E”
a.
: 3‘ L

LE... .1

.1. 1.3.1
":23.
.

km
"3

E. .1.

. . ‘ . .1 A,
11 \ﬂhnud.xvs.nvf§ I...
I» «kc. l...
39).: 1..

 

‘ . 4.7:): ¢ .
155:: (1.1;...
1’ . )1. .3
dt .3”

P...

. u 5.5!. . 9\ ‘0
3.31.5.12345‘

.a. It?! 1&1 {$431.}
.12... ii... } .lti

. «I. 1..

«ENE

huh" .4. ‘ ‘
. 1

. 5!}; if b.2211.

 

LIBRARY
Michigan §tate
University

This is to certify that the
dissertation entitled

EXPLORING THE INFLUENCE OF LAND-USE AND CLIMATE ON
REGIONAL HYDROLOGY AND GROUNDWATER RECHARGE

presented by
DUSHMANTHA HELAPRIYA JAYAWICKREME

has been accepted towards fulﬁllment
of the requirements for the

Ph.D. degree in Geological Sciences

 

 

Ma70r Professor’s Signature
12 , I I 4,008

Date

 

MSU is an Afﬁrmative Action/Equal Opportunity Employer

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/06 K:IProj/Acc&Pres/CIRCIDateDue.indd

 

EXPLORING THE INFLUENCE OF LAND-USE AND CLIMATE ON REGIONAL
HYDROLOGY AND GROUNDWATER RECHARGE

By

Dushmantha l-lelapriya Jayawickreme

A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Geological Sciences

2008

ABSTRACT

EXPLORING THE INFLUENCE OF LAND-USE AND CLIMATE ON REGIONAL
HYDROLOGY AND GROUNDWATER RECHARGE

By

Dushmantha Helapriya Jayawickreme

Global energy, water, and bio-geochemical cycles are strongly linked to
land-use and land-cover characteristics. Land-use and land-cover is also a
component of the environment extensively impacted and continuously altered by
human activity. Population growth, food, energy and other needs coupled with
human ingenuity has greatly altered and will continue to change the terrestrial
biosphere across the globe. With additional concerns for signiﬁcant and
widespread land-use/land-cover transformations due to global climate change, a
need to better understand the effects of these transformations on environmental

systems from local, regional, to global scale has emerged in the recent decades.

In this dissertation I have investigated the impacts of land-use and land-
cover (i.e. vegetation) on groundwater recharge, which is a critical component of
the hydrologic cycle. The dependence of people and many sensitive ecosystems
around the world on groundwater alone warranted a closer look at how changing
land-use/cover and climate are affecting the quantity and quality of groundwater.
By evaluating streamﬂow, climate, land-cover, and other attributes in Michigan’s

watersheds we showed that intense agriculture reduced summer time

streamﬂow, hence groundwater recharge in such watersheds. By quantifying
baseﬂow discharges relative to precipitation in July-September peak growing
season over multiple years we found that recharge in primarily agricultural (>70
agricultural uses by land area) watersheds was only one third of the recharge in

watersheds that are mix use (<50% agricultural uses).

To better grasp how land-cover; speciﬁcally vegetation, vegetation
differences, and vegetation dynamics affect recharge we adopted geophysical
techniques, a step beyond the traditional uses of geophysical methods as well as
an unorthodox approach to terrestrial ecosystem investigations. It was
hypothesized that there would be observable differences in the way vegetation
interacts with the shallow subsurface and such interactions could be quantiﬁed
with geophysical methods. Based on multi-year geophysical monitoring of soil
moisture at a forest-grassland ecotone we found large seasonal and long-term
differences in the way vegetation affects groundwater recharge as well as
shallow groundwater environments. Apart from water use differences, we show
that soil temperature as well as salt dynamics even at very local scales are
affected by vegetation differences. For example we observed that forests in
shallow groundwater regions are likely to increase groundwater salinity
compared to grasslands in similar settings. These ﬁndings contribute to
developing greater insights into the functioning of the natural environment and
how anthropogenic forcings through land-use change may imperil or help protect

the health of hydrologic systems in a range of regions.

ACKNOWLEDGEMENTS

First and foremost I express my sincere gratitude to Dr. David Hyndman,
my dissertation advisor, from whom I received knowledge, guidance, and
unwavering support throughout my time as a graduate student at Michigan State
University. Members of my committee, Dr. Phanikumar Mantha, Dr. Jiaguo Qi
offered valuable advice and their time at many occasions. Their insights have
vastly improved the quality of my work and shaped my scientiﬁc thinking. Dr.
Remke van Dam, my fourth committee member generously poured his time and
knowledge in to this work. It is with his effort and many time consuming trips to
ﬁeld sites especially when I couldn’t keep up with it that I was able to complete

my dissertation research.

During the ﬁve years of my graduate work many others contributed to the
successful outcome l have experienced with this research. Dr. Jeff Andreasen,
Dr. Dennis Gilliland, Dr. Alvin Smucker, and Dr. Jan Stephenson are thanked for
their support with data and ﬁeld equipment. Dr. Bryan Pijanowski, Dr. Jim Butler,
and Dr. Jan Hendirx are acknowledged for their support with manuscripts and
presentations. Michael Mores, Matt Malkowski, Samer Hariri, Cheryl Kendall,
Matt Parsons, Dave Szymanski, Abby Norton, Randy Klevikas, and Paul Bloes
provided many hours of arduous ﬁeld and lab support. Without their help very

little would have been accomplished.

Karen Tefend and Anthony Kendal became the most generous of hosts
and best of friends at times that were the most difﬁcult. Along with them Colleen
McLean, Nick Welty, Amy Lansdale, and many others are greatly acknowledged

for their years of friendship.

Udeni, my wife and our daughter Hiru made the most sacriﬁces and
provided utmost support so that I could see the end and reach a goal I had set

for myself some years ago. For them I dedicate this dissertation.

TABLE OF CONTENTS

List of Tables ...................................................................................................... viii
List of Figures ....................................................................................................... x
Chapter One: General Introduction ....................................................................... 1
Objective .................................................................................................... 6
References ................................................................................................ 9

Chapter Two: Evaluating the inﬂuence of land cover on seasonal water budgets

usingNext Generation Radar (NEXRAD) rainfall and streamﬂow data ............... 15
Abstract .................................................................................................... 16
Introduction .............................................................................................. 1 7
Approach ................................................................................................. 19
Study Sites ............................................................................................... 23
Evaluation of NEXRAD Rainfall Data ....................................................... 25
Results and Discussion ............................................................................ 33
Summary and Conclusions ...................................................................... 48
Acknowledgments .................................................................................... 51
References .............................................................................................. 53

Chapter Three: Subsurface imaging of vegetation, climate, and root-zone

moisture interactions .......................................................................................... 57
Abstract .................................................................................................... 58
Introduction .............................................................................................. 59
Methods ................................................................................................... 61
Results ..................................................................................................... 65
Conclusions ............................................................................................. 70
Acknowledgements .................................................................................. 71
References .............................................................................................. 72

vi

Chapter Four: Integrating geophysics and hydrologic models to evalueate land-

use impacts on groundwater resources ..................................................... 75
Introduction .............................................................................................. 75
Study sites ............................................................................................... 79
Field observations .................................................................................... 83
Electrical Resistivity data acquisition and processing .............................. 86
Electrical resistivity and soil moisture ....................................................... 94
Soil Moisture Dynamics and Relevance to Recharge .............................. 95
Hydrologic simulations ............................................................................. 99
Discussion and Conclusion .................................................................... 112
References ............................................................................................ 1 14

Chapter Five: Evaluating a range of approaches to derive hydrogeological

information from electrical resistivity data ......................................................... 118
Introduction ............................................................................................ 1 18

. Electrical Resistivity Imaging (ERI) ........................................................ 120
ERI and environmental variability .......................................................... 122
Electrical conductivity and volumetric soil moisture ............................... 122
Soil temperature and electrical conductivity ........................................... 128
The effect of pore-water conductivity ..................................................... 132
ERI and time-lapse differencing ............................................................. 137
Soil moisture estimates with resistivity ratios ......................................... 144
Conclusions ........................................................................................... 146
References ............................................................................................ 148

vii

List of Tables

Table 2-1. Spearman’s Ranked Correlation Statistics for Mean July to September
(2002-2004) Volume Ratios and Other Watershed Variables for the 40 Selected
Watersheds. Statistical signiﬁcance (p<0.05) is indicated by bold typeface. ...... 35

Table 2-2. Spearman's Nonparametric Correlation Statistics for Land Cover and
Other Watershed Variables for the 40 Selected Watersheds. Statistical
signiﬁcance is indicated by bold typeface. .......................................................... 39

Table 2-3. Multiple Regression Statistics for July September Mean (2002-2004)
Streamtloszainfall and Base F loszainfall Ratios and Watershed Attributes.
The residuals from both ﬁts were tested for normality with the Anderson-Darling
test. Normality was not rejected; p values are 0.27 and 0.35, respectively. ....... 40

Table 24. Estimated Annual Growing Season Recharge Amounts (July-
September); in terms of annual rainfall and actual amounts in centimeters. ...... 42

Table 4-1. Estimated soil hydraulic parameters for the three soil layers in the
forest with standard errors of regression coefﬁcients (S.E. Coeff) and conﬁdence
intervals. ........................................................................................................... 106

Table 4-2. Estimated soil hydraulic parameters for the three soil layers in the
grassland. ......................................................................................................... 106

Table 4-3. Sensitivity of simulated cumulative transpiration Z T and bottom
pressure head (h) in the forest to different root distributions. ERI Root is root
distribution estimated with resistivity data, deRosany model used with a
distribution coefficient of -0.9 (de Rosnay and Polcher 1998). ......................... 110

viii

Table 4-4. Sensitivity of simulated evaporation (2E), transpiration (Z T), and
model bottom pressure head (h) in the forest to a 110% adjustment in root water
abstraction parameters (Feddes et al. 1974). ZE, and Z T are given as percent
changes relative the base model. h1-minimum pressure head for active
transpiration, h2-pressure- head for maximum transpiration efﬁciency, h3-limiting
pressure head where transpiration ceases to occur at maximum efﬁciency, and
h4-wilting point .................................................................................................. 110

Table 4-5. Sensitivity of cumulative inﬁltration (ZI), transpiration (Z T),
evaporation (2E), and bottom pressure head (h) in the forest to a t10%
adjustment in soil parameters. 2|, 2 T, and {E shown are percent changes
compared to the base model with optimized soil hydraulic parameters. ........... 111

Table 5-1. Parameters for Archie’s equation from laboratory measurements of
soils at the study site. ps is obtained from ﬁeld resistivity data (for sand estimated
with resistivity data from below the water table, and for clay estimated from
datasets collected when the site was extremely wet). ...................................... 124

List of Figures

Figure 2-1. Locations of our study watersheds (shaded) with NEXRAD radar and
rain gauge locations. Dashed lines show the effective range of the radar
detectors. ............................................................................................................ 23

Figure 2-2. RMS differences between NEXRAD and ground-based precipitation
measurements at various sites with data spanning the 2002, 2003, and 2004
growing seasons ................................................................................................. 27

Figure 2-3. Map of the percent absolute differences (calculated relative to
observed gauged precipitation) between the 2004 (April-November) gauge and
NEXRAD rainfall data. The Grand Rapids, Lansing, Flint, and Alpena gauges are
known to be used by the WSR-88D system for real-time calibration of radar
rainfall estimates (Ann Arbor and Grayling are missing 1 month of data, Harbor
Beach is missing 2 moths, and Montague is missing 3 months, and thus these
months were not used in this analysis. ............................................................... 30

Figure 2-4. RMS differences for eight gauge locations using 2002, 2003, and
2004 monthly radar and ground-based gauge precipitation data. May RMSD
calculation is based only on data from 2002 and 2004 due to missing NEXRAD
grid data in MAY 2003. ....................................................................................... 31

Figure 2-5. Comparison of monthly (July-September) observed gauge and
NEXRAD rainfall for 2002, 2003, and 2004. RMSE and bial (average NEXRAD
rainfall-average observdd gauge rainfall) are given in mm. ................................ 33

Figure 26 Percentage of agricultural land cover in watersheds compared to
streamﬂowzrainfall ratios for July to September periods of 2002-2004. Each data
point represents a watershed, and three watersheds with greater than 30% clay
sediments are marked with a gray oval. ............................................................. 34

Figure 2-7. Average 2002-2004 quarter year base ﬂowzdrainage area rations in
high-intensity (>70°/o) and moderate-intensity (<50%) agricultural watersheds.
The difference of the ﬂow between the two types of watersheds are statistically
signiﬁcant (p value <0.05) in all quarters except January-March. ....................... 42

Figure 2-8. Average of 2002-2004 January to May streamﬂowzdrainage area
ratios (cm). Higher values likely indicate greater snowmelt inﬂuence on the
watershed hydrology. The contour map was generated by kriging the ratio
calculated from data at the 113 USGS streamﬂow gauges shown on the map.
Images in this dissertation are presented in color ............................................... 43

Figure 2-9. The 2002-2004 monthly base ﬂowzdrainage area ratios and average
monthly rainfall for 40 selected watersheds in Michigan. The peak growing
season is clearly marked by higher leaf area indexes (LAI), which corresponds
well with the decline in stream base ﬂow during the same period. ..................... 45

Figure 2-10. Measured water table depths in two shallow wells in the northern
portion of Michigan’s Lower Peninsula (see Figure 2-1for locations). A steady
decline in water level is evident during the growing season (land use percentages
in the watershed containing B12 are forests ~65%, shrubs/open land ~35%; G13,
forests ~66%, agriculture ~28%). Note that the minimum water level is delayed
signiﬁcantly from the peak LAI and minimum base ﬂow levels shown in Figure
2-9. ..................................................................................................................... 46

Figure 2-11. Recharge estimates for the July to September period, as a percent
of NEXRAD bias adjusted rainfall in high-intensity agricultural (>70%) and
moderate-intensity agricultural (<50%) watersheds. The difference in
baseﬂowzrainfall ratios between the two types of watersheds are statistically
signiﬁcant (p value <0.05) in all years ................................................................. 48

Figure 2-12. Total monthly streamﬂow, and overland ﬂow volumes normalized by
watershed area along with rainfall averaged across the 40 study watersheds. .. 49

Figure 3-1. Soil moisture from probes at the site and the average of the 20 and
86.5 cm depth ERI moisture estimates for all datasets corrected with site
temperatures. Precipitation data is shown for a nearby gage. The color bar
shows the state of the forest canopy, vertical lines mark data collection dates,
and shaded areas above the bar are differential inversion periods for Figure 3-2.
Images in this dissertation are presented in color ............................................... 63

Figure 3-2. Differential resistivity panels for approximately one-month periods. a)
Early fall (10/18 — 11/15), b) late fall (11122 - 12/20), c) winter (1/5 — 2/9), d)
spring (3/16 - 4/20), and e) summer (6/10 - 7/2). Trees are not to scale; 80 cm of
relief along the array is included in the inversion. Red triangles locate moisture
probes and temperature arrays. A decrease in resolution with depth and
associated smoothing artifacts can cause resistivity differences below the water
table. Images in this dissertation are presented in color. .................................... 66

xi

Figure 3-3. Spatial and temporal distribution of soil moisture estimated across the
ecotone. Images in this dissertation are presented in color. ............................... 69

Figure 3-4. Laterally averaged changes in soil moisture below the forest (36-51
m) and the grassland (92-107 m) between early-growing (April to early May) and
peak-growing periods (July to August). Images in this dissertation are presented
in color. ............................................................................................................... 71

Figure 4-1. Locations of East Lansing and Traverse City, Michigan ﬁeld sites
where soil moisture dynamics were imaged with ERI. ........................................ 79

Figure 4-2. East Lansing ﬁeld site. A-B: electrode array with 84 electrodes at
1.5m separation. E3, E26, E70 are groundwater observation wells.
Temperature/soil moisture sensor locations have moisture probes at 20 & 80 cm
depth. Temperature probes are at 5, 10, 20, 40, 80,117, and 147cm depth at
eachlocaﬁon. ..................................................................................................... 81

Figure 4-3. Traverse City study site. A-B: electrode array with 84 electrodes at
1.5m separation. B3 is the groundwater observation well. Moisture sensor
locations shown have probes at 20 and 80 cm depths, and temperature sensors
are at 5, 10, 20, 40, 80, 117, and 147 cm depths. .............................................. 82

Figure 4-4. Groundwater observations since October 2007 at the East Lansing
study site. Depth to water from the surface is shown for each well. Since the
surface elevation at the forest well (E3, Figure 4-2) is 99 cm lower than the
surface elevation at the grassland well the two vertical axis are offset by a similar
amount. The wells are approximately 90 meters apart. ...................................... 84

Figure 4-5. Groundwater observations at the Traverse City study site. Annual
recharge-discharge cycles related to the growing season and spring snowmelt
are evident in this well record. ............................................................................ 85

Figure 4-6. Soil moisture measurements below forest and the grassland at the
East Lansing site. ............................................................................................... 86

Figure 4-7. Cross-sectional view of the East Lansing study site with a
representative late-summer subsurface resistivity distribution. Images in this
dissertation are presented in color ...................................................................... 88

xii

Figure 4-8. Cross—sectional view of the Traverse City study site with a
representative late-summer subsurface resistivity distribution. Images in this
dissertation are presented in color ...................................................................... 89

Figure 4-9. The temporal variability of subsurface (0-6m, surface to below
seasonal water table) resistivity distribution below the forest and the grassland at
the East Lansing site on selected dates. Mean, 25th & 75th percentiles, and the
range for each date are shown. .......................................................................... 91

Figure 4-10. Mean resistivity values in the 0-2m and 2-4m depths through time.
Similar resistivities in the 2-4m range during the latter part of the study period
shown coincide with higher observed groundwater elevations at the research
site. ..................................................................................................................... 93

Figure 4-11. ERI estimated soil moisture and the observed moisture in the East
Lansing study site (Forest). The arithmetic average of the 20 and 80cm cm
moisture observations are shown. ...................................................................... 95

Figure 4-12. ERI calculated transient soil moisture (20-80cm cm arithmetic
average) in the grassland and the forest. Higher moisture levels in the grassland
are observed on each of the ERI data collection dates. ..................................... 96

Figure 4-13. Temporal soil moisture changes in ~15 wide zones below the
grassland and the forest. Cooler colors indicate moisture increases and warmer
colors are moisture decreases. C is the main fall recharge period, followed by
spring snowmelt (D). A, B, and E are growing season moisture deﬁcits due to
plant water use. Images in this dissertation are presented in color. ................... 98

Figure 4-14. A simpliﬁed illustration of the soil texture and stratigraphy (from soil
cores) in the forest and the grassland. Different soil texture units identiﬁed in ﬁeld
soil cores were combined to three representative layers in the vadose zone ﬂow
model (right). Images in this dissertation are presented in color ....................... 100

Figure 4-15. Conceptual framework for integrating geophysical soil moisture
estimates with hydrologic modeling to estimate groundwater recharge. Soil
parameters 63, a, n, Ks are optimized in stage 2, and plant water stress function
parameters h1, h2, h3, h4 , are calibrated in stage 3. ...................................... 101

Figure 4-16. Laterally averaged changes in soil moisture below the forest and the
grassland between early-growing (April to early May) and peak-growing periods
(July to August) (from Jayawickreme et al., 2008). ........................................... 103

xiii

Figure 4-17. Cumulative precipitation throughfall (a), representative leaf area
indexes (b) from MODIS, calculated cumulative potential transpiration (c) and soil
evaporation (d) inputs for hydrologic simulations from ILHM. ........................... 104

Figure 4-18. Simulated vs. ERI estimated soil moisture in the forest. Data is from
10 depths within the ﬁrst 3m of the subsurface ................................................. 105

Figure 4-19. Simulated vs. ERI estimated soil moisture in the grassland with data
from nine depths in the ﬁrst 3m of the subsurface. ........................................... 105

Figure 4-20. Measured and simulated groundwater table in the forest with
optimized soil hydraulic parameter sets in the 1D-Hydrus model. These
observation data were not used in the optimization for hydraulic parameters in
the forest. ......................................................................................................... 107

Figure 4-21. Measured and simulated groundwater table in the grassland with
optimized soil hydraulic parameter. .................................................................. 107

Figure 4-22. Observed (ERI calculated) vs. simulated moisture in the forest
during summer (June-October, 2007) ............................................................... 109

Figure 4-23. Observed vs. simulated soil moisture in the grassland from June to
October, 2007. .................................................................................................. 109

Figure 5-1 Electrode positions for Wenner (top) and Dipole-dipole (bottom)
electrode conﬁgurations. A, B are current electrodes and M, N are potential
electrodes. ‘a’ is the distance between electrodes used for measurements, and n
is an integer. ..................................................................................................... 121

Figure 5-2. Data coverage for Wenner and dipole-dipole electrode conﬁgurations
for an 84 electrode array with 1.5m electrode separations. Higher data density
near the electrode positions is obtained with the dipole-dipole conﬁguration
compared to Wenner, where data coverage is evenly distributed. 1453 and 1134
measurements are obtained with these dipole-dipole and Wenner arrays
respectively ....................................................................................................... 1 21

Figure 53 Soil moisture estimates from ERI (dipole-dipole) data compared to
other independent measurements of soil moisture at the study site. ................ 125

xiv

Figure 5-4. The sensitivity of ERI derived saturation to m, and p8 estimates. The
high (grassland-98 Ohm-m) and low (forest-65 Ohm-m) values for saturated
resistivity were derived from ﬁeld resistivity data. Some of the differences in
saturated resistivity between the grassland and forest are due to observed
groundwater conductivity differences. .............................................................. 126

Figure 5-5. Water table elevation observations and resistivity measurements in
the forest (a), grassland (b) from November 2007 to June 2008. The correlation
between water table elevation and contours of the resistivity estimates is much
clearer in the forest relative to the grassland. Images in this dissertation are
presented in color. ............................................................................................ 127

Figure 5-6. Seasonal grassland soil temperature ﬂuctuations at the study site.
Inset shows diurnal soil temperature differences between the grassland and the
forest stand over a span of three days in June. ................................................ 129

Figure 5-7. The inﬂuence of temperature on resistivity and the resulting
difference in calculated soil moisture with Archie’s equation. ........................... 130

Figure 5-8. The inﬂuence of m on soil resistivity-temperature dependence at two
different soil temperatures (28°C and 12°C). The sensitivity to m is important at
higher temperature ranges. .............................................................................. 131

Figure 5-9. Percent change in resistivity from August 10, 2007 to January 11,
2008 without (a) and with (b) temperature correction. Prominent changes in the
grassland (a) do not appear once the temperature correction is added (b).
Images in this dissertation are presented in color ............................................. 131

Figure 5-10. Groundwater conductivity measured at the research site in the
grassland and forest observation wells (a). Pore-water and bulk electrical
resistivity relationship calculations based on Archie’s equation for various
degrees of saturation (b). A soil with 30% porosity, ﬁtting constant, m=1.16 and
an n value of 2, characteristic of sandy soils, was used in the calculation ........ 133

Figure 5-11. Potential inﬂuence of groundwater conductivity differences below
the grassland and forest. Higher resistivity values below water table (~4 m from
surface) in the grassland is likely due to the lower measured groundwater
conductivities in the grassla grassland. Data shown is temperature corrected and
the soils are similar on either side of the ecotone (the forest-grassland boundary).
Images in this dissertation are presented in color ............................................. 135

XV

Figure 5-12. Measured bulk electrical resistivity difference between the grassland
and the forest (pgrassland-pforest). Higher resistivities below the water table in

the grassland are consistent with the lower measured groundwater conductivities
there. The shaded area indicates the approximate range of water table change in
the grassland between March and June 2008. ................................................. 136

Figure 5-13. Data characteristics of three resistivity data sets from the study site,
which were used for difference inversion method analysis. Images in this
dissertation are presented in color .................................................................... 139

Figure 5-14. Comparisons of base dataset effects. B1, 32, and BB are three
separate inverted ﬁeld datasets. The notation B1>>B3 refers B3 calculation with
B1 as the base dataset in the difference inversion. Only the data from the top 5m
were selected for the analysis. ......................................................................... 141

Figure 5-15. Comparison of computed B1, 82, B3 with different with respect to
others in the difference inversion. ..................................................................... 142

Figure 5—16. Spatial distribution of differences from difference inversions with B1,
32, and BS datasets. Correlations between each dataset are shown in Figure
5-15. The choice of base dataset has a larger impact on the resistivity based
interpretations in the forest. Images in this dissertation are presented in color. 143

Figure 5-17. Laboratory measure resistivity and soil water contents for soil
samples collected from multiple locations at the study site. Images in this
dissertation are presented in color .................................................................... 144

Figure 5-18. Soil moisture ratios observed with point gauges and soil moisture
ratios calculated with ERI near the location where point gauges are located. .. 145

xvi

Chapter One

General Introduction

Land-use and land cover (LULC) changes have important consequences
on the functioning of regional to global environmental systems. Signiﬁcant land
use changes may seriously impact a region’s water resources by altering surface
water, groundwater, and soil moisture patterns, and perturb groundwater
recharge processes (Zhang et al. 2001; Walker et al. 2002; Jayawickreme and
Hyndman 2007; Pijanowski et al. 2007) and water quality (Lenat and Crawford
1994; Paul and Meyer 2001; Tong and Chen 2002; Wayland et al. 2003). LULC
also has a strong inﬂuence on land-atmosphere interactions and energy
exchange characteristics at the surface-atrnosphere boundary layer with global
climate consequences. Land-use characteristics affect the partitioning of surface
incident energy into latent and sensible heat components. Alterations to these
surface ﬂuxes of water vapor and heat can affect regional atmospheric
temperatures, precipitation, and other climate variables (Pitman et al. 2004;
Pielke 2005). Local- to regional-scale land-use changes and global climate
teleconnections are also well documented (Marland et al. 2003; Koster et al.

2004; Werth and Avissar 2005).

Continued global population growth and demand for food and energy
security are likely to locally accelerate the rate of LULC change in many parts of
the world. Additionally, regional scale changes will be brought about by an

increasingly warming global climate with elevated atmospheric COz

concentrations and other environmental changes (Houghton 1994). Beyond the
impacts on ecosystems and bio-diversity, LULC change may have signiﬁcant
world-wide socioeconomic and geopolitical consequences. Understanding,
managing and mitigating the implications of LULC change is therefore imperative

for environmental and societal sustainability.

Throughout human history, people have been affecting the
terrestrial biosphere through land-use changes for agriculture, energy, natural
resources and other purposes. However, an interest in understanding the
consequences of these land-use changes on the hydrologic cycle, particularly the.
subsurface components has emerged only recently. As a result substantial
knowledge gaps exist in our understanding of both short and long-terrn impacts
of LULC and LULC change on water resources. The potential impacts of these
changes on subsurface hydrology and hydrologic processes are signiﬁcant
(Scanlon et al. 2005). Alterations to groundwater recharge processes due to
LULC change can damage the viability of groundwater resources that are a
critical source of fresh water around the world. At current consumption and
population growth rates (United Nations, 1999), signiﬁcant conversions of natural
ecosystems to agriculture will occur in the next half century (Tilman et al. 2001).
Population growth will also result in conversion of substantial natural lands to
urban land-uses. In the US alone the proportion of developed land base is
projected to approximately double by 2025 (Alig et al. 2004). While similar global
projections are not readily available, current trends in economic and social

development in Asia, South America, and other regions is likely to increase the

current urban/suburban land base substantially (Lambin et al. 2001).
Compounding the impacts of these issues are population migrations that will
result from climate change driven sea level rises and increased aridity in some

regions of the world (Byravan and Rajan 2006).

Changes in LULC can alter soil characteristics, thereby affecting runoff
and inﬁltration properties of soils. Soil tillage with agricultural land-uses for
1 example affect groundwater recharge by changing the soil structure (Oleary
1996; Leduc et al. 2001). Increased percentages of impervious surfaces with
urbanization are known to generally increase runoff, evaporation, and decrease
groundwater recharge (Grove et al. 2001; Rose and Peters 2001; Burns et al.
2005; Jayawickreme and Hyndman 2007). Signiﬁcant and dynamic changes in
recharge are also accompanied by changes in vegetation where one or more
species is supplanted by another in a landscape (Engel et al. 2005; Nosetto et al.
2007; Scanlon et al. 2007). Such changes can lead to modiﬁcations of fractional
vegetation cover and hence precipitation interception, solar insolation, and wind
turbulence characteristics. Perennial, deciduous, or transient traits of vegetation
species can introduce important seasonal forcings on the environment.
Replacing perennial vegetation with annual crops and crop rotations for example
have lead to increased recharge during fallow periods (Oconnell et al. 1995;
Zhang et al. 1999). Changes in rooting depths with crop rotations can affect both
the quantity and quality of groundwater. Increased recharge and solute
mobilization have caused soil salinization and water quality degradation in large

areas of southwestern Australia after land clearing (Schoﬁeld and Ruprecht

1989; Schoﬁeld 1992; Pierce et al. 1993; Petheram et al. 2002). In regions where
deep rooted tress and other woody plants are gradually replacing shallow rooted
grasslands, decreased recharge and increased discharge of groundwater by
trees reaching deeper into the subsurface have been observed (Jobbagy and
Jackson 2004). Woody plant encroachment of native grasslands has also
increased in North America during the last century with ﬁre suppression and
other land management practices (Van Auken 2000; Coppedge et al. 2001;
Heisler et al. 2003; Pielke et al. 2007). In northern latitudes, vegetation changes
can alter frozen soil dynamics by affecting snow accumulation and snowpack
stability during the winter months (van der Kamp et al. 1999; van der Kamp et al.
2003). With deeper frozen soil columns, recharge potential is likely to be reduced
during the spring snowmelt, which is an important period of groundwater

recharge in these regions (Hayashi et al. 2003).

The vadose zone is the critical link between groundwater and the land
surface. The partitioning of precipitation into runoff and inﬁltration is largely
governed by the physical characteristics of the near surface soils or other
geologic materials. The fraction of inﬁltrated precipitation volume that eventually
reach the water table depends on the properties and processes of the vadose
zone. Principal among properties are the soil water retention and hydraulic
conductivities of the constituent materials. Where these physical conditions are
favorable, a signiﬁcant portion of the inﬁltrated water can reach the water table.
In natural settings however, vegetation has evolved to effectively utilize the

reservoir of water in the vadose zone, and hence these dynamics play an

important role in groundwater recharge. It is often recognized that vegetation
density and diversity across landscapes is a reﬂection of soil moisture and
nutrient dynamics (Baillie et al. 1987; Swaine 1996). Central to understanding the
inﬂuence of vegetation on subsurface water ﬂuxes is plant root distribution and
dynamics. This is also a topic of substantial importance for understanding
ecosystem carbon and water ﬂuxes (Rodriguez-lturbe 2000; Koster and Suarez

2001; Katul et al. 2007).

Root distributions and water abstraction characteristics (i.e. wilting point)
of vegetation vary signiﬁcantly between species and regions (Canadell et al.
1996; Jackson et al. 1996). Despite being studied for over two centuries
(Jackson et al. 1996), there are signiﬁcant knowledge gaps about the
spatiotemporal functions of plant roots. Existing insights are primarily from
studies that have employed techniques from excavations to radio-isotopes
(Dambrine et al. 1997; Boutton et al. 1999). While these have been immensely
helpful for conceptualizing root zone geometries and some aspects of root zone
dynamics, the static and localized nature of these. observations limit their use in
large scale models. Newer approaches that can link subsurface processes with
above-ground remotely-sensed large scale observations from space and air
borne platforms can contribute signiﬁcantly to improving and incorporating

vegetation dynamics in regional hydrologic, climate, and other models.

Objective

The broad objective of this research is to identify and quantify the potential
impacts of land-use (the manner in which the land is used), land-cover (the
physical and biological cover over land surface), and vegetation on groundwater
recharge. The research presented focuses on the temperate American mid-west
where land use is dominated by agriculture and vegetation being the primary
land cover. Groundwater recharge in the region primarily occurs through diffuse
recharge processes and is strongly inﬂuenced by vegetation and seasonal
climate dynamics. The vadose zone, through which recharge primarily occurs, is
an important focus of this research. This zone is also the region where plants
obtain most or all of their nutrients and water required for growth and
maintenance. Gaining insights into the vadose zone and its processes is
therefore necessary to understand land-use impacts on groundwater recharge. In
this research, a novel approach is used to explore and quantify dynamic vadose

zone interactions between vegetation, climate, and soil moisture.

The research in this dissertation is presented in ﬁve chapters. Chapter two
examines the implications of LULC on groundwater recharge at watershed scales
by analyzing streamﬂow across a range of watersheds. Baseﬂow (groundwater
contribution to streamﬂow), overland ﬂow, (surface ﬂow due to inﬁltration excess
precipitation), and total ﬂow (sum of baseﬂow and overland ﬂow) components of
streamﬂow in watersheds ranging from 20 to >1000 km2 are linked with land-
uses (forests, agriculture, urban etc.), soil, and morphological attributes using

statistical measures. Various spatial and non-spatial data sources, both

customary and non-traditional, were evaluated and utilized to quantify the
watershed hydrologic budgets and identify how LULC affect various ﬂux

components of the water budget.

In chapter three, the vadose zone interactions between climate, soil
moisture, and vegetation is explored. A novel approach combining geophysics
and petrophysical models is utilized to image subsurface soil moisture dynamics
across a forest-grassland ecotone. Electrical resistivity data collected at the
ecotone over multiple seasons is ﬁrst converted to soil moisture using
petrophysical relationships based on laboratory analysis of soils from the study
site. The two dimensional soil moisture distributions obtained are then correlated
with the vegetation across the ecotone. Seasonal soil moisture differences
identiﬁed and correlated with vegetation at the study site highlight potentially
signiﬁcant implications of large scale biological land cover changes on hydrology

and groundwater recharge.

Chapter four integrates geophysical measurements with hydrologic
modeling to quantify groundwater recharge differences that result from above
ground vegetation differences. The geophysical estimates of soil moisture are
used to derive soil hydraulic parameters for a one dimensional ﬂow model to
represent the ﬁeld site. Evidence of root zone geometries and distributions from
geophysical data is used to deﬁne the inﬂuence of vegetation on subsurface soil
moisture dynamics. However, coupling subsurface geophysical observations with
process-based hydrologic models remains a difﬁcult task and as such the

integration made in this research is only presented as a ﬁrst approximation.

The potential of electrical resistivity method for investigating and acquiring
subsurface hydrologic characteristics and quantities is explored in Chapter ﬁve of
this dissertation. The strong dependence of electrical conductivity on soil
moisture provides the basis for its use in hydrologic investigations. However,
deriving soil moisture from resistivity data involves several other considerations.
Data acquisition and processing, as well as the effects of other environmental
variables all inﬂuence the ability to derive accurate estimates of soil moisture
from resistivity data. The potential implications of these are discussed and
possible improvements are suggested in chapter ﬁve with ﬁeld examples from

our research site.

References

Alig, R.J., J.D. Kline, et al. 2004. Urbanization on the us landscape: Looking
ahead in the 21st century. Landscape and Urban Planning 69:219-234.

Baillie, I.C., P.S. Ashton, et al. 1987. Site characteristics and the distribution of
tree species in mixed dipterocarp forest on tertiary sediments in central
sarawak, malaysia. Journal of Tropical Ecology 32201-220.

Boutton, T.W., S.R. Archer, et al. 1999. Stable isotopes in ecosystem science:
Structure, function and dynamics of a subtropical savanna. Rapid
Communications in Mass Spectrometry 13:1263-1277.

Burns, D., T. Vitvar, et al. 2005. Effects of suburban development on runoff

generation in the croton river basin, new york, usa. Journal of Hydrology
311:266-281.

Byravan, S., and SC. Rajan. 2006. Providing new homes for climate change
exiles. Climate Policy 6:247-252.

Canadell, J., R.B. Jackson, et al. 1996. Maximum rooting depth of vegetation
types at the global scale. Oecologia 108:583-595.

Coppedge, B.R., D.M. Engle, et al. 2001. Landscape cover type and pattern
dynamics in fragmented southern great plains grasslands, usa. Landscape
Ecology 16:677-690.

Dambrine, E., M. Loubet, et al. 1997. Localisation of mineral uptake by roots
using sr isotopes. Plant and Soil 192:129-132.

Engel, V., E.G. Jobbagy, et al. 2005. Hydrological consequences of eucalyptus
afforestation in the argentine pampas. Water Resources Research 41.

Grove, M., J. Harbor, et al. 2001. Impacts of urbanization on surface hydrology,
little eagle creek, indiana, and analysis of lthia model sensitivity to data
resolution. Physical Geography 22:135-153.

Hayashi, M., G. van der Kamp, et al. 2003. Focused inﬁltration of snowmelt water
in partially frozen soil under small depressions. Journal of Hydrology
270:214-229.

Heisler, J.L., J.M. Briggs, et al. 2003. Long-terrn patterns of shrub expansion in a
c-4-dominated grassland: Fire frequency and the dynamics of shrub cover
and abundance. American Journal of Botany 90:423-428.

Houghton, RA. 1994. The worldwide extent of land-use change. Bioscience
44:305-313.

Jackson, R.B., J. Canadell, et al. 1996. A global analysis of root distributions for
terrestrial biomes. Oecologia 108:389-41 1.

Jayawickreme, DH, and D.W. Hyndman. 2007. Evaluating the inﬂuence of land
cover on seasonal water budgets using next generation radar (nexrad)
rainfall and streamﬂow data. Water Resources Research 43:W02408.

Jobbagy, E.G., and RB. Jackson. 2004. Groundwater use and salinization with
grassland afforestation. Global Change Biology 10:1299-1312.

Katul, G., A. Porporato, et al. 2007. Stochastic dynamics of plant-water
interactions. Annual Review of Ecology Evolution and Systematics 38:767-
791.

Kearey, P., and M. Brooks. 1991. An introduction to geophysical exploration.
Secon edition ed. Blackwell Science.

Koster, R.D., P.A. Dirrneyer, et al. 2004. Regions of strong coupling between soil
moisture and precipitation. Science 305:1138-1140.

10

Koster, RD, and M.J. Suarez. 2001. Soil moisture memory in climate models.
Journal of Hydrometeorology 2:558-570.

Lambin, E.F., B.L. Turner, et al. 2001. The causes of land-use and land-cover
change: Moving beyond the myths. Global Environmental Change-Human
and Policy Dimensions 11:261-269.

Leduc, C., G. Favreau, et al. 2001. Long-term rise in a sahelian water-table: The
continental terminal in south-west niger. Journal of Hydrology 243:43-54.

Lenat, DR, and J.K. Crawford. 1994. Effects of land-use on water-quality and

aquatic biota of 3 north-carolina piedmont streams. Hydrobiologia
294:185-199.

Marland, G., R.A. Pielke, et al. 2003. The climatic impacts of land surface change
and carbon management, and the implications for climate-change
mitigation policy. Climate Policy 32149-157.

Nosetto, M.D., E.G. Jobbagy, et al. 2007. The effects of tree establishment on
water and salt dynamics in naturally salt-affected grasslands. Oecologia
152:695-705.

Oconnell, M.G., G.J. Oleary, et al. 1995. Potential groundwater recharge from
fallowing in north-west Victoria, australia. Agricultural Water Management
29:37-52.

Oleary, G.J. 1996. The effects of conservation tillage on potential groundwater
recharge. Agricultural Water Management 31 :65-73.

Paul, M.J., and J.L. Meyer. 2001. Streams in the urban landscape. Annual
Review of Ecology and Systematics 32:333-365.

Petheram, C., G. Walker, et al. 2002. Towards a framework for predicting
impacts of land-use on recharge: 1. A review of recharge studies in
australia. Australian Journal of Soil Research 40:397-417.

11

Pielke, RA 2005. Land use and climate change. Science 310:1625—1626.

Pielke, R.A., J. Adegoke, et al. 2007. An overview of regional land-use and land-
cover impacts on rainfall. Tellus Series B-Chemical and Physical
Meteorology 59:587-601.

Pierce, L.L., J. Walker, et al. 1993. Ecohydrological changes in the murray-
darling basin .3. A simulation of regional hydrological changes. Journal of
Applied Ecology 30:283-294.

Pijanowski, 8., OK. Ray, et al. 2007. Using backcast land-use change and
groundwater travel-time models to generate land-use legacy maps for
watershed management. Ecology and Society 12.

Pitman, A.J., G.T. Narisma, et al. 2004. Impact of land cover change on the
climate of southwest western australia. Journal of Geophysical Research-
Atmospheres 109.

Rodriguez-lturbe, l. 2000. Ecohydrology: A hydrologic perspective of climate-soil-
vegetation dynamics. Water Resources Research 36:3-9.

Rose, 8., and NE. Peters. 2001. Effects of urbanization on streamﬂow in the
atlanta area (georgia, usa): A comparative hydrological approach.
Hydrological Processes 15:1441-1457.

Scanlon, B.R., R.W. Healy, et al. 2002. Choosing appropriate techniques for
quantifying groundwater recharge. Hydrogeology Journal 10:18-39.

Scanlon, B.R., I. Jolly, et al. 2007. Global impacts of conversions from natural to
agricultural ecosystems on water resources: Quantity versus quality.
Water Resources Research 43:W03437.

Scanlon, B.R., R.C. Reedy, et al. 2005. Impact of land use and land cover
change on groundwater recharge and quality in the southwestern us.
Global Change Biology 11:1577-1593.

12

Schoﬁeld, N.J. 1992. Tree planting for dryland salinity control in australia.
Agroforestry Systems 20:1 -23.

Schoﬁeld, N.J., and J.K. Ruprecht. 1989. Regional-analysis of stream salinization
in southwest western-australia. Journal of Hydrology 112:19-8.

Swaine, MD. 1996. Rainfall and soil fertility as factors limiting forest species
distributions in ghana. Journal of Ecology 84:419-428.

Tilman, D., J. Fargione, et al. 2001. Forecasting agriculturally driven global
environmental change. Science 292:281-284.

Tong, S.T.Y., and W.L. Chen. 2002. Modeling the relationship between land use
and surface water quality. Journal of Environmental Management 66:377-
393.

Van Auken, O.W. 2000. Shrub invasions of north american semiarid grasslands.
Annual Review of Ecology and Systematics 31 :197-215.

van der Kamp, G., M. Hayashi, et al. 2003. Comparing the hydrology of grassed
and cultivated catchments in the semi-arid canadian prairies. Hydrological
Processes 17:559-575.

van der Kamp, G., W.J. Stolte, et al. 1999. Drying out of small prairie wetlands
after conversion of their catchments from cultivation to permanent brome
grass. Hydrological Sciences Journal-Journal Des Sciences
Hydrologiques 44:387-397.

Walker, G.R., L. Zhang, et al. 2002. Estimating impacts of changed land use on
recharge: Review of modelling and other approaches appropriate for
management of dryland salinity. Hydrogeology Journal 10:68-90.

Wayland, K.G., D.T. Long, et al. 2003. Identifying relationships between baseﬂow
geochemistry and land use with synoptic sampling and r-mode factor
analysis. Journal of Environmental Quality 32:180-190.

13

Werth, D., and R. Avissar. 2005. The local and global effects of african
deforestation. Geophysical Research Letters 32.

Zhang, L., W.R. Dawes, et al. 1999. Estimating episodic recharge under different
crop/pasture rotations in the mallee region. Part 2. Recharge control by
agronomic practices. Agricultural Water Management 42:237—249.

Zhang, L., W.R. Dawes, et al. 2001. Response of mean annual
evapotranspiration to vegetation changes at catchment scale. Water
Resources Research 37:701-708.

14

Chapter Two

Jayawickreme, D. H., and D. W. Hyndman (2007), Evaluating the inﬂuence of
land cover on seasonal water budgets usingNext Generation Radar (NEXRAD)
rainfall and streamﬂow data, Water Resour. Res., 43, W02408,

doi:10.1 029/2005WR004460.

15

Abstract

Accurate estimates of ﬂuxes between different components of the
hydrosphere are needed for water resources management at the watershed
scale. Runoff and evapotranspiration are critical ﬂuxes that are heavily inﬂuenced
by land cover characteristics; however, our understanding of the interactions
between land over attributes and these ﬂuid ﬂuxes is generally limited by
inadequate regional data to capture variations in both climatic conditions and
landscape characteristics. This limitation is largely avoided by integrating data
from Next Generation Radar (NEXRAD) precipitation systems with widely
available streamﬂow, land cover, and other Geographic lnforrnation System data
sets. Such data integration facilitates development of rapid and reliable methods
for estimating hydrologic ﬂuxes at desirable temporal and spatial scales. In this
study, we calculate ratios of streamﬂow to NEXRAD rainfall over the peak
growing season for 40 watersheds across Michigan and use these to evaluate
the landscape factors that inﬂuence groundwater recharge rates. Results indicate
that ratios of streamﬂow and baseﬂow to rainfall are strongly inﬂuenced by land
cover attributes. Stream baseﬂow analyses indicate that approximately 5% of the
July to September rainfall becomes recharge in high-intensity agriculture (>70%)
watersheds compared to 15% in moderate intensity agriculture (<50%)
watersheds across our study region during the same period. A strong negative
correlation was also found between intensive agriculture and the

streamﬂowzrainfall ratio during summer and early fall periods.

16

Introduction

Understanding and evaluating water budgets in small to regional
watersheds is critical for a range of resource management decisions. Water
ﬂuxes between different components of the hydrosphere change over a range of
temporal and spatial scales due to variations in climatic conditions and landscape
characteristics. Surface hydrological properties and processes such as soil
moisture, runoff, and evapotranspiration are heavily inﬂuenced by land use and
land cover characteristics, which we will refer to as land cover for simplicity. In
addition, human activities and the accompanied land cover changes have the
potential to signiﬁcantly alter the local hydrology and cause long-terrn
environmental changes (Dow and DeWalle, 2000; Walker et al., 2002; Costa and
Foley, 2000; Pielke et al., 1998). Despite this potential, our understanding of the
interactions between land cover and hydrology has been limited by inadequate

data collection and integration at regional scales.

Runoff, evapotranspiration (ET), groundwater recharge, and changes in
aquifer storage due to groundwater outﬂows and pumping are the main
processes that redistribute precipitation in hydrologic systems in addition to
snowmelt in colder regions. While runoff and aquifer storage can often be
evaluated with a reasonable degree of accuracy, the most important components
for watershed management, ET and groundwater recharge, are difﬁcult to
quantify. Commonly used methods including hydrograph separation,
groundwater budget analysis, and tracer analysis coupled with groundwater

models can be used to estimate recharge rates at various spatial scales (Scanlon

17

et al., 2002). However, large data requirements and other limitations have made
it difﬁcult for watershed managers and other decision makers to develop accurate
estimates using such methods. Partly in response to these limitations, relatively
simpler techniques using water balance models and Geographic lnforrnation
Systems (GIS) have recently been introduced (e.g., Cherkauer and Ansari, 2005;
Szilagyi et al., 2004). Nevertheless, most such approaches do not describe the
impact of land cover on watershed hydrology and groundwater recharge at short
timescales. In regions such as Michigan, with distinct seasonal changes in land
cover (vegetation) and spatially diverse land use practices, vegetation dynamics
are a key component of any hydrologic analysis. This paper examines the
inﬂuence of land cover attributes on hydrologic ﬂuxes at watershed scales on a
seasonal basis. We use commonly available data sources to obtain watershed
attributes and ﬂux budgets for 40 watersheds, and we statistically link the ﬂux
budget characteristics with the watershed attributes. Since evapotranspiration
(ET) is a key component of the water budget during the growing season and it is
directly related to land cover characteristics (Mo et al., 2004; Finch, 1998), we
also compare the streamﬂow and groundwater recharge differences of high-
intensity and moderate-intensity agricultural systems. This analysis provides

insight into the potential impacts of land use decisions on watershed hydrology.

18

Approach

Seasonal analysis of the ratio between total water input (precipitation) and
outﬂows (streamﬂow and its components; overland ﬂow and base ﬂow) to a
watershed during speciﬁed time intervals can provide insight into the
mechanisms that redistribute moisture (e.g., runoff, recharge, ET). We evaluated
correlations between streamﬂow: rainfall ratios and watershed characteristics
(e.g., land cover, geologic materials, and soils) to help understand how these
attributes inﬂuence hydrologic processes at watershed scales. We calculated the
streamﬂowzrainfall ratios over the approximate peak growing season for 40
different watersheds using hourly Next Generation Radar (NEXRAD)
precipitation data from the National Weather Service (NWS) and streamﬁow from
US. Geological Survey (USGS) daily records. Prior to this analysis, the accuracy
of NEXRAD data was evaluated by comparing it to observations from ground-

based gauges, as described in more detail below in section 4.

The July-September period was chosen for the water budget calculations
both to minimize the effects of snowmelt recharge on the analysis and to capture
the inﬂuence of vegetation on watershed hydrology. This period corresponds to
both the low streamﬂow period based on USGS records and the peak growing
season of the region based on mean leaf area index (LAI, one-sided green leaf
area or projected needle leaf area per unit ground area). The LAI values for the
region were obtained from the 1-km Moderate Resolution Imaging
Spectroradiometer (MODIS) 8 day composite data product (version 4) from the

National Aeronautic and Space Administration (NASA). We used the MODIS

19

Reprojection Tool (USGS, 2004) to convert MODIS data in hierarchical data
format (HDF) into Georeferenced Tag Image File Format (GEOTIF F) and ESRI
grids. The spatial analyst extension and zonal statistics tools in ESRI ArcGIS
software were used to extract mean LAI values for watersheds by overlaying a

watershed boundary coverage on the LAI grids in ESRI ArcMap (version 9.1).

The spatially averaged NEXRAD precipitation values for each study
watershed were calculated using the ArcMap zonal statistics tool on 100 ‘m
resampled NEXRAD grid cells within each watershed boundary. Daily, monthly,
and longer-period precipitation estimates were subsequently derived from the
hourly precipitation estimates. We automated the hourly precipitation extraction
process using custom scripts (included in the auxiliary materiali) written in Visual
Basic for Applications (VBA) integrated with ESRI ArcGIS, which load and unload
data from ArcMap, and calculate mean watershed precipitation amounts with
zonal statistics tools. This signiﬁcantly improved the efﬁciency of processing
_2000 individual NEXRAD grids for each of the 40 watersheds. Correction
coefﬁcients obtained from the regressions between July—September monthly
NEXRAD and monthly gauged precipitation were used to adjust for the bias from

NEXRAD data before utilizing it in mass balance analyses.

Total monthly base ﬂow and overland ﬂow volumes were estimated from
the daily mean streamﬂow records using the PART computer program (Rutledge,
1998), which estimates base ﬂow from daily streamﬂow records based on
antecedent streamﬂow recession. This approach assumes spatially diffused

recharge to the water table, uniform aquifer thicknesses, uniform hydraulic

20

conductivities and storage characteristics as well as minimal regulation and
diversion of streamﬂow within the gauged watershed. Although these
assumptions are essentially never met in real aquifer systems, this provides an
objective tool to evaluate base ﬂow. In addition, heterogeneities in soil properties
tend to be at a much smaller scale than that of the watersheds we analyzed in
this study, thus the localized impacts likely average out. Base ﬂow estimates
obtained with PART have been shown to be comparable to that obtained with
various other manual methods (Rutledge, 1998). Linear interpolation of base
ﬂows during times that do not ﬁt the antecedent criteria used in PART would lead
to base ﬂow estimation errors. However, errors resulting from the linear
interpolation have been shown to be minimal for monthly or longer timescales

(Rutledge, 1998).

Land cover percentages (croplands, forests, urban areas, wetlands, etc.)
for watersheds were calculated based on the National Land Cover Dataset
(NLCD) (USGS, 1999) and Anderson level I classes (Anderson et al., 1976). The
forest cover percentage for each watershed was obtained by aggregating the
deciduous, evergreen and mixed forest classes in NLCD. Pasture/hay, row
crops, and small grain classes were combined to obtain the total percentage of
agricultural uses. Total urban land cover percentages were obtained by
combining low-intensity residential, high-intensity residential, and

commercial/industrial/transportation classes.

The distribution of Quaternary geologic materials was obtained from a

digital coverage of Farrand and Bell (1982) (data available at

21

http:/Mwwmcgistate.mi.us/qul/), and were aggregated in to ﬁve broad classes
(glacial tills, end moraine tills, outwash sand and gravel, lacustrine clay and silt,
and lacustrine sand and gravel) for each watershed. On the basis of State Soil
Geographic (STATSGO) database for Michigan (available at I
http://www.ncqc.nrcs.usda.qovl products/data sets/statsgo/data/index.html), we
also categorized the watershed soils into three different drainage classes that
were expected to inﬂuence streamﬂow characteristics in the study region
(extremely to somewhat extremely well drained, well to moderately well drained,
and poorly to very poorly drained). Statistical correlations between watershed
attributes (land cover, Quaternary geology, soil drainability, and watershed
morphology) and the average 2002-2004 volume ratios (streamﬂowzrainfall,
overland ﬂowrrainfall, and base ﬂowzrainfall) were evaluated using nonparametric
Spearman’s correlations. The Kruskal-Wallis (nonparametric) test along with the
chi- square approximation for its two sided p value was used to compare base
ﬂow and recharge differences in high intensity (>70%) and moderate-intensity
(<50%) agricultural watersheds. Two sided p values at level 0.05 were used to
test statistical signiﬁcance. Nonparametric statistical methods were used in this
study to minimize the effects of assumptions associated with parametric
correlation methods. We also used stepwise multiple regressions and standard
least squares ﬁts as exploratory methods to evaluate variability of
streamﬂowzrainfall and base ﬂowzrainfall ratios between watersheds, and to
explain the variability of the ratios in terms of land cover and other watershed

attributes.

22

 

Figure 2-1. Locations of our study watersheds (shaded) with NEXRAD radar and
rain gauge locations. Dashed lines show the effective range of the radar
detectors.

Study Sites

Forty Michigan watersheds with drainage areas ranging from 20 km2 to
1000 km2 with an average of 312 km2 (USGS station IDs and names are provided
in the auxiliary material) were chosen that span a range of land cover

characteristics and have sufﬁcient data for our analysis (Figure 2-1). The primary

23

land cover types in the selected watersheds are agriculture, forests, and urban,
followed by the minor proportions of wetlands, grasslands, and open water. The
percentage of agricultural land in the study watersheds ranges from about 6 to
85%, with a mean of 51%. Corn, beans, and alfalfa are the main crops cultivated
in the state, which have growing seasons ranging from late April/early May to
about mid-November. The percentage of forest cover in the study watersheds
ranges from about 8 to 55, with a mean of 26%; and nearly all of the forests in
our study watersheds are deciduous. The amount of urban area in the study
watersheds ranges from nearly 0 to 70%, with a mean of 11%. Urban and
suburban land covers are more common in the southeastern part of the state
with most of the forest land in the northwestern regions of Michigan’s Lower (or

southern) Peninsula.

Quaternary glacial advances and retreats shaped the regional geology of
Michigan. The surﬁcial deposits of the state are mostly glacial outwash and till
deposited during the Pleistocene continental glaciation. Glacial outwash deposits
are more abundant in the north/northwestern and southwestern portions of
Michigan’s Lower Peninsula, while till deposits are prominent in the central and
eastern half of the Lower Peninsula, extending from the Saginaw Bay area to the
Ohio border. Lacustrine clays are common in the Saginaw Bay area and along
the eastern fringe of the southern half of the state where artiﬁcial irrigation
management practices, such as tile drains, are common. 78% of our study

watersheds had less than 10% clay and only 17% had more than 20% clay.

24

Evaluation of NEXRAD Rainfall Data

Starting in 1980, the National Weather Service (NWS) established the
nationwide NEXRAD network of Doppler radar stations (Weather Surveillance
Radar (WSR) —1988 Doppler (88D)). There are approximately 158 operational
WSR-88D stations throughout the US, with some overseas locations. Information
from these radar stations is commonly used to issue warnings of severe weather
and ﬂash ﬂoods to the public, and provide information for air trafﬁc safety, water

management, and outdoor activities.

The detailed spatial and temporal coverage of NEXRAD data makes it a
useful input to hydrologic models and provides an invaluable resource where
ground-based rain gauges are scarce. The Army Corps of Engineers, US.
Department of Agriculture, and National Weather Service all use radar data in
hydrologic models. Within the research community, radar rainfall data have
mainly been used to simulate streamﬂow response to storm events. Some
examples include Neary et al. (2004), who used radar rainfall data to derive basin
averaged hourly precipitation to simulate streamﬂow using a HEC-HMS model.
Di Luzio and Arnold (2004) used the NEXRAD hourly grids in Soil and Water
Assessment Tool (SWAT) model to predict hourly streamﬂow in response to
storm events. In a similar manner, Carpenter et al. (2001) used NEXRAD
precipitation data in a spatially distributed hydrologic model to simulate runoff
and streamﬂow to evaluate the use of distributed hydrologic models in an
operational environment. In this study, we are mainly interested in quantifying the

inﬂuence of watershed characteristics on water balances, which could not be

25

accurately evaluated based on ground-based gauges alone due to the general
sparse nature of the rain gauge networks. We overcame this obstacle by
calculating basin-averaged hourly precipitation rates from NEXRAD grids, which
were then summed into monthly and growing season volumes for each study
watershed. The primary radar rainfall product from the WSR-88D, called the
Digital Precipitation Array (DPA), is generated by processing the radar
information using a Precipitation Processing System (PPS). The PPS is a set of
algorithms that use information from two external functions for precipitation
detection (effective within a 230 km radius from the radar station) and rain gauge
data acquisition, and ﬁve internal functions for data preprocessing, radar to
rainfall rate conversions, rainfall accumulation calculations, gauge-radar
adjustments, and product generation (Fulton et al., 1998). The hourly
precipitation products used in this study have a 4 km x 4 km spatial resolution
and are generated from base Doppler radar data (reﬂectivity, mean radial doppler

velocity and spectrum width) using the PPS.

26

Si

 

0
30« +2002
02003
25, ° + A2004
«- +
E
EZO‘ A 9 o
315 0 0 A +
E 1‘ + A A A + 6
10« A A +
5 «i
O

 

 

 

Apr t May'June' July' Aug‘ Sep' Oct' Nov
Month

Figure 2-2. RMS differences between NEXRAD and ground-based precipitation
measurements at various sites with data spanning the 2002, 2003, and 2004
growing seasons.

There are several sources of systematic and random error associated with
radar rainfall estimates (Seo et al., 1999). The uncertainties in reﬂectivity (the
quantity measured by the radar) to rainfall conversion, which is highly nonlinear,
is recognized as one of the main sources of error (Neary et al., 2004).
Differences in radar instrument calibration from station to station, distance from
radar stations or the range effect (Sharif et al. 2002), radar scan angles, local
topography (Young et al., 1999), and climate conditions (Smith et al., 1996) can
all cause signiﬁcant error in radar rainfall estimates. We evaluated the accuracy
of NEXRAD data calculated on the basis of the three WSR- 88D weather radar
stations in Michigan (Figure 2-1) relative to point observations of precipitation at

NWS and other independent stations at daily and monthly timescales. This

27

involved comparing 4 km x 4 km spatial rainfall grids provided by the Michigan
State University Geography Department to point rainfall data for 28 locations in
Michigan’s Lower Peninsula with sufﬁcient data from April to November of 2002-
2004 (Figure 2-1). Root-mean-square differences (RMSD) between radar and
direct measurements were computed for the radar and gauge pairs. We also
assessed temporal variations in NEXRAD errors over monthly and growing

season timescales.

Comparison of NEXRAD and ground-based precipitation data from May—
November indicate that August has the largest monthly RMSD in the 2002 and
2003 data sets (Figure 2-2). This is mainly due to relatively large differences
between NEXRAD and ground-based precipitation at a few gauge locations:
Detroit, Bellaire, Grayling, and Ypsilanti in 2003; and Detroit, Bellaire, Grand
Haven, Muskegon, and Howell in 2002. Nearly all of the monthly RMS

differences remained below 25 mm throughout each of these growing seasons.

The largest percent absolute difference (calculated relative to gauged
precipitation) between the radar estimated and gauged rainfall within the study
region for the 2004 growing season (May-November) was 13% at the Grand
Haven gauge location (Figure 2-3). This amounts to a 50 mm difference between
the radar and gauge systems for the entire growing season. The largest absolute
differences were concentrated in the northwestern corner of Michigan’s Lower
Peninsula, where the highest errors ranged from about 25% to 38% in 2002 at
ﬁve locations (North Port, Traverse City, Cadillac, Kalkaska, and Houghton). All

watersheds from the northwestern region were thus excluded from this study.

28

The lower absolute differences in the southern half of the state are partially due
to the use of at least some of 'Grand Rapids, Lansing and Flint station data by the
WSR-88D system for real-time corrections of the precipitation predictions. At
most locations, NEXRAD estimates tend to be smaller than the precipitation

recorded by the ground-based gauges.

Although most of the discrepancy between observed and NEXRAD
precipitation is likely due to errors in radar estimates, some of the differences can
also be attributed to inaccuracies in the ground-based precipitation observations.
Mechanical failures associated with tipping bucket gauges often give rise to
random errors, and the aerodynamic design of the gauges frequently result in
systematic error in rainfall measurements (Sevruk, 1996; Heinemann et al.,
2002). Habib et al. (2001) have shown that sampling frequency, bucket size and
precipitation characteristics also contribute to errors in tipping bucket rainfall
data. Errors in ground-based observations discussed here would generally result
in an under estimation of actual rainfall, particularly during heavy precipitation
events. These errors contribute to a larger RMSD and would count the same as

radar overestimates.

29

 

2.8 .
AW! .
8.7 ‘
Graylhal
i 5
Houghton Lake .
A 0.1
Harbor Beach .

a if.» V,
Morita mag,” 11. 5 $3
Muskegonll13 1& . 9.1
Flint
Grand Haven 6. 9 |
Grandliwld s '5 Lansing 1%]
0

0 W 1.4
‘ Ann Arbor'% 31:38?
Detrolt

 
  

 

Figure 2-3. Map of the percent absolute differences (calculated relative to
observed gauged precipitation) between the 2004 (April-November) gauge and
NEXRAD rainfall data. The Grand Rapids, Lansing, Flint, and Alpena gauges are
known to be used by the WSR-88D system for real-time calibration of radar
rainfall estimates (Ann Arbor and Grayling are missing 1 month of data, Harbor
Beach is missing 2 moths, and Montague is missing 3 months, and thus these
months were not used in this analysis.

30

 

80 ‘ 5 Alpena A Flnt o Lansing
a a Detroit + Muskegon o Pelston
... x Grand Rapids 1: l-bughton lake

 

 

 

 

 

251 °
X n + g
A e
20. x “‘ +
g + ‘ x 0 9
g 151 2 ‘ +. +
C 3 o
10.1 & + X
O § . x O a *
51 X * x . g T
a
0 K o
O I’ V I 1 3 ‘T V j
Apr May Jun Jay Am Sop Oct Nov
Morin

Figure 2-4. RMS differences for eight gauge locations using 2002, 2003, and
2004 monthly radar and ground-based gauge precipitation data. May RMSD
calculation is based only on data from 2002 and 2004 due to missing NEXRAD
grid data in MAY 2003.

The accuracy of the NEXRAD estimates relative to ground-based
observations for the eight gauges across the region with continuous precipitation
records from 2002 to 2004 is shown in Figure 2-4 (see Figure 2-1 for locations).
On the basis of this analysis, there does not appear to be any signiﬁcant spatial
trend in NEXRAD precipitation estimates except for the already mentioned larger
error to the northwest corner of the Lower Peninsula. The Muskegon gauge was
the only location that showed consistently higher RMSD values throughout the
analysis period. In general there seems to be relatively higher variability at all
gauge locations beginning in May and continuing through August. This may be
due to intense precipitation events associated with convective weather systems

that are relatively common during this period. Under such conditions, both

31

ground based gauges (Heinemann et al., 2002) and radar systems (Krajewski
and Smith, 2002) are known to be less accurate, which contributes to the

relatively large RMSD during such periods.

As is commonly the case, the total radar rainfall estimates for the
evaluation period are generally lower than the ground-based precipitation
recorded during the same period at the gauge stations. A limited evaluation of
event scale data (hourly NEXRAD versus observed precipitation) showed that
radar system performed poorly during very small precipitation events. However,
only a small percentage of the differences in monthly precipitation totals between
radar and gauge data in Michigan can be attributed to such small events.
Another source of error is the comparison of precipitation derived from relatively
large NEXRAD grid cells (4 km x 4 km) with point gauges. Event-scale NEXRAD
data were not directly compared with ground based gauge data in this study
because of difﬁculty in obtaining hourly precipitation data for a sufﬁcient number
of ground-based gauges. The 2004 data show the highest degree of correlation
between monthly gauge and NEXRAD rainfall from July to September, our main
study period for water budget evaluations (Figure 2-5), while 2002 and 2003 also
had reasonable correlations to gauged data. Data from the 2002 to 2004 period
were used for water budget calculations in this study after adjusting for the bias

in NEXRAD precipitation using correction coefﬁcients shown in Figure 2-5.

32

 

 

   

 

 

 

 

 

 

 

 

 

 

 

'
. ” ’l’
A 1 “Ex-W...
E It: i
gim- “T”
3 00- ..........
40- * Symbol Year 92 [RMSE Bu
‘ +5 . 2002 o.77|17.r ea
‘ 2003 osglias 52
0*, . , . ‘ . . . 2004 can“? -1.e
0 4o 80 12° ‘50
NEXRADRQI‘HMIM

Figure 2-5. Comparison of monthly (July-September) observed gauge and
NEXRAD rainfall for 2002, 2003, and 2004. RMSE and bial (average NEXRAD
rainfall-average observdd gauge rainfall) are given in mm.

Results and Discussion

A signiﬁcant decrease in the July September streamﬂowzrainfall ratio was
observed with increasing agricultural land cover above about 60% in all years
(Figure 2-6). This relationship has a smaller correlation in watersheds with lower-
intensity agricultural land uses, likely due to the heterogeneity in land cover and
morphological attributes that are characteristic of the lower-intensity agricultural

systems, but it is still statistically signiﬁcant across our 40 watershed sample
according to the nonparametric correlation coefﬁcients with a p value of 2.0x104

(Table 2-1). Comparatively, high-intensity agricultural systems tend to be
relatively homogeneous with respect to vegetation and morphology. An additional

factor that could contribute to the low streamﬂowzrainfall ratio in intensely

33

agricultural systems in some environments is the presence of tiled drains.
However, such engineered drainage systems mainly exist in areas with
signiﬁcant proportion of clay rich soils. Only three watersheds with >30% clay rich
soils were included in this analysis (circled on Figure 2-6), and two of these three
have lower streamﬂowzrainfall ratios than other watersheds with similar

agriculture percentages (Figure 2-6).

 

 

 

40
am _9... 112.076, ya-0.94x+79
"" zoos 2-9.... 112-0.55. y—O.49x+43
3°13 + 2004 --..’c--.n’-o.ee. y=-O.95x+87
+ + Ave —L-i-'1’=0.71, y=-o.7sx+es

Streamﬂow Vol.:Rainfall Ratio Vol. (96)

 

 

 

 

% Agricultural Land In Watershed

Figure 2-6. Percentage of agricultural land cover in watersheds compared to
streamﬂowzrainfall ratios for July to September periods of 2002-2004. Each data
point represents a watershed, and three watersheds with greater than 30% clay
sediments are marked with a gray oval.

.8080 28 .3 02855 0 agave 50:85:00 .802ch 822555 028.50 ow 9: .e 83%; 852905
.050 new 8me oE:_o> $00N-N00NV ESEoﬁow 2 >3... cows. .8 000%me 00:20:00 000.com Mcmctmoaw .TN 030...

 

 

 

 

«0.950 5.0 5.3.0 no.0 F0.mm.~ 9.0 @003 cows.
noﬂmd 9.0 500.0 no.0. 500.? «N0 $5.923 592:0
«0.2.6 3.0 3.3.0 00.0. mom : 0N0 m__om 00:56 =03 30:8.on 023058389wa
00.3.0 00.0. Sand and. 3&0... 00.0. $00A923000<
«0&0... 3.0. 27mm." 3.0. 3.m0.~ 00.0. 23.30000 .2
00.36 3.0 F0.wm.m «0.0. «0.90 8.0 $20“.
5.3.0 3.0 «0&0; 5.0 amoimns 0N0 cue:
os_m> 0 9E :mEmoam o:_m> a 05... 55.0on o:_m> 0 05¢ :2:me
oE:_o> .3.—50m oE:_o> _Esmm 0E:.o> _Ecﬁm

\0E:_o> 26¢ 33 Emobm \oE:_o> 30E 05:90 \oE:_o> 300E025

 

35

An expected signiﬁcant increase of overland flow in watersheds with urban
land cover percentage is also evident from the correlation statistics (Table 2-1),
even in this case where only 20% of the watersheds in this study had more than
10% urban land cover. A positive correlation was also observed between base
ﬂowzrainfall ratios in forested watersheds. The small p values (<0.05) in Table 2-2
indicate that the correlations between these land cover attributes and volume
ratios in the analyzed watersheds are unlikely to be random. Other factors that
show signiﬁcant relationships with the volume ratios are the distribution of certain
surﬁcial glacial geologic materials and soil drainability classes. The amount of
base ﬂow is positively correlated with both the percent glacial outwash sand and
gravel deposits and the extremely to somewhat extremely well drained soils, yet
there was no statistically significant correlation between soil drainability and
outwash deposits. Glacial outwash deposits were present in 90% of the
investigated watersheds with varying degrees of abundances. Positive
correlations also exist between glacial outwash deposits and forest cover
percentages in the study watersheds, and between agricultural land uses and
both poorly drained and low slope areas (Table 2-2). Seventy-three percent of
the variability in the streamflow: rainfall ratio was explained by three watershed
attributes (percentages of agriculture, open water, and glacial outwash sand and
gravel) using a multiple linear regression analysis (Table 2-3). The proportion of
agricultural land cover is the most signiﬁcant attribute among the pool of
watershed variables used for the exploratory multiple regression analysis. Poorly

drained soils failed to explain a signiﬁcant portion of the variability observed with

36

streamﬂow: rainfall ratio, indicating that low streamﬂow conditions in watersheds
with high-intensity agriculture is more likely related to the land use and cover
attributes, rather than simply the soils. A negative coefﬁcient for open water
(lakes, ponds, wetlands etc.) indicates that these areas are associated with
reductions in streamﬂow, which is likely associated with direct evaporation from
open water surfaces. In a similar manner as the streamﬂowzrainfall ratio, 75% of
the variability associated with base ﬂowzrainfall ratio in the study watersheds was
explained by four watershed attributes (Table 2-3). Agriculture, open water, and
glacial outwash sand and gravel explained most of the variability in base I
ﬂowzrainfall ratio, followed by extremely to somewhat extremely well drained
soils. One of the 40 watersheds was removed from the multiple regression
analysis as an outlier due to unusually high streamﬂow conditions, likely related
to urbanization effects since it was in the Detroit suburban area. Watersheds with
high-intensity agriculture (>70%) tend to have lower base ﬂows during the
growing season than those with moderate-intensity (<50%) agriculture. Kruskal-
Wallis tests had p values that indicate that April— June, July-September, and
October—December base ﬂow differences between high-intensity and moderate-
intensity agricultural watersheds are statistically signiﬁcant (Figure 2-7). A
considerable drop in base ﬂow is evident during the peak growing season (July to
September) in both classes of watersheds. However, the decline is relatively
larger in watersheds with intense agriculture. Watersheds with over 10% high-
intensity urban land uses (NLCD classiﬁcations “high-intensity residential” and

“commercial/industrial/ transportation”) were removed from the data set prior to

37

the base ﬂowzrainfall comparison in Figure 2-7, to minimize the urban effects that
tend to be hydrologically different from forest and agricultural systems. High ET
demands by active crops and anthropogenic abstraction of water for irrigation are
two factors that could contribute to low base ﬂow in high-intensity agricultural
watersheds. However, the persistence of low base ﬂow across the range of
studied watersheds suggests that ET demand is a major component of the water
budget during the growing season. According to US. Department of Agriculture
(USDA) statistics, only a small percentage (~5%) of Michigan’s croplands are
irrigated (Economic Research Service, 2004). Countywide surface and shallow
water withdrawals in 2000 were nearly uniform across the study watersheds, thus
it is unlikely that anthropogenic abstraction of water for irrigation is the main
cause of the observed lower streamﬂows associated with high-intensity

agricultural watersheds.

38

.8080 23 .3 8.8.2.. 0 8:80:00 _800000 0020.20; 02800
00 00. .0. 00.00..0> 0000.000; .050 000 .0>00 000.. .00 0000005 0000.980 00006000002 000000000 .00 0.00...

 

 

00.w00.0 00.0 5.m00.0 00.0 00.306 3.0. 0.0.0.0 00.000004
5000.0 00.0 5000.? 00.0. 00.3.0.0 00.0 00.90000 000350
00.50% 90.0. 000.000 00.0 5.m00.0 00.0. 0.... 0.00.0
00000.0 00.0 00.m0 «.0 10.0. 50.00; 00.0 0:00 000.0.0 :03 220.000.00.03
5.w00._. 00.0. 5.5.0.0 000 00.5.0.0 00.0 0:00 000.0.0 :03 20.00.08 «050.000.00.003
00.m.00.0 00.0. 00000.: 00.0 00.m00.0 0.0.0. 0:00 00500.0 >000...
5.w00.0 5.0. 00.m00.0 00.0. 00.50.: 00.0 000.0 000.2
0:_0> 0 000. 00000000 0:.0> 0 001 00000000 0:.0> 0 05... 00000000

 

0000.00.00 0095 0000.00.00 0.3.0002 003000.00 00.0“.

 

39

202000000. .000 000 00.0 0.0 00:.0> 0 ”00.000. .00 003 3:00:02 .000. 00.000
000.0000. 0... 0:; 2:00:00 .0. 00.00. 0.03 0.: £00 E0: 0.00000. 00... 0050...? 09.0.00; 000 0000”. _.0.0.0m§0_n.
000m 000 ._0.0.0m§o=0.00:w A500-0000. 000.2 .00E0.00w 2:... .0.— 00..0..0.w 00.000500. 0.00.0.2 .00 0.00...

 

 

 

2.0.00.0 5.0 _.0.0 $0.000.8 00.90000 0002000
oo.m0...0 5.0 00. .... 003000.00 .0003 0000
5000.0 00.0 05. 003000.00 0.2.0002
5000.0 2.0. 00.0 0:00 000.0.0 :03 20.00.00 000300.00 000 20.00.00
mménwwsz... . . . 000.000.
00.000 000.00 0 00 0 0 0..
03000> 00000000. ..0.0..0m...sot 000m
00000.: 00.0 00.0 e0.0050000 000.200
8.08.0 3.0 00.0- $.20; :80
00.080 3.0 0.0. $92.85..
5.0uwmsE - . . . 000.000.
0 0.0.00. 0.. mom 0 00 0 0 00
030.003 00000000. ..0.0..0&.§0:E00:m
0.. .0 000.050 _._Aa__3320 00 0.2.000 5.3

 

00.00.00m 00050.00

 

4O

The differences in annual base ﬂow between high intensity and moderate-
intensity agricultural watersheds can also be attributed to early spring frozen soil
conditions. High-intensity agricultural areas are more susceptible to frozen soils
during the winter and early spring months, which would tend to lower the
recharge rates during the important snowmelt period. In the absence of a
persistent snowpack, soils in the region have been shown to freeze to about 5
cm depth even in warm winters (lsard and Schaetzl, 1998). Intermittent
snowpack conditions in winter months are more likely in bare and exposed
farmlands resulting in a thicker frozen soil layer. An increasing trend in
groundwater recharge toward the northwestern and western parts of the state
reported by Holtschlag (1997) is consistent with the larger snowpack and denser
forest cover along the northwestern and western fringes of the state as observed
with mean January through May streamﬂowzdrainage area ratio (Figure 2-8).
Spatial patterns in volume ratios for watersheds in this study, however, did not

show any speciﬁc east-west trend during the July to September period.

Culmination of the growing season, as depicted by a sharp decrease in
mean monthly LAI in Figure 2-9, initiates a period of steady increase in base
ﬂow. Correspondence between the LAI and the base ﬂowzwatershed area ratio is
consistent with the expectation that transpiration signiﬁcantly reduces the

recharge rates in this region during the summer period.

41

a
a:

 

 

 

d
N
1

 

 

 

 

a

b

 

 

O
0

Quarter Year Baseflowzorainage Area Ratio (cm)

 

 

Jan-Mar Apr-June I July-Sop ' Oct-Dec

Figure 2—7. Average 2002-2004 quarter year base ﬂowzdrainage area rations in
high-intensity (>70%) and moderate-intensity (<50%) agricultural watersheds.
The difference of the ﬂow between the two types of watersheds are statistically
signiﬁcant (p value <0.05) in all quarters except January-March.

 

 

Parameter 2002 2003 2004 Average
Agriculture <50%
Percent of rainfall 14.1 9.9 21.7 15.2
Actual amount, cm 2.6 2.4 3.9 3
Agriculture >70%
Percent of rainfall 4.6 3.3 7.8 5.3
Actual amount, cm 0.7 0.8 1.5 1

 

 

Table 2—4. Estimated Annual Growing Season Recharge Amounts (July—
September); in tems of annual rainfall and actual amounts in centimeters.

42

 

Figure 2-8. Average of 2002-2004 January to May streamﬂowcdrainage area
ratios (cm). Higher values likely indicate greater snowmelt inﬂuence on the
watershed hydrology. The contour map was generated by kriging the ratio
calculated from data at the 113 USGS streamﬂow gauges shown on the map.
Images in this dissertation are presented in color.

The vegetation density associated with agricultural uses, calculated based
on LAI, is signiﬁcantly lower than that of forests, which is the most frequent land
cover type in moderate-intensity agricultural watersheds. However, Figure 2-7
suggests that the high-intensity agriculture has a much more signiﬁcant effect on
hydrology than moderate-intensity agriculture during the peak growing season.

The observed temporal changes in streamtlow and volume ratios are also

43

evident in the region’s groundwater system. Data from continuous water level
transducers installed in a shallow aquifer in the Grand Traverse Bay Watershed
located to the northwestern part of the state reveal a steady decline in water
levels through late October followed by an increase thereafter (Figure 2-10).
There is only minimal irrigation according to county statistics for the area where
the wells are located, and the primary land cover within their watersheds is forest
(Economic Research Service, 2004). This is further evidence of low recharge
rates during the growing months, which are too small to compensate for the
deﬁcit created by elevated water abstraction and transpiration by plants and base
ﬂow discharges to streams. Because of the damping effects of the subsurface
materials, the lowest groundwater levels in the region are generally observed a
few months after the peak growing period. This highlights the difﬁculty in using
alternative approaches that analyze only groundwater levels and budgets to

estimate transient groundwater recharge rates.

 

    

250
Total Rainfall 20 5,
:l2002 '
022003 -15
'20“ .10‘
.05
. E
E BaseﬂovrArea Ratio
“2.5.. LealArealndex
g Forest ’5
X
$2.0. r4 12
in. 0
a . ..
D
3 ~23
1.0. .................
é _ .1
0.5

 

 

 

Mayr Jun? Julyf Aug' Sept 00' Nov

Figure 2-9. The 2002-2004 monthly base ﬂowzdrainage area ratios and average
monthly rainfall for 40 selected watersheds in Michigan. The peak growing
season is clearly marked by higher leaf area indexes (LAI), which corresponds
well with the decline in stream base ﬂow during the same period.

45

pallynaml
(mm)
{A 0

 

 

 

 

50.
3.95 me
$13 3 E
8 7:; 9 g
E5 in
925 6.35

 

 

Jul ‘Aug 'Sep'ocl ‘Now 'oeclJan 'Feb
2003 2004
Date

Figure 2-10. Measured water table depths in two shallow wells in the northern
portion of Michigan’s Lower Peninsula (see Figure 2-1 for locations). A steady
decline in water level is evident during the growing season (land use percentages
in the watershed containing B12 are forests ~65°/o, shrubs/open land ~35%; G13,
forests ~66%, agriculture ~28%). Note that the minimum water level is delayed
signiﬁcantly from the peak LAI and minimum base ﬂow levels shown in Figure
2-9. .

The base ﬂowzrainfall ratio for July September period of each year was
analyzed to evaluate differences in recharge between high-intensity and
moderate-intensity agricultural watersheds in each of the years from 2002 to
2004. Base ﬂowzrainfall ratio is a good measure of groundwater recharge given
the base ﬂow estimates are reasonably accurate. According to our analysis, the
average growing season recharge from 2002-2004 in high-intensity agricultural
watersheds was ~5% of the total rainfall compared to 15% in moderate-intensity
agricultural watersheds (Figure 2-11). According to the Kruskal-Wallis test, the

difference of the mean base ﬂowzrainfall ratio between the two types of

46

watersheds in each of the years is statistically signiﬁcant (Figure 2-11). The
above percentages amount to ~1.0 cm of growing season recharge in our high-
intensity agricultural watersheds, compared to ~3.0 cm in moderate intensity
agricultural watersheds (Table 2-4). The variability of recharge percent across
the studied watersheds within a given year is likely related to both differences in
land cover percentages and precipitation characteristics. The relatively higher
ratios in 2004 are likely related to late spring snowmelt discharge effects and
precipitation which contribute to higher streamﬂow conditions that sometimes
extend to late June (Figure 2-12). In areas where times between recharge events
and resulting discharge are relatively small (days to a few weeks), the ratio
method presented in this paper could perhaps be applied at shorter temporal
scales. Both the starting point and ending point for the analysis should include no
signiﬁcant rainfall events or recent increases in streamﬂow for several days to
weeks, depending on the response time of the watershed of interest. However,
larger uncertainties associated with both radar rainfall and base ﬂow estimates at
small time intervals would tend to increase the uncertainty in short time period

recharge estimates with this approach.

47

 

 

 

 

 

 

 

 

Baseflow Vol.:Rainfall Vol. Ratio (‘36)

 

 

 

 

Figure 2-11. Recharge estimates for the July to September period, as a percent
of NEXRAD bias adjusted rainfall in high-intensity agricultural (>70%) and
moderate-intensity agricultural (<50%) watersheds. The difference in
baseﬂowzrainfall ratios between the two types of watersheds are statistically
signiﬁcant (p value <0.05) in all years.

Summary and Conclusions

Despite recent advances that have been made to quantify groundwater
recharge rates, existing methods are generally limited by insufﬁcient data.
Increasing concerns over the likelihood of unsustainable water resources in
many regions of the world emphasize the need for simple approaches that use
readily available data for water budget assessment. Many of the existing
approaches for groundwater recharge assessment require long-term monitoring,
cumbersome and complex watershed models, accurate subsurface parameter

estimates that are difﬁcult to acquire, and signiﬁcant time commitments. As a

48

result these approaches often fail to deliver rapid water budget estimates for
watersheds over critical time periods. To address these difﬁculties, we introduce
a method that can be used to rapidly estimate water budgets and recharge rates
over various temporal and spatial scales. We believe this approach can be
adopted for rapid preliminary assessment of seasonal and longer term recharge
conditions in most humid regions with relatively small unsaturated zones and no
large artiﬁcial diversions of water. Data extraction and processing for this

approach can be easily performed using GIS systems and simple database

 

 

 

 

 

 

 

 

schemes.
- 25
5 «ram .
g +Streamlow Volumezwmelshed Area I
E 20‘ —Overland Flow Volume:WatershedAree :
l
g. 15‘ E
l
l l
505 l :
l l
g 0 uL.\%—W—i>—I=-/ : g—
0“? $090900 @y"9\>‘%y° od‘eﬁ‘l‘lﬂeywﬁea

2002 2003 2004

Figure 2-12. Total monthly streamﬂow, and overland ﬂow volumes normalized by
watershed area along with rainfall averaged across the 40 study watersheds.

We statistically analyzed ratios of streamﬂow, stream base flow and

overland ﬂow volumes to precipitation volume to examine factors that inﬂuence

49

the water mass balances for 40 watersheds across Michigan during the growing
season. Observed differences are primarily attributed to land cover
characteristics. Our analysis indicates that there is minimal growing season
recharge in all of the studied watersheds, with high-intensity agricultural
watersheds receiving essentially one third of the growing season recharge of
moderate-intensity agricultural watersheds. Low runoff and base ﬂow conditions
and low streamﬂowzrainfall ratios in agricultural watersheds provide insight into
the signiﬁcant growing season water demands for intensive croplands. While this
has signiﬁcant implications for managing water resources, further analysis is
required to interpret and quantify the detailed processes leading to these trends

in terms of ET and other forms of water use related to agriculture.

Statistical evaluation of streamﬂow and its component volumes (base ﬂow
and overland ﬂow) as a percentage of total precipitation can be correlated with
land cover and other watershed attributes. This provides a fairly simple and
efﬁcient approach to characterize watershed behavior across a range of temporal
and spatial scales. When accurate precipitation and ﬂow data are available for
watersheds, they can be used to evaluate seasonal or longer-term potential
recharge, ET, and runoff volumes. These estimated ﬂuxes provide critical inputs
to transient hydrologic models. Relatively large uncertainties associated with
base ﬂow estimates as well as radar rainfall estimates at event and other short
timescales (i.e., daily, weekly), however, makes it difﬁcult to apply the ratio

method to evaluate ﬂux budgets in watersheds over short time windows.

50

The use of radar-derived precipitation estimates can simplify and
potentially improve the quality of water resource analyses for some watersheds.
This is especially true when large watersheds are involved because there are
often signiﬁcant spatial and temporal variations in precipitation at these scales.
Spatial characteristics of precipitation are extremely difﬁcult to capture solely
from ground-based gauges; thus use of NEXRAD data is likely to improve the
accuracy of regional water mass balances. While there are numerous merits to
using NEXRAD precipitation data in hydrogeological studies, the temporal and
spatial accuracy needs to be evaluated for any study region due to the known
uncertainties currently associated with these data. The continued effort of the
National Oceanic and Atmospheric Administration (NOAA) to improve the
accuracy of the radar rainfall estimates is likely to make NEXRAD an

indispensable resource for many hydrologic applications.

Acknowledgments

We greatly appreciate the funding provided by a grant from the National
Science Foundation (EAR-0233648) and thank the Associate Editor and three
anonymous reviewers for their constructive comments. Our sincere gratitude is
extended to Dennis Gilliland of the Center for Statistical Consulting at Michigan
State University for suggestions that greatly improved the quality of this
manuscript. Special thanks are also extended to Jeffrey Andresen, James T.

Brown, and Tracy Aichele of the Department of Geography, Michigan State

51

University for converting the NEXRAD data into arc grids and archiving these for
extended time periods. We appreciate the suggestions and contributions of
Bryan Pijanowski (Purdue), M.S. Phanikumar, and students in the hydrogeology
research group at Michigan State University. Any opinions, ﬁndings, and
conclusions or recommendations expressed in this material are those of the
authors and do not necessarily reﬂect the views of the National Science

Foundaﬁon.

52

References

Anderson, J. R., E. H. Hardy, J. T. Roach, and R. E. Vtﬁtmer (1976), A Land Use
and Land Cover Classiﬁcation System Wlth Remote Sensor Data, U. S.
Govt. Print. Off., Washington, D. C.

Carpenter, T. M., K. P. Georgakakos, and J. A. Sperfslagea (2001), On the
parametric and NEXRAD-radar sensitivities of a distributed hydrologic
model suitable for Operational use, J. Hydrol., 253, 169-193.

Cherkauer, D. S., and S. A. Ansari (2005), Estimating ground water recharge

from topography, hydrogeology. and land cover, Ground Water, 43, 102-
112.

Costa, M. H., and J. A. Foley (2000), Combined effects of deforestation and
doubled atmospheric CO: concentrations on the climate of Amazonia, J.
Clim., 13, 18-34.

Di Luzio, M., and J. G. Arnold (2004), Formulation of a hybrid calibration
approach for a physically based distributed model with NEXRAD data
input, J. Hydrol., 298, 136—154.

Dow, C. L., and D. R. DeWalle (2000), Trends in evaporation and Bowen ratio on
urbanizing watersheds in eastern United States, Water Resour. Res., 36,
1835- 1843.

Economic Research Service (2004). State fact sheet for Michigan, U. S. Dep. of

Agric., Washington, DC (Available at nn92/lmw.ers.gsda.ggv/
StateFacts/Ml.htm).

Farrand, W. R., and D. Bell (1982), Quaternary geology of Michigan, Michigan
Geological Survey map, scale 12500000, Mich. Dep. of Nat. Resour.,
Lansing.

53

Finch, J. W. (1998), Estimating direct groundwater recharge using a simple water
balance model-sensitivity to land surface parameters, J. Hydrol., 211,
1 12-125.

Fulton, R. A., J. P. Breidenbach, D. J. Seo, D. A. Miller, and T. O’Bannon (1998),
The WSR-88D rainfall algorithm, Weather Forecasting, 13, 377- 395.

Habib, E., W. F. Krajewski, and A. Kruger (2001), Sampling errors of tipping-
bucket rain gauge measurements, J. Hydrol. Eng., 6, 159- 166.

Heinemann, A. B., G. Hoogenboom, and B. Chojnicki (2002), The impact of
potential errors in rainfall observation on the simulation of crop growth,
development and yield, Ecol. Mod., 157, 1 -21.

Holtschlag, D. J. (1997), A generalized estimate of groundwater recharge rates in
the Lower Peninsula of Michigan, U. S. Geol. Surv. Water Supply Pap.,
2437.

lsard, S. A., and R. J. Schaetzl (1998), Effects of winter weather conditions on
soil freezing in southern Michigan, Phys. Geogr., 19, 71- 94.

Krajewski, W. F., and J. A. Smith (2002), Radar hydrology: Rainfall estimation,
Adv. Water Resour., 25, 1387- 1394.

Mo, X., S. Liu, 2. Lin, and W. Zhao (2004), Simulating temporal and spatial

variation of evapotranspiration over the Lushi basin, J. Hydrol., 285, 125-
142.

Neary, V. S., E. Habib, and M. Fleming (2004), Hydrologic modeling with
NEXRAD precipitation in middle Tennessee, J. Hydrol. Eng., 9, 339- 349.

Pielke, R. A., R. Avissar, M. Raupach, A. J. Dolman, X. B. Zeng, and A. S.
Denning (1998), Interactions between the atmosphere and terrestrial
ecosystems: Inﬂuence on weather and climate, Global Change Biol., 4,
461— 475.

Rutledge, A. T. (1998), Computer programs for describing the recession of
ground-water discharge and for estimating mean ground-water recharge
and discharge from streamﬂow data-update, U. S. Geol. Surv. Water
Resour. Invest. Rep., 98-4148.

Scanlon, B., R. Healy, and P. Cook (2002), Choosing appropriate techniques for
quantifying ground-water recharge, Hydrogeol. J., 10, 18- 39.

Sea, D. J., J. P. Breidenbach, and E. R. Johnson (1999), Real-time estimation of
mean ﬁeld bias in radar rainfall data, J. Hydrol., 223, 131-147.

Sevruk, B. (1996), Adjustment of tipping-bucket precipitation gauge
measurements, Atmos. Res., 42, 237—246.

Sharif, H. 0., F. L. Ogden, W. F. Krajewski, and M. Xue (2002), Numerical
simulations of radar-rainfall error propagation, Water Resour. Res., 38(8),
1140, doi:10.1029/2001WR000525.

Smith, J. A., D. J. Seo, M. L. Baeck, and M. D. Hudlow (1996), An
intercomparison study of NEXRAD precipitation estimates, Water Resour.
Res., 32, 035—2045.

Szilagyi, J. S., F. E. Harvey, and J. F. Ayers (2004), Regional estimates of total
recharge to groundwater in Nebraska, Ground Water, 43, 63- 69.

U. S. Geological Survey (1999), Michigan land cover data set, Sioux Falls, 8.0.

(Available at t ://e . r. . v r r ms/lc In tion llan
html)

55

U. S. Geological Survey (2004), MODIS Reprojection Tool user’s manual,release
3.2a, USGS EROS Data Center, Sioux Falls, SD.

Walker, G. R., L. Zhang, T. W. Ellis, T. J. Hatton, and C. Petheram (2002),
Estimating impacts of changed land use on recharge: Review of modeling

and other approaches appropriate for management of dryland salinity,
Hydrogeol. J., 10, 68- 90.

Young, C. B., B. R. Nelson, A. A. Bradley, J. A. Smith, C. D. Peters- Lidard, A.
Kruger, and M. L. Baeck (1999), An evaluation of NEXRAD precipitation
estimates in complex terrain, J. Geophys. Res., 104, 19,691-19,703.

56

Chapter Three

Jayawickreme, D. H., R. L. Van Dam, and D. W. Hyndman (2008), Subsurface
imaging of vegetation, climate, and root-zone moisture interactions, Geophys.
Res. Lett., 35, L18404, doi:10.1029/2008GL034690.

57

Abstract

Changes in global climate and land use affect important processes from
evapotranspiration and groundwater recharge to carbon storage and biochemical
cycling. Near surface soil moisture is pivotal to understand the consequences of
these changes. However, the dynamic interactions between vegetation and soil
moisture remain largely unresolved because it is difﬁcult to monitor and quantify
subsurface hydrologic ﬂuxes at relevant scales. Here we use electrical resistivity
to monitor the inﬂuence of climate and vegetation on root-zone moisture, bridging
the gap between remotely-sensed and in-situ point measurements. Our research
quantiﬁes large seasonal differences in root-zone moisture dynamics for a
forest-grassland ecotone. We found large differences in effective rooting depth
and moisture distributions for the two vegetation types. Our results highlight the
likely impacts of land transformations on groundwater recharge, streamﬂow, and

land-atmosphere exchanges.

58

Introduction

Modiﬁcation of temperatures and precipitation patterns due to global
climate change will cause signiﬁcant transformations of ecosystems and plant
physiological functions (Betts et al. 2007). Climate changes will also affect water
availability and plant productivity (Field et al. 1995; Nemani et al. 2003), and
hence will impact soil moisture dynamics in many regions of the world.
Compounding these effects are rapid changes in land cover, driven by factors
including urbanization, energy demands, and food production (Evans and Kelley

2004; Jayawickreme and Hyndman 2007; Scanlon et al. 2007).

Vadose zone soil moisture is a key driver of climate, energy and carbon
cycles (Niklaus et al. 2003; Koster et al. 2004; Seneviratne et al. 2006), as well
as ecosystem dynamics (Rodriguez-lturbe 2000). The amount and distribution of
soil moisture also inﬂuences important processes from soil microbial activity to
nutrient ﬂuxes and groundwater recharge (Rushton et al. 2006; Hyndman et al.
2007). Characterizing the dynamic interactions between vegetation and soil
moisture is critical to forecast global water resources and improve land-
atmosphere feedback models. However, these interactions remain largely
unresolved due to our inability to characterize transient subsurface water ﬂuxes

at high resolution under natural conditions.

There are signiﬁcant gaps in resolution and sampling volume between
common approaches to monitor root zone moisture. Remote sensing provides

regional-scale estimates of water content, but has limited spatial and temporal

59

resolution and marginal depth penetration (Dubois et al. 1995; Jackson 2002;
Entekhabi and Moghaddam 2007; Wagner et al. 2007); (Scott et al. 2003). In
contrast, time domain reﬂectometry and other probes provide accurate point-
scale estimates of soil moisture with high temporal resolution, but can not readily
be up-scaled (Robock et al. 2000). In addition, remote sensing and point-based
methods provide only limited information about moisture percolation below the
root zone and thus groundwater recharge rates, which are essential for water

resources management.

Electrical resistivity imaging (ERI) is an alternative approach to monitor
subsurface hydrologic conditions and processes across a range of materials and
spatial scales with high temporal resolution. In ERI, direct current is injected into
the ground and potential differences are measured between a series of electrode
pairs along an array to obtain 2D resistivity images; these can then be converted
into soil moisture estimates based on petrophysical relationships between
resistivity and pore-water content (Lesmes and Friedman 2005). Until now,
hydrological applications of ERI have largely focused on solute transport and
inﬁltration monitoring with little emphasis on vegetation effects (Daily et al. 1992;
Slater et al. 2000; Berthold et al. 2004; LaBrecque et al. 2004; Singha and

Gorelick 2005; Al Hagrey 2007).

Although ERI has been used to characterize temporal changes in moisture
content (Binley et al. 2002; Amidu and Dunbar 2007), we explore for the ﬁrst time
natural interactions between seasonal soil moisture dynamics, climate variability,

and vegetation differences using time-lapse ERI. For this analysis, we equipped

60

a forest-grassland ecotone with a suite of hydrogeophysical equipment. Our
observations from this ecotone demonstrate that ERI can be used to accurately
quantify the spatiotemporal distribution of root-zone moisture content, bridging
critical gaps between remotely-sensed and in-situ point measurements. This
information is essential to project the inﬂuence of changing climate and land

covers on hydrologic ﬂuxes and ecosystem sustainability.

Methods

At a ﬁeld site near East Lansing, Michigan, USA, we instrumented an
ecotone separating a mature Maple forest from a grassland. Data were collected
at this site from October 2006 through September 2007. The ~4 m thick
unsaturated zone has 40-60 cm of clay loam underlain by medium to ﬁne sand
across the site, with respective porosities of 0.47 and 0.39. Maximum expected
rooting depths are ~7m for forest and ~2.5 m for grass (Jackson et al., 1996).
Along a 124.5 m transect centered on the forest-grassland boundary, 84 equally-
spaced, 30 cm long graphite electrodes were installed. The permanent electrode
array improves data reproducibility, maximizing the ability to identify changes in
soil moisture. Capacitance-type soil moisture loggers (20 and 80 cm depth) and
vertical temperature arrays (5, 10, 20, 40, 80, 117, and 147 cm depth) were

installed under both land covers.

Climate data for the site were obtained from a weather station 1.5 km from
the site; average daily air temperature ranged from -19°C to 28°C during the

61

study period. As expected, ﬂuctuations in soil temperature were increasingly
damped and phase-shifted with depth. The soils were insulated by continuous
snow cover from mid January to mid March, and based on our soil temperature
observations only the top 5-10 cm of the soils froze during the January to early
February period. The trees lost their leaves in mid November and leafed out
again in late April (See color bar in Figure 3-1). In contrast, the grass remained

green for the entire study period except from mid January to mid March 2007.

Twenty-seven resistivity data sets were collected using a Wenner
conﬁguration during the study period, spanning a full cycle of seasons (Figure
3-1). Differential inversions were used to calculate changes in resistivity,
minimizing the need to extensively characterize site stratigraphy. The differential
inversion algorithm used in this study (LaBrecque and Yang 2001) ﬁrst estimates
the resistivity distribution for a base dataset and then calculates the resistivity
change to a second data set. March 30, 2007 was selected for our base data set,
because the site had relatively uniform soil moisture across the ecotone after
spring snowmelt. To obtain the absolute resistivity distribution for each date, the
estimated resistivity change is added to the inverted base dataset. Resistivity
changes for each time interval were then obtained by subtracting the two

absolute resistivity images.

62

 

-——— 20cm soil racist!“ recs)

 

 

 

 

-ms:m°ém< 0
4o~ ESRI mgmwczoaémm. forest) mgr“?
E l f M i
g... l . -\.’I\'K\. “m «g
g N ‘ e I‘ g
320‘ 0.1 3
10~ o
o

 

 

 

 

 

 

 

OctNovDecJanFebMarApr'Mame Ju AmSep

m 2007

 

 

Figure 3-1. Soil moisture from probes at the site and the average of the 20 and
86.5 cm depth ERI moisture estimates for all datasets corrected with site
temperatures. Precipitation data is shown for a nearby gage. The color bar
shows the state of the forest canopy, vertical lines mark data collection dates,
and shaded areas above the bar are differential inversion periods for Figure 3-2.
Images in this dissertation are presented in color.

We corrected our inverted resistivity data for soil temperature differences

using data from the vertical sensor arrays (Hayley et al. 2007). Average summer
soil temperatures at 20 cm depth below the forest were ~3°C cooler than below

the grassland, largely due to shading by the forest canopy. After leaf-off in
November 2006, the near surface temperature difference became negligible. The
temperatures were assumed to be laterally uniform below each land cover,
except for the ~20 m wide portion of the grassland seasonally shaded by the

adjoining forest, where we interpolated the temperatures. Seasonal ﬂuctuations

63

were less signiﬁcant at 147 cm, thus the temperature was linearly interpolated to
a constant 10°C at 10 m depth based on observations below the water table.

Temperature distributions preceding the installation of temperature sensors were

estimated with multi-depth soil temperatures from the nearby weather station.

To estimate moisture contents from the resistivity data, site- and material-

speciﬁc relationships between resistivity (p) and volumetric water content (0)

were developed following ASTM standard G57-95. After oven drying at 105°C for
24 hours, ten soil samples from the ﬁeld site were wetted in ~4% water content

increments, homogenized, and placed in a test box for resistivity measurements.
For these samples, as for most soil materials, the p-6 relationship is well

approximated by a power function with coefﬁcient m (sand = 1.16; clay loam =

0.67), which is used in (Archie 1942):

S = ($)% (3-1)

where Sis saturation (volumetric water content / porosity), and ps is bulk
resistivity of the soil at 100% saturation, obtained from the ﬁeld data (sand =
71.53 am, clay loam = 68.15 Om). Since annual precipitation in Michigan greatly
exceeds evapotranspiration (ET), we assume constant ﬂuid conductivity. We
calculated the soil moisture content for each resistivity value obtained from the

differential inversions.

Results

Near the beginning of the study period, a signiﬁcant growing-season soil
moisture deﬁcit existed below the shallow forest soils, which recovered after rain
in November 2006. Beneath the snow, soil moisture declined until mid March,
when snowmelt brought in a pulse of water. Both areas then experienced a
steady moisture decline during the 2007 growing season, with a sharp increase

following sustained rain in late August (Figure 3-1).

Important processes in the vadose zone are highlighted in Figure 3-2
using panels that show the profound seasonal inﬂuence of vegetation and ‘
climate on resistivity during the study period. The effects of different rooting
depths between the forest and grassland on soil moisture are clear in early fall of
2006 (Figure 3-2a). A large rain four days prior to the end of the measurement
interval initiated an inﬁltration pulse across the site (Figure 3-1). The shallow
resistivity values below the forest then rapidly declined, yet values at similar
depths below the grassland increased slightly. This implies that most of the
inﬁltrated water below the grassland drained deeper within the proﬁle because
soils there were much closer to ﬁeld capacity. Continued ET then caused a slight
increase in shallow resistivity below the grassland. Increased resistivity at depth
below the forest likely indicates that the trees were actively transpiring. In
contrast, deep resistivity decreased below the grassland as water drained below

the effective rooting depth.

65

Forest _g— Grassland
nnl shadow zone

0, ¥’*3¥y !J 62 Zitwmwwwwmu. t u \l 1 Minimum

UIO

Depth (m)

 

 

 

, . ' a,“ - -. - 'lll"'l “WNWHW ‘ ' (d)
v h.“ Lﬁl Spring
“ 5"” Summer

% Change in resistivity

 

-35 -25 -15 -5 5 15 25 35

Figure 3-2. Differential resistivity panels for approximately one-month periods. a)
Early fall (10/18 - 11/15), b) late fall (11/22 — 12/20), c) winter (1/5 - 2/9), d)
spring (3/16 - 4/20), and e) summer (6/10 — 7/2). Trees are not to scale; 80 cm of
relief along the array is included in the inversion. Red triangles locate moisture
probes and temperature arrays. A decrease in resolution with depth and
associated smoothing artifacts can cause resistivity differences below the water
table. Images in this dissertation are presented in color.

Following the 2006 leaf senescence, resistivity below the forest decreased
(Figure 3—2b), consistent with increases in soil moisture during a period of
minimal transpiration and reduced evaporation under an insulating layer of leaf
litter. In contrast, the shallow soils below much of the grassland, where the
vegetation was still green, show no signiﬁcant change in resistivity. However, the
shallow soils below the grasses that were previously shaded by the trees show a

small increase in resistivity during this period, suggesting a late period of active

66

ET in response to increased solar radiation. The resistivity of the deeper soils
below the grass decreased, indicating that precipitation continued to percolate

below the zone of root water uptake.

During the winter, consistent below-freezing air temperatures and snow
cover reduced ET and surface inﬁltration, as indicated by the temperature-
corrected resistivity increase (Figure 3—2c) and moisture probe data (Figure 3-1).
Beneath the forest, the resistivity data show moisture redistribution deeper within
the soil proﬁle. The lack of signiﬁcant resistivity changes below the grassland

indicate that the moisture conditions there remained stable during this period.

The spring period is characterized by snowmelt inﬁltration, which resulted
in a continued decrease in resistivity below the forest (Figure 3-2d). In contrast,
moisture contents below most of the grassland remained largely stable as
indicated by an absence of resistivity changes. This suggests that these soils
were at or above ﬁeld capacity during the spring months, and most snowmelt

likely became recharge in the absence of signiﬁcant runoff at the site.

Early in the summer, high ET reduced the root-zone soil moisture content,
increasing the resistivity of the near surface soil layers (Figure 3-2e). Vegetation
differences are prominently highlighted, with larger resistivity increases below the
forest. The forest canopy intercepts more precipitation than the grasses, reducing
inﬁltration. Greater transpiration in the forest is another likely contributor to the
higher resistivity increases. Across the site, the resistivity increase was mostly

limited to the upper 2 m, in contrast to the observations during early fall of 2006

67

(Figure 3-2a). This illustrates preferential uptake of near surface moisture,

caused by the low suction potential of the relatively moist early-summer soils.

The soil moisture calculated from ERI data with Archie’s equation (3-1)

strongly correlates with the point-scale observations (ecajcujated = 1-04’90bserved

- 0.04; R2 = 0.92, Spearman’s ranked correlation = 0.88; p-value < 0.0001). The

small underestimation of the calculated moisture contents is likely due to a
preferred sensitivity of ERI to high resistivity layers as shown by the excellent
match between ERI estimated values and the smallest of the two measured
water contents in the forest (Figure 3-1). The small residual variation is partly due
to differences in resolution and ﬂuid conductivity, but this conductivity effect is

minimal in the unsaturated zone where water saturation governs bulk resistivity.

The spatial and temporal evolution of moisture content through the study
period shows a clear contrast across the ecotone (Figure 3-3). The largest
contrast in moisture contents beneath the two vegetation types was observed in
October 2006. The soil moisture deﬁcit was much larger and extended deeper
below the forest (Figure 3-3a), which implies that the forest has less recharge
than the grassland. As expected, an increase in moisture contents is observed
through early spring prior to signiﬁcant transpiration by the trees. This resulted in
the minimum observed contrast in moisture contents across the ecotone during
the early growing season (Figure 3-3e). During the summer of 2007, the soil
moisture deﬁcit again began to build preferentially below the forest (Figure 3-3t).

In contrast to the late growing season of 2006, most of the moisture is extracted

68

from the shallow zone during this period. The large increase in soil moisture
below 4 m depth, from October 2006 to April 2007, is partly due to a rising water

table.

0 62.25 124.5 m
J

O
.—
._

r_.__rr_._rr_._1r_._rL.._._Il_._r
. .
f

(a)
10118/06

0|

(1))
11/15m
~1- . 1 ~ 1 .7 .- .v

(C)
2/20/08

(d)
02/09/07

Depth (m)

(0)
04/20/07

(0
07/02/07
Volumetric moisture (cm’lcm’)

0 0.1 0.2 0.3 0.4 0.5

Figure 3-3. Spatial and temporal distribution of soil moisture estimated across the
ecotone. Images in this dissertation are presented in color.

We computed spatially averaged subsurface soil moisture differences
between peak-growing (July-August) and early-growing (April-May) periods for
zones entirely within the forest and the grassland. These data clearly show both
the effective rooting depth and contrasts in moisture abstraction characteristics of
the two vegetation types (Figure 3-4). The total moisture change is much smaller
below the grassland than the forest, and the zone of effective root water uptake
is considerably shallower for the grass. The deep moisture depletion below the

forest suggests that the tree roots tap into readily available water in the capillary

69

fringe, similar to phreatophites in riparian areas. However, the largest moisture
changes occur in the top 2 m of the subsurface in both vegetation types,
consistent with ﬁeld and modeling studies of root water abstraction (Jackson et

al. 1996).

Conclusions

Our results quantify differences in root-zone moisture uptake below
contrasting vegetation types and the evolution of vadose zone moisture in
response to seasonal climate and vegetation processes using ERI. The results
demonstrate the value of ERI for quantifying soil moisture distributions and
understanding unsaturated zone processes. We believe that with such
geophysical methods, signiﬁcant insight can be gained about interactions
between the atmosphere, hydrosphere, and biosphere. Understanding these
interactions is essential for identifying the potential impacts of climate and land
cover changes on the hydrologic cycle. Our ﬁndings show that land
transformations would alter near surface soil moisture distribution patterns,
impacting groundwater recharge rates, land-atmosphere energy exchange
characteristics, and streamﬂow. For example, partial reforestation of the
midwestern United States would likely reduce regional groundwater recharge
rates. Until now, such tangible evidence at spatial and temporal scales

considered in our study has been unavailable. Our analysis provides signiﬁcant

70

new insights into the impacts that accompany ongoing and anticipated global

changes in land use and climate due to anthropogenic and natural stressors.

 

 

 

 

 

0
1 .
E 2'
.2
ti
8 a-
4 . ........... ;. ..... .l ......
; +Grassland
i +Forest
5 , ; :
-0.05 0 0.05 0.1 0.15

Moisture difference (om3/cm3)

Figure 3-4. Laterally averaged changes in soil moisture below the forest (36-51
m) and the grassland (92-107 m) between early-growing (April to early May) and
peak-growing periods (July to August). Images in this dissertation are presented
in color.

Acknowledgements

This research was funded by NSF (EAR-0233648) and MSU Center for
Water Sciences. We thank M. Malkowski, M. Morse, P. Bloese, and R. Klevickas

for ﬁeld support, and J. Butler, J. Hendrickx, and A. Kendall for discussions.

71

References

Al Hagrey, SA. (2007), Geophysical imaging of root-zone, trunk, and moisture
heterogeneity, J. Exp. Bot, 58(4), 839—854.

Amidu, SA, and J.A. Dunbar (2007), Geoelectric studies of seasonal wetting
and drying of a Texas vertisol, Vadose Zone J., 6(3), 511-523.

Archie, GE. (1942), The electrical resistivity log as an aid in determining some
reservoir characteristics, T. Am. Inst. Min. Metal. Eng., 146, 54—61.

Berthold, 8., LR. Bentley, and M. Hayashi (2004), Integrated hydrogeological

and geophysical study of depression-focused groundwater recharge in the
Canadian prairies, Water Resour. Res., 40(6), W06505.

Betts, RA, 0. Boucher, M. Collins, PM. Cox, P.D. Falloon, N. Gedney, D.L.
Hemming, C. Huntingford, C.D. Jones, D.M.H. Sexton, and M.J. Webb
(2007), Projected increase in continental runoff due to plant responses to
increasing carbon dioxide, Nature, 448(7157), 1037-1041.

Binley, A., P. Winship, L. J. West, M. Pokar, and R. Middleton (2002), Seasonal
variation of moisture content in unsaturated sandstone inferred from
borehole radar and resistivity proﬁles, J. Hydrol., 267(3-4), 160—172.

Daily, W., A. Ramirez, D. Labrecque, and J. Nitao (1992), Electrical-resistivity

tomography of vadose water movement, Water Resour. Res., 28(5),
1429-1442.

Hayley, K., LR. Bentley, M. Gharibi, and M. Nightingale (2007), Low temperature
dependence of electrical resistivity: Implications for near surface

geophysical monitoring, Geophys. Res. Lett., 34(18), L18402.

72

Jackson, R.B., J. Canadell, J.R. Ehleringer, H.A. Mooney, O.E. Sala, and ED.
Schulze (1996), A global analysis of root distributions for terrestrial
biomes. Oecologia, 108(3), 389—41 1.

Jackson, T.J. (2002), Remote sensing of soil moisture: implications for
groundwater recharge, Hydrogeol. J., 10(1), 40—51.

Jayawickreme, DH, and D.W. Hyndman (2007), Evaluating the inﬂuence of land
cover on seasonal water budgets using Next Generation Radar
(NEXRAD) rainfall and streamﬂow data, Water Resour. Res., 43, W02408.

LaBrecque, DJ, and X. Yang (2001), Difference inversion of ERT data: a fast
inversion method for 3-D in situ monitoring, J. Env. Eng. Geoph., 6(2), 83—
89.

Lesmes, DP, and SP. Friedman (2005), Relationships between the electrical
and hydrogeological properties of rocks and soils, in Hydrogeophysics,
edited by Y. Rubin and SS. Hubbard, pp. 87-128, Springer, Dordrecht.

Nemani, RR, 0.0. Keeling, H. Hashimoto, W.M. Jolly, S.C. Piper, C.J. Tucker,
R.B. Myneni, and SW. Running (2003), Climate-driven increases in global
terrestrial net primary production from 1982 to 1999, Science, 300(5625),
1560-1563.

Robock, A., K.Y. Vinnikov, G. Srinivasan, J.K. Entin, S.E. Hollinger, N.A.
Speranskaya, S.X. Liu, and A. Namkhai (2000), The global soil moisture
data bank, Bull. Am. Met. Soc, 81(6), 1281-1299.

Rodriguez-Iturbe, l. (2000), Ecohydrology: A hydrologic perspective of climate-
soil-vegetation dynamics, Water Resour. Res., 36(1), 3-9.

Rushton, K.R., H.M. Eilers, and RC. Carter (2006), Improved soil moisture
balance methodology for recharge estimation, J. Hydrol., 318(1-4), 379-
399.

73

Scanlon, B.R., l. Jolly, M. Sophocleous, and L. Zhang (2007), Global impacts of
conversions from natural to agricultural ecosystems on water resources:
Quantity versus quality, Water Resour. Res., 43(3), W03437.

Scott, C.A., W.G.M. Bastiaanssen, and M.U.D. Ahmad (2003), Mapping root
zone soil moisture using remotely sensed optical imagery, J. lrr. Drain.
Eng., 129(5), 326-335.

Seneviratne, S.l., D. Luthi, M. Litschi, and C. Schar (2006), Land-atmosphere
coupling and climate change in Europe, Nature, 443(7108), 205—209.

Singha, K., and SM. Gorelick (2005), Saline tracer visualized with three-
dimensional electrical resistivity tomography: F ield-scale spatial moment
analysis, Water Resour. Res., 41(5), W05023.

Slater, L., A. Binley, W. Daily, and R. Johnson (2000), Cross-hole electrical
imaging of a controlled saline tracer injection, J. Appl. Geoph., 44(2-3),
85—102.

74

Chapter Four

Integrating Geophysics and Hydrologic Models to Evaluate Land-use
Impacts on Groundwater Resources

Introduction

Land-use and land cover (LULC) changes have important consequences
on the hydrologic cycle. Transformation of vegetation that often results during
LULC change can lead to hydrological shifts at local to regional scales (Oconnell
et al. 1995; Pitman et al. 2004). The contrasts between plant species in their
ability to intercept, access and transpire water affect the atmospheric and
subsurface components of the hydrologic cycle (Nosetto et al. 2005; Pielke 2005;
Beighley et al. 2008). A major concern related to LULC change is its potential
effects on groundwater recharge, and hence the sustainability of a primary
source of freshwater around the world (Scanlon et al. 2005). LULC and LULC
change impacts not only the quantity (Farley et al. 2005; Scanlon et al. 2005), but
it is also closely linked to quality of the groundwater (Blinn and Bailey 2001;
Wayland et al. 2003; Jarvie et al. 2008). With changes in groundwater recharge
patterns, the interlinked systems of streams, wetlands, and riparian environments
along with their respective ecosystems are likely to be permanently altered (Rood

et al. 2008).

The concern for potentially rapid land-use changes due to climate,

population dynamics, social and economic development policy decisions both at

75

national and international arenas have increased in recent decades. The Clean
Development Mechanism-Afforestation/Reforestation provisions of the KYOTO
protocol for example could have unintended consequences on water resources
of the local environments while only minimally supporting the intended objective

(Trabucco et al. 2008).

Above ground vegetation changes lead to below ground alterations of
plant root structures. While the above ground implications of vegetation change
(e.g., precipitation interception, albedo, phenology, and leaf density) are readily
identiﬁable, below ground consequences are often not. However it is necessary
to accurately characterize and represent these changes in hydrologic,
ecosystem, and biological models for reliable assessments of land-use impacts
on our environment and natural resources. Globally, most land-use changes are
associated with deforestation for agriculture, afforestation of open lands or
natural conversion of grasslands to forests (Bryant et al. 1997). On average
woody species such as trees are more deeply rooted than grasses (Canadell et
al. 1996), but the bulk of the root mass (>~45%) is contained within the ﬁrst 30
cm for both woody and grass species (Jackson et al. 1996). Signiﬁcant
differences in root biomass distributions within a species and differences driven
by environmental aspects such as soil characteristics, nutrient availability, and
climate are known (Vogt et al. 1996; Coomes and Grubb 2000). However,
subsurface roots are generally described in models with limited parameters; i.e.
maximum rooting depth and the vertical root distribution. The two parameters

allow for the partitioning of water uptake based on the relative density of roots at

76

each depth (Jackson et al. 2000). It is clearly important to validate the
parameters of root distribution models at appropriate scales. However, such
validations have been rare because there haven’t been many practical
approaches to measure below ground impacts of vegetation over large spatial
and temporal scales. These ﬁndings illustrate the need to understand variability
in root distributions and dynamics in space and time based on geological,

biological, and hydrological properties.

The utility of geophysics for characterizing the shallow subsurface is well
recognized (Reynolds 1997). In recent years the value of certain methods for
monitoring and quantifying moisture changes in the shallow subsurface has been
reported by a number of researchers (Daily et al. 1992; Zhou et al. 2001; French
et al. 2002; Hanafy and al Hagrey 2006). The sensitivity of electrical conductivity
to changes in soil moisture contents allows the use of Direct Current Electrical
Resistivity Imaging (ERI) method for hydrologic investigations. Employing the
technique in a time-lapse mode, where the apparent resistivities are repeatedly
measured with the electrodes at the same locations, provides valuable data for
estimating transient characteristics of subsurface water ﬂuxes (Looms et al.
2008). Such information provides an excellent platform to describe and
understand recharge processes at large spatial scales, which has been difﬁcult
with traditional point observations. Moreover, the spatially explicit nature of
subsurface geophysical measurements enables direct comparisons of these
observations with above ground distributions of vegetation and their uptake

processes. The depth sensitivity of the measurements, with either commonly

77

practiced surface electrode resistivity measurements or borehole techniques can
help deﬁne the spatial distributions, and dynamics of root water abstraction
processes at ﬁeld scales. This approach avoids the practical difﬁculties and
invasive nature of trenching and other methods that have been traditionally
practiced for measuring root attributes to implement root water abstraction in

hydrologic and ecosystem models.

Here we present evidence of signiﬁcant differences in land-use driven
subsurface ﬂuxes that are important for groundwater resource sustainability. Soil
temperature, moisture, water quality, and groundwater observations for different
land-covers collected over multiple seasons are compared and contrasted. We
evaluate a simpliﬁed approach to quantify the broader impacts of land-use
differences on groundwater recharge by integrating spatially explicit soil moisture
observations at ﬁeld-scales with hydrologic modeling. Our ﬁeld observations
provide insight into below ground vegetation impacts at ﬁeld scales. These
observations along with the explored hydrologic and geophysical data integration
are expected to help the community develop a scientiﬁc consensus on the effects
and impacts of LULC on water resources (Petheram et al. 2002; Scanlon et al.

2005)

78

Study sites

 

0 75 150 Km

g TreverseCity
Q
:5
f
.2
3

East Lansing

Michigan

 

 

 

Figure 4-1. Locations of East Lansing and Traverse City, Michigan ﬁeld sites
where soil moisture dynamics were imaged with ERI.

Two ﬁeld sites were established in Michigan to monitor the spatiotemporal
interactions between climate, vegetation, and vadose zone soil moisture (Figure
4-1). The East Lansing study site is characterized by ~4m deep unsaturated zone
composed of 40-60 cm of clay loam underlain by a layer of medium to ﬁne sand
extending to at least 5m depth. The soil distribution is laterally similar across the
studied site transect. The vegetation at the site consists of a mature Maple forest
(Acer Saccharum) and a grassland, which are separated by a sharp managed
boundary. The grassland was established in 2004, prior to which it had been a

Honeylocust (Gleditsia Tn'acanthos) plantation. The forest side of the ecotone

79

develops a dense canopy during the growing season (~May-October), which
shades the ground during this time period. The trees are approximately 20-30
years old and are about 30m tall. The grassland is mowed once every year in

late-summer or early-fall.

The site was initially equipped in September 2006 to monitor climate and
vegetation impacts on near surface soil moisture distribution. Eighty four graphite
electrodes (CD12 cm x 30 cm) were permanently installed at 1.5m separations
along a 124.5m long transect, centered on the forest-grassland boundary for
time-lapse electrical resistivity measurements (Figure 4-2). The permanent
nature of the electrodes improves data quality, maximizing the potential to
identify soil moisture changes. Capacitance-type soil moisture probes were
installed at 20 cm and 80 cm depths in the forest in October 2006, and in the
grassland in February 2007 (Figure 4-2). Soil temperature beneath the respective
land-covers was monitored with two vertical temperature arrays with sensors at
5, 10, 20, 40, 80, 117, and 147 cm depths. The temperature sensors were
installed in the forest in December 2006 and the grassland in February 2007.
Additional temperature sensors were installed at 5 cm depth every 15m along the

array at the beginning of the study period.

Groundwater elevations were monitored in ~ 5m deep observation wells
installed below each land cover type in September 2007. Water table ﬂuctuations
and temperatures were logged every two hours with data logging

pressure/temperature transducers. Climate data for the site were obtained from a

80

Michigan Automated Weather Network (MAWN) station approximately 1.5km

from the site.

E n'l‘xult- .I'my

Temperature rnoixlun- Qullxﬁl“.

 

Figure 4-2. East Lansing ﬁeld site. A-B: electrode array with 84 electrodes at
1.5m separation. E3, E26, E70 are groundwater observation wells.
Temperature/soil moisture sensor locations have moisture probes at 20 8 80 cm
depth. Temperature probes are at 5, 10, 20, 40, 80,117, and 147cm depth at
each location.

The Traverse City study site is located within a glacial outwash plain. The
vadose zone consists of medium to ﬁne clean sands in the ﬁrst meter of the sub
surface and based on available USGS well logs, the ﬁrst 10 meters of the
subsurface (the approximate unsaturated zone) consists of a system of sands
and clays followed by sand and gravel to approximately 30 m depth. A sharp

managed boundary separates an approximately 20 year old pine plantation from

81

a recreational grassland (Figure 4-3). The trees have a thin canopy, but 10 to 15
m tall trees are closely spaced (~3-4 m), resulting in signiﬁcant shading. The
needles that accumulate on the forest ﬂoor result in a thick litter layer that is

signiﬁcantly different from the one that develops at the East Lansing study site.

lilt’cﬂ'k‘dl‘ awn

Ternpcraturelmoismre sensors

’
I

.l'

 

Figure 4-3. Traverse City study site. A-B: electrode array with 84 electrodes at
1.5m separation. B3 is the groundwater observation well. Moisture sensor
locations shown have probes at 20 and 80 cm depths, and temperature sensors
are at 5, 10, 20, 40, 80, 117, and 147 cm depths.

For long-term monitoring of soil moisture dynamics, the site was equipped
similarly to the East Lansing study site with a permanent 84 electrode (1.2 cm
diameter x 30 cm long) array for resistivity measurements. Capacitance-type soil
moisture probes were installed in the forest at 20 and 80 cm depths, in addition
to moisture probes under a canopy drip line and a tree line (close to tree stems)

in July, 2006. The grassland soil moisture was monitored since October, 2006.

82

Data was logged at least every 30 minutes in both land-covers. Sensor arrays for
monitoring soil temperature, constructed similar to those at East Lansing site,

were installed below both land-covers in December 2006.

Field observations

At the East Lansing site, the deciduous Maple trees leaf out in late April
and develop a dense canopy by mid to late June. The forest canopy insulates the
forest ﬂoor resulting in only sparse undergrowth that is generally no more than a
few centimeters tall. The trees shed their leaves in mid November, which
deposits a litter layer on the forest ﬂoor each year. In contrast the adjacent
grassland remains green until the ﬁrst snowfall or the onset of below freezing
temperatures which generally arrives in late December. Grasses again become

active in late March after snowmelt with warm spring temperatures.

The evergreen trees at the Traverse City site are much smaller in
diameter. The needle leafs are prominent only in the tree crowns, but a
signiﬁcantly thicker (5-15 cm) litter layer is present on the forest ﬂoor throughout
the year. The adjacent grassland here is frequently mowed during summer
months, resulting in ~<5 cm tall grasses during the growing season. The
Traverse City study site has a shorter growing season than East Lansing site

because of its relatively northern latitude.

The two groundwater observation wells at the East Lansing site show a

steady rise in water levels beginning in late fall (Figure 4-4). This corresponds to

83

the end of growing season and cessation of water use by the vegetation. The
rising trend continues through March, where a signiﬁcant rise in water table
corresponds to the primary snowmelt of the year. The difference in water table
elevations between the land covers disappear following snowmelt (Figure 44).
While the overall long term water table behavior is similar within both land
covers, the high frequency water table ﬂuctuations observed in the grassland
throughout the year are signiﬁcantly damped in the forest. After May, a steady
decline of water levels began, coinciding with the reemergence of leaves and

active transpiration.

 

 

 

350 250
E --Grasslend
0 --Forest A
T.’ E
8 3
O h-
: 400 ' #300 3
O O
* 3
5 8
3 5
'D n.
‘O 0
g 450 - #350 B
a n
§ g
o U-

500 i v v T f v v I T 400

 

 

Oct Nov Dec Jan Feb Mar Apr May Jun Jul

Figure 4-4. Groundwater observations since October 2007 at the East Lansing
study site. Depth to water from the surface is shown for each well. Since the
surface elevation at the forest well (E3, Figure 4-2) is 99 cm lower than the
surface elevation at the grassland well the two vertical axis are offset by a similar
amount. The wells are approximately 90 meters apart.

The Traverse City site has one USGS observation well located

approximately 100 m northeast of the electrode array (Figure 4-5). Water levels

84

monitored there since June 2006 show the annual recharge-discharge cycles.
Drops and rises in water table approximately coincide with the growing seasons
and snowmelts respectively. Stable winter snowpack with no intermittent melting
likely reduced recharge in 2007-2008 causing drainage and withdrawals to

continuously lower the water table from November to early April.

 

 

12
€12.51
8
a
3
.9
*5.
o 131
O

13.5 . - -

 

 

May '06 Oct‘06 Mar. '07 Sep.‘07 Feb. '08 Jul. '08

Figure 4-5. Groundwater observations at the Traverse City study site. Annual

recharge-discharge cycles related to the growing season and spring snowmelt
are evident in this well record.

Soil moisture measurements from the East Lansing site show substantial
losses during the summer, marked by large declines in soil moisture below both
land covers (Figure 4-6). Grassland moisture depletion however is signiﬁcantly

smaller than that below the forest.

85

 

Soil moisture (emf/err?)

 

 

 

 

o . .. . . T
s a I § § s 25 g s

 

2006 2007

Figure 4-6. Soil moisture measurements below forest and the grassland at the
East Lansing site.

Electrical Resistivity data acquisition and processing

Resistivity data were collected at both study sites using the Wenner and
dipole-dipole electrode conﬁgurations. Data acquisition, on each visit, resulted in
1134 subsurface apparent resistivity measurements with the Wenner array,
which was used for all the analysis and evaluations in this work. At the East
Lansing study site, 53 datasets were collected from October 2006 to July 2008.
Five datasets were collected at the Traverse City study site between July 2006
and June 2008, coinciding with important periods to capturing the impact of

vegetation and climate processes on soil moisture ﬂuctuations. Figure 4-7 and

86

Figure 4-8 illustrate two cross sectional views of the resistivity distributions in the

subsurface and land cover characteristics at the two sites in summer.

The resistivity data acquired at the East Lansing site were inverted for
resistivity changes using a difference inversion scheme (LaBrecque and Yang
2001 ), to focus on areas of the subsurface where soil moisture changes occur.
All difference inversions were performed with respect to a single reference
(background) dataset to preserve our ability to compare changes in soil moisture
through time. A dataset collected on March 30, 2007 was selected as the base
dataset, because the site had relatively uniform soil moisture across the ecotone
after spring snowmelt. We ﬁrst difference inverted all datasets relative to the
base dataset, and then performed a pixel to pixel recalculation of each dataset
using the percent difference and the base dataset. This approach minimizes
uncertainties associated with inverting each dataset separately, especially when
the goal is to identify differences in measured resistivity due to soil moisture

changes (Daily et al. 2005).

87

5.8 E 02:80:. 9m cozmtommﬁ £5 5 893: .5339sz

b_>=m_m9 oomtzmnaw .oEEswég 026E893. a 5:5 26 225 9:98.. “mum 05 Co 32> _mcozoomémeo N4. 059“.

 

 

(U1) uidao

allﬂ 1n... 4 [alt 5.381] 3.111111 . .5... till . . admit — OF
. .. . 9”.»
0!

RI {F 4... III HymrlIlIPl II I'll! I'D/II
4 u _.

em— 8. g a Q 8 ms ow mm 9 o:
a 2...... ewwﬁﬁﬁeﬁﬁw 4

88

.._o_oo :_ P2535 0...... 5.5536 £5 5 women: .cozsnEmﬁ
33550.. moatawnsm .oEEsmég 028E393. m 5:5 mam 63¢. E0 88%; 05 so 32> _mcozoowémEO .wé 9:9“.

9|!
mmmmm
Ego.

5.7!:

(w) llldiiCl

 

 

 

 

NR
Banana

89

Interpreting resistivity data

Subsurface resistivity distribution below the two land covers at the East
Lansing study site over time is shown in Figure 4-9. Distinct differences are
evident below the two land-covers in both space and time. The groundwater data
(Figure 4-4), which showed steady declines from June through November, and
rises starting November are also evident in the vadose zone resistivity
distributions. The within land-use resistivity variability is signiﬁcantly higher in the
forest compared to the grassland. A combination of potentially higher
transpiration and canopy interception contribute to this higher resistivity variability
observed in the forest. The grassland remained relatively homogeneous with
respect to resistivity, reﬂecting apparently stable soil moisture characteristics.
Slight increases in resistivity are evident during late summers in the grassland.
Signiﬁcantly lower interception here results in frequent soil wetting after
precipitation events, making the grassland moisture behavior more dependent on
precipitation frequency. Forest moisture dynamics in comparison are largely a
function of vegetation dynamics during the growing season. A similar behavior
was observed at the Traverse City study site between the grassland and the pine

forest.

90

.5505 Be Emu some .2 once.
05 new £22523 Emu .w 5mm dam—2 .mosmu c.2023 co 0% 95:3 "mom 05 um ncﬂwmma 05 new $28 05 26.3
5:45:56 32562 $33 .053 .9883. 26.3 2 33:5 .Eoé oomtzmnam so 55.2.9 .9an3 0...... .m-v 05mm

 

 

 

88_ Son _ 88— 88 _
WW SMWMFN MPOSM WPN
mmwmamm mm m wmanmwwm
muuuammmmm mnuwwmmmmw
III-PP'DF Flhillllbo
mi :ﬁ. w s
.
ﬂ.
.. 53w
0.
_. 53m
u
w
.. 531m...
.5390 use...
Foam

 

 

 

 

 

91

The lowest resistivities observed below the grassland and the forest is
approximately uniform across the land covers (Figure 4-9). A relatively higher
pore water conductivity in the forest however results in minimum resistivities in
the forest that are slightly lower than those below the grassland. Depth averaged
(0-2m and 2-4m) resistivities shows the potential contrast in rooting zones
between the two land covers and therefore the potential water withdrawal
differences from the vadose zone (Figure 4-10). Resistivity changes occurred
much deeper below the forest in 2006, potentially indicating the ability of trees to
withdraw water from deeper parts of the subsurface (Figure 4-10). A similar

increase in the 2-4m depth however did not occur in 2007.

92

.23 genome. 05 5 2265.0 5326505 602030 .029; £3, 02058 5505 totem 33m 05 .6 tea are. 05 acts“.
once. EYN 05 E 352362 55:5 .08: 3:25 mﬁaov EYN vcm Emé 05 5 $29 @2330. cams. .oTv oSoE

 

 

 

 

 

 

88 _ Bow _ 88 _ 88 _

r. «I N S “I W N
MMWmmmmemmMmmmmwmmm
mnuuwmmmmwmﬂuawwwmmw
D I F b P P I b I L P P I I I h P b I on

Tom—m.

n.

.ccwM

. M

..cmNO

.. 53w

ETN icon III .- ENIO 320.... III [Own m
SYN c.3380 Ir End 9.5.320 1... cc?

 

 

 

 

 

93

Electrical resistivity and soil moisture

To estimate moisture contents from the resistivity data, site and material-

speciﬁc relationships between resistivity (p) and volumetric water content (0)
were developed following ASTM standard 657-95. After oven drying at 105°C for

24 hours, ten soil samples from the ﬁeld site were wetted in ~4% water content
increments, homogenized, and placed in a test box for resistivity measurements.
For these samples, as for most soil materials, the p-6 relationship is well
approximated by a power function with coefﬁcient m (sand = 1.16; clay loam =

0.67), which is used in (Archie 1942):

1

S = (‘3‘?)3 (4-1)

where, S is saturation (volumetric water content / porosity), and p9, is bulk
resistivity of the soil at 100% saturation (sand = 71.53 Qm, clay loam = 68.15
Om; based on ﬁeld data). Since annual precipitation in Michigan greatly exceeds
evapotranspiration (ET), we assume constant ﬂuid conductivity values. We
calculated the soil moisture content for each resistivity value obtained from the
differential inversions. All resistivity data were corrected for temperature effects
using an empirical relationship (Hayley et al. 2007). The soil moisture calculated
with ERI is slightly underestimated compared to the point observations from the

site (Figure 4-11).

94

 

0.3

 
  
  
      

P
N
I;

.°
_0
I

Calculated moisture (aha/an?)

yet .04x-0. 04
R2-0.92

 

 

0 w r
0 0 . 1 0 .2 0.3
Observed moisture (cmalcms)

Figure 4-11. ERI estimated soil moisture and the observed moisture in the East

Lansing study site (Forest). The arithmetic average of the 20 and 80cm cm
moisture observations are shown.

Soil Moisture Dynamics and Relevance to Recharge

The calculated changes in soil moisture at the East Lansing site (Figure
4-12) show that the upper meter of the grassland subsurface has greater soil
moisture retention than the forest. Field sampling of soils at the study site
indicated that the soil texture characteristics within the two land-covers are
generally similar. Therefore it is likely that this difference is due to water use
differences between vegetation types, and potentially the relative compaction of

the soils in the grassland. Resistivity data from the Traverse City study site also

95

showed a similar contrast in resistivity distribution between the forest and the

grassland, with lower resistivities in the grassland.

 

 

 

 

a? 0.4
S H '
EON "' 00.. . .ee '. ...
é '0 °
‘6
E 0.2« . s'
.8 ° 9 0° 00° . o °o
g 00 ° 0 e o°°o° o
o

- cm 9 0 °
8 Ograssland O Q
E 0forest 0
LL!

0‘- V V V V 'F ‘7- ' V

a. s. s s s g ,3 s s
I 2007 2008

Figure 4-12. ERI calculated transient soil moisture (20—80cm cm arithmetic
average) in the grassland and the forest. Higher moisture levels in the grassland
are observed on each of the ERI data collection dates.

Knowledge of deep percolation is particularly important for groundwater
recharge studies. Such information is often indirectly inferred from water table
ﬂuctuations and other localized observations. In contrast, broader spatial aspects

of recharge characteristics can be better imaged with hydrogeophysical methods.

Figure 4-13 presents temporal soil moisture changes over the same time
interval below the forest and the grassland at the East Lansing study site. These
images show averages of an approximately 15m wide section of inverted data

below each respective land cover for each collection date. The important times

96

when changes pertinent to groundwater recharge are identiﬁed using letters on
the diagram. The early to mid growing seasons are represented by soil moisture
losses in the shallow subsurface (A), followed by deeper moisture losses in the
later part of the growing season (B), due to deeper drainage and surface
inﬁltration during the growing season. Moisture depletion is higher in the forest.
The main fall and spring recharge periods at the study site are represented by C
and D when deep wetting is prominent. In the fall, deep wetting occurs after

active transpiration. In spring, deep wetting is the result of snowmelt.

97

6.8 c. 0060006

0.0 cozmt0mmi 0...: :. w0m0E. .00: .062, Eu... 2 0.6 0.0.006 0.30.9: 60000 9.265 0.0 m 9.0 .m .< .on =0Ez6cm
0560 .3 0032.8 62.0.... 099.00. =8 50:. 05 m. 0 00000600 0.306E 0.0 06.8 BEE; 6:0 0000065 0.30.9:
0606:. 06.8 0.80 60.6.. 0... 0:0 96.0006 05 26.06 033 0.23 m T. :. 009.020 0.30.2: .60 .066E0... .m Tv 050.”.

 

 

 

98

Hydrologic simulations

We integrated the ERI derived soil moisture with the one dimensional
unsaturated zone ﬂow model, HYDRUS-1 D (Simi’rnek et al. 1998) to quantify the
recharge differences between the grassland and the forest. The modeling code

implements the well known Richard’s equation for variably saturated media;

60(h) _ 3 g
_at — az (K(h) 32 + K(h)) (4-2)

where 8(h) is the volumetric moisture content, t is time, 2 is elevation, and
K is the hydraulic conductivity. The K(h) and 6(h) functions in Equation 4-2 are
represented by van Genuchten parametric models (van Genuchten 1980). An
eight meter deep soil column was chosen for the simulations, which keeps the
model bottom boundary below the observed water table for the entire simulation
period from October 2006 to June 2008. The ﬂow domain was discretized into
801 cells of1 cm thickness. Separate model domains for the forest and the
grassland were developed to better integrate the broad soil and stratigraphic
characteristics observed within each land cover (Figure 4-14). The bottom
boundary of the model was deﬁned as a zero ﬂux boundary as a ﬁrst
approximation, but can be relaxed into a speciﬁed ﬂux boundary based on the
drainage during the snow cover period. We assume that lateral ﬂows are
insigniﬁcant in the system based on measured temporally consistent pore water
conductivity differences (~100uS/cm) in the two wells at the site which are

separated by ~90m.

99

Forest Grassland

 

 

 

 

Depth (cm)
0 0
so _ °'" '°"“ so - d" m
. 0: ﬁne sand . ii Fine and
215 -- '-' '
248 “
I ‘ a 30° - Loemy and
370
my loernlcleyloem
520
llne sand/loom
loemlflnd and
m m
Fm Model Field Model

Figure 4-14. A simpliﬁed illustration of the soil texture and stratigraphy (from soil
cores) in the forest and the grassland. Different soil texture units identiﬁed in ﬁeld
soil cores were combined to three representative layers in the vadose zone ﬂow
model (right). Images in this dissertation are presented in color.

The conceptual framework for integrating geophysical estimates of spatial
soil moisture with HYDRUS-1 D is illustrated in Figure 4-16. The subsurface soil
moisture distributions obtained using the ERI data in step 1 were ﬁrst used with

an inverse ﬂow model to obtain optimized soil parameters; 63, a, n, Ks (van

Genuchten 1980) for each of the soil texture layers in the one dimensional
models. Soil moisture estimates from late October to mid May of 2006/2007 and
2007/2008 were used in this step, because it should be possible to get better
estimates of the soils alone during this period of low soil evaporation and

vegetation water uses. Spatially averaged ERI derived soil moisture below the

100

two land-covers were used as calibration data for soil hydraulic parameter
optimization in each model. The starting soil moisture for the modeled soil
columns were obtained from a resistivity dataset collected on the same day the

simulations were started (October 18, 2006).

.I.‘0.C...“CK‘O‘CI‘OO‘II‘OC‘CC‘J.“IuI‘CI‘CI‘OI‘. IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. 1 :
. . Archie‘s . Soil moisture j
(ERI) fonnule. (time series) 3
............................................................................. \
Mstudypenod‘ """""" '3 ”season‘
Vadose zone E ' i
5 ﬂow model {-5 h1.h2.h3,h4 :
(menus-1D) g
g orwa , 5 aparameteni‘»
E \Mm/ i 5 (Feddes et 01/1974)
' : \hdose zone
5 Gm?“ 5 flow model
: '9 (menus-1o)
'. ................................... 5...................................5 L..................................3

 

 

Figure 4-15. Conceptual framework for integrating geophysical soil moisture
estimates with hydrologic modeling to estimate groundwater recharge. Soil
parameters as, a, n, Ks are optimized in stage 2, and plant water stress function
parameters h1, h2, h3, h4 , are calibrated in stage 3.

To deﬁne the root distribution within the model domains, we use
information extracted from ERI data collected at the research site. The root zone
distribution was interpreted based on the soil moisture changes computed

between early-growing seasons and peak growing seasons with ERI data from

101

2007 (Figure 4-17). This is likely a better representation of the actual root
distribution at the site. The sensitivity of the simulated hydrologic ﬂuxes to root

distribution was evaluated with a commonly used root distribution model;

——L'(z) = e“cz (4-3)
Lprofile

where Lr(z)/meﬁ/9 is the normalized root length density (0-1), 2 is depth,

and c is a biome dependent empirical parameter, - 0.9 (forest) (de Rosnay and
Polcher 1998). An initial set of root water abstraction parameters were selected
from the HYDRUS-1 D database (Simi’rnek et al. 1998), and the same parameter
set was used in both forest and grassland models.

Potential evapotranspiration rates at the site were calculated separately
for the grassland and the forest (Figure 4-18) using the modiﬁed Penman-
Monteith method (Monteith 1965), implemented in the Integrated Land Hydrology
Model (ILHM) (Hyndman et al. 2007). Snowpack and snowmelt conditions under
the two land-covers were assumed to be the same due to inadequate snowpack

observations at the site.

102

 

: ‘~,
1 ’.
.l . .. . .. - ’0 ..-
rt"
r"?
O
:é: 2‘ --...”’_-... ..
f I, '
g , e
I
8 3 - I «—
I
-o- Grassland
+ Forest

 

 

 

 

005 0 0.05 0.1 0.15
Moisture difference (om’lcm’)
Figure 4-16. Laterally averaged changes in soil moisture below the forest and the

grassland between early-growing (April to early May) and peak-growing periods
(July to August) (from Jayawickreme et al., 2008).

With the soil parameters estimated with inverse modeling, only limited
agreement between simulated and ERI estimated soil moisture was achieved
(Figure 4-19 and Figure 4-20). However, the soil hydraulic parameters obtained
are relatively representative of the soil texture types identiﬁed in ﬁeld cores
(Table 4-1 and Table 4-2). Simulations of hydrologic ﬂuxes for the entire study
period from October, 2006 to June, 2008 show that measured groundwater table
elevations are fairly well estimated in the forest (Figure 4-20), but the simulated
heads are relatively different in the grassland (Figure 4-21) due to among others

an inadequate description of boundary conditions.

103

Precipitetlon througfell (m)
LAI

 

 

 

 

Potential transpiration (m)
Potential soil evaporation (m)
0
‘1’

8335538855

 

Figure 4-17. Cumulative precipitation throughfall (a), representative leaf area
indexes (b) from MODIS, calculated cumulative potential transpiration (c) and soil
evaporation (d) inputs for hydrologic simulations from ILHM.

104

 

Simulated moisture (an’lan‘)

 

0.1 0.2 0.3 0.4

ERI estimated moisture (011301113)

Figure 4-18. Simulated vs. ERI estimated soil moisture in the forest. Data is from
10 depths within the ﬁrst 3m of the subsurface.

0.4

 

0.2

0.1-

Simulated moisture (crn3lcm3)

 

0.3 3"

     

0.47): + 0.12
R2 a 0.56 ‘t{

 

 

 

0

*

0.1 0.2 0.3 0.4

ERI estimated moisture (cchm’)

Figure 4-19. Simulated vs. ERI estimated soil moisture in the grassland with data
from nine depths in the ﬁrst 3m of the subsurface.

105

 

 

 

 

 

5.5. 95% Cl
Layer Parameter Estimate
C03“ Lower Upper
85 0.30 0.06 0.19 0.42
1 6.32 4.43 -2.40 15.03
n 1.50 0.11 1.28 1.72
Ks 1.46 2.16 -2.79 5.71
Os 0.29 0.03 0.23 0.35
2 5.40 2.61 0.25 10.55
n 1.36 0.08 1.19 1.53
Ks 4.00 5.86 -7.55 15.55
85 0.30 0.01 0.28 0.33
3 a 2.80 0.66 1.50 4.10
Ks 0.20 0.10 0.01 0.39

 

Table 4-1. Estimated soil hydraulic parameters for the three soil layers in the
forest with standard errors of regression coefﬁcients (S.E. Coeff) and conﬁdence

intervals.

 

 

 

 

 

5.5. 95% CI
Layer Parameter Estimate
C09“ Lower Upper
85 0.36 0.06 0.24 0.48
1 a 7.84 9.21 -10.33 26.00
n 1.20 0.08 1.04 1.36
Ks 0.87 2.65 -4.35 6.09
85 0.30 0.04 0.22 0.37
2 a 4.40 4.77 -5.00 13.80
n 1.20 0.07 1.07 1.33
Ks 8.00 20.97 -33.35 49.35
85 0.30 0.01 0.28 0.33
3 a 2.60 1.23 0.17 5.03
Ks 0.20 0.18 -0.15 0.55

 

Table 4-2. Estimated soil hydraulic parameters for the three soil layers in the

grassland.

 

 

 

Bottom pressure head (m)

-m easu red
—-sim ulated

 

 

3 . .. l . - ﬂ . . . I
séﬁ‘s‘fé yfrt'yeyﬁdb yﬁf 4‘9

Figure 4-20. Measured and simulated groundwater table in the forest with
optimized soil hydraulic parameter sets in the 1D-Hydrus model. These
observation data were not used in the optimization for hydraulic parameters in
the forest.

 

 

 

5.5
g 51
'0
Q
2
a 4.5‘
3
so
3 4
a l
E
:3 3.54
m

 

—-m easu red
W —simulated
3

is" ukftzssé‘ "6‘? \fm‘ySy'J’y Jys'f {9

Figure 4-21. Measured and simulated groundwater table in the grassland with
optimized soil hydraulic parameter.

 

107

The simulated summer growing season (June-October, 2007) soil
moisture compares reasonably well in the forest (Figure 4-22), but the agreement
is poor in the grassland where the simulated soil moisture content is lower than
the ERI estimates (Figure 4-23). The cumulative transpiration in the forest and
the grassland during the entire simulation period is 407 and 419mm respectively.
Based on the discrepancy between ERl estimated and model simulated summer
soil moisture in the grassland (Figure 4-23), it can be suggested that the
simulated grassland transpiration is likely higher than the actual amount. The
sensitivity of the simulated transpiration to root distribution (Table 4-3), root water
abstraction parameters (Table 4-4), and soil hydraulic parameters (Table 5-1)
indicate that simulated transpiration is likely to be most affected by the root water
abstraction parameters M and h2 (Feddes et al. 1974). These parameters (h1,
h2, h3, M) were initially assumed the same for both grassland and the forest.
Based on the above analysis however, it is evident that this was an over
simpliﬁcation. But with the available ERI estimates of root zone soil moisture, root
water uptake parameters for the grassland can be reﬁned until a reasonable
match between the simulated root zone soil moisture and ERI estimates is
obtained. This process should also improve the simulated hydrologic ﬂuxes in the

grassland.

108

0“—

y- tux-0.02
R’a 0.64

“l

0.24

Observed m oistu re

 

 

 

O 0.1 0.2 0.3 0.4
Sim u lated m olstu re

Figure 4-22. Observed (ERI calculated) vs. simulated moisture in the forest
during summer (June-October, 2007).

 

 

 

 

 

0.4
y-0.71x+0.06 0
R’= 0.28 ,

3 03¢
a
U)
'6
E
a 0.2'
d)
E
3’. .
8 o

0.11

0
0 0.1 0.2 0.3 0.4

Sim u lated m oistu re

Figure 4-23. Observed vs. simulated soil moisture in the grassland from June to
October, 2007.

109

 

Parameter deRosany ERI Root A96

2T (mm) 394.17 407.69 -3.32
h 5.47 5.28 3.61

Table 4-3. Sensitivity of simulated cumulative transpiration 7. T and bottom
pressure head (h) in the forest to different root distributions. ERI Root is root
distribution estimated with resistivity data, deRosany model used with a
distribution coefﬁcient of -0.9 (de Rosnay and Polcher1998).

 

 

 

 

 

Parameter {E {T h Base;)alue i 13:“:856

h 1 0.43 -9.71 4.33 -0.8 -0.88
-0.44 9.10 -3.88 -0.72

l12 0.44 -7.72 2.80 -2.5 -2.75
-0.59 8.54 -3.16 -2.25

h 3 0.00 0.00 0.00 -10 -11
0.00 0.00 0.00 -9

h 4 0.01 0.36 -0.15 -80 -88
0.01 0.43 0.18 -72

 

Table 4-4. Sensitivity of simulated evaporation (ZE), transpiration (Z T), and
model bottom pressure head (h) in the forest to a 110% adjustment in root water
abstraction parameters (Feddes et al. 1974). ZE, and X T are given as percent
changes relative the base model. h1-minimum pressure head for active
transpiration, h2-pressure- head for maximum transpiration efﬁciency, h3-limiting

pressure head where transpiration ceases to occur at maximum efﬁciency, and
h4-wilting point.

110

 

 

 

 

 

 

 

 

 

 

 

 

 

Model layer Parameter Base value 1: 10% base value {I 2T {E h
as 0.3 0.27 0.56 0.60 -2.44 1.21
0.33 -0.50 -0.11 2.21 -1.26
Alpha 6.32 5.688 0.01 5.31 0.06 -2.23
6.952 0.02 -4.24 -0.10 1.60
Layer 1
n 1 5 1.35 -0.55 8.39 2.41 -5.80
' 1.65 0.61 -9.25 -2.68 5.57
Ks 1.46 1.314 -0.06 -1.56 0.28 0.44
1.606 0.07 1.32 -0.32 -0.47
as 0.29 0.261 0.01 -8.20 -0.04 6.66
0.319 0.00 4.84 0.01 492
mpha 5.4 4.86 0.07 8.69 0.33 -4.41
5.94 -0.07 -9.28 0.31 5.04
Layer 2
n 1 36 1.224 -0.03 2.31 0.15 -6.56
' 1.496 -0.05 -14.00 0.21 10.34
Ks 4 3.6 0.02 -3.10 0.10 1.35
4.4 0.04 2.28 0.17 -1.17
as 0 3 0.27 0.15 -9.35 0.67 19.18
' 0.33 0.00 6.65 0.01 -18.99
Npha 2.8 2.52 0.02 2.82 -0.07 -1.79
3.08 0.01 -2.72 0.02 1.50
Layer 3
n 1 69 1.521 0.00 3.04 0.00 -1.87
' 1.859 0.00 -2.97 0.02 1.59
0.18 0.00 -0.85 0.01 0.44
Ks 0.2
0.22 0.00 0.53 0.00 -0.20

 

Table 4-5. Sensitivity of cumulative inﬁltration (2|), transpiration (Z T),

evaporation (ZE), and bottom pressure head (h) in the forest to a 110%

adjustment in soil parameters. 2|, 2 T, and 2E shown are percent changes

compared to the base model with optimized soil hydraulic parameters.

111

Discussion and Conclusion

In this chapter evidence for signiﬁcant hydrologic differences driven by
LULC characteristics at study sites in Michigan were presented. These
differences are readily identiﬁed using geophysical methods that are both non-
invasive and spatially explicit. Using time-lapse approaches, the spatial and
temporal changes in soil moisture were identiﬁed and were related to vegetation
and climate processes. Independent measurements of point soil moisture and
groundwater table data substantiated the validity of geophysical measurements
and subsequent interpretations.

Electrical resistivity measurements made at the study sites provided
important information about subsurface consequences of above ground
vegetation. Soil moisture and soil moisture change information derived from ERI
data can be used to asses subsurface root distributions as well as to quantify
spatial and temporal soil moisture use differences related to vegetation. Such
spatially explicit information is generally difﬁcult to obtain with point observations
alone. By integrating this type of information with hydrologic models, better
estimates of groundwater recharge and transpiration ﬂuxes at any site can be
obtained.

The approach devised in this study to directly integrate ERI estimated soil
moisture is straight forward and simple. However the subsurface systems
involved were inherently complex in terms of their properties and processes.
Therefore the reasonable estimates of soil moisture and other hydrologic ﬂuxes

obtained with the forest model in particular are promising. Improvements to the

112

one dimensional models with better descriptions of boundary conditions could
further improve their performance. Additionally, other approaches to integrate the
geophysical estimates with the hydrologic models should be explored. The
uncertainties of ERI estimated soil moisture in the grassland is particularly high,
especially during the summer months as a result of greater variability of
temperature and other environmental conditions compared to the forest.
Evaluating these uncertainties prior to integrating ERI estimated grassland soil
moisture with the hydrologic models could improve the relatively poor estimates

of hydrologic quantities obtained.

113

References

Archie, GE. 1942. The electrical resistivity log as an aid in determining some
reservoir characteristics. T. Am. Inst. Min. Metal. Eng. 146254-61.

Beighley, R.E., T. Dunne, et al. 2008. Impacts of climate variability and land use
alterations on frequency distributions of terrestrial runoff loading to coastal
waters in southern california. Journal of the American Water Resources
Association 44:62-74.

Blinn, D.W., and P.C.E. Bailey. 2001. Land-use inﬂuence on stream water quality
and diatom communities in Victoria, australia: A response to secondary
salinization. Hydrobiologia 466:231-244.

Bryant, D., D. Nielsen, et al. 1997. The last frontier forests: Ecosystems and
economies on the edge. p. 16. World Resources Institute, Washington,
DC.

Canadell, J., R.B. Jackson, et al. 1996. Maximum rooting depth of vegetation
types at the global scale. Oecologia 108:583-595.

Coomes, DA, and P.J. Grubb. 2000. Impacts of root competition in forests and

woodlands: A theoretical framework and review of experiments. Ecological
Monographs 70:171~207.

Daily, W., A. Ramirez, et al. 2005. Electrical resistance tomography-theory and
practice. p. 525. In D.K. Butler (ed.) Investigations in geophysics. Society
of Exploration Geophysics.

Daily, W., A. Ramirez, et al. 1992. Electrical resistivity tomography of vadose
water movement. Water Resources Research 28:1429-1442.

de Rosnay, P., and J. Polcher. 1998. Modelling root water uptake in a complex
land surface scheme coupled to a gcm. Hydrology and Earth System
Sciences 2239-255.

114

Farley, K.A., E.G. Jobbagy, et al. 2005. Effects of afforestation on water yield: A
global synthesis with implications for policy. Global Change Biology
11:1565-1576.

Feddes, R.A., E. Bresler, et al. 1974. Field-test of a modiﬁed numerical-model for
water uptake by root systems. Water Resources Research 10:1199-1206.

French, H.K., C. Hardbattle, et al. 2002. Monitoring snowmelt induced
unsaturated ﬂow and transport using electrical resistivity tomography.
Journal of Hydrology 267:273-284.

Hanafy, S., and SA. al Hagrey. 2006. Ground-penetrating radar tomography for
soil-moisture heterogeneity. Geophysics 71 :K9-K18.

Hayley, K., L.R. Bentley, et al. 2007. Low temperature dependence of electrical
resistivity: Implications for near surface geophysical monitoring.
Geophysical Research Letters 34118402.

Hyndman, D.W., A.D. Kendall, et al. 2007. Evaluating temporal and spatial
variations in recharge and streamﬂow using the integrated landscape
hydrology model (ilhm). In D.W. Hyndman, F.D. Day-Lewis, and K. Singha
(ed.) Subsurface hydrology: Data integration for properties and processes.
American Geophysical Union.

Jackson, R.B., J. Canadell, et al. 1996. A global analysis of root distributions for
terrestrial biomes. Oecologia 108:389-411.

Jackson, R.B., H.J. Schenk, et al. 2000. Belowground consequences of
vegetation change and their treatment in models. Ecological Applications
10:470-483.

Jarvie, H.P., P.J.A. Withers, et al. 2008. Inﬂuence of rural land use on
streamwater nutrients and their ecological signiﬁcance. Journal of
Hydrology 350:166-186.

115

LaBrecque, D.J., and X. Yang. 2001. Difference inversion of ert data: A fast
inversion method for 3-d in situ monitoring. Journal of Environmental &
Engineering Geophysics 6:83-89.

Looms, M.C., K.H. Jensen, et al. 2008. Monitoring unsaturated ﬂow and transport
using cross-borehole geophysical methods. Vadose Zone Journal 7:227-
237.

Monteith, J.L. 1965. Evaporation and environment. Symp. Soc. Exp. Biol. 19:19.

Nosetto, M.D., E.G. Jobbagy, et al. 2005. Land-use change and water losses:
The case of grassland afforestation across a soil textural gradient in
central argentina. Global Change Biology 11:1101-1117.

Oconnell, M.G., G.J. Oleary, et al. 1995. Potential groundwater recharge from
fallowing in north-west Victoria, australia. Agricultural Water Management
29:37-52.

Petheram, C., G. Walker, et al. 2002. Towards a framework for predicting
impacts of land-use on recharge: 1. A review of recharge studies in
australia. Australian Journal of Soil Research 402397-417.

Pielke, RA. 2005. Land use and climate change. Science 310:1625-1626.

Pitman, A.J., G.T. Narisma, et al. 2004. Impact of land cover change on the
climate of southwest western australia. Journal of Geophysical Research-
Atmospheres 109.

Reynolds, J.M. 1997. An introduction to applied and environmental geophysics.
John Wiley & Sons Ltd.

Rood, 8.3., J. Pan, et al. 2008. Declining summer ﬂows of rocky mountain rivers:
Changing seasonal hydrology and probable impacts on ﬂoodplain forests.
Journal of Hydrology 349:397-410.

Scanlon, B.R., R.C. Reedy, et al. 2005. Impact of land use and land cover
change on groundwater recharge and quality in the southwestern us.
Global Change Biology 11:1577-1593.

116

Trabucco, A., R.J. Zomer, et al. 2008. Climate change mitigation through
afforestation/reforestation: A global analysis of hydrologic impacts with
four case studies. Agriculture Ecosystems & Environment 126:81-97.

van Genuchten, MT 1980. A closed-form equation for predicting the hydraulic
conductivity of unsaturated soils. Soil Sci. Soc. Amer. J. 44:892-898.

Vogt, K.A., D.J. Vogt, et al. 1996. Review of root dynamics in forest ecosystems
grouped by climate, climatic forest type and species. Plant and Soil
187:159-219.

Wayland, K.G., D.T. Long, et al. 2003. Identifying relationships between baseﬂow
geochemistry and land use with synoptic sampling and r-mode factor
analysis. Journal of Environmental Quality 322180-190.

Zhou, Q.Y., J. Shimada, et al. 2001. Three-dimensional spatial and temporal
monitoring of soil water content using electrical resistivity tomography.
Water Resources Research 37:273-285.

Simﬂnek, J., M. Sejna, et al. 1998. Code for simulating the one-dimensional
movement of water, heat, and multiple solutes in variably saturated porous
media.

117

Chapter Five

Evaluating a range of approaches to derive hydrogeological information

from electrical resistivity data
Introduction

Geophysical methods have often been used to develop spatially extensive
estimates of subsurface hydrologic properties and characteristics (Daily et al.
1992; Hyndman and Gorelick 1996; Hyndman et al. 2000; Looms et al. 2008). In
some cases, time-lapse geophysical approaches can help discern temporal
changes that occur in the subsurface due to climate, vegetation, and other
processes. Recent advances in instrument technology, as well as data
acquisition and processing methods have resulted in many recent geophysical
applications in environmental research (Slater et al. 2000; Pellerin 2002; al
Hagrey 2007; Amidu and Dunbar 2007; Looms et al. 2008). Direct current
electrical resistivity imaging (ERI) is a promising geophysical method that is
commonly used for subsurface hydrologic characterizations (Binley et al. 2002;
Israil et al. 2006; al Hagrey 2007; Nosetto et al. 2007). The sensitivity of electrical
conductivity to changes in soil moisture and solute concentrations, and the
potential ability of ERI to reasonably capture and quantify these changes in
space and time with minimal disturbance are some of the major reasons for the
popularity of ERI within the hydrologic science community (Kemna et al. 2002;

French and Binley 2004; Israil et al. 2006).

118

Despite the signiﬁcant use of ERI method for environmental
investigations, its practical uses outside of experimental settings are complicated
by a variety of factors. Survey techniques, measurement physics, inversions and
other data processing methods are still being developed and improved. Wlth
time-lapse measurements necessary to image transient and dynamic subsurface
processes, further uncertainties may emerge due to inconsistent data acquisition
strategies and effects of natural environmental variability. These are inherently
difﬁcult to fully capture and describe in ﬁeld settings. Soil temperature and
moisture are two of the principal variables affecting changes in soil conductivity.
Other biological, geochemical, and physical processes also have the potential to
inﬂuence the measurements, allowing for potential misinterpretation of resistivity
observations. Conversion of electrical resistivity estimates to other quantities of
interest, such as soil moisture is generally done with two commonly accepted

models (Archie 1942; Topp et al. 1980).

This chapter provides an overview of the ERI method based on our
experience with imaging long-term subsurface soil moisture dynamics at a ﬁeld
site in East Lansing, Michigan. The literature does not provide extensive
information on the opportunities provided by this non-invasive method for gaining
insights into subsurface environments or the practical challenges to its effective
long term use in natural settings. Resistivity data collected for two years at our
East Lansing research site are used to examine the effects of data acquisition
and processing strategies for quantifying the natural variability of subsurface

electrical resistivity. We also evaluate the inﬂuence of these choices on our ability

119

to interpret and quantify soil moisture dynamics with the electrical resistivity data.
Finally, we discuss approaches to further enhance the reliability of hydrologic

quantities extracted from time-lapsed data.

Electrical Resistivity Imaging (ERI)

ERI is a method that estimates the subsurface distribution of electrical
resistivity by introducing a known current into the ground using a pair of
electrodes, and measuring the resulting potential difference between separate
pairs of electrodes. The measured resistance is a function of electrical
conductivity of the subsurface constituents (grain matrix, porosity, water content
and conductance). By deploying arrays with multiple electrodes, many
independent measurements of resistivity can be quickly collected in a repeatable
manner. Several electrode conﬁgurations can be used for data acquisition with
varying resolutions and sensitivities (Reynolds 1997; Dahlin and Lake 1998; Lake
2000; Dahlin and Zhou 2004). At our research site, we acquired data with the
common Wenner and Dipole-Dipole electrode conﬁgurations (Figure 5-1)
because of their robustness. The spatial data coverage for these array types at

the research site are illustrated in Figure 5-2.

120

Dipob-dipolo

 

 

 

 

8 I'll I
A +-—-—-’ B <— ——————— —> M«—-—--—-> N
—. 1 O *—
Wenner
a a a
A e— ————— >M<— ————— > "4— ————— ) B
+ O——*-—. 0'—

Figure 5-1 Electrode positions for Wenner (top) and Dipole-dipole (bottom)
electrode conﬁgurations. A, B are current electrodes and M, N are potential
electrodes. ‘a’ is the distance between electrodes used for measurements, and n
is an integer.

0 31 62 03 L21 (m)

 

OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO

 

Apparent depth (m)

1 5 — . . .o...'.:.'.‘.:.‘.‘.:.:.‘.t.‘.:.‘.‘.:.‘3333333333.;‘3‘... . .
. 0.00.00.00.00. .
3 0 _ OW . . . .
(m )

 

 

3o- Wenner
(M)

Figure 5-2. Data coverage for Wenner and dipole-dipole electrode conﬁgurations
for an 84 electrode array with 1.5m electrode separations. Higher data density
near the electrode positions is obtained with the dipole-dipole conﬁguration
compared to Wenner, where data coverage is evenly distributed. 1453 and 1134
measurements are obtained with these dipole-dipole and Wenner arrays
respectively.

121

ERI and environmental variability

The solid constituents of shallow soils are generally good insulators. As a
result, the electrical conductivity measured with ERI is primarily from ionic
conduction through water in the pore spaces of soils, sediments, and rocks. This
dependency between electrical conductivity and pore-water can thus be exploited
to obtain quantitative estimates of soil moisture. While soil moisture variability is
often the primary reason for temporal changes in subsurface resistivity, pore-
water conductivity and soil temperature may also have signiﬁcant affects on ERI
measurements that may limit the ability to compare time-lapse measurements in
some environments. These variables also vary spatially, and thus their effects
tend to be different along the electrode array where the data is collected. Soil
structure, porosity, and other physical characteristics can also affect conductivity
over time, but these are relatively less important, particularly for seasonal or

annual estimates.

Electrical conductivity and volumetric soil moisture

A general approach to estimate soil moisture from ERI data is to relate the
measured electrical conductivity to water saturation through petrophysical
models. The most general is Archie’s law (Archie 1942); which describes the
electrical conduction of a material, related to the conductivity and water
saturation in the pore spaces. The relationship, derived through experimental

results, can be written in terms of electrical resistivity as;

122

pi = aprmSJ.“ (54)

where pw is the resistivity of the pore ﬂuid, <19 is porosity, and S is fractional water
saturation; S=0/CD where 9 is the water content. The exponential constant m,
known as the cementation factor, depends on the interconnectedness of pore
spaces; 1.3 is commonly used for unconsolidated clean sand formations (Knight
and Endres 2005). The constant a is a ﬁtting parameter with values ranging
between 0.62 and 2.45, however it is often considered unity in practice. The
exponent n, called the saturation exponent, accounts for the connectedness of
water in the partially saturated soils (Knight and Endres 2005). Archie’s model
assumes the primary conduction occurs through the pore ﬂuid. Therefore in
places where clay and other conductive materials are present, the effective
contributions of these materials to the bulk conductivity should be assessed by
integrating other empirical components with Archie’s equation (Waxman and
Smits 1968). Alternatively, the contribution of clay and other conductive
components can be accounted for by measuring the resistivity-soil moisture

relationship for site speciﬁc soils in a laboratory setting.

A common approach to estimating Archie’s parameters for a particular soil
is to perform laboratory measurements using standard procedures (ASTM
standard 657-95). Such methods however suffer from a number of practical
difﬁculties such as the need to extensively disturb the soil samples to facilitate
even wetting and packing in a test box. However, a general relationship obtained

with multiple sample tests in combination with other independent estimates of

123

saturated soil resistivities and porosities from undisturbed field samples are often
sufﬁcient to describe the resistivity-soil moisture relationship for a particular soil.
In this study, a relationship was constructed between laboratory measurements
of soil saturation and resistivity to estimate m for the simpliﬁed form of the

Archie’s equation.

S = (5")E (5-2)

This equation was used to convert ﬁeld resistivity images into soil moisture based

on estimates of porosity from soil cores collected at the site and saturated

resistivity values from ﬁeld datasets. p3 here is saturated bulk resistivity.

 

 

 

Parameter Mean Stdev Max Min Soil

(0 0.47 0.03 0.51 0.43 I

m 0.67 0.09 0.84 0.6 C 3"
("=6)

ps 68.15 - - -

(D 0.39 0.03 0.43 0.36

m 1.16 0.11 1.29 0.99 53""
ln=6l

p, 71.53 . - 98 65

 

Table 5-1. Parameters for Archie’s equation from laboratory measurements of
soils at the study site. p. is obtained from ﬁeld resistivity data (for sand estimated
with resistivity data from below the water table, and for clay estimated from
datasets collected when the site was extremely wet).

124

 

0:4

—200m soil moisture

—800m soil moisture
o»EFuInuinouldsoﬂrnoﬂnuntGN>412cnlaMgJ

 

 

 

 

930.3-
E
3
E s
3 .
9 1"? \‘ﬁ’l "w
a . _. v. « R o
.9; 1 ° " o
E " l a -
i; 0.1! ‘\'~ ' e
U) Q
«J H
o v I ' v i I U
o .-
3 = s 8 s z 8 e g
25 $2 25 :5 23 :3 S; 2; c:

2006 2007 I 2008

Figure 5-3. Soil moisture estimates from ERI (dipole-dipole) data compared to
other independent measurements of soil moisturelat the study site.

The ERI estimated soil moisture, based on laboratory derived parameters
(Table 5-1) and equation 5-1, correlates well with point soil moisture
measurements made at the site with automated data loggers (Figure 5-3). Figure
5-4 illustrates the potential uncertainty in ERI estimated soil moisture due to
uncertainty in Archie’s parameters. The ﬁtting parameter m has a stronger
inﬂuence on soil moisture estimates for higher resistivities or lower soil moisture
contents. Estimation error of saturated resistivity has more impact on the
computed soil moisture at lower resistivity ranges, which may complicate

attempts to identify the elevation of the capillary fringe from ERI (Figure 5-5).
125

Clay loam

 

 

 

 

1 1
g m
90.76
“3‘ 00.87
“0.58
.3 0.e< ;
S .
0.4 1 0
II 3
’ +
' . o 3
-_§§ i
0 V I T j
0 100 200 300 4.00
Send
1 1 . m
+1.27
0'. ‘ 3 01.16
C105
0.6« 3:
%
0.4 1 a
.3
0.2 1
o 1' T 1' 1
0 200 400 600 800
Resistivity (Ohm-m)

 

 

 

 

 

1- 4
p8
0 8 ° 474.97
‘ ‘ - 000.10
*, -01.34
0.0- .
" 4»
0.4- o
- O
. o
' O
0.2 1 " § §
0 § . ‘
0 V f V V
0 100 200 300 400
1 ' .1.. p‘
' +0800
0-3‘ 3' 071.53
- 6 -06.00
0.61
0.4-
0.21
0 V 1 V 1
0 200 400 000 800

Figure 5-4. The sensitivity of ERI derived saturation to m, and p8 estimates. The
high (grassland-98 Ohm-m) and low (forest-65 Ohm-m) values for saturated
resistivity were derived from ﬁeld resistivity data. Some of the differences in
saturated resistivity between the grassland and forest are due to observed
groundwater conductivity differences.

126

 

 

1

01/13
02/02
02/22
03113
04102
04122
05/12

9 §

2007 2008

"/14

Figure 5-5. Water table elevation observations and resistivity measurements in
the forest (a), grassland (b) from November 2007 to June 2008. The correlation
between water table elevation and contours of the resistivity estimates is much
clearer in the forest relative to the grassland. Images in this dissertation are
presented in color.

Figure 5-5 illustrate the temporal changes in resistivity measured near the
grassland (spatially averaged data within a ~3 zone near the well) and forest

(spatially averaged data within ~3m zone, ~10m away from the well) groundwater

127

observation wells at the study site. The changes in water table elevation are
similar to the pattern of resistivity changes in the forest, but this relationship is
relatively less clear in the grassland. The inability of ERI to image the water table
in the grassland could result from a number of different factors. The presence of
the shallow high conductivity zone (low resistivity) in the grassland is one
contributing factor that likely affects the resolution of ERI measurement below it
resulting in a lower vertical resistivity contrast. Alternatively, it is also possible
that the vertical soil moisture contrast is low in the grassland due to the lower soil
moisture deﬁcit there at the end of the growing season, which recovers relatively

quickly with fall precipitation.

Soil temperature and electrical conductivity

The electrical conductivity of pore ﬂuids is inﬂuenced by soil temperatures.
Temperature variations are largest in the shallow subsurface (0-1 m), especially
over relatively short time frames. However, over seasonal time scales,
temperature also vary down to the water table at this site. Additionally, soil
temperatures can differ spatially, particularly during warm summer months

depending on land-cover characteristics and associated shading (Figure 5-6).

Observations made at our study site show that summer diurnal soil

temperatures in the top 10 centimeters change by ~ 20°C in the grassland but
only ~ 10°C in the forest (Figure 5-6, inset). At the same site, consistent long-

terrn groundwater temperature differences of ~2°C were also recorded between

128

the land-covers. Our temperature observations show that overall, the subsurface
in the grassland is ~2~3°C warmer than the forest throughout the year due to

differences in solar insolation.

 

 

35

Tm
3,, ‘ ..l at“ ’31:.“

{Ml II“ .. 11'.
tl‘ '
.Jl

 

 

 

 

 

    

Soil temperature (’C)
8 z;

.x 1‘1

 

 

 

1 1 1 1 1 1' 1 ' 1'6 '
4° 3°40 11° 4° 9 1° 191°
0’111 0““ 01"0 0“"; 09“ «‘9‘ «11% 011“ of“ 0409“

Figure 5-6. Seasonal grassland soil temperature ﬂuctuations at the study site.
Inset shows diurnal soil temperature differences between the grassland and the
forest stand over a span of three days in June.

There are several models to describe the effect of temperature on

electrical conductivity of soils. In the 0-25°C range, the relationship can be

approximated using

0' _ m('I‘§ g—25)+1 (5'3)
Std_( m(Tti-25)+1 ) i

(Hayley et al. 2007); where Tstd is a reference temperature. ostd is conductivity at

the reference temperature Tstd. Ti and o; are measured temperature and

129

electrical conductivity values, and m is a material dependent temperature
coefﬁcient (Hayley et al. 2007). Figure 5-7 illustrates the effect of different soil
temperatures on the resistivities typically measured at the research site. Over
seasonal scales and at higher soil resistivity ranges the temperature induced
resistivity variations could be relatively high (~25%). The effect of temperature
however is more important in lower resistivity ranges near soil water saturation.
The resistivity-soil moisture relationship computed with Archie’s equation shows

that a ~12°C temperature change leads to a ~20% difference in the calculated

soil moisture at lower resistivities (a saturated resistivity of 730m, m coefﬁcient of
1.16 used in the calculation). The effect of m in the temperature-resistivity
relationship is less important at lower measured resistivities. However, at the
4000m and 28°C illustrated, m (0.018-0.022) range results in 3 ~13% resistivity

difference in the correction (Figure 5-8).

      
     

 

 

 

 

500- 500-
-—28°C
A 4001 A 4004 _24°C
5 E --- 20°C
" 9 - . 18°C
E 300‘ g 300-
15 '2 seem 12°C
8 __ . E
2 200‘ 2° c g 200-
‘ “III-24°C l
3 --- 20°C 3
100‘ - . 16.0 1001 I“...
m... 12°C
0 U V T 1 o I T V T I
0 200 400 800 800 0 0.2 0.4 0.8 0.8 1
Tom p. corrected bulk resistivity (0m) Saturation (ﬁle)

Figure 5-7. The inﬂuence of temperature on resistivity and the resulting
difference in calculated soil moisture with Archie’s equation.

130

Bulr resistivity (an)

 

 

 

0 v . . .
0 200 400 600 000
Temp. cetrected built resistivity (0m)

Figure 5-8. The inﬂuence of m on soil resistivity-temperature dependence at two
different soil temperatures (28°C and 12°C). The sensitivity to m is important at
higher temperature ranges.

 

0 31 62 93 124(m)

 

 

—.g i I."
“01001151000010
0.10.4‘7‘7 ‘ v-v-NN

Figure 5-9. Percent change in resistivity from August 10, 2007 to January 11,
2008 without (a) and with (b) temperature correction. Prominent changes in the
grassland (a) do not appear once the temperature correction is added (b).
Images in this dissertation are presented in color.

131

The differences computed between two resistivity datasets from the East
Lansing ﬁeld study site are compared in Figure 5-9. The noticeable change in
grassland resistivity (Figure 5-9a), prior to temperature correction do not appear
in the difference computed with the temperature corrected data (Figure 5-9b). In
general the temperature variability is larger in the grassland. With limited spatial
measurements of temperature the uncertainty due to this variation is likely to be

higher in the grassland.

The-effect of pore-water conductivity

Dissolved ion concentrations in the subsurface depend on a number of
factors. Biological degradation and ion leaching processes along with
concentration of solutes due to evaporation and transpiration are important
factors that may introduce land-use driven pore water conductivity differences in
the vadose zone. Anthropogenic inputs such as road salt and agricultural
fertilizers are other potential considerations. Seasonal differences in biological
processes and variations in ion concentrations in the precipitation (Rein et al.

2004) can also introduce temporal variations in water conductivity.

Groundwater conductivity measured in two observation wells at the
research site show that the groundwater conductivity differs between the land
covers (Figure 5-10a). The difference is fairly consistent over the course of the
measured period. The relative inﬂuence of this difference on measured resistivity

in the grassland and the forest can be evaluated with the Archie’s equation (5-1).

132

.cozmSoﬁo 05 5 new:
we; .m__om 25mm *0 36220220 .N .6 m:_m> : cm can 9. WE .2928 use“: .3628 $9.” 53> :8 < .SV 5:233

be 385% 30:? .2 5:23 m.e_:2< :0 women 22.3.8.8 95:250.. $2652 .8536 3:45 new $53.20.".
.3 2.0; 9.0st030 528 new ecsmmﬁm 05 E 96 52¢me 05 “a noSwmoE £28328 5533390 .o _..m 059“.

 

 

 

 

Esme 3.2.32.8 6.25 23
0mm oak one can cme one» 32. Sn Em
E s O l I 09‘
I I I I I 95.320101
IIIIIIII 8 320...}
4444444 53m. 53%
Q C 4 up. U
o ‘2 n
m. - III. e.
4 4 58 m. 58 m...
e e e e M H
e e e 101 m.
e U are.
4 . noon w 2 :69” m
nonms e e w .1
2.9.64 (
.5 Saw- 3
new can

 

 

 

 

 

133

The effect of pore water conductivity on bulk electrical conductivity is
largest at lower water saturations and much smaller in the saturated zone. When
pore-water conductivities are similar along with similar subsurface soil and
sediment characteristics, the measured resistivities below the water table are
expected to be similar. However, at the study site, with comparable soil
conditions, a consistent discrepancy in resistivities between the grassland and

the forest is observed below the water table (Figure 5-11).

134

3.8 c. coucmmoa 9m cozmtomwﬁ £5 E momma... .3353 23399598 9.: 0:280

9: ho 02m 555 :0 .m__E_m 9m 28m 9: new umﬁotoo 9322...:2 m_ 5505 Sac .ucsmmﬂm «.855 05 5 8:35:28
5632305 35¢.on .026. 9: 2 can 29.: m_ ucmﬁmﬂm 05 5 $8.26 Ea: E Viv 033 .995 26.3 mo=_m>

DEERE 3:9: .592 “Em c5335 05 26.3 moocoaotﬁ 33:03:00 5526505 Io 3:255 33:20.“. .36 2:9”.

%%%%@%%ﬁ%
Ed I ,lrel

 

 

 

 

 

“IOIQI. u . A . . t
I , . . .....

EEE§ﬁ= ng =:o

Basso moan.

135

The measured bulk electrical resistivities are approximately 30-500m
higher in the grassland (Figure 5-12), which is consistent with the measured
lower groundwater conductivity. An adjustment for the conductivity difference
(~137pS/cm) would result in a resistivity adjustment of ~26%, which lowers the
measured saturated zone grassland resistivity (~980m) to ~720m,
approximately equivalent to that measured in the saturated zone of the forest
(~70-750m). In the unsaturated zone a similar adjustment would further reduce
the measured resistivity in the grassland. However, no measurements of pore

water conductivity in the vadose zone are available for this purpose.

a? O uJunO‘l_08

 

D a _1 AMay18_08
'3 ‘4 AMay02_08
‘3 “ A _2 oMar22_oa
E N oMar14_08
5 “ .3
g water table (max)
watertabmminj" I:
.200 .100 o 100

Pgrassland - Pforest

Figure 5-12. Measured bulk electrical resistivity difference between the grassland
and the forest (pgrassland-pforest). Higher resistivities below the water table in
the grassland are consistent with the lower measured groundwater conductivities
there. The shaded area indicates the approximate range of water table change in
the grassland between March and June 2008.

136

ERI and time-lapse differencing

The primary objective of time-lapse ERI is to obtain temporal variations of
subsurface hydraulic quantities of interest. Two approaches are often adopted for
computing the differences between time-lapse ERI datasets. The simplest is to
invert each ERI dataset independently and then compute the difference with
respect to a reference background dataset. This approach however is often
considered unreliable because data related errors tend to mask small variations

in the amount of moisture or solute (Daily et al. 2005).

An alternative is to use a difference inversion algorithm (LaBrecque and
Yang 2001), where the ratios of two data sets or the differences in the data itself

are associated with the inversion. The ratio approach can be expressed as;

d
dn = aid“ (54)

where dn the new data vector with normalized data, dt is monitoring or time lapse

data, do is data for the reference state or base data, and dh is an arbitrary
homogenous conductivity representative of data that would be observed in a

homogeneous subsurface. The dn data set is then inverted to obtain changes

relative to the reference dh (Daily and Owen 1991).

All data processing in this study were based on the difference inversion
approach above. One contentious issue related to this method is the selection of

an appropriate base dataset t for the difference computations. The choice of

137

base dataset was complicated by the fact that the resistivity distributions were
highly variable at the research site both in space and time, driven by the sharp
vegetation difference. How these contrasts embedded in a base dataset would
affect and propagate through subsequent datasets processed relative to it were
initially difficult to evaluate. A simple analysis of the potential errors and
uncertainties related to base dataset selection is therefore presented in this

section as a foundation for further work.

The following analysis evaluates three resistivity datasets, B1 (collected
on October, 2006, 82 (July 2007), and B3 (January 2008) from the study site,
and difference inversion related computations made with them as base datasets.
The characteristics of the three datasets as acquired from the ﬁeld site are
illustrated in Figure 5-13. Dataset B1 has a signiﬁcant spatial coverage of higher
resistivities and a larger contrast with the forest. 82 in comparison has a sharper
contrast between the shallow and deep soils below the forest, and B3 has a
relatively uniform resistivity distribution across the ecotone, which is most

uniformly wet compare to B1 and BZ (Figure 5-13).

138

.86.... E 8282a 2m cozmtowwﬁ £5 5 892:. airman 850E
5552.. 858:6 .2 com: 2m; 52; 6% >35 9: E0: £3 2% 33:39 925 he 85:93.20 Eon. .m Tm 9:9“.

 

 

 

no A n. e e I 1 1 1 z
00 oo 00 on» 00 0o. 00. a.» 06 00
E 0 [genre gll'll
«.988 6.3%.. . _
amﬁooméaq

 

I. 31.3.5-5 .. [655 1 r

F as;

2388 £528 ., E

 

 

. . it: mﬁm
WI . .- ,,,,, ii .,;,.ll. .3.»! . its}... 0

139

The comparisons between independent inversions of the three datasets
with the same datasets derived using difference inversions and the other two
datasets as base data is illustrated in Figure 5-14. The two approaches clearly
lead to relatively different outcomes. Overall the direct inversions appear to result
in lower inverted resistivities compared to difference inversions. Figure 5-14 b &
c shows that difference inversion of B1 with BB as the base result in a better
comparison with independently inverted B1. Using B3 however for computing 82
results in a poor correlation between the two inversions. B3 which has the lowest
resistivity range among the three datasets is relatively poorly computed with the
difference inversion when datasets with larger resistivity ranges (B1, 82) are
used as base datasets (Figure 5-14, 9 & h). Overall, all selected datasets lead to
different inverted resistivities with no particular dataset performing better or

worse as a suitable base dataset.

140

 

 

 

  
     

 

 

 

 

 

 

 

 

 

 

 

 

    
  

   

    

 

 

 

 

 

 

 

 

 

 

 

 

    

   

 

 

 

 

 

 

 

 

 

506 500 500
R’=o.93 n’aoss
coo~ .. goo- RMSE'W-N .. 400-1 RMSE-13.04
. . ..a I"
300— Jam ..v‘ F300-
3 To ‘4‘" in
zoo- aoo- ;_ 200-
100- roo~ 100-
a b c
o I I l I o I I I I o I I I I
O 100 200 300 400 500 0 100 200 300 400 600 0 100 200 300 400 500
81 >> 81 82 » at B3 » 81
500 500 500
R’-0.9e _ . R5031 .
- RMSE-12. “ 4 ’ ,. RMSE-19.17 -'
400 3:” ‘ , 400 / 400 .1553“
300- ‘ -, ‘ 300- aoo~
B . 8 8 -
200- 200- 200~
100~ 100' 100-
d e f
o I I I I o I I I I o I I I I
O 100 200 300 400 500 O 100 200 300 400 500 O 100 200 300 400 500
B1>>82 32>>B2 B3>>32
500 500 500
R’IOJO III-0.77
400‘ RMSE=14.02 400+ RMSE-13.76 400-
300- 1300- 300d
8 8 B ,"
200- 200- 200-
100- 100.. 1004
g l
o I I I I o I I I I 0 I I I I
0 100 200 300 400 500 0 100 200 300 400 500 0 100 200 300 400 600
81 >> 83 82 >> 83 83 >> 83

Figure 5-14. Comparisons of base dataset effects. B1, B2, and BB are three
separate inverted field datasets. The notation B1>>B3 refers BB calculation with
B1 as the base dataset in the difference inversion. Only the data from the top 5m
were selected for the analysis.

A similar comparison where one particular dataset was derived using the
other two as the base dataset also was performed for this analysis (Figure 5-15).
Most differences among the three are associated with computing the dataset 82,

which has a signiﬁcant high-low resistivity contrast in the forest (Figure 5-13).
141

The spatial distribution of these differences illustrate that the choice of base

dataset may be an important consideration in the forest, where large resistivity

heterogeneities is common throughout the year (Figure 5-16)

 

 

 

soc .
rat-0.93 3
“0‘ Ruse-usa.?
‘-
r '0
“£30m .~
A C
3200‘ -.
100-
0 I I I I
o 100 200 300 400 500

 

33 >> 81

600

 

400—
¥aoo~
A

3200«

100‘

 

Rf-oai
RMSE-1&1

 

 

0
0

 

160 260 350 450 500
33>>32

 

 

 

 

 

500
RtuﬂJa

‘00. RMSE 115.00
3
R300—
8two-1

100'

0 I r r I
0 100 200 300 400 800

31 » 33

Figure 5-15. Comparison of computed B1, 82, B3 with different with respect to
others in the difference inversion.

142

.560 E 02:30:. 2m cozmtommﬁ m_£ E wommg .392 05 5 2332925
woman E2662 9: cc 8595 :55. m mm; «82% $3 ho 8.9.0 9:. .m_.-m 2:9... E 5505 2m “33% some comics
2.23950 .3898 mm can .Nm .5 53> 2.22mi: 855:5 EB.— wmocmEE—u .o 5:39sz _mzmaw .36 059m

/. \ \ s s A

 

 

smasasmmﬁ

 

 

 

-1ﬂw.....§-Hl- .. - ..-,-,l.,.l-.,;.l inn.” ll “lurifrllrlllf st
wmmnmmmvﬁmxsa- . l .. . . I, i . -
m... I I :- .,Il . .m.ﬂ...riliu.l.t,l.mwh mural-MPH”.- .Ii I in m I i a inninﬂIuﬁI I In!!- ., III I I I..- an . MIA II. ., I I “am
:mxxmaémxxwa

III! 11.15)). i . . l. Ill [I'lﬂllﬂl‘ delizad rb ﬂirting“. $333.11..

 

a .
.

u
I
l I r u
i will lulu-Ii in ., h. ll. .,
F...II:E¢4!:T .rIFAiIIAII.‘ tr... uh I. , I,.I titan-I l. :1: III-Ir ...... , l . , l n I ‘II W.G.ulil. MI I, I F l. tn. , éiﬂ‘ In

143

Soil moisture estimates with resistivity ratios

An alternative approach to derive soil moisture change information from
ERI is to use resistivity ratios in combination with laboratory derived
petrophysical relationships. By focusing only on resistivity changes at a location,
the uncertainties related to spatial variability of soil texture characteristics could
be largely avoided. It also is a potentially useful method to make comparisons
between for example the grassland and the forest at our study site where
conductivity differences result in biased estimates of absolute soil moisture
between the two land covers. The ratio method is also affected by temporal

variability of soil water conductivity and temperature.

 

1000-l

Resistivity (Ohm-m)

100

I III]

 

 

 

 

Soil Water Content (cm3/cm3)

Figure 5-17. Laboratory measure resistivity and soil water contents for soil
samples collected from multiple locations at the study site. Images in this
dissertation are presented in color.

144

Based on the similar behavior of the resistivity-soil water content
relationship (Figure 5-17) for a number of soil samples from the study site, the

resistivity and water content change can be derived as;

"4%)
-9—1 e m (5-5)

2

where 61 and 62 are volumetric water contents at two different data collection
dates, p1, p2 calculated resistivities from differential inversions for the respective
dates, and m is the average slope of the log-log relationship between

temperature corrected resistivity and water content for soil samples from the site.

1.7

 

y = 1.05x- 0.06
R’ = 0.89

 

 

 

Observed moisture ratio

 

 

 

 

 

0.3 1 1 .7
Calculated moisture ratio

Figure 5-18. Soil moisture ratios observed with point gauges and soil moisture
ratios calculated with ERI near the location where point gauges are located.

145

The calculated moisture changes (61/ 92) were compared with soil
moisture changes observed in automated soil moisture loggers installed in the
forest. The comparisons yield a reasonable correlation indicating the potential
viability of the approach for understanding and quantifying soil moisture change

with ERI data (Figure 5-18).

Conclusions

Many practical challenges still remain in the use of ERI for environmental
monitoring and hydrologic characterizations. Natural environmental variability
related to temperature and pore-ﬂuid conductivity is a major concern for obtaining
accurate temporal soil moisture estimates from ERl. As highlighted in this study,
such variability is to be anticipated in any natural setting. If the effects of these
variables are not well constrained, signiﬁcant uncertainty may exist in the
estimated hydrologic quantities. Applying petrophysical relationships to obtain
soil moisture infomation from resistivity data is another potential source of
uncertainty. The best approach to minimize the uncertainties related to these is
to develop site speciﬁc relationships with laboratory experiments, regardless the
various practical limitations associated with the existing laboratory methods.
Alternatively, when only information on subsurface moisture change is required,
the potential of the resistivity ratio approach discussed can be explored. Soil
moisture computed with both types of ERI datasets (Wenner and dipole-dipole)

compared relatively well with the point soil moisture measurements made with

146

other probes at the site. However, data errors were more signiﬁcant with the
dipole-dipole measurements. As shown in this research, resistivity estimated soil
moisture often tend to under predict or closely relate to the lowest measured
(with other instruments) moisture in a given soil volume. While ERI is inherently
an estimate of a larger volume than point measurements, the general tendency

to be biased towards the lowest estimates need to be further explored.

The simple analysis of uncertainty resulting from difference inversion of
ERI data suggests that the choice of base dataset inﬂuences the resistivity data
interpretations and parameter derivations. All difference inversions with the
selected base datasets resulted in different outcomes, with largest differences in

areas with signiﬁcant heterogeneity such as in the forest portion of the study site.

147

References

al Hagrey, SA. 2007. Geophysical imaging of root-zone, trunk, and moisture
heterogeneity. Journal of Experimental Botany 58:839-854.

Amidu, SA, and J.A. Dunbar. 2007. Geoelectric studies of seasonal wetting and
drying of a texas vertisol. Vadose Zone Journal 62511-523.

Archie, GE. 1942. The electrical resistivity log as an aid in determining some
reservoir characteristics. Transactions of the American Institute of Mining
and Metallurgical Engineers 146254-61.

Binley, A., P. Winship, et al. 2002. Seasonal variation of moisture content in
unsaturated sandstone inferred from borehole radar and resistivity
proﬁles. Journal of Hydrology 267:160-172.

Dahlin, T., and M.H. Loke. 1998. Resolution of 2d wenner resistivity imaging as
assessed by numerical modelling. Journal of Applied Geophysics 38:237-
249.

Dahlin, T., and B. Zhou. 2004. A numerical comparison of 2d resistivity imaging
with 10 electrode arrays. Geophysical Prospecting 52:379-398.

Daily, W., and E. Owen. 1991. Cross-borehole resistivity tomography.
Geophysics 56: 1 228-1 235.

Daily, W., A. Ramirez, et al. 2005. Electrical resistance tomography-theory and
practice. p. 525. In D.K. Butler (ed.) Investigations in geophysics. Society
of Exploration Geophysics.

Daily, W., A. Ramirez, et al. 1992. Electrical resistivity tomography of vadose
water movement. Water Resources Research 28:1429-1442.

148

Day-Lewis, F.D., K. Singha, et al. 2005. Applying petrophysical models to radar
travel time and electrical resistivity tomograms: Resolution-dependent
limitations. Journal of Geophysical Research-Solid Earth 110.

French, H., and A. Binley. 2004. Snowmelt inﬁltration: Monitoring temporal and
spatial variability using time-lapse electrical resistivity. Journal of
Hydrology 297:1 74-1 86.

Hayley, K., L.R. Bentley, et al. 2007. Low temperature dependence of electrical
resistivity: Implications for near surface geophysical monitoring.
Geophysical Research Letters 34118402.

Hyndman, D.W., M.J. Dybas, et al. 2000. Hydraulic characterization and design
of a full-scale biocurtain. Ground Water 38:462-474.

Hyndman, D.W., and SM. Gorelick. 1996. Estimating lithologic and transport
properties in three dimensions using seismic and tracer data: The
kesterson aquifer. Water Resources Research 32:2659-2670.

Israil, M., M. aI-hadithi, et al. 2006. Groundwater-recharge estimation using a
surface electrical resistivity method in the himalayan foothill region, india.
Hydrogeology Journal 14:44-50.

Kemna, A., J. Vanderborght, et al. 2002. Imaging and characterisation of
subsurface solute transport using electrical resistivity tomography (ert) and
equivalent transport models. Journal of Hydrology 267:125-146.

Knight, R.J., and AL. Endres. 2005. An introduction to rock physics principles for
near-surface geophysics. Investigations in geophysics. Society of
Exploration Geophysics.

LaBrecque, D.J., and X. Yang. 2001. Difference inversion of ERT data: A fast
inversion method for 3-d in situ monitoring. Journal of Environmental &
Engineering Geophysics 6283-89.

Loke, M.H. 2000. Electrical imaging surveys for environmental and engineering
studies: A practical guide to 2-d and 3d surveys.

149

Looms, M.C., A. Binley, et al. 2008. Identifying unsaturated hydraulic parameters
using an integrated data fusion approach on cross-borehole geophysical
data. Vadose Zone Journal 7:238-248.

Nosetto, M.D., E.G. Jobbagy, et al. 2007. The effects of tree establishment on
water and salt dynamics in naturally salt-affected grasslands. Oecologia
1 52:695-705.

Pellerin, L. 2002. Applications of electrical and electromagnetic methods for
environmental and geotechnical investigations. Surveys in Geophysics
23:101-132.

Rein, A., R. Hoffmann, et al. 2004. Inﬂuence of natural time-dependent variations
of electrical conductivity on dc resistivity measurements. Journal of
Hydrology 285:215-232.

Reynolds, J.M. 1997. An introduction to applied and environmental geophysics.
John Wiley & Sons Ltd.

Slater, L., A.M. Binley, et al. 2000. Cross-hole electrical imaging of a controlled
saline tracer injection. Journal of Applied Geophysics 44:85-102.

Topp, G.C., J.L. Davis, et al. 1980. Electromagnetic determination of soil-water
content - measurements in coaxial transmission-lines. Water Resources
Research 16:574-582.

Waxman, M.H., and L.J.M. Smits. 1968. Electrical conductivities in oil-bearing
shaly sands. Society of Petroleum Engineers Journal 8:107-122.

150

 

 

 

 

 

  

      
      

lllllllljlllljjlllllllllllll

G
1 956 9443

VERS ARIES