect is greatest on the pressure head and water content profiles with a negligible effect
on the concentration profile. This treatment is repeated in this study to evaluate the
effects of hysteresis on solute flow through a variably saturated heterogeneous media.
A slug of water is introduced containing a non-reactive tracer followed by a slug of
solute free water and allowed to redistribute.
Generally, the objective is to simulate the transport of a solute through variably
saturated media integrating the steps necessary to achieve this goal. It is the intention to
convey a unified overall approach to modeling the transport of a solute fi'om estimation
of the various properties required to model this process, the development of a
conceptual model to describe the system being modeled, and how this conceptual model
impacts transport predictions.
Specifically, the objectives of this paper are to; evaluate location specific and
spatially integrated inversion methods on the ability to characterize the transport
parameters, to investigate the effects of spatial averaging of the estimated transport
89
properties and the impact on transport predictions when used in transport models, and
finally to investigate the effects of hysteresis and the retardation coefficient on solute
transport predictions.
4.2. Transport Models
4.2.1. Parameter Estimation and Analytical Solutions to Solute Transport
Parameter estimation was accomplished by using CXTFIT 2.1 [Toride et al.,
1995] in the inverse mode with observed concentration data from the SMI and NMI
locations. These estimated parameters were then averaged and used as first
approximations in analytical and numerical models. The analytical model is based on
an analytical solution to the one-dimensional CDE [van Genuchten and Alves, 1982].
HYDRUS I-D [Simunek et al., 1998] was used to numerically simulate flow and
transport using the estimated parameters. The results of the analytical and numerical
models were then compared to observed data and to each other to optimize the transport
parameters. The transport parameters were modified by an iterative comparison of the
results from the analytical and numerical models to the measured data.
The retardation factor and the dispersion coefficient were estimated using
CXTFIT 2.1 [Toride et al., 1995]. This code uses a nonlinear least-squares parameter
optimization method to estimate the solute transport parameters using any of three one-
dimensional transport models: the deterministic equilibrium CDE, the deterministic
90
nonequilibrium CDE, and the stream tube model for field-scale transport. CXTFIT 2.1
numerically evaluates analytical solutions to these transport models.
The observed data were analyzed in terms of the conventional equilibrium CDE
to estimate the solute transport parameters and to simulate the transport of bromide and
is given by:
where v is the average pore-water velocity ( y = q / 0 ) (cm d ') and R is the retardation
factor (dimensionless). R is given by:
r = 1 + (4.2)
where pb is the soil bulk density (mg cm'^), and Kd is the distribution coefficient (cm^
mg ') relating solute adsorption on the solid phase to the solute concentration in the
liquid phase and was assumed to obey a linear isotherm equilibrium. No decay or
production terms were included for the tracer. For this particular system, the inverse
boundary-value problem was solved by minimizing an objective function describing the
tit of a mathematical solution of equation (4.1) to the observed concentration data
[Toride et al., 1995] for the R and D parameters.
A third-type boundary condition was used at the soil surface and a zero
concentration boundary was applied at the groundwater table. The initial and boundary
conditions for the solute transport models are given by:
c(z,0) = c,(z) = 0 /=/„ (4.3)
91
\ d z j
+ qc = 31.6 mg /1 r=lIJ
-l9D
dc dz
dc^ = 0
r=llJ
= 0
0 < / < 15 days
/ >15 days
t > 0 days
(4.4)
(4.5)
(4.6) r=0
where D is the dispersion coefficient (cm^ d"') and q is the fluid flux density (cm d ').
The plot was irrigated at a rate of 1.85 cm per day for a total of 24 days; for the first 15
days the irrigation water contained 31.6 mg /1 of bromide (average concentration).
The following solution of the deterministic equilibrium CDE, with no
production or decay terms [van Genuchten and Alves, 1982] is used to estimate the
solute transport properties and to simulate the transport of bromide:
\ c i + { c „ - c , ) A i z , t ) 0 < t < t o
) ^ ( 2 ' , 0 - A ( z , t - t j t > L (4.7)
with
A { z , t ) = 0 . 5 e / f c Rz-vt
2 (£>/?/)"
^ 2, V t
^tvDR J exp
{ R z - v t f
ADRt
-0.5 V
2 . \ , V Z V t 1 + — +
D DR exp
D erfc
J
Rz + vt
2 { D R t ) OJ
(4.8)
where c, is the initial solute concentration in the soil profile, Co is the input
concentration, and to represents the length of the solute pulse after which the input
concentration is 0. This solution is subject to the condition where the volumetric water
content and flux are constant in time and space (steady-state condition).
92
4.2.2. Numerical Modeling of Solute Transport
HYDRUS 1-D [Simunek et ai, 1998] was utilized to simulate the transport of
bromide through a variably saturated soil profile. Separate governing equations were
solved for the solute transport and water flow when solving coupled problems in
HYDRUS. HYDRUS couples Richards' equation with the CDE to solve problems of
flow and transport through variably saturated media. For this study, two previously
developed flow models (Section 3.3.4, Cases II and III) were coupled to the transport
model to simulate the transport of the solute, bromide. The first flow model does not
incorporate hysteresis into the soil hydraulic functions while the second does.
The governing solute transport equation, solved in HYDRUS 1-D for this
system, is given by:
d0Rc _ d f
d t 5 z V & dz (4.9)
The retardation factor, R, and dispersion coefficient, D, are given by:
R = (4.10) 0 (l + 7c^)
D = av^D^T^ (4.11)
where, (412)
where/is the fraction of exchange sites assumed to be in equilibrium with the solution,
ks, P, and rj are empirical constants, a is the dispersivity (cm), Dy, is the binary diffusion
coefficient in free water for a particular solute (cm^ d"'), and Tw is a tortuosity factor
93
developed by Millington and Quirk [1961]. Wben^ = 1, adsorption occurs according to
the Langmuir model, when»/ = 0, adsorption occurs according to the Freundlich model,
and when P = I and 7 = 0, adsorption occurs according to a linear adsorption isotherm.
Setting ks = 0 specifies no adsorption (/? = 1). The values of water content, 0, and the
flux density, q, are required to solve this equation and are determined by solving a
mixed form of Richards' equation. Note that R cannot be fixed to a constant value
when solving coupled systems since ^is solved by Richards' equation and is a function
of the soil hydraulic functions.
The initial concentration throughout the 11.5 m profile was specified as 0 mg /
cm^. The initial water contents were calculated for each modeled depth by linear
interpolation of the observed water contents (Figure 4.1). The initial and boundary
conditions for solute transport are given by Equations (4.3) - (4.6).
The initial distribution of pressure head was calculated from the average water
content measured with the neutron probe at each depth (Figure 4.1) using the water
retention model (3.9).
94
0
1200
-0.5
0.1 Water Content (cm^ cm'^)
0.2 0.3 0.4
-0.25 0 Concentration (mg cm"^)
0.25
0.5
0.5
Figure 4.1. Model initial conditions and observed initial conditions. The circles represent the mean observed water contents. The horizontal lines represent ± one standard deviation and the solid line represents the interpolated water contents used in the model (HYDRUS). The dashed line represents the initial conditions for the bromide used in the model (HVDRUS).
95
4.3. Measured Bromide Concentrations
The bromide concentrations used were measured at the 50 - 300 cm depths (50 cm
intervals) at the NMI and SMI locations (Figure 4.2 - Figure 4.5). Only the SMI, west
side experienced breakthrough of bromide at all depths within the irrigation time period,
with the bromide breaking through at approximately 19 days at the 300 cm depth (Table
4.1). At the 300 cm depth, bromide breakthrough occurred at 29 days for the NMI, east
side, and 37 days for the SMI, east side locations. The NMI, west side location
exhibited a very dispersed breakthrough at 200 cm, with this behavior continuing at the
250 and 300 cm depths (relative concentration below 0.5). The SMI, east side location
also demonstrated some dispersive behavior at a depth of 200 cm and below with the
bromide breakthrough curves flattening somewhat as compared to the breakthrough
curves at the shallower depths. The NMI, east side and SMI, west side locations
experienced very strong bromide breakthrough even at the 300 cm depth with the
breakthrough curves exhibiting sharp bromide interfaces with both locations
experiencing greater than 80 percent of the bromide breaking through within 4-10
days. Table 4.1 lists the breakthrough times of the bromide at the monitoring island
locations. This breakthrough time is defined as when relative concentration of the
bromide reaches 0.5 or larger. Locations where relative concentrations exceeded one
can be explained in part by the fact that an average input concentration of 31.6 mg /1
was used to normalize the concentrations when in fact input concentrations exceeded 40
mg /1 on some irrigation days (Table 2.2).
96
_o
nj CJ c o CJ o >
o QC
s o
s o u c o o u >
o 2sj:
c o
c o u c o CJ >
a:
1-2 o 50 cm
0.8
JO
o
200 cm
oo °c5b o ̂ o o
O O o o
no
^ 0.4 o o
0
1.2
^ftrtCnpinnn o! O !Q ^ °!o
100 cm 250 cm
°c^
° o °
0.8 •— o'
0.4 ^—o °°o
ooo oo
Oo po,
0 oj O ,o o o ,o
o o
o o
LO
1.2 :— CP o
°150 cm
0.8 o 6*
o o
300 cm
o — o o o O
O r\ oo o° o
S 0.4 o
o
0
o o O !
(JB> I
0 20 40 60 80 0 20 40 60 80 100
Time (Days) Time (Days)
Figure 4.2. Relative bromide concentrations from the NMI, east side.
97
c o 1.2 50 cm 200 cm
c O U c o u o >
cs:
c
"cS b. c o o c o o o >
o a:
"S u. C a> o c o o o >
o
0.8 -O o o o
S 0.4
o o
O o o o
o o
0 j § : ^oo°o o o; O o o ° o
OD o
1.2 -
0.8 H
100 cm
o O O O
o o •« 0.4 r o
o
O O O
0
1.2
0.8
n 0.4
150 cm
o Oo°oo° ^ o
o o -
o
o o o ,
0
250 cm
O O
poo°0 O O
300 cm
o o O
CQp OQO
Oo °
0 20 40 60 80 0 20 40 60 80 100
Time (Days) Time (Days)
Figure 4.3. Relative bromide concentrations from the NMI, west side.
98
c o
o u c o CJ o
G O CJ C o CJ >
c
13 Urn
c o u c o u u >
1.2
0.8
50 cm
0.4 "-Q"
0
200 cm
° o O o ° " o
o o
;0
o o '^6^0r,'0^ O oi o O 6 ^ o
S 1.2 100 cm 250 cm
0.8 ~o O o 5_i
O Q O O
S 0.4 - cP o oO o o ^ o o
R- O
jS-0
1.2
0.8 i—
OO
CP p '
150 cm o o 0°Oo o ° n O o O O O o ^
o ° o
1 0.4 O
QC
6>
cP 2̂° ̂I
300 cm
o o
^Oo o o
CD O
0 20 40 60 80 0
Time (Days)
20 40 60 80 100
Time (Days)
Figure 4.4. Relative bromide concentrations from the SMI, east side.
99
1.6
c _o « W c
c o u o >
OC
c _o
b. c o u c o ZJ a >
OL
c
"S c o u c o u o > "5 o
_o o
1.2 50 cm 200 cm
0.8 o
0.4
o Q
°o o O o
O ̂ O (5—
0 9y»yO nUn o Q! O o ri O io O
1.2 ;— i On
0.8
100 cm — 6'
250 cm
o° o o^o O o ° O o
•« 0.4
— ° ' o
o o
0 <9,
o °0 0 ' i '
OOP o oi n o ri Q o o
idlTl
1.2 —
0.8
150 cm 300 cm o cP
O O
Oq O O O O O o ° o ~
o oO
0.4 — o o
0 (9 s ol LSJgl
0 20 40 60 80 0 20 40 60 80 100
Time (Days) Time (Days)
Figure 4.5. Relative bromide concentrations from the SMI, west side.
100
Table 4.1. Breakthrough times for the monitoring island locations
Depth (cm) North monitoring island South monitoring island
Depth (cm) West East West East
50 6.5 4 3 5 100 12.5 11.5 9 11 150 19 19 15 21 200 37 21 15 25 250 27 18 39 300 29 19 37
101
Two locations, NMI (west) and SMI (east), exhibit what may be preferential
transport. The locations show what are perceived as the breakthrough of bromide at
two different times. This behavior can be observed at depths of 150, 200 and 250 cm at
the NMI (west) locations and the 150 and 200 cm at the SMI (east) locations. Possible
causes for this behavior are discussed below.
4.4. Methods
4.4.1. Mass Balance
A mass balance was calculated for bromide. Experiment I. The mass balance
compared the amount of bromide in the water applied to the upper surface of the plot
through the trickle system to the amount of bromide in the water contained within the
soil profile at the monitoring island locations at specific times during the experiment.
The bromide mass present in the soil profile was determined by integrating the soil
solution mass profile with depth for specific times during the experiment. The amount
of bromide applied to the plot through the trickle system was calculated as a fxmction of
time as the average KBr input concentration multiplied by the volume of water
measured fi-om water meters located at the inlet water pipes of the irrigated plot. A
mass balance was calculated according to the following equations;
Mass balance = (4.13)
= (4.14) 0
102
A(^c) = (^cX_-(5'c),̂ o
(4.15)
where M is ±e mass of bromide (mg cm'^), Az is the vertical depth interval represented
by a given solution sampler (cm), A^is the change in water content from the initial
condition, Ac is the change in concentration from the initial condition (mg cm'^), and i
represents a measurement interval.
4.4.2. Approach
The approach for achieving the objectives involves multiple steps. First, the
measured data used for estimating transport properties and evaluating predictions from
different transport models are presented and summarized. Secondly, two approaches
are investigated for estimating the transport properties using CXTFIT 2.1 {Toride et al.,
1995]. Thirdly, averaging schemes using the estimated transport properties from the
two approaches are evaluated by using analytical and numerical models in the forward
mode. Fourth, numerical models are used to quantify the effects of hysteresis on solute
transport. Finally, numerical models are used to quantify the effects of possible
mechanisms, such as anion exclusion or preferential flow, which require the adjustment
of the retardation coefficient on solute transport.
103
4.4.2.1. Parameter Estimation
Two approaches similar to that of Porro et al. [1993] were taken for estimating
the transport parameters, a local approach and a global approach. For both approaches
the inverse model estimates the retardation factor and dispersion coefficient and
requires that the average pore water velocity be defined. The classical form of the
equilibrium CDE used for estimating the transport properties (Equation (4.1)), requires
that the water content and average pore water velocity are constant in both time and
space (steady-state flow). Based on these restrictions, the average pore water velocity
was calculated differently for each method. The two approaches were evaluated using
analytical models in the forward mode. The estimated transport properties were then
averaged using two different schemes. The averaging schemes were also evaluated
using analytical models in the forward mode. Six test cases were conducted to
accomplish these tasks. Table 4.6 summarizes all the test cases for the readers benefit.
In the local approach (Case 1), the parameters were estimated for each individual
depth and monitoring island location for which the breakthrough of bromide occurred
within the irrigation time period (Table 4.2). A total of 16 out of a possible 24 locations
fulfilled this criterion. Because of the above named restrictions, an average pore water
velocity was computed for each location by averaging the water contents behind the
wetting front for each particular depth interval and location and using an applied upper
flux rate of 1.85 cm d"' {y = q I Oavg).
For the global approach (Case II), the observed data from all six depths were
fitted simultaneously to estimate the solute transport parameters at each sampling
104
Table 4.2. Locations where bromide breakthrough occurred within the irrigation timeframe
Depth (cm) North monitoring island South monitoring island
Depth (cm) West East West East
50 X X X X 100 X X X X 150 X X X X 200 X X 250 X 300 X
105
location (i.e. four locations, NMI and SMI, east and west sides). The average pore
water velocity was computed for the upper 300 cm of the soil profile using the average
water content measured behind the wetting front from all the depths above 300 cm and
the applied flux rate of 1.85 cm d '. Table 4.3 shows the input parameters used in the
inverse models for Cases I and II.
4.4.2.2. Predictions Using an Analytical Model
The estimated transport properties were evaluated using analytical models in the
forward mode. The estimated transport properties were not averaged for these cases.
Cases III and IV investigated the localized and global methods for estimating the
transport properties. Case III solved the equilibrium CDE using transport parameters
estimated for each depth, for the SMI west location (Case I). Case FV solved the
equilibrium CDE using transport parameters estimated for the SMI west location using
the data from all depths collectively (Case II). This location was chosen because it is
the only location where the breakthrough of bromide was measured down to the 300 cm
depth within the irrigation timeframe (Table 4.2). Table 4.4 lists the input parameters
used in Cases III — VI.
The estimated transport properties were averaged according to the method by
which they were estimated. Cases V and VI evaluated which averaging scheme
provided the best results when compared to the measured bromide concentrations.
106
Table 4.3. Input parameters used in the inverse models. Cases I and II
Depth (cm) Case I Depth (cm) a * vavc (cm / d) R D (cm" / d) 50 0.26 7.12 1.00 10.00 100 0.25 7.40 1.00 10.00 150 0.24 7.71 1.00 10.00 200 0.23 8.04 1.00 10.00 250 0.22 8.60 1.00 10.00 300 0.22 8.60 1.00 10.00
Location Case II NMI, west 0.235 7.87 1.00 10.00 NMI, east 0.235 7.87 1.00 10.00 SMI, west 0.235 7.87 1.00 10.00 SMI, east 0.235 7.87 1.00 10.00
* average water content behind wetting front, in the upper 300 cm for Case II ^ v = q / davg, <1= 1.85 cm / d
Table 4.4. Transport parameters for the analytical model. Cases III - VI
Depth (cm) R £>(cm^ d"') V (cm d"') a (cm) Case III, SMI west, properties from localized approach
50 0.50 4.74 7.12 0.67 100 0.66 3.03 7.40 0.41 150 0.76 2.06 7.71 0.27 200 0.61 3.90 8.04 0.48 250 0.61 3.65 8.60 0.42 300 0.56 3.54 8.60 0.41
Case IV, SMI, west, properties from global approach 0-300 0.57 24.47 7.87 3.11
Case V, properties averaged from local approach 0-300 0.78 8.89 7.64 1.18
Case VI, properties averaged from global approach 0-300 0.87 21.71 7.87 2.76
107
Case V solved the CDE using transport parameters formed by averaging all 16 locations
(properties estimated from Case I, Table 4.4). This approach was taken, rather than
averaging the averages of each individual depth, because of biased weighting that
would result for the depths where breakthrough occurred at only one or two locations.
Case VI used transport parameters produced by averaging the parameters estimated
globally from Case II. The four sets of transport properties using the global method
were averaged to yield one set to describe transport through the entire soil profile.
It should be noted that both schemes could introduce a certain bias into the
transport properties. Depending on what fraction of locations experience dispersed or
"plug" flow transport processes at a specific location could be misrepresented by the
mean transport properties. Four analytical models were tested (Cases III - VI) to
evaluate the two approaches for estimating the transport properties and the averaging
schemes of these estimated properties.
4.4.2.3. Numerical Solute Transport Simulations
Numerical models were used to investigate the effect of hysteresis and the
retardation coefficient on the transport of bromide. The results of the numerical model
were also used to evaluate how well the average transport properties, estimated using
analytical models, represented the average behavior of bromide when used in a
numerical model. In addition, the results from the numerical models and the " best fit"
analytical model (Case VI) were compared to evaluate the strengths and weaknesses of
108
each approach. Three cases were simulated to explore these issues. Cases VII and VIII
used the average estimated transport properties from Case VI, which provided the best
results with respect to the MSE. The effect of hysteresis on the breakthrough of
bromide, as well as on the bromide profiles, was tested in Cases VTI and VIII. Case VII
used the flow model without hysteresis incorporated into the soil hydraulic functions.
Case VUI utilized the results from a flow model that incorporated hysteresis into the
soil hydraulic fiinctions. The results of Case VIII, which incorporated hysteresis into
the unsaturated flow model, was compared to the breakthrough of bromide with no
hysteresis specified in the unsaturated flow model (Case VII).
Case IX used transport parameters estimated by Young et al. [1999] assuming
bromide is conservative and is not subject to adsorption {ks = 0) (Table 4.5). The results
from the flow model incorporating hysteresis were used in this case. The predictions
from this case are compared to predictions from cases that use a retardation coefficient
less than one (Cases VII and VIII). Because of the manner in which R is defined,
(4.10), ks had to be estimated using a single average water content representative of the
soil profile behind the wetting front. A linear adsorption isotherm was used for this
case with negative adsorption {R< 1). It is further assumed that 100 percent of the
exchange sites were in equilibrium with the fluid.
Solutes or dissolved species subject to attenuation travel at a rate that is retarded
relative to the average pore-water velocity. Solutes may also travel at a rate that is
faster than the average pore-water velocity due to mechanisms such as negative
adsorption (anion exclusion) or preferential flow. The retardation coefficient is
109
Table 4.5. Mean transport properties used in the numerical models, Cases VII - IX
Case R or (cm) ks (mg cm'^)' f P n Flow Model
VII 0.87 2.76 -1.95 X 10'^ 1 1 0 No hysteresis
VIII 0.87 2.76 -1.95 X 10'̂ 1 1 0 Hysteresis
IX I.OO 3.1 0.0 1 1 0 Hysteresis
X^ 1 ^ ,
0.87 2.76 -1.95 X 10'̂ 1 1 0 Hysteresis Calculated using a constant water content of0.225 behind the wetting front
^ Variable input concentration boundary condition (Figure 2.4)
110
typically used to account for the net effects of a solute or dissolved species that is
attenuated in some manner with R values typically greater than 1. In cases where
solutes or dissolved species travel faster than the average pore-water velocity R values
typically less than 1 are used. Many mechanisms that are difHcult to quantify have been
identified by investigators that affect the transport of solutes. These mechanisms include
variations in local unsaturated hydraulic conductivities, variable local pore-water
velocities, preferential or macropore flow, and anion exclusion. Variations in local
unsaturated hydraulic conductivities has been attributed to variability in local pore-
water velocities creating a scale dependency in the dispersion coefficient [Ellsworth et
al., 1996]. Variability in local pore-water velocity fields in turn are attributed to
concentration hotspots, irregular concentration distributions, and the observation of
local scale dispersion coefficients that are much smaller then global scale dispersion
coefficients [Hammel and Roth, 1998]. Preferential flow has also been found to have a
significant impact on the transport of solutes. Roth et a/. [1991] found that mechanisms
they attributed to preferential flow caused a single uniform solute pulse to split into
several pulses of varying velocities causing a depletion of the main pulse with respect to
mass. Due to the number of unsaturated variables affecting transport and the non-
linearity of these properties they were unable to delineate the details of preferential
flow. Anion exclusion has been found to have a significant impact on the transport of
anions with respect to solute arrival times and concentrations [James and Rubin, 1986;
Porro et al., 1993]. Anion exclusion is the process whereby the anion, carrying a
negative charge, is repelled by the electrostatic double layer of clay particles within the
I l l
soil. The repulsion of anions away from the interface of soil particles forces the anions
into the center of the pores. Anions traveling in the center of the pores, i.e. the point in
the velocity profile where the fluid velocity is greater than the average pore water
velocity, travel farther than predicted using the average pore water velocity (i.e. /? = 1).
An additional case (Case X) was made to evaluate how the boundary conditions
as defined affect the predicted bromide breakthrough curves. Instead of using an
average input bromide concentration of 31.6 mg /1 (Equation (4.4)) a variable input
concentration was used as shown in Figure 2.4. This was done to assess if the
variability of the input bromide concentrations could help describe the observed
behavior of dual bromide breakthrough curves discussed in Chapter 4.3. Table 4.6
summarizes all the test cases for field scale solute transport for the readers benefit.
The nine models were evaluated according to their ability to accurately predict
the breakthrough time of the solute front and the mean square errors (MSE) of the
predicted versus observed solute concentrations at each depth. The MSE for one depth
is given by:
where c,, c., are the measured and predicted solute concentrations for bromide
respectively, for a time interval, i.
2
(4.16)
112
Table 4.6. Sununary of test cases for field scale solute transport
Case Description I Transport properties estimated using local approach II Transport properties estimated using global approach III Evaluation of local approach using analytical model in forward mode IV Evaluation of global approach using analytical model in forward mode V Evaluation of average transport properties from local approach using
analytical model in forward mode VI Evaluation of average transport properties from global approach using
analytical model in forward mode VII Evaluation of excluding hysteresis on solute transport using numerical models
(anion exclusion included) VIII Evaluation of incorporating hysteresis on solute transport using numerical
models (anion exclusion included) IX Evaluation of ignoring anion exclusion on solute transport using numerical
models (hysteresis incorporated)
113
4.5. Results and Discussion
4.5.1. Mass Balance
A mass balance calculation was performed using equations (4.13) - (4.14).
Figure 4.6 shows the bromide mass measured in the soil profile at the two monitoring
islands for each day during the time of bromide application. At the SMI (west side), the
amount of bromide recovered was on average only 51 percent of the amount applied; 35
percent of the applied mass was recovered on the east side. At the NMI, 46 and 29
percent was recovered from the east and west sides. At the four locations the amounts
of bromide recovered follow a trend with, on average, 41 percent of the bromide
recovered, which falls on the 59 percent error line shown. Possible sources of this error
are anion exclusion, preferential transport, bias introduced by the measurement device
(i.e. solution samplers), integration of the soil water profile, and analytical error in
measuring the bromide concentrations.
Analysis by various investigators have found a large variation in mass recovery
of bromide using solution samplers. Ellsworth et al. [1996] found that solution
samplers consistently under-estimated the applied mass for three tracers, bromide,
nitrate, and chloride. The consistency between the breakthrough curves suggested that
the poor mass recovery was not attributable to spatial variability or an inadequate
number of measurements. Rather, the solution samplers measure the concentration only
at the cup location and therefore do not provide a complete spectrum of the solute flow
paths [Ellsworth et al., 1996]. Adams and Gelhar [1992] found similar results when
114
1 ^
5 0.8 CO
o E 0.6 0 a.
1 0.4 •o o « eb 0.2
• SMI, west • SMI, east • NMI, west o NMI, east
' 59 % error
Excess recovery
Incomplete recovery
I
0
0.2 0.4 0.6 Amount applied (mg cm ")
0.8
Figure 4.6. Amounts of bromide recovered at the monitoring islands from day 5 to day 15, the solid line represents 100 percent recovery.
115
analyzing data from a 20-month natural gradient tracer study in the saturated zone of a
highly heterogeneous aquifer. The solution samplers recovered 45 to 300 percent of the
bromide mass which was attributed to spatial variability in hydraulic conductivity and
the sampling methodology. The poor mass recovery reflects the effects of plume
truncation, sampler bias, and possibly some type of tracer adsorption mechanism.
Butters et al. [1989] refer to solution samplers as flux concentration detectors and also
measured some variation in the recovery of bromide mass using solution samplers. The
variation was attributed to variations in local flux rates, imprecision of solution
samplers as flux concentration detectors, transient solute application rates and error
introduced from sample extraction rates that are higher or lower than the drainage flux
which can cause an over- or under-estimation of the bromide mass [Butters et al.,
1989]. Roth et a/. [1991] used data from a chloride tracer test that was monitored using
solution samplers to quantify preferential flow. Using predictions from a fitted
convection-dispersion model it was found that due to preferential flow the main pulse
would contain less than half of the applied solute mass while the remaining mass is
contained in several smaller solute pulses of varying velocities that split from the main
solute pulse.
Two potential mechanisms attributed to anion exclusion could contribute to the
bromide mass balance error. Due to anion exclusion the bromide is located only in the
center of the pore channels. The pore water extracted using the solution samplers is
from the entire pore channel (excluding of course the adhesive water around the soil
grain) which includes the water from the sides of the pore channels that contain no
116
anions, an excluded volume. Extraction of the excluded volume water in addition to the
volume containing the bromide anion results in a diluted measured concentration of
bromide. In addition, due to the anion traveling faster than the average pore-water
velocity some of the bromide may have traveled further than the bottom solution
sampler of the monitoring island during the sampling time interval resulting in a
truncation of the solute concentration profile. This effect could be further enhanced by
local average pore-water velocities that are greater than the calculated values.
Preferential transport could have led to bromide bypassing the area around the
lysimeter resulting in incomplete recovery of bromide. The truncation of the solute
concentration profile could be caused by the solution samplers measuring the solute
concentration of either a depleted main solute pulse or, a smaller solute pulse that split
from the main pulse due to a mechanism such as macropore flow or anion exclusion.
As Roth et a/. [1991] point out due to the number of variables and the complexities of
the relationships for unsaturated flow and transport it is difficult to determine the details
of preferential transport.
Another source of error can be attributed to the integration of the soil water
profile. On average, the calculated water applied under-predicted the actual amount
applied by 17 percent over the time period examined (day 5 — day 15). The under-
prediction of the applied water would reduce the calculated bromide mass according to
Equation (4.14). Experimental error in measuring the bromide concentration in the
samples could also account for some of the calculated discrepancies. Finally, Destouni
[1992] makes the observation that the random character of the average pore-water
117
velocity in natural fields implies that the mass flux and mass accumulation cannot be
evaluated deterministically. This concept could be applied analogously to the sampling
methodology which is deterministic in that the measurements are point measurements.
4.5.2. Modeling
4.5.2.1. Parameter Estimation, Cases I and II
The results for the local method (Case I) yielded good results for all locations
with correlation coefficients ranging between 0.72 and 0.98 (Table 4.7). The global
method (Case II) yielded somewhat poorer results with correlation coefficients ranging
between 0.66 and 0.95. The results from both approaches clearly show that transport
occurs at a rate faster than the average pore-water velocity, R values less than 1 were
found for most locations. As expected, the approach using the local method yielded
slightly better results as reflected by the lower r^ values. The poorer results of the
global method are most likely due to a combination of factors: the observed
breakthrough of bromide occurring almost simultaneously at 2 depths at the same
location, or in some cases, a deeper depth experiencing an earlier breakthrough than the
depth above it (Table 4.1); the presence of outliers at each depth having more weight
when considered collectively, and incomplete or faulty recovery of bromide combined
with experimental error in measuring the bromide concentrations, causing the
breakthrough curves to appear either more or less dispersed than they are in reality.
Porro et al. [1993] found a larger variation in dispersion than in retardation factors in
Table 4.7. Estimated transport properties from Cases I and II
Case / Location R £)(cm^/d) a (cm) r^ MSB Case I, 50 cm
NMI, west 1.03 8.34 1.17 0.80 0.0179 NMI, east 0.51 7.14 1.00 0.94 0.0124 SMI, west 0.50 4.74 0.67 0.87 0.0286 SMI, east 0.79 10.96 1.54 0.72 0.0195
Case I, 100 cm NMI, west 0.94 18.13 2.45 0.96 0.0067 NMI, east 0.84 14.07 1.90 0.96 0.0056 SMI, west 0.66 3.03 0.41 0.81 0.0188 SMI, east 0.81 5.77 0.78 0.95 0.0077
Case I, 150 cm NMI, west 1.00 18.03 2.34 0.98 0.0021 NMI, east 0.96 7.09 0.92 0.98 0.0028 SMI, west 0.76 2.06 0.27 0.93 0.0126 SMI, east 1.11 19.02 2.47 0.94 0.0028
Case I, 200 cm NMI, east 0.84 12.83 1.60 0.98 0.0012 SMI, west 0.61 3.90 0.48 0.88 0.022
Case I, 250 cm SMI, west 0.61 3.65 0.42 0.95 0.0121
Case I, 300 cm SMI, west 0.56 3.54 0.41 0.98 0.004
Mean 0.78 8.89 1.18 0.91 0.0111 CV (%) 24.63 65.80 66.00
Case II NMI, west 1.05 16.61 2.11 0.95 0.0072 NMI, east 0.85 25.45 3.23 0.80 0.0361 SMI, west 0.57 24.47 3.11 0.66 0.0633 SMI, east 1.00 20.32 2.58 0.88 0.0202
Mean 0.87 21.71 2.76 0.82 0.0317 CV (%) 24.87 18.72 18.71
119
parameters estimated using the local and global approaches in uniform and layered
laboratory columns. Similar results regarding the increased variability of dispersion
were found in this study, however, only when estimating the transport properties using
the local approach. Also, larger values of dispersion were estimated using a global
method than the local method, this is supported by the Porro et al. [1993] study. No
scale dependency with depth is observed in the dispersivity values (Table 4.7). hi
addition, the dispersivity values of 1 (local) to 3 (global) are small for field scales but in
agreement with lab data.
4.5.2.2. Evaluation of the Local and Global Approaches Using Analytical Models, Cases III - VI
The Uvo sets of transport properties estimated using the SMI west side measured
data were incorporated into the analytical model (Cases III and IV) and compared to the
observed data and to each other (Figure 4.7). The model based on transport properties
estimated for each depth and location (Case III) clearly yielded superior results as
compared to Case IV. Case III has lower MSE values (average MSE is 0.0149) than
Case IV (average MSE is 0.0670, Table 4.8). Additional cases that were simulated (not
shown) yielded similar results. As expected the models using parameters from the local
approach performed better than models using parameters from the global approach at
specific locations.
The model using average transport properties estimated globally (Case VI)
yielded better results than the model using average properties derived locally, from the
120
c o
c u u c o o o > w w
c o
1.2 — o 50 cm
0.8 r
/ 0.4 — ,
0 , ; /
1 •J
" o o
0 ' /' I
1.2 r
b
T
\0\ , O 1
100 cm
200 cm
o— / /o ° o d/
/; / o
^r\ n ^ ? ,q/
250 cm
o o 5 o u > _ra u Q[£
c c
o u c o u o >
u Qi.
0.8
0.4
0
1.2
0.8
Is O O O \ // o o o o o o ° I
\ o —r ! o
I ^^ ' J !
150 cm
Y ° o ° a ° /
« 0.4 ^— / L
0
/ o? / \ o ° J \
-L
o o
0/
^6 /
ojp
300 cm
O o
'O O f t 2/
0 / 1
I / / / o //
0 5 10 15 20 0
Time (Days)
10 15 20 25
Time (Days)
Figure 4.7. Comparison of measured bromide data and the predicted bromide concentrations at the SMI, west location. The solid line represents model predictions from Case III based on transport properties estimated locally (Case I), the dashed line represents model predictions from Case IV based on transport properties estimated globally (Case II).
121
Table 4.8. MSE values for Cases III - VI
NMI SMI Case Depth (cm) West East West East
50 0.0450
III Local
Approach
100 0.0167 III Local
Approach
150 200 250 300
0.0107 0.0037 0.0102 0.0034
Mean 0.0149 50 0.0567
IV Global
Approach
100 0.0658 IV Global
Approach
150 200 250 300
0.1352 0.0183 0.0413 0.0844
Mean 0.0670 50 0.1026 0.0861 0.1392 0.0238
v 100 0.0565 0.0183 0.0442 0.0133
Averaged from Case I
150 0.1146 0.1160 0.0275 0.2265 Averaged from Case I 200 0.1225 0.0049 0.2519 0.1171 Averaged
from Case I 250 0.0011 0.0005 0.4676 0.0011 300 0.0010 0.0015 0.3431 0.0013
Mean 0.0664 0.0379 0.2122 0.0639 50 0.0587 O.I 160 0.1730 0.0182
VI Averaged
from Case 11
100 0.0231 0.0060 0.0633 0.0115 VI Averaged
from Case 11
150 0.0457 0.0423 0.0751 0.1210 VI
Averaged from Case 11 200 0.0480 0.0059 0.3050 0.0424
VI Averaged
from Case 11 250 0.0006 0.0005 0.4680 0.0006 300 0.0009 0.0015 0.3420 0.0013
Mean 0.0295 0.0287 0.2377 0.0325
Table 4.9. Mean transport properties from Cases I and II used in Cases V and VI
Parameter Case V Case VI R 0.78 0.87
D (cm^ d"') 8.89 21.71 V (cm d"') 7.64 7.87
a (cm) 1.18 2.76
122
16 breakthroughs (Case V), for every location except the SMI west location (Figure
4.8). Case V had a mean MSE value of0.0951. Case VI had a mean MSE value of
0.0821 with a difference between the two cases of 15 percent. Case VI provides
significantly better results than Case V at the NMI (east and west) and SMI (east)
locations (Table 4.8). These locations yielded better results using the averaged
transport properties from Case II because the above named locations all behaved in a
very similar manner as opposed to the SMI (west) location that experienced a much less
dispersed breakthrough than the other locations.
4.5.2.3. Numerical Solute Transport Simulations, Cases VII - X
4.5.2.3.1 Effect of Hysteresis on Bromide Transport
Both Cases VII and VIII use a retardation coefficient less than 1. Because of the
manner in which R is defined, (4.10), ks had to be estimated using a single average
water content representative of the soil profile behind the wetting front. A linear
adsorption isotherm was used for this case with negative adsorption (anion exclusion).
Incorporating hysteresis into the soil hydraulic frmctions had little effect on the
bromide breakthrough curves (Figure 4.9); Case VIII (hysteresis) yielded marginally
better results than Case VII (no hysteresis)(Table 4.10). Figure 4.9 shows the measured
123
c
c o CJ o >
OC
c o
c u u c o
>
0^
ca K-c o c o C-)
>
o Q:i
r r
2 1.2 r © ° o
o ' _ z'
50 cm
0.8 — ° '/
§° 2 ° oSa> 9§8' o
0.4
O , ^ .' / d
//o
0 ^
1.2 I—
n
0.8 i— !
0.4 ;—
/ I L' ! -yQ i
O o \
o
J ! ^ o a la 0 I !
0
2 1.2 ;—
0.8 ^
0.4 r
0 0
100 cm
8
/o ^89o!8^ '
° I ° 1 I
150 cm
' n n g
200 cm O
O NO 0° o°e :
! O nifi e I-J^a o
250 cm
O O i ° O
I O t-» n ala a a a o^b a A.o-8i° 8
300 cm
Oo i
o °1
5 10 15 20 0
Time (Days)
J i n o?a Q J) 8 8!o a e 0 8!^ '
5 10 15 20 25
Time (Days)
Figure 4.8. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations. The dashed line represents model predictions from Case V based on average transport properties estimated locally (Case 1), the solid line represents model predictions from Case VI based on transport properties estimated globally (Case II).
124
C o
c o c o CJ ZJ >
QJ OC
c .2
c a> o c o CJ > "5
0^
c .2 "S U* C o o c o u o >
o cd
50 cm 200 cm
o o " 6 b ° - 8
° °°! o O ol o ° ° ° °
100 cm 250 cm
_ o ° 150 cm CP o 300 cm
20 40 60 80
Time (Days)
20 40 60 80 100
Time (Days)
Figure 4.9. Comparison of all measured bromide data from around the monitoring islands and the predicted bromide concentrations. The solid line represents model predictions with hysteresis in the hydraulic properties (Case VIII); the solid line with solid diamonds represents model predictions with no hysteresis in the hydraulic properties (Case VII).
125
Table 4.10. Mean square error for Cases VI - IX using average measured bromide concentrations
Depth (cm) Cases
Depth (cm) VI VII VIII, R ~ 0.87 Vni, R ~ 0.80 I X , R = 1
50 0.0195 0.0165 0.0145 0.0106 0.0273 100 0.0373 0.0044 0.0043 0.0177 0.0316 150 0.2555 0.0237 0.0232 0.0555 0.0087 200 0.1664 0.0528 0.0536 0.0676 0.0281 250 0.1439 0.0388 0.0331 0.0606 0.0282 300 0.1624 0.0319 0.0382 0.0149 0.1463
Mean 0.1308 0.0280 0.0278 0.0378 0.0451
126
bromide concentrations from all monitoring island locations at each depth along with
the simulated results from Cases VII and VUI. At most depths, both cases predicted the
mean behavior of the measured bromide and yielded predictions that are difRcult to
distinguish from each other. Only at late times can the predictions from the two cases
be delineated. At the deeper depths, the numerical solutions failed to accurately predict
the bromide concentrations at the SMI and NMI, west locations because of diverging
effects. At the SMI west location, the measured bromide breakthrough curves had a
sharp interface that indicates primarily advective transport with little or no dispersion.
At this location the numerical solutions under predicted the magnitude of the bromide
breakthrough as well as over predicted the arrival time of the solute front. At the NMI,
west location, the solute front followed a very dispersed pattern and the numerical
solutions over predict the bromide concentrations. The MSE values (Table 4.10)
supported the observation that the model without hysteresis in the soil hydraulic
functions (Case VII) yielded very similar results as compared to the model
incorporating hysteresis into the soil hydraulic functions (Case VIII).
The predicted bromide concentration depth profiles also support the conclusion
that hysteresis does not play a major role in the transport of bromide (Figure 4.10 -
Figure 4.12). No significant effect from incorporating hysteresis into the soil hydraulic
functions could be detected in the predicted concentration profiles from Cases VII and
VIII. Both solutions yield good predictions with respect to the position of the measured
solute front as well as the shape of the bromide concentration profiles. These results
127
0 0.2 Relative concentration
0.4 0.6 0.8 1.2
SMI, west
SMI, east
NMI, west
NMI, east
— —Case VII
Case Vin
Figure 4.10. Measured bromide concentrations and model predictions for day 16, irrigation day 16.
128
0 0.2
Relative concentration
0.4 0.6 0.8 1.2
SMI, west SMI, east NMI, west NMI, east
— — Case VII Case vm
Figure 4.11. Measured bromide concentrations and model predictions for day 24, the last day of irrigation.
129
0 0.2 Relative concentration
0.4 0.6 0.8 1.2
c o
a. u Q
SMI, west • SMI, east • NMI, west • NMI, east
"Case vn 'Case Vni
Figure 4.12. Measured bromide concentrations and model predictions for day 93, the last measured day of redistribution.
130
hold even for late time predictions (Figure 4.12). These results are supported by similar
studies conducted by others [Mitchell and Mayer, 1998; Pickens and Gillham, 1980].
4.5.2.3.2 Preferential Transport
Numerical predictions for a nonsorbing, nonreactive, solute (/? = 1.0, ks = 0.0,
Case IX) were compared to the results from Case VIII {R = 0.87) to determine the
effects of ignoring potential mechanisms such as preferential transport or anion
exclusion. These mechanisms, preferential transport and anion exclusion, are used here
to describe accelerated solute transport. Because the cause of the accelerated transport
is unknown the retardation coefficient is used as an effective parameter to capture the
average preferential transport behavior rather than a specific process. As mentioned
previously, the transport parameters used in the numerical models represent the
properties fi-om the forward analytical models that had the lowest MSE's. The mean
values from the global approach yielded the best results in the forward analytical
models as compared to mean transport properties from the local approach.
Case VIII yielded the best overall results in terms of the MSE's with respect to
predicting the average bromide concentrations at all the locations (Table 4.10).
However, the predictions from the numerical models yielded mixed results in terms of
the specific locations. At the 50 cm depth, all of the models accurately predicted the
measured bromide concentrations (Figure 4.13). However, at eight of the remaining 20
locations. Case IX yielded better predictions than Case VIII (Figure 4.14 -Figure 4.17).
131
c o
c u u c o CJ u >
<u o£.
NMI, west NMI, east
p Q o p j
SMI, west
20 40 60 80
Time (Days)
iFfirnnriuianu'ft iS'u'^—; SMI, east i
grniarea
20 40 60 80 100
Time (Days)
Figure 4.13. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 50 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VIII (R ~ 0.87), the dashed line represents results from Case IX (/? ~ 1.0), the solid line with solid circles represents results from Case VIII (R ~ 0.8).
132
NMI, east
c o
u C o o o >
0^
NMI, west
SMI, west SMI, east
20 40 60
Time (Days)
20 40 60 80 100
Time (Days)
Figure 4.14. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 100 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VIII (R ~ 0.87), the dashed line represents results from C a s e I X ( R ~ 1 . 0 ) , t h e s o l i d l i n e w i t h s o l i d c i r c l e s r e p r e s e n t s r e s u l t s f r o m C a s e V I I I ( R ~ 0.8).
133
c .0
Urn c CJ r— o u >
oNMI, west
SMI, east SMI, west
20 40 60
Time (Days)
20 40 60
Time (Days)
100
Figure 4. IS. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 150 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VIII (R ~ 0.87), the dashed line represents results from Case IX (/? ~ 1.0), the solid line with solid circles represents results from Case VIII {R ~ 0.8).
134
c o
o o c o u o > "5 u
o NMI, east fi NMI, west
o o
SMI, east SMI, west
20 40 60 80 0
Time (Days)
20 40 60 80 100
Time (Days)
Figure 4.16. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 200 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VIII (R ~ 0.87), the dashed line represents results from C a s e I X ( / ? ~ 1 . 0 ) , t h e s o l i d l i n e w i t h s o l i d c i r c l e s r e p r e s e n t s r e s u l t s f r o m C a s e V I I I ( R ~ 0.8).
135
NMI, west NMI, east
§) SMI, west SMI, east
20 40 60
Time (Days)
20 40 60 80 100
Time (Days)
Figure 4.17. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 250 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VIII (R ~ 0.87), the dashed line represents results from C a s e I X ( / ? ~ 1 . 0 ) , t h e s o l i d l i n e w i t h s o l i d c i r c l e s r e p r e s e n t s r e s u l t s f r o m C a s e V I I I ( R ~ 0.8).
136
Six of these locations are at the NMI, west location where measured bromide
concentrations adhere to a more dispersed pattern than the other locations (Figure 4.19).
Predictions from the conservative solute model (/? = 1, Case IX) also yielded better
results at the 150, 200, and 250 cm depths with regard to the average bromide behavior.
At the deeper depths (below 250 cm) Case IX failed to accurately predict the arrival
time of the solute front as well, and predicted a much more dispersed breakthrough
curve. This is obvious at the 300 cm depth (Figure 4.18). The retardation factor of
Case VIII (R ~ 0.87) was adjusted toR~ 0.8 to establish whether or not a better fit
could be achieved. Case VIII provided better results at all depths, except at the 300 cm
depth, using an /? ~ 0.87.
The fact that using an /? ~ 0.80 in Case VIII provided better results at the 300 cm
depth indicates that solutes are transported faster than the average pore-water velocity
above this depth. These results suggest that at the deeper depths, a mechanism such as
preferential transport or anion exclusion (or a combination of mechanisms) play a more
significant role in the transport of bromide. This may be attributed to the
heterogeneities of the soil profile due to layering or spatial variability in the velocity
field. The SMI, west location in particular exhibited transport significantly faster than
that predicted using an average pore-water velocity (Figure 4.21). Predictions clearly
show that it is necessary to use an /? < I although it is not clear what mechanism causes
this preferential solute transport. Adams and Gelhar [1992] attribute incomplete mass
recovery on solute transport that occurs at a rate greater than the average pore-water
velocity resulting in the solute traveling farther than would be predicted. This
137
_o rt c u o o o <u >
u cd
NMI, west NMI, east q O o o O
CO/ a
^ o SMI, west Q cP ̂ o o o o
o ^ o
SMI, east
o o
&
20 40 60 80 0
Time (Days)
20 40 60 80 100
Time (Days)
Figure 4.18. Comparison of measured bromide data and the predicted bromide concentrations at the monitoring island locations, 300 cm depth. The solid line with triangles represent results from Case VI (analytical model, R = 0.87), the solid line represents results from Case VIII (R ~ 0.87), the dashed line represents results from Case IX (/? ~ 1.0), the solid line with solid circles represents results from Case VIII (/? ~ 0.8).
138
c o
c <u u o u o >
o
o o c o u o >
o a:
c .2 eS u. C u c c u o >
<u
200 cm 50 cm
^iWooOo ^ _ o o o o
250 cm 100 cm
7 ' O •/ ̂ ponOO O O
300 cm 150 cm
0 20 40 60 80
Time (Days)
20 40 60 80 100
Time (Days)
Figure 4.19. Comparison of measured bromide data and the predicted bromide concentrations at the NMI, west location. The solid line represents model predictions from Case VIII (R ~ 0.87), the dashed line represents model predictions from Case IX (R = 1.0, Young et al., 1999), and the solid line with solid circles represents model predictions from Case VIII (R ~ 0.80).
139
c o
c o o c o
c o C3
(U o c o u u >
c o a
o u c o o >
(D
50 cm 200 cm
o ^ O o
•I '
I'
1.2 i—
K-^rr- rr^n—Q ^-O
100 cm 250 cm
°150 cm 300 cm
o O no 0°
20 40 60 80
Time (Days)
20 40 60 80 100
Time (Days)
Figure 4.20. Comparison of measured bromide data and the predicted bromide concentrations at the NMI, east location. The solid line represents model predictions from Case VIII (R ~ 0.87), the dashed line represents model predictions from Case IX (R = 1.0, Young et al., 1999), and the solid line with solid circles represents model predictions from Case VIII (R ~ 0.80).
140
c _o « k-c o u c o u u >
u Qi
c _o b. c u u c o u o > w u
Qi.
c o
c u u o u u >
u Qi
200 cm 50 cm
250 cm 100 cm
° oioo o o o 'o n o !o 300 cm 150 cm
0 20 40 60 80 0
Time (Days)
20 40 60 80 100
Time (Days)
Figure 4.21. Comparison of measured bromide data and the predicted bromide concentrations at the SMI, west location. The solid line represents model predictions from Case VIII {R ~ 0.87), the dashed line represents model predictions from Case IX (/? = 1.0, Young et al., 1999), and the solid line with solid circles represents model predictions from Case VUI {R ~ 0.80).
141
explanation is convenient in that it can describe both the observed accelerated transport
at the deeper depths as well as the incomplete bromide mass recovery. Another
explanation that would explain the observations of incomplete bromide recovery as well
as the accelerated transport is described by Roth etal. [1991] as preferential flow. They
attributed incomplete bromide mass recovery to preferential flow where several smaller
solute pulses of varying velocities split from the main solute pulse. There may be some
evidence of this preferential transport at the NMI (west) and SMI (west) locations
where a dual bromide breakthrough can be detected (Chapter 4.3, Measured Bromide
Concentrations) although, as Vanderborght et al. [2000] points out, transport through
macro pores is barely apparent in time series measured in-situ. However, whatever the
mechanism, it is clear from model predictions that it is impossible to predict the
breakthrough of bromide at the deeper depths without using a retardation coefficient
less than one.
The results of the analytical solution (Case VI) are also shown and compared to
the numerical solutions (Case VII - VIII) at the monitoring island locations (Figure 4.13
- Figure 4.18). At the 50 cm depth the analytical solution yielded results similar to that
of the numerical solution (Table 4.10). This is supported by the MSE that was
computed using the average measured bromide concentrations for each depth against
the simulated bromide concentrations. It should be noted that the results of the
analytical model (Case VI) compare favorably with the measured bromide data at the 50
and 100 cm depths although the solution is only valid for times within the irrigation
period. As expected, at the 150 and 200 cm depths, the predictions of the analytical
142
model provided good results only at early times. For the 250 and 300 cm depths, as
well as the mid to late time solutions for the 150 - 200 cm depths, the analytical
solutions were not valid and failed to accurately predict the bromide concentrations.
This is because of the inability of the analytical model to account for solute transport
due to drainage and displacement processes.
Figure 4.19 through Figure 4.22 show the measured and predicted breakthrough
of bromide at each of the monitoring island locations. Viewing the data in this manner
more clearly shows the effect of dispersion at the individual locations. The SMI, west
side location clearly shows the strongest anion exclusion or preferential transport with a
sharp breakthrough and the bromide breaking through at all depths by day 20. The
simulated bromide breakthrough curves closely match the observed data at the SMI,
east side and the NMI, west side locations.
Varying the input bromide concentrations, instead of using an average bromide
concentration of 31.6 mg/1, did not improve model predictions with respect to the
average bromide behavior. Figure 4.23 shows the predicted bromide breakthrough
curves for the model using an average bromide concentration of 31.6 mg / 1 for the
upper boundary condition (Case VIII) and the model using a variable bromide
concentration (Case X, Figure 2.4) for the upper boundary condition. Defining the
bromide concentrations for the upper boundary explicitly did not improve the
predictions with respect to capturing the details of the preferential transport.
143
0 20 40 60 80 0 20 40 60 80 100
Time (Days) Time (Days)
Figure 4.22. Comparison of measured bromide data and the predicted bromide concentrations at the SMI, east location. The solid line represents model predictions from Case VIII (R ~ 0.87), the dashed line represents model predictions from Case IX (R = 1.0, Young et al., 1999), and the solid line with solid circles represents model predictions from Case VIII (R ~ 0.80).
144
0 20 40 60 80 0 20 40 60 80 100 Time (Days) Time (Days)
Figure 4.23. Comparison of all measured bromide data from around the monitoring islands and the predicted bromide concentrations. The solid line represents model predictions using a uniform input concentration (C© = 31.6 mg / 1, Case VIII); the dashed line represents model predictions using variable input concentrations (Figure 2.4, Case X).
145
Predictions actually worsened with respect to the average measured bromide
concentrations with percent difference between the measured and predicted bromide
concentrations increasing from 15.9 percent for Case VII (average bromide
concentration) to 21.5 percent for the model that explicitly define the bromide
concentrations for each irrigation day (Case X).
4.6. Conclusions
Numerical inversion techniques were used with measured bromide concentrations
at the monitoring island locations to estimate the transport parameters, R and a. The
transport of bromide was then simulated using analytical and numerical models and the
results compared to measured bromide concentrations at the monitoring island
locations.
Numerical inversion was utilized to estimate the transport properties using
measured bromide concentration data from the monitoring island locations. The
properties were estimated using two approaches introduced by Porro et al. [1993]: a
local approach where the measured concentration data are "fit" at each individual
location, and a global approach where the measured concentration data from six depths
at a location are "fit" simultaneously. These estimated transport parameters were then
used in a forward analytical model to assess which approach yielded the best results.
Examination of the correlation coefficients (r^) reveal that the local approach produced
the best results with respect to the "fit" of the data (Table 4.7). The estimated transport
146
properties from the SMI, west location were also used in an analytical model to assess
the two approaches. As expected, the local approach clearly offered superior results
compared to the global approach. These results held for other locations as well.
The parameters estimated fi'om these two approaches were then averaged and
incorporated into analytical models to investigate how averaging affects the previous
results. Based on previous results, it would appear reasonable to assume that the
estimated transport properties averaged from the local approach would yield superior
results. However, the analytical models using average transport properties from the
global approach yielded better results than averaging the transport properties obtained
from the local approach at all locations except the SMI, west location (Table 4.8). At
the SMI, west location the results were slightly better with an 11.3 percent difference in
the MSB values. At the other locations the results of the analytical model using average
transport properties from the global method yielded more significant differences with
the MSE's of the two methods differing by 28 to 77 percent.
Based on the results of the averaging schemes, the average transport properties
from the global approach were used in the numerical models. Numerical models were
developed to investigate the effects of hysteresis and anion exclusion on the transport of
bromide. In addition, the results from the numerical models were compared to the
results obtained from the best analytical model (Case VI) and used to provide insight to
how well the average transport properties represent the measured bromide behavior and
to compare the solutions. Incorporating hysteresis into the soil hydraulic functions had
little effect on the predicted bromide breakthrough curves. The models incorporating
147
hysteresis only yielded marginally better results than the models with no hysteresis
(Table 4.10). At most depths, both models predict the mean behavior of the measured
bromide and yield predictions that are difficult to distinguish from each other. Only at
late times can the predictions from the two cases be delineated. The predicted bromide
concentration profiles also support the conclusion that hysteresis does not play a major
role in the transport of bromide. No significant effect from incorporating hysteresis into
the soil hydraulic functions could be detected in the predicted concentration profiles
from the two models. Both solutions yield good predictions with respect to the position
of the measured solute front as well as the shape of the bromide concentration profiles.
These results also hold for late time predictions. These results are supported by similar
studies conducted by others [Mitchell and Mayer, 1998; Pickens and Gillham, 1980].
The results of the models using an /? < 1 demonstrate that while the mechanisms
which cause the solutes to travel faster than the average pore-water velocity catmot be
delineated, the process or processes must be accounted for and are lumped into the
retardation coefficient (Table 4.10). At the deeper depths ( below 250 cm), the
importance of using an /? < 1 becomes obvious as the model that uses R = 1 completely
fails to accurately predict the arrival time of the solute front and predicts a much more
dispersed breakthrough curve. Adjusting the retardation factor toR~ 0.8 failed to yield
better results compared to numerical models using a ~ 0.87, except at the 300 cm
depth. This result may indicate that the system experiences enhanced variations in the
velocity fields above this depth interval. At the deeper depths faster velocity fields play
a more significant role in the transport of bromide. The results of the analytical solution
148
(Case VI) were compared to the numerical solutions at the monitoring island locations.
The analytical solution yields results similar to those of the numerical solution at the 50
cm depth (Table 4.10). It should be noted that the results of the analytical model (Case
VI) are only valid for times within the irrigation period because of the assumption of
steady-state flow . As expected, at the late times and the deeper depths, the analytical
solutions are not valid and fail to accurately predict the bromide concentrations. This is
because of the inability of the analytical model to account for solute transport due to
drainage and displacement processes.
149
5. CONCLUSIONS
The objective of this dissertation was to determine how changes to the
parameterization of the conceptual model affects the flow and transport predictions.
The sub-objective of this study was to simulate the water flow and the transport of a
solute through a variably saturated heterogeneous field soil.
The infiltration and redistribution of water into a large (50 m by 50 m), controlled
and monitored field plot was modeled and the results compared with measured data.
Effects of the inclusion of hysteresis in the retention and hydraulic conductivity
functions were determined, and the ability of laboratory derived hydraulic parameters
from core samples to accurately predict the propagation of the wetting front under a
field setting was examined. Both the wetting front arrival and the changes in water
content using laboratory derived hydraulic properties incorporating hysteresis were
modeled. The arrival of the wetting front was simulated and compared to observed data
from two successive field infiltration events. Several models were developed using
hydraulic properties derived from drying and wetting experiments. These models were
then compared and the parameters adjusted if necessary, against data taken from the
first infiltration experiment. A post-audit of the final models from Experiment I were
conducted using data from a second experiment. Using the adjusted hydraulic
properties from Experiment I, simulations were performed and the results compared to
the measured data from Experiment II to assess the quality of the final model as well as
the assumption regarding using a homogeneous representation of the soil profile.
150
Simulations using hydraulic properties estimated in the laboratory with core
samples as first approximations yielded poor results with regards to estimating the
arrival of the wetting front at various depths as well as predicting the drainage phase of
the experiment until adjustment was made to the Ks parameter. The satmated hydraulic
conductivity, Ks, was adjusted to correctly predict the time of the wetting front arrival
and the changes in water content observed in the field during and after infiltration.
Without increasing Ks tenfold from the value of 46 cm d"', determined in the laboratory,
to 450 cm d ' acceptable predictions could not be made using the given set of hydraulic
properties and /Tj as a fitting parameter. The Ks adjustment is justified by computing
unsaturated hydraulic conductivity values determined in the field from tensiometer
observations at a range of depths from 0.5 to 3 m from around the monitoring islands
and average water content values from neutron probe data in a modified version of the
scaled-predictive method. Thus Ks became an effective saturated hydraulic
conductivity, Ks.eff, which incorporates uncertainties not included in the laboratory
determined Ks. The modified version of the scaled-predictive method introduced by
Mace et al. [1998] offered a valuable tool for determining how much to adjust Ks to
adjust the predictive hydraulic conductivity relationship. The adjustments lead to
reasonably good results for estimating the arrival of the wetting front and changes in
water content. The results also support the use of a homogeneous vertical soil profile
for this particular field site.
Laboratory derived hydraulic properties provided a much better first
approximation to the field properties than hydraulic properties based on soil texture
151
taken from an internal, general database in HYDRUS. It was not possible to adjust
either of the two models using hydraulic properties taken from the internal database as a
first approximation to the same degree of quality as the models using the laboratory
derived properties. Predictions based on properties taken from a database were unable
to capture the behavior of the changes in the water content in terms of the magnitude of
the water contents behind the wetting front and the redistribution phase of the
experiment. These results may, in part, stem from the use of a homogeneous profile.
The laboratory derived hydraulic properties are, in effect, based on a lumped parameter
profile where, although the soil core represents a small volume (or point measurement)
of the entire profile, they capture the average behavior in terms of the properties.
Models with hydraulic properties based on textural considerations used in a soil profile
with properties not conforming to a particular texture would be unable to capture the
average behavior. Capturing more distinct characteristics of the system by delineating
more layers of the system may correct this. Based on these results the laboratory-
derived parameters provide a unique and superior first approximation for predicting the
wetting front arrival and the shape of the curves as compared to predictions using
hydraulic properties from the internal database.
The simulations utilizing properties taken from wetting experiments and the case
incorporating hysteresis into the retention and hydraulic conductivity functions yielded
the best results as compared to the results from utilizing hydraulic properties taken from
drying experiments. Comparison of the results from these two cases indicates that
although hysteresis was not a dominant mechanism during these experiments the model
152
incorporating hysteresis does provide only marginally better results for predicting water
content changes. These results may be because the soil profile remained in the "wet"
region of the retention curve during most of the experiment. Incorporation of hysteresis
into the soil hydraulic functions, however, yielded superior predictions of the soil water
content profile as compared to predictions without hysteresis a finding which is
supported by similar observations by Pickens and Gillham [1980]. In addition,
incorporation of hysteresis resulted in the use of a more realistic value for the saturated
hydraulic conductivity as compared to the case without hysteresis (Case II). Another
factor affecting these results is that while the model incorporating hysteresis follows a
separate main drying curve fi-om the non-hysteretic model (Figure 3.18), the maximiun
difference between the drying curve for these two cases is approximately 7 percent at
100 cm of suction; this would indicate that a large difference should not be expected
between the two cases.
A post-audit of the adjusted parameters conducted by simulating a second
irrigation event in the same plot, yielded good results, with the MSE (Table 3.5) of the
same order as the results obtained fi'om the non-hysteretic and hysteretic models fi-om
Experiment I. In addition, using the results fi-om the models collectively, it would
appear that the assumption regarding the use of a homogeneous soil profile is supported
when trying to capture the average field behavior of processes such as the wetting fi-ont
arrival and the drainage behavior. However, it is very clear that the use of a
homogeneous profile to represent the real system fails when looking at detail orientated
processes such as water content changes at a specific point.
153
The transport of a solute through a variably saturated field soil was simulated
using analytical and numerical models. Location specific and spatially integrated
inversion methods were compared using analytical and numerical models, to assess
their ability to characterize the flux and transport properties. The effects of spatial
averaging of point measurements {9, R, D, a) on the transport predictions were
investigated. And finally, numerical models were used to investigate the effects of
hysteresis and the retardation coefficient on solute transport.
Transport properties were estimated using two approaches, a local and global
method. In the local approach the measured concentration data are "fit" at each
individual location, whereas in the global approach the measured concentration data
from six depths at a location are "fit" simultaneously. The local approach produced the
best results with respect to the correlation coefficients. The estimated transport
parameters from the two methods were then used in a forward analytical model to
assess which approach yielded the best results. The transport properties estimated at the
SMI, west location was used in an analytical model to assess the two approaches. As
expected, the local approach clearly offered superior results compared to the global
approach. These results held for other locations as well.
The parameters estimated from these two approaches were then averaged and
incorporated into analytical models to investigate how averaging affects the previous
results. The analytical models using average transport properties from the global
approach yielded better results than models using average transport properties from the
local approach at all locations except the SMI, west location. This is counter-intuitive
154
based on the previous results since the analytical models using transport properties
estimated from the local approach yielded superior results for specific locations. At the
other locations the results of the analytical model using average transport properties
from the global method yielded predictions that were 28 to 77 percent better in terms of
the MSE's than the models using properties from the local method.
The average transport properties from the global approach were used as first
approximations in the numerical models. Transport properties from the global approach
were chosen because predictions from analytical models using these properties yielded
better results in terms of the MSE's as compared to models using average properties
from the local approach.
Incorporating hysteresis into the soil hydraulic functions had little effect on the
predicted bromide breakthrough curves. The models incorporating hysteresis only
yielded marginally better results than the models with no hysteresis. At most depths,
both models predict the mean behavior of the measured bromide and yield predictions
that are difficult to distinguish from each other. Only at late times can the predictions
from the two cases be delineated. The predicted bromide concentration profiles also
support the conclusion that hysteresis does not play a major role in the transport of
bromide. No significant effect from incorporating hysteresis into the soil hydraulic
functions could be detected in the predicted concentration profiles from the two models.
Both solutions yield good predictions with respect to the position of the measured solute
front as well as the shape of the bromide concentration profiles. These results also hold
155
for late time predictiotis. These results are supported by similar studies conducted by
others {Mitchell and Mayer, 1998; Pickens and Gillham, 1980].
The results of the models using an /? < 1 demonstrate that while the mechanisms
which cause the solutes to travel faster than the average pore-water velocity cannot be
delineated, the process or processes must be accounted for and are lumped into the
retardation coefficient (Table 4.10). Numerical models using an /? ~ 0.87 yielded the
best results compared to numerical models that used retardation factors of 0.80, and 1.0
with respect to the MSE's. At the deeper depths ( below 250 cm), the importance of
using an /? < 1 becomes obvious as the model that uses R = 1 completely fails to
accurately predict the arrival time of the solute front and predicts a much more
dispersed breakthrough curve. Adjusting the retardation factor to R ~ 0.8 failed to yield
better results as compared to numerical models with ~ 0.87 with regard to the arrival
time of the bromide breakthrough except at the 300 cm depth. This result may indicate
that the system experiences enhanced variations in the velocity fields above this depth
interval. A possible explanation for this result is that at the deeper depths the local
pore-water velocity is greater due to some mechanism such as preferential flow or anion
exclusion and plays a more significant role in the transport of bromide. The results of
the analytical solution were compared to the numerical solutions at the monitoring
island locations. The analytical solution yields results similar to that of the numerical
solution at the 50 cm depth. It should be noted that the results of the analytical model
are only valid for times within the irrigation period because of the assumption of steady-
state flow . As expected, at the late times and the deeper depths, the analytical solutions
156
are not valid and fail to accurately predict the bromide concentrations. This is because
of the inability of the analytical model to account for solute transport due to drainage
and displacement processes.
In conclusion, for this field site, and under these conditions, the use of a
relatively simple one-dimensional model is suitable for predicting processes such as
wetting front advance and the transport of a solute at the scale of tens of meters. The
results also show that although laboratory derived hydraulic properties may be limited
in their applicability to field settings, for this field setting they clearly offer a superior
first approximation as compared to hydraulic properties based on texturai considerations
taken from a database. However, the laboratory derived hydraulic properties needed to
be adjusted in order for the numerical model to yield good predictions of the wetting
front arrival and changes in water content. These adjustments were made following an
iterative approach where the results of the numerical model reveal inconsistencies of the
hydraulic properties and the field data are used to adjust the conceptual model. This
iterative approach is used to achieve many of the objectives outlined above.
157
APPENDIX: CHARACTERIZATION
A variety of approaches were applied to characterize the MAC site. Most of
these approaches provided some quantitative information of varying quality with
respect to deriving hydraulic properties. However, these approaches did provide
valuable information with respect to the hydraulic and transport properties and observed
field processes. This section outlines some of these approaches to provide for a clearer
understanding of why some of the later approaches were taken.
Deriving hydraulic properties using the measured field data was an important
goal at the onset of the project. The primary types of data available for this task were
neutron probe data that yielded calculated water contents, tension data from installed
tensiometers, and textural data (sand, silt, clay, and gravel) at selected locations within
the plot. Ideally, water retention curves are generated from the observed data and soil
hydraulic properties are estimated from these curves for use in a soil hydraulic model.
The following sections outline the various approaches taken, as well as the processes
examined, to characterize the site's hydraulic properties at the preliminary stage of the
project. The approaches outlined are: development of field based water retention
curves, the assessment of variability in the measured neutron probe data in support of
deriving field based water retention curves, and examination of the unsaturated
hydraulic conductivity using the internal drainage method. In addition, it should be
recognized that what the data may lack in resolution it makes up for in volume (number
of point measurements) and the synergistic manner in which the data are examined!
158
Analysis of the Neutron Probe Data
Many of the approaches and methods used to estimate hydraulic properties, such
as the instantaneous profile method and use of field derived water retention curves
require that the in-situ water contents and potentials are defined. Because of variability
in the neutron probe measurements (hence variability in the calculated water contents),
various mechanisms were examined to account for this variability and various methods
were examined to reduce the variability the neutron probe data.
Water content values for infiltration experiments conducted at the MAC field
site were indirectly determined using a neutron probe. The neutron probe (neutron
source and detector) was placed in the soil in access tubes and high-energy neutrons
were emitted fi-om a Ra-Be or Am-Be source. Neutrons then collide with nuclei of
atoms in the surrounding soil. When the mass of the nuclei the neutron collides with is
much larger than the neutron mass, the neutron does not lose any appreciable energy
and the neutrons' velocity is retained. When the neutrons strike nuclei of a similar mass
(such as that of hydrogen nuclei), much of the neutrons' energy is lost. After a number
of these collisions, the velocity of the neutron becomes characteristic of the thermal
motion of hydrogen atoms in the soil [Jury et al., 1991]. This velocity is referred to as
"thermal velocities" (velocities characteristic of the hydrogen atoms in the soil). The
neutron-probe detector is sensitive only to neutrons moving at these thermal velocities.
The thermalized neutron count is proportional to the density of hydrogen atoms in the
159
vicinity of the source. A calibration curve can be developed "giving the number of
counts per unit time received by the detector versus the amount of hydrogen present"
[Jury etal., 1991]. The plotted neutron readings are not absolute counts but instead
count ratios {CR), which are a ratio of absolute counts to a shield count.
A polynomial equation is used to fit the data collected and calculate the
volumetric water content:
<51 = Co + C,(C/?) (A.I)
where Ov is the percent moisture content, Co and C/ are constants, and CR is the count
ratio determined as the count divided by the standard count (5C).
Volumetric water contents for the infiltration experiments conducted at the
MAC site were determined as follows. A standard count was taken in a vertical neutron
probe access tube isolated from the irrigated plot, approximately 20 m to the west,
before the actual count was taken in the irrigated plot. The standard count was taken at
a 2 m depth. The standard count for each day was then used to calculate the CR for that
day and then subsequently used in Equation (A. I).
The resulting calculated water contents were somewhat variable. This
variability can be attributed to the methodology underlying the determination of the
constants (Co andC/) in Equation (A.I), and a possible temperature dependence in
neutron probes.
The neutron probe used in this study was calibrated utilizing a steady-state
method. A small plot (12.2 by 6.1 m) containing 3 access tubes was irrigated 6 days
and then allowed to redistribute. Neutron probe readings were taken in 25 cm depth
160
increments during the data collection period. Following the data collection period, a
Giddings probe was used to collect soil core samples for water content analysis. The
calibration curve was constructed by correlating the neutron probe counts to the water
content determined by direct method and the constants, Co andC/ were subsequently
derived from the curve. Co andC/ were calibrated as -0.05830 and 0.21853,
respectively. Because the polynomial equation constants were developed for a plot
located off the experimental plot, and because they represent averaged values for the
calibration plot (taking into account heterogeneities for the entire calibration plot) they
1) carmot account for heterogeneity's on a very localized scale and 2) may not be
representative of the plot because the plot's physical properties may have been altered
during site preparation and installation of instnmients. No attempt was made to address
these complications.
Personal communication with Dr. Wierenga (Professor, Department of Soil,
Water and Environmental Science, University of Arizona, 1999) indicated that there
may be some temperature dependence associated with neutron probes. To explore
temperature dependence, the C/?'s, thus the water contents, were recalculated using
different techniques to determine the standard counts. Three approaches were used to
adjust the standard counts: a mean standard count value was calculated for each
individual experiment, a mean standard count was calculated representing both
experiments, and a regression was applied to the standard counts. Each approach was
evaluated by recalculating the water contents using the "new" standard count and then
161
quantifying whether there was an improvement in the "fit" of a regression applied to the
water content versus time curve.
The water contents were recalculated using four methods: the experimental day
SC (using a SC for the specific day), a mean standard count for the specific experiment,
a mean lumped SC (mean SC is computed using all experimental data), and a SC based
on a regression applied to the SC versus time curves. The "new" water contents were
then plotted versus time and regressions applied to them to check if the correlation
between the points had improved.
In every case, the correlation coefficients of the regressed water contents versus
time improved marginally for the water contents that were computed using either the
experimental mean, the lumped mean, or the regressed SC, as compared to the original
method. Also in every case, the method using the regressed SC resulted in marginally
better correlation coefficients than the methods using the means. This is surprising
considering the regressed SCs correlation coefficient for Experiment II was very low
(0.0001). Because of the very low regressed correlation coefficients, the improved
results were observed as being artificial and the method abandoned. It should also be
noted that despite the elaborate methods by which a "new" SC, was derived the end
result on the actual water content values was minimal. Based on these findings a closer
look was taken at the data used to calibrate the neutron probe.
Analysis of Field Derived Water Retention Curves
162
The field derived parameters (FDP), specifically van Genuchtens' curve fitting
parameters alpha and m(n = l/(l-m)), were determined by correlating the water content
to the tension data through time and processing the resulting tension versus water
content data with the code RETC [van Genuchten et al., 1991]. The residual water
content, saturated water content, and saturated hydraulic conductivity were estimated by
examining the raw data. Because of the questionable reliability of the tension data in
the 300-400 cm dry range (Wierenga, Personal Communication, 1999), only data fi-om
the wet range was input into RETC. The measured tension data fi-om the monitoring
islands was used. RETC was used to solve the inverse problem and generated an alpha
and n parameter for each depth (50 - 300 cm, 50 cm intervals) for seven locations
within the irrigated plot. Basic statistics were then performed on the parameters to
determine a mean value and examine the variability. The parameters were normally
distributed, however it should be noted the sample size was small. The residual and
saturated water contents were estimated by examining the raw data, the derived core
values, and the expected range of values given by the textural composition. Because of
a lack of reliable in-situ K values, the mean saturated hydraulic conductivity value was
taken fi-om laboratory experiments conducted on the core samples collected fi-om the
plot.
As indicated in the Analysis of Neutron Probe Data, the variability in the
calculated water contents resulted in highly variable water retention curves. In addition,
when fitting only two (a and m) of the four parameters defined in the inverse problem
{Or, 6s, a and m), the correlation coefficients were poor. This is attributable to the
163
variability of the calculated water contents and the limited range over which the
measured data extended. Because of these results, it was decided to use hydraulic
properties estimated in the laboratory from soil cores collected from the field site.
Estimation of the Unsaturated Hydraulic Conductivity Relationship
The internal drainage method as described by Hillel et al. [1972] was applied to
field data collected from the MAC site to determine the imsaturated hydraulic
conductivity relationship. The internal drainage method is derived from the
instantaneous profile method developed by Watson [1966], a method originally
developed for application to soil columns. Much of the following information is
summarized or abbreviated to allow the approach of the analysis to be understood.
Detailed descriptions of the experimental setup, procedures, and data collected can be
found in Young et al., [1999].
Application of the internal drainage method to a soil profile provides a means to
measure the unsaturated hydraulic conductivity function and other hydraulic
characteristics in situ. According to Hillel et al. [1972], the lot (irrigated plot) should be
large enough so that boundary effects are not experienced at the center of the plot. A
vertical neutron probe is installed within the plot to a depth usually exceeding two
meters to provide water content data. In proximity to the neutron probe (near enough to
measure the same soil mass), a series of tensiometers are installed at intervals not
exceeding 30 cm to provide tension data. The plot is then irrigated until effective
164
saturation is achieved, at which time irrigation stops and the plot is covered to prevent
evapotranspiration. As the redistribution process commences, measurements of water
content and tension are taken throughout the soil profile. Water content data can be
gathered using either time domain reflectometry (TDR) or neutron probe measurements.
Tensiometers are typically used to provide tension data. The water content and tension
measurements should be taken concurrently and frequently during the drainage period.
The resulting data are used to determine instantaneous potential gradients and fluxes
with in the soil profile.
The water content data for numerous locations throughout the irrigated plot were
graphed to see if there were any trends in the data. While the water content data
followed an overall trend of decreasing through time (at any given depth) during the
drainage period, there are specific points where the water content increases. These
variations were invariant with location and depth of measurement. Because of this
invariance, the trends were not attributed to heterogeneities in the material properties
but rather to calibration and measurement technique of the neutron probe, which would
affect the constants in the calibration polynomial. In addition, the manner in which the
standard counts are handled in computing the water contents can affect the variability in
the water content (see Analysis of Neutron Probe Data). An effort was made to
determine if some of this "noise" in the data could be removed by adjusting the
calibration values (i.e. averaging the standard counts). This resulted in marginally
better "fits" of regressions applied to the water content versus time data. No effort was
made to smooth the data by the process of elimination because there is no way to
165
ascertain which data points are "correct" or "incorrect." Following the advice of Ahuja
et al. [1980], regressions were applied to the water content and tension data to allow
values to be interpolated at specific times (i.e. to allow the tension and water contents to
be correlated at common points in time for calculating fluxes and gradients from the
regression curves). Regressions were applied to generate best-fit equations. The
equation with the highest correlation coefficient and a continuous first derivative was
chosen to interpolate the needed data.
The tension data for the NMI and SMI, east and west sides. Experiments I and
II, was plotted versus time to determine the most suitable data set to use in the method.
Experiment I data from the east side of the SMI was utilized for the internal drainage
method.
The flux for each depth and the cumulative flux for the entire drainage period
was calculated from the generated moisture content curves. The unsaturated hydraulic
conductivity for each depth through time was then calculated by dividing the fluxes by
the gradient. To evaluate whether the entire profile could be represented by a single
hydraulic conductivity curve, and to determine the functional relationship of the
hydraulic conductivity, the unsaturated hydraulic conductivity and the natural log of the
unsaturated hydraulic conductivity values was plotted against water content. A
regression was applied to the hydraulic conductivity values for all depths yielding
calculated saturated hydraulic conductivity's of 43.9 and 26.0 cm per day for two
locations within the plot. Regressions were also applied to the hydraulic conductivity
values for individual depths as well as smaller subsets of the entire data set with no
166
improvements in the fit. Examination of the functional relationship of hydraulic
conductivity revealed that it followed an exponential relationship as given by K(6) =
ae*^. These values are within the ranges foimd from laboratory tests run on soil cores
collected from the site. It should be recognized that these values are somewhat artificial
because of the large amount of smoothing of the measured data. The effect of
smoothing on the calculated hydraulic conductivity values is also observed and
considered by Ahuja et al., [1980].
167
REFERENCES
Abbaspour, K.C., M.T. van Genuchten, R. Schulin, and E. Schlappi, A sequential uncertainty domain inverse procedure for estimating subsurface flow and transport parameters, Water Resources Research, 33 (%), 1879-1892, 1997.
Adams, E.E., and L.W. Gelhar, Field study of dispersion in a heterogeneous aquifer 2. Spatial moments analysis. Water Resources Research, 28 (12), 3293-3307, 1992.
Ahuja, L.R., R.E. Green, and S.K. Chong, A Simplified Functions Approach for Determining Soil Hydraulic Conductivities and Water Characteristics In Situ, Water Resources Research, 16 (5), 947-953, 1980.
Aziz, K., and A. Settari,, in Petroleum Reservoir Simulation, pp. 395-401, Applied Science, Barking, United Kingdom, 1979.
Baker, F.G., P.L.M. Veneman, and J. Bouma, Limitations of the Instantaneous Profile Method for Field Measurement of Unsaturated Hydraulic Conductivity, Soil Science Society Of America Proceedings, 38 (6), 885-888, 1974.
Bronswijk, J.J.B., W. Hamminga, and K. Oostindie, Field-scale solute transport in a heavy clay soil. Water Resources Research, 31 (3), 517-526, 1995.
Celia, M.A., and P. Binning, A Mass Conservative Numerical Solution for Two-Phase Flow in Porous Media With Application to Unsaturated Flow, Water Resources Research, 28 (10), 2819-2828, 1992.
Cislerova, M., J. Simunek, and T. Vogel, Changes of Steady-State Infiltration Rates in Recurrent Ponding Infiltration Experiments, Journal of Hydrology, 104, 1-4, 1988.
168
Destouni, G., Prediction uncertainty in solute flux through heterogeneous soil. Water Resources Research, 28 (3), 793-801, 1992.
Eching, S.O., and J.W. Hopmans, Optimization of hydraulic functions from transient outflow and soil water pressure data. Soil Science Society of America Journal, 57(5), 1167-1175, 1993.
Eching, S.O., J.W. Hopmans, and W.W. Wallender, Estimation of in situ unsaturated soil hydraulic functions from scaled cumulative drainage data. Water Resources Research, 30 (8), 2387-2394, 1994.
Eching, S.O., J.W. Hopmans, and O. Wendroth, Unsaturated hydraulic conductivity from transient multistep outflow and soil water pressure data. Soil Science Society of America Journal, 58 (3), 687-695, 1994.
Ellsworth, T.R., W.A. Jury, F.F. Ernst, and P.J. Shouse, A three-dimensional field study of solute transport through unsaturated, layered, porous media 1. Methodology, mass recovery, and mean transport. Water Resources Research, 27 (5), 951-965, 1991.
Ellsworth, T.R., P.J. Shouse, T.H. Skaggs, J.A. Jobes, and J. Fargerlund, Solute transport in unsaturated soil: Experimental design, parameter estimation, and model discrimination. Soil Science Society of America Journal, 60 (2), 397-407, 1996.
Evett, S.R., and J.L. Steiner, Precision of neutron scattering and capacitance type soil water content gauges from field calibration. Soil Science Society of America Journal, 59 (4), 961-968, 1995.
Feyen, J., D. Jacques, A. Timmerman, and J. Vanderborght, Modelling water flow and solute transport in heterogeneous soils: A review of recent approaches, J. Agric. Eng. Res., 70(3), 231-256, 1998.
Forrer, I., R. Kasteel, M. Flury, and H. Fluhler, Longitudinal and lateral dispersion in an unsaturated field soil. Water Resources Research, 35 (10), 3049-3060, 1999.
169
Gardner, W.R., Field Measurement of Soil Water Diffiisivity, Soil Science Society Of America Proceedings, 34 (5), 832-833, 1970.
Gardner, W.R., D. Hillel, and Y. Benyamini, Post-Irrigation Movement of Soil Water: 1. Redistribution, Water Resources Research, 6 (3), 851-861, 1970.
Gardner, W.R., D. Hillel, and Y. Benyamini, Post-Irrigation Movement of Soil Water 2. Simultaneous Redistribution and Evaporation, Water Resources Research, 6 (4), 1148-1153, 1970.
Hammel, K., and K. Roth, Approximation of asymptotic dispersivity of conservative solute in unsaturated heterogeneous media with steady state flow. Water Resources Research, 34 (4), 709-715, 1998.
Hillel, D., and W.R. Gardner, Measurement of Unsaturated Conductivity and Diffusivity by Infiltration through an Impeding Layer, Soil Science, 109 (3), 149-153, 1970.
Hillel, D., V.D. Krentos, and Y. Stylianou, Procedure and Test of an Internal Drainage Method for Measuring Soil Hydraulic Characteristics in Situ, Soil Science, 114 (5), 395-400, 1972.
Hills, R.G., K.A. Fisher, M.R. Kirkland, and P.J. Wierenga, Application of flux-corrected transport to the Las Cruces trench site. Water Resources Research, 30 (8), 2377-2385, 1994.
Hills, R.G., D.B. Hudson, I. Porro, and P.J. Wierenga, Modeling One-Dimensional Infiltration into Very Dry Soils; 2. Estimation of the Soil Water Parameters and Model Predictions, Water Resources Research, 25 (6), 1271-1282, 1989.
Hills, R.G., I. Porro, D.B. Hudson, and P.J. Wierenga, Modeling One-Dimensional Inflltration into Very Dry Soils: 1. Model Development and Evaluation, Water Resources Research, 25 (6), 1259-1269, 1989.
Hills, R.G., P.J. Wierenga, D.B. Hudson, and M.R. Kirkland, The Second Las Cruces Trench Experiment: Experimental Results and Two-Dimensional Flow Predictions, Water Resources Research, 27 (10), 2707-2718, 1991.
170
Hopmans, J.W., Comparison of Various Methods to Scale Soil Hydraulic Properties, Journal of Hydrology, 93, 3-4, 1987.
Hopmans, J.W., and J.H. Dane, Temperature Dependence of Soil Water Retention Curves, Soil Science Society of America Journal, 50 (3), 562-567, 1986.
Jacques, D., J. Kim Dong, J. Diels, J. Vanderborght, H. Vereecken, and J. Feyen, Analysis of steady state chloride transport through two heterogeneous field soils, Water Resources Research, 34 (10), 2539-2550, 1998.
Jacques, D., A. Timmerman, and J. Feyen, Measuring water flow and solute transport under natural boundary conditions in a loamy soil profile, Annales Geophysicae, 16 {2), 479, 1988.
Jacques, D., J. Vanderborght, D. Mallants, J. Kim Dong, H. Vereecken, and J. Feyen, Comparison of three stream tube models predicting field-scale solute transport, Hydrology and Earth System Sciences, I (4), 873-893, 1997.
James, R.V., and J. Rubin, Transport of Chloride Ion in a Water Unsaturated soil Exhibiting Anion Exclusion, Soil Science Society of America Journal, 50, 1142-1150, 1986.
Jury, W.A., W.R. Gardner, and W.H. Gardner, Soil Physics, 328 pp., John Wiley &. Sons, Inc., 1991.
Kool, J.B., and J.C. Parker, Development and Evaluation of Closed-Form Expressions f o r H y s t e r e t i c S o i l H y d r a u l i c P r o p e r t i e s , W a t e r R e s o u r c e s R e s e a r c h , 2 3 ( I ) , 105-114, 1987.
Kool, J.B., J.C. Parker, and M.T. Van Genuchten, Determining soil hydraulic properties from one-step outflow experiments by parameter estimation. I. Theory and numerical studies. Soil Science Society of America Journal, 49 (6), 1348-1354, 1985.
171
Kool, J.B., J.C. Parker, and M.T. Van Genuchten, Parameter Estimation for Unsaturated Flow and Transport Models - A Review, Journal of Hydrology, 91 (3), 255-293, 1987.
Land, C.S., Calculation of imbibation relative permeability for two- and three-phase flow from rock properties. Trans. AIME, 243, 149-156, 1968.
Lehmann, F., and P. Ackerer, Determining soil hydraulic properties by inverse method in one-dimensional unsaturated flow, 7. Environ, ^ai, 26 {I), 76-81, 1997.
Libardi, P.L., K. Reichardt, D.R. Nielsen, and J.W. Biggar, Simple Field Methods for Estimating Soil Hydraulic Conductivity, Soil Science Society of America Journal, 44 {\), 3-7, 1980.
Mace, A., D.L. Rudolph, and R.G. Kachanoski, Suitability of Parametric Models to Describe the Hydraulic Properties of an Unsaturated Coarse Sand and Gravel, Ground Water, 36 (3), 465-475, 1998.
Mallants, D., D. Jacques, P.H. Tseng, M.T. Van Genuchten, and J. Feyen, Comparison of three hydraulic property measurement methods. Journal of Hydrology Amsterdam, 199, 3-4, 1997.
Mallants, D., P. Mohanty Binayak, D. Jacques, and J. Feyen, Spatial variability of hydraulic properties in a multi-layered soil profile. Soil Science, 161 (3), 167-181, 1996.
Millington, R.J., and J.M. Quirk, Permeability of Porous Solids, Trans. Faraday Soc., 57, 1200-1207, 1961.
Mitchell, R.J., and A.S. Mayer, The significance of hysteresis in modeling solute transport in unsaturated porous media. Soil Science Society of America Journal, 1506-1512, 1998.
172
Moldrup, P., T. Yamaguchi, D.E. Rolston, and J.A. Hansen, Predicting wetting front advance in soils using simple laboratory derived hydraulic parameters. Water Resources, 30(6), 1471-1477, 1996.
Mualem, Y., A New Model for Predicting the Hydraulic Conductivity of Unsaturated Porous Media, Water Resources Research, 12 (3), 513-522, 1976.
Mualem, Y., A modified dependent domain theory of hysteresis. Soil Science, 137, 283-291, 1984.
Padilla, I., Y., T.C.J. Yeh, and M. Conklin, H., The effect of water content on solute transport in unsaturated porous media. Water Resources Research, 55 (11), 3303-3313, 1999.
Parker, J.C., J.B. Kool, and M.T. Van Genuchten, Determining soil hydraulic properties from one-step outflow experiments by parameter estimation. II. Experimental studies. Soil Science Society of America Journal, 49 (6), 1354-1359, 1985.
Parker, J.C., and M.T. Van Genuchten, Flux-Averaged and Volume-Averaged Concentrations in Continuum Approaches to Solute Transport, Water Resources Research, 20(7), 866-872, 1984.
Parlange, J.Y., Simple Method for Predicting Drainage From Field Plots, Soil Science Society of America Journal, 46 (4), 887-888, 1982.
Pickens, J.F., and R.W. Gillham, Finite Element Analysis of Solute Transport under Hysteretic Unsatiu^ted Flow Conditions, Water Resources Research, 16 (6), 1071-1078, 1980.
Porro, I., P.J. Wierenga, and R.G. Hills, Solute Transport Through Large Uniform and Layered Soil Columns, Water Resources Research, 29 {A), 1321-1330, 1993.
Resources, A.D.o.W., Arizona Water Resources Assessment, Vol. 2, Hydrologic Summary, 243 pp., Arizona Department of Water Resources, 1993.
173
Rockhold, M.L., R.E. Rossi, and R.G. Hills, Application of similar media scaling and conditional simulation for modeling water flow and tritium transport at the Las Cruces trench site. Water Resources Research, 32 (3), 595-609, 1996.
Roth, K., W.A. Jury, H. Fluhler, and W. Attinger, Transport of chloride through an unsaturated field soil. Water Resources Research, 27 (10), 2533-2541, 1991.
Russo, D., Determining Soil Hydraulic Properties by Parameter Estimation: On the selection of a Model for the Hydraulic Properties, Water Resources Research, 24 (3), 453-459, 1988.
Scott, P.S., G.J. Farquhar, and N. Kouwen, Hysteretic effects on net infiltration. Advances in infiltration. Am. Soc. of Agric. Eng., Publ. 11-83, 163-170, 1983.
Simunek, J., M. Sejna, and M.T. van Genuchten, The HYDRUS-ID Software Package for Simulating the One-Dimensional Movement of Water, Heat, and Multiple Solutes in Variably-Saturated Media, Version 2.0, 178 pp., U.S. Salinity Laboratory, USDA, ARS, Riverside, CA, 1998.
Sisson, J.B., Drainage from Layered Field Soils: Fixed Gradient Models, Water Resources Research, 2i (11), 2071-2075, 1987.
Sisson, J.B., A.H. Ferguson, and M.T. van Genuchten, Simple Method for Predicting Drainage from Field Plots, Soil Science Society of America Journal, 44 (6), 1147-1152, 1980.
Sisson, J.B., W.M. Klittich, and S.B. Salem, Comparison of Two Methods for Summarizing Hydraulic Conductivities of a Layered Soil, Water Resources Research, 24 (S), 1271-1276, 1988.
Sisson, J.B., and M.T. Van Genuchten, An improved analysis of gravity drainage experiments for estimating the unsaturated soil hydraulic functions. Water Resources Research, 27 (4), 569-575, 1991.
174
Toorman, A.F., P.J. Wierenga, and R.G. Hills, Parameter Estimation of Hydraulic Properties from One-Step Outflow Data, Water Resources Research, 25 (II), 3021-3028, 1992.
Toride, N., F.J. Leij, and M.T. van Genuchten, The CXTFIT Code for Estimating Transport Parameters from Laboratory or Field Tracer Experiments, Version 2.1, Research Report No. 137, 121 pp., U.S. Salinity Laboratory, USDA, ARS, Riverside, CA, 1995.
van Genuchten, M.T., A Closed-Form Equation for Predicting the Hydraulic Conductivity of Unsaturated Soils, Soil Science Society of America Journal, 44 (5), 892-898, 1980.
van Genuchten, M.T., and W.J. Alves, Analytical Solutions of the One-Dimensional Convective-Dispersive Solute Transport Equation, Tech. Bulletin 1661, 151 pp., U.S. Department of Agricultural, 1982.
van Genuchten, M.T., F.J. Leij, and S.R. Yates, The RETC Code for Quantifying the Hydraulic Functions of Unsaturated Soils, 93 pp., U.S. Salinity Laboratory, USDA, ARS, Riverside, CA, 1991.
Vanderborght, J., C. Gonzalez, M. Vanclooster, D. Mallants, and J. Feyen, Effects of soil type and water flux on solute transport. Soil Science Society of America Journal, 61 (2), 372-389, 1997.
Vanderborght, J., D. Jacques, and J. Feyen, Deriving Transport Parameters from Transient Flow Leaching Experiments by Approximate Steady-State Flow Convection Dispersion Models, Soil Science Society of America Journal, 64 (4), 1317-1327, 2000b.
Vanderborght, J., A. Timmerman, and J. Feyen, Solute Transport For Steady-State and Transient Flow in Soils with and without Macropores, Soil Science Society of America Journal, 64 (4), 1305-1317, 2000a.
175
Vogel, T., K. Huang, R. Zhang, and M.T. van Genuchten, The HYDRUS code for simulating one-dimensional water flow, solute transport, and heat movement in variably-saturated media. Version 5.0, Research Report No. 140, U.S. Salinity laboratory, USD A, ARS, Riverside, CA, 1996.
Warrick, A.W., P.J. Wierenga, M.H. Young, and S.A. Musil, Diurnal fluctuations of tensiometric readings due to surface temperature changes. Water Resources Research, 34 (11), 2863-2869, 1998.
Watson, K.K., An instantaneous profile method for determining the hydraulic conductivity of unsaturated porous materials, fVater Resources Research, 2, 709-715, 1966.
Watson, K.K., V.A. Sardana, and G.C. Sander, Comparison of analytical and numerical results for constant flux inflltration, J. Hydrol. Amst., 165, 1-4, 1995.
Wierenga, P.J., R.G. Hills, and D.B. Hudson, The Las Cruces Trench Site: Characterization, Experimental Results, and One-Dimensional Flow Predictions, Water Resources Research, 27 (10), 2695-2705, 1991.
Wierenga, P.J,, and M.T, Van Genuchten, Solute Transport Through Small and Large Unsaturated Columns, Ground Water, 27 (1), 35-42, 1989.
Wosten, J.H.M., and M.T. Van Genuchten, Using Texture and Other Soil Properties to Predict the Unsaturated Soil Hydraulic Functions, Soil Science Society of America Journal, 52 (6), 1762-1770, 1988.
Yates, S.R., M.T. van Genuchten, A.W. Warrick, and F.J. Leij, Analysis of measured, predicted, and estimated hydraulic conductivity using the RETC computer program. Soil Science Society of America Journal, 56 (2), 347-354, 1992.
Young, M.H., P.J. Wierenga, A.W. Warrick, L.L. Hofinann, S.A. Musil, M. Yao, C.J. Mai, Z. Zou, and B.R. Scanlon, Results of field studies at the Maricopa Environmental Monitoring Site, Arizona, NUREG/CR-5694, 260 pp., U.S. Nuclear Regulatory Commission, Washington, D.C., 1999.
176
Zarba, R.L., E.T. Bouloutas, and M.A. Celia, General Mass-Conservative Numerical Solution for the Unsaturated Flow Equation, Water Resources Research, 26 (7), 1483-1496, 1990.
Zhang, R., and M.T. Van Genuchten, New models for unsaturated soil hydraulic properties. Soil Science, 158 (2), 77-85, 1994.