-31i’;?‘.lt(.

$539;
.13‘

a”?

 

:10" -

.13 Mm”

' Kev-.04-

1"." 1-43:-

v {may

on! 143:?!

.- m
. 4n.

VO"

£0.15

'p-v J .'.‘
r4 ...,A
.

'vs

$543!“)-

.\\1

. .J

r- . l"

' ~=v~*.‘.‘va‘¢:‘,\ T.
{H '

 

THESIS
5; 01M“? 50

This is to certify that the
thesis entitled

ESTIMATING SEASONAL AGRICULTURAL IRRIGATION
WATER USE IN MICHIGAN: FIELD-LEVEL EVALUATION OF
THE MICHIGAN WATER USE REPORTER

presented by
COLIN R. NUGENT

has been accepted towards fulfillment
of the requirements for the

Master of Arts

 

 

      

.an r Pro ssor’s Signature

\

 

Date

MSU is an Affirmative Action/Equal Opportunity Institution

 

LIBRARY
Michigan State
University

 

 

 

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DateDuepes-pJS

ESTIMATING SEASONAL AGRICULTURAL IRRIGATION WATER USE IN
MICHIGAN: FIELD-LEVEL EVALUATION OF THE MICHIGAN WATER USE
REPORTER
By

Colin R. Nugent

A THESIS

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

MASTER OF ARTS
Department of Geography

2004

ABSTRACT
ESTIMATING SEASONAL AGRICULTURAL IRRIGATION WATER USE IN
MICHIGAN: FIELD-LEVEL EVALUATION OF THE MICHIGAN WATER USE
REPORTER
By

Colin R. Nugent

The Michigan Water Use Reporter (MWUR) model is a simulation designed to
estimate water use from irrigated agriculture in the state of Michigan. The model was
developed by Moen (1999) but had never been evaluated against actual grower reported
irrigation amounts. The evaluation of this model took place with data from the 2002 and
2003 growing season. Twenty-one fields across central and southern lower Michigan
were used as study sites. Volumetric soil moisture and seasonal irrigation water depths
were recorded from each site and used to test the simulation. Validation of the simulatiOn
was conducted in two stages. First, seasonal irrigation water volumes were compared,
using descriptive statistics, to simulated season irrigation output. Second, simulated
volumetric soil moisture were validated using field measurements from a capacitance
probe. A sensitivity analysis of managerial and crop physiological parameters was
conducted after validation. Depth per irrigation, irrigation trigger level, planting date, and
root growth rate were analyzed. The simulation tended to overestimate both seasonal
irrigation water depth and volumetric soil moisture across all crops. The sensitivity
analysis found depth per irrigation and trigger level were by far the most sensitive
parameters. These tests indicate the model, while demonstrating sound hydrology, does
not properly characterize the methods grower use to decide when to irrigate. Better

parameterization of these methods will result in a more robust simulation.

Acknowledgements

I would like to give my thanks and gratitude to my advisor, Dr. Jeff Andresen, for
his support and guidance throughout this project. Without him, none of this work would
be completed. Next, I would like to thank my committee members, Dr. Bill Northcott, Dr.
Bob von Bernuth, and Dr. Julie Winkler for their questions and suggestions in the design
and execution of this study.

To Dr. Nothcott and Dr. von Bemuth, I wish to express a special thanks for
helping me grow as a person, student, researcher, and engineer in the last eight years and
Michigan State. I have the utmost respect for you as teachers and engineers. It has been a
real pleasure to have the opportunity to learn fi‘om you and work with you during my
time at this university.

I would like to also thank my family for their love and support while I have been
a student here at MSU. It has been wonderfiII to have a family so close, even with a large
distance between us.

Finally I would like to thank Tracy Aichele, Costanza Zavalloni, and the
Geography graduate students for giving wonderfiil suggestions and support during the
learning process as a graduate student. I have thoroughly enjoyed working with you over

the years.

iii

Table of Contents

 

 

 

 

 

 

 

 

List of Figures - - _ v
List of Tables vii
Statement of Problem 1
Literature Review 3
Introduction ..................................................................................................................... 3
Next Generation Radar ................................................................................................... 7
Irrigation Scheduling and Crop Modeling ...................................................................... 9
Other calculations in MWUR ....................................................................................... 12
Use of GIS in Crop Modeling ....................................................................................... 14
Soil Moisture Monitoring and Calibration .................................................................... 18
Regional Modeling and Consideration of Scale ........................................................... 22
Conclusion and Summary ............................................................................................. 28
Methods 30
Overview ....................................................................................................................... 30
Measurement of Soil Moisture ...................................................................................... 32
Model Validation ................................................................. .. ........................... . ............ 40
Model Sensitivity Analysis ........................................................................................... 43
Results 47
Model Validation .......................................................................................................... 47
MWUR Model Sensitivity Analysis ............................................................................. 61
MWUR Performance with Changes to Managerial Variables ...................................... 76
Conclusions 84
Appendix---- 86
References 89

 

iv

List of Figures
Figure 1: A visualization of the layering taking place within the MWUR simulation
(courtesy of Tracy Aichele, 2002). ......................................................................... 15

Figure 2: The NEXRAD 4 km grid network across lower Michigan with point locations
of study sites. ......................................................................................................... 24

Figure 3: The NEXRAD 4 km grid network at the county size, with field sizes delineated.
.............................................................................................................................. 24

Figure 4: Four NEXRAD 4 km grid cells and three study fields (in black). .................... 25

Figure 5: Visualization of model layer depths (left) and capacitance probe measurement
layer depths (right) with tube placement in soil profile. .......................................... 42

Figure 6: Simulated vs. Reported Seasonal Irrigation Depth (mm) for all reporting fields
in 2002 and 2003. Simulated results calculated using default model settings. ......... 51

Figure 7a: Simulated volumetric soil moisture, as a percent of observed soil moisture by
depth for com in Mecosta County, 2003. Simulated values calculated using default
model settings. ....................................................................................................... 55

Figure 7b: Simulated volumetric soil moisture, as a percent of observed soil moisture by
depth for corn in Montcalm County, 2002. Simulated values calculated using default
model settings. ....................................................................................................... 56

Figure 7c: Simulated volumetric soil moisture, as a percent of observed soil moisture by
depth for corn in St. Joseph County, 2003. Simulated values calculated using default
model settings. ....................................................................................................... 57

Figure 7d: Simulated volumetric soil moisture, as a percent of observed soil moisture by
depth for corn in Saginaw County, 2003. Simulated values calculated using default
model settings. ....................................................................................................... 58

Figure 8a: Simulated and reported seasonal irrigation depth accumulations (mm), by day
of year. A carrot field in Mecosta County, 2002. .................................................... 62

Figure 8b: Simulated and reported seasonal irrigation depth accumulations (mm), by day
of year. A soybean field in St. Joseph County, 2002 ............................................... 63

Figure 9: Difference between simulated and reported seasonal irrigation depth vs. total
seasonal drainage and change in soil moisture across all study sites, 2003. ............ 65

Figure 10a: Seasonal trend of simulated and observed volumetric soil moisture (cm3/cm3)
for the 0-30 cm soil profile layer in corn, Mecosta Co. 2003. Observed values are
average of three samples per field, with maximum and minimum values reported

with upper and lower error bars, respectively. Simulated values calculated using
default model settings. ........................................................................................... 67

Figure 10b: Seasonal trend of simulated and observed volumetric soil moisture (cm3/cm3)
for the 30-60 cm soil profile layer in corn, Mecosta Co. 2003. Observed values are
average of three samples per field, with maximum and minimum values reported
with upper and lower error bars, respectively. Simulated values calculated using
default model settings. ........................................................................................... 68

Figure 10c: Seasonal trend of simulated and observed volumetric soil moisture (cm3/cm3)
for the 60-90 cm soil profile layer in corn, Mecosta Co. 2003. Observed values are
average of three samples per field, with maximum and minimum values reported
with upper and lower error bars, respectively. Simulated values calculated using
default model settings. ........................................................................................... 69

Figure 11: Simulated vs. reported seasonal irrigation depth (m) for all study fields in

2002 and 2003. Simulated values calculated using altered model settings (to improve
mean differences of seasonal irrigation depth). ...................................................... 80

vi

List of Tables

Table 1: Volumetric soil moisture calculated with regression equation developed from
field soil cores, Saginaw Co. 2003. ........................................................................ 36

Table 2: Volumetric soil moisture calculated with regression equation developed from
field soil cores, Montcalm Co. 2003 ....................................................................... 36

Table 3: Factory calibration curve values by code letter. Constants are pre-loaded into the
probe datalogger and automatically calculate volumetric soil moisture from raw
probe readings. (ISM, 1999) .................................................................................. 38

Table 4: Mean and mean absolute difference between factory calibrated capacitance
probe and volumetric soil core measurements of volumetric soil moisture (cm3/cm3)

by depth for all fields, 2003. .................................................................................. 38

Table 5: Mean and mean absolute difference between simulated and reported seasonal
irrigation depth (m) and simulated seasonal irrigation depth as a percent of
reported irrigation, calculated by crop for seasons 2002, 2003, and the two years
combined. Simulated depths calculated using default model settings ...................... 48

Table 6: Simulated and reported seasonal irrigation depth standard deviation for seasons
2002 and 2003 combined. ...................................................................................... 48

Table 7: Simulated, reported, differences, and absolute differences of seasonal irrigation
depth for individual study sites, 2002 and 2003 ...................................................... 50

Table 8: Mean and mean absolute differences between simulated and observed soil
moisture (cm3/cm3), calculated by crop and soil profile depth for years 2002 and
2003 combined. Simulated soil moisture calculated using default model settings.
Significance tested at or = 0.01 level ....................................................................... 53

Table 9: Mean and mean absolute difference between simulated and reported irrigation
depths (mm) and simulated seasonal irrigation depth as a percent of reported
irrigation, calculated by grower for years 2002, 2003, and the two years combined.
Simulated depths calculated using default model settings. ...................................... 60

Table 10: Standard deviations of simulated and reported seasonal irrigation depth, by
grower for seasons 2002 and 2003. Simulated irrigation depths calculated using
default model settings. ........................................................................................... 60

Table 11: Mean and mean absolute differences between simulated seasonal irrigation
depths of altered model settings and default model setting (50% of plant available
soil moisture), by crop for years 2002 and 2003 combined. Changes in model
settings made to the irrigation trigger level, which is based upon the percent of plant
available soil moisture. Default setting is 50% of plant available soil moisture. ..... 71

vii

Table 12: Mean and mean absolute differences between simulated seasonal irrigation
depths of altered model settings and default model setting (25mm per irrigation
event), by crop for years 2002 and 2003 combined. Changes in model settings made
to the irrigation depth per event default of 25 mm per irrigation event. .................. 73

Table 13: Mean and mean absolute differences between simulated seasonal irrigation
depths of altered model settings and default model setting (varying by crop), by crop
for years 2002 and 2003 combined. Changes in model settings made to the root
development rate, which based upon either growing degree units or calendar days
after planting, depending on crop type. .................................................................. 74

Table 14: Mean and mean absolute differences between simulated seasonal irrigation
depths of altered model settings and default model setting (planting date on day of
year 135), by crop for years 2002 and 2003 combined. Changes in model settings
made to the planting date default ofDOY 135. ...................................................... 75

Table 15: Mean and mean absolute difference between simulated and reported seasonal
irrigation depth (m), calculated by crop for years 2002, 2003, and the two years
combined. Simulated depths calculated using altered model settings (to improve
mean differences of seasonal irrigation depth). ...................................................... 78

Table 16: Mean and mean absolute difference between simulated and reported seasonal
irrigation depth (m), calculated by grower for years 2002, 2003, and the two years
combined. Simulated depths calculated using altered model settings (to improve
mean differences of seasonal irrigation depth). ...................................................... 78

Table 17: Mean square error for default and altered simulated seasonal irrigation depth
for corn, potato, soybean, and all crops for 2002 and 2003. .................................... 81

Table 18: Mean and mean absolute differences between default and altered simulated
volumetric soil moisture (cm3/cm3), by profile depth and crop for the years 2002 and
2003 combined. ..................................................................................................... 81

Table 19: Mean and mean absolute differences between simulated and observed
volumetric soil moisture (cm3/cm3), by soil profile depth and crop for 2002 and 2003
combined. Simulated values calculated using altered model settings (to improve
mean differences of seasonal irrigation depth). ...................................................... 83

viii

Statement of Problem

Water use in the state of Michigan has not been of great concern until recently.
The visibility of the Great Lakes and abundant groundwater supply has, in the past, given
citizens of the state a sense of security regarding the availability of fresh water. In recent
years, though, pressures for water use have been increasing fiom industry, residents and
agriculture. Concerns over the export of Michigan water and legal disputes between
irrigators and residents have put the issue of water availability in the limelight and the
state legislature is ready to act on regulating water use within the state.

Agriculture has historically benefited from legislation that exempts the industry
from many state regulations. Also, much of the agricultural water use in the state goes
toward irrigation of crops during the growing season. Moen (1999) developed the
Michigan Water Use Reporter (MWUR) model that estimates the amount of water
applied for irrigation, at a 4 km resolution but reported at a county level, across the state
on an annual basis. While the MWUR model has been written, it has not yet been
validated against field reported or measured results.

This study proposes to address the outputs of the model by statistically comparing
the modeled total seasonal water volume and seasonal soil volumetric water content to
similar data collected at cooperator sites across the state. If a positive correlation is found,
these comparisons will allow legislators to put their trust in the scientifically validated
model output when making policy decisions. This model will eventually allow both the
state government to monitor water use in a non-intrusive fashion and the growers to

justify their reasonable water use for irrigation under current practices.

Objectives

The MWUR model, while potentially very powerfirl, has yet to have its output
compared with data collected from actual study sites across the state. The model output
includes both the total volume of water applied and volumetric soil moisture per day for a
growing season at a 4 km resolution. The objectives for this study are therefore as
follows:
1) Collect volumetric soil moisture and amount of water applied in irrigation for study
sites across the state.
2) Run the model for the Study Site locations.
3) Compare the model output for volumetric soil moisture trends and total seasonal water
use.
4) If the model output significantly deviates from any of the data collected from the study
Sites, perform a sensitivity analysis to explain why these deviations occurred and suggest

possible ways they may be resolved.

Background and Literature Review

Introduction
Irrigation of agricultural crops is a vital component of food production worldwide,

allowing growers to reduce production risk associated with crop water shortages and to
improve commodity quality. Land under irrigation comprised approximately 22 Mha in
the United States in 1997 and Michigan accounted for approximately 0.15 Mha of the
total (NASS 1999). Overall, the seasonal average irrigation depth per season in the US.
has decreased fiom 650 mm in the 1970's to 500 mm by 1996, indicating improved water
efficiency. At the same time, the use of center pivot irrigation systems has increased to
over 30 percent of all irrigation systems (Howell 2001).

In recent years, water use has moved to the forefront of public awareness in the
state of Michigan. Highly publicized lawsuits involving groundwater rights have been
brought against water bottlers, mining companies, and agricultural irrigators. With all of
this publicity, state lawmakers have begun to take notice and are looking closely at water
use issues. They have passed two pieces of legislation regulating large consumptive users
of Great Lakes Basin water. PA 148 is a reporting and regulation bill and PA 177 is a
conflict resolution bill.

The Council of Great Lakes Governors adopted the Great Lakes Charter in
February 1985 in an agreement to outline ways to protect the water of the Great Lakes
from environmental degradation and exportation of water outside the basin. The Great
Lakes Basin is defined to include all bodies of water, rivers, stream, connecting channels,

and groundwater within the watershed of the Great Lakes and St. Lawrence River.

Michigan is the only Charter member state of the nine US. states and two Canadian
provinces that lies completely within the Great Lakes Basin.

The original Charter calls for, among many things, a common database of
information on water withdrawals within the Basin. This database includes data such as
the volume and uses of the withdrawn water. Another agreement was signed in August
2001, known as Annex 2001, to reaffirm many of the ideals of the original charter. Six
directives were added to the Charter at this time. One directive calls for the establishment
of a new decision-making standard for the approval process of new user withdrawals,
such as large scale irrigation pumping wells. The second directive calls for the
development of a decision support system to ensure the best data are available regarding
the state and uses of the Basin water (COGLG 1985; COGLG 2001)

The Michigan Water Use Reporter (MWUR) model estimates the seasonal
volume of irrigation water demand in the state of Michigan. The MWUR simulation
upholds many of the ideals of the Charter and the directives in Annex 2001. It is designed
to provide regulators with detailed information regarding water use for irrigated
agriculture in Michigan and to provide quality data for the common database of
lmowledge called for in the Charter. It can provide science-based information for
decision supports systems, as well (COGLG 2001).

While the Charter is a semi-binding agreement between the Great Lakes states
and provinces, it is not a legally binding document because the federal governments of
the United States or Canada have not ratified it. Therefore, the Michigan state legislature

recently passed two documents, signed by Gov. Jennifer Granholrn in November 2003,

relating to water withdrawal reporting and regulation, and conflict resolution of water use
disputes.

The reporting and regulation bill passed is PA 148, from the original Senate Bill
289. This act calls for the reporting of well data for large scale agricultural users, among
others. Agricultural users must comply with reporting regulations if their well pumping
capacity exceeds 100,000 gallons of water per day for a 30-day period. If the grower falls
into this category, they must report the source, volume, and use of the water withdrawn.
Part (2) of Section 32708 states: “The Department (of Environmental Quality) and the
Department of Agriculture, in consultation with Michigan State University, shall validate
and use a formula or model to estimate the consumptive use of withdrawals made for
agricultural purposes consistent with the objectives of Section 32707.” The MWUR
model from Michigan State University is the system referred to in the bill. This
Simulation estimates the volume of water consumed by irrigated agriculture in Michigan,
reporting at the county level.

Let us first consider the definition of beneficial use of water. Burt et al. (1997),
writing for the American Society of Civil Engineers (ASCE), defines beneficial water use
in agriculture as water which “supports the production of crops: food, fiber, oil,
landscape, turf, omamentals, or forage.” This includes water use for crop
evapotranspiration (ET), maintaining or improving soil quality (removal of salts), climate
control (frost protection), and plant emergence among others. He does qualify that the top
priority of water is human consumption but that agriculture is still considered a beneficial

use of water to society.

Another important definition to this argument is that of consumptive use of water.
Agricultural water use for irrigation fits the definition of consumptive use. Burt et al.
(1997) define consumptive use as “irrigation water that ends up in the atmosphere
(evaporation or ET) or in the harvested plant tissues (either as molecular water, notably in
watermelon or tomatoes, or in organic compounds and is considered irrecoverable, that
is, it is consumed”. It is assumed by the Michigan Department of Environmental Quality
(MDEQ) that if the water reaches the atmosphere through evapotranspiration, 90 percent
of the original amount will be transported out of the basin, in this case the Great Lakes
Basin, and is lost fi'om the hydrologic system of the Great Lakes (R. van Til, MDEQ,
personal communication).

The agriculture industry in Michigan has relied on farmer surveys in the past to
estimate the amount of irrigation water used by the industry. Michigan Agricultural
Statistics Service (MASS) and the National Agricultural Statistics Service (NASS)
conduct these surveys and publish results once every four years. With the advent of new
weather and climate monitoring technologies, a computer model was developed to
provide government agencies a way of estimating irrigation water use in a less intrusive
way (Moen 1999). While these estimates are not official, they do give these agencies
some idea of the annual water use by irrigators.

The MWUR water use simulation employs the well-tested method of a soil water
balance (Ritchie 1985; Knox et al. 1996; Ejieji and Gowing 2000; George et al. 2000).
While other simulations have estimated water use across regional areas (Knox et al.
1996), none have been developed or tested for an entire state in the Great Lakes region of

the United States. Estimates from initial simulation runs were found to be in agreement

with state-level government estimates, although the Michigan Water Use Reporting
(MWUR) model has yet to be validated with in-field data (Moen 1999). The model
cannot be considered reliable until this validation takes place.

The model uses commonly measured weather data, more specifically temperature,
solar radiation, and precipitation, in conjunction with available soil texture maps to
estimate crop water demands through the growing season across a large area of thousands
of square kilometers. The technological breakthrough making this model possible is the
implementation and improvement of National Weather Service WSR—88 radar rainfall
estimates, also known as the Next Generation Radar (NEXRAD). This product provides
hourly growing season precipitation estimates, a key input variable, at a 4 km resolution
across a continuous grid for the entire state and region.

Next Generation Radar

The National Weather Service first released its WSR-88 (NEXRAD) product in
1988. The radar beam scans the atmosphere at a small angle above horizontal and
measures the reflectivity of the return beam, Z (mm6/m3). Algorithms then relate the
reflectivity to a rainfall amount, R (m) (Fulton et a1. 1998). This relationship is
dependent upon the intensity of precipitation, size of the hydrometeors, and form of
precipitation (rain, sleet, snow, etc.). Biases do occur when precipitation intensities
increase, when the size of the hydrometeors increase, or when the precipitation is
partially or totally frozen. Atmospheric particulates also can have adverse affects on these
estimations (Krajewski and Smith 2002).

Operational post-processing is utilized by the National Weather Service to aide in

the reduction of errors of the precipitation estimates. Krajewski and Smith (2002) found a

reduction in errors when the original estimates were corrected using rain gauge network
data. The output of the second stage of processing is referred to as Stage II NEXRAD
data. Unfortunately, there are errors associated with rain gauge networks as well, and care
must be taken when using these products. Both Steiner et al. (1999) and Krajewski and
(2002) reported poor data quality in historical rain gauge data. When high quality data
were used, Steiner et al. (1999) were able to Obtain a root mean square error for the radar
rainfall estimation of about 10 percent.

The final post-processing stage involves overlaying Stage H estimations fi'om
nearby radar stations over one another. The possibility for error increases the firrther the
particle is from the radar station (Fulton et al. 1998). Also, errors occur as the angle from
horizontal increases (Borga 2002; Sharif et a1. 2002). The mosaicing process helps to
reduce both of these sources of error (Borga, 2002). The Stage 111 data product used in
the MWUR model comes {Tom the NWS and the Michigan Climatological Resources
Program (MCRP). The NEXRAD Stage III product is only available during the warm
season, as it has yet to correctly estimate frozen precipitation events. This limitation does
not necessarily affect the MWUR model because it simulates crop water use during the
growing season. The MCRP found the frequency in which the NEXRAD correctly
sensed precipitation was on the order of 95.6 percent when using the Michigan
Automated Weather Network (MAWN) as a baseline (Andresen and Aichele 2003).
Mean differences between radar estimated and ground measured (MAWN) hourly
precipitation was 0.01 mm, with a mean absolute difference of 0.11 mm. When these

statistics are calculated using National Oceanographic and Atmospheric Administration

(NOAA) first order weather stations, the mean difference in precipitation was -0.1 mm
and had a mean absolute difference of 0.61 mm.

While there are errors related with this new rainfall data product, there are also
great advantages. Chief among them is a Spatially continuous dataset with a resolution of
4 km (Figure 2). This allows for larger scale study areas where rain gauges may be
limited. Koren et al. (1999) found these data to be useful for lumped hydrological
modeling, the soil-water balance being one method tested, for the Arkansas-Red River
basin. A finer resolution was preferred because of grid overlaying, but evapotranspiration
was positively correlated with scale. Carpenter et al. (2001) also found the Stage HI
NEXRAD to be applicable and useful for hydrologic modeling of larger scale
catchments.

Irrigation Scheduling and Crop Modeling

The soil water balance model used in the MWUR scheme is a well tested method
that has become a standard in physical modeling of crop systems (Jensen et al. 1970;
Wright and Bergsrud 1991; Knox et al. 1996; Prajamwong et a1. 1997; Ejieji and Gowing
2000; Panigrahi and Panda 2003). A soil-water balance method is utilized to calculate
plant available soil water, based on the work of Joe Ritchie (Ritchie 1972; Richardson
and Ritchie 1973; Ritchie 1985; Ritchie 1998). A soil water balance sums water inputs
and outputs fiom the system for a specific area:

AS=Pe+IRR+GW—DP—ET—RO-ASS
where the change in soil water storage, AS, is the sum of the effective precipitation, Pe;

irrigation, IRR; ground water upward flux, GW; deep percolation, DP;

evapotranspiration, ET; surface runoff, R0; and change in surface storage, ASS
(Prajamwong et a1. 1997; Moen 1999).

The method requires commonly measured meteorological data on a daily basis
and knowledge of physical soil characteristics. Meteorological data, in this model
includes precipitation, maximum and minimum temperatures, and solar radiation.
Temperature and radiation are used directly to calculate crop potential
evapotranspiration, which will be discussed later. Soil water holding capacity, including
the drained upper and lower limits, is necessary for the calculation of the net plant
available water stored in the soil profile.

The profile is ofien broken into distinct horizontal layers and the soil water
balance is calculated for each. These layers may have different characteristics based upon
location and the depth of each may change with root development during the growing
season (Burt et a1. 1997; Prajamwong et al. 1997). Also, soil types may change through
the profile, with different properties such as plant available water capacities (Mahmood
and Hubbard 2003; Panigrahi and Panda 2003; Starks et al. 2003).

Within the soil profile, water may move in many directions. The greatest
depletion in the rooting zone occurs as a result of evapotranspiration (ET) (Jensen et a1.
1971). This is the combination of evaporation from the soil surface and transpiration of
water through plants. ET may be directly measured through the use of a lysirneter or
estimated indirectly fi'om the calculation of potential evapotranspiration (ETD) using one
of many commonly-used equations. Over the years, many different methods for
estimating ETp have been developed. There are a handfirl ofwell-tested methods, each

useful depending upon the regional climate and climate data available. In most cases, a

10

reference ET is calculated for a standard well-watered canopy, usually grass or alfalfa.
The estimated crop ETp is a product of the reference ET and an empirical crop coefficient
(Kc) (Jensen et al. 1990; Allen et al. 1998). The Kc value is based upon the specific crop
of interest and its growth stage. A drawback to the estimation of crop ETp is the
requirement of a relatively large amount of detailed meteorological and agronomic data.

When meteorological data are limited, a number of estimation approaches are
available. For monthly temperature based estimates of potential evapotranspiration (ETp),
the Thomthwaite method is reasonable. Thomthwaite (1948) developed a model based
upon mean monthly temperature, day length, days per month, and a heat index derived
from the sum of a 12-month index. This method has obvious shortcomings if a sub-
monthly time period is used. Short-term mean temperature does not relate well with
incoming radiation, and therefore leads to serious errors using this method (Rosenberg et
al. 1983)

The J ensen-Haise method uses mean daily air temperature and the daily solar
radiation equivalent of evaporated water to estimate ETp. Jensen and Haise (1963)
evaluated this method with lysimetric measurements in arid regions of the western United
States. They found a good correlation, but only under non-advective conditions
(Rosenberg et al. 1983).

Probably the most widely used method to estimate ETp are the Penman and
modified Penman-Monteith methods (Rosenberg et al. 1983; Allen et al. 1998). The
Penman method is a combination of an energy component, solar radiation, wind speed
and duration. No temperature component is used. The method was developed using a

linear regression of evaporation rate over vapor pressure deficit versus wind speed.

11

Evaporation was measured fi‘om a evaporation pan, surrounded by a grass canopy, at Fort
Collins, CO (Jensen et a1. 1990). Monteith (1981) later modified the Penmen equation to
accommodate aerodynamic and crop canopy resistance (Hatfield 1990).

For well-watered or humid conditions, the Priestly-Taylor method is a simplified
and very useful form of the Penman-Monteith method (Jensen et al. 1990). It takes the
form:

m, = aIs/(s + m * (R. + S)
where ETp is the potential evapotranspiration, s is the slope of the saturation vapor
pressure at the mean wet bulb temperature, 7 is the psychrometric constant, Rn is the flux
density of net radiation, and S is the soil heat flux. The or term is considered the ratio
ETp/ET,q and is an empirically derived constant. ETeel is the equilibrium
evapotranspiration. Stewart and Rouse (1977) found values of or varied Slightly around
the value 1.26 for temperature ranges of 15 to 30°C (Rosenberg et al. 1983). The equation
is used in the Ritchie water balance and the MWUR simulation because of its utility
under humid conditions, which best describes the growing season climate in Michigan.
Also, meteorological data are more readily available for this ETp calculation. Within
MWUR, ET is calculated from the ETp by multiplication of a crop-specific coefficient.
Other calculations in M WUR

Calculation of ET is the first major step of the soil water balance model within
MWUR, followed by root grth and development. New root growth is a function of
daily solar radiation and existing leaf area or days after planting, depending upon crop.

Vertical root grth distribution is later used to calculate potential plant water uptake for

12

each soil layer. Water is extracted from any layer with a root length distribution value of
0.05 or greater (Moen, 1999).

Next, volumetric soil water values are calculated. Ponding values are determined
based upon hourly NEXRAD rainfall rates, irrigation events, and soil hydraulic
conductivity. When rainfall intensity and volume are greater than soil infiltration rate,
ponding occurs. This routine calculates the daily infiltration, runoff, and changes to depth
of ponded water. Downward water movement through each layer is calculated when
water is in excess of the soil drained upper limit. The potential drainage calculation
relates the layer’s drained upper limit (DUL), saturation level (SAT), hydraulic
conductivity (KS), and infiltration from the layer above. Finally, soil evaporation is
calculated. This routine calculates upward movement of water by capillary action as well
as evaporation. Evaporation is a fimction of leaf area index and potential
evapotranspiration (Ritchie 1972; Moen 1999).

The determination of irrigation events is then based upon the ratio of extractable
soil water (EWS) to the potential extractable soil water (PEWS) in the rooting zone of the
soil profile. The extractable soil water and potential soil water equations take the form:

ESW = 2(SWi — LLi)*DI for i = 1...n
PESW = 2(DULi — LLi)*DI for i = 1...n

D1 = ESW/PESW
Where

SWi = Current volumetric soil water content in layer i
LLi = Lower limit of volumetric soil water in layer i

DULi = Drained upper limit of volumetric soil water in layer i

D1 = Drought index

13

n = Number of layers
The default drought index, which effectively triggers an irrigation event, is set at a value
of less than or equal to 0.5, but can be altered by the user to reflect differing water
management strategies (Moen 1999).

Use of GIS in Crop Modeling

A geographic information system (GIS) is a tool to store, analyze, and display
Spatially referenced data. It allows the user to link information stored in a database to a
location in space and compute new data for that location (Maracchi et al. 2000) It also
has the capability to incorporate remotely sensed data for analysis and display purposes.
These data can include satellite land cover data, satellite-derived soil moisture estimates,
or radar precipitation estimates. The use of GIS in hydrologic modeling has increased in
recent years because of the spatial analysis capabilities of these programs (Engel et al.
1997; Knox et al. 1997; Sousa and Pereira 1999; Ogden et al. 2001; Heinemann et al.
2002; Ines et al. 2002; McKinney and Cai 2002; Martin de Santa Olalla et a1. 2003;
Rowshon et al. 2003; Rowshon et al. 2003). Figure 1 is a visualization of the input-
layering taking place within the MWUR model.

Ogden et al. (2001) describe a number of different GIS interfaces for hydrologic
watershed modeling. In his review, he notes the importance of temporal variability. Not
only is there a need for spatial consistency within inputs and outputs, but also temporal
consistency such as daily or annual averages of variables. He also states the need for
spatially continuous data, such as the NWS NEXRAD radar precipitation estimates. The
NEXRAD product is useful because the data are available in a raster grid network

covering the entire state and relatively short temporal resolution (hourly). As the quality

14

   
 

   
   
 

Landscape Features

Daily Weather Data
(Watersheds and Soils)

(Solar Radiation,
Maximum and

 
   

Crop Distribution

Hourly
Rainfall Data

hr

     
   

Irrigation Water Use
in Michigan

Figure 1: A visualization of the layering taking place within the MWUR simulation (courtesy of
Tracy Aichele, 2002).

and availability of products such as NEXRAD precipitation estimates increase, use of
hydrologic models will continue to increase within the GIS platform.

A GIS is also very capable of analyzing the water needs for agriculture. Knox et
al. (1997) wrote a computer model that estimates irrigation water requirements for
potatoes in England and Wales, which is very similar to the MWUR program. They were
able to create maps of water use at a 5 km resolution for analysis at county and river
basin scales. All necessary datasets were available in digital form fiom government
agencies. These included regional soils maps, weather data, and land use maps. This
program provided water use maps for catchment managers and planners, as well as
providing politicians with a tool to more effectively litigate their water resource. Inputs
and calculation methods were very similar to those used by Moen (1999), except for the
precipitation data. Knox et al. used data fiom 11 automated weather stations instead of
remotely sensed precipitation and solar radiation data. To avoid modeling many soil
textures, they used three representative soils, high, medium, and low available water
capacities, for the entire region instead of the numerous soils used in MWUR. Total water
demand for maincrop potatoes in 1990 was determined and compared with the
governmental survey estimation at a county level. The simulated water depth applied was
greater than reported governmental values by about 16 percent.

Engel et al. (1997) developed a program, AEGIS/WIN, that linked the Decision
Support System for Agrotechno logy Transfer (DSSAT) with a GIS interface. This
allowed for the creation of thematic maps of field management practices, such as final
yields, irrigation requirements, and nitrogen leaching for an entire farmsted, which can be

used by growers as part of a precision agriculture practice or by planners as part of a

16

regional resource management program. Heinemann et al. (2002) had a newer version of
this program, assumed it was calibrated for their region, and used it to estimate the spatial
water requirements for counties in the Brazilian state of Parana. Thematic maps were
developed for annual irrigation withdrawal and runoff. The authors were very satisfied
with the results when such limited input data were available.

Other investigators have developed irrigation monitoring systems with GIS
platforms. Martin de Santa Olalla et al. (2003) and Rowshon et al. (2003a) both
developed reliable programs for system managers to better monitor and distribute
irrigation water. Rowshon et al. monitored furrow irrigation in Malaysia, outputting
weekly water needs for rice production in maps, graphs, and tables. While their irrigation
scheduler underestimated water needs, they were satisfied with the weekly monitoring
and map development for water use. de Santa Olalla et al. used LANDSAT satellite
imagery to delineate irrigation areas and crop types in southeastern Spain. The GIS was
able to link these locations, crop types, and field areas with government estimations of
water use for individual crops to monitor groundwater withdrawal for irrigation in the
aquifer. The system passed beta testing after successfiilly outputting reasonable water
volume estimates compared to an exhaustive field study and is now fully operational for
estimating irrigation withdrawals in the Mancha Oriental aquifer of southeastern Spain.
He (1999) performed an analysis of irrigation water needs for the Saginaw Bay basin in
lower Michigan. He used a GIS in conjunction with four crop growth models to overlay
soil series maps, multi-station climate data, and multi-season management strategies
(planting date, harvest date, etc.) to calculate the average irrigation demand for the Cass

River watershed.

17

Sousa and Pereira (1999) validated a regional irrigation water requirement model
for maincrop potatoes in northeast Portugal by first validating the simulation under local
conditions, running a 19 year time series of historical weather data, and finally creating
spatial water requirement maps using kriging techniques within a GIS interface. They
chose the geostatistical method of kriging to overcome spatial heterogeneity problems.
Soil moisture was monitored in 2 ha plots using a neutron probe, gravirnetric samples,
and tensiometers. Monitoring of soil moisture was utilized in the validation of their
model to local conditions. The subsequent 19-year irrigation water estimation, for 106
locations, resulted in a mean depth of 290 mm of water per year. Surface maps of
irrigation requirements for the entire region were later created from these 106 locations
using kriging techniques.

Soil Moisture Monitoring and Calibration

Some of the above irrigation scheduling and monitoring programs were validated
with measured volumes of water flowing through a monitored system (Rowshon et al.
2003) or from governmental survey estimates (Fanning et a1. 2001). These methods are
not common, though. The majority of irrigation models based upon the soil water balance
were validated through a combination of soil moisture probes and gravirnetric field
samples.

One of the many methods for measuring soil moisture in the field is the time
domain reflectometry (TDR) method. This method is based upon the relationship
between the soil water capacitance, dependent upon water content, and the time shift of a
1 MHz signal sent by two metallic probes in the soil (Lane and Mackenzie 2001). The

technology is usefiil because it provides a continuous output signal and can be automated

18

with a datalogger. The academic community has until recently, mainly utilized it. Once
calibrated, it can be a reliable method for continuous, in situ measurement of soil
moisture (Jackson 2003). Many times this technique is used as a comparison for the
validation of models (Starks et a1. 2003), remote sensing of soil water content (Wilson et
al. 2003), or comparison for other field-based measurement technologies (Tomer and
Anderson 1995; Lane and MacKenzie 2001; Wilson et al. 2003). Chief among the
limitations of this method is the high costs for installation to achieve the desired spatial
coverage. Also, TDR probes are not portable and each must be buried to the desired
depth. Finally, the sphere of influence that the TDR measures is based upon the length of
metallic probe, which can vary between units (Starks et aL 2003).

The neutron scattering probe was used to validate models of Knox et al. (1996),
Sousa and Pereira (1999), George et a1. (2000), and Panigrahi et al. (2003). Each set up a
sampling scheme for their respective studies and sampled soil moisture on a routine basis.
The neutron probe requires aluminum access tubes installed within the field. Readings
are taken at a soil depth as to give an average of soil moisture for each soil layer, usually
about 15-cm in depth. Measurement frequency ranged from twice per day (George et al.
2000), to daily (Panigrahi and Panda 2003), and finally weekly (Sousa and Pereira 1999).
These soil moisture values were used to compare soil water balance output from their
respective irrigation models to physically measured root zone water content.

Portability and depth of measurement are two of the main reasons to use this
technology. Measurements can also be taken relatively quickly. There are limitations to
the neutron scattering method of measuring soil moisture, though. First, the neutron

probe does use radioactive material and requires special licensing and safety equipment

19

to use. Second, the resolution of soil moisture is coarser for this method than some others
due to averaging of a greater soil area.

Gaze et al. (2002) performed a study to assess the accuracy and utility of the
neutron probe for measuring changes in soil moisture in a potato field. They placed
aluminum access tubes on the ridge, side, and furrow of a potato field to a depth of 100-
cm. The probe was calibrated against gravirnetric soil samples for each reading depth of
each tube under bare soil conditions. Sample readings were taken prior to and 2-4 hours
after an irrigation event. There were a total of nine irrigation events spread over three
plots. Another test was set up in the laboratory in which an access tube was placed in a
vat of soil, with measurements taken prior to and after watering events. In both the field
and lab studies, they found the neutron probe underestimated soil moisture immediately
after a wetting event, but was reasonable during soil drying fi'om field capacity. They also
applied equal amounts of water to dry and wet soils in the field and the probe could
account for more of the applied water when soil was initially drier. There was also no
difference in general trends between tube locations. They concluded that the neutron
probe has difficulty measuring water present near the soil-atmosphere interface and that
the device is inconsistent in its measurement of soil water storage under large water input
settings. Therefore, its reliability and utility for the measurement of soil water deficits
with large irrigation amounts must be questioned.

While there are both positive and negative studies regarding the utility of the
neutron probe, other technologies are available. In particular, capacitance-type probes use
similar construction and principals of soil water content measurement to the neutron

probe and TDR, respectively. An electrical field signal is generated between 2 annular

20

electrodes placed in the soil profile (Lane and Mackenzie 2001). This Signal penetrates
the surrounding soil profile, which returns a signal at a similar frequency. Some of the
energy is trapped in the water present and the probe then measures the shift in return
frequency (Tomer and Anderson 1995).

Tomer and Anderson (1995) conducted a field evaluation of a capacitance type
soil moisture probe against neutron and TDR technologies in sandy textured soils. They
chose the Troxler Sentry 200-AP frequency domain reflectometer, the same probe used
for the validation of the MWUR model. Soil cores were taken for calibration purposes of
the neutron, capacitance, and TDR probes. A linear regression equation was fit to the
frequency shift values vs. calculated volumetric water content. The calibration resulted in
a good correlation, particularly at depths greater than 1.0-m. The capacitance probe
tended to sense water near the surface, which the neutron probe could not. But they also
reported the capacitance probe had difficulty detecting frequency shifts in dry, coarse
soil. They believe this bias is accentuated when air pockets result fi'om poor soil-to-tube
interface occurs. These air pockets are important because of the large difference between
the dielectric constant of water (80) and air (1). The authors concluded the probe was
satisfactory for relative measurements of soil moisture, but cautioned the user about the
need to be deliberate in tube installation. In a similar experiment, Ould Mohamed et al.
(1997) and Khosla and Persaud (1997) were in agreement with Tomer and Anderson
(1995) and found the calibration and use of a capacitance probe suitable for in situ soil
moisture measurement. The soils tested were different textural classes, silt clay loam
(Ould Mohamed et al. 1997) and loamy sand (Khosla and Persaud 1997), but similar

conclusions were found. Both agreed with the utility of this method but report the probe

21

has difficulty properly measuring soil moisture under dry soil moisture conditions and
cautioned users about tube installation. Khosla and Persaud (1997) used a Marquard
family of equations to find a regression equation for each site tested.

Work has continued in testing of the capacitance method for measuring and
monitoring in situ soil moisture. Wu (1998) calibrated a probe for heterogeneous soil
profiles in Nepal and demonstrated a single regression curve could be calculated for each
field. By far the most reported problem with this technology is the need for deliberate
installation of the access tube. de Rosny et al. (2001) and Lane and MacKenzie (2001)
both reported the introduction of errors in frequency shifts due to air gaps between the
soil and access tube. Lane and MacKenzie concluded the utility of capacitance probe
technology was questionable because of these installation problems. Also, errors in probe
measurements increased as volumetric soil moisture rose above 35 percent. Chanzy et al.
(1998) concluded soil moisture could be accurately monitored, after calibration, in a field
using one to three access tubes. This method was chosen to validate the MWUR because
of its consistent volume of aggregation, safety, and speed of measurement.

Regional Modeling and Consideration of Scale

The scale of inputs and outputs must be considered whenever modeling any plant-
soil-atmosphere system, as errors can be introduced when aggregating or disaggregating
variables or inputs to fit the desired scale. Hansen and Jones (2000) and Anderson et al.
(2003) both discuss the problems and common pitfalls of upscaling or downscaling crop
and crop/climate models. Errors are often introduced when trying to aggregate or weigh
heterogeneity within an input pixel. Using linear areal averages can be problematic if the

inputs are related in a nonlinear fashion. Also, the model-driving inputs may change as

22

the scale of the model increases. Anderson et al. (2003) reports an example of this for the
calculation of ET. As the model scale moves from canopy to landscape, ET becomes
more dependent on feedbacks from the atmospheric boundary layer than net radiation
receipt. Hansen and Jones (2000) describe the phenomena as a shift from high-fiequency
disturbance sensitivity to low frequency disturbance sensitivity.

Hansen and Jones also suggest input sampling when aggregating data. This
involves "repeatedly using different sets of inputs sampled in a manner that captures
enough heterogeneity to reduce aggregation errors to an acceptable level". This is
currently being done for the MWUR NEXRAD Stage III precipitation estimates at the
Michigan Climatological Resources Program at Michigan State University. A GIS is a
suitable tool for this sort of task because of the ability to analyze data in different input
layers and at various scales. Raster data are already in evenly sectioned grid cells, while
vector data can take the form of odd shaped and sized polygons. These layers can be
overlayed and aggregation algorithms calculate outputs (Figure 1). Hansen and Jones
note a GIS is also a good tool for the data processing stage for similar reasons. The
meteorological and soil data available for the MUWR model are at similar scales and
extents. Figure 2 is the statewide NEXRAD grid network spatial distribution, which is the
same as the MWUR output. All have sub-county resolution and regional extent (Figure
3). In the validation phase of this study, soil moisture and water use reports are at a much
smaller scale. The area of a NEXRAD grid is 16 kmz, while a typical study field is 0.32
to 0.64 km2 (Figure 4). Plus, soil heterogeneity within a single field can be large (Basso

et al. 2001).

23

u -

mum-u

I. .muu
u u-
n

 

Figure 2. The NEXRAD 4 km grid network across lower Michigan with point locations of study sites.

 

Figure 3: The NEXRAD 4 km grid network at the county size, with field sizes delineated.

24

 

 

 

 

 

 

 

 

Figure 4: Four NEXRAD 4 km grid cells and three study fields (in black).

25

Soil moisture is probably the most highly variable parameter for a model such as
MWUR, and one of the most important to characterize. It is highly variable both spatially
and temporally and has non-linear influences on many environmental processes (Western
et al. 2002). Western et al. review many works regarding the scaling techniques of soil
moisture. Soil texture, topography, and vegetation all have effects on the spatial
distribution of soil moisture at the local scale, while variations in rainfall or even climate
may affect moisture at the regional scale (Paz et al. 2001; Batchelor et al. 2002).
Therefore, it is not necessarily desirable to quantify the spatial pattern of soil moisture,
but rather to quantify the Spatial statistical structure. Western and B10 schl (1999) were
able to capture the switch in process as scale was increased fi'om vertical water
movement at smaller scales to lateral movement at larger scales by representing the
overall effects of soil moisture processes through spatial continuity.

Western et a1. (2002) give two examples of interpretation of soil moisture
measurements with small support, or area of aggregation, and large spacing. The first
method relies on a dense sampling structure with a high resolution of data points. The
second method is the development of a relationship between point measurements and
areal soil moisture. The limitation to this method is the need for a time-stable relationship
between point measured soil moisture and the spatial mean.

One must be keep in mind the different environmental variables that drive
physical processes at different scales when validating soil moisture. The processes most
influential at the field scale are not necessarily the same at larger ones. In the MWUR
simulation, variability lies in larger processes such as precipitation and crop water

demands over a multi—field area. The point measurements used to validate this simulation

26

can quantify variability within a single field. The upscaling of the measurements to that
of the model is necessary and still has relevance. Field scale measurements have been
used previously to validate regional models (Knox et al. 1996; Sousa and Pereira 1999).

Two schools of thought dominate in the argument of the complexity of models as
scale is increased. One argues for Simpler models, hence reducing the number of inputs
and the potential for effects of bias in the data. The second group believes larger scale
models are built upon the smaller scale models, just with more complexity. While input
data can become arduous, more real world functions are taken into account (Hansen and
Jones 2000). However, input data may also be error prone. Soil maps at a regional scale,
such as STATSGO, are aggregated to a map unit that may lose individual, field level soil
characteristics. Hansen and Jones report the median CV of plant available water within
soil associations in the STATSGO data set range from 40-60%. This can have dramatic
effects on potential ET and plant water uptake, with reports of a 16% underprediction of
ET and a 17% underprediction of mean production (yield) in the Hansen and Jones
Simulation. It must be noted, though, that increasing the detail in datasets does not
necessarily mean perfect data. The SSURGO data set, a newer and more detailed soil
survey, is not available for all areas and can still result in limited detail at the field level
(Hansen and Jones 2000).

Weather data can be problematic as well. Interpolation between rain gauges may
not describe the variability of amount or intensity over short distances (Hansen and Jones
2000). The NEXRAD Stage HI radar precipitation estimates, with a spatial resolution of
4 km (Fulton et al. 1998), are better suited to sense differences in rainfall amount over

smaller distances than a rain gauge network of typical density within the United States.

27

The MWUR simulation requires a daily precipitation estimate, which is the sum of hourly
NEXRAD estimates, reducing the effect of rainfall intensity on the landscape.

Management decisions are also difficult variables to pararneterize. Decisions like
crop cultivar and planting date may have impacts on model outputs (Hansen and Jones
2000). Crop cultivar is not considered at all in the MWUR simulation, but planting date is
set for a single date across the state for each crop. Distribution of crop acreages across the
state is also necessary.

The issue of scale is a difficult aspect of modeling and model validation. It is
important to characterize the level of detail required to adequately describe the desired
process. Data availability, processing capabilities, and spatial and temporal extent all
must be considered when making the modeling decisions. Weather data (particularly
precipitation), soils data, and crop physiology are inputs of the greatest concern for
MWUR. Characterizing the sensitivity of the model to these inputs is of great interest
because it describes the driving parameters at the regional scale and the level of
aggregation for output optimization.

Conclusion and Summary

The simulation of environmental physical processes is continuing to better
characterize the ‘real world’ for growers, regulators, and decision makers. This thesis on
modeling crop water use from irrigation is concerned with just one of the many important
physical processes in agricultural production. While it does not take many cultural or
societal issues into account, the model does have the potential to complement a suite of

models that do better characterize this social aspect.

28

Bland (1999) discusses the need for agricultural modelers to begin working on
integrated assessment, and specifically the integrated assessment models. The goal of
integrated assessment is to produce scientifically-based information for environmental
debates introduced in public and governmental forums, while respecting the human
value. He proposes this modeling work focus on creating a system of diverse models
interacting with one another so interaction between systems not usually considered
together can be assessed. Specifically, models with a predictive capability are needed.
System models, composed of small physical models describing farm—level functions, can
be aggregated to describe and predict farming operations, and the broader agricultural
system, depending upon various social choices. Bland calls this new paradigm “Agrarian
Systems Modeling”. This term describes the construction of ISMS for food-production
systems. His example of the SIVI model for the potato and vegetable industry in central
Wisconsin links agronomic practices, spatial distribution of fields, groundwater flow,
crop yield and value, income derived from that yield, and linkages between the industry
and public infiastructure. The goal of this model is to inform the public about the
environmental and economic impacts the industry has on the region.

I believe, with firrther work, the MWU R model could become an integral part of a
similar simulation for the state of Michigan, and perhaps the Upper Great Lakes Region.
When integrated with a suite of models, MWUR could give important contextual

information on the volume and distribution of irrigation water use.

29

Methods
Overview

The validity and sensitivity of the Michigan Water Use Reporting model,
developed at Michigan State University (Moen 1999) was tested. The model estimates
seasonal water use at a 4 km resolution for irrigated agriculture across the state. Farm
locations and crop acreages are the necessary inputs into the simulation. While the
resolution of the model is 4 km, the results must be aggregated to county level because of
privacy issues.

Validation involved calculation of mean differences and mean absolute
differences between simulated results and data collected from cooperating growers across
lower Michigan. Data collection included recording volumetric soil moisture and grower
reports of irrigation volume applied to each crop during the 2002 and 2003 growing
seasons. The crops studied are the largest irrigated crops by acreage and volume of water
applied. They include com, potatoes, and soybeans, along with specialty crops. The farm
locations were located in counties with large acreages under center pivot irrigation. This
type of irrigation sprinkles water from above the crop canopy through a boom fixed at
one end and free at the other, giving it the ability to rotate across the field. This was the
most widespread irrigation delivery system in the state in 1997 (MASS 1998).

Monitoring soil moisture involved contacting potential cooperators, choosing
field locations, installing observation tubes, and taking bi-weekly measurements of
volumetric soil moisture values with a capacitance probe. Potentially willing growers
were contacted regarding cooperation with the study. Of those willing, field level sites

were chosen based upon location within the state, crop type, and distribution of crops in

30

the area (Figure 3). An effort was made to be as comprehensive as possible in the
selection of crop type and distribution in the major irrigated areas of the state. Field level
spatial resolution was chosen because of data availability from the growers and the soil
moisture measurement device. Attempts were made to monitor the same fields in both
seasons of the study (years 2002 and 2003), but some compromises had to be made to
ensure a more representative crop distribution.

Field sampling involved physically carrying the FDR unit to each access tube,
lowering the probe to the proper depth, and taking a reading with the aid of a datalogger.
A number of considerations were taken into account before final field selections were
made. First, a number of irrigation delivery systems are utilized in Michigan. Center
pivot systems are associated with the greatest irrigated area and volume of water applied
in the state in 1997 (NASS 1999). Therefore, these systems were the focus of this study.
Drip, furrow, and subsurface irrigation, among others, account for a small percentage of
the total volume of irrigation water.

General farm locations were considered based upon the amount of center pivot
irrigation taking place in their region. Two of the counties chosen, St. Joseph and
Montcalm, were the largest irrigating counties in the state by acreage in 1997. These
counties account for approximately 19 percent of the irrigated acreage of the state.
Saginaw County, in eastern Lower Peninsula, has been the location of past residents vs.
grower disputes. The three counties mentioned, plus Mecosta County, account for
roughly 21.3 percent of the irrigated acreage in the state in 1997 (MASS 1998). St.
Joseph County is located in the southwest portion of the Lower Peninsula of Michigan,

while Montcalm and Mecosta Counties are all located in the west-central part of the

31

Lower Peninsula. Figure 2 shows the relative location within the state of the study sites.
Agricultural soils in all three counties tend to be loamy sand to sandy and have high
infiltration rates. Saginaw County soils tend to be higher in clay content with poor to very
poor infiltration and drainage (Survey 1960; Survey 1983; Survey 1984; Survey 1994).

Simulation runs occurred after all the seasonal weather data was made available.
Temperature, solar radiation, and radar precipitation estimates were the meteorological
data used to drive the model. These variables are available as raster data sets.
Temperature and solar radiation data are at a 20 km resolution, while radar rainfall
estimates are at 4 km resolution. The model calculates water use at the 4 km resolution.
Figure 2, Figure 3, and Figure 4 show the relative size of the fields at the state, county,
and local scales, repectively, to the NEXRAD grid network. Other inputs include a
statewide soil texture data set, the STATSGO data set, crop acreages, and farm locations.
Measurement of Soil Moisture

Soil moisture was monitored in the field using a Troxler Sentry 200-AP
capacitance probe (Irrigation Scheduling Methods, Inc. (ISM), Malaga, WA). This probe
relates oscillation fiequency to soil water content, based upon the capacitance, in relation
to the dielectric constant, of the surrounding material (Robinson et al. 1998). Water has a
much greater dielectric constant air (Wu 1998). The probe requires calibration for each
soil type being measured (Tomer and Anderson 1995; Khosla and Persaud 1997; Ould
Mohamed et al. 1997; Chanzy et al. 1998; Wu 1998; Lane and Mackenzie 2001; Mandal
et al. 2002) because soil is not a perfect dielectric. The soil and material surrounding a

tube is an electrical conductor (Robinson et al. 1998). The probe readings are an

32

integration of a cylinder of soil 30 cm high and 10 cm radius from the wall of the tube
(Wu 1998; de Rosny et al. 2001).

Calibration of the capacitance probe began with the installation of access tubes in
the field. These tubes are 5 cm Schedule 40 polyvinyl chloride (PVC) pipe. Installation
began when the 1.7 m tubes were pounded vertically into the ground using a sand filled
mallet. Soil was then augered out of the middle of the tube. Care must be taken to ensure
a tight soil-tube interface. The tubes were driven to a maximum depth of 1.5 m, but many
times obstacles in the soil profile, like stones, prevented installation to the maximum
depth. A minimum depth of 0.9 m was achieved in all fields. This depth is satisfactory
because it encompasses most of the rooting zone for all crops considered in this study.
The excess PVC above ground was clipped to a height of 0.15 m. Sample probe readings
were taken at 0.3 m intervals when installation was complete (ISM1996). Three access
tubes were placed in transect across each field, about 75 m apart. The three tubes were
installed per field (Chanzy et al. 1998), totaling 39 tubes per season for the entire study.
Care was taken to try and line the access tubes with the irrigation pivot in a radial line to
reduce the delay in the time between which individual tubes saw irrigation events.

Immediately after installation, bulk density cores were taken near the access tube
at the median depth of the probe readings in the soil pro file because the probe integrates
an entire 0.3 In section. These samples were placed in tins and sealed in plastic bags to
minimize evaporative loss. The procedure was repeated for each tube installed.
Approximately once every two weeks, bulk density samples were again taken for each
depth, near one tube. As many samples were taken as time permitted for each field

through the season.

33

Once in the lab, bulk density cores were weighed, dried in a 105-degree C oven,
for 24 hours, and re-weighed. Volumetric soil moisture and bulk density were calculated
and recorded. A regression curve was fitted to find an equation relating the oscillation
frequency shift probe values, or “raw values”, to the volumetric soil moisture values.
Marquard’s non-linear least squares algorithm was used in statistical software to calculate
the equation values (Khosla et a1. 1997). They calibrated the same Troxler 200-AP probe
and found favorable results with this type of equation. The equation takes the form of:

D = xleM”‘2+x3 (1)
Where D is the frequency shift reading from the probe, M is the volumetric soil moisture,
and x1, xz, and x3 are constants (Khosla and Persaud 1997; ISM 1999). For calibration
purposes, the volumetric soil samples fi'om bulk density cores were used as the
independent variable, M. The probe readings taken immediately after a bulk density
sample were the dependent variable, D. The statistical program calculated constants x1,
x2, and X3. These constants were then used to calculate volumetric soil moisture for each
field observation through the season by rearranging equation 1 and solving for M.

M = (Ln {})/X2 (2)

Tomer and Anderson (1995) attempted to use a linear regression equation to
calibrate a Troxler ZOO-AP. The volumetric soil samples provide a second method of
calculating soil moisture and were meant for the development of field level soil moisture
equations for use with the capacitance probe. This method resulted in very poor 1’2 values
for fields in which convergence could not be attained using the Marquard family

equation.

34

The results of the curve fitting exercise were not positive. In some cases,
convergence requirements were not met after 100 iterations of the regression. Even in
cases where convergence requirements were met, the equations would result in unfeasible
soil moisture values for bi-weekly field observations. Table 2 is an example of erroneous
field observed soil moisture. Some values are negative (volumetric soil moisture) and
some result in numbers approaching infinity. Table l is an example of reasonable results
from the calibration process for another field observed soil moisture. Unfortunately, this
was only successfirl for 5 out of 13 fields, three in Saginaw County and two in Mecosta
County. This trouble in the development of some field equations altered the use of the
oven dried volumetric soil sample data. The large amounts of soil texture heterogeneity
and lack of data points, too few bulk density samples per field, are most likely to blame
for these problems. Instead, the data were utilized as a check of the capacitance probe
data calibrated using the manufacturer’s pre-developed equations.

Basso et al. (2001) reported soil moisture is a function of both soil depth and
texture. They also found that soil texture was highly variable in a 7 ha field near Durand,
MI, within the study area of this research. Sand content was reported to vary from 40 to
80 percent and clay content 8 to 25 percent across the field. Soil water was reported in
terms of potential extractable soil water and the range of values was 140mm, in low
areas, to 70 mm, in high areas. A correlation was found between topography and soil
texture. Clay content was highest in lower elevations, and sand content greater in higher
elevations.

ISM offers calibration software as part of the probe package. The software

requires an initial soil textural classification (i.e. sand, loamy sand, etc.) and initial water

35

 

.33 60 8.3235 £23 :8 20¢ E9: cone—33. nexus—co 5.89%.... 5.? 33.3.3 9.322: :8 8.52:2...» "N 2...;-

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

3... a... 8.8 8.8 8.8 8.» 8r .8 8.8
2.2 8.2 5.2 2.2 8.2 2.2 8. 2- .8 8.8 a 83
8.2 8.2 8. 8 8.2 8.2 8. 2 8.8- so 8...

8.8 8.8 s 2.2. a 8.8 8... .8 8.8
8.2 8.2 8.2 8.2 8.2 2.2 8. 2- .8 8.8 u 33
8... 8... 8.2 8.. 2.... 88 8...- .8 8...
oo oo 8 oo oo 8 mV. PN Eu Owns
8.2 8.8 8.2 8.2 8.2 8.: 8. 2- .8 8.8 2 8.:
8.2 8.2 8.8 8.2 8.2 8.2 2.8- so 8...
88.2: 88.3... 88.2... 88.... 88...: 88.8... 88.2...
, . . . w .w...§.....so 2:88: =8 2853.5. 82188888 ., . . __ , m ‘

.mcan .GU Bag—3am 60.30 __am 30¢ 59: 62—0—993. Gaza—~70 flcugouuoh :33 Baa—52.3 0.533:- =¢m ovGoEESr “n 033—.
E. 8 8.8 2. 8 2.8 8.8 8.8 2.8 .8 8.8
8.2 8.2 8.2 8.2 2.2 8.2 8.2 .8 8.8 n 83
2.... 8.... 8.. 2. 2 8.8 8... 4.8 .8 8...

8.8 8.8 8.8 8.8 8.8 88 m 2.8 .8 8.8
8.2 3.2 2.2 8.2 8.3 8.2 8.2 .8 8.8 u 83
8... 8.. 8.2 8.2 8.. 8... 8... .8 8...
2.8 2.8 8.8 8.8 :8 8.8 8.8 .8 8.8
8.8 8.2 8.2 8.2 8.2 2.2 8.2 .8 8.8 F 8.:
8... a... 8. 2 8.2 8. 2 8. 2 2.... .8 8...
88.8.. 88: 2... 88.3.. 88... E 88...... 88...: 88.8...
rec..so.eau_oa=8§§2$82a8=§880

 

 

 

 

 

6
3

content as a percent of field capacity. The initial water content was determined with
litmus paper and color chart provided with the installation package. With soil texture,
initial water content, and initial frequency shift, or “raw value” from the capacitance
probe, the software determines the calibration equation for each layer. Each calibration
equation has an associated letter code, which the software outputs for each layer of each
tube. The letter output by the software was programmed into the datalogger for the
associated tube and soil profile depth. Each letter has its corresponding equation pre-
prograrnmed in the datalogger. These pre-determined “equations” are nothing more than
equation 1 with each letter code having been solved for the set of constants, X1, X2, and X3.
See Table 3.

Mean differences between the factory calibration and volumetric calibration show
an increase in soil moisture using the volumetric calibration method (Table 3). The
increase in mean difference of water content also increases by depth and was greatest at
the third layer (60-90 cm). The factory calibration method gives each layer a
predetermined equation, based upon soil texture and antecedent water content. In Table 4,
the mean differences between probe measured and volumetrically determined soil
moisture is small through the profile. With this evidence, along with the lack of data
points to determine equations for individual fields, it was decided to use the
manufacturer’s calibration methods and codes.

Observation tube installation began in early June for 2002 and early May for

2003. The late start in 2002 was due to shipping delays fiom the soil moisture monitoring

37

.88 8.2.. 8 L... 58.. 3 959.25. 25%...
=8 9.52.5.9» .e 3:25.532: 9.8 :8 952:2?» .2... 2.9... 92.8.2.5... 3.95:3 .282... .5952. o2.o..o....... 338.... 5.2.. F...“ 53.). .v 0...“...

 

 

 

 

 

 

 

38

 

 

 

 

. . m... on.
ovo o moo o ..<
oooo mood Eu oo-oo
Fmoo mood- Eu oo-om
wood wood- Eu on-o
”Eur—:0 _ Earn...
, 3:20.30 coco-BEA.
83.3.... 53: cue-.2

 

A33 .556 89.53.. 2.9:. 3.... :8... 9.322..
:8 2.52.5.2. 22.5.3 2.3.3.5....— ...=. .5323; 2.9... 2.. 3... 3.2.2.9... 9:. 253230 :23. 2.3 .3 8......» 2E3 3.3.5:: PBS... 5 935.

 

 

 

 

 

 

 

 

 

mmodmmv v Foo Fm.» Nodomv oomv 9.9. m Fmv owmv vav ommv ovmv on Fv ox
mvmoo- owvod- oomod- mmmod- o 50.0. momod- omvoo- wovoo- Fomoo- Novoo- mwmoo- «x
am... now. Now- ooo- omo F- Fm F F- mom F- mom F- vow F- mom F- who F- .x

... _ I 0 n. m n. U U<u_ m <

 

 

 

 

 

 

 

 

 

equipment distributor. Installations in 2003 began shortly after planting. Work started in
the southern most county, St. Joseph, and was concluded in Mecosta County to the north
three weeks later. Installation of the tubes was completed in early July for 2002 and early
June for 2003. Afterwards, each field was visited twice per week for sampling. Fields in
Montcalm, Mecosta, and Saginaw counties were visited on the same days (Monday and
Thursday) while sites in St. Joseph County were visited separately (Tuesday and Friday).
Observations continued through the growing season until either the crop was harvested,
in the case of some potato fields, or irrigation was stopped for grain dry-down just prior
to harvest. The tubes were removed fi'om each field immediately prior to harvest so no
damage would be caused to any harvest equipment.

Probe readings are an integration of the volumetric soil moisture for a 0.3 m
depth of soil (Tomer and Anderson 1995; ISM 1996; Khosla and Persaud 1997). Soil
moisture data were analyzed for depths to 0.9 m because all tubes were driven to a
minimum of this depth and for consistency across all study areas. The 0.9 m tube depths
resulted in three 0.3 m volumetric soil moisture values for each tube. The capacitance
probe was lowered to the proper depth, a measurement was taken, the reading was saved
to a file, and the probe was lowered to the next depth until all depths had been measured.
Each of these depth measurements were made at all three tubes in an individual field and
each field was visited twice per week. Before a field was re-visited, data was downloaded
to a PC in the laboratory. While all the individual fields could fit in the memory, only one
visit per field could be stored on the datalogger. A RS-232 connection linked the
datalogger to a PC running the download program (ISM). Data was later compiled and

analyzed using spreadsheet software. The volumetric soil moisture observations were

39

taken to validate the MWUR simulation volumetric soil moisture results and were also
saved in the spreadsheets. MWUR model runs began when the temperature/ solar
radiation data grids and NEXRAD radar precipitation estimates for the year were made

available fiom NWS and the Michigan State Climatological Resources Program.

Model Validation
The goal of this portion of the study was to assess the validity of soil moisture and

irrigation water volume applications of the MWUR model when compared to field level
data. The model was validated in two parts. First, the simulated volumetric soil moisture
output was compared to volumetric soil moisture measured in test fields around the state
of Michigan. Second, growers supplied the values of irrigation water volume applied on
test fields. The reported volumes were compared to the irrigation water volume output by
the model for the grid cells encompassing each test field.

The robustness of the model was evaluated through calculation of mean
differences between MWUR irrigation water volume application estimates and values of
irrigation water for individual study fields from cooperators. Descriptive statistics, such
and mean difference and mean absolute difference, averaged by crop and by grower,
were used to describe the relationships because of the desire to find the degree to which
the two datasets are different and to find any bias between them (Ould Mohamed et al.
1997; Chanzy et al. 1998; Sousa and Pereira 1999; Basso et al. 2001). Given the limited
number of data points available, a maximum of 26 fields for the two years, descriptive
statistics best describe the relationships than other tests. Unfortunately, some cooperators
failed to provide seasonal irrigation records for certain fields. These fields were not

included in any part of the statistical analysis. Records submitted left 21 of the 26 fields

40

with complete records. Paired t-tests were utilized to test the null hypothesis that states
the simulated and observed volumetric soil moisture and irrigation depths have the same
mean. Biases due to within field and regional soil heterogeneity were overcome by
plotting simulated values as a percent of observed over time. Discrepancies between
model irrigation amounts and observed irrigation amounts applied were described by
plotting the difference between the irrigation amounts against potential drainage.

To begin, model outputs were compared to both field measured and grower
supplied data. Calibrated probe volumetric soil moisture data for each field were first
entered into a spreadsheet program. Measured soil moisture values for each depth on each
date were averaged to lessen the effects of within- field heterogeneity. The model was run
for each field based upon field location in relation to the NEXRAD grid network for
years 2002 and 2003. Model parameters were initially set to default values (Moen 1999).
The results for seasonal irrigation depth, irrigation dates, and daily volumetric soil
moisture were written to a database and, after simulation executions, individual locations
were queried and the resultant outputs saved in a spreadsheet. Queries by crop were also
saved. Simulated soil moisture is calculated in layers through the soil profile to a depth of
160 cm. There are 10 layers, consisting of 2, 5, 8, 11, 14, 20, 23, 25, 25, and 25 cm
depths below the surface, respectively (Figure 5). These values for each layer must be
aggregated to layers of 30.48-cm in size, corresponding to the layer depth of the
capacitance probe measurements. A weighted average of these values was taken to
convert to layer depths of the capacitance probe.

After the daily values of soil moisture were calculated at each depth for the

growing season, another filter was applied to match dates of measurement with calculated

41

Tube Height
1524cm

MWUR Model Layers

all

 

 

 

Layer 1 (2.0 cm)
Layer 2 (5.0 cm)

 

Layer 3 (8.0 cm)

 

Layer4 (11.0 cm)

 

 

Layer5 (14.0 cm)

 

Layer 6 (17.0 cm)

 

 

Layer 7 (20.0 cm)

 

Layer 8 (25.0 cm)

 

 

 

 

C-PFDDB MBBSUI'EU L3 BIS
Soil-Atmosphere If'IIB 8C8

Layer 1 (30 .48 cm)

Layer 2 (30.48 cm)

Layer 3 (30.48 cm)

 

 

Figure 5: Visualization of model layer depths (left) and capacitance probe measurement layer depths

(right) with tube placement in soil profile.

42

dates. Mean differences and mean absolute differences were calculated for each crop, by
layer and for years 2002, 2003, and the two years combined. Paired t-tests were run by
crop for the two years combined. Similar tests were performed when data were grouped
by grower.

The grower supplied daily irrigation volumes for individual crops were entered
into a spreadsheet. Model output of the daily irrigation and irrigation events were queried
and saved into the spreadsheet. Timelines of depth accumulation were plotted for
modeled and reported values. Seasonal irrigation volumes are calculated by summing
daily amounts and the mean difference and absolute mean difference were calculated for
each crop and each county on a yearly basis. Mean difference and mean absolute
differences were calculated for each crop for 2002 and 2003 combined. Finally, paired t-
tests were conducted by crop on the irrigation volume amounts and soil moisture for 2002

and 2003 together.

Model Sensitivity Analysis
Four grower managerial and crop physiological parameters were adjusted to test

model reaction. These tests are meant to compliment the previous sensitivity analysis of
Moen (1999). He tested the sensitivity of the model to changes in soil map units of the
STATSGO data set. His results indicate fewer total map units may be used across the
state and more aggregation of soil texture classes within 4 km grid cells can occur. He
also tested propagation of errors in the NEXRAD data set. The results indicate cells
further from radar stations have greater errors. Neither of these model parameters was

tested here because of the results fi'om the previous study.

43

Instead, four crop and managerial model parameters were tested. Choice of these
parameters was based upon results of the validation portion of the study. Daily irrigation
accumulation led to questions regarding timing, amount per irrigation, and season length.
Analysis of volumetric soil moistures at depths led to questions of root growth and
development. The altered parameters include root growth rate, planting date, amount of
water per irrigation event, and soil moisture irrigation trigger (Oc). As roots grow faster
and deeper, there is more potential for plant water extraction at different depths. Planting
date changes may affect annual soil moisture cycles and when the irrigation season
begins and ends. Amount per irrigation and Oc changes affect the frequency and duration
of irrigation cycles. Results can be seen with larger or smaller and more or less frequent '
irrigations.

Rooting growth rate was adjusted for each crop, and similar methods were used
for analysis as with the validation portion. Mean and mean absolute differences, as well
as paired t-tests were calculated for each parameter change. Similar methods were used to
test soil moisture irrigation trigger (Oc), irrigation amounts, and planting date parameters.
These parameter changes were run for all the study fields in 2002 and 2003. These
statistics were calculated upon the differences between changed parameter and default
simulated volumetric soil moisture and irrigation depths.

The variable (9c is the fraction of water withdrawn fi'om plant available soil water
that triggers an irrigation event. The variable was changed within the range of 0.40 to
0.60 of plant available water in increments of 0.05. The default value is 0.50. The output
values of soil moisture and irrigation events were exported to a spreadsheet, where the

soil moistures were averaged to 30.48 cm layers and the seasonal irrigation amount for

44

 

each location was calculated. The values for each location were placed in a table and
differences between the adjusted Oc results and default results were found. Mean
differences in volumetric soil moisture were calculated from planting date, day of year
(DOY) 135, until the model ended calculation for the particular crop. The total irrigation
amount mean differences were calculated by crop for 2002, 2003, and the two years
combined.

Irrigation amounts per even were changed fi'om a default value of 25 mm to a
maximum of 38 mm and a minimum of 12 mm, in 6 mm increments. The root grth
rate look-up table in the model relates root depth to either accumulated growing degree
units or days after planting, depending upon the crop type. The depth of the root at the
given steps was increased by 10 and 25 percent and decreased by 10 and 25 percent for
all crops. The model ran for each of these scenarios and descriptive statistics and paired t-
tests were calculated. Planting of crops does not occur on the same day across the entire
state. There can be a period of weeks between far southern and northern Michigan. The
model sets the planting date of most crops at DOY 135, but a few of the specialty crops
are set at DOY 150. The model ran with adjusted planting dates ranging from -10 days to
+10 days, in 5 day intervals. Summary statistics and t-tests were performed on the soil
moisture and irrigation volumes.

Once these summary statistics were calculated for the initial run and parameter
sensitivity runs a “best fit” characterization of physiological and managerial parameters
were made. After the most sensitive parameters were determined, these model variables
were adjusted until the smallest mean difference between seasonal simulated and reported

irrigation depth for each crop were found. These “best-fit” parameter values were

45

recorded for each crop. Irrigation amount per event, trigger soil moisture deficit, and

season length were adjusted. Summary statistics were calculated for each of these runs.

46

Results

Model Validation

The first and most important objective of this study was to determine how close
the MWUR irrigation depth and soil moisture simulated outputs are to those reported and
measured on commercial farms in Michigan. In an effort to quantify any differences,
volumetric soil moisture was monitored in field during the 2002 and 2003 growing
seasons for 6 different commercial crops: corn, potatoes, soybean, sugarbeet, carrot, and
pepper. Corn, potato, and soybean crops were monitored across both years of the study.
Overall, a total of seven com, nine potato, two soybean, one carrot, one sugar beet, and
one sweet pepper fields were monitored during the study. Grower-supplied irrigation data
were not available in all the fields monitored, and these fields were not included in the
statistical analysis.

Mean and mean absolute differences between simulated and reported seasonal
irrigation depths by crop are given in Table 5. Simulated irrigation depths were greater
than those reported by the growers for the individual 2002 and 2003 seasons and for each
crop type considered in the comparison. The model also tended to overestimate seasonal
irrigation across all crops, with a mean difference of 54 mm for potatoes, 59 mm for corn,
and 117 mm for soybeans. Mean absolute differences, a better indicator of the magnitude
of individual differences, ranged from 16.8 mm for carrots during the 2002 growing
season to 159.0 mm for peppers in 2003. Simulated seasonal irrigation depths were on the
order of 135 percent of the reported seasonal totals for com, 124 percent for potato, and

197 percent for soybean over the duration of the study. From Table 6, model standard

47

1' .. 91"

.3533 noon, .23 Sen 2338 3. 3.3.3.. 9333.. 5...... 3.33.... 3338 33.2.9. .23 33.255 .o v.3...

 

 

 

 

 

 

 

m. 3 v. 8 m 8 am 8:83.
2.2 v.8 4.8 cm .822
32.3.... 033.. :30

 

 

8953... .222:

2.33.. 3.2. 33.3.3 2.2.2. 33.2.25 62.2.2.3 939» 95 2.. .23 £ch .33 .2338 3.. 3.3 .3 33.3.3 .333... 33.2.9. .3 239.2. a
3 53.. 33“.... .3338 33.2.2.8 .23 A225 5...... 33“.... .3332. 3.22.9. .23 33.3.... 3952. 333...... 22.8.3 .32.. .23 23.). .m 0.3...

 

 

 

 

 

 

  

 

 

2.3.3.... .A

 

f. -. .. .. u..-.

....\l.

 

. . . . . . . . . . . . 3.33m
momF vvwF vmmF «mum mem momF FmFF an? meF ovoF mowF meF 333.3.
.25.
. . . . . . . . . . . . 33.2.5
moFF me mm.- ommF meF «5 m Fm mmF Nmm m3 F8 mmF 23.32
:35.
.EE.
m.oF F Yam Fem odm F deF «.8 F.mm m.oF «.3 QB tam was 333:5
:3:
333m 03.3. _ .500 3.2.9... _, 32.3w. _ 033m. ,_ E00, .230 __ 32.395 Enogom F 0.33... _ :30
....,__._,,7 “’> --< . mum.“ -..-......m...w......,.... DIP ma...- ,..,.,.. .. H..... A .,... .I .2 . “n... .x: -, .. , .. , N8“ ., . g...“

 

 

48

deviations of irrigated depth of water for the two years ranged fi’om 17.7 mm in soybean
to 37.4 mm in com. Reported irrigation depth standard deviations for the same period
ranged from 44.9 mm in soybean to 87.4 mm in potato. Across individual crop types that
were replicated during the same season, there was large variation among the differences,
ranging from 79.1 mm for com in 2002 to 249.3 mm for potato in 2002. Individual field
differences can be seen in Table 7. This data suggests there is large variability in
irrigation depths between individual growers across the state.

A scatterplot of simulated versus reported seasonal irrigation depths is given in
Figure 6. The broad scatter of the points and low r2 of the fitted regression line indicate
relatively poor agreement between estimated and observed and fiirther illustrate the
tendency of the model to overestimate irrigation. Regardless of the many differences
between the simulated and observed irrigation totals, when they were compared
statistically with a paired t-test, none were found to be significantly different at the a =
0.05 level, suggesting non-rejection of the null hypothesis that the means of the two
populations are equal.

Comparisons were also made of simulated versus observed volumetric soil
moisture. Soil moisture is a critical variable at the heart of the simulation. The values
calculated for each layer of the model determine the water content available to the plant
and strongly influence the frequency of simulated irrigation throughout the growing
season. The soil moisture values the simulated by the model were aggregated to layers of
a depth in the profile equal to that monitored by the capacitance probe, in this case three

layers of 30.5 cm depth.

49

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Model Reported Absolute
Irrigation Irrigation Difference Difference
Depth (mm) Depth (mm) (mm) (mm)
Location Crop 2002
Mecosta Corn 225 203.2 21 .8 21 .8
Montcalm Corn 200 1 1 1 .8 88.2 88.2
Saginaw Corn 200 99.1 100.9 100.9
St Joe Corn 250 157.5 92.5 92.5
Mecosta Potato 325 241 .3 83.7 83.7
Montcalm Potato 300 157.5 142.5 142.5
Montcalm Potato 325 247.7 77.4 77.4
Saginaw Potato 300 231 .1 68.9 68.9
St Joe Potato 325 431.8 -106.8 106.8
Saginaw Beets 200 144.8 55.2 55.2
St Joe Soybean 250 152.4 97.6 97.6
Mecosta Carrot 300 283.2 16.8 16.8
2003
Mecosta Corn 175 158.8 16.3 16.3
Saginaw Corn 250 106.7 143.3 143.3
St Joe Corn 150 234.2 -84.2 84.2
Mecosta Potato 275 266.7 8.3 8.3
Saginaw Potato 325 137.2 187.8 187.8
St Joe Potato 250 182.9 67.1 67.1
St Joe Potato 300 294.6 5.4 5.4
Saginaw Pepper 286 127.0 159.0 159.0
St Joe Soybean 225 88.9 136.1 136.1

 

 

 

 

 

 

Table 7: Simulated, reported, differences, and absolute differences of seasonal irrigation depth for

individual study sites, 2002 and 2003.

50

 

 

  

250 7
y = 0.3073x + 199.46

r2 = 0.2289

   

é
o

8

so 1—‘ — -- _,
50 100 150 200 250 300 350 400 450 500
Reported Seasonal Irrigation Depth (mm)

 

 

"ii‘T " ' __fi- _——_fi___ ' __’T—'_ '

___, _ _, __ ,,_____..___._T.

Simulated Seasonal Irrigation Depth (mm)
l0
8
O
O
0

Figure 6: Simulated vs. Reported Seasonal Irrigation Depth (mm) for all reporting fields in 2002 and
2003. Simulated results calculated using default model settings.

51

 

Mean and mean absolute differences of volumetric soil moisture, averaged in the
30.5 cm layers and across replicates for each crop type, are given in Table 8. The
differences in many of the layers were positive (indicating overestimation of soil
moisture) with the exception of carrots, corn, and soybean in the top 0-30.5 cm layer, in
which the negative difference suggested underestimation. For comparison, a typical
range of plant available water content, approximately the difference between field
capacity and wilting point of a soil, values are generally in the range of 0.08-0.10 in/in
for the sands and loamy sands found at most study sites (Survey 1960; Survey 1983;
Survey 1984; Survey 1994). The model calculates total volumetric soil water content and
not plant available water, but the values fiom the soil survey give an indication of the
range of values (from dry to wet conditions) expected from the model. The negative
values in the top layer ranged from —0.040 cm3/cm3 for soybeans to —0.004 cm3/cm3 for
corn. The differences for potatoes, sugarbeet, and peppers are positive and somewhat
larger ranging from 0.002 cm3/cm3 for potatoes to 0.120 cm3/cm3 for sugarbeet. Paired t-
test results indicate significant differences between the populations of modeled and
observed soil moisture for most crops. The total number of samples was 108 for com, 98
for potato, and 32 for soybean.

Lower in the soil profile, simulated soil moisture in the second and third 30 cm
layers of the profile was greater than observed moisture for all crops. The mean
differences increased with increasing profile depth for all crops. For example, the second
layer ranged from 0.011 cm3/cm3 in soybean to 0.061 cm3/cm3 in potatoes over the two
years of observations. Mean differences, and the output data itself, are consistent across

most crops in the third layer. Mean absolute differences for all layers were of the same

52

 

Mean Difference Mean Absolute

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(cm’lcm’) Difference (cm’lcm’) Significance
0-30 cm Soil Profile Depth
Corn -0.004 0.049 *
Potato 0.002 0.032
Soybean -0.040 0.048 *
Carrot -0.023 0.027 "’
Sugarbeet 0.121 0.121 *
Pepper 0.057 0.057 *
All Crops 0.007 0.048
30-60 cm Soil Profile Depth
Corn 0.018 0.061
Potato 0.061 0.064 '
Soybean 0.01 1 0.043
Carrot 0.051 0.051 *
Sugarbeet 0.119 0.119 *
Pepper 0.148 0.148 *
All Crops 0.048 0.068
60-90 cm Soil Profile Depth
Corn 0.044 0.051
Potato 0.050 0.053 “'
Soybean 0.053 0.053 '
Carrot 0.056 0.056 "’
Sugarbeet -- -- _.
Pepper 0.123 0.123 *
All Crops 0.054 0.058
All Layers
Corn 0.019 0.054 *
Potato 0.037 0.050 '
Soybean 0.008 0.048
Carrot 0.028 0.045 *
Sugarbeet 0.120 0.120 "
Pepper 0.110 0.110 *
All Crops 0.036 0.058

 

 

 

 

 

Table 8: Mean and mean absolute differences between simulated and observed soil moisture
(cm’lcm’), calculated by crop and soil profile depth for years 2002 and 2003 combined. Simulated
soil moisture calculated using default model settings. Significance tested at a = 0.01 level.

53

magnitude as mean differences, suggesting the level of individual differences were
similar.

Differences between simulated and observed soil moisture were also analyzed
geographically, as each field and each sample location may have unique soil properties,
based upon soil type, parent material, topography, and other factors (Bechini et a1. 2003).
The ratio of simulated and observed soil moisture was calculated, as a percentage, over
time during the growing season at each field sampled in an attempt to overcome any
biases introduced by these geographical differences. An interesting trend was found for
all layers of corn during the two-year period (Figure 7a, b, c, d). The simulated
percentages were relatively high at the location in Saginaw County, similar to the
observed values in Mecosta and Montcalm Counties, and less than the observed values in
St. Joseph County. These differences suggest the possibility of spatial biases within the
model, perhaps due to variation in physical properties of soils. At lower layers, these
general trends, greater or less than observed, were similar within each of the fields.

Finally, the management strategy of the grower must be considered. In the
MUWR system, the assumption is made that every grower applies an equal amount of
water to supply plant needs at exactly the most opportune time. This assumption has
obvious shortfalls because each grower may utilize different strategies or methods to
determine when to irrigate. The numerous commercial and government irrigation
schedulers and scheduling recommendations, including the new GAAMP (2003) protocol
in Michigan, are just an indication of the variety of ways growers determine when to
irrigate. Growers also face practical problems, such as the minimum time needed for the

irrigation system to complete one watering cycle in a given field or a limited maximum

54

200

 

180-

 

160-

 

140

 

 

 

 

120 -

100

 

 

8

 

 

 

40 ~-

Percent of Observed Soil Moisture

20.—_——_~ .. _.__ -._.___* ._ .. _ .%

 

 

 

 

NNNN

Day of Year

 

f- o — Mecosta 00.. MO cm Depth +Meoosta 00.. 30-60 cm Depth—
A- - Mecosta 00.. 60-90 cm Depth

 

 

Figure 7a: Simulated volumetric soil moisture, as a percent of observed soil moisture by depth for
corn in Mecosta County, 2003. Simulated values calculated using default model settings.

55

200

 

180

 

160 ,

140»

 

120 ,

100

 

 

 

40 u.- -~__ *fi. ,.___ %.7-___. 2_ .. ___ _ ___ ,_,#.___. A .. _

Percent of Observed Soil Moisture

20 _. _._ .______i ___ - __ ___ ._2__ _ __ ___.-. , ___

 

 

 

‘0 N «0'5 Q‘L‘J‘bkhflo'bb‘b’lv‘oQ:
4Medeeedeeeeeeeeeeeece
Dayonear

 

:5: Montcalm 00.. 3-50 cm Depth +nMontcalTnflCo..i30-60” W “SID—em,
C _-__A- - - Mon_tcalm Co., 60-90cm Depth *

 

 

 

Figure 7b: Simulated volumetric soil moisture, as a percent of observed soil moisture by depth for
corn in Montcalm County, 2002. Simulated values-calculated using default model settings.

56

_.L
b
o

 

 

 

a... _.l
8 B

 

 

.——a

 

 

 

 

 

 

 

 

 

 

2

3

a

.2

O

E

'8

0)

2 8° ‘

2

O

3 6°"

0

h-

2 40

C

O

8

n. 20 _. -» —- -—————~e —7-
0 T 1rfYT- VT #77? VTITTTjIY‘YYVerthTTYYYTV' lYlY‘TYY if
Q N to O N 0 N CO V Q N N
22::e§§e§§§sss§aaa§§§§s§§§§

Day of Year

 

711; — St Joseph 00.. 0-30 cm Depth +Stdoseph 00.. 30-67 "DEB—Depmi
L-__:*~StJosephCo..60—90cmDepth . !

 

Figure 7c: Simulated volumetric soil moisture, as a percent of observed soil moisture by depth for
corn in St. Joseph County, 2003. Simulated values calculated using default model settings.

57

600

 

500 .-,_ ._ ‘. ___. a -77 -7- - . ._fl-- -7. .5-

300 .--

 

 

 

 

1oo -~ ~e-~—-———-—~ — -- —-

 

Percent of Observed Soil Moisture

 

 

 

{90

see eeeeeeeee 0.» 0.»
'3'3\‘94‘3’9@ffo‘fitrflffio'g’ohofifififoé’
'Dayonear

 

 

i- O — Saginaw 00., 0-30 Eni Depth +Saginaw Co.. 30-60 cm Depth

i--.___‘p__. :Saginaw (30., 60-90 cm Depth“ _

 

 

Figure 7d: Simulated volumetric soil moisture, as a percent of observed soil moisture by depth for
corn in Saginaw County, 2003. Simulated values calculated using default model settings.

58

rate of water application. In Michigan, some growers may use various forms of irrigation
scheduling models, while others simply schedule with the number of calendar days since
precipitation (Wright and Bergsrud (1991), grower personal communication). In an
attempt to determine any bias introduced fiom multiple managerial processes, mean
differences and mean absolute differences were calculated for each grower. All growers
except one reported less water use than estimated by the MWUR model, with the
differences over both seasons ranging from -41 .3 mm to 119.2 mm, in Table 9. With the
exception of grower 3, the differences were greater during the 2002 season. The
relatively higher rankings of the differences for growers 3 and 4 each year suggest some
consistency of management scheme. Standard deviation of seasonal irrigation depths
within individual grower’s fields ranged from 47.1 mm to 137.4 m (Table 10). Not all
growers had the same crop type or distribution, but this does suggest a large amount of
variability within an individual farm.

Taken collectively, the results of the direct comparisons indicate relatively poor
agreement between the MWUR estimates of soil moisture and subsequent irrigation totals
and those observed at the field level. This is somewhat surprising, given the satisfactory
results of Andresen et al. (2000) on the only previous attempt to validate the model for
this type of application a larger statewide scale. It is also surprising given a number of
successful previous field-level applications of the base soil moisture algorithm, which
serves as the base of the MWUR model (Ritchie 1985; Ritchie 1998; Andresen et al.
2002). Arguably the most important source of potential source of error is the National
Weather Service Stage III precipitation estimates used as input by the MWUR model

system (Moen (1999), pp. 138-148). While possibly related to some of the observed

59

.3538 .23... :33: 9...: 33.3.3

2.2.9.. scent... 33.556 .33 23 «can 2.838 .8. .6393 .3 .59.... 5.33:. 3.338 18.2.9. E... 33.2.... .3 2.6.3.5.. 13.53% "a 0.3...

 

 

 

 

 

 

 

 

 

 

. . . . . 5.332.
e .m. . 8 m t. m 8 . 3 page...” 8.3%..
. . . . . .5325
w E m .m m we . 8 m .8 Bunsen 52235
c v n N F .2520
28> =<

 

 

 

6.9.58

.25... ..=a..o.. ME... 33.5.3 2.2.0.. 193385 62.3.55 ".39. 95 o... E3 .33 .83 9:3.» .8 .53....» .3 33.3.3 633$... 3.3%.. ..o 33.2.
a mu 5%.. 3.33.... .2838 33.5.5 .33 .85. 2.2.2. 5.39.... cor—2.9. .Ea ..o3.:E_u 5932. 02.9.95: 93.83 ES... 33 .30.). x. 93:.

 

 

 

 

 

 

 

 

. . . . . notomom
e 8 . me 9.. me. om. 2. e 8 5. «8 e me. n m. 8. .e. .m. 8. em. 3 .52...
.EE.
. . . . . . . - . . . . 8.3.2.5
o 8 Ne. . 2 2. e 8 e um . m. a E 8. m N. B. we. . mm o m. e 8 m 3 328.2
cues.
.25.
m. .e. we. 3. 2. ”.8 3m 8.... 3. 8. ON. 8.- N... .3 0.2 m8 3... 35.2.5
:35.
e m _ e n N . e _ e n . . e _ m _ e m _ N _ . .235
was. .2 «8... «SN

 

 

 

60

differences between simulated and observed water content and use, the errors in
precipitation associated with the NEXRAD estimates for sites in Michigan during the
same time frame were only found to be on the order of 0.01 mm (Andresen and Aichele
2003), which even accumulated on a seasonal basis are far less than the observed field
level differences. It was concluded that these differences in irrigation depth must be due
to improper or inadequate parameterization of one or more variables associated with the
MWUR model, possibly related to spatial heterogeneity on the field level scale (Basso et

a1. 2001; Anderson et al. 2003).

M WUR Model Sensitivity Analysis

Some of the discrepancies between modeled and reported irrigation depths may
be simply explained as matter of timing. Figure 8a is an example of the simulated
seasonal irrigation events well timed with grower reported events. Figure 8b is an
example where simulated and reported irrigation depths are impacted by timing and
amount of individual irrigation events. These untimely irrigations and excessive depth
applied per event suggest the model is not characterizing the individual locations
perfectly.

A question rising from these seasonal water accumulations relates to the
reliability of reported seasonal irrigation depths to be the amount of water the crop needs.
By plotting potential drainage against the difference between modeled and reported
seasonal irrigation depths, an indication of water loss through the profile can be seen. The

potential drainage was calculated fi‘om a simple daily water balance for each field. Daily

61

 

350

 

300 a»--- i ,

§

 

>—— ___

 

 

200 ..

_.L
l

 

100 —. -

Irrigation Depth (mm)

 

50

 

 

 

 

 

o .

omeuoom‘b some: con, 0

(Pkg'xb'xq’é~$$$3w°i®q>rfifiriywhwh9$$rfl
Dayonear

 

 

L— 0; Simulated Irrigation Depth ——I— Reported Irrigation Depth 1

 

Figure 8a: Simulated and reported seasonal irrigation depth accumulations (mm), by day of year. A
carrot field in Mecosta County, 2002.

62

 

300

 

 

Irrigation Depth (mm)

 

 

 

 

 

Day of Year

 

 

. + — Modeled Irrigation-Depth _._A‘ Reported Irrigation Depth 9

 

Figure 8b: Simulated and reported seasonal irrigation depth accumulations (mm), by day of year. A
soybean field in St Joseph County, 2002.

63

model ET estimates are subtracted from reported irrigation and daily NEXRAD
precipitation estimates to calculate change in soil water storage and potential drainage.

This simple drainage water balance is determined for each field. Assumptions had
to be made about the drained upper and lower limits of the soil profile. Initial soil
moisture values were derived from gravimetric soil samples in 2003 and initial
capacitance probe values for 2002. The 2002 season does not contain a complete record
of soil moisture because of the late starting date for tube installation.

The seasonal drainage is plotted against the difference between modeled irrigation
amount and field reported irrigation. From Figure 9, a negative correlation is apparent;
when the growers apply more water than the model, the estimated drainage tends to be
greater. A stronger relationship can be seen in 2003 as opposed to 2002, which can be
found in the Appendix. The result suggests the growers putting on much more water than
the model may be over-irrigating their crops. It also suggests the model is a reasonable
estimator of plant water needs under strict decision assumptions.

After determining drainage, seasonal trends of irrigation depth for individual
fields in can best be seen through graphics in Figure 8a and Figure 8b. Seasonal water
accumulation trends show the model applying more water less fi'equently than any of the
growers, regardless of the total water applied. Also, the model tends to have a longer
irrigating season than do the growers. In nearly all of the crops, the model is irrigating
well after the grower ceases applying water. In many cases, the model is simulating
irrigation events earlier in the year and continuing later in the season than are the

growers.

64

200 «
150 i
E
E, 100 -«'
C
.9.
fl
3
“5 5° '
<1
0 .
o
-50
-100 --_..-, , ,

100

  

 

 

 

Drainage + A Soil Moisture (mm)

y - 0.7237): + 206.94
r’ . 0.8208

350

Figure 9: Difference between simulated and reported seasonal irrigation depth vs. total seasonal
drainage and change in soil moisture across all study sites, 2003.

65

Figure 10 (a, b, and c) plot the seasonal timeline of volumetric soil moisture for
both the model and observed values. The points in the observed plot are an average of the
three samples within the field for each sample day. Maximum and minimum values are
seen with the observed values as upper and lower error bars, respectively. This allows for
the visualization of the range of values measured within the field. In many cases,
especially Figure 10a and Figure 10b, the modeled soil and observed soil moistures show
similar trends and mimic one another for major increases and decreases. Figure 10c
shows the model having much greater soil moisture than observed, but closing the gap
later in the season. This large difference between simulated and observed soil moisture in
the lowest soil layer (60-90 cm) suggest the model is keeping the layer at or near field
capacity, while in reality it is not. Simulated roots may not be taking up as much water in
this layer as others either because they have sufficient water above 60 cm or the roots
may not develop as much into layers below 60 cm. The figures for other locations
indicate many of the same trends and can be found in the Appendix.

In an attempt to determine potential sources of error in the MWUR model system,
several model variables describing both physiological and managerial aspects of the
model were systematically adjusted in a series of simulations to assess its sensitivity.
After studying the seasonal irrigation depth trends (Figure 8 a and b) and high simulated
volumetric soil moisture in the lowest layer (Table 8), the variables include the depth of
water applied per irrigation event (mm), root growth rate (mm/ growing unit), irrigation
trigger level (9,), and planting date. When looking at seasonal irritation events and total
depths for the study sites, differences between simulated and reported values could

possibly be explained through timing (6),), depth per irrigation event, planting date,

66

 

0.25 -

 

s. , . .

50.15- V , ‘ ’ \ IQ /.
\ I -~ \ /

% ’ I ’ \ Q

2 ‘ I ’ I

= \‘ ‘ \

g 0.1-

0

'5

E

20.05j

O

>

0' ‘T r l r 1 r .— . ‘1‘r r—"‘ r

 

 

 

to Q: to ‘L Q Q) in Q ’\ '\
<\ e e as '19 q> v9 «1:5 «a e at?
Day of Year

 

 

 

r _
|
|

.- a, - 695% Mode'flsaiimsfle_+— __9-30 91039295'33 S9M9ist9E

 

 

Figure 10a: Seasonal trend of simulated and observed volumetric soil moisture (cm’lcm’) for the 0-30
cm soil profile layer in corn, Mecosta Co. 2003. Observed values are average of three samples per
field, with maximum and minimum values reported with upper and lower error bars, respectively.
Simulated values calculated using default model settings.

67

0.25

 

”s
.7." 0.2 —

E O-

3 l

2 /

,3 0.15 —

'5 \

a .‘i ‘
8 0.1

fi

0

E 0.05 ~

9.

§

0 "'_—“'"____'_"—_ T ‘ .' _ *_— T—"f- “ 7 ‘ ' ‘ 7 " T__ *fi‘w '7 ' ' ' "’"""’T_ —_ ‘—'_'——ki

to b '1, 9 b h 0 '\ '\
e <6" e re «a e «t» e «a e at?
Day of Year

 

l? E5595" M05915 5°"W _+___ _ T555 5'5 5¥9r9§55555°ist5r9§

 

Figure 10b: Seasonal trend of simulated and observed volumetric soil moisture (cm’lcm’) for the 30-
60 cm soil profile layer in corn, Mecosta Co. 2003. Observed values are average of three samples per
field, with maximum and minimum values reported with upper and lower error bars, respectively.
Simulated values calculated using default model settings.

68

0.25 ~

g Joooooo.o."
g 0.15 _. “i
'5 ‘C.
2 [ ‘0.
g 0.14}
0 w
'5
O
5 0.05 5
O
>
o . .* e‘ .+AA..-_ . -_- _-_fi_.-._... Y.“ - 7+ .. f V. 7*“ True—7.7.

9: ‘b ‘3 'L 9 Q) bu Q ’\ ’\
o ,3. «5 «.5 v.5 o rt» «£5 rs» qr rt?
Day of Year

 

:5 47599EM®§I§¢§1M§¢E+ 3053'? Qafiisafiowej
Figure 10c: Seasonal trend of simulated and observed volumetric soil moisture (cm’lcm’) for the 60-
90 cm soil profile layer in corn, Mecosta Co. 2003. Observed values are average of three samples per
field, with maximum and minimum values reported with upper and lower error bars, respectively.
Simulated values calculated using default model settings.

69

and/or root development (affecting depth and amount of root water uptake). Root growth
rates were altered in an attempt to explain the large differences between simulated and
observed soil moisture in the lowest layer (60-90 cm) of the study profile. During the
sensitivity analysis runs, all other variables in the model were held constant at default
values and the locations were the same as those sampled in the field portion of the study
during 2002 and 2003. All default and adjusted model parameters may be found in the
Appendix.

The triggering mechanism, (~96, is the fraction of plant available water (PAW) in
the rooting zone at which water is added by irrigation. By default, the value is set at 0.50
of PAW; so when the plant available water in the root zone drops below 0.50, the model
initiates an irrigation event. When the trigger levels are lowered, a reduction of seasonal
irrigation volume is expected, as more plant available water is consumed before the next '
irrigation is required. The opposite response would be expected as the trigger is
increased. Four different trigger levels surrounding the default level of 0.50 were used in
the sensitivity analysis for each of the crops and locations of the reported fields: 0.40,
0.45, 0.55, and 0.60 of PAW.

Differences between the adjusted and default values of the irrigation trigger for all
crops over both seasons are given in Table 11. The changes in trigger level did have the
expected results for seasonal irrigation volume for each crop, with smaller irrigation
totals for lesser irrigation trigger levels and vice versa. By crop, the greatest range of
mean differences was found for potato, which also has the shallowest rooting depth The
lowest trigger level, 0.40 of PAW, resulted in a mean reduction of 95 mm of irrigation

depth to a high of an increase of 150 mm of water with the 0.60 of PAW trigger level.

70

6.53.2: =8 min—.9... Eu... .8 Ram a. ”5:3 «.389 6.53.0... :8 0.32.“: :8... he «:38.— o... :2...
.82... m. 5...: .85. .83.... 8.83:. a... S v.8... unison .28.: E 89:30 62.353 m2." 1.... 83 9:3» 8.. no.5 .3 A3538... :8 8.52.3... =8...
.8 exam. ”5:2 .28.: .382. v.8 maize. .28... .88.... .8 32.0.. 5.83.... .2883 BEBE? .8352. 8258...... 328.... E3... E... .88.). "a 2...“...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

m.oo. Woo. QB 95 9m. 0.2 0.8. odm. 99. 99. 5 $8

.3... Q: msm msn Qmm 9mm ad. ad. a. .N m. .N we $3

Nov Nov. m...” msm- 0.8 0.8. are... ms..- gm fiww- "6 $3

2K hon- mdm mNm- odo. odoT N .m n. _.m- mdm «.8. 5 $3.

Es. Es. Ea. Es. .5... is. .5... is. Es. is. 3.
3.8.3.5 3.88:5 8088...... 3.8850 3.88:5

3.88:5 3.8850 3.8850 3.8850 388:5 88......

33.3.3 :85. 23.82 cums. 8.3.32 585. 23.8.3 .88.... 23.3.2 :35. .8. «9:

.822 .885. cues. cues. .88.... .. . .

2.9.0 ..< .980 coon>ow 0.80.. :80

 

 

71

For com, the differences ranged from —56 mm to 44 mm. Overall for both seasons for all
crops, the increase in the difference with increasing trigger level was 66.3 mm of
irrigation water depth While the direction of the changes in irrigation with different
triggers was consistent across crops and seasons, the magnitude of the differences varied.

A second important input variable related to grower management strategy is the
amount of water applied per irrigation event. The model default value of this variable is
25 mm and is considered constant throughout the season. In the sensitivity analysis, this
variable was altered in seasonally constant 6 mm increments from a minimum of 12 mm
per irrigation to a maximum of 38 mm per irrigation (values of 12, 19, 32, and 38 mm).
Differences in seasonal irrigation totals between these altered and default values are
given in Table 12. Potatoes were, by far, the most sensitive crop to changes in this
variable, with a maximum increase in seasonal irrigation volume of 127 mm for the 38
mm application level during the combined seasons and a reduction of 80 mm for the 12
mm application. In contrast, the differences for corn ranged from a deficit of 8.25 mm (at
12 mm application) to an excess of 7.5 mm (at 38 mm application). Similar patterns were
observed for the other crops. The ranges of differences were greater for soybeans, but less
than that of potatoes.

The two other parameters changed were root growth rate and planting date. Root
growth rate alterations did have an effect on soil moisture values (Table 13), especially in
deeper layers. The seasonal irrigation amounts increased with a reduction in growth rate
and a decrease in water with increases in growth rate (Table 13). The new water volumes
do not show as great a range as with amount per irrigation or 6),. Planting date changes

resulted in no sensible changes for either soil moisture or seasonal irrigation volume.

72

4.3.3 35%....
3.— 3... ma .3 :33: 23>» .3.— 53v 353...... 2: 8 23.: 35:8 3.3.: a. 39:25 62.553 gen E... «can 9...?» 3.— .....o .3 6.33 gamut... 3.—

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

  

 

 

    

 

 

 

c239...

:33. «5:8 .28.: 2.52. E... 35:8 .25... 3.8... .8 2.32. 3.33.... 2.8.3“ 3.22:: :3an 89:28.5. 258.... :3... E... :32 .3 2..."...

m. 5 can Qom 0.8 m. w: w. 3 9mm. mdmr 0.3 w.

New adv Qmm odw m. V: m. V” 2.. c m. w «:3 we

Ewm 0mm- m9. mar- o. w: 2w- 2 a man. Va 3

0.8 Nam? o. 3 n. :1 ms: 3h- n. . Noe- mdF Mum-

ASE. ASE. AEEV REE“ AEEV AEEV A .5 REE. ~EE. AEEV

3.33:5 3.33:5 3.33:5 3.33:5
33355 8282 3.3355 8232 3.3355 30.33.55 2232 333:5 23.0.0.2 3.3355
.322 :35. .322 c a 3.2 :35. :35. :35. :32 :32 :35.
. , . . ; ,3 ,MW.:kunaam..fl .H...h.m,fl _ . m 380m 2. . .. . EGO”.

 

  

 

 

 

 

 

 

  

 

 

 

 

73

2.5 .33 3 "_._—32.3 .953... 3a.. 92:. .3233“. 3 8...: 2.33 «.33..»
35.9 3.... .32... 3....» .2... 2333—23.. 3.... 2.. 8 32.. 35:3 .32.. 5 39.2.0 62.538 33 .2... Son “3% .3.— .33 .3 A33 .3 ”=33...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

9:38 .23... 53.3.. .2... 3.5.8 .03... .33. a 3 2.2.2. 32.3.3. .2338 33.2.... .3232. 8233...... 35.8.... .32.. .2... :32 n: 2...;
23:.
.20 03
men .2- o...“ 9mm- of. 9mm- .3“ m2- .8 m. a- 2.3.. a
23+
.33
-no 33
3. .3. 3.. 98. m...” 3... I: 3- mm. o... 2.6.... a
22+
.coE
.n 33
m2 3. m2 3. o... o... 3. N... v... 2. who”. a
$8.
.3...
.20 33
EN 3m m... m...” 0.0 3. 08 3.8 3. I. Lox o
«.3.
AEE. A .33. .23.. .33. ASE.
SE. .33. .33. .23... .EE. .3...
3.33:5 3.33:5 333:5 3.33:5 333:5 3.33:5 3.33:5 333:5 3.33:5 333:5 A3335
33.32 :35. 23.39.. .335. 83.32 :35. 83.9.23 .33.... 83.093 .322 30¢
:35. :35. 53.2 .33.... .322
«33 .2 350 53.6.... 839. 58

 

 

74

.mm. >OQ .3 333.. 8.... 9....3... a... 3 0.3... 89.2.8 .23... E 39.2.0 ...o........3 8.." E... «can .39. 3.. .....3 .3 Ann. 39. 3 3.. .3 8.... 95.3....
9:38 .23... 333.. ...3 895.8 3.3... .33.... 3 85...... 3.39.... .2338 33.2.... .3952. 8233...... 03.8.... .32.. ...3 .332 "N. 2.3.!

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I: 03. 9mm 93. Qmm 0.0 w. l 0.3- 93. mv . >05
3”. :. .- 9mm 9mm- md. 99. e. 1 m... Ym- o3 >On_
v. C :. .. o.o 0.0 9mm 9mm- ... t ed _..m on . >05
mm. 3...- o.mm Qmm. Qmu 9mm- ... w. m..- 06 mm . >05
.58. .5... .EE. . 3..
.EE. 3.33:5 .EE. 3.33:5 .EE. .EE. 3.33:5 .EE. 3.33:5 .EE. 9.3.3...
3.33:5 3.33:5 3.33:5 3.33:5 3.33:5 3.33:5
.322 2232 .322 223.... .322 .322 warm“... .322 o......v....m2..< .322
.322 .322
,. £20.31... a. _ . . 3:8 .U......§8.om , . . . 63¢... , Eco.

 

75

 

Mean absolute differences were of the same magnitude for all four tests performed and
suggest little individual difference within each test. These results suggest the model is
most sensitive to @c and irrigation depth per event, of the four parameters tested. Both of
these variables deal directly with water availability and are directly related to decisions

growers must make on their own farms.

M WUR Performance with Changes to Managerial Variables

From the preceding discussion, there is ample evidence to suggest that a major
limitation of the MWUR system is the lack of representative input information relating to
a few key managerial variables. Given the data taken from the field observations and the
previous model sensitivity results, it is possible to adjust the input variables and rerun the
model with the hopes of improving its performance. While insufficient data were
available for both model redevelopment and test validation, this procedure may still
provide an estimate of the potential performance.

In the earlier sensitivity analysis, four physiological and managerial model
parameters were examined for mode output sensitivity. The most sensitive variables
found were the trigger level, 9c, and the depth of water per irrigation. These variables
were then altered in combination to minimize the mean differences between simulated
and reported seasonal irrigation water depths. Mean and mean absolute differences were
calculated in the same manner as in previous sections of the study. Regardless of the
drainage results, the assumption is made that the growers apply an appropriate amount of
water for their crops and what they are applying is the “reality of the real world”. Amount

per irrigation, @c, and season length for potatoes (personal communication) were altered

76

to reduce the mean difference between modeled and reported seasonal irrigation depths
for individual crops to a minimum. An irrigation depth of 19 mm was used because this
value is approximately the average depth per irrigation growers across the study use.
Simulation runs were conducted with varying values of (9c, until the smallest mean
differences in seasonal irrigation depth were achieved. The default season length of
potatoes (124 days) was noticed to be long for the varieties grown in this study. Typical
season lengths range from 90-100 days, according to Chris Long, potato specialist in the
Department of Crop and Soil Science at Michigan State University (personal
communication) but most of the growers in this study had varieties with season lengths
on the order of 100 days (grower personal communication). Also, The Ohio State
University Extension bulletin number 672-03 shows season lengths for some of the
varieties grown in Michigan to range from 90-115 days (Smith 2003). Therefore, an
arbitrary season length of 101 days was selected.

Corn mean differences could be reduced to —1.02 mm, with a mean absolute
difference of 49.9 mm for years 2002 and 2003 combined, see Table 15. While the mean
difference was greatly reduced, mean absolute difference remained large and suggests
overall improvement of model performance was the result of large over and under
predictions at different locations or years. In this case the year 2002, saw too much water
and 2003 saw too little. The trigger level was reduced to a value of 40 percent of plant
available water, which is within the range suggested by Rhodes and Bennett (1990) for
simulating corn development.

Potato mean differences were reduced to ~2.75 mm for the two years combined

with a mean absolute difference of 62.6 mm. Again, mean absolute difference was large,

77

 

.332. 55%.... .2838 .o 82.95.... :3... 0.5.9:. 8. unit... .28... 3.3... ”in: 3.23.: 2.2.0.. 3.2.5.5 62.388 Ea»... PE 2.. E... .33
.33 9.3.. .8 .33...» .3 33.3.3 ABS. 5%.. grant... .2338 totem... _._... ..8a...:..m .3952. 853...... 828.... 5.2.. _._... 56$. .3 2......

. . . . . . . . . . . . . . . . tau—homo".

O: _.Nm Vrow mmmr voww mam own man o9; vow m we _.Nm mo: mmow voww mam won—.0930.
.EE.

. . . . . . . . . . . . . . . . 3:28....

o MNF N NVN o 9. N Va 0 av m mm N X: 0 mm wow 5 3.. w _.mv n NVN o mN v w 0 av V m 320mg

cues.

.EE.

mdw- odmm 0.». r. N...“ me NNN- me- mdN- _._oo N. TV. goow o.wNN Qmm Wm mfim QM- 00:93:3—

:35.
m _ m _ v _ n _ N _ F o _ v _ n H F c _ m _ v _ n _ N _ _. .0395

28> =< , SON N8N

 

 

 

 

 

 

 

 

 

 

.23.... cougar... .2598 _.e «353...... :3... 9555. 3. $538 .23... 19.2... 9...: 33.3.3 2.2.9.. 33.555 ...o........3 9.3.. 95 o... .2...
.83 .38 23.. .5. :98 .3 33.3.3 ABE. 5%.. 5.33.... .2338 93.2.9. E... .533...“ .5282. 3:95.... 838.... :3... E... 53.). “n. 9.9....

 

 

 

 

 

 

 

 

 

 

 

 

 

mo
v.8. Q8 mdm EX: ado. QB. QR o.mo_. Nam mdm mNm Yo: “Ammo.“
on 0.8 me. o... 2. m8 v.2 to N: N. 3N 0.8 .mnwmuuduwfi
. . . . . . . . . . . ~ v
QN m N- o F- o o S N t m 8- v o- N t- N N E: v NN 855%.”. :35.
:aogom 820.". _ Eco .ommwn. _ :wonmom 250m _ Eco caonxmw _ .250 _ 33.... :m 82°... . Eco meo
”W , 33> =< .. noon ,. . _. «can

 

 

78

but in this instance 2002 and 2003 saw approximately the same mean difference, 18.1 and
17.2 mm, respectively. Therefore, the overall mean difference was the result of over and
under predictions, rather than across the board improvement in model performance. The
season length was changed from 124 days to 101 days, 9c increased to 0.55 of plant
available water, and irrigation amount decreased to 19 mm. Soybeans decreased their two
season mean difference to 2.85 mm and mean absolute difference to 3.25 mm. The
irrigation trigger and amount per irrigation were decreased to 38 percent plant available
water and 19 mm, respectively. These values are similar to those reported by Wright and
Stark (1990) for simulation of potato development.

The improved seasonal irrigation depth results from the altered management
variables showed a much improved scatter plot of modeled verses observed seasonal
irrigation volume (Figure l 1). The r2 values are increased from 0.020 in for default model
runs to 0.366 in for best-fit model runs. Also, the slope of the trend line is closer to one
than that of the original default trend line slope. The overall root mean square error for all
crops in both seasons improved fi'om 97.8 mm for default to 64.4 mm for altered model
runs (Table 17).

While improvements were made to the mean differences of seasonal irrigation
water depth, the same cannot be said for simulated volumetric soil moisture. The largest
differences, by crop, between the adjusted and default model runs were for potato, with a
mean difference of -0.033 cm3/cm3 Table 18) in the 0-30 cm layer. Yet the 30-60 cm
layer was wetter by a mean difference of 0.066 cm3/cm3. Corn was drier through the
entire profile, with the largest change in the second layer. The mean differences show the

altered simulations result in drier soil moisture than default simulations in the top two

79

350

 

300 --
250

200 .
150

100 i .

Simulated Seasonal Irrigation Depth (mm)

 

50

  

y - 0.4402: + 105.34
r' - 0.3662

   

 

 

50 1 00

1 50 200 250

300

350 400

Reported Seasonal Irrigation Depth (mm)

450

500

Figure 11: Simulated vs. reported seasonal irrigation depth (m) for all study fields in 2002 and
2003. Simulated values calculated using altered model settings (to improve mean differences of

seasonal irrigation depth).

80

 

 

 

 

 

 

 

Root Mean Square Error
Default Irrigation Altered Settings
Depth (mm) Irrigation Depth (mm)
Corn 88.6 64.2
Potato 99.6 80.1
Soybean 131.9 4.2
All Crops 97.8 64.4

 

 

 

 

Table 17: Mean square error for default and altered simulated seasonal irrigation depth for corn,
potato, soybean, and all crops for 2002 and 2003.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Mean Difference Mean Absolute
(cm’lcm’) Difference (cm’lcm’)

0-30 cm Soil Profile Depth

Corn -0.013 0.015
Potato -0.033 0.044
Soybean -0.017 0.017
Sularbeet -0.012 0.016
Carrot 0.006 0.018
Pepper -0.005 0.012
30-60 cm Soil Profile Depth

Corn -0.019 0.027
Potato 0.016 0.034
Soybean -0.016 0.016
Sugarbeet -0.01 1 0.01 1
Carrot -0.007 0.008
Pepper 0010 0.010
60-90 cm Soil Profile Depth
Corn -0.005 0.005
Potato 0.099 0.101
Soybean 0.182 0.182
flarbeet -- --
Carrot -0.001 0.001
Pepper -0.001 0.001
All Layers

Corn -0.012 0.016
Potato 0.027 0.060
Soybean -0.011 0.011
Sugarbeet -0.012 0.014
Carrot 0.000 0.009
Pepper -0.005 0.008

 

 

Table 18: Mean and mean absolute differences between default and altered simulated volumetric soil
moisture (cm’lcm’), by profile depth and crop for the years 2002 and 2003 combined.

81

layers. These results are reasonable; with less irrigation water applied, more of the
moisture must come from the soil profile. The unexpected caveat in Table 18 is the wetter
soil conditions for the second layer in potatoes. By increasing @c, the plant was able to
use more water from the higher layer.

The alterations actually resulted in slightly worse mean and mean absolute
differences between the simulated soil moisture after alterations and observed values
(Table 19). Yet, improvement between simulated and observed soil moisture was seen in
the 30-60 cm layer. The lowest layer studied showed little difference between soil
moisture fi'om altered and default model runs. Mean differences are very similar for all
crops and layers in Table 8 and Table 19. This is consistent with the results in Table 18
and would generally indicate the alterations made to model settings result in more water

extraction from the top two layers.

82

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Mean Difference Mean Absolute
(cm’lcm’) Difference (cm’lcm’)

0-30 cm Soil Profile Depth

Corn -0.017 0.050
Potato -0.003 0.031
Soybean -0.057 0.060
Carrot -0.017 0.028
Sugarbeet 0.109 0.109
Pepper 0.053 0.053
All Crops -0.004 0.046
30-60 cm Soil Profile Depth

Corn -0.001 0.065
Potato 0.051 0.056
Soybean -0.004 0.047
Carrot 0.044 0.044
Sugarbeet 0.108 0.108
Pepper 0.139 0.139
All Crops 0.037 l 0.067
60-90 cm Soil Profile Depth

Corn 0.039 0.051

Potato 0.041 0.043

Soybean 0.053 0.053

Carrot 0.056 0.056
Suga_rbeet -- ---

Pepper 0.122 0.122

All Crops 0.049 0.054

All Layers

Corn 0.007 0.055
Potato 0.029 0.043
Soybean -0.003 0.053
Carrot 0.028 0.043
Sugarbeet 0.108 0.108
Pepper 0.105 0.105
All Crops 0.027 0.056

 

 

 

 

Table 19: Mean and mean absolute differences between simulated and observed volumetric soil
moisture (cm’lcm’), by soil profile depth and crop for 2002 and 2003 combined. Simulated values
calculated using altered model settings (to improve mean differences of seasonal irrigation depth).

83

Conclusions

The first main objective of this study was to test the MWUR model output against

field measured and reported values of field level volumetric soil moisture and seasonal

irrigation depth. Secondary objectives in this portion of the study were to develop a

sampling scheme to adequately describe the simulation.

The MWUR model did not perform as well as expected, after initial tests by
Moen (1999) and Andresen (2003), in the estimation of seasonal irrigation water
use. Simulated water use for corn was 35 percent greater than reported, 24
percent greater for potatoes, and 97 percent greater for soybean.

The simulation adequately estimated volumetric soil water in the top 30 cm of the
soil profile under most cropping types. The difference between simulated and
observed soil moisture for all crops was approximately a positive 7 percent of the
typical plant available water for the model. Yet the simulation greatly
overestimated volumetric soil moisture in the subsequent two layers, resulting in
overestimations in the range of 48 percent of typical plant available moisture in
the second layer to 56 percent of typical plant available water in the third layer.
Tests to determine any bias introduced by varying grower scheduling practices
indicate the model adequately estimates water needs for crops under ideal

conditions, but cannot account for the variation in grower scheduling methods.

The second main objective of this study was to test model sensitivity to multiple

managerial and physiological parameters. The four variables chosen, base upon the

validation portion of the study and a review of the literature, were depth of water per

irrigation, soil trigger level (9c), root growth rate, and planting date.

84

o The simulation was most sensitive to the depth of water per irrigation and @c.
Mean differences fiom default ranged between a decrease of 76 mm to an
increase of 100 mm of seasonal in'igation depth for 6% changes. Mean differences
from default ranged between 49 mm less to 78 mm more seasonal irrigation
depth for changes to depth per irrigation.

0 These sensitive parameters were then altered to optimize model seasonal
irrigation water depth to reported values by attaining the smallest mean
difference possible for various crops. This exercise resulted in the model
accounting for 99 percent of the reported depth for com, 99 percent for potatoes,
and 102 percent for soybean.

0 These changes, most times, resulted in lower volume and more frequent irrigation
events. They also indicate, again, the model has problems strategizing irrigations
the same as the growers in this study and suggest potential problems with
initialization of crop and management parameters.

The simulation has shown that it is possible to properly estimate seasonal
irrigation water depth for a variety of crops across a large region. Work should continue
to properly initialize model parameters to the strategies of the majority of growers in
Michigan. Also, more work should be done to determine causes of positive simulation

biases of volumetric soil moisture at the lower layers of the model soil profile.

85

Appendix

86

.2733 558 5 v8: 395 3.. 82.; .23.: 2.3%: "an «Bah

£90 .2 :80 .o £90

£90 ._2 E00 .0 £90

 

87

or

 

393 n x
8%..» + x385. n >

 

9.30.02 :8 < .7 000520

 

mums—20 .m> 5.39:. <

uogeflpu V

88

References

Allen, R. G., L. S. Pereira, D. Raes and M. Smith (1998). FAO Irrigation and Drainage
Paper No. 56. Crop Evapotranspiration. Rome, Italy, FAQ, Water Resources,
Development and Management Service: 300.

Anderson, M. C., W. P. Kustas and J. M. Norman (2003). "Upscaling and downscaling -
A regional view of the soil-plant-atmosphere continuum." Agronomy Journgl 95(6):
1408-1423.

Andresen, J. A. and T. M. Aichele (2003). Simulation of plant disease risl_( on a regional
basis. Proc. American Society of Agronomy Annual Meeting, Denver, CO, American
Society of Agronomy. Madison, WI.

Andresen, J. A., T. N. Moen, J. T. Ritchie and R. van Til (2002). Estimated water use for
agricultural irrigation in Michigan: 1997-1999. Lansing, MI, Michigan Department of
Environmental Quality, Surface Water Quality Division.

Basso, B., J. T. Ritchie, F. J. Pierce, R. P. Braga and J. W. Jones (2001). "Spatial
validation of crop models for precision agriculture." Agricultural Systems 68: 97-112.

Batchelor, W. D., B. Basso and J. O. Paz (2002). "Examples of strategies to analyze
spatial and temporal yield variability using crop models." Eurqpean Journgl of Agronomy
18: 141-158.

Bechini, L., S. Bocchi and T. Maggiore (2003). "Spatial interpolation of soil physical
properties for irrigation planning. A simulation study in northern Italy." Eurgoean Joumgl
ongronomy 19(1): 1-14.

Bland, W. L. (1999). "Toward integrated assessment in agriculture." Agg'cultural
Sfitems 60: 157-167.

Borga, M. (2002). "Accuracy of radar rainfall estimates for streamflow simulation."
Journ_al of Hyrdolggy 267: 26-39.

Burt, C. M., A. J. Clemmens, T. S. Strelkoff, K. H. Solomon, R. D. Bliesner, L. A. Hardy
and T. A. Howell (1997). "Irrigation performance measures: Efficiency and uniformity."
Journgl of Irrigation and Drainage Engineering 123(6): 423—442.

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." Journal of Hydrolo gy 253: 169-193.

Chanzy, A., J. Chadoeuf, J. C. Gaudu, D. Mohrath, G. Richard and L. Bruckler (1998).
"Soil moisture monitoring at the field scale using automatic capacitance probes."
European Journ_al of Soil Science 49: 637-648.

89

Council of Great Lakes Governors (COGLG) (1985). The Great Lakes Charter.
Principles for the Management of Great Lakes Water Resources. Chicago, 111. Council of
Great Lakes Governors: 20.

Council of Great Lakes Governors (COGLG) (2001). The Great Lakes Charter Annex. A
Supplementary Agreement to the Great Lakes Charter. Chicago, 111. Council of Great
Lakes Governors: 12.

de Rosny, G., A. Chanzy, M. Parde, J .-C. Gaudu, J .-P. Frangi and J .-P. Laurent (2001).
"Numerical modeling of a capacitance probe response." Soil Science Society of America
Journal 65(1): 13-18.

Ejieji, C. J. and J. W. Gowing (2000). "A dynamic model for responsive scheduling of
potato irrigation based on simulated water use and yield." Journal of Agricultural Scierga
135(2): 161-171.

Engel, T., G. Hoogenboom, J. W. Jones and P. W. Wilkens (1997). "AEGIS/WIN: A
computer program for the applicaiton of crop simulation models across geographic
areas." gronomy Journal 89: 919-928.

Fanning, J. L., G. E. Schwarz and W. C. Lewis (2001). A field and statistical modeling
study to estimate irrigation water use at benchmark farms study sites in southwestern
Georgia, 1995-96. Atlanta, GA, United State Geological Survey: 32.

Fulton, R. A., J. P. Breidenbach, D.-J. Seo and D. A. Miller (1998). "The WSR-88D
rainfall algorithm." Either Ed Forecasting 13(2): 377-395.

Gaze, S. R., M. A. Stalham and E. J. Allen (2002). "Accuracy of the neutron probe for
measuring changes in soil water storage under potatoes." Journal of Agricultural Science
138: 135-152.

George, B. A., S. A. Shende and N. S. Raghuwanshi (2000). "Development and testing of
an irrigation scheduling model." Agriculth Water Management 46(2): 121-136.

Hansen, J. W. and J. W. Jones (2000). "Scaling-up crop models for climate variability
applications." Agricultural Systems 65: 43-72.

Hatfield, J. L. (1990). Methods for estimating evapotranspiration. Irrigation of
agricultural crops. B. A. Stewart and D. R. Nielsen. Madison, WI, American Society of
America Inc., Crop Science Society of America Inc., Soil Science Society of America
Inc. 30: 436-474.

He, C. S. (1999). "Assessing regional crop irrigation requirements and streamflow
availability for irrigation development in Saginaw Bay, Michigan." GEOGRAPHICAL
ANALYSIS 31(2): 169-186.

90

Heinemann, A. B., G. Hoogenboom and R T. de Faria (2002). "Determination of spatial
water requirements at county and regional levels using crop models and GIS. An example
for the State of Parana, Brazil." Agricultural Water Management 52: 177-196.

Howell, T. A. (2001). "Enhancing water use efficiency in irrigated agriculture."
Agronomy Journal 93: 281-289.

Ines, A. V. M., A. D. Gupta and R. Loof (2002). "Application of GIS and crop growth
models in estimating water productivity." Agricultural Water Management 54: 205-225.

ISM, I. S. M. I. (1996). PRISM User's Manual. Malaga, WA: 61.

Jackson, S. H. (2003). "Comparison of calculated and measured volumetric water content
at four field sites." Agricultural Water Management 58: 209-222.

Jensen, M. E., R. D. Burman and R. G. Allen (1990). Evapotranspiration and Irrigation
filter Requirements. New York, American Society of Civil Engineers.

Jensen, M. E. and H. HR. (1963). "Estimating evapotranspiration fi‘om solar radiation."
Journflf Irrigatim and Drainage Engineering, ASCE 89: 15-41.

Jensen, M. E., D. C. N. Robb and C. E. Franzoy (1970). "Scheduling irrigations using
climate-crop-soil data." Journal of the Irrigation and Drainage Division. Proceedings of
the American Society of Civil Engineers 96(IR1): 25-38.

Jensen, M. E., J. L. Wright and B. J. Pratt (1971). "Estimating soil moisture depletion
fiom climate, crop, and soil data." Trans_action_s of the ASAE 14(5): 954-959.

Khosla, R. and N. Persaud (1997). "Performance of a Non-nuclear Resonant Frequency
Capacitance probe. I. Calibration and Field Testing." Communicatioas in Soil Science
and Plant Analysis 28(15&l6): 1333-1345.

Khosla, R. and N. Persaud (1997). "Performance of a non-nuclear resonant frequency
capacitance probe. 11. Measurement of dryland crop water use." Communications in Soil
Science and Plant Analysis 28(15&l6): 1347-1357.

Knox, J. W., E. K. Weatherhead and R. 1. Bradley (1996). "Mapping the spatial
distribution of volumetric irrigation water requirements for maincrOp potatoes in England
and Wales." Agricultural Water Management 31: 1-15.

Knox, J. W., E. K. Weatherhead and R. 1. Bradley (1997). "Mapping the total volumetric

irrigation water requirement in England and Wales." Aggicultural Water Maflgement 33:
1 -1 8.

91

Koren, V. I., B. D. Finnerty, J. C. Schaake, M. B. Smith, D.-J. Seo and Q.-Y. Duan
(1999). "Scale dependencies of hydrologic models to spatial variability of precipitation."
Journal of Hydrology 217: 285-302.

Krajewski, W. F. and J. A. Smith (2002). "Radar hydrology: rainfall estimation."
Adva_n_ces in Water Resources 25: 1387-1394.

Lane, P. N. J. and D. H. Mackenzie (2001). "Field and laboratory calibration and test of
TDR and capacitance techniques for indriect measurement of soil water content."
Australian Journal of Soil Resear_ch 39: 1371-1386.

Mahmood, R. and K. G. Hubbard (2003). "Simulating sensitivity of soil moisture and

evapotranspiration under heterogeneous soils and land uses." Journal of Hydrology 280:
72-90.

Mandal, U. K, K. S. S. Sarma, U. S. Victor and N. H. Rao (2002). "Profile water balance
model under irrigated and rainfed systems." figonomy Journal 94: 1204-1211.

Maracchi, G., V. Perarnaud and A. D. Kleschenko (2000). "Applications of geographical
information systems and remote sensing in agrometeorology." Agricultural and Forest
Meteorology 103: 119-136.

Martin de Santa Olalla, F., A. Calera and A. Dominguez (2003). "Monitoring irrigation
water use by combining Irrigation Advisory Service, and remotely sensed data with a
geographic information system." AgriculturaliVater Management 61: 111-124.

MASS, M. A. S. S. (1998). Michigan Agricultural Statistics 1997-98. Lansing, MI,
Michigan Department of Agriculture: 147.

McKinney, D. C. and X. Cai (2002). "Linking GIS and water resources management
models: an object-oriented method." Environmental Modeling & Sofiwafi 17: 413-425.

MDA, M. D. o. A. (2003). Generally Accepted Agricultural and Management Practices
for Irrigation Water Use. Lansing, MI, Michigan Department of Agriculture: 21.

Moen, T. N. (1999). Assessment of a modeling approach for the estimation of
agrucultural irrigation water use in Michigan. Department of Resource Development.
East Lansing, Michigan State University: 189.

Monteith, J. L. (1981). "Evaporation and surface temperature." Quarterly Joumal of the
Royal Meteorological Society 107: 1-27.

NASS, N. A. S. S. (1999). 1997 Census of Agriculture Farm and Ranch Irrigation
Survey. Washington, DC, United State Department of Agriculture: 148.

92

Ogden, F. L., J. Garbrecht, P. A. DeBarry and L. E. Johnson (2001). "FID and
Distributed Watershed Models. 11: Modules, Interfaces, and Models." Journal of
Hydrol‘ogic Eng'meering 6(6): 515-523.

Ould Mohamed, S., P. Bertuzzi, A. Bruaud, L. Raison and L. Bruckler (1997). "Field
evaluation and error analysis of soil water content measurement using the capacitance
probe method." Soil Science Society of America Journal 61(2): 399-408.

Panigrahi, B. and S. N. Panda (2003). "Field test of a soil water balance simulation
model. " Agricultural Water Management 58(3): 223-240.

Paz, J. O., W. D. Batchelor, G. L. Tylka and R. G. Hartzler (2001). "A modeling
approach to quantify the effects of spatial soybean yield limitig factors." Transactions of
the ASAE 44(5): 1329-1334.

Prajamwong, S., G. P. Merkley and R. G. Allen (1997). "Decision support model for

irrigation water management." J oumal of Irrigation and Drainage Engeering 123(2):
1 06- 1 1 3.

Rhodes, F. M. and J. M. Bennett (1990). Corn. Irrigation of agricultural cropa. B. A.
Stewart and D. R. Nielsen. Madison, WI, American Society of Agronomy inc., Crop
Science of America Inc., Soil Science Society of America Inc. 30: 569-596.

Richardson, C. W. and J. T. Ritchie (1973). "Soil water balance for small watersheds."
Transactioas of the ASAE 16: 72-77.

 

Ritchie, J. T. (1972). "Model for predicting evaporation from a row crop with incomplete
cover." Water Resources Resea_rc_h 8(5): 1204-1213.

 

Ritchie, J. T. (1985). A user-oriented model of the soil water balance in wheat. Wheat
growth and modelling. Series A: Life Sciences Vol. 86. W. a. A. Day, R.K New York,
Plenum Press: 293-305.

Ritchie, J. T. (1998). Soil water balance and plant stress. Understanding options for
agricultural production. G. S. Tsuji, Hoogenboonr, G., and Thornton, P.K. Boston,
Kluwer Academic Publishers: 41-54.

Robinson, D. A., C. M. Gardner, J. Evans, J. D. Coper, M. G. Hodnett and J. P. Bell
(1998). "The dielectric calibration of cpacitance probes for soil hydrology using an

oscillation frequency response model." Hydrology and Earth Systems Sciences 2(1): 111-
120.

Rosenberg, N. A., B. L. Blad and S. B. Verma (1983). Microclimate: The biological
environment. New York, Wiley and Sons.

93

Rowshon, M. K, C. Y. Kwok and T. S. Lee (2003). "GIS-based scheduling and
monitoring irrigation delivery for rice irrigation system Part H. Monitoring." Agg'cultural
fitter Management 62: 117-126.

Rowshon, M. K, C. Y. Kwok and T. S. Lee (2003). "GIS-based scheduling and
monitoring of irrigation delivery for rice irrigation system Part 1. Scheduling."
Agricultural Wafi' Management 62: 105-116.

Sharif, H. O., F. L. Ogden, W. F. Krajewski and M. Xue (2002). "Numerical simulations
of radar rainfall error propagation." Water Resources Reseaflla 38(8): 15-1-15-14.

Smith, K. L. (2003). 2003 Ohio vegetable production guide, bulletin 672-03, The Ohio
State University Extension.

Sousa, V. and L. S. Pereira (1999). "Regional analysis of irrigation water requirements
using kriging: Application to potato crop (Solanum tuberosum L.) at Tras-os-Montes."
Agricultgal Water Management 40: 221-233.

Starks, P. J ., G. C. Heathman, L. R. Ahuja and L. Ma (2003). "Use of limited soil
property data and modeling to estimate root zone soil water content." Journal of
Hydrology 272: 131-147.

Steiner, M., J. A. Smith, S. J. Burges, C. V. Alonso and R. W. Darden (1999). "Effect of
bias adjustment and rain gauge data quality control on radar rainfall estimation." Water
Resources Reseaflh 35(8): 2487-2503.

Stewart, J. B. and W. R. Rouse (1977). "Substantiation of the Priestly and Taylor
parameter a = 1.26 for petential evapotranspiration in high latitudes." Journal of Applied
Meteorology 16: 649-650.

Survey, N. C. S. (1960). Soil Survey of Montcalm County Michigan. Washington, DC,
United States Department of Agriculture: 102.

Survey, N. C. S. (1983). Soil Survey of Saint Joseph County Michigan. Washington,
DC, United States Department of Agriculture: 177.

Survey, N. C. S. (1984). Soil Survey of Mecosta County Michigan. Washington, DC,
United States Department of Agriculture: 239.

Survey, N. C. S. (1994). Soil Survey of Saginaw County Michigan. Washington, DC,
United States Department of Agriculture: 329.

Thomthwaite, C. W. (1948). "An approach toward a rational classification of climate."
Geograply Review 38: 55-94.

 

94

Tomer, M. D. and J. L. Anderson (1995). "Field Evaluation of a soil water-capacitance
probe in a fine sand." Soil Science 159(2): 90-98.

Western, A. W. and G. Bloschl (1999). "On the spatial scaling of soil moisture." Journal
of Hydrology 217: 203-224.

Western, A. W., R. B. Grayson and G. Bloschl (2002). "Scaling of soil moisture: A
hydrologic perspective." Annual Review of Earth and Planetm Science 30: 149-80.

Wilson, D. J ., A. W. Western, R. B. Grayson, A. A. Berg, M. S. Lear, M. Rodell, J. S.
Famiglietti, R. A. Woods and T. A. McMahon (2003). "Spatial distribution of soil
moisture over 6 and 30 cm depth, Mahurangi River catchment, New Zealand." Journal of
Hydrology 276: 254-274.

Wright, J. and F. Bergsrud (1991). Irrigation Scheduling: Checkbook Method. University
of Minnesota, Minnesota Extension Service: 12.

Wright, J. L. and J. C. Stark (1990). Potato. Irrigation of agricultural crops. B. A. Stewart
and D. R. Nielsen. Madison, WI, American Society of Agronomy Inc., Crop Science
Society of America Inc., Soil Science Society of America Inc. 30: 860-888.

Wu, K. (1998). "Measurement of soil moisture change in spatially heterogeneous
weathered soils using a capacitance probe." Hydrological Processes 12: 135-146.

95

   

uljljjjjnjjjjjm1111121er