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

 

DATE DUE DATE DUE DATE DUE

 

0 05
gwg

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ma chlRC/DdoOmst-p.“

ASSESSMENT OF A MODELING APPROACH FOR THE ESTIMATION OF
AGRICULTURAL IRRIGATION WATER USE IN MICHIGAN

Thomas N. Moen

A DISSERTATION

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

DOCTOR OF PHILOSOPHY

Department of Resource Development

1999

ABSTRACT

ASSESSMENT OF A MODELING APPROACH FOR THE ESTIMATION OF
AGRICULTURAL IRRIGATION WATER USE IN MICHIGAN

By

Thomas N. Moen

Monitoring the use of critical natural resources such as water is an important component of
effective resource management. In Michigan, water use data are collected for four sectors;
industrial, municipal, electrical power generation plants, and agriculture. In terms of
agricultural use, Michigan’s water use reporting law requires water users who have the
capacity to withdraw over 100,000 gallons per day averaged over any 30-day period during
a year to report their water use to the Department of Environmental Quality. From 1993 to
1995, agricultural water use data were collected using a direct reporting method. The survey
method was time-consuming and prone to errors and biases in reporting. With the
availability of high-resolution precipitation data from the National Weather Service in 1996,
a modeling approach based on remotely-sensed weather data, a state-wide soil database, and
a soil water balance model was developed. The soil water balance method (SWBM) offers
many potential benefits compared to the survey approach; non-invasive data collection,
detailed estimates of water use over time and space during the course of a growing season,
and a defensible and scientific methodology for estimating water use. The SWBM method
is dependent on the availability of spatially and temporally accurate data inputs for soil and
weather data. In this research, the potential for the use of simplified soil data inputs was

examined using 30 year simulation runs that test for irrigation differences between soil map

units. Precipitation data from the NEXRAD radar system of the National Weather Service
was compared to rain gauge observations as a partial validation of the NEXRAD data.
Range-dependent biases were found in the NEXRAD data for all three years of the study
(1996 — 1998). The correlation of weekly precipitation totals between rain gauge sites was
calculated to assess the potential for the use of rain gauge data as a precipitation source for
the model. Low spatial correlation was found between rain gauge sites. It was concluded
that NEXRAD precipitation data is the best source for detailed estimates at the sub-county
level, although a dense rain gauge network could potentially be used for precipitation inputs
at a county or watershed level. Large discrepancies between NEXRAD estimated
precipitation and ground measured (gauge) precipitation were found for some sites, with
NEXRAD precipitation levels generally lower than gauge observations. Further research
must be conducted to more accurately assess the reliability and validity of the NEXRAD
precipitation estimates. This research describes the successful integration of a biophysical
model in a geographic information system (GIS) environment to provide estimates of
agricultural water use in an operational setting. With continued advancements in data
accuracy and resolution in the future, the SWBM approach is expected to become a valuable

tool for the effective monitoring, management, and analysis of water resources.

Copyright by
THOMAS NATHAN MOEN
1999

ACKNOWLEDGEMENTS

I would like to thank all of my graduate committee members for their support and
contributions to this research project. Dr. Joe T. Ritchie of the Department of Crop and
Soil Sciences was my graduate research advisor and provided unfailing support, wise
guidance, and critical modeling expertise for the duration of the project. Dr. Eckhart
Dersch, Department of Resource Development and chair of the graduate committee
provided wise advice, insightful questions, and academic contributions throughout the
course of my academic program. Dr. Jeffrey A. Andresen of the Department of
Geography and Dr. David P. Lusch of the Center for Remote Sensing & Geographic
Information Science served as committee members as well as contributors to the research
project in the areas of agricultural meteorology and remote sensing/spatial analysis,

respectively. To all: thanks for your time and expertise.

A number of people were involved in this research project, and each provided valuable
contributions. I would especially like to thank Ron Van Til of the Michigan Department
of Environmental Quality whose forward-looking management and whose perseverance
saw the project through from initiation to successful operational status. Brian D. Baer of
the Department of Crop and Soil Sciences and Dr. Yung-Tsung Kang also should be
acknowledged for their computer programming contributions in the areas of soil water

balance modeling and weather data processing, respectively.

Special thanks to my parents, Dr. Aaron N. Moen and Sharon R. Moen, and to family and

friends, for your love and support.

TABLE OF CONTENTS

LIST OF TABLES ................................................................................... ix
LIST OF FIGURES ................................................................................. xi
LIST OF ABBREVIATIONS .................................................................... xiv
CHAPTER I INTRODUCTION ................................................................. 1
1.1 Introduction ............................................................................. I
1.2 Problem Statement ..................................................................... 4
1.3 Objectives and Methodology ......................................................... 5
1.4 Research Assumptions and Limitations ............................................. 9
1.5 Water Resources in a Global Context ............................................... 1 l
1.6 Summary .............................................................................. 12
CHAPTER II LITERATURE REVIEW ...................................................... 14
2.1 Soil Water Balance Modeling ...................................................... 14
2.2 GIS and Regional Scale Studies ................................................... 20
2.3 Rain Gauge Measurement of Precipitation ....................................... 23
2.4 Radar Measurement of Precipitation .............................................. 29
2.5 Spatial Scale and Regional Modeling ............................................. 33
2.6 Summary .............................................................................. 36
CHAPTER III MODEL AND DATA DESCRIPTION ...................................... 38
3.1 The Soil Water Balance Model ..................................................... 38
3.2 Temperature Data ..................................................................... 44
3.3 Solar Radiation Data ................................................................. 45
3.4 NEXRAD Precipitation Data ....................................................... 46
3.5 Soil Data ............................................................................... 48
3.6 Other GIS Data Layers ............................................................... 54
3.7 Irrigation in Michigan ............................................................... 56
3.8 Summary .............................................................................. 57
CHAPTER IV SOIL DATA ANALYSIS ...................................................... 59
4.1 Objectives ............................................................................. 59
4.2 Methodology .......................................................................... 61

vi

4.3 Results and Analysis ................................................................. 63

4.4 Outlier Soil Analysis ................................................................. 72
4.5 Composite Soil Analysis ............................................................ 79
4.6 Conclusions and Recommendations ............................................... 81
CHAPTER V PRECIPITATION DATA ANALYSIS ....................................... 83
5.1 Objectives ............................................................................. 83
5.2 Weather Station Data: Statistical Summary ...................................... 86
5.3 Spatial Correlation of Gauge Measured Precipitation ........................... 91
5.4 NEXRAD Data: Statistical Summary (1996) .................................... 96
5.5 NEXRAD Data: Statistical Summary (1997) ................................... 106
5.6 NEXRAD Data: Statistical Summary (1998) .................................. 114
5.7 Correlation Between Gauge and NEXRAD Precipitation ..................... 121
5.8 Irrigation Estimates Using Gauge vs. NEXRAD Precipitation ............... 138
5.9 Summary, Conclusions, and Recommendations ............................... 146
CHAPTER VI SUMMARY, FINAL ANALYSIS, AND CONCLUSIONS ............ 149
6.1 Research Objectives ............................................................... 149
6.2 Soil Data Analysis .................................................................. 151
6.3 Precipitation Data Analysis ....................................................... 152
6.4 Model Sensitivity to Management Parameters ................................. 155
6.5 Total Water Requirements ....................................................... 157
6.6 Irrigation Amount Versus Precipitation ......................................... 161
6.7 Options for the Estimation of Irrigation Water Use by Agriculture .........166
6.8 Study Limitations .................................................................. 169
6.9 Issues for Further Research ....................................................... 171
6.10 Final Comments .................................................................... 173
APPENDIX A. THE RITCHIE SOIL WATER BALANCE MODEL ................... 175

APPENDIX B CROPS AND DEFAULT MANAGEMENT PARAMETERS 1 82

REFERENCES ................................................................................... 1 84

vii

LIST OF TABLES

Table 3.1. Soil layers and depth (cm.) used in the SWBM.

A = top depth of layer, Z = bottom depth of layer. .................................... 51
Table 3.2 Soil parameters derived from the STATSGO database ........................... 51
Table 3.3 Values for KsMatrix as a function of Clay and Sand % ........................... 52
Table 4.1 Results of the test for paired differences (by map unit count) ..................... 66

Table 4.2 Results of the test for paired differences
(by percentage of land area) .............................................................. 68

Table 4.3 Soil map units with irrigation counts less than -I .96 SD
from the mean, by crop type ............................................................. 74

Table 4.4 Component soils of soil map unit ‘MIOO8’ showing the name,
ID, percentage, and surface texture of each component. ............................. 75

Table 4.5 Surface layer textural class composition (percentage) of the eight
outlier soil map units ...................................................................... 76

Table 4.6 Surface layer textural class composition (percentage) of five
soil map units with high irrigation need. ............................................... 77

Table 4.7 Range of irrigation estimates for counties reporting more than
1,000 acres of irrigation in 1997. ........................................................ 80

Table 5.1 Summary of maximum daily rainfall reported for May -— August,
1996 — 1997 by all rain gauges (cm.). ................................................... 86

Table 5.2 Minimum, maximum, mean, and standard deviation of total
precipitation (cm) reported by all stations,May — August, 1996 — 1998 ............ 87

Table 5.3 Minimum, maximum, mean, and standard deviation of total

precipitation (cm) reported by Lansing and Flint stations for

May — August, 1961 — 1990. ............................................................. 88
Table 5.4 Minimum, maximum, mean, and standard deviation of

May — July precipitation totals (cm.) for all NEXRAD cells

1996, 1997, and 1998. .................................................................... 97

Table 5.5 Minimum, maximum, mean, and standard deviation of precipitation
totals by month for all NEXRAD cells, 1996. ........................................ 100

Table 5.6 Minimum, maximum, mean, and standard deviation of precipitation

viii

totals by month for all NEXRAD cells, 1997. ....................................... 109

Table 5.7 Minimum, maximum, mean, and standard deviation of precipitation
totals by month for all NEXRAD cells, 1998. ........................................ I 15

Table 5.8 May — August precipitation totals (cm.) for NEXRAD Radar (R)
and Rain Gauge (G) and R/G ratio for all stations reporting in
1996, 1997, and 1998. ................................................................... 123

Table 5.9 Summary of differential between radar and gauge-measured
precipitation for a) May — August totals, 1996 — 1998
and b) July —August totals. n = number of stations ................................. 125

Table 5.10 Summary of linear regression of 1997 weekly NEXRAD
precipitation totals on Gauge totals, ordered by r2 (descending). .................. 133

Table 5.11a Results of the simulations using NEXRAD (R) and
gauge precipitation (P) data for corn. .................................................. 141

Table 5.11b Results of the simulations using NEXRAD (R) and
gauge precipitation (P) data for soybeans. ............................................ 141

Table 5.12 Minimum, maximum, mean, and standard deviation of
May — August precipitation totals for all NEXRAD cells by County. ............ 145

Table 5.13 SWBM irrigation estimates using 1997 reported acreage
with a) NEXRAD precipitation (R) and
b) Nearest Rain Gauge precipitation (G) as precipitation input. .................. 146

Table 6.1 Total irrigation requirements under different management strategies
using 1997 irrigator records and NEXRAD precipitation data. ................... 156

LIST OF FIGURES

Figure 3.1 20 kilometer temperature and solar radiation cells for the

state of Michigan used in 1996. .......................................................... 45
Figure 3.2 16 kilometer temperature and solar radiation cells for the

state of Michigan used in 1997. .......................................................... 46
Figure 3.3 4 kilometer NEXRAD cells for the state of Michigan ........................... 47
Figure 3.4 STATSGO Map Units for Michigan. ............................................... 49
Figure 3.5 STATSGO Map Units for Barry County, Michigan with 4 km.

NEXRAD grid cells superimposed. ..................................................... 54
Figure 3.6 Counties of Michigan. ................................................................ 55
Figure 3.7 Watersheds of Michigan. ............................................................. 56
Figure 4.1 Soil groups with non-different irrigation amount for corn. ..................... 70
Figure 4.2 Soil groups with non-different irrigation amount for soybeans. ................ 70
Figure 4.3 Soil groups with non-different irrigation amount for strawberries. ............ 71
Figure 4.4 Soil groups with non-different irrigation amount for sod. ...................... 71

Figure 4.5 Standard deviation groupings for corn computed from the 30 year
simulation runs showing soil areas with irrigation +/- one or more
standard deviations from the mean irrigation amount. ............................... 73

Figure 5.1a Spatial distribution of May — August 1996 total precipitation
(size indicates amount of precipitation. Range = 11.86 to 56.03 cm. .............. 89

Figure 5.1b Spatial distribution of May — August 1997 total precipitation
Range = 18.01 to 48.92 cm. .............................................................. 89

Figure 5.1c Spatial distribution of May — August 1998 total precipitation
Range = 13.00 to 46.05 cm. .............................................................. 90

Figure 5.2a Spatial correlation of weekly precipitation between Lansing
and all other stations, May — August 1996. ............................................ 92

Figure 5.2b Spatial correlation of weekly precipitation between Lansing
and all other stations, May — August 1997. ............................................ 92

Figure 5.2c Spatial correlation of weekly precipitation between Lansing

and all other stations, May — August 1998. ........................................... 93

Figure 5.3a Spatial correlation of weekly precipitation between Allegan

and all other stations, May — August 1996. ........................................... 94
Figure 5.3b Spatial correlation of weekly precipitation between Allegan

and all other stations, May — August 1997. .......................................... 95
Figure 5.3c Spatial correlation of weekly precipitation between Allegan

and all other stations, May — August 1998. .......................................... 95
Figure 5.4 Frequency distribution of 1996 NEXRAD precipitation (cm)

for the period May 1 — August 31. .................................................... 99
Figure 5.5a Total precipitation by NEXRAD cell for May, 1996 ........................ 101
Figure 5.5b Total precipitation by NEXRAD cell for June, 1996 ........................ 102
Figure 5.5c Total precipitation by NEXRAD cell for July, 1996 ........................ 103
Figure 5.5d Total precipitation by NEXRAD cell for August, 1996 .................... 104
Figure 5.5e Total precipitation by NEXRAD cell for May - August, 1996 ............ 105
Figure 5.6 Frequency distribution of 1997 NEXRAD precipitation (cm)

for the period May 1 — August 31. ..................................................... 108
Figure 5.7a Total precipitation by NEXRAD cell for May, 1997 ........................ 110
Figure 5.7b Total precipitation by NEXRAD cell for June, 1997 ........................ 111
Figure 5.7c Total precipitation by NEXRAD cell for July, 1997 ......................... 112
Figure 5.7d Total precipitation by NEXRAD cell for August, 1997 ..................... 113
Figure 5.7e Total precipitation by NEXRAD cell for May - August, 1997 ............ 114
Figure 5.8 Frequency distribution of 1998 NEXRAD precipitation (cm)

for the period May 1 -- August 31. ..................................................... 1 16
Figure 5.9a Total precipitation by NEXRAD cell for May, 1998 ........................ 1 17
Figure 5% Total precipitation by NEXRAD cell for June, 1998 ........................ l 18
Figure 5.9c Total precipitation by NEXRAD cell for July, 1998 ......................... 1 19
Figure 5.9d Total precipitation by NEXRAD cell for August, 1998 ..................... 120

xi

Figure 5.9e Total precipitation by NEXRAD cell for May - August, 1998 ............ 121

Figure 5.10a Frequency distribution of R/G ratio, 1996. .................................... 126
Figure 5.10b Frequency distribution of R/G ratio, 1997. ................................... 126
Figure 5.10c Frequency distribution of R/G ratio, 1998. ................................... 127
Figure 5.11a Spatial distribution of the R/G ratio, 1996. ................................... 128
Figure 5.11b Spatial distribution of the R/G ratio, 1997. ................................... 128
Figure 5.11c Spatial distribution of the R/G ratio, 1998. .................................... 129
Figure 5.12a Rain gauge totals versus NEXRAD totals for May - August 1996.........130
Figure 5.12b Rain gauge totals versus NEXRAD totals for May - August 1997. . . . .....l3l
Figure 5.12c Rain gauge totals versus NEXRAD totals for May - August 1998. . . . . l 32

Figure 5.13a Spatial distribution of r2 values for correlation of weekly radar
precipitation on gauge measured precipitation, 1996. ............................... 136

Figure 5.13b Spatial distribution of r2 values for correlation of weekly radar
precipitation on gauge measured precipitation, 1997. ............................... 136

Figure 5130 Spatial distribution of r2 values for correlation of weekly radar
precipitation on gauge measured precipitation, 1998. ............................... 137

Figure 5.14a Distribution of total May - August 1997 precipitation (cm.)
in St. Joseph County by NEXRAD cell using NEXRAD data. ................... 143

Figure 5.14b Distribution of total May - August 1997 precipitation (cm.)
in St. Joseph County by NEXRAD cell using
nearest station data (SI). ................................................................ 143

Figure 5.14c Distribution of total May - August 1997 precipitation (cm.)

in St. Joseph County by NEXRAD cell using

weighted multiple station data (S2). ................................................... 144
Figure 6.1 Total water requirements for corn on soil M1001 for the 30 year

simulations using weather data from the

Lansing, MI station 1961 — 1990. ...................................................... 158

Figure 6.2 Total water requirements for corn on soil M1001 for the 30 year

xii

simulations using weather data from the
Lansing, MI station 1961 — 1990. ...................................................... 159

Figure 6.3 Total water requirements for soybeans on soil M1001 for the 30 year
simulations using weather data from the Lansing, MI station 1961 — 1990 ...... 160

Figure 6.4 Total simulated irrigation (acre-inches) versus scaled NEXRAD
precipitation with scalar applied of .5, .7, 1.0, 1.3, 1.5 and 2.0. ................... 162

Figure 6.5 Irrigation applied (cm.) versus NEXRAD precipitation
for field com, 1997 ...................................................................... 164

Figure 6.6 Irrigation applied (cm.) versus July — Aug. NEXRAD precipitation
for field com, 1997. ...................................................................... 165

xiii

cm.

GIS

MDA
MDEQ
MI

NEXRAD
NRCS
NWS

STATSGO
SWBM

USDA

LIST OF ABBREVIATIONS

Centimeters
Geographic Information System

Kilometers

Michigan Department of Agriculture

Michigan Department of Environmental Quality
Michigan

Next Generation Radar

Natural Resources Conservation Service

National Weather Service
State Soil Geographic Data Base
Soil Water Balance Model

United States Department of Agriculture

xiv

CHAPTER I

INTRODUCTION

1.1 Introduction

The waters of the Great Lakes Basin are a resource of enormous ecological and economic
value. A 1985 study conducted by the International Joint Commission projected significant
increases in industrial, power generation, agricultural, public supply, and other water uses in
the Basin beyond 2000 (U C, 1985). During the past twenty years, several proposals have
been made to divert Great Lakes water outside of the Great Lakes Basin to alleviate water
shortages. The need to meet current and future water needs within the Great Lakes Basin
and to oppose unwarranted water diversions to other regions of the country prompted the
eight Great Lakes states and two Canadian provinces to adopt the Great Lakes Charter in
1985 (Great Lakes Governors Task Force, 1985). The Charter is a vital cooperative
agreement intended to guide the protection and management of the water resources of the
Great Lakes Basin. While the Charter serves as a basic framework for preserving Great
Lakes waters for use within the basin, long-term and effective Great Lakes protection will
require cooperative efforts by the states and provinces to manage this internationally

significant resource.

Following formal adoption of the Great Lakes Charter, the seven Great Lakes states passed
laws requiring water use reporting. In Michigan, Public Acts 326 and 327 of 1990 were
enacted (now Part 327, Great Lakes Preservation, Michigan Natural Resources and
Environmental Protection Act, 1994 PA 451, as amended) to authorize the Michigan Water
Use Reporting Program. The overall goal of the program is to establish an inventory of
major irrigation, industrial, power generation, and public water supply uses of water on a
statewide basis. The program helps Michigan establish the strongest possible legal defense
against Great Lakes water diversions to other areas of the country, while providing essential

water use information for state and regional water resources planning and management.

Michigan’s water use reporting law requires water users who have the capacity to withdraw
over 100,000 gallons per day averaged over any 30-day period during a year to report their
water use to the Department of Environmental Quality (MDEQ). The law includes a special
provision for agricultural irrigation reporting that authorizes the Department of
Environmental Quality and the Department of Agriculture (MDA) to work jointly with
county soil conservation districts to collect water use information. Under the law, farmers

are required to report their irrigated crops, acreage, and water sources on an annual basis.

From 1993 to 1995, data for the Water Use Reporting Program were collected using a direct
reporting method. At the end of each growing season, a form was sent to inigators for
reporting crops grown, acreage, and water use estimates for the year. MDEQ project staff
estimated that about 2/3 of the total irrigated crop acreage in the state was reported using the

direct reporting method; many irrigators did not return the form. In addition, the process of

data collection and verification was time consuming both from the standpoint of project
staff and for irrigators filling out the lengthy form. Biases in water use estimates were also
a concern of project administrators. It was felt that irrigators might tend to underestimate

water use in anticipation of potential regulatory actions.

In 1996 the MDEQ contracted with the Department of Crop and Soil Sciences at Michigan
State University to develop an alternative approach for estimating agricultural irrigation
water use. The recently available precipitation data from the Next Generation Weather
Radar System (NEXRAD) of the National Weather Service provided spatially referenced,
hourly estimates of precipitation at a resolution of four kilometers. It was hypothesized that
NEXRAD precipitation data, combined with other spatially referenced input data, could be
used to run a physically-based soil water balance model (Ritchie, 1985) to simulate soil
water conditions and irrigation requirements over the course of a growing season. Using the
new method, irrigators would be required to report only the crops grown and the acreage of
each crop for each year. Geographic Information System (GIS) and relational database
technology would be used to integrate spatial and temporal data with the soil water balance

model to provide simulated estimates of irrigation water requirements.

The soil water balance model (SWBM) method was developed during 1996 and 1997. This
development involved the programming of a data entry application for acquiring and storing
irrigator information, acquisition and conversion of soil and meteorological data layers, and
conversion of the soil water balance model (Ritchie, 1985) to run in an integrated modeling

environment (IME) designed for the Microsoft Windows operating environment. The [ME

was designed to incorporate the input data, model, and GIS data layers into a single
application for efficient access, execution, and analysis. Results for the 1997 growing
season were reported to MDA and MDEQ in October of 1998, using the irrigator records
provided by MDA, which was a sample of the total population of irrigators. An advantage
of the SWBM method is that data collection from irrigators is greatly simplified. Recent
discussions at MDEQ and MDA have suggested use of the US. Department of Agriculture
census data to identify irrigated acreage, which would eliminate the need for any direct data
collection. In addition to county level reporting, the SWBM method also allows for results
to be aggregated and analyzed on a watershed basis, consistent with the recent emphasis on

watershed-based water quality management (Grant, 1997).

1.2 Problem Statement

The SWBM method provides the potential for a more efficient and more accurate reporting
process for agricultural irrigation water use. The completion of the 1997 irrigation
estimation demonstrated the ability of the SWBM method to simulate levels of irrigation
water use in the range of expected values. Further analysis and validation is necessary
before adoption and use on a wider scale is considered, however. The future adoption and
use of the SWBM method by Michigan and other Great Lakes states could be negatively
affected by the dependence of the SWBM on detailed input data requirements. Detailed
precipitation data (both temporally and spatially) and detailed soil parameters are required

by the SWBM. In addition, the location and acreage of irrigators must be known to a

resolution of 4 km. in order to associate NEXRAD precipitation data with irrigated acreage.
These data are often not readily available to government agencies in the format required for
running the model. In some cases, the data may be obtainable, but the government agency
may not have the technical expertise available or funding available to prepare and maintain

data for use in the SWBM.

The SWBM currently uses hourly precipitation data during the course of the growing season
at a spatial scale of 4 km. Most government agencies will have ready access only to daily
estimates of precipitation at a larger spatial scale (county or larger area). The adaptation of
the SWBM to use currently existing (but coarser resolution) soil and weather data would
make implementation and adOption in other states and regions more likely. It is not known,
however, if the use of coarser resolution input data would provide water use estimates
within an acceptable level of precision and accuracy as compared to the use of fine-scale
resolution input data. The accuracy of the NEXRAD precipitation data relative to rain
gauge measured precipitation is also not known and is the subject of chapter V of this

dissertation.

1.3 Objectives and Methodology

The overall goal of this research is to provide a thorough assessment of various factors
affecting the use and accuracy of the SWBM method for the estimation of agricultural

irrigation water use. The research focuses on the sensitivity of the irrigation estimates to

soil data and precipitation data. The main objective of the analysis is to assess the potential
for the use of coarse-scale input data and it’s potential impact on the accuracy of results, as
compared to the use of fine-scale data. A secondary, but no less important, objective is to
compare NEXRAD precipitation estimates with precipitation levels reported from a

network of rain gauges throughout the state.

For this project, the SWBM uses the STATSGO soil data (NRCS, 1994) to provide
estimates of local soil parameters. The STATSGO database defines 190 soil map units for
Michigan, each consisting of 2 to 21 soil series or soil phases. A more detailed description
of the STATSGO database can be found in chapter III. The variability of irrigation
requirements between different soil map units is not known. To determine the variability of
irrigation requirements by soil map units a historical database of daily weather for central
Michigan will be used to represent the range of climatic conditions expected. For each year
simulated, climate will be assumed to be the same across the entire state, so the only
difference in results will be due to differences in soil characteristics of the STATSGO soil
map units. Simulations will be run for four crops for each map unit-year combination. For
each soil map unit, the result of the simulations will be a record of total irrigation amount
(cm.) applied for the year. A paired t-test will be used to determine if the difference in
irrigation amount between any pair of soil map units was different than zero over the 30
year simulation period. Non-different soil map units will be grouped by crop type and the
results analyzed. Results of this analysis will provide information on the potential to create
major groups of soils from the 190 Map Units. The use of major soil groupings would

greatly simplify the data inputs for soil, if the results show that this can be done without a

loss in estimate accuracy.

Meteorological data inputs to the SWBM are precipitation, daily minimum temperature,
maximum temperature, and daily solar radiation. Of these variables, precipitation plays a
major role in determining the soil water balance, and is the most variable over space and
time. For these reasons, the weather data analysis will focus on the NEXRAD precipitation

data. Analysis of NEXRAD data will consist of the following:

1. Detection of outliers.
2. Detection/assessment of systematic bias.

3. Comparison and correlation with gauge measured weather station data.

Given the highly variable and somewhat random distribution of precipitation over space
and time, it could be hypothesized that as the time period of analysis is increased, the
values of cumulative precipitation for adjacent cells or cells in close proximity will
converge to similar values. That is, distinct differences for adjacent cells are expected on
a daily basis, but summed over the 4 month summer period, one would expect
convergence to similar values for adjacent or nearby cells. Outlier cells, those that differ
significantly from the regional mean, could be a result of consistent bias (ground clutter)
or random errors in the radar estimation of precipitation. The relative occurrence and
spatial and temporal variation of outlier cells for Michigan NEXRAD data is currently not
known. Analysis to identify outliers will be performed for each of the three years of

weather data. Outliers will be identified as those cells that have irrigation amounts

greater than 1.96 standard deviations from the mean irrigation amount for the analysis
period. Some researchers have shown changes in mean values for cells conditional upon
distance from the radar site (Smith et. al., 1996). In this study, systematic range-
dependent bias of the NEXRAD data will assessed by visual analysis of the monthly and

seasonal precipitation values.

To determine the correlation between NEXRAD data and rain gauge measured data, daily
precipitation data will be obtained from all reporting weather stations in the Michigan
Agricultural Cooperative reporting network for the 3 year period 1996 — 1998. Linkage
analysis of daily rainfall using the correlation coefficient will be performed to assess the
degree of correlation (spatially and temporally) between the 64 stations. This information
will be used to gain a better understanding of variation in rainfall over the state and
potentially identify regions of strong correlation within the state. Significant differences,
if any, between the NEXRAD precipitation data and the rain gauge data will be identified

and discussed.

Results of the soil and precipitation analysis will be used to assess the potential for use of a
coarse-resolution input data set consisting of aggregated soils and weather station data. The
results will allow for an estimation of the magnitude of the difference in irrigation estimates
obtained using fine-scale versus coarse-scale data. In addition to the soil and precipitation
data analysis described above, the sensitivity of the SWBM to changes in management
parameters will be assessed. There are two management parameters that can be

manipulated by the operator of the SWBM that correspond to management decisions made

by irrigators. The first management parameter is the decision of when to irrigate, as a
function of soil moisture. The second management parameter is the amount of water to
apply during each irrigation event. Simulations will be run to determine the sensitivity of
model results to the range of reasonable and customary values for these two management

parameters.

1.4 Research Assumptions and Limitations

A key assumption of this research is that irrigators use some logical method to determine
when and how much to irrigate a crop, based on some desired objective. This method can
be a computerized soil water budget, monitoring/measurement of actual soil water,
observation of plant indicators, or some combination of these (Hill, 1991). It is assumed
in this research that the desired objective is to maintain the soil water balance for optimal
development of the crop. No constraints on water use, either physical or economic, are

considered.

A limitation of this research is the lack of available data for which to validate the SWBM
method results. Because of this, comparison of the SWBM method results to actual
irrigator records will not be undertaken as part of this research. A partial validation of the
SWBM method will be done by comparing SWBM results to those previously reported in
other Michigan irrigation studies. Due to data, time and logistical constraints, it is not

possible to comprehensively validate the SWBM method for all irrigated crops, all

management strategies, and all possible climate and soil conditions that exist in

Michigan.

The judgement of ‘accuracy’ is compounded by the differences in scale between the
model and potential records used to validate the results. The finest resolution of the
model is 4 km., while irrigator records pertain to the field level. Strict validation of the
SWBM model over a wide range of climatic conditions and crops would require detailed
data at the field level over many growing seasons and many crop types. Such validation
will only be possible after the SWBM is implemented in an operational setting, with
volunteered records used for validation on an on-going basis, as such data becomes
available. Validation of the soil water balance model in a research setting is discussed in

Chapter II.

The direct reporting method was conducted for the years 1993 to 1995, prior to the
availability of NEXRAD precipitation data. The direct reporting method was suspended
in 1996, with the SWBM method planned to be implemented in 1997. Because there is
not a single year in which the Survey method and the SWBM method are used
simultaneously, it is not possible to compare results of each method directly. Because of
this, it will not be possible to analyze differences in water use estimates obtained using

the two methods.

10

1.5 Water Resources in a Global Context

It is important to consider, in any research setting, the context of the research and it’s
objectives in relation to a global perspective. The basic goal of this research is to
improve our ability to estimate and monitor agricultural irrigation water use. In this case,
the monitoring function is performed by a government agency, the Michigan Department
of Environmental Quality (MDEQ). MDEQ will use these results, in addition to
information on water use by other sectors to analyze and monitor water use levels

throughout the state.

While the results of this research will be directly used by MDEQ, there are a number of
other potential beneficiaries. The water use reporting program is national in scope. If the
SWBM method is shown to be reliable, accurate, and efficient, the potential exists for
other states or regions to use this method in their water use reporting programs. Our
understanding of irrigation water use impacts and potential for irrigation throughout the
study area should be greatly enhanced. A by-product of the analysis will be the
generation of model—simulated irrigation requirements that can be compared to actual
water use by irrigators. Implementation of the SWBM at the field level could assist
irrigators in determining optimal irrigation amounts. The efficient use of water in
agriculture is critical. Deficiencies in water use result in failed or lowered crop yields,
leading to inconsistent production levels and unstable economic returns for the irrigator.
Over-use of water can result in excessive infiltration, and corresponding problems such as

increased nitrogen leaching. Researchers, resource managers, and producers should all

11

learn from this project.

The earth’s population continues to expand and requires more and more water for
municipal, industrial, agricultural, environmental, recreational, and other needs (Bouwer,
1994). In addition to more direct consumption of water, more water will be needed for
irrigation of crops to produce food for the expanding population. In water—scarce areas,
competition for water will become increasingly intense, and regulation of water use will
become greater. As an example of this, in 1980 the state of Arizona passed the
Groundwater Management Act (Coupal and Wilson, 1990). Under this regulatory
legislation, the Department of Water Resources is empowered to reduce farm—level water
use by imposing higher irrigation efficiencies on farmers. In less developed countries,
where much of the projected population expansion is expected (U.N. Population Fund,
1993), the problems of water quantity and quality are more acute. It is currently
estimated that half of the population of the Third World does not have access to safe
drinking water, that one billion get sick each year from water-borne diseases, and that 12
million die, 80% of which are children (Bouwer, 1994). Water resource management
must be implemented on a local or regional scale, with different solutions to different
problems in different areas. We should all be driven by a common goal however:

promotion of the conservation, protection, and efficient use of water resources.

1.6 Summary

This chapter has discussed the historical development of this research project and the use of

its results in a management setting. Objectives of the research have been discussed, and

methodology introduced. The context of the research in a global setting has been discussed.

While this research is driven by an applied problem, in a more general sense it pertains to
the application of advanced information technology for resource management. Most, if not
all, resource management problems involve time and spatial components, encompassed and
interrelated within a complex system. Better decisions can be made if one has timely access
to pertinent information, and information technology can provide this information. This
research offers the opportunity to bridge the gap between theory and practice that often
exists between the research institution and managers of resources. This research will
demonstrate that results of basic science (theory) can be used in a management setting

(practice) to assist in our understanding and management of natural resources.

The remainder of this study is organized as follows. In Chapter II, a discussion of literature
related to this research is presented. Chapter III discusses the soil water balance model, data
conversions, and source of all data used in the research. Chapter IV presents and discusses
the results of the soil data analysis. Chapter V presents and discusses the results of the
precipitation data analysis. Chapter VI summarizes and discusses the results and provides

recommendations for future research.

13

CHAPTER II

LITERATURE REVIEW

This chapter presents a review of literature related to this study. The literature review
has been categorized according to the disciplines or fields of study represented in the
study. Studies pertaining to soil water balance modeling are discussed in section 2.1,
followed by a discussion of the use of Geographic Information Systems (GIS) for
regional scale modeling in section 2.2. Studies on precipitation measurement using
rain gauges and radar are discussed in sections 2.3 and 2.4. Section 2.5 presents issues
relating to spatial scale and regional modeling. Section 2.6 summarizes the literature

review.

2.1. Soil Water Balance Modeling

An understanding of the soil water storage profile is important for many hydrological
problems, including irrigation management. Accordingly, factors affecting the soil water
balance have been the focus of many researchers. In an agricultural setting, precipitation
and irrigation provide the major water inputs to the hydrological system. Evaporation,

transpiration, surface runoff, and drainage by percolation are the major water losses.

The Soil Water Balance Model (SWBM) used in this research is the current version of
the Ritchie model as reported by Ritchie (1972), Richardson and Ritchie (1973), Ritchie
(1985), and Ritchie (1998). Ritchie (1972) presented a model for calculating the daily
evaporation rate from a crop surface. In this article, evaporation from the soil surface and
evaporation from the plant surface are considered separately. Test of the model showed
very good agreement between model estimated evaporation and measured evaporation

using a weighing lysimeter for a 37 day period on a grain sorghum test plot.

Richardson and Ritchie (1973) evaluated the SWBM on a watershed basis. This work
recognized the effect of soil water content on runoff, as discussed in Knisel and Baird
(1969). In the 1973 study, Richardson and Ritchie used three years of data from a 20
acre watershed to test the model. Rainfall was measured with a rain gauge, and runoff
measured at the watershed outlet. Effective rainfall was calculated as precipitation —
runoff, and used as a model input. By removing an unknown from the equation (runoff)
and replacing it with measured runoff, the SWBM predicted total soil water content with
high accuracy (correlation coefficient = .99). The SWBM is a critical component of the
CERES (Crop Estimation through Resource and Environmental Synthesis) family of
crop-soil-atrnosphere models, which are used in the Decision Support System for
Agrotechnology Transfer (DS SAT), a product of the International Benchmark Sites
Network for Agrotechnology Transfer (IBSNAT, 1986). Gerakis and Ritchie (1998)
used the SWBM in the simulation of atrazine leaching and showed good agreement
between simulated results and observed soil water content at three depths (13 cm., 26

cm., and 67 cm.).

The soil water balance approach has been applied in a number of different research areas
at different scales. Yates (1996) presented a water balance model (WatBal) that was used
to assess the potential impact of climate change on a river basin. This study used the
Priestly-Taylor (1972) method for estimating potential evapotranspiration. Two case
studies were used to test the model, with historical data used to calibrate and validate the
model for each. Hypothetical scenarios of climate change were then used to estimate the
river basin discharge response to changes in temperature and precipitation levels. This
study is an example of modeling at the watershed basin scale. Parameters for the water
balance model such as catchment holding capacity (Smax) are given as basin-scale
estimates. Changes to the soil water are also calculated at a basin scale. While this
empirical approach is useful for understanding general rainfall/runoff/temperature
relationships within a particular basin, it is necessary to calibrate the model for the basin
being analyzed. The need to calibrate for specific locations renders the model less

suitable for use in locales other than the study site.

To assess the impact of tillage practices on soil water, Shanhoultz and Younos (1994)
also used a water balance model approach. Their model used an empirical formula to
estimate evaporation that was derived from field measurements of pan evaporation, soil
water content, precipitation, and runoff. In this formula, evaporation was estimated as a
function of the plant available water in the top 30 cm. of soil, the average time since the
soil was at field capacity, and an estimate of pan evaporation for the day. As in the Yates

(1996) study, a number of parameters required calibration using the test data. These

parameters included potential plant interception, an infiltration parameter, an evaporation
recession constant, a stress factor, and potential depression storage. As in most reported
studies, the simulated soil water content had reasonable agreement with the measured
plant available soil water. Results were only reported for years for which the model was

calibrated, there was no attempt to verify the model for other years.

Water balance calculations are often a critical sub-component of many agriculturally
related studies. For example, studies that examine pesticide contamination of
groundwater often use a soil water balance approach to estimate the amount of water
percolating to groundwater. Peralta et al. (1994), as an example, present a
simulation/optimization model for preventing pesticide contamination of groundwater
while maximizing irrigated crop yield. A fundamental assumption of this model is that
chemical moves only in the liquid phase in response to soil water movement. A similar
study concerning the modeling of water transport and nitrogen dynamics is presented by
Lafolie et al. (1997). In studies such as these, the accurate modeling of soil water

content and soil water flow is a necessary condition for accurate model results.

Models such as those presented by Yates (1996) and Shanhoultz and Younos (1994)
required calibration using test data sets. Interpretation of the results, therefore, should
only be done within the context of the test sites and test years. Models that are more
physically-based, such as that of Black et al. (1969) and Ritchie (1972) should be more
adaptable to a GIS-based modeling environment. Physically-based models require less

calibration for individual sites or regions, which is an important consideration for detailed

modeling in a GIS environment. As detailed GIS data becomes available, the potential
for incorporating physical. models and analyzing larger spatial areas will become greater,
but the models must use available and measurable independent variables in order to be

easily integrated in a GIS environment.

The development and validation of soil water balance models has led to their use as sub-
models in other modeling efforts. As an example Dierckx et. al (1988) used a soil water
balance model (SWATRE) developed by F eddes et al. (1978) in conjunction with the
SUCROS crop model (van Keulen et. al., 1982) to evaluate the ability of the combined
models to estimate soil water levels (by depth) and crop yield under different irrigation
strategies. Parameters for the model parameters were derived from published literature,
including the work of Black et al. (1969) and Ritchie (1972). In this study, good
agreement was obtained between simulated and measured soil water for fully irrigated
and zero-irrigated corn. Predicted and measured grain yield under both conditions also
showed good agreement. As suggested by the authors, these results indicate that such a
modeling system might have the capability of predicting corn yield in response to a given
irrigation sequence so that economic criteria can be used to schedule irrigation. In a
similar study related to irrigation planning in India, Singh and Singh (1996) used
simulation modeling to estimate the optimum irrigation schedule for cotton resulting in
minimum percolation losses. They found the calibrated model to be an effective tool in

evaluating the performance of different on-farm irrigation management scenarios.

Most, but not all, of the water balance models discussed in the literature run on a daily
time-step. Victor et al. (1988) present the results of a study that used a simplified soil
water balance model developed by Frere and POpov (1979), to estimate pearl millet yields
in India. In this study, the soil water balance was computed on a weekly basis. A Water
Requirement Satisfaction Index (WRSI) was used to indicate the extent to which the
water requirements of the crop have been satisfied. Yield was found to be exponentially
related to WRSI, with a high correlation (R2 = .88) between estimated and observed
yields. In this study the WRSI is a cumulative index, and is designed to be an easily
calculated index value that can be used to estimate final yield during as the growing
season progresses. Researchers have shown that moisture stress at a certain critical
periods in crop development is more highly correlated with yields than stress at other
times (Baier and Robertson, 1968, Mack and Ferguson, 1968). This work highlights the
need for the dynamic and accurate modeling of soil water content in order to accurately
estimate crop yields. Studies that involve the use of a water balance model to estimate
other variables (such as crop yield, irrigation requirements, or runoff) are dependent upon

accurate results of the soil water balance model.

The Dierckx et al. (1988) study is a good example of an integrated model, based partly on
prior studies, used to reasonably estimate soil water and crop yield for a field test plot.
Due to time and other constraints, such studies are often limited by space (one field) and
time (one to five growing seasons). Further validation for many growing seasons,
different soil types, and different irrigation strategies would be necessary before models

such as these can be implemented in a management setting. As Lafolie et al. (1997) point

out, it is essential to test models against various experimental conditions in order to
improve them so that they can be applied to a broad range of soil and climatic conditions.
Very few data sets are available for this purpose however. With the increased use of
Geographic Information Systems (GIS) for modeling purposes, the potential to model
over a broad range of soil and climatic conditions has increased. With this ability comes
the enhanced ability to ‘validate’ models over a broader range of soil and climatic
conditions, a necessary step in the integration of physical models in a GIS. In the next
section, studies that have used GIS technology and data in a modeling context are

discussed.

2.2 GIS and Regional Scale Studies

A number of studies have been conducted to map soil water balance or irrigation
requirements on a regional or country scale. Madsen and Holst (1990) divided Denmark
into four climatic zones and derived relationships between the root zone water holding
capacities and the mean irrigation need for grass and barley in each zone. Maps for each
crop showed the mean irrigation need for 36,000 soil profiles. These were further
mapped as areas of low, medium, and high irrigation need for barley. These maps did not
take into account current land use, they were based solely on soil properties. Kerkides et
al. (1996) used measurements from 31 stations throughout Greece to define soil moisture

deficit isolines for the country. Long term average monthly precipitation,

20

evapotranspiraton, and combined soil and vegetation characteristics were used to estimate

the soil water balance using the method of Thornthwaite and Mather (1955).

Knox et al. (1996) mapped the spatial irrigation water requirements for potatoes in
England and Wales. In this study, an irrigation scheduler computer model was used to
calculate the irrigation needs at 11 sites chosen to represent the typical range of climatic
conditions experienced across England and Wales. A regression model within a GIS was
then used to correlate the model results with existing soils, land use, and climate
databases. A series of irrigation need maps at resolution of 5 km. were generated. This
study was conducted in part to provide insights for potential policies for water
conservation in England and Wales. In a follow-up study, Knox et al. (1997) mapped the
total volumetric irrigation water requirements in England and Wales at a resolution of 2
km. using 1994 cropping data. In terms of output, these studies are similar to the outputs
that are the objective of this study; irrigation water requirements on an annual basis for a
region (State/Country). Knox et al., however, did not use actual precipitation when
modeling and mapping irrigation water requirements, instead a ‘design’ dry year was
used, based on simulations using 20 years of rainfall data. Comparison of the model
estimates (adjusted for 1990 cropping patterns) to government reported results for 1990
showed no significant differences. As pointed out by the authors, a number of
simplifying assumptions were made in these studies that are a potential source of error.
These include the classification of soils into 3 categories, the assumption that all the crop
is grown on the dominant soil within each 1 km pixel, and inaccuracies in the exact

location of some farms due to the need for data confidentiality.

Thomas (1992) mapped the agricultural water balance for rice and maize in Yunnan
Province, PR China using a monthly water balance model and GIS data for soils and
elevation. The result of this study was a province-level map showing areas that are fully
suitable, suitable, moderately suitable, and unsuitable for rice and maize. The use of 1:5
million scale soils maps and 600 M contour intervals in this study, however, leads to only
generalized conclusions. Lal et al. (1993) reported on the use of crop simulation models
and GIS for regional productivity analysis in a study that simulated bean cultivation in
western Puerto Rico. Studies such as these are important in developing and
demonstrating technological capabilities. However, higher resolution input data is often

necessary before the results are useful for agricultural planners beyond a general sense.

The use of coarse-scale spatial data is currently necessary because, for many areas of the
world, high-resolution digitized data sets are not yet available. Remote sensing of
climatic data and soil characteristics should provide more detailed data in the future for
many areas of the world, however. As an example, Stewart et al. (1999) report on the use
of satellite data to estimate radiation and evaporation for northwest Mexico. This area
has only a sparse network of instruments measuring climatic variables. Estimates of solar
radiation from satellite data closely matched measured hourly solar radiation. The
maximum spatial resolution of the satellite data was .8 km. The authors indicate that
both radiation and rainfall data were available from the satellite, but only report on
radiation estimates. Validation of rainfall estimates would require a much denser

network of ground-based weather stations (only three were used in the study).

22

To summarize the current literature regarding the use of GIS and remotely-sensed data
for detailed spatially explicit modeling, it is fair to say that this is a field of research and
technology in early developmental stages. Current research is often hampered by the lack
of high-resolution data. Because of this, assumptions must be made and results are
general, based somewhat on the researcher’s assumptions. In the case of remotely-sensed
data, data may be available at high resolutions, but dense networks of ground-based data
stations do not exist for the validation of the remotely-sensed data. In this study this is
the situation for precipitation data. Rainfall is reported at a resolution of 4 km on an
hourly basis to the nearest 1/100th mm. The network of weather stations, located many
kilometers apart, report precipitation on a daily basis to the nearest l/100th inch. The
relatively sparse network of rain gauges makes comprehensive validation of the
NEXRAD data a difficult task. Validation of the NEXRAD precipitation data used in

this study, to the extent possible, is discussed in Chapter V.

2.3 Rain Gauge Measurement of Precipitation

Precipitation is an important variable affecting the soil water balance. Because of its
affect on agriculture and other human activities, the measurement and spatial variation of
precipitation (and other weather variables) is also a topic of importance to climate
researchers. Many studies have been conducted concerning the accuracy of

measurement. Other researchers are concerned with the optimum density of sensors (rain

23

gauges) in a network with the goal of identifying the minimum network necessary to
provide sufficient information on rainfall amounts for a given area. Other researchers
focus on the spatial and temporal variability of precipitation and the movement of storms.
Each of these areas of study are relevant to this research effort. It should be recognized
that precipitation patterns are variable by region or smaller scales and therefore results of

studies reported below are not in all cases directly applicable to Michigan.

As discussed by Neff (1977), early studies (late 1800’s and early 1900’s) of rain gauge
accuracy demonstrated that wind influenced rain-gauge catch, and catch decreased as
wind velocity increased with height. Neff (1977) conducted a study to determine the
difference in rainfall measurement between standard US. Weather Bureau rain gauges
normally exposed (1 M above ground) and rainfall measured in control gauges at the
ground surface. The study was conducted at four locations; Pullman, Washington,
Reynolds Creek, Idaho, Sidney, Montana, and Ekalaka, Montana. The average error for
all locations combined was found to be —10%. That is, rain gauges l M above the ground
caught 10% less rain than the control gauges. The range was 5 — 15%. As Neff points
out, whether this difference is important depends upon the intended use of the rain gauge
data. The difference may be important in a network whose purpose is to provide
quantitative estimates of precipitation for detailed hydrological modeling. It is not
possible to use a simple adjustment (10%) to correct rain gauge records because of the
relationship of error to wind velocity. Errors were zero for storms with little or now

wind, but as much as 75% for storms with high wind. Other factors besides wind

24

velocity influence wind effects on rain gauge catch. The effect is related to drop size

distribution and the relative timing of wind activity and rainfall intensity (Neff, 1977).

In terms of validation of the NEXRAD precipitation data, results of the Neff study
discussed above should be kept in mind. Rain gauge estimates are point estimates, with
some degree of measurement inaccuracies. Radar-estimated rainfall is generalized over a
continuous surface and has other sources of inaccuracy, as discussed later in this chapter.
Because of potential inaccuracies and differences in spatial and temporal resolution of the
measurements, a direct comparison and comprehensive validation is not possible.
However, we should expect some degree of correlation between gauge-measured and

NEXRAD-estimated precipitation.

Hubbard (1994) points out that confidence in network measurements are more than a
question of sensor accuracy. Measurements represent conditions at a station but are also
often used to infer conditions between sites, in order to report basin or regional level data.
Many studies have been performed to investigate the correlation of meteorological data
between stations. Hendrick and Comer (1970) reported on space variations of
precipitation and implications for rain gauge network design. In this study, 23 rain
gauges in a 43 square mile watershed were used to examine the correlation of daily
precipitation. They concluded that a 9-gauge network would be necessary for correlation
of 90% or more for them more variable summer storms. This roughly translates to a
distance of 2 miles (3.2 km) between stations. Note that this distance is slightly less than

the resolution of NEXRAD data (4 km.) used in this research. Hendrick and Comer

25

(1970) also point out that, because of the spatial variability of storms, substantial errors
will occasionally occur when a rain gauge is used to estimate rainfall at a nearby
ungauged point regardless of the high correlation that may exist between rainfall at the
two points. The time period of interest is an important factor to consider when assessing
the accuracy and need for rain gauges. For long term (annual) estimation of rainfall data,
Eagleson (1967) concluded that only two stations were necessary in a 1250 square mile
watershed located in Australia. Long-term averages have little relevance to the subject of
this dissertation, except to indicate ranges of precipitation totals that one might expect.

Of more relevance are daily and weekly variations in rainfall.

In order to quantify the spatial variability for a number of daily meteorological variables
from automated weather stations, Hubbard (1994) examined 5 years of data from 24
stations in the High Plains Automated Weather Data Network. The coefficient of
determination (r2) was used as the statistical measure to quantify spatial variability.
Analyses were centered on 0rd, Nebraska, with correlation calculated between the 0rd
station and all other stations. These results were then used to create contour maps of
correlation fields (using a kriging technique). Variograms were prepared by plotting the
r2 between station pairs and their separation distance. When plotted by month, the annual
variograms show the degree of separation (km) needed to achieve a certain level of
correlation (r2). The results show the seasonal variation in spatial variability for each
weather variable. Generally, higher correlation was observed in spring and fall and lower
correlation in middle summer and middle winter. Hubbard found that in order to achieve

a correlation of 90% for maximum temperature, a station separation of 60 km or less

26

would be required. For minimum temperature, relative humidity, solar radiation, and
evapotranspiration, the separation distance for 90% correlation was found to be 30 km.
For precipitation, a separation distance of 5 km or less would be required for a 90%
correlation (10 km for 75% correlation). These results are indirectly applicable to this
study, if summer precipitation patterns are somewhat similar between the High Plains
area and Michigan. The NEXRAD resolution of 4 km. is slightly less than the separation

distance of 5 km for 90% correlation found by Hubbard.

Huff and Shipp (1969) investigated the spatial correlation of different storm types in
three Illinois rain gauge networks. Similar to other studies, they found that a greater
density of rain gauges is needed in the warm season (May to September) than in the
colder months of the year to achieve the same level of correlation. For all storms in May
- September, a distance of 2 miles (3.2 km) or less was found to be necessary for an r2 of
.90. Stol (1972) also reported monthly differences in correlation coefficients between
rain gauges in a study conducted in the Netherlands. A distance of roughly 4 km. was

necessary to achieve a correlation of .90 in the summer months.

In a Texas study, Lyons (1990) examined monthly precipitation at 46 stations throughout
the state over a continuous period from 1923 — 1984. As in the Hubbard (1994) study,
higher correlation between gauge sites was found for winter and spring, and lower
correlation in summer months. Large positive and negative anomalies were seen, and it
was concluded that monthly precipitation anomalies could not be predicted or anticipated

based on time-series or spectral analysis. Results of this study are not directly applicable

27

to Michigan because of differences in location and climate regimes between Texas and
Michigan. The lower correlation of summer rainfall between stations is again
substantiated by this study, however. The study points out the spatial and temporal
variability in monthly precipitation estimates over a large region. Berndtsson (1988)
found that correlation fields defined as the area within a .7 correlation isoline were
usually less than 8 km2 in a study in Tunisia. Less than 50% correlation was found at

distances of 50 km or more.

The use of rain gauge data to reconstruct the movement and spatial distribution of a storm
was investigated by Niemczynowicz (1987). In this study, a network of 12 rain gauges
in Lund, Sweden was used to assess the use of cross-correlation techniques to determine
storm movement. Niemczynowicz concluded that an objective and reliable storm
tracking method does not yet exist. Objective methods failed to accurately track storms if
more than one rainfall cell existed over the network. Radar or other remote sensing

technologies offer the best long-term solution to this problem.

Most of the studies reviewed used daily or monthly data in their evaluation of
precipitation patterns. Using daily data, the general conclusion for rain gauge spacing is
that the distance should be 4 km or less to achieve high correlation between sites.
Experience suggests that convective rainfall (thunderstorms) often deposit rainfall in
localized areas much smaller than 4 km. This has been confirmed by researchers using a
dense rain gauge network in Spain (Lorente and Redano, 1990). If the pattern of rainfall

deposited by thunderstorms is somewhat random in time and space, however, then it

28

should be the case that as the time period of analysis is lengthened, the values of
precipitation amounts should be more closely aligned. In this study, correlation of
weekly precipitation data is likely more important than correlation of daily data. That is,
irrigation decisions are likely based more on a weekly or greater time frame, not on the
presence or absence of rain on a particular day. Chapter V present results of a correlation

analysis of precipitation data for Michigan using weekly totals of precipitation.

2.4 Radar Measurement of Precipitation Data

The precipitation data used in this study was obtained from the National Weather Service
NEXRAD (Next Generation Weather Radar) system of WSR-88D (Weather Surveillance
Radar — I988 D0ppler) radar. This system is expected to provide high-quality, high-
resolution precipitation data for the United States that meet a wide range of hydrologic
applications (Smith et al., 1996). Because of the importance of precipitation in the
SWBM, the scale of NEXRAD precipitation data (4 km) is the scale at which the model

is run, with other data inputs aggregated or disaggregated to this scale.

In a study done prior to the introduction of the NEXRAD system, Austin (1987)
discussed the complexity of the relation between measured radar reflectivity and surface
rainfall. In the 1987 study, Austin compared rain gauge and radar measured data for
twenty storms in New England. Results varied by storm, in seven of the twenty storms,

the radar total precipitation was more than 20% below the gauge-measured precipitation.

29

In four storms the radar precipitation exceeded the gauge by 20%. For frontal storms, the
radar measurements were consistently low compared to the gauge measurements. For
convective stroms, radar measurements were generally higher. Austin (1987) points out
that for highly convective storms, radar may be the only reliable data source for
precipitation estimates. Austin (1987) also suggests that a scientist should use all
available data (from both radar and gauges) if estimates of areal rainfall amounts or
small-scale distribution of rain are needed. Austin found large and apparently random
discrepancies between the amounts of rain collected by individual gauges and the radar-

indicated amounts for the same area (at a resolution of 2 km.).

Smith et al. (1996) compared one year of NEXRAD data with rain gauge data in
Oklahoma. Biases were examined for range dependent sampling, systematic differences
between two radar sites observing the same area, and differences between radar and rain
gauge estimates of rainfall. Range dependent biases were found. Mean rainfall increased
from the radar out to a range of 100 km. and decreased from the 100 to 230 km. range.
Kitchen and Jackson (1993) also found degradation of radar rainfall estimates at far
range. To examine radar-radar biases, paired analysis of NEXRAD rainfall estimates
from two stations with overlapping coverage was examined. Systematic differences were
found, with one station showing a consistently larger rain area and amount than the other.
Smith et al. (1996) indicate that the radar-radar differences are consistent with differences
that could occur due to differences in radar calibration. When comparing rain gauge
precipitation to NEXRAD, Smith et al. found rain gauge observation to be 48% larger in

the range of 0 — 40 km., 18% in the 40 — 160 km. range, and 40% higher in ranges greater

30

than 160 km. NEXRAD data was found to be much more capable for delineating areas
of heavy rainfall during convective storms, however. Numerous storms producing hourly
rainfall accumulation of 50 mm. were completely missed by the rain gauge network.
Smith et al. suggest that the spatial analysis of heavy rainfall illustrate the fundamental

advantage of NEXRAD estimates over rain gauge estimates.

It is clear from the studies reviewed that biases or inaccuracies in measurement are
unavoidable in both rain gauge and radar measurements. Despite the potential
inaccuracies, the use of NEXRAD estimates of precipitation would appear to be
preferable to rain gauge estimates, given the highly variable spatial patterns of summer
precipitation in the study area. Accuracy of radar estimates of precipitation can be
improved by calibration with rain gauge data, as discussed by Collier (1986). For this
reason, the scientist should make use of all available data, both radar and gauge
measurements, when assessing precipitation levels for a region. With continued
improvement in NEXRAD calibration and measurement capabilities, accuracy of
estimates should continue to improve. Accuracy of a radar-based estimate is partially
dependent on the features of the precipitation, with frontal storms more easily measured,
and convective storms much less reliably measured. The scale of analysis should be
considered when considering precipitation data. At the field or farm scale, (smaller than
the 4 km. resolution of NEXRAD data), a network of rain gauges would undoubtedly
provide more accurate estimates of precipitation than NEXRAD. This is impractical for
large—scale implementation however, for obvious reasons. At the regional or statewide

level, as in this study, NEXRAD estimates calibrated to rain gauge data should provide

31

the best estimate of precipitation levels. Any assessment of the accuracy of a modeling
effort should consider the difficulty and potential inaccuracies in measuring precipitation

levels, given their high spatial and temporal variability.

Looking to the future, satellite measurement of precipitation may provide estimates of
environmental variables on a much larger scale. As an example of the use of remote
sensing of climatic variables, Hsu et al. (1996) report on precipitation estimation from
remotely-sensed information using artificial neural networks. They state that with the
current rapid growth in remote-sensing technology, we will soon be able to monitor the
global distribution of rainfall. The system reported by Hsu, et al. however, currently
requires ground-based measurements for system calibration. In two case studies, they
found that it is much more difficult to produce accurate estimates of hourly rainfall than
of monthly accumulated rainfall, due to random estimation errors. They present some
positive results for monthly estimation, but admit that a great deal of work is still required
to develop and strengthen the methods they propose. The spatial scale of rainfall
estimates was .25° latitude by .25° longitude, (approximately 30 km. resolution). Remote
sensing of rainfall from satellites could potentially provide accurate data at larger scales
(30 km.), but this seems to be a technology very much in development and experimental

stages at this time.

Future developments in remote sensing capabilities may also provide direct estimates of

soil moisture that could be used to derive irrigation needs. Ulaby et al. (1996) show good

agreement between estimated soil moisture and measured soil moisture using data

32

collected from both a DC-8 aircraft (AIRSAR instrument) and the space shuttle’s Shuttle
Imaging Radar C (SIR-C) instrument. Data were collected for a watershed in
southwestern Oklahoma. Ulaby et al. (1996) state that it is conceptually possible to
image the terrain with a 30 m spatial resolution using SIR-C. SIR-C is currently an
experimental system, however, so it will likely be many years before technology such as

this could be used in an operational setting.

2.5 Spatial Scale and Regional Modeling

Easterling et al.(l998) examined the relationship between modeled and observed crop
yields as a function of the spatial scale of input data. In this study, the EPIC model
(Williams et al., 1990) was used to simulate crop yields in Missouri, Iowa, Nebraska, and
Kansas using different scales of input weather data. The largest scale used corresponded
to the scale typically used by global circulation models (GCMs). This is a 3 degree or
larger grid cell (approximately 250 km x 350 km). The climate variables were calculated
as averages of all stations within a cell, at the spatial resolution being modeled. County
level yields from the National Agricultural Statistical Service were used to test the model
results. EPIC simulations were performed for various levels of disaggregation of soils

and climate data in order to examine scale effects on modeled and observed yields.

Easterling et al. (1998) found that at the GCM (250 km x 350 km) level, observed yields

explained only 43% of the variation in simulated crop yields over the 1984 — 1992 period.

33

The r2 was improved to .658 when a resolution of 79 km x 104 km was used for input
data. No improvement in goodness of fit was found at higher resolutions (the finest
resolution tested was 47 km. x 62 km.). Their results suggested that the disaggregation
of climate data, not soils data, was the only factor that improved agreement between
simulated and observed yields. This study points out the potential errors in using very
coarse scale input data for modeling purposes. Results of such modeling can certainly be

within reasonable ranges, but correlation with ‘reality’ is potentially low.

Klemes (1983) discussed the issue of conceptualization and scale in hydrology. He
points out that we cannot impose scales on nature, but rather must search for those that
exist and try to understand their interrelationships and patterns. Klemes contends that in
nature, scales of things are not arbitrary, they arise as a function of their material
substance and of the balance between the interacting forces. Scale, in this case, should be
considered both in a spatial and temporal sense. The points made by Klemes (1983) are
important to consider. What is the appropriate scale to model precipitation, variability in
soil moisture, or cropping patterns? Currently, scale is constrained and imposed by the
technology available. For example, the 4 km. resolution of NEXRAD data is the most
detailed available, yet research and experience has shown that summer rainfall especially
can occur at much smaller scales. The time dimension must be considered as well. Four
km. may be a more appropriate scale if the time scale is measured in days or weeks, but
not in hours. Theoretically, spatial correlation should be increased as the time dimension
for measurement is increased. The modeling scale is also restricted by the resolution of

digitized soil data (1:250,000 scale). In this study, scale is dictated by current technology

34

and data availability, but we should always keep in mind the search for appropriate scales
and the implications of our decisions in regards to the choice of scale used in a modeling

environment.

Bergstrom and Graham (1998) discuss the issue of scale in hydrological modeling. They
generally classify hydrological models as either physically or conceptually (empirically)
based. The physically-based model is generally oriented towards detailed, small-scale
modeling of processes, requiring high resolution input data. Expansion to a larger (basin)
scale would require detailed information on the variability in the basin. Bergstrom and
Graham point out that conceptual models generally have more straightforward water

balance or runoff modeling as a goal, and treat a large basin as a sum of smaller ones.

The SWBM used in this study was developed as a functional model with both empirical
and mechanistic elements. The goal of this approach is to provide the most detailed
estimated possible for irrigation water use estimates using generally available data. As
finer resolution input data becomes available, it will be possible to incorporate greater
detail in the modeling effort. Variability and appropriateness of scale over space and time
must be considered as the modeling approach used in this study is further developed or as

new fine—scale data becomes available.

35

2.6 Summary

In multi-disciplinary studies such as this it is not possible to comprehensively review all
literature related to the many components of the study. There is, for example, a vast
amount of literature on each component of the general water balance equation;
precipitation, evaporation, transpiration, and infiltration. An attempt has been made to
review the literature most pertinent to this research project as it applies to the assessment
of the use of different data sources, scales, and interpretation of model results. For the
modeling of a soil water balance, the literature reviewed was directly related to the
SWBM used in this study or provided examples of uses of the SWBM approach in other
settings. The studies by Knox et al. (1996) that mapped the irrigation need for England
and Wales are most closely related to this study in terms of objective and approach. The
literature reviewed pertaining to meteorological variables focused on precipitation and
it’s variability over time and space. This was done because of the importance of
precipitation as a driving variable in the SWBM. Most studies supported the need for
high-resolution data (4 km. or less) to accurately represent precipitation patterns over

space on a daily basis.

A basic premise in modeling should be that the more detailed the model inputs, the more
realistic are the results. Review of the literature relating to spatial scale and model
accuracy (Easterling et al., 1998) supports this premise. The increasing capabilities and
development of GIS and remote sensing technology should only increase our ability to

effectively model or monitor environmental processes. Many of the studies currently

36

using GIS to for large-scale modeling or simulation use correspondingly large-scale
(coarse-resolution) data inputs. This is largely a function of current data availability and
computing capabilities. In the future, data availability and computing capabilities will
increase, and spatially distributed models and simulations will be run using high-
resolution data inputs. This study is a step towards the future, as are many of the current

studies integrating GIS technology and biophysical models.

37

CHAPTER III

MODEL AND DATA DESCRIPTION

This chapter contains 7 sections. Section 3.1 discusses the Soil Water Balance Model
(SWBM) used in this research. Section 3.2 discusses the source and derivation of soil
data used by the SWBM. Section 3.3 discusses the temperature data used in the study.
Section 3.4 discusses the solar radiation data used. Section 3.5 discusses the
NEXRAD precipitation data. Section 3.6 discusses the database of historical weather

data used in the soil simulation runs. Section 3.7 summarizes this chapter.

3.1. The Soil Water Balance Model

The core of the irrigation modeling component of this research is the soil water balance
model (SWBM). References pertaining to the development and validation of the SWBM
were cited in chapter 11 (Ritchie, 1972, Richardson and Ritchie, 1973, Ritchie, 1985, and
Ritchie, 1998). This section contains a summary of the SWBM. A more detailed
summary of the SWBM components is given in Appendix A. Additional details of the

SWBM are given in Ritchie (1972), Ritchie (1985), and Gerakis and Ritchie (1998).

38

The basic soil water balance equation, without respect to changes in vertical distribution

of water, can be written as:

where

dS/dt=P+I—R—E,—Et—D (3.1)

dS/dt = the change in water storage (S) in time period t

P = precipitation

I = irrigation

R' = runoff

E5 = evaporation from the bare soil surface
E = transpiration by plants

D = drainage to the sub-surface soil layers.

Precipitation and irrigation provide the water input to the system. In this research,

precipitation is a direct input (source described below) and irrigation is a model output.

That is, the variable ‘I’ in equation 3.1 is solved for based on a management strategy to

maintain sufficient levels of soil water. Management options for controlling irrigation

amount and frequency are described below. The SWBM is a multiple layer model that

allows calculation of the vertical distribution of soil water as well as the total soil water

content. In this research, 11 soil layers are defined. The depth of each of these layers is

shown in Table 3.1. The change in volumetric soil water content for each layer is

calculated on a daily basis. The routines involved in the soil water balance simulation are

summarized below, with more detail given in Appendix A.

39

Potential evaporation (E0) is calculated as a function of the air temperature during
daylight hours and solar radiation levels. Air temperature during daylight hours (TD) is
approximated by a weighted mean of daily minimum (TMm) and maximum (TMax) air
temperatures (°C) . Solar radiation (SR) is a direct input (source described below), in
units of MJ rn'2 day". Potential soil evaporation (EOS) is a function of the potential
evaporation (EO) and the current leaf area index (LAI). LAI is the ratio of leaf area to
ground area. As LAI increases during the growing season, potential soil evaporation is
lessened'because of the shading and protection from wind offered by the leaf cover. In
this research, LAI values are generated for each crop as a function of time (days after
planting). This is a simplified approach, meant to approximate the typical leaf
development for each crop over the course of a growing season. In a more detailed
simulation, a crop growth model could simulate LAI as a function of growing degree

days.

Root mass and distribution are important parameters for the estimation of the amount of
water taken up by the roots. New root growth (NRG, grams m'2 day"’) is estimated as a
function of solar radiation (SR) and the current leaf area index (LAI). The root length
distribution is calculated for each soil layer as a function of the new root growth, layer
water content and a root water uptake coefficient (RWUCON). Root length distribution
is used later in the calculation of water uptake.

Ponding refers to the accumulation of standing water on the soil surface. The ponding

routine calculates ponding based on the rainfall intensity and a function for hydraulic

4o

conductivity. It calculates the daily amounts of infiltration, runoff, and changes to the
height of the ponded water. The ponding routine also calls a routine for soil compaction
due to rainfall action. The NEXRAD hourly rainfall rates are used in this routine to
determine the maximum rainfall intensity during the precipitation event. This is the only
routine in which hourly values are used, in all others the time step is daily. The result of
the ponding routine is a distribution of any incoming precipitation to runoff, infiltration,
or additional ponding. If irrigation occurs for the day, the irrigation amount is added to

infiltration.

The downward movement of water is calculated next. Two major factors affect the
downward movement of water: the hydraulic force (mostly gravity for downward flow)
and the hydraulic conductivity of the soil (Brady, 1990). Hydraulic conductivity is a
basic property of each soil layer (see derivation below). Based on the soil water content
at each layer in relation to the layers drained upper limit (DUL), saturation level (SAT),
and hydraulic conductivity (KsMatrix), the amount of water percolating to each soil layer

is calculated, with upper layers providing water to lower layers.

Water moves upward in a soil as a function of the same two factors controlling the
downward flow described above (Brady, 1990). Evaporation at the soil surface dries out
soil in the top layers, creating a potential gradient between the dry and moist soil areas.
Water will move in the direction of the drier soil, in this case upwards. The SWBM uses
the available energy for soil evaporation (EOS) to estimate the amount of water moving

upward and the amount evaporated from the surface layer.

41

The amount of water taken up by the roots is a function of the potential plant evaporation
and transpiration for the day, bounded by potential evaporation minus soil evaporation.
Potential plant evaporation (EOp) is a function of the potential evaporation and the leaf
area index. Water is removed from each soil layer as a function of the soil water content,
the layer lower limit (LL), the root length distribution of the layer, and the root uptake
coefficient (RWUCON). The total amount of water taken up by roots is equal to EOp if
the plant can transpire at potential levels. Otherwise the total amount of water taken up

by the roots is equal to E0 — EOS, the potential evaporation minus soil evaporation.

The decision of whether or not to apply irrigation water is based on the ratio of

extractable soil water to the potential extractable soil water, calculated as:

ESW = z (swi — LL,)* Di for i = 1 n. (3.2)

PESW = X (DUL; — LLi)* Di for i= 1 11.

D1 = ESW/PSW

Where
SWi = Current volumetric soil water content, layer i
DUL, = Drained upper limit, volumetric soil water, layer i
LL; = Lower limit, volumetric soil water, layer i
D1 = Drought Index
n = the total number of layers tested (see below).

42

The current soil water content is the current value for the day being simulated. The
drained upper limit (DUL) is the highest field-measured water content of a soil after
thorough wetting and draining until drainage becomes negligible. The DUL corresponds
to water content at ‘field capacity’. The lower limit is the lowest, field-measured
volumetric water content of a soil after plants stop extracting water due to premature
death or dormancy as a result of water deficit. LL corresponds to the ‘permanent wilting

point’ commonly referred to in the literature.

The number of layers tested (n) in Equation 3.2 is a variable that can be set by the
operator. The default value is 6 (depth = 57 cm.). If the value for root length distribution
is greater than .05, however, that layer is also included. The Drought Index (DI) is the
variable used to test for the irrigation threshold value (Tc). The default value for Tc is .5.
If D1 is less than or equal to .5, an irrigation event is triggered. The irrigation amount is a
specified value that differs by crop, see Appendix B for a complete listing. A typical
value is 2.5 cm. for each irrigation event. Both the threshold value (Tc) and the irrigation
amount are management parameters. The sensitivity of the SWBM to changes in these

management parameters was tested as part of this research and is discussed in chapter VI.

To summarize the SWBM, it is a model that has been thoroughly tested and validated in a
research setting over the past 25 years. It has also been used in an operational setting to
provide estimates of soil moisturein the Midwestern United States (Kunkel, 1990). The
use of readily available data for soil and climatic data inputs enhances it’s potential for

use in a variety of settings. Some soil parameters are not obtainable directly from readily

43

available sources, but their derivation is well documented. ‘Calibration’ for a particular

crop or region should be possible by adjustment of the irrigation management parameters.

In the implementation of the SWBM for this research, simplified but realistic
assumptions pertaining to leaf area development (LAI) and root growth have been used.
In future development, it is possible that these components could be simulated using a
crop growth model. In view of the relatively coarse resolution of other data inputs (such
as soils and precipitation) it is probably not necessary to incorporate detailed crop growth
models at this time. This could be done when the resolution of other input data becomes
finer and there is a need to estimate water use in finer scales (temporally and spatially)

than is currently possible.

3.2 Temperature Data

Temperature data consisted of daily minimum and maximum temperature (degrees
centigrade) obtained from the Michigan State University Agricultural Climatology
Division. A kriging procedure was used to create a continuous surface of temperature for
the state of Michigan. Resolution of the temperature data was 20 km. in 1996 and
approximately 16 km in 1997 and 1998. There were 271 cells in the 1996 grid (Figure

3.1 ), and 550 cells in the 1997 grid (Figure 3.2)

44

3.3 Solar Radiation Data

Solar radiation data for the State of Michigan was also obtained from the Michigan State
University Agricultural Climatology Division. Daily solar radiation estimates were
provided in units of MJ/Mz. The spatial resolution of the solar radiation data matched the

temperature data; 20 km. in 1996 and 16 km. in 1997 (Figures 3.1 and 3.2).

 

 

 

 

 

Figure 3.1. 20 kilometer temperature and solar radiation cells for the state of Michigan
used in 1996. Note: Isle Royale is excluded on this and other maps in this text.

45

 

 

 

 

 

Figure 3.2. 16 kilometer temperature and solar radiation cells for the state of Michigan
used in 1997.

3.4. NEXRAD Precipitation Data

Hourly precipitation data at a resolution of 4 km. from the NEXRAD (Next Generation
Weather Radar) system (Hudlow, 1988; Hudlow et al. 1991) was obtained from the
National Weather Service. There are a total of 7,446 NEXRAD cells covering the land

area of Michigan (Figure 3.3).

46

 

 

 

 

 

Figure 3.3. 4 kilometer NEXRAD cells for the state of Michigan (Isle Royale excluded).

Significant processing was involved in converting the raw NEXRAD data to a format
used in the integrated modeling environment (IME) in which the SWBM was
implemented. These processing steps involved processing of the data on a UNIX system,
which created ASCII files that were read into an MS Access database on the MS

Windows platform.

47

3.5 Soil Data

The State Soil Geographic (STATSGO) database (USDA, 1994) provided the statewide
soils data for this research. The STATSGO soil association maps are compiled from
generalized soil survey maps at 1:250,000 scale. A soil association, or Map Unit,
represents mixtures or pattern of soils that occur together on the landscape. Individual
soils in the Map Unit are referred to as Components of the Map Unit. The Component
soils may be similar to each other or they may be very different. A STATSGO Map Unit
has a minimum resolution of 625 ha (1544 acres) and, for Michigan, can contain from 2
to 21 soil series. The average number of soil series in a Map Unit is 13. The percentage
of each Component in the Map Unit is known, but the spatial distribution of each soil
within each Map Unit is not known. A total of 468 individual soils are identified in the

STATSGO data for Michigan, organized into 190 distinct Map Units (Figure 3.4).

48

 

 

 

 

 

Figure 3.4. STATSGO Map Units for Michigan. 190 unique map units are defined for
2,235 polygons. Colors on this map identify unique soil map unit areas. Source: State
Soil Geographic Database (USDA, 1994)

STATSGO Map Units are of relatively coarse resolution compared to the size of an
irrigated field. The USDA cautions that the level of mapping of the STATSGO data is
designed for broad planning and management uses covering state, regional, or multi-state
areas. A more detailed database called the Soil Survey Geographic (SSURGO) database
is available from the USDA. SSURGO mapping scales generally range from 1:12,000 to
1:63.300, and each Map Unit contains one to three soil series. SSURGO data are only
available for selected counties however. At the time of this writing, only 7 of the 83

Michigan counties were available. Because of the unavailability of the SSURGO

49

database on a statewide basis, the STATSGO database was used for this research. The
minimum size of the STATSGO Map Unit (625 ha) corresponds to an area of about 6.25
km.2. The area of the NEXRAD cell (discussed below) is approximately 4 kmz, slightly
smaller than the smallest STATSGO soil map unit. Most soil map units are much larger
than 6.25 km.2, however, because of the homogeneity of soils and the groupings of many

soil series in to a single soil map unit.

Soil parameters required by the SWBM were not in all cases directly available from the
STATSGO data. It was necessary to derive some parameters based on fields available in
the STATSGO database. Conversion of STATSGO soil data to SWBM soil data was a
two step process. In step 1, STATSGO layers were converted to the 11 layers used by
the SWBM (Table 3.1). STATSGO defines from 1 to 6 layers for each soil series, with
depth generally defined to 152.5 cm (60 inches). Some STATSGO soils are defined to a
depth of 251 cm. (99 inches). To create the 11 SWBM layers, values were weighted by
percent contribution in those cases where an SWBM layer contained values from more
than one STATSGO layer. If an SWBM layer corresponded to only one STATSGO
layer, values for the STATSGO layer were used without weighting. The result of this
step was a standardized set of soil layers for each soil series in the Map Unit,

corresponding to the layers in Table 3.1.

In step two of the STATSGO processing, a Composite soil for each STATSGO map unit

was calculated as the average of all Component soils in the Map Unit, weighted by area

50

(percent composition). Table 3.2 lists the soil parameters that were derived from the

STATSGO database and the formulas used in their derivation.

Table 3.1.

Soil layers and depth (cm.) used in the SWBM. A = top depth of layer, Z =

bottom depth of layer.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

, Layer ' A,(Cm.) 2 (cm)
1 0 2
2 2 7
3 7 15
4 15 26
5 26 40
6 40 57
7 57 77
8 77 100
9 100 125
10 125 150
11 150 175
Table 3.2 Soil parameters derived from the STATSGO database.
Field Formula or Value
30 (bdh + bdl)/2
0C (omh + oml)/2
Clay (clayh + clayl)/2
Sieve10 (no10h + no10l)/2
Sieve200 (n0200h + n0200|)l2
Sand ((Sieve10 - Sieve200) / Sieve10) * 100
Inch10 (inch10h + inch10l) / 2
lnch3 (inch3h + inch3l) / 2
Stones lnch10 + lnch3
Soith (Sieve10 * (100 - Stones)) / 100
BDM 2.65
Mfactor (Soith/ BD) / ((Soith / 80) + (1 - Soith) / BDM)
LL .015 + (.0037 * Clay) * BD
PESW .138 - (1 .6E-11 * (Sand - 25)5
SAT (1 - (BD/ BDM)) * .92
DUL IF (SAT - (LL + PESW)) > .01, DUL = LL + PESW ELSE DUL = SAT - .01
LL LL * Mfactor

 

 

 

51

 

 

Table 3.2 (Cont’d)

 

 

 

 

 

 

 

 

 

Field ‘ ' Formula or Value ,

DUL DUL * Mfactor

SAT SAT * Mfactor

AD .44 * DUL2

FIowUCO .63 / 22 * .495 * e(2.804 — 10.76 * DUL) if layer = 1 THEN adjust by
multiplyingby .82 — 4.7 * (.45 — DUL2)

Silt 100 — (Clay + Sand + Stone)

KsMatrix See Table 3.3.

SettledSWCN (3 — (Sand * 015)) * KsMatrix

SWCN SettledSWCN

 

 

 

 

Table 3.3. Values for KsMatrix as a function of Clay and Sand %

 

Sand ‘

. . . °/o . . j . .

0 20 30 40 50 60 70 80 90 100
O 12 24 72 84 90 96 96 96 106 110
10 9.6 19.2 60 72 78 84 90 96 106
Clay% ‘ 20 7.2 14.4 48 60 67.2 72 84 96
30 4.8 9.6 36 48 56.4 60 72
40 3.6 4.8 6 7.2 8.4 9.6
50 2.4 3.12 3.12 3.12 3.12

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

The result of the soil data processing was a standardized Composite soil for each soil
Map Unit, based on the Component soils defined for the Map Unit, with a standardized
set of layers (Table 3.1). The Map Unit Composite soil is used, during the model run, to
calculate soil characteristics at the level of the NEXRAD cell. Some NEXRAD cells
contain only one soil Map Unit, others contain several. Figure 3.5 shows the STATSGO

map units for Barry County with the NEXRAD cells superimposed. The NEXRAD cells

52

are recognized by their rectangular pattern. Because the SWBM is run at the resolution
of the NEXRAD cell, soil parameters must be provided at the same level. Visual
inspection of Figure 3.5 shows that most NEXRAD cells in this area contain 3 or 4
STATSGO map units. The soil parameters for each NEXRAD cell is calculated as the

average (weighted by area) of the Map Units contained within the NEXRAD cell.

Calculation of the NEXRAD soil parameters is done dynamically during the model run
for two reasons. First, to minimize storage space for the soils database. It is not
necessary to store and maintain soil characteristics for each of the 7,446 NEXRAD cells
since they are composed of only 190 unique Map Unit Composite soils. Disk space
requirements are lessened by storing records for the 190 Map Units rather than for each
NEXRAD cell. Second, data for the STATSGO map units may be periodically updated,
or formulas for the derived parameters (discussed above) may be modified. When this
occurs, the operator only needs to recalculate the Map Unit Composite soil table (190
records). Modifications to the NEXRAD soils are made during the next model run, when
modified Map Unit Composite soils are used to dynamically build a set of soil parameters

at the NEXRAD cell level.

53

 

 

 

 

 

 

 

 

 

 

Figure 3.5. STATSGO Map Units for Barry County, Michigan with 4 km. NEXRAD
grid cells superimposed.

3.6 Other GIS Data Layers

The WEQ must report water use estimates at the county and watershed level on an
annual basis. Data layers in GIS format corresponding to these two reporting scales were
obtained. Because the borders of NEXRAD cells do not coincide with county or
watershed borders, it is necessary to associate a county and watershed with each irrigator
record. That is, a NEXRAD cell ID is not sufficient to permit aggregation to the county

and watershed levels because more than one county or watershed may be contained in a

particular NEXRAD cell. For reference, counties and major watersheds of Michigan are
shown in Figures 3.6 and 3.7. The watersheds used are the United States Geological

Survey (USGS) HUC 8 digit watersheds. There are 83 counties and 65 major watersheds

in the state.

 

 

 

   
 
    

WRON

ruscoul 3»;ng

   
 
 
 

SAGI MW

    
 
 
   
 
   
 
   
 

KENT
IONIA

MEGAN BARRY EATON

NREN

CASS

 

 

Figure 3.6 Counties of Michigan.

55

 

 

 

 

 

 

 

Figure 3.7. Watersheds of Michigan (USGS 8 digit).

3.7 Irrigation in Michigan

Bedell and Van Til (1977) reported that irrigation was practiced in every county in
Michigan in 1977. In that year, more than 2,800 irrigators applied an estimated
2,852,000 acre-inches of water on approximately 325,000 acres. 85.5% of the total
irrigated acreage was agricultural. Bedell and Van Til (1977) compared the 1977
numbers to a 1970 report and concluded that the total irrigated acreage increased by

217% between 1970 and 1977. In addition, the total quantity of water applied increased

56

by 268% and the number of irrigators increased by 8%. In relation to other water uses,
Bedell and Van Til (197 7) report that irrigation represented 1.4% of the total water used
in the state in 1977 and accounted for 33.3% of the water withdrawn and 66.8% of the
water diverted during the summer months from inland surface and ground-water sources.
Sweat and Van Til (1987) reported that in 1985 agriculture accounted for 2.1% of all
water withdrawals and 39.3% of total consumptive use of water. Thermoelectric power
plants withdraw the largest amounts of water in the state (73.5% in 1985) but account for
less consumptive loss than irrigation. The 1992 Federal Agricultural Census showed

3,331 irrigators (irrigating more than 10 acres) and a total of 365,578 acres.

Center-pivot sprinkler irrigation is the predominant method of irrigation in Michigan
(Sweat and Van Til, 1987). Although irrigation is practiced in all counties, a large
portion occurs in the southwestern part of the state. Major irrigated crops include com,
soybeans, potatoes, and vegetables. In the limited survey done by the MDA in 1997,
corn accounted for 43% of the reported irrigated acreage, with soybeans and potatoes

each approximately 10% of the total.

3.8 Summary

This chapter has provided a summary of the calculation steps of the SWBM, the source
and derivation of data inputs, and irrigation in Michigan. The SWBM method is a data-
intensive approach, especially when implemented over large spatial areas such as the

state of Michigan. GIS technology was used to prepare and view the spatially related

57

data, but is not necessary for the operation of the SWBM, which is implemented as a
Microsoft Windows application. In this research every attempt has been made to use the
most detailed data available. As data of higher resolution becomes available, this data

can be incorporated in the modeling effort.

58

CHAPTER IV

SOIL DATA ANALYSIS

This chapter contains 5 sections. Section 4.1 discusses the objectives of the soil data
analysis. Section 4.2 discusses the methodology used. Section 4.3 presents the results
and analysis. Section 4.4 presents further analysis pertaining to mapping irrigation need
and characteristics of outlier soils. Section 4.5 summarizes this chapter and discusses

conclusions from this analysis.

4.1 Objectives

In the ideal modeling situation, soil parameters would be defined on a site-specific basis.
In this study, however, the location of irrigators is known only at a resolution of 4 km
(the NEXRAD grid cell). Because of this, it is necessary to estimate soil characteristics
by creating a composite soil for the 4 km. cell, as discussed in the previous chapter. For
this research, the STATSGO soil database (described in chapter 111) provided the base
soil data inputs. For Michigan, there are 190 map units defined, each map unit consisting
of from 2 to 21 soil series. The composite percentage of each series in the map unit is

given, indicating the percentage of the map unit that is composed of the particular soil

59

series. Because the SWBM is run at the resolution of the NEXRAD cell (4 km), it is
necessary to calculate a composite soil for each NEXRAD cell based on the STATSGO
data. The composite soil data are calculated dynamically during the model run as the
average (weighted by area) of the soil map units within the NEXRAD cell. The
composite NEXRAD soil is based on a composite map unit soil that was calculated from

weighted averages at the STATSGO map unit level (as discussed in chapter 111).

The purpose of the soil data analysis is to determine the sensitivity of the SWBM, in
terms of irrigation amounts, to changes in soil characteristics at the STATSGO map unit
level. The composite soil generated for the 4 km. NEXRAD cell does not represent a
specific soil series, rather it is a weighted combination of soil map units, which
themselves are a weighted combination of soil series mapped at a relatively coarse
(1:250,000) scale. It is relevant to question whether a simpler soil database can be used
without loss of accuracy as compared to the use of NEXRAD level soil parameters. That
is, if irrigation amounts are similar (not different) for many of the soil map units, these
map units could be combined and treated as a single soil by the SWBM. If however,
there is high variability between map units, then the calculation and use of composite
soils, as is currently done, is justified. The result of the analysis will be used to assess the
potential for using simplified soil data inputs for the SWBM. The results of this analysis
also have implications for use of the SWBM in areas where detailed soil data is not

available.

60

4.2. Methodology

To test for differences in irrigation estimates by soil map unit, a matched-pairs
experiment was run. 30 years (1961 — 1990) of daily weather data from the Lansing,
Michigan station was used to provide a representative range of climatic conditions. Four
crops (corn, soybeans, strawberries, and sod) were simulated to assess differences
between crop type. The SWBM was run for each of the 190 soil map units, for each of
the 4 crops, for each of the 30 years of historical data. Total irrigation amount was

recorded for each year/crop simulated.

The 30 year data set from Lansing, Michigan provided precipitation estimates on a daily
basis, not hourly as with the NEXRAD data set. Before running the simulations, it was
necessary to convert the SWBM to run using daily precipitation amounts. The ponding
sub-routine was the only portion of the model affected by this difference. Rather than
change the model code, the daily data were converted to hourly during the model run

according to the following methodology:

A threshold value of .254 cm./hour was used to distribute the daily rainfall to hourly. If
total rainfall for the day equaled or exceeded 6.1 (.254 * 24) cm, rainfall was distributed

as:

Precipi = Total/24, i = 1 .. 24 (4.1)

where

61

Total = total precipitation for the day

Precip; = hourly precipitation for i = l to 24
If total rainfall for the day was less than 6.1 cm., rainfall was distributed as
Precip; = Total/NumHours, i = 1 .. NumHours (4.2)

where
Total = total precipitation for the day
NumHours = Total/.254

Precipi = hourly precipitation

The irrigation model was tested for differences using hourly vs. daily precipitation. The
model was run using the 1997 MDA reported irrigators using 1) Hourly and2) Daily
precipitation amounts. Results using hourly precipitation showed a total application of
1,137,534 acre-inches of water. Using daily precipitation records, the total was
1,138,883 acre-inches of water. There was slightly more irrigation (12%) using daily
precipitation records but the difference is an insignificant amount. These results show
that SWBM irrigation estimates obtained using daily precipitation will be nearly identical

to those obtained using hourly precipitation.

62

The SWBM was run for each map unit, for each of 4 crops, using the Lansing 30 year
data set as weather input. It was assumed that the daily weather (temperature,
precipitation, and solar radiation) was experienced uniformly across the entire state.
Differences in model results for each crop were therefore due only to differences between
soil map units. The result of this simulation run were, for each of the 4 crops, a set of
190 records (one for each map unit), with each map unit record consisting of a series of
30 irrigation totals (one for each year). Each irrigation total is the total amount of

irrigation applied for each crop for year 11 (n = l to 30).

4.3. Results and Analysis

Results of the matched-pairs simulation were tested for significant differences in

irrigation amounts between pairs of soil map units. To test for significant differences,

the mean difference of the paired irrigation amounts was tested. The number of

combinations to be tested was potentially a very large number, since the number of

combinations of n objects taken r a time is:

n!/r!(n-r)! (4.3)

In this case, n = 190 (soil map units) and r = 2. Equation 5.1 evaluates to 17,955.

63

Not all of the possible 17,955 combinations needed to be tested, however. If map units
were found to be similar, they were grouped and no further testing was necessary. For
example, if map unit 3 was found to be the same as map unit 1, then no further
combinations using map unit 3 needed to be tested, since map unit 3 was ‘assigned’ to the
map unit 1 group. An example is presented below (Example 4.1) to illustrate the
methodology used to test for differences in irrigation counts between map units. The test
involved determining if the mean difference of the paired observations for each map unit

was significantly different than zero.

Tests for differences between pairs of all data series were conducted according to the
methodology described in Example 4.1. Groups of map units with no significant
difference were created. In testing map unit A vs. map unit B, if map unit B was found to
not differ from A, then B was added to the “A” group and excluded from further testing.
Results are summarized in Table 4.1, which shows the number of groups and number of

soil map units in each group for each of the four crops.

64

Example 4—1. Testing the Paired Differences.

 

The mean difference for each year is tested between two map units soil A and B. Assume

that the results show the following irrigation amounts:

Year _A _B Difference

l 73 68 5
2 85 79 6
3 64 60 4
4 90 84 6
5 69 65 4

The mean difference, denoted up, is tested. The test statistic is

1: (X0 - lib) / (SD / 3911010)) (4.4)
where

x0 = the sample mean

sD = the sample standard deviation

in) = the number of paired observation

This test requires the assumption that the differences are normally distributed. The test
statistic has nD- 1 degrees of freedom. The test is

H0: up = 0

H A: no <> 0

The rejection region for t025, 4 (from the Student t distribution) is |t| > 2.776
For this example,
t=(5 —0)/(l /sqrt(5)) = 11.18
11.18 > t
Conclusion: Reject Ho.
There is sufficient evidence to conclude that a difference exists in the mean irrigation

count of soil map unit A and soil map unit B.

 

65

Table 4.1. Results of the test for paired differences (by map unit count).

 

‘-Count'Of Soil ‘ ‘Units ' 4-
Com Strawberries
133 127 129
13 16 35
13 13 1

12

 

   

\J
A—LN
00031:.

01
b

0
4
3
2
1
1
1
1
1
1
1

A—l—lA—KNNNNNOOO)
AAAAAANNOJA
_l—A—‘L-A—ANNNQ

 

Table 4.1 is interpreted as follows. For com, 133 of the 190 soil map units showed no
significant differences in irrigation amount over the 30 year simulation period. Of the
soil map units remaining, they could be aggregated to groups of 133, l3, l3, 7, etc. For
corn, 5 soil map units showed results that could not be grouped with any other soil map
unit. In terms of number and distribution of groupings, results were similar for the other
crop types. These results indicate that, for a given crop type, the majority of soil map

units in the state will not show significant differences in irrigation amount, if each map

66

unit were to experience the same weather conditions during the growing season.
However, there are exceptions in every case. Some soil map units show different

irrigation totals and could not be grouped with any other soil map unit.

The results from Table 4.1 should be interpreted by area as well. Table 4.2 shows the
breakdown of soil groups as a percentage of land area in Michigan. For corn, 80% of the
land area in Michigan is covered by the first soil group. The land area considered
includes all land cover types. If forest and other non-tillable land were excluded, the
percentage would be higher. The corresponding land coverage is 69%, 63%, and 58% for
soybeans, strawberries, and sod, respectively. If the top five soil groups are considered,
then the percentage of land coverage is over 90% for all crops simulated (the range is
90.7% to 95.0%). Based on this analysis, it can be concluded that the majority of soil
map units in Michigan (and correspondingly the majority of land area) will give similar
(non-different) results in terms of total irrigation for a growing season when simulated for
a number of seasons for a particular crop. This analysis considers only total irrigation
over a season, not the timing of irrigation, which could potentially differ among similar

soil map units.

Results of the map unit grouping should also be viewed in their spatial context. This
gives an indication of the differences in location of the major soil groups by crop type.
Figures 4.1 — 4.4 show the spatial patterns of the grouped map units for each for each of
the crops simulated. For reference, the county boundaries for Michigan are shown on the

maps. Note that many of the grouped map units cover areas larger than one county.

67

Review of Figures 4.1 — 4.4 indicate the differences in groupings by crop type. That is,
the spatial pattern of soil groupings is not consistent across crops. The same two pairs of
soil map units may show similarity for one crop, but not for another. This is an important

result in terms of the potential for creating major soil groups.

Table 4.2. Results of the test for paired differences (by percentage of land area).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

—- Percentage of Land Area -

Soil Group Com Soybeans Strawberries Sod
1 80.1 69.4 63.0 58.1
2 5.1 12.2 26.2 16.0
3 3.2 5.7 3.9 9.5
4 2.3 3.9 1.9 8.3
5 2.1 0.7 1.3 1.9
6 0.4 1.8 0.7 1.4
7 0.7 1.8 0.3 0.7
8 1.2 0.7 0.1 1.2
9 0.1 1.3 0.2 1.3
10 1.3 0.1 0.7 0.3
11 0.6 0.7 1.1 0.1
12 0.5 1.1 0.1 0
13 0.1 0.1 0.4 0.7
14 0.7 0.2 0.4

15 0.5 0.4

16 1.1

Total 100 100 100 100

 

 

 

 

 

 

 

Based on these results, 4 major soil groups could be created to capture over 90% of the
variability in irrigation amount, but the soil characteristics for each major group and the
spatial distribution and area for each group would differ by crop type. There are over 30
different crops potentially irrigated in Michigan. ‘Simplification’ of the soil data would

potentially require the creation of unique soil groups for each of the 30+ crops. each with

68

four or five soil groupings. The variation of irrigation counts between crop types
essentially makes the creation of major soil groups impractical for this application.
Information would be lost in attempting to create four or five major soil groups that apply
to all crops. The results in Figures 4.1 — 4.4 provide useful information beyond viewing
the spatial distribution of similar soil groups on a statewide basis. Some general patterns
can be seen. On a regional basis, the soils in southeast Michigan show more variability
than soils in other parts of the state. With the exception of strawberries, soils in Huron
County (the thumb area) show differences with respect to other soil groups. On a county
basis, some counties show little or no variation in irrigation counts. This would indicate

the similarity of soils in the county from the standpoint of irrigation requirements.

69

 

 

 

 

 

 

 

 

 
   

 

:1 Counties
Soil Group

1
-3
-7

 

 

 

 

    

|:] Counties
Soil Group
- 1

3
_12

 

 

 

 

 

Figure 4.2 Soil groups with non-different irrigation amount for soybeans.

70

 

    
 

:1 Counties
Soil Group
- 1

3
-6

 

 

 

 

 

 

 

 

 

 

 

 

ECounties I 2 ' 3 -
Soil Group - _f: A
- 1 .
3 .
-6 x'» ,t- . L
o , Jr

 

 

 

 

Figure 4.4 Soil groups with non-different irrigation amount for sod.

71

4.4 Outlier Soil Analysis

The results of the 30 year simulations were used to further investigate the potential
irrigation differences between soil map units. For each year and crop type, the mean
irrigation count, mean irrigation amount, and standard deviation was calculated. The
results of each soil map unit were compared to the mean for the year by calculating the
deviation from the mean count, deviation from the mean amount, and deviation from the
standard deviation (in terms of number or fraction of SD). This provided a relative
measure for each map unit compared to the “average” irrigation requirement over the 30
year simulation period. The mean standard deviation was calculated to identify soil map
units that differed significantly from other soils (termed ‘outlier’ soils). The results for
corn are shown in Figure 4.5. Soils showing standard deviations of greater than +/- one
from the mean are shown. The sandy soils of Huron County are notable in their
increased irrigation requirement compared to the rest of the state. The mean standard
deviations for all map units ranged from a minimum of —4.633 to a maximum of +1 .936.
The lower value of —4.633 indicated that a few soil map units had significantly lower
irrigation amounts than the majority of the soil map units. If the distribution of irrigation
amount by soil map unit is considered to be normal, or nearly so, then ‘outlier’ soils can
be identified as those that are less than or greater than 1.96 standard deviations (SD) from
the mean. In the case of corn, no soils are outliers on the positive side, but 8 map units
showed negative differences of less than -1 .96 SD. A similar analysis was done for the
other 3 crops simulated. Results are tabulated in Table 4.3. Soil map units that were less

than -1 .96 SD from the mean were consistent across crop type, although strawberries and

72

sod included additional map units than corn and soybeans. In total, there were 11 map
units that showed standard deviation less than -1.96 from the mean. The 8 common map
units were analyzed for common characteristics that would indicate a physical basis for

decreased irrigation need.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

[:1 Comties
Standard Deviation
- < - 3
- ~2.99 - -2
- -1.99 - -1

-1 - 1

1 - 2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2» l

 

 

 

 

 

 

Figure 4.5. Standard deviation groupings for corn computed from the 30 year simulation
runs showing soil areas with irrigation +/- one or more standard deviations from the mean
irrigation amount.

73

Table 4.3. Soil map units with irrigation counts less than -1.96 SD from the mean, by

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

crop type.
- Outlier Soil Map Units by Crop --
Corn Soybeans Strawberries Sod
M1008 M1008 M1008 M1008
M1009 M1009 M1031 M1031
M1097 M1031 M1019 M1009
M1176 M1097 M1009 M1019
M1031 M1019 M1097 M1055
M1019 M1176 M1055 M1097
M1055 M1055 M1176 M1006
M1006 M1006 M1006 M1176
M1184 M1185 M1185
M1185 M1184 M1184
M1096
Count 8 10 10 11

 

 

 

 

 

 

 

Soil map unit ‘M1008’ showed the highest deviation from the mean for all crops. The
map unit name for ‘M1008’ is LENAWEE-TOLEDO-FULTON. This map unit is
composed of 14 soil series, as shown in Table 4.4. The dominant soil in this map unit
(LENAWEE) has a surface texture classified as ‘clay’, and other major soil components
are either silty clays or clays. The lower infiltration rates of a clay soil should result in
greater water retention and lower irrigation need, and the model results are consistent
with these expectations. For five of the eight common map units, the surface texture of
the predominant soil is classified as either clay or silty-clay. For the remaining three,
‘M1055’ is composed of both silt and silty-clay soils, M1006 is predominately loam soil

with 5% muck, and M1076 is predominately silty-loam with 10% muck. The presence of

74

muck soils in a soil map unit should decrease irrigation need estimates, due to the higher

organic matter, higher water holding capacity, and poor drainage of these soils.

Table 4.4. Component soils of soil map unit ‘M1008’ showing the name, ID, percentage,
and surface texture of each component.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

COMPNAME SSID COMPPCT SURFTEX
LENAWEE M10002 45 CL
FULTON OH0080 12 SICL
TOLEDO OH0081 12 SICL
PERT M10197 8 CL
DEL REY |L0022 7 SICL
CORUNNA M10088 5 SL
GRANBY M10029 2 LFS
PEWAMO M10042 2 CL
LENAWEE M10314 2 CL
OAKVILLE M10349 1 FS
OAKVILLE M10038 1 FS
PIPESTONE M10257 1 S
SELFRIDGE M10081 1 LS
ARKONA M10264 1 LS

 

 

 

 

 

 

Examination of the textural classification of the surface layer of the eight outlier soils
showed that they generally are higher in clay, with correspondingly less sand than other
soils. The mean clay percentage for all 190 soils at the surface layer is 10.6%. Table 4.6
shows the breakdown by major textural class of the eight outlier soils. With the
exception of M1097, all have clay levels of 20% or higher. M1097 appears to be an
exception, but further examination revealed high clay content (> 33%) starting at a depth

of 26 cm. This illustrates the need to examine soil characteristics by depth when

75

evaluating soils for irrigation requirements. As seen here, the surface layer will generally

indicate potential irrigation requirements, but not in all cases.

Table 4.5. Surface layer textural class composition (percentage) of the eight outlier soil
map units (Other = larger material such as stones and pebbles).

 

-— Textural Class ..

 

 

 

 

 

 

 

 

 

 

Map Unit Sand Silt Clay Other
M1008 25.9 45.4 28.5 0.2
M1009 15.4 58.6 26 0
M1097 58.4 27.1 12 2.5
M1176 20.6 59.3 19.5 0.6
M1031 12.8 59.5 26 1.7
M1019 23.8 44.6 29.5 2.1
M1055 15.9 56.8 25.5 1.8
M1006 13 60.1 24 2.9

 

 

 

 

 

 

 

Analysis of the soil map units with consistently higher irrigation need was also done.
Five soil map units that showed higher irrigation need for each of the four crops
simulated were analyzed (M1130, M1173, M1180, M1189, and M1208. The surface
texture of the major component of M1130 is defined as a somewhat-poorly or poorly
drained ‘cobbly-loam’. M1173 is predominately a silt-loam (moderately well-drained)
with 10% sand, M1180 is a well drained silt-loam, M1189 is silty-clay and loam (well-

drained), and M1208 is predominately silt (well-drained).

The STATSGO soil drainage class of the predominant soil is ‘well-drained’ or
‘moderately well-drained’ for four of the five soils analyzed. For these cases, the model

results are consistent with the general STATSGO classification of soil texture. Based on

76

the STATSGO soil drainage classification, the presence of M1130 (classified as
somewhat-poorly drained) as a high irrigation need soil was inconsistent with
expectations. The amount of coarser soil material (cobbles) in M1130 was very high
(19.2%) as shown in Table 4.6. If sand % is calculated excluding the larger fragments,
then the effective percentage is 61%, closer to the sand percentage of other high irrigation
soils (M1173 and M1180, for example), which would explain the model estimation of
higher irrigation need. Whether or not this correlates with reality is subject to further

investigation.

The average sand percentage at the surface layer for all soils is 56.2%. Three of the five
“high-irrigation” soils (Table 4.6) are below the average. It is evident from this that the
sand % at the surface layer cannot be used, by itself, as an indicator of potential irrigation
need. In addition, STATSGO classification of soil drainage and model results may not

necessarily coincide in all cases, as demonstrated by soil M1130.

Table 4.6. Surface layer textural class composition (percentage) of five soil map units
with high irrigation need. (Other = larger material such as stones and pebbles).

 

- Textural Class -

 

 

 

 

 

 

 

Map Unit Sand Silt Clay Other
M1130 49.3 19.5 12 19.2
M1173 61.2 26.2 8.5 4.1
M1180 62.1 25.2 8.5 4.2
M1189 29.5 49.1 19.5 1.9
M1208 28.5 50.8 16 4.7

 

 

 

 

 

 

 

77

To summarize the outlier analysis, it can be stated that the model results are generally
consistent with a priori expectations of irrigation requirements based on general soil
descriptions. That is, soil map units showing lower irrigation needs are generally high
clay soils with few or no coarser fragments (stones and pebbles). Those soils showing
higher irrigation needs are less identifiable based on soil textural classification, although
the clay % is generally lower than other soils. It is not possible to assess the potential
irrigation need of a soil map unit based on the texture at the surface layer of the
predominant component. Changes in soil characteristics with depth play a large role in
determining irrigation need. It is also shown that model results may be inconsistent with

the STATSGO soil drainage classification (M1130) in some cases.

This discussion is based on the general premise that the composition of a soil by major
textural class has some correlation with the soil’s potential irrigation requirements. The
use of soil textural classes to estimate porosity has been a subject of discussion in the soil
science literature. Saxton et a1. (1986) present equations for estimating hydraulic
conductivity as a function of sand, silt, and clay percentages. Schuh and Bauder (1986)
found that sand to silt ratio was a better predictor of hydraulic conductivity than soil
particle size percentages alone. Helalia (1993), however, reported that the textural
variables (clay %, silt %, and clay pus silt) showed weaker correlation with the
infiltration rate than the structural variables (total porosity, effective porosity, and bulk

density).

78

4.5 Composite Soil Analysis

The results of the soil map unit analysis discussed above indicate that the majority of soil
map units show insignificant differences in terms of estimated irrigation amounts. This
lends support to the use of a composite soil at the 4 km. NEXRAD cell level in the
modeling process. Since each irrigator is only identified at the 4 km. cell resolution, the
soil series under irrigation could be any soil series within the 4 km. area. To further
examine the potential range of irrigation estimates by soil series, an additional simulation
was run using the actual 1997 reported irrigated acreage. In this simulation, the SWBM
was run for each soil series in the NEXRAD cell for each crop irrigated. This provided a
potential range of values for irrigation estimates for those cells in which more than one
soil series was present. Irrigation amounts were aggregated at the county level and
reported as “low” and “high” estimates. The low estimate represents the irrigation
estimate if all crops were raised on the soil series in each NEXRAD cell requiring the
least amount of irrigation. The high estimate, conversely, represents the irrigation
estimate if all crops were raised on the soil series requiring the highest amount of
irrigation. Results of the simulation for those counties reporting more than 1,000

irrigated acres are shown in Table 4.7.

79

Table 4.7. Range of irrigation estimates for counties reporting more than 1,000 acres of
irrigation in 1997. Low = irrigation estimate (cm.) using soil series with least irrigation
requirement. High = irrigation estimate (cm.) using soil series with highest irrigation
requirement. Difference (em) = High - Low.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

County Acres Low High , Difference
St. Joseph 36211 14.10 14.50 0.41
Montcalm 27888 10.64 11.43 0.79
Branch 17367 15.01 15.16 0.15
Kalamazoo 12540 16.56 17.20 0.63
Van Buren 12363 11.94 12.17 0.23
Cass 10789 14.33 14.66 0.33
Ottawa 9224 5.49 6.12 0.64
Allegan 6975 10.11 10.34 0.23
Calhoun 6157 15.93 16.10 0.18
Berrien 4963 8.31 10.72 2.41
Monroe 4589 7.92 8.18 0.25
Mecosta 4398 9.55 10.54 0.99
Bay 4183 7.98 8.03 0.05
Kent 3410 13.21 13.44 0.23
Oceana 3358 11.20 11.51 0.30
Kalkaska 3035 17.53 18.08 0.56
Genesee 2850 11.68 11.68 0.00
Hillsdale 2798 12.88 13.31 0.43
Leelanau 2797 14.99 15.39 0.41
Sanilac 1545 2.97 2.97 0.00
Huron 1537 11.13 11.53 0.41
Lapeer 1528 1.78 2.03 0.25
lonia 1461 13.46 15.44 1.98
Newaygo 1455 4.83 5.72 0.89
Osceola 1450 13.51 14.33 0.81
Saginaw 1348 12.75 12.75 0.00
Mason 1217 15.06 15.62 0.56
Washtenaw 1181 12.01 16.36 4.34
Tuscola 1151 9.91 10.31 0.41
Gratiot 1109 14.25 14.50 0.25
Antrim 1049 21.49 21.56 0.08
Lenawee 1040 3.89 11.35 7.47

 

 

 

 

 

 

 

For most counties, the difference in irrigation estimates between the low and high
estimates was less than 1 cm. Lenawee and Washtenaw counties showed greater

variation, with differences of 7.47 cm. and 4.34 cm., respectively. For all counties

80

combined, the low estimate for 1997 was 973,166 acre-inches and the high estimate was
1,023,738 acre-inches. These results indicate that for most counties, identification of the
site-specific soil series would not significantly alter the irrigation estimate as compared to
the use of a composite soil. As with most results of this research project however, there
are exceptions to the general rule, as evidenced by Lenawee and Washtenaw counties.
The total irrigation amount for the 1997 data set estimated using a composite soil was
1,008,205 acre-inches, which as expected lies between the low and high estimates of

973,166 and 1,023,738.

4.6 Conclusions and Recommendations

The test for differences between means of paired time series data was used to group soil
map units into groups of soils with similar irrigation requirements. Groupings proved to
be dependent on crop type. Because of the differences by crop type and the number of
unique groups, results of the time series analysis could not be used to create a simplified
‘coarse-resolution’ data set that would apply for all crops. Analysis of outlier soils
showed general agreement between surface layer textural class and irrigation need, but
not in all cases. While major textural class can be used to generally classify soils by
irrigation need, exceptions will exist. Because of these exceptions, it is recommended
that all available soil information be used in the modeling process. In the state of
Michigan, the similarity of many soil series in regards to irrigation requirements allows

for the use of a composite soil in the modeling process. As discussed above, most

81

counties showed little variation in irrigation estimates obtained between low and high
irrigation soils. This is not true in all cases, however. For those counties that show a
high variation between low and high estimates, accuracy could be improved by obtaining

site-specific soil data.

Ultimately, the need for detailed soil information is dependent on the degree of
heterogeneity in the regional soils and the required degree of accuracy for the modeling
effort. As shown in this chapter, the SWBM is sensitive to changes in soil characteristics.
Any attempt to simplify the soil data would potentially result in a loss of accuracy, if we
assume that the results obtained with the most accurate data represent the best estimate.
The estimate using ‘detailed’ soil information must be viewed not as the ultimate ‘truth’,
however, but rather as a best estimate. If a confidence interval could be associated with
the best estimate, what might this interval be at the 95% level? The answer to this could
only be obtained by additional research. It is quite possible that simplified soil groupings
could result in estimates that are within an acceptable range (+/- x%) of the best estimate
using detailed soil data. As a general rule, if detailed information is available, it should
be used, as this will provide the best estimate regardless of the confidence level
associated with it. Site-specific data, if available, is the best option. Researchers must
always be aware of the degree of precision and accuracy by which input data is measured,

as this is correlated with the confidence interval one can attach to a best estimate.

82

CHAPTER V

PRECIPITATION DATA ANALYSIS

This chapter contains nine sections. Section 5.1 presents the research objectives for the
precipitation data analysis. Section 5.2 presents a statistical summary of the rain gauge
precipitation data used for NEXRAD validation. The spatial correlation of rain gauge
data is discussed in section 5.3. Sections 5.4 - 5.6 present a statistical summary of the
NEXRAD data for 1996 - 1998. The correlation between NEXRAD data and rain gauge
data is discussed in section 5.7. A comparison of SWBM estimates of irrigation
requirements using gauge data versus estimates using NEXRAD data is presented in

section 5.8. Section 5.9 discusses conclusions that can be made from this analysis.

5.1 Objectives

The availability of NEXRAD precipitation data was the major force for the initiation of

this project (as explained in chapter I). The use of NEXRAD data, however, requires the

model operators to be responsible for the conversion, validation, and maintenance of this

data. Technical expertise in relational database design and management, as well as

83

expertise in GIS-based data conversions is required. In addition, significant storage space
(on a personal computer) is required to physically store the NEXRAD data. Similar to
the STATSGO soils data discussed in the previous chapter, NEXRAD data requires a
significant amount of processing to be converted to the format used by the SWBM. In
addition to the processing and storage requirements, the availability of NEXRAD data is
an issue to be considered. The data used in this project is available through an informal
agreement between the National Weather Service (N WS) and Michigan State University,

but is not a standard NWS product at the time of this writing.

Chapter II reviewed literature pertaining to the spatial and temporal variability of
precipitation. Most studies examined daily or storm-level variability in precipitation
levels. Irrigation decisions, on the other hand, are likely to be made with a longer time
frame in mind. It is reasoned that irrigators do not respond to hourly or daily changes in
precipitation but rather to weekly or longer time intervals. That is, the decision to irrigate
is based on the current soil water status (a function of the season to date) and the forecast
or expectations for precipitation in the future (one week or more). It is firrther reasoned
that spatial variance in precipitation should be lessened as the time frame for analysis is
expanded. Monthly and weekly totals of adjacent NEXRAD cells should be more highly
correlated than daily totals, given the somewhat random nature of precipitation on a daily
basis. If the spatial correlation of NEXRAD cells is high, and the correlation of these
cells to nearby weather station data is high, then the potential exists for the use of a less-

detailed (but more readily available) data set as the precipitation input to the SWBM.

84

In view of the technical, storage, and availability issues regarding the NEXRAD data, the
investigation of the use of a surrogate data set for precipitation estimates (gauge data) is
pertinent to this research. This is even more evident when considering the potential for
expansion of the SWBM methodology to other states in the Great Lakes region, some of

which may not have NEXRAD data available.

The objectives of the weather data analysis are:

1. Partial validation of the NEXRAD data. NEXRAD precipitation estimates
will be compared to rain gauge measurements. Outlier data will be

detected (Section 5.3).

2. Determination of the degree of spatial correlation between gauge-
measured precipitation on a weekly basis and comparisons of totals on a

seasonal basis. (Section 5.4)

3. Determination of the degree of correlation between NEXRAD data and
gauge data on a weekly basis and comparisons of seasonal totals. (Section

5.7)

4. Determination of potential differences in irrigation estimates by the
SWBM using NEXRAD hourly and nearest-gauge daily data. (Section

5.8)

85

5.2 Weather Station Data: Statistical Summary

Daily reported weather station rain gauge data for the years 1996 — 1998 were obtained
for use as a basis for partial validation of the NEXRAD precipitation data and also to
assess the potential for the use of rain gauge data from existing weather stations in the
SWBM. The number and location of reporting gauges differed slightly in each year.
There were 86, 91, and 96 rain gauge stations in 1996, 1997, and 1998 respectively. A
complete listing of the stations for each year is given in Appendix B. Included in
Appendix B is a listing of the highest daily rainfall for each station. The maximum daily
rainfall reported is summarized in Table 5.1. Maximum daily rainfall was used to
identify a threshold value for detection of outliers in NEXRAD precipitation. Table 5.1
is interpreted as follows. For 1996, 1.42 to 12.19 represents the range of maximum daily
precipitation reported for all 86 stations for all days during the period May 1 to August

31.

Table 5.1. Summary of maximum daily rainfall reported for May — August, 1996 — 1997
by all rain gauges (cm.).

 

 

 

 

 

 

Year 4 Min. Max Mean
1996 1.42 12.19 5.43
1997 1.85 14.73 4.06
1998 1.35 12.19 4.02

 

 

 

 

 

 

86

Table 5.2. Minimum, maximum, mean, and standard deviation of total precipitation
(cm) reported by all stations for May — August, 1996 — 1998.

 

 

 

 

 

Yea? Min (Max Mean 7 SD
1996 11.86 56.03 32.97 8.30
1997 18.01 48.92 31.35 6.14
1998 13.00 46.05 25.02 7.25

 

 

 

 

 

 

 

 

Table 5.2 summarizes the seasonal precipitation totals for all reporting stations for May —
August, 1996 — 1998. The range of total precipitation for the four month period is quite
large. The lowest reported total precipitation by any one station was 11.86 cm., while the
highest was 56.03 cm. There was an average of 32.97 cm. of precipitation for the four
month period. Comparing across years, it can be seen in Table 5.2 that 1996 and 1997

had similar (on average) precipitation totals, and 1998 was slightly drier.

To compare the 1996 — 1998 station data to long term averages, summary statistics were
computed for the Lansing and Flint stations for the period 1961 — 1990. These stations
were chosen because they had a complete data set for the period 1961 - 1990. Other
stations in Michigan had incomplete or missing data for some portion of the 30 year
reporting period. The maximum daily reported rainfall over the 30 year period during the
4 month period May to August was 12.57 cm for Lansing and 11.30 cm for Flint. This
compares favorably with the maximum reported by all stations for 1996 - 1998 (Table

5.1).

87

A summary of May — August precipitation totals for the 30 year period is shown in Table
5.3. The mean precipitation over the 30 year period was essentially the same (30.6 cm.)
for Lansing and Flint. The values are within the range of the reported average values for

1996 — 1998 station data (Table 5.2), with 1998 being a slightly drier year than normal.

Table 5.3. Minimum, maximum, mean, and standard deviation of total precipitation
(cm) reported by Lansing and Flint stations for May — August, 1961 — 1990.

 

 

[ . Station Min (cm) Max (cm) Mean (cm.) ‘ SD

 

Lansing 18.74 46.17 30.60 7.34

 

(1965) (1975)

 

 

Flint 19.24 54.24 30.64 7.45

 

 

 

 

 

 

 

(1966) (1975)

 

 

88

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.1a. Spatial distribution of May — August 1996 total precipitation (size indicates
amount of precipitation. Range = 11.86 to 56.03 cm.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.1b. Spatial distribution of May — August 1997 total precipitation (size indicates
amount of precipitation. Range = 18.01 to 48.92 cm.

89

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.1c. Spatial distribution of May — August 1998 total precipitation (size indicates
amount of precipitation. Range = 13.00 to 46.05 cm.

The spatial distribution of total May — August precipitation for the years 1996, 1997, and
1998 is shown in Figures 5.1 (a — c). A pattern that is apparent from visual inspection of
these three years is the generally higher levels of precipitation in southwest Michigan. In
addition, lower precipitation values are seen in the central part of the state in 1997 and
1998. The occurrence of small and large values in close proximity to each other is an
indication of either high variability of rainfall in nearby locations, differences in

measurement precision and techniques between stations, or some combination of both.

90

5.3 Spatial Correlation of Gauge Measured Precipitation

The correlation of weekly precipitation totals between rain gauge sites was computed in
order to determine if the correlation of rainfall was related to distance between sites. The
coefficient of determination (r2) was calculated for each pair of gauge sites for each year.
For example, for the Lansing station in 1996, the correlation of weekly precipitation
values between the Lansing site and each of the other 85 sites was calculated. This
resulted in 85 r2 values, showing the degree of correlation of weekly precipitation values.
For Lansing, the r2 values ranged from a low of .00002 to a high of .88. The range of
values covers the spectrum from essentially no correlation to very high correlation of
weekly precipitation values. Figures 5.2 (a-c) show the results from a spatial standpoint

for Lansing in 1996, 1997, and 1998.

91

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.2a. Spatial correlation of weekly precipitation between Lansin (square) and
all other stations, May — August 1996. (Size = value of r2, light color = > .50).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.2b. Spatial correlation of weekly precipitation between Lansing (square) and
all other stations, May — August 1997. (Size = value of r2, light color = r2 > .50).

92

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.2c. Spatial correlation of weekly precipitation between Lansing (square) and
all other stations, May — August 1998. (Size = value of r2, light color = r2 > .50).

Differences in the number of stations that show significant correlation (r2 > .50) with the
Lansing station are evident between years. In 1996, 26 of the 85 stations had r2 values of
.50 or greater. Only 6 and 4 stations had significant correlation in 1997 and 1998,
respectively. The figures show a general, although certainly not conclusive, relationship
between r2 and distance. That is, higher r2 values are generally seen closer to the station
being analyzed. This is not consistent in all cases, however. For example, in 1997, the
Muskegon station had higher correlation (.57) with Lansing than the MSU Horticultural
station (.42). Muskegon is located 127 miles from the Lansing station, while the MSU
Horticultural facility is 9 miles from the Lansing station. Muskegon and Lansing were

certainly not affected by the same rainstorms (at least for summer convective storms), so

93

results such as this are due to the random component of rainfall distribution. In this case,
total precipitation for the four months was nearly identical for the Lansing and MSU
Horticultural stations; 27.1 and 27.4 cm. A relatively low r2 value indicates that the
timing of the precipitation differed between stations. The Muskegon station reported a

total of 26.1 cm in 1997.

To illustrate the difference in results obtained for other stations, correlation with the

Allegan station, located in southwest Michigan, are shown in Figures 5 .3 (a-c).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.3a. Spatial correlation of weekly precipitation between Allegan (square) and
all other stations, May — August 1996. (Size = value of r2, light color = r2 > .50).

94

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.3b. Spatial correlation of weekly precipitation between Allegan (square) and
all other stations, May -— August 1997. (Size = value of r2, light color = r2 > .50).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.3c. Spatial correlation of weekly precipitation between Allegan (square) and
all other stations, May — August 1998. (Size = value of r2, light color = r2 > .50).

95

The results for Allegan show the generally negative relationship between r2 value and
distance between stations. Stations in southwest Michigan are likely to be more highly
correlated with Allegan precipitation than stations in other parts of the state. As in the
Lansing case, none of the Upper Peninsula stations were highly correlated with Allegan.
The number of high correlation stations also declined from 1996 to 1998 as in the
Lansing case. 47, 17, and 7 stations had r2 values greater than .50 for 1996, 1997, and

1998 respectively.

Analysis of the rain gauge data has shown the high degree of variability in measured
rainfall amounts for the May — August period between rain gauge site located throughout
Michigan. In terms of weekly precipitation totals, it can be concluded that higher
correlation generally exists between closer stations, but not in all cases. The exceptions to
this show that there can be high variability of total precipitation on a weekly basis
between stations located a short distance apart. This would lend support for the use of
more detailed precipitation estimates (such as the NEXRAD data) in the modeling

process.

5.4 NEXRAD Data: Statistical Summary (1996)

The NEXRAD data used in this study were made available on an experimental basis to

MSU Agricultural Climatology by the Midwest office of the National Weather Service.

96

NEXRAD precipitation data were reported at hourly intervals in units of 1/100’h mm for
cells of approximately 4km x 4km resolution for the state of Michigan. In all, there are
7,446 cells in the State (see Figure 3.3). Because of the experimental nature of this data,
it was necessary to perform a partial validation of the data by comparing to rain gauge
measurements. The gauge-observed precipitation data from weather stations discussed in
the previous sections was used as a basis for comparing NEXRAD and gauge

precipitation amounts.

Table 5.4. Minimum, maximum, mean, and standard deviation of May — July
precipitation totals (cm.) for all NEXRAD cells, 1996, 1997, and 1998. (* = data
corrected for outliers.)

 

 

 

 

 

 

 

Year 1996 1997 1997* 1998
Min 0.009 0.09 0.09 0.03
Max 135.93 197.39 52.59 45.55
Mean 20.81 28.17 21.45 16.34
SD , 9.15 19.11 8.42 5.47

 

 

 

 

 

 

 

To begin the analysis, precipitation totals for each NEXRAD cell were summed for the 4
month period May 1 — August 31. Table 5.4 summarizes the seasonal totals for all
NEXRAD cells. In 1996, the minimum value for total May - August rainfall was .009
cm., and the maximum reported was 135.93 cm.. The mean value for 1996 was 20.81
with a standard deviation of9. 15. Total precipitation values were compared to data from

the weather stations. In 1996, the highest reported NEXRAD-reported precipitation total

97

was 135.93 cm., much higher than the highest value by all weather stations (56.03 cm.,
see Table 5.2). The cell with the extreme value of 135.93 was located in Alger County
on the south shore of Lake Superior. The next highest cell had a total of 69.89 cm. of
precipitation, and was located adjacent to the highest cell. The third highest cell had a
value of 61 .04 and was located in Monroe County (Southeast Michigan) along the Lake
Erie shoreline. Precipitation totals for all other cells were below 52 cm.. The value of
135.93 cm. is extremely high, more than double it’s neighboring cells. The higher
probability of fog and cloud cover along Great Lakes shorelines may contribute to these
high values. While the three highest values are all higher than expected, they cannot be

discounted as they are within potentially feasible ranges.

A number of factors can contribute to errors in radar estimation of precipitation,
including ground clutter and instrument calibration. Given the inexact nature of this
measurement, we can expect some values to be in the extremes of the distribution, while
most should be within expected ranges. This is the case for 1996 NEXRAD data. Figure
5.4 shows a frequency distribution of 1996 precipitation totals for all NEXRAD cells.
The frequency distribution illustrates the presence of a few outlier cells in the upper

range, and only one cell with a very low value (.09 cm).

98

 

Frequency Distribution: 1996 NEXRAD Total Precipitation

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

700
600 7 7 7 7 7 77 7 7
E 500 7 7 7 7 7 7 7 7 77
O
U
D
400 7 7 7
3
IIIJ
a. 300 7 7 7 7 -
'5
O
0 200 _ . ___- 77.. . 7777 - ._7. 77
100 7|, II II 7 _7
.1111 ., (1111...... .

 

 

COMODNQEFNMOONQIDFNVOCOM
Pv-NODI‘D WDONNmmmCOI-NNM

Total Precipitation (cm.)

 

 

 

Figure 5.4. Frequency distribution of 1996 NEXRAD precipitation (cm) for the period
May 1 — August 31.

In terms of daily rainfall amounts, for 1996 the maximum reported daily precipitation
from the 383,157 records was 30.52 cm, with the next value being 16.14, and all other
values below 15 cm. While the value of 30.52 cm. is likely not valid, all other values lie
within a reasonable (i.e., potential) range of daily precipitation. No data deletions or
adjustments were made to the 1996 NEXRAD precipitation data, it was concluded that

there were no clearly identifiable outlier values.

99

Table 5.5. Minimum, maximum, mean, and standard deviation of precipitation totals by
month for all NEXRAD cells, 1996.

 

J
0
57.9

6.2
3.9

 

 

Maps were produced showing the spatial distribution of rainfall by month over the four
month period and seasonal totals based on the NEXRAD data. These maps not only
provide a visual means of assessing rainfall patterns but also can be used to compare
rainfall distribution patterns between NEXRAD and station data. Figures 5.5 a-e show

total rainfall by NEXRAD cell for May, June, July, August, and Season Totals for 1996.

100

 

 

 

 

 

Figure 5.5a. Total precipitation by NEXRAD cell for May, 1996 (Darker color
indicates more precipitation).

101

 

 

 

 

 

Figure 5.5b. Total precipitation by NEXRAD cell for June, 1996 (Darker color
indicates more precipitation).

102

 

 

 

 

 

Figure 5.5c. Total precipitation by NEXRAD cell for July, 1996 (Darker color
indicates more precipitation).

103

 

 

 

 

 

Figure 5.5d. Total precipitation by NEXRAD cell for August, 1996 (Darker color
indicates more precipitation).

104

 

 

 

 

 

Figure 5.5c. Total precipitation by NEXRAD cell for May — August , 1996 (Darker
color indicates more precipitation).

Higher precipitation levels in southwest Michigan are evident from the monthly and
seasonal totals shown in Figures 5.5 (a-e). For 1996, spatial distribution of the areas with
heaviest rainfall differed by month. In July, for example, the Upper Peninsula
experienced relatively more rainfall compared to the rest of the state than in May and
June. Figure 5.5e can be compared to Figure 5.1a to visually assess the correlation
between NEXRAD precipitation estimates and station totals (rain gauge measurement)

for the four month period. The general pattern of higher precipitation in a band from the

105

southwest to the Saginaw Bay area can be seen in both the NEXRAD and the rain gauge
precipitation totals. Correspondingly, lower precipitation is seen in other areas of the
state, that is, the Upper Peninsula, northern Lower Michigan, and southeast Michigan. A
more detailed discussion on the correlation of NEXRAD data and rain gauge
measurements will be given later (section 5.6). Figure 5.5a shows the total monthly
precipitation by NEXRAD cell for May, 1996. A demarcation of the outer range of the
NEXRAD radar appears to be visible in the southeast and southwest portions of the state.
These range lines are also visible in the four month totals (Figure 5.5e). The same range
line can be seen in the totals for May, 1997 (Figure 5.7a), but not in other months for

1997. The reason for this apparent range-dependent bias for May is currently not known.

5.5 NEXRAD Data: Statistical Summary (1997)

1997 NEXRAD precipitation data were summarized in the same manner as for 1996.
Initial analysis showed a maximum cell value of 197.39 cm. for the 4 month totals, a
value much higher than any of the reporting weather stations during the 3 year period
1997 — 1998. Inspection of the daily rainfall amounts showed a high value of 171.52 cm.,
an unrealistically high daily rainfall amount. For comparison, the highest daily rainfall
reported for the Station data over the 3 year period was 14.73 cm. This clearly indicates

the presence of outlier values containing incorrect values in the 1997 data.

106

Further investigation revealed that all NEXRAD data with unrealistic precipitation
amounts for 1997 was associated with day 182 (July 1, 1997). Most of the high
precipitation values for day 182 were in Southeast and North-Central MI but high values
(over 50 cm) appeared randomly in some of the Upper Peninsula counties as well. It was
initially thought that the outliers on day 182 might be caused by incorrect calibration or
data processing that resulted in a systematic shift of the numbers to a higher order of
magnitude. A visual comparison with gauge data for day 182 however, revealed little
correlation between the NEXRAD data and ground-measured gauge data. Except for
isolated high reports, most gauges in Lower Michigan reported 0 or trace rainfall for this
day. Lapeer, Dryden, and Chesaning were the highest reporting gauges, with 2.33, 2.06,
and 1.83 cm of reported rainfall. 41 of the 91 gauges reported 0 precipitation, 70 of the
91 reported less than .254 cm for this day. Based on the gauge data, this appears to be a
day of isolated thunderstorms with localized heavy rainfall in a few areas. For purposes
of this analysis, it was decided to eliminate the NEXRAD data for this day on the
grounds that precipitation values were unrealistically high and could not be attriuted to
invalid calibration or data processing. According, all NEXRAD precipitation values for
day 182, 1997 were set to 0. All original values were stored in a table called
DATA_PRECIP_ERRS, in the event that they could be useful for tracing errors in the
future. The adjusted 1997 NEXRAD data was used for all further analysis. Differences

between the original and adjusted 1997 data are shown in Table 5.4.

A frequency distribution of NEXRAD data precipitation totals is shown in Figure 5.6.

The frequency distribution shows a bimodal distribution, unlike the more normally

107

distributed values for 1996 (Figure 5.4). While the mean cell value is very similar for the
two years (20.8 in 1996, 21.4 in 1997) the distribution across cells is very different. The
range of values is much greater for 1996 data, as shown in Table 5.4. In the upper range,
the 1996 maximum was 135.93 cm. while the 1997 maximum was 52.59 cm. The
distribution of total May — August precipitation differed between 1996 and 1997. In
1996 (Figure 5.5e) more rain was reported in the central part of the state, while in 1997
(Figure 5.7e) the heaviest concentration was in the western part of the state and in
portions of Menominee County in the Upper Peninsula. The May totals for 1997 (figure

5.7a) show a range-dependent bias as does the 1996 May totals (Figure 5.5a).

 

Frequency Distribution: 1997 NEXRAD Total Precipitation

 

250 it

§
1

 

 

 

 

 

Count (NEXRAD Cell)
8
1

§
1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

50 1 1|, ,
0 TITlrlrlljrllll'l'i'illl II 7' . I ‘ T I III" IITIII’JTITIA'I'I'l'l‘; 1., 7.177%
O m I!) [N O N In N O N In N O) N g N 0') N V’ (D O) V-
a- v- v- v- N N N N N O") t') (‘0 v V V V In

Total Precipitation (cm)

 

 

 

Figure 5.6. Frequency distribution of 1997 NEXRAD precipitation (cm) for the period
May 1 — August 31.

108

Table 5.6. Minimum, maximum, mean, and standard deviation of precipitation totals by
month for all NEXRAD cells, 1997.

 

 

 

Maps were produced showing the spatial distribution of rainfall by month over the four
month period and seasonal totals based on the NEXRAD data for the 1997 data. Figures
5.7 (a—e) show total rainfall by NEXRAD cell for May, June, July, August, and Season

Totals for 1997.

109

 

 

 

 

 

Figure 5.7a. Total precipitation by NEXRAD cell for May, 1997 (Darker color
indicates more precipitation).

110

 

 

 

 

 

Figure 5.7b. Total precipitation by NEXRAD cell for June, 1997 (Darker color
indicates more precipitation).

Ill

 

 

 

 

 

Figure 5.7c. Total precipitation by NEXRAD cell for July, 1997 (Darker color
indicates more precipitation).

112

 

 

 

 

 

Figure 5.7d. Total precipitation by NEXRAD cell for August , 1997 (Darker color
indicates more precipitation).

113

 

 

 

 

 

Figure 5.7c. Total precipitation by NEXRAD cell for May — August , 1997 (Darker
color indicates more precipitation).

5.6 NEXRAD Data: Statistical Summary (1998)

The range of May - August precipitation totals for 1998 NEXRAD data was .03 to 45.55
cm (Table 5.4). Because the maximum value of 45.55 cm. is within feasible amounts, no
outlier correction was necessary for 1998 NEXRAD precipitation data. Comparison of
the mean value for all NEXRAD cells in 1998 with 1996 and 1997 reveals the lower

amount of summer precipitation for 1998. Mean values were 20.8, 21.4, and 16.3 cm. for

114

1996, 1997, and 1998 respectively (Table 5.4). Table 5.7 shows a statistical summery by
month for 1998, which shows that the lower 1998 precipitation totals were largely due to
a very dry July period. Monthly averages for the other three months are comparable to
1996 and 1997 values, however July is significantly lower. The frequency distribution of
1998 NEXRAD cell totals is shown in Figure 5.8. Unlike the 1997 bimodal distribution
(Figure 5.6) the 1998 distribution more closely approximates the classical normal
distribution. As for 1996 and 1997, the distribution of monthly and seasonal precipitation

totals for 1998 are shown in Figure 5.9 (a-e).

Table 5.7. Minimum, maximum, mean, and standard deviation of precipitation totals by
month for all NEXRAD cells, 1998.

 

J
0 0
18.5 12.7

5.4 2.2
2.8 1.7

 

 

115

 

Fnaqunwuartlsbflbuficur:1998liE!fliA[)Fhemipflaflkxn(Ilay-nnuunlil

 

 

 

 

Count (NEXRAD Cell)
9
l
l

 

 

 

100 77 . . __7.7_.77, -777 .__,

,0 ll.

- II 77 ,
0 _Illlllllllllllll Illllllllinl...-- __-- ...._
11111 7’1111 1 ' ‘ 11' 11 '1 u11I11IIIIIIIIIITIJ'IIrIITIIIIIIIIrIIjIIITTIIIl'I-I—
O N g (D
(‘0 0‘) (‘0

owvcooo miniscna—coiooo
FPFFNNNN ('0

Tot: Precipitation (cm)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

F
P

 

 

 

Figure 5.8. Frequency distribution of 1998 NEXRAD precipitation (cm) for the period
May 1 - August 31.

116

 

 

 

 

 

Figure 5.9a. Total precipitation by NEXRAD cell for May, 1998 (Darker color
indicates more precipitation).

117

 

 

 

 

 

Figure 5.9b. Total precipitation by NEXRAD cell for June, 1998 (Darker color
indicates more precipitation).

118

 

 

 

 

 

Figure 5.9c. Total precipitation by NEXRAD cell for July, 1998 (Darker color
indicates more precipitation).

119

 

 

 

 

 

Figure 5.9d. Total precipitation by NEXRAD cell for August, 1998 (Darker color
indicates more precipitation).

120

 

 

 

 

Figure 5.9e. Total precipitation by NEXRAD cell for May — August, 1998 (Darker
color indicates more precipitation).

5.7 Correlation Between Rain Gauge and NEXRAD Precipitation Data

The literature reviewed in chapter 11 pertaining to the validation of radar data showed
generally good agreement between rain gauge measured data and radar estimates,
although radar estimates tended to show less precipitation than rain gauges. These

studies were conducted in an experimental setting with greater control and a higher

121

 

density of rain gauges than is often encountered in an operational setting. Readily
available precipitation estimates are usually only available from a relatively sparse
network of weather stations, as is the case for Michigan. Because of this, a complete
validation of the NEXRAD data is not possible in this study. However, assessment of the
degree of association and correlation between NEXRAD data and available station data
provides valuable information. If the two data sources are highly correlated and show
similar values, then the potential exists for using station data as a surrogate for NEXRAD
data, if NEXRAD data are not available. If they are not highly correlated or do not show
good agreement, then further investigation and more detailed research will be necessary

to explain the discrepancies.

In the first phase of this analysis, the seasonal (May — August) totals for each rain gauge
station are compared to the NEXRAD seasonal totals for the NEXRAD cell in which the
station is located. It should be a fair assumption that seasonal rainfall totals will not be
highly variable within a 4 km area of a given point (or rain gauge). Given this
assumption, and the assumption that rain gauge measurement is accurate, then we would
expect to see a high correlation between the rain gauge measured precipitation and the
NEXRAD measured precipitation for the same area. Table 5.6 shows total May - August
precipitation measured by each of the 79 common stations for the two years 1996 - 1998.
The R/G ratio, calculated as Radar Total/ Gauge Total, is also shown. The R/G ratio
gives an indication of the relative discrepancy between the two estimates. Table 5.7
summarizes the results of this comparison. For 1996, the difference between radar and
gauge reported precipitation totals ranged from -24.1 to +9.2 cm. That is, the radar

estimate for one cell was 24.1 cm less than the rain gauge estimate in the cell, while

122

another radar estimate was 9.2 cm. more than reported by the rain gauge. The average
discrepancy was -10.6 cm. in 1996, -5.5 cm. in 1997, and —7.5 cm in 1998. The average

R/G ratio was .68, .83, and .75 for 1996, 1997, and 1998, respectively.

Frequency distributions of the R/G ratio for 1996 — 1998 are shown in Figures 5.8a, 5.8b,
and 5.8c, respectively. The frequency distributions shown in Figure 5.8 illustrate the
tendency for NEXRAD estimates of precipitation to be lower than gauge measured
precipitation. Assuming that the range of .7 to 1.3 for the RIG ratio is a range of
reasonably good correlation, in 1996, 29 of 83 stations (35%) were within this range. For

1997, 58 of 89 (65%) and for 1998 41 of 88 (47%) were within the range of .7 to 1.3.

Table 5.8. May — August precipitation totals (cm.) for NEXRAD Radar (R) and Rain
Gauge (G) and R/G ratio for all stations reporting in 1996, 1997, and 1998.

 

 

123

Table 5.8 (Cont’d)

 

124

Table 5.8 (Cont’d)

 

 

Table 5.9. Summary of differential between radar and gauge-measured precipitation for
a) May — August totals, 1996 — 1998 and b) July —August totals. n = number of stations,
Min = minimum of radar — gauge (cm.), Max = maximum of radar — gauge (cm.), Mean =
mean of radar — gauge, R/G = radar/gauge, and SD = standard deviation of R/G.

 

a) May — August

 

b) July - August

 

 

125

 

Frequency Distribution of RIG Ratio, 1996

 

(a)

O
r
T

Count

 

    

. l - l 1 r T‘-_"i
0.05 0.24 0.43 0.63 0.82 1.01 1.20 1.39 1.58 1.77
RIG Ratio

 

 

 

 

 

Figure 5.10a. Frequency distribution of R/G ratio, 1996.

 

Frequency Distribution of RIG Ratio, 1997

 

Count
3

 

 

 

 

0.17 0.32 0.48 0.64 0.80 0.96 1.12 1.28 1.44 1.60
RIG Ratio

 

 

 

Figure 5.10b. Frequency distribution of R/G ratio, 1997.

126

Frequency Distribution of RIG Ratio, 1998

 

20 , 4+

 

15 -—~——~

Count

10 ~94

 

 

 

0.18 0.32 0.46 0.60 0.74 0.89 1.03 1.17 1.31 1.45 1
RIG Ratio ‘

Figure 5.10c. Frequency distribution of R/G ratio, 1998.

Maps were produced to assess the spatial pattern of correlation between the radar (R) and
rain gauge (G) precipitation totals for 1996 — 1998 (figures 5.11 a-c). Values of the R/G
ratio in the range of .7 — 1.3 are shown as the darker color. The size of the circle indicates
the value of the R/G ratio. Two patterns emerge from these maps; consistently lower
NEXRAD precipitation estimates in the Upper Peninsula and northern Lower Michigan
compared to rain gauges, and, consistently better agreement between NEXRAD and rain
gauge estimates in southern and central Michigan. In Figure 5.11b it can be seen that
most of the R/G ratios in the southern and central part of the state (where most irrigation
takes place) fall within the .7 — 1.3 range. Of note are the values for Huron and Sanilac

counties (the ‘thumb’ area), which show low R/G ratios for both 1996 and 1997.

127

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

R/GRatio Z. 0 ’ ,
. 0.05-0.21 {-1

0.21-0.38

0.38-0.51
0.51-0.6 .
0.6-0.69 ° 0 T' '
(1.7-074 0. .
0.74-0.86 Q: o ’ '

81332139 1' - 1

7 . 1
an...“ fi' . -1. 1.1T? -/

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

cocoooo- -

 

 

 

 

 

 

 

 

Figure 5.11a. Spatial distribution of the R/G ratio, 1996.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.11b. Spatial distribution of the R/G ratio, 1997.

128

 

 

 

RIG Ratio

3

O

O
[:1

 

0.18 - 0.27
0.27 - 0.5
0.5 - 0.6
0.6 - 0.69
0.7 - 0.74
0.74 - 0.8
0.8 - 0.88
0.88 - 1.02
1.02 - 1.3
1.3 - 1.45
Counties

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

i}
l
r”

 

 

 

 

 

 

 

o
0 .

 

 

 

 

 

 

 

 

 

 

 

 

 

/

 

 

 

 

 

based errors in the NEXRAD data.

NEXRAD totals for 1996 — 1998.

129

Figure 5.11c. Spatial distribution of the R/G ratio, 1998.

R/G ratios for 1998 indicate better agreement between radar and gauge estimates in the
central part of the state, with lower radar estimates for most sites in the southern and
northern parts of the state. Spatial grouping of the R/G ratio can be seen for all three

years, but this pattern differs each year. Such patterns may be indicative of calibration-

To visually assess the correlation between rain gauge observed precipitation levels and

radar estimates, figures 5.12 a — c show the rain gauge totals super-imposed on the

 

 

 

 

 

 

Figure 5.12a. Rain gauge totals versus NEXRAD totals for May - August 1996. (Size
indicates value of the gauge observation)

130

 

 

 

 

 

Figure 5.12b. Rain gauge totals versus NEXRAD totals for May - August 1997. (Size
indicates value of the gauge observation)

131

 

 

 

 

 

Figure 5.12c. Rain gauge totals versus NEXRAD totals for May - August 1998. (Size
indicates value of the gauge observation)

The above analysis has been concerned with the correlation of seasonal totals (May —
August) of precipitation between rain gauges and NEXRAD. Another important
consideration is the timing of rainfall over the course of a season. In order to assess this,
the coefficient of determination (r2) was calculated for weekly precipitation totals
between each station and the nearest NEXRAD cell. Results of this analysis are shown in

Table 5.8. Simple linear regression was used to estimate the equation:

132

y=a+bx (Eq.5.1)
Where
y = Radar (N EXRAD) precipitation
a = y intercept
b = slope

x = Rain gauge observation.

For 1996, 55 of the 83 sites had r2 values greater then .50. In 1997, only 11 of the 89
stations had r2 values greater then .50 (Table 5.10) and in 1998, 19 of 88 sites had r2
values greater then .50. While the correlation between radar and gauge seasonal totals for
1997 appeared to be best for the three analysis years, correlation of the weekly totals was
the weakest for 1997. For 1997 and 1998, the general conclusion can be made that for
most rain gauge sites, changes in gauge-measured precipitation on a weekly basis had
little correlation with the NEXRAD measured weekly precipitation for the surrounding

area. Results are shown for 1996 — 1998 in Figures 5.13 a — c.

Table 5.10. Summary of linear regression of 1997 weekly NEXRAD precipitation totals
on Gauge totals, ordered by r2 (descending).

 

 

 

 

 

 

 

 

 

 

 

 

 

..'ta,tion a I " b 'r—s‘q‘uared',
DRYDEN 0.221 0.493 0.68
KALKASKA 0.107 0.623 0.66
GRANT 0.122 0.889 0.57
FREMONT 0.055 0.782 0.56
GULLLAKE 0.164 0.731 0.54
IONIA 0.097 0.706 0.54
FLINT 0.542 0.553 0.52

 

133

 

Table 5.10 (Cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Station , , J , r-squared
WHEELER 0.236 0.865 0.51
MTCLEMENS 0.431 0.469 0.51
SAGVALLEY 0.072 0.741 0.5
NILES 0.291 0.624 0.5
LUDINGTON 0.28 0.684 0.48
PELLSTON 0.35 0.528 0.48
MUNISING 0.205 0.516 0.46
CHATHAM 0.106 0.535 0.45
LANSING 0.589 0.517 0.44
HTNLAKE 0.508 0.343 0.41
PEACHRIDG 0.711 1.022 0.4
NUNICA 0.449 0.608 0.38
SALINE 0.397 0.622 0.38
BEAVERTON 0.54 0.422 0.36
HART 0.185 0.845 0.35
IRONMTN 0.323 0.768 0.35
DETROIT 0.73 0.613 0.35
ROMEO 0.541 0.36 0.34
ROGERCITY 0.223 0.265 0.32
SAGINAW 0.194 0.558 0.31
OLDMISSION 0.5 0.296 0.31
BIGRAPIDS 0.581 0.605 0.3
ENTRICAN 0.371 0.78 0.29
ALPENA 0.405 0.163 0.27
ALLENDALE 0.351 1.135 0.27
MSUHORT 0.517 0.547 0.26
KENTCITY 0.424 0.846 0.26
CHESANING 0.495 0.479 0.25
WESTBRANCH 0.283 0.287 0.25
MORENCI 0.673 0.725 0.24
NWMIHORT 0.437 0.309 0.23
STANDISH 0.65 0.288 0.23
MUSKEGON 0.733 0.439 0.23
CORNELL 0.28 0.393 0.22
LAPEER 0.58 0.459 0.21
LAKECITY 0.679 0.261 0.2
HASTINGS 1.389 0.442 0.19
BADAXE 0.718 0.207 0.18
OSSINEKE 0.371 0.091 0.18
CENTREVL 0.914 0.297 0.18
OWOSSO 0.783 0.237 0.17
VANDERBILT 0.351 0.17 0.17
KINDE 0.667 0.367 0.17
BELDING 0.675 0.456 0.16

 

 

 

 

134

 

Table 5.10 (Cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Station . ’ 7 7 r-squared

VESTABURG 0.862 0.4 0.15
HOUGHTON 0.08 0.081 0.14
IRONWOOD 0.514 0.108 0.13
CLARKSVL 0.905 0.325 0.13
BEULAH 0.474 0.678 0.13
HOLLAND 1.313 0.379 0.13
STAMBAUGH 0.452 0.296 0.12
DUNDEE 1.034 0.314 0.12
BEARLAKE 0.796 0.271 0.11
BROOKLYN 0.935 0.396 0.1
FENNVILLE 1.079 0.48 0.1
HUDSON 0.996 0.323 0.1
SSMARIE 0.423 0.52 0.09
GLENDORA 1.14 0.182 0.09
EATNRAPIDS 1.069 0.253 0.08
MILFORD 1.034 0.293 0.07
SANDUSKY 0.878 0.184 0.07
PORTSANILA 0.857 0.24 0.07
HUDSNVLLE 1.343 0.289 0.07
ALLEGAN 1.49 0.166 0.06
SWMIHORT 1.173 0.173 0.06
WATERVLIET 1.183 0.114 0.05
BALDWIN 1.315 0.498 0.05
PIGEON 1.18 0.275 0.05
OXFORD 0.987 0.194 0.04
ALMA 0.987 0.235 0.04
MARQUETTE 0.304 0.083 0.04
HELL 0.948 -0.098 0.03
GRANDJUNC 1.166 0.439 0.03
GRAPIDS 1.609 0.317 0.03
WHITE LAKE 1.822 -0.203 0.02
PAWPAW 1.345 0.241 0.02
STEPHENSON 1.928 0.14 0.01
TIPTON 1.471 0.131 0.01
COLDWATER 1.364 -0.062 0.01
FENTON 1.471 0.044 0
THREERIVER 1.691 -0.033 0
NORTHPORT 0.788 0.011 0

 

 

 

 

 

135

 

 

 

 

R-squ

1:1

 

 

ared

0.003 - 0.152
0.152 - 0.309
0.309 - 0.434
0.434 - 0.49
0.5 - 0.606
0.606 - 0.67
0.67 - 0.721
0.721 - 0.81
0.81 - 0.85
0.85 - 0.975
Counties

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6.1

Q .~ .IJ.( 5M7

 

 

 

O
v
O

 

 

 

 

 

O
O
_—

 

 

 

 

 

 

Figure 5.13a. Spatial distribution of r2 values for correlation of weekly radar
precipitation on gauge measured precipitation, 1996. (Size indicates value of r2).

 

 

 

R-squ

o
[:1

 

 

ared

0 - 0.025
0.025 - 0.49
0.05 - 0.102
0.102 - 0.148
0.148 — 0.211
0.211 - 0.271
0.271 - 0.356
0.356 - 0.49
0.5 - 0.567
0.567 - 0.676
Counties

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.13b. Spatial distribution of r2 values for correlation of weekly radar
precipitation on gauge measured precipitation, 1997. (Size indicates value of 1'2).

136

 

 

 

 

 

 

 

 

 

 

 

  
 

 

 

 

 

 

R-squared

0 - 0.038
0.038 - 0.098
0.098 - 0.151
0.151 - 0.218
0.218 - 0.281
0.281 - 0.308
0.308 - 0.385
0.385 - 0.49
0.5 - 0.628
0.628 - 0.777

ECountias oi . .11 .o. 1.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

......0.0o

 

 

 

 

 

 

 

 

 

 

 

Figure 5.13c. Spatial distribution of r2 values for correlation of weekly radar
precipitation on gauge measured precipitation, 1998. (Size indicates value of r2).

The spatial distribution of r2 values reported in Table 5.8 is shown in Figure 5.13b.
Higher values indicate better correlation, values of greater than .5 are shown as a darker
color. Spatial patterns of correlation for 1997 and 1998 are not evident based on visual
inspection of Figures 5.13 b and c. Stations that show a high correlation in terms of total
seasonal precipitation do not in all cases show a high correlation in weekly totals. This
indicates the need to assess the timing of precipitation, as well as total amounts, when

comparing NEXRAD and rain gauge observations.

137

5.8 Irrigation Estimates using Gauge Vs. NEXRAD Preciptiation

If rain gauge observations and NEXRAD estimates are highly correlated in terms of both
seasonal and weekly totals, then the potential for using rain gauge observed precipitation
in the SWBM exists. Based on the above analysis, however, it can be concluded that a
high degree of correlation, in terms of seasonal distribution of rainfall, does n_ot exist
between rain gauge observations and NEXRAD estimates. Assuming that systematic
biases in NEXRAD data are corrected by calibration to rain gauge observations and the
random error component is minimized, the use of NEXRAD data should be preferable to
other sources of data because of it’s ability to capture localized variations in precipitation
amounts. In cases where NEXRAD data are not available, rain gauge measured
precipitation data may be a suitable substitute if the gauge network is dense enough to

capture regional variation in precipitation totals.

To assess the potential impact of the use of gauge data rather than NEXRAD data, three
sets of simulations were run, varying only the source of precipitation data. Irrigation
requirements for 100 acres of corn and 100 acres of soybeans were simulated for all
NEXRAD cells in five counties. The five counties simulated were Allegan, Huron, Ionia,
Grand Traverse, and St. Joseph. These counties were selected because of their potential
for irrigation and location in various regions throughout Michigan. Temperature and
solar radiation data were from the same data source. The three sets of precipitation data

were:

138

1. NEXRAD data
2. Nearest rain gauge data. Precipitation values from the nearest rain gauge site.

3. Weighted average of all rain gauge sites within 48 km. (30 miles).
Precipitation values from all rain gauges within 48 km. were used. Weights to
each value were assigned based on the squared inverse of rain gauge distance

from the NEXRAD cell. See Example 5.1.

Example 5.1. Calculation of precipitation weighted by inverse of squared distance.

 

In situations where more than one rain gauge was located in close proximity (<30 miles),
the precipitation value was calculated as a weighted average of all nearby stations,
weighted by the inverse of squared distance. For example, suppose there are three

stations within 30 miles, with the following values:

 

Station Distance Precipitation
1 2 l 0

2 4 5

3 6 1 1

In step 1, the total of the squared inverse distance is calculated, in this case,
1/22 +1/42 +1/362 = .34

In step 2, estimated precipitation is calculated as a sum of the precipitation from each

station times the weight for the station. This is:

((1/22)/.34) * 10) + ((1/42)/.34) * 5) + ((1/62)/.34) * 11) =9.17.

In cases where no gauges existed within 30 miles, the value for the nearest gauge was

used.

 

139

Results of the simulations are presented in Tables 5.11a and 5.11b for corn and soybeans,
respectively. The total irrigation requirements (cm.) assume that 100 acres of corn and
100 acres of soybeans were grown in each NEXRAD cell in each county. The result of
interest is the Pl/R and P2/R ratios, which show the ratio of irrigation estimates using the
nearest station (P1) and multiple stations (P2) to the irrigation estimate using radar (R)
precipitation. For example, in Allegan County, values of .96 and .94 are shown for corn
(Table 5.9a). This indicates that using the nearest gauge data for precipitation inputs, the
irrigation estimates were .96 of the estimates obtained using NEXRAD data. Using a

weighted average of all gauges within 48 km., the ratio was .94.

As expected, total irrigation requirements were less using rain gauge precipitation data
than using NEXRAD data for all counties for both crops. This is due to the consistently
lower NEXRAD precipitation values compared to rain gauge measurements, as discussed
above. R/G ratios (Table 5.8) are as follows; Allegan, .8, Ionia, .8, Three Rivers (St.
Joseph) .6, NWMI Hort, .6, Old Mission .7, Pigeon 1.0 and Bad Axe .6. This indicates
that, on average, gauge observed precipitation is 20 — 40% higher than NEXRAD
estimates in the simulated counties. Correspondingly, irrigation totals using rain gauge
precipitation are lower (by 10 —15%) than totals using NEXRAD precipitation. In all but
one case (Allegan), the use of multiple stations to estimate the precipitation value resulted

in an improvement in the P/R ratio.

140

Table 5.11a. Results of the simulations using NEXRAD (R) and gauge precipitation (P)
data for corn. P1 = nearest station, P2 = all stations within 48 km. Total irrigation water
applied is shown (cm.) for the three data sources. Ratio of P/R is shown for both P1 and

P2.

 

 

 

 

 

 

 

 

 

 

County R (cm.) P1 (cm.) P2 (cm.) P1/R P2/R
Allegan 64680 62230 60564 0.96 0.94
Huron 55566 40474 43414 0.73 0.78
lonia 45178 34594 35084 0.77 0.78
Grand Traverse 34888 30282 30282 0.87 0.87
St. Joseph 41454 36652 37828 0.88 0.91
Total 241766 204232 207172 0.84 0.86

 

 

 

 

 

 

 

 

Table 5.1 lb. Results of the simulations using NEXRAD (R) and gauge precipitation (P)
data for soybeans. P1 = nearest station, P2 = all stations within 48 km. Total irrigation
water applied is shown (cm.) for the three data sources. Ratio of P/R is shown for both

P1 and P2.

 

 

 

 

 

 

 

 

 

 

County R (cm.) P1 (cm.) P2 (cm.) P1/R P2/R
Allegan 110642 108290 107114 0.98 0.97
Huron 153370 98000 130046 0.64 0.85
lonia 100744 98784 102900 0.98 1.02
Grand Traverse 107800 98000 99372 0.91 0.92
St. Joseph 73402 60564 60956 0.83 0.83
Total 545958 463638 500388 0.85 0.92

 

 

 

 

 

 

 

 

The results reported in Tables 5.9a and 5% lend support for the potential use of daily
station data from rain gauges to provide irrigation estimates consistent with estimates

computed using NEXRAD data. Differences in the estimates provided can be explained

141

by differences in NEXRAD and rain gauge precipitation estimates. In addition to
assessing county totals, however, it is important to view the results at the NEXRAD cell
level. Biases in estimations could be introduced, for example, as a function of distance
from the weather station. Cells closer to the station should be in closer agreement than
those cells located farther away. Table 5.12 shows the range of P/R values in each
county for each crop simulated. In Allegan County, for example, values for corn ranged
from .14 to 2 using the nearest station (Pl/R), and .57 to 3 using all stations within 48
km. (P2/R). On a cell by cell basis great variation can be seen between irrigation
estimates using NEXRAD precipitation and estimates made using rain gauge data. When
aggregated to the county level however, individual cell variations disappear. This
analysis assumes that crops are grown in all NEXRAD cells in each county, which in
reality is not the case. If only a small percentage of the cells in a given county contain
irrigated acreage, then the potential exists for irrigation estimates made using nearest rain

gauge data versus estimates made using NEXRAD data to differ substantially.

Table 5.12. Range of P/R ratios by County and Crop.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

County Crop Min P1/R Max P1/R Min P2IR Max P2IR
Allegan Corn 0.14 2 0.57 3
Soybeans 0.3 3 0.73 2.5
Grand Traverse Corn 0.71 2 0.6 1.33
Soybeans 0.5 1.18 0.69 1.2
Huron Corn 0.38 1.67 0.5 1
Soybeans 0.57 1.44 0.64 1.2
lonia Corn 0.33 2 0 1.33
Soybeans 0.62 6 0.8 1.57
St. Joseph Corn 0.71 1.25 0.71 1.25
Soybeans 0.6 1.3 0.6 1.25

 

 

142

Because the P1 and P2 estimates are based on the same data source (nearest station or
stations), the variability seen in P/R ratios must be a function of the variability in
NEXRAD precipitation estimates. This is illustrated in Figures 5.14a, 5.14b and 5.140,
which show May — August precipitation totals for St. Joseph County using a) NEXRAD

data, b) nearest station (P1), and c) multiple stations (P2).

 

 

. Weather Station
Precipitation (cm.)
25.9 7. 03

 

   

 

 

Figure 5.11a. Distribution of total May - August 1997 precipitation (cm.) in St. Joseph
County by NEXRAD cell using NEXRAD data.

 

. Weather Station
Precipitation (cm.)
- 24.13
- 35.992
1% 38.456
- 48.461

 

 

 

 

Figure 5.11b. Distribution of total May - August 1997 precipitation (cm.) in St. Joseph
County by NEXRAD cell using nearest station data (SI).

143

 

 

0 Weather Statlon
Precipitation (cm.)

   

 

 

 

Figure 5.14c. Distribution of total May - August 1997 precipitation (cm.) in St. Joseph
County by NEXRAD cell using weighted multiple station data (S2).

The effect of smoothing the precipitation data using a weighted average of multiple
stations is apparent in Figures 5.14b and 5.14c. Data values from four stations are shown
in Figure 5.14b, with each cell assigned a value according to the nearest station.
Precipitation values are ‘smoothed’ when all stations within 48 km. are used to compute a
precipitation estimate, as seen in Figure 5.14c. In terms of range of values, the NEXRAD
data ranges from 25.98 to 36.6 cm., the SI data ranges from 24.13 to 48.46, and $2 data
ranges from 35.17 to 48.38. A high degree of variation in seasonal totals by NEXRAD
cell is seen for all Counties. Table 5.12 shows the minimum, maximum, mean, and
standard deviation of May — August precipitation for the five Counties simulated. The
range for these five Counties is from 10 to 20 cm. of precipitation. That is, there can be a
difference of 10 to 20 cm. in precipitation estimates between cells in the same county for
a given season. This variation, if indeed true, can not be captured by rain gauge

measurements because rain gauges are generally located many kilometers apart.

144

Table 5.12. Minimum, maximum, mean, and standard deviation of May — August
precipitation totals for all NEXRAD cells by County.

 

 

 

 

 

 

 

 

 

 

 

 

 

County Min (cm.) Max (cm.) Mean (cm.) SD
Allegan 26.1 46.3 35.8 4.6
Grand Traverse 14.3 27 19.9 3.1
Huron 16.4 30.6 22 2.8
Ionia 21 35.3 26.5 3.4
St. Joseph 26.4 36.6 30.9 2.1

 

 

To conclude the precipitation data analysis, the SWBM was run using the 1997 irrigation
acreage reported to MDA using the nearest rain gauge site for precipitation data. Results
compared to the results obtained using NEXRAD precipitation data are shown in Table
5.13. The irrigation estimate using rain gauge data was 94% of the amount obtained
using NEXRAD precipitation data. This is consistent with the findings reported in
Tables 5.9 (a and b). The use of rain gauge measured data resulted in aggregate irrigation
estimates that were slightly lower than those obtained using NEXRAD data, which is
largely a function of the higher rainfall amounts reported by rain gauges. The difference
in terms of the average depth of water applied was less than one centimeter (12.6 versus
11.8 cm.). Given the uncertainty in the measurement of site—specific driving variables
(soil and weather) of the SWBM in addition to the potential variation in management
practices, a difference of one centimeter of water applied is an insignificant difference.

Consider that if the SWBM were to model irrigation applications within +/- one irrigation

145

event, then the interval of estimation would be 12.6 +/- 2.5 cm. The use of daily rain
gauge data, then, is a viable option for the SWBM if the scale of estimation is at the

county, watershed, or state level.

Table 5.13. SWBM irrigation estimates using 1997 reported acreage with a) NEXRAD
precipitation (R) and b) Nearest Rain Gauge precipitation (G) as precipitation input.

 

 

Precipitation Source Total (Acre-Inches) Avg. Amount (cm.)

 

NEXRAD 1008205 12.6

 

 

 

 

 

Gauge 945022 11.8

 

 

5.9 Summary, Conclusions, and Recommendations

This chapter has presented an analysis of the NEXRAD precipitation data and it’s
correlation with rain gauge observed precipitation amounts. As a basis for comparison,
weather station data from a network of volunteer weather reporters was obtained and
summarized. The spatial distribution of station data was shown for each of the three
years 1996 -— 1998. Statistical summaries and spatial distribution of the NEXRAD data

was also presented.

Some general conclusions can be drawn from this analysis.

I. NEXRAD data for May 1996 and May 1997 appears to show some systematic bias as

a function of distance from the radar site. A clear range line can be seen at the outer

146

range of the two radar ranges in southeast and southwest Michigan.

In comparison to rain gauge data, NEXRAD estimates were consistently lower for all
three years (1996 — 1998). There is also a spatial pattern to this bias. NEXRAD
estimates for the Upper Peninsula and northern lower Michigan are much lower
compared to rain gauge estimates than the same ratio for southern and central
Michigan sites. The gauge to NEXRAD ratio averaged 1.4 for 1997, with most of the
southern and central Michigan stations range from 1.2 to 1.3. This indicates a 20 to
30% higher precipitation measurement by rain gauges as compared to NEXRAD

values.

Five counties were simulated to assess the potential for the use of station data as a
source of precipitation data for the SWBM. At the county total level, results showed
good agreement estimates made using NEXRAD estimates. The slightly lower
estimates obtained using station data are a function of the higher precipitation
estimates from stations compared to NEXRAD. If station data and NEXRAD totals
are similar in value, then irrigation estimates should also be very similar between the

two data sources.

A data smoothing procedure was used to assess the impact of using multiple stations
for precipitation estimation, rather than the nearest station. Results (compared to
NEXRAD estimates) were better using smoothed data than simply using the

precipitation from the nearest station.

147

5. For county or larger scale estimates, the use of daily rain gauge data from the nearest
rain gauge will provide irrigation estimates similar to NEXRAD data. Results may be
slightly improved if a kriging procedure is used to incorporate data from multiple
stations. Given a reasonably dense rain gauge network, however, adequate results
can be obtained using the nearest rain gauge as the precipitation source. This is a
viable option for states or regions that do not have access to NEXRAD data or

technical capabilities for processing and storing this data.

6. If estimates are to be done at the sub-county level, the only logical source of
precipitation data is NEXRAD, due to the potentially high variations in seasonal
totals seen in distances of only a few kilometers. Further analysis is recommended,
however, to evaluate the accuracy of the NEXRAD data in terms of quantity
estimates and the spatial distribution of these estimates. The high spatial and
temporal resolution of this data source makes comprehensive ‘validation’ a difficult
task. Because of the many potential uses for this data, however, continued
assessment and improvement is expected as the NEXRAD product is further

developed.

148

CHAPTER VI

SUMMARY AND FINAL ANALYSIS

In this concluding chapter, section 6.1 presents a brief review of the research objectives.
Section 6.2 summarizes the soil data analysis and section 6.3 summarizes the
precipitation data analysis. Section 6.4 presents an analysis relating to model sensitivity
to management parameters. Section 6.5 discusses total water requirement as a means of
assessing irrigation need. Section 6.6 presents results relating to the correlation of
irrigation amount and precipitation levels. Section 6.7 discusses the advantages and
disadvantages of various water use estimation methods. Study limitations are discussed
in section 6.8, and issues for further research are discussed in section 6.9. Section 6.10

presents final comments.

6.1 Research Objectives

Michigan’s water use reporting law currently requires that irrigation water use by
agriculture is monitored and reported at county, watershed, and statewide levels. The
Michigan Department of Environmental Quality (MDEQ) administers the water use

reporting program. Under the law, water users who have the capacity to withdraw over

149

100,000 gallons per day averaged over any 30-day period during a year to report their water

use to the MDEQ.

In 1994 and 1995, water use information was collected by surveying all to irrigators in the
state. This process was time-consuming, and the survey response rate was less than desired.
With the availability of high-resolution precipitation data from the NEXRAD radar system
of the National Weather Service in 1996, MDEQ decided to pursue a different approach to
water use estimation for agricultural irrigation. The use of a soil water balance model
(Ritchie, I972, 1986, and 1998) to model water use requirements was proposed. It was
hypothesized that NEXRAD precipitation data, combined with other spatially referenced
input data, could be used to run the soil water balance model to simulate soil water
conditions and irrigation requirements over the course of a growing season. Using the
revised method, irrigators are required to report only the crops grown and the acreage of
each crop for each year. This greatly lessens the reporting requirements, and makes it
possible to collect acreage data during the current growing season, rather than at the end of

each agricultural year.

As an initial test, the SWBM method was used to estimate irrigation water for the partial
set of inigator forms that were returned for the 1997 season. A number of questions
arose during the development and implementation of the SWBM method. While the
SWBM method offers great potential in terms of efficiency and accuracy improvements,
it also requires detailed input data from multiple sources. Detailed soil and precipitation

may not be available for other US. states that could potentially use the SWBM method

150

for water use estimation. The primary objective of this research was to assess the
potential for using ‘simplified’ input data for soil and precipitation to run the SWBM.
This analysis concentrated on assessing the sensitivity of individual soil map units to
irrigation requirements and the use of readily available rain gauge data for precipitation
estimates. The result of this analysis, it was felt, would be useful to current users of the
SWBM as well as for resource managers in other states or regions who are responsible
for water resource management and analysis and could potentially use the SWBM for
their state or region. A secondary objective of this research was to assess the degree of
correlation between NEXRAD estimated precipitation amounts and rain gauge

observations.

6.2 Soil Data Analysis

Chapter IV discussed the soil data analysis. The STATSGO database (N RCS, USDA,
1994) provided a data set of spatially referenced soil information on a statewide basis. A
number of data fields required by the SWBM were derived from the STATSGO database,
as discussed in chapter 3. To assess the variability of each soil map unit in terms of
irrigation requirements, a 30 year data set of historical weather was used to simulate
irrigation requirements for 4 crops for each of the 190 soil map units in Michigan. Tests
for non-different results of the paired time series data were carried out to identify map

units showing insignificant differences in irrigation requirements.

151

The simulation results showed that for a given crop, irrigation requirements were similar
for many map units. Four or five soil groupings could be created that would include
about 95% of the land area in Michigan, with most soils (about 90%) in the major group.
Groupings differed by crop, however, meaning that, in an operational setting, it would be
necessary to maintain soil groups for each crop or crop type. The need to maintain soil
groupings by crop type is inconsistent with the original goal of simplifying the soil data
parameters, and therefore it was recommended that major soil groupings only be used in

situations where more detailed soil data is not available.

6.3 Precipitation Data Analysis

The main objective of the precipitation data analysis in chapter V was to compare
precipitation data obtained from NEXRAD radar with rain gauge observations from a
network of reporting weather stations around the state. Temperature and solar radiation
data, because of their general availability and reliability, were not explicitly considered in
this research. Because of the various technical, storage, and availability issues regarding
the NEXRAD data, the investigation of the use of a surrogate data set for precipitation
data (gauge data) was considered pertinent to this research. In addition to it’s potential as
a data source, the rain gauge data also provided a means of testing the NEXRAD data for

systematic biases.

152

Statistical summaries of the station and NEXRAD data were presented. Maps showing
the spatial distribution of rain gauge data were produced for comparison to the NEXRAD
data. To assess the spatial variability and correlation of weekly rainfall, the coefficient of
determination (r2) for each gauge in relation to all other gauges was calculated and
assessed. Comparison of the NEXRAD data to rain gauge data revealed a generally
lower level of precipitation estimated in the NEXRAD data compared to the observed
rain gauge levels. The bias was more pronounced in northern lower Michigan and the
Upper Peninsula of Michigan (Figure 5.1 lb). Although most stations reported more
rainfall than NEXRAD, some reported less, so the bias was not evident for all rain gauge
sites. Spatial bias was evident for the three years examined, but the extent and pattern of

spatial biases differed in each year.

Within a defined region, such as a county, a high degree of variability in seasonal totals
of precipitation was reported by NEXRAD (Table 5.12). Although not yet validated, if
this variability is indicative of reality, then there is a definite need for high-resolution
precipitation data such as NEXRAD for detailed modeling efforts. Such variability is
expected on a daily basis, due to the predominance of rainfall from isolated
thunderstorms, but was not expected by the researcher for seasonal totals. As the time
frame of analysis is increased, it was thought that values of adjacent or nearby cells
should be closer together assuming that the occurrence of rain in any one cell is a random
event for small storms or a related event for larger storms. Further research is necessary
to investigate the variance of precipitation over small areas to determine the true nature of

this phenomenon.

153

SWBM simulations using rain gauge precipitation data were run for five counties and
results compared to the estimates using NEXRAD precipitation data. It was found that
the irrigation estimates for county level totals were quite close to results obtained using
NEXRAD precipitation estimates. Differences were largely a function of the differences
in rainfall estimates between the two data sources. A smoothing technique that used
precipitation data from all nearby stations to compute a weighted average precipitation
value was found to provide better results than using the nearest station data. At the
County and watershed levels, it was concluded that the use of daily precipitation data
from a well-designed network of weather stations would provide irrigation estimates very
similar in value to those provided using NEXRAD data. Because of the variability in
precipitation at the 4 km. scale, however, station data should only be used for large-scale

(county, watershed, or state) estimates.

In summary, systematic biases were found in the NEXRAD precipitation data compared
to rain gauge precipitation amounts. More research is necessary to assess the NEXRAD
estimates both in terms of amount (cm. of rain) and in terms of the spatial distribution of
the precipitation. The decision to use NEXRAD or station data for precipitation input to
the SWBM should be a function of many factors. The availability of NEXRAD data is
the first major factor. If NEXRAD data are available, then the cost and effort involved in
processing, storing, and maintaining the NEXRAD data are factors to consider. If spatial
and temporal accuracy at sub-county levels is desired, then NEXRAD data are the only

viable option if the spatial variability in precipitation amounts suggested by the

154

NEXRAD data is true, and if this variability is reflected in differences in irrigation
amounts between irrigators. If estimates are to be made at the county or greater level on
a seasonal basis, then the use of precipitation data from a well-designed rain gauge
network is a viable option. An important consideration is the level of accuracy that can
be associated with the NEXRAD precipitation estimates. Comparison with rain gauge
data would indicate that a high level of accuracy of NEXRAD data should not be

assumed, but rather be the subject of further investigation.

6.4 Model Sensitivity to Management Parameters

Differences in management strategy may also impact the total irrigation water applied to
a crop. Aversion to risk, labor availability, and economic factors impact an irrigator’s
decision of when and how much to irrigate a crop. Swaney et a1. (1983) evaluated
irrigation strategies for soybeans with the goal of profit maximization. Maximum profit
was found at an irrigation threshold of 70% and 1 cm. irrigation per application. Near-
optimal results were found with an irrigation threshold of 60% and 3 cm. applications.
The irrigation threshold and amount of water applied are management parameters that are
defined for each crop in the SWBM. The default values used in this research are listed in
Appendix B. In the default settings, the irrigation threshold is 50% for all crops and the
irrigation amount varies from 1.5 to 2.5 cm. per application. To examine the sensitivity
of the SWBM to changes in these management parameters, simulations were run for the

1997 data set using the two irrigation strategies defined above (70% and 1cm. and 60%

155

and 3 cm.). The results of these simulations were compared to the results obtained using

the default values (Table 6.1).

The sensitivity of irrigation estimates to different management strategies can be seen in
the results shown in Table 6.1. The default strategy is the most conservative in terms of
water use. Strategies 11 and III represent the strategies found by Swaney et al. (1983) to
be optimal for soybeans. Increasing the irrigation threshold from 50% to 60%, with a
fixed application rate of 3 cm. resulted in a 23% increase in irrigation amount. Strategy
111 shows a 45% increase in irrigation amounts compared to the default strategy. While
results obtained by researchers for soybeans on a sandy soil in Florida are not directly
applicable to all Michigan crops and soils, these results show the potential differences in
irrigation amount estimations based on changes in irrigation management strategies. The
yield response to different water management strategies must be known in order to
determine economically optimal management strategies. In terms of implementation of
the SWBM method, these results indicate the importance of knowing the management

strategies employed by irrigators in the region being modeled.

Table 6.1. Total irrigation requirements under different management strategies using
1997 irrigator records and NEXRAD precipitation data.

 

 

 

 

 

 

StrategL Threshold (%) Amount (cm.) Irrigation (cm.) Ratio
1 (Default) 50 Variable 1 108026 1.00
|| 60 3 1358467 1.23
111 70 1 1605303 1.45

 

 

 

 

 

 

156

 

6.5 Total Water Requirements

If the timing of precipitation during the course of a growing season is not a critical factor
in determining irrigation requirements, then it could be hypothesized that the total water
requirements of a particular crop/soil combination would be relatively constant on an
annual basis. That is, in dry years, more irrigation would generally be required than in
wet years, but the gal water applied (precipitation + irrigation) would be approximately
the same. If the timing of precipitation is a critical factor, however, we would expect to
see variation in irrigation amounts, even in years with similar precipitation levels. Figure
6.1 shows the total water requirements for corn on soil M1001 for the 30 years of weather
data from the Lansing station (1961 — 1990). The yearly data is sorted in ascending order
of total seasonal (May — August) precipitation. Total seasonal precipitation and total

irrigation water applied, as simulated by the SWBM is shown.

157

Total Water Requrements: Corn on Soil M1001

7O ,

1 60

1 50 1
l l
l . s ‘
. 40 1 . .

1 - illmgation1
‘ : ‘aPrecip. “'

20

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Amount (cm )
_l (A)
O O O
1 E22:—

21
23

25 .4 ..
27

29 .-»~

 

Figure 6.1. Total water requirements for corn on soil M1001 for the 30 year simulations
using weather data from the Lansing, MI station 1961 — 1990. Year is sorted in
ascending order of precipitation amount. Precipitation is May - August precipitation.

Most of the irrigation of corn in Michigan occurs in the months of July and August, when
potential evaporation and transpiration are high and rainfall can be sporadic. Irrigation
amounts, therefore, should be more highly correlated with July and August precipitation
totals than with total seasonal (May — August) totals. Figure 6.2 shows July and August
precipitation in ascending order on the x axis and total irrigation for each of the 30 years

simulated.

158

 

Total Water: Corn on Soil M1001

4o ' I Irrigation

in July and Aug.
1 Precip.

Amount(cm.)
—x _. N N 0.)
01 o 01 o 01 o
7 T
l
l
l

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

25
28

Year

 

 

 

Figure 6.2. Total water requirements for corn on soil M1001 for the 30 year simulations
using weather data from the Lansing, MI station 1961 - 1990. Year is sorted in
ascending order of precipitation amount. Precipitation is July and August precipitation.

The results shown in Figures 6.1 and 6.2 illustrate the generally negative relationship
between precipitation totals and irrigation amount resulting in a somewhat constant total
water requirement, with slightly higher amounts for wetter than average years. In Figure
6.2, for example, it can be seen that July and August precipitation plus irrigation
generally ranges from 25 — 30 cm for corn on soil M1001. There are exceptions to every

general rule, however. Year 18 in Figure 6.2 shows greater irrigation amounts than other

159

years with similar precipitation amounts. The results shown in Figures 6.1 and 6.2 apply
only for a specific crop (corn) on a particular soil group (M1001). To show the potential
differences between crops, Figure 6.3 shows the same results for soybeans on soil
(M1001) using July and August precipitation. The shallower rooting soybeans extract
more water from the upper soil layers than com, and therefore total water irrigation water
requirements, as simulated by the SWBM, are higher. Less variability in total water
requirements is seen for soybeans as compared to corn. Variations in irrigation water

applied of 5 cm. for similar precipitation levels are common however

 

Total Water: Soybeans on Soil M1001

1 I irrigation

l

l

40 July and Aug. Precip. 1
1 1

Amount (cm.)
8

20

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

21
23
25
27
29

 

 

 

Figure 6.3. Total water requirements for soybeans on soil M1001 for the 30 year
simulations using weather data from the Lansing, MI station 1961 — 1990. Year is sorted
in ascending order of precipitation amount. Precipitation is July and August totals.

160

The variations in irrigation levels seen in years that experienced similar precipitation
levels illustrates the potential advantages of a dynamic, physically-based soil water
balance model for accurate assessment of irrigation water requirements. Simplified
empirical approaches may provide reasonable results, but unless the empirical model was
properly and completely specified, estimates could be incorrect in outlier years with

irregular precipitation patterns.

6.6 Irrigation Amount Versus Precipitation

The discussion in the previous section showed the generally negative correlation between
seasonal precipitation totals and irrigation water requirements for a given crop/soil
combination. To assess the impact of consistent systematic bias in the NEXRAD
precipitation data and it’s impact on irrigation water use estimation a series of simulation
were run in which the NEXRAD precipitation estimated were multiplied by a scalar. The
scalar used was .5, .7, l, 1.3, 1.5, and 2.0. The irrigation acreage used was the 1997

irrigator data set. Results are shown in Figure 6.4.

161

 

 

 

 

 

Total Irrigation Vs. Scaled Precipitation
1600000

8 1400000 ~4
8
g 1200000 .1
0
6 1000000
5.
1: 800000 ..
8
.5, 600000
-'_= 400000 1»
6 200000*
.—

0

0.5 0.7 1 1.3 1.5 2
Precipitation Scalar

 

 

 

Figure 6.4. Total simulated irrigation (acre-inches) versus scaled NEXRAD
precipitation with scalar applied of .5, .7, 1.0, 1.3, 1.5 and 2.0.

As expected, total simulated irrigation requirements decrease as precipitation amounts are
increased, but not in direct proportion to the change in precipitation. A 30% increase in
precipitation, for example, resulted in an 18% increase in total irrigation amount. A 50%
increase in precipitation resulted in a 27% decrease in total irrigation. Conversely, a 50%
decrease in precipitation resulted in a 36% increase in irrigation, and a 30% decrease in
precipitation resulted in a 21% increase in irrigation. Total irrigation estimates then, are

inelastic and negatively correlated with respect to changes in precipitation amounts.

162

Results in the preceding paragraph should not be taken to imply that a simple relationship
exists between SWBM irrigation estimates and precipitation amounts. A scalar applied
to all NEXRAD cells as in the previous paragraph does not alter the timing of
precipitation, another important factor in determining irrigation amounts. In addition,
analysis of aggregated irrigation amounts masks the variation in irrigation due to

differences in soil characteristics.

To examine the impact of precipitation timing and differences in soil characteristics on
the relationship between irrigation estimates and precipitation amounts, scatter plots of
1997 irrigation requirements versus precipitation at the NEXRAD cell level were
prepared for field corn. Results are shown for May to August precipitation totals (Figure
6.5) and July and August precipitation (Figure 6.6). In these figures, each point
represents the amount of irrigation (cm.) applied for the given crop in a NEXRAD cell.
At the NEXRAD cell level, differences in the timing of precipitation and differences in
soil characteristics are isolated. To a lesser extent, differences in temperature and solar

radiation also impact the simulated irrigation requirement.

163

lrrigation Amount Versus Precipitation (Field Corn) 1

 

 

 

 

 

 

1 20 g
18
E JeeWeSo» T" .
16 7 77 7 V7 -
A e’ “We. .
1 E 14 f 7 - 7 -
3
E12 "7.-“ ””*-?9~4~'°’ 3 H
1 3
I 2 10 .m,“ “More on.
C
O 8
i a ° 9 0‘ O 90 e
l a
l E 6
— OO O .
4-~—— _______ _
, 2 - - l
O l = T
0 5 10 15 20 25 30 35 40 45

May - August Precipitation (cm.)

 

Figure 6.5. Irrigation applied (cm.) versus NEXRAD precipitation for field com, 1997.
Each point represents a NEXRAD cell. Precipitation is May to August seasonal total.

Visual inspection of Figure 6.5 reveals that there is not a discernable negative correlation
between simulated irrigation amount and May-August precipitation totals at the
NEXRAD cell level. A strong negative correlation is seen, however, when irrigation

amount is plotted against July and August precipitation totals (Figure 6.6).

164

Irrigation Amount Versus July and August Precipitation

 

 

 

 

(Corn)

250 7
_. 20.0 7 .
S. M e 1
E ,
.5 10.0 , " . f.

i g ’ on»

E 5.0 _ 7 777+ 7* . .

0.0 1 ‘ l

0 5 10 15 20 25 30

July and August Precipitation (cm.)

Figure 6.6. Irrigation applied (cm.) versus NEXRAD precipitation for field com, 1997.
Each point represents a NEXRAD cell. Precipitation is July and August totals.

As seen in Figure 6.6, a considerable range of precipitation levels can apply to a given
irrigation level. For example, irrigation of 10 cm. was applied over a range of
precipitation from 12 to 22 cm. These results show the impact of local differences in the
timing of precipitation, other climate factors, and soil characteristics on the estimation of
irrigation water requirements. The impact of a scalar applied to precipitation on these
results would be to shift the locus of points on the scatter plot in a horizontal direction.
These results again underscore the potential advantages of the use of a detailed and

dynamic modeling approach for the estimation of irrigation water requirements. A more

165

simplified empirical approach would not be able to incorporate the many local factors

affecting irrigation water needs.

6.7 Options for the Estimation of Irrigation Water Use by Agriculture

Four clear options exist for the estimation of agricultural irrigation water use in

Michigan. These options are:

1. Direct survey of all irrigators, as done in 1994 and 1995.
II. Reporting of county level irrigation levels by MDA/Extension agents.

III. Detailed modeling using the SWBM

IV. Simplified empirical approach using multi-year SWBM results.

Of these options, the first two are ‘low-technology’ approaches and the third and fourth
are ‘high-technology’ approaches. Option 11 is introduced here as an alternative to the
labor-intensive and bias-prone direct survey method. In option 11, field personnel of the
Michigan Department of Agriculture or County extension agents would be responsible
for reporting to MDEQ estimates of irrigation levels in their respective regions. These
estimates would be based on the local knowledge of field personnel and their interactions
with irrigators in the region. This is a viable option if field personnel are intimately
familiar with irrigation practices in their local area and prepare the estimation of

irrigation amounts for each crop in a serious manner with due regard for accuracy. A

I66

positive feature of option 11 is that it is much less labor-intensive and less invasive an
approach than option 1. Mailing of surveys to irri gators is not necessary under option 11,
rather, a survey form would be sent to each responsible field personnel.
Administratively, option [I is much preferable to option I. From the standpoint of
estimate accuracy, option II is also preferable i_f field personnel strive to prepare unbiased
and accurate estimates for all crops irrigated in their region. For these reasons, Option II

is recommended over option I as a low-technology method.

In terms of simplified approaches (option IV), the use of simplified soil data groups was
discussed in chapter IV, the use of rain gauge observed precipitation was discussed in
chapter V, and the suggestion of an empirical approach using total water requirements
was made earlier in this chapter. As discussed in previous sections, however, any
attempt to simplify data inputs or the water use model will result in potentially less
accurate estimates than a detailed modeling approach. This discussion assumes that the
simpler empirical methods would be based on results of multi-year runs of the detailed
SWBM and that ‘detailed’ data is more accurate than ‘simplified’ data. Any simpler
empirical method (option IV) would be less accurate than the use of a detailed model
(option III) because of the potential for the detailed model to handle variable weather and
soil conditions more accurately. In addition, from a scientific standpoint, the continued
development of a sound functional model is much preferable to the development of
empirical (regression-based) models of water use. Therefore, option 111 (detailed

modeling) is recommended as a high-technology approach.

167

In comparing the recommended low-technology approach (option 11) versus the
recommended high-technology approach (option 111) it is fair to say that both methods
could be used to provide ‘reasonable’ estimates of agricultural irrigation water use on a
county and watershed level. The continued development of a modeling approach (option
III), however, potentially offers many more benefits than survey approaches (option 11).
Irrigation estimates are possible at a 4 km. resolution using option III, but limited to
county and watershed level resolution using option II. In addition, estimates can be made
during the course of a growing season using option III, while option 11 is limited to year-
end estimates. Option III, therefore, has far greater potential for use as a monitoring and
management support tool than other options. Derivative benefits exist for option 111 as
well. Consider, for example, the use of the SWBM to estimate the potential for nitrate
leaching. While survey approaches are possible in the case of irrigation water use, it
certainly is not feasible in the case of the estimation of nitrate leaching. Irrigation water

use is only one of many potential products of a detailed modeling approach.

An additional and important feature of a detailed modeling approach (option III) is that
such an approach provides a documented and defensible scientific methodology for the
estimation of irrigation water use. This makes possible the use of the scientific method
for the improvement of the model, in which hypotheses can be tested against observable
evidence in a disciplined manner, with each step in the process made explicit. Future
advancements of the theoretical knowledge base through basic research coupled with the
knowledge and experience of applied practitioners will serve to improve the accuracy and

precision of the SWBM irrigation water estimates. In addition to advancement in

I68

scientific theory, the continued development and enhancement of new technology will
increase the availability of high-resolution data. Availability of such data will allow for
the testing and validation of models over a broader range of conditions. As mentioned
above, it will be possible in the future to run the SWBM throughout the growing season
with near real-time data acquisition, allowing for a more thorough validation of this
approach, and ultimately better monitoring of water use and more responsive water

management.

Of the four options presented, option 111 (detailed modeling using the SWBM) is the
recommended approach. This approach, however, requires a long-term commitment to
the continued development and use of technology for resource management by the
participating organizations. The potential benefits of this approach are many if such a
commitment is made. In the absence of support or funding for a high-technology
approach, option II (modified survey) is recommended as a viable low-technology

alternative.

6.8 Study Limitations

The processes being modeled in the SWBM are very complex and highly dependent on
soil and weather conditions that are also highly variable over time and space. The
researcher must segment these continuous processes into discrete units in time and over

space. In some cases, such as the soil data in this study, the segmentation scale is defined

I69

by available data. As discussed in chapter II, the characteristic scale of the processes
being modeled is an important consideration when designing and developing models. The
physical laws that govern natural systems, such as the force of gravity on water flow,
make it possible to build mathematical models that closely approximate ‘reality’. It must
be recognized, however, that, any model, by definition, is a simplification of reality.
Simplification occurs in the mathematics of the model itself and in the values of

independent variables used by the model.

In the case of this study, the soil water balance model used has been shown to produce
good results in a number of experimental settings. Accurate results of the SWBM model
are dependent on accuracy of the data inputs to the model, especially pertaining to soil
characteristics and precipitation amounts. STATSGO soil data and NEXRAD
precipitation data, despite their relatively high resolution when compared to a county or
watershed, do not capture data at the resolution of the individual irrigator (the field level).
Accordingly, differences will exist between individual irrigators and model results,
simply due to differences in the measurement scale of input data. A direct comparison of
the SWBM method results with individual irrigator results is therefore not possible, due
to these differences. Because of the scale of the data inputs, the results of the SWBM
method in terms of irrigation requirements are not meant for site specific management.
Levels of irrigation applied should be interpreted as estimates based on the best sources
of input data available, but are not meant to imply recommended irrigation amounts for a

particular crop or county.

I70

Integrated modeling such as performed in this research is by nature a multi-disciplinary
effort. It is not possible to assess, in great detail, the many factors and research issues
associated with each component of this study. Because of the statewide scale and many
variations in crop and soil types, ‘conclusions’ of this research are often limited to
statements that identify general patterns and relationships in the data and model results.
The issues raised in this research, however, should provide a basis for further research in

many areas.

6.9 Issues for Further Research

Because of the pre-operational status of the SWBM method, comparison of SWBM
estimates with actual irrigator records on a county level was not possible during this
research. To better assess the SWBM results, it is recommended that voluntary
submissions of irrigation records be obtained from irrigators who maintain accurate
records during the first three years of implementation. As discussed above, SWBM
results can vary significantly based on variations in management strategies. Comparison
of SWBM results with irrigator records is necessary to determine if there is any
systematic bias in the SWBM estimates. If so, adjustment of the management parameters
to more closely reflect those of irrigators may be necessary. When assessing the
accuracy of the SWBM, the discontinuous nature of irrigation application must be kept in

mind. If, for example, irrigation is applied in amounts of 2.5 cm. with each application,

171

then if irrigation counts are modeled to +/- 1, irrigation amounts are modeled to +/- 2.5

cm., a range of5 cm.

In the absence of site-specific data for a more detailed validation, results of the SWBM
method should be monitored and compared to estimates of irrigation amounts obtained
from other sources. County extension agents involved with irrigation, for example, can
provide assessments of the model results in relation to their knowledge of local
operations. On a statewide basis for 1997, the average irrigation amount estimated by the
SWBM for corn was 13.5 cm. (5.3 inches) with a range of 6.8 to 17.8 cm. at the county
level. For soybeans in 1997, the average amount of irrigation was 17.5 cm. (6.9 inches)
with a range of 9.6 - 20.3 cm at the county level. Aggregated numbers such as these can

be used as a partial validation of the SWBM method on an on-going basis.

Continued assessment of the accuracy of NEXRAD precipitation data should be an
important topic of research that would benefit not only this research effort but all that use
NEXRAD data. Verification of the high degree of variability in precipitation estimates
between adjacent or nearby cells would provide conclusive evidence for the need to use
NEXRAD data in any modeling effort that attempts to accurately estimate at a sub-
county spatial resolution. The systematic bias in NEXRAD data and discrepancies
between NEXRAD and rain gauge precipitation levels, as discussed in chapter V, should

also be a topic of further investigation.

I72

In terms of the SWBM model itself, the current implementation uses a simplified
approach for the estimation of leaf area index (LAI). Because of the importance of LAI
in determining potential evaporation, the enhancement of this portion of the SWBM
would potentially lead to more accurate results. The use of remotely-sensed or field-
measured LAI values and the calibration of the SWBM to these values over the course of

a growing season is a possible method of improving LAI estimation.

6.10 Final Comments

The implementation of the SWBM in an integrated modeling environment represents an
effort to utilize a field-scale soil water balance model for large-scale water use estimates
using the most detailed data available as model inputs. In this research, the SWBM is run
at the finest resolution allowable by the data (4 km.) with larger scale estimates (county
and watershed) made by aggregating results obtained at the finer scale. Accuracy is a
function of two main factors; the ability of the model to estimate changes in dependent
variables using all relevant factors (i.e. correct specification of the model), and the
measurement precision and accuracy of the independent variables. Given a theoretically
sound underlying model, the accuracy of data inputs plays the major role in determining
the accuracy of the modeling effort. The state-wide scale of this research project made it
necessary to substitute 4 km. resolution NEXRAD precipitation data, 1:250,000 scale
soils data, and general assumptions regarding management practices for variables and

processes that, in reality, differ at the field or sub-field scale. Results of the SWBM

I73

should therefore be viewed in the context of the precision with which the independent

variables can be measured.

The ‘information age’ of the 21St century will offer many new opportunities for resource
managers and researchers. The ability to monitor, analyze, and model natural resources
at high resolutions over large spatial areas will be greatly enhanced as we develop and
utilize these new technologies. The integrated modeling environment and research
described in this dissertation is a small, but hopefully positive step towards the
development of greater capabilities for resource monitoring, modeling, and management.
There are many challenges facing resource managers and researchers, but each challenge
should offer an opportunity... an opportunity to more fully understand and better

manage, conserve, and protect the earth’s natural resources.

I74

APPENDICES

APPENDIX A

APPENDIX A

THE RITCHIE SOIL WATER BALANCE MODEL

The core of the irrigation modeling component of this research is a soil water balance
model (SWBM). References pertaining to the development and validation of the SWBM
were cited in chapter 2 (Ritchie, 1972, Richardson and Ritchie, 1973, Ritchie, 1985, and
Ritchie, 1998). Mathematical documentation of the SWBM is given in this Appendix.
Additional details of the SWBM are given in Ritchie (1972), Ritchie (1985), and Gerakis

and Ritchie (1998).

The SWBM is a multiple layer model that allows calculation of the vertical distribution

of soil water as well as the total soil water content. The routines involved in the soil

water balance simulation are described below in the order in which they are performed.

Step 1. Calculate potential evaporation and soil evaporation.

Potential evaporation (E0) is a function of the air temperature during daylight hours and

solar radiation levels. Air temperature during daylight hours (TD) is approximated by a

weighted mean of daily minimum (TMin) and maximum (TMaX) air temperatures (°C)

175

To = (TMax * 0.7) + (TMin * 0.3) (Al-l)

E0 is calculated as a function of two components, a radiation component and an

aerodynamic component:

Where

RadCom = (1 - GODPG) * Net24 / LatHeat (Al-2)
AeroCom = (2.8 * vpd + 3 * vpd " 2) * GODPG

E0 = (RadCom + AeroCom) / 10

RadCom = Radiation component

AeroCom = Aerodynamic component

E0 = potiential soil and plant evaporation (cm day—l)

SR = Solar radiation (MJ 111‘2 day")

Tnewc = 44 'Dew point temperature constant
Net24 = SR * 0.61 - 1.5

LatHeat = 2.501 - TD / 423 'Latent heat

Tpewg = 20.5 - 0.01184 * (TM,n - TDcwc) A 2 'Estimated dew pt. temp

.7209-3806/ T 273 - 2 /T + 7 "
Vpday=0.l "‘e‘18 (0+ >22153 (D 23) 2)

“8.7209 - 3806/ (T + 273) - 222153 / (T + 273) A 2)
1 *
- e DewE DewE

vpe = 0
vpd = vpday — vpe

Press = 97.8

176

GODPG = (0.598 - 0.00003 * Press) - 0.017 * TD + 0.000149 * To A 2

Potential soil evaporation (E05) is a function of EO and the current leaf area index (LAI).

130S = E0 * e“ ‘ LA” (Al-3)

Where

LAI = leaf area index

Step 2. Calculate root growth for the day.

Root mass and distribution are important parameters for the estimation of the amount of

water taken up by the roots. In this version of the SWBM, new root growth (N RG, grams

m'2 day") is estimated as a function of solar radiation (SR) and the current leaf area index

(LAI):

NRG = (SR * 0.21) * (1 - e“°-" " LAl’) (Al-4)

The root length distribution is calculated for each soil layer as a function of the new root

growth, layer water content and a root water uptake coefficient (RWUCON). Root

length distribution is used later in the calculation of water uptake (see step 7).

I77

Step 3. Calculate ponding amount.

Ponding refers to the accumulation of standing water on the soil surface. If precipitation
occurs or the current ponding depth is greater than zero, this routine is called. The
ponding routine calculates ponding based on the rainfall intensity and a function for
hydraulic conductivity. It calculates the daily amounts of infiltration, runoff, and changes
to the height of the ponded water. The ponding routine also calls a routine for soil
compaction due to rainfall action. The NEXRAD hourly rainfall rates are used on this
routine to determine the maximum rainfall intensity during the precipitation event. This
is the only routine in which hourly values are used, in all others, the time step is daily.
The result of the ponding routine is a distribution of incoming precipitation to runoff,
infiltration, or additional ponding. If irrigation occurs for the day, the irrigation amount

is added to infiltration.

Step 4. Calculate the downward flow of water.

Based on the soil water content at each layer in relation to the layers drained upper limit

(DUL), saturation level (SAT), and hydraulic conductivity (KsMatrix), this routine

calculates the amount of water infiltrating to each soil layer from layers above.

178

Step 5. Calculate soil evaporation and upward water flow.

This routine uses the available energy for soil evaporation (130,) to estimate the amount

of water moving upward and the amount evaporated from the surface layer.

Step 6. Calculate root water uptake.

The amount of water taken up by the roots is a function of the potential plant evaporation
and transpiration for the day, bounded by potential evaporation minus soil evaporation
(see step 1). Potential plant evaporation (EOp) is a function of the potential evaporation

and the leaf area index:

(Bop) = .4 * E0 * LAI (Al-S)

Water is removed from each soil layer as a function of the soil water content, the layer
lower limit (LL), the root length distribution of the layer (step 2), and the root uptake
coefficient (RWUCON). The total amount of water taken up by roots is equal to EOp if
the plant can transpire at potential levels. Otherwise the total amount of water taken up

by the roots is equal to E0 — E05, the potential evaporation minus soil evaporation.

Step 7. Determination of irrigation amounts.

I79

The decision of whether or not to apply irrigation water is based on the ratio of

extractable soil water to the potential extractable soil water, calculated as:

ESW = 2 (swi - LL,)* Di for i=1 n. (Al-6)

PESW = Z (DUL, — LLi)* Di for i = l n.

D1 = ESW/PSW

Where
SW, = Current volumetric soil water content, layer i
DUL, = Drained upper limit, volumetric soil water, layer i
LL, = Lower limit, volumetric soil water, layer i
D1 = Drought Index
n = the total number of layers tested (see below).

The current soil water content is the current value for the day being simulated. The
drained upper limit (DUL) is the highest field-measured water content of a soil after
thorough wetting and draining until drainage becomes negligible. The DUL corresponds
to water content at ‘field capacity’. The lower limit is the lowest, field-measured
volumetric water content of a soil after plants stop extracting water due to premature
death or dormancy as a result of water deficit. The LL corresponds to the ‘permanent

wilting point’.

I80

The number of layers tested (n) in Equation A1-6 is a variable that can be set by the
operator. The default value is 6 (depth = 57 cm.). Any soil layer with a value for root
length distribution of greater than or equal to .05 is included however. The Drought
Index (D1) is the variable used to test for the irrigation threshold value (Tc). The default
value for Tc is .5. That is, if D1 is less than or equal to .5, then an irrigation event is
triggered. The irrigation amount is a specified value that differs by crop, see Appendix b
for a complete listing. A typical value is 2.5 cm. for each irrigation event. Both the

threshold value (Tc) and the irrigation amount are management parameters.

l8I

APPENDIX B

APPENDIX B

CROPS AND DEFAULT MANAGEMENT PARAMETERS

Default settings for management parameters used in the SWBM are listed below for each

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

crop.
‘ ID Crop , Planting DOY MaxRootDepth (cm) ThetaC' lrrAmOunt (mm.)]
1 Seed Corn 135 100 50 25
2 Commercial Corn 135 100 50 25
3 Popcorn 135 100 50 25
4 Sweet Corn 135 100 50 25
5 Dry Beans 135 75 50 25
6 Green (Snap) Beans 135 75 50 25
7 Navy/Kidney Beans 135 75 50 25
8 Navy Beans 135 75 50 25
9 Soybeans 135 75 50 25
10 Barley 121 100 50 25
11 Oats 121 100 50 25
12 Potatoes 135 50 50 25
13 Rye 121 100 50 25
14 Sorghum 135 100 50 25
15 Hay/Alfalfa 121 100 50 25
16 Wheat 121 100 50 25
17 Asparagus 150 100 50 25
18 Broccoli 150 35 50 20
19 Cabbage 150 35 50 20
20 Cauliflower 150 35 50 20
21 Celery 150 60 50 25
22 Carrots 150 60 50 25
23 Cucumbers 150 25 50 15
24 Eggplant 150 50 50 25
25 Gourds 150 60 50 25
26 Lettuce 150 25 50 15
27 Melons 150 25 50 15
28 Onions 150 50 50 25
29 Peas 150 40 50 22
3O Peppers 150 40 50 22

 

 

 

 

 

 

 

I82

Appendix B (eont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

lD Crop Planting DOY MaxRootDepth (cm) ThetaC lrrAmount (mm)
31 Pumpkins 150 50 50 25
32 Radishes 150 25 50 15
33 Rhubarb 150 100 50 25
34 Squash 150 70 50 25
35 Tomatoes 150 50 50 25
36 Turnips 150 25 50 15
37 Zucchini 150 50 50 25
38 Apples 135 100 50 25
39 Grapes 135 100 50 25
40 Peaches 135 100 50 25
41 Pears 135 100 50 25
42 PIumbs/Prunes 135 100 50 25
43 Sweet/T art Cherries 135 100 50 25
44 Tart Cherries 135 100 50 25
45 Chestnuts 135 100 50 25
46 Blackberries 135 100 50 25
47 Blueberries 135 100 50 25
48 Cranberries 135 100 50 25
49 Raspberries 135 100 50 25
50 Strawberries 135 35 50 20
51 Sod 121 25 50 15
52 Nursery/Floral 135 25 50 15
53 Trees/Shrubs 121 100 50 25
54 Mint 121 100 50 25
55 Parsley 125 60 50 25
56 Mustard 125 25 50 15
57 Currents 135 100 50 25
58 Vegetable Group 1 135 30 50 18
59 Vegetable Group 2 135 50 50 20
60 Vegetable Group 3 135 50 50 20
63 Sugar Beets 135 100 50 25

 

 

 

 

 

 

183

 

REFERENCES

REFERENCES

Austin, PM. and RA. Houze, Jr. 1972. Analysis of the structure of precipitation
patterns in New England. J. Appl. Meteor. 11: 926-935.

Austin, PM. 1987. Relation between Radar Reflectivity and Surface Rainfall. Monthly
Weather Review 115: 1053-1071.

Baier, W. and G.W. Robertson. 1968. The Performance of Soil Moisture Estimates as
Compared with the Direct Use of Climatological Data for Estimating Crop Yields.
Agricultural Meteorology 5: l7-31.

Bedell, DJ. and Van Til, R. 1979. Irrigation in Michigan, 1977. Water Management
Division, Department of Natural Resources, State of Michigan.

Bergstrom, S. and LP. Graham. 1998. On the scale problem in hydrological modelling.
Journal of Hydrology 211: 253-265.

Bemdtsson. R. 1988. Temporal variability in spatial correlations of daily rainfall. Water
Resources Research. 24(9): 1511-1517.

Black, T.A., W.R. Gardner, and G.W. Thurtell. 1969. The prediction of evaporation,
drainage, and soil water storage for a bare soil. Soil Sci. Soc. Amer. Proc. 33:
655-660.

Bouwer, H. 1994. Irrigation and global water outlook. Agricultural Water Management
25: 221-231.

Brady, NC. 1990. The Nature and Properties of Soils. Macmillan Publishing Company,
New York.

Collier, G.C. 1986. Accuracy of Rainfall Estimates By Radar, Part I: Calibration By
Telemetering Raingauges. Journal of Hydrology 83: 207-223.

Coupal, RH. and RN. Wilson. 1990. Adopting Water-Conserving Irrigation
Technology: The Case of Surge Irrigation in Arizona. Agricultural Water
Management 18: 15-28.

Cushman, J .H. 1986. On measurement, scale, and scaling. Water Resources Research.
22: 129-134.

Dierckx, J ., J .R. Gilley, J. Feyen, and C. Belmans. 1988. Simulation of soil-water
dynamics and corn yields under deficit irrigation. Irrigation Science. 9: 105-125.

I84

Eagleson, P. S. 1967. Optimum density of rainfall networks. Transactions of the ASAE.
3: 1021-1033.

Easterling, W. E., et al. Spatial scales of climate information for simulating wheat and
maize productivity: the case of the US Great Plains. Agricultural and Forest
Meteorology. 90: 51-63.

Feddes, R.A., P.J. Kowalik, and H. Zaradny. 1978. Simulation of field water use and
crop yield. Pudoc, Wageningen.

Frere, M. and GP. Popov. 1979. Agrometeorological crop monitoring and forecasting.
FAO Plant production and protection paper No. 17, Food and Agricultural
Organization, Rome, Italy. 64 pp.

Gerakis, A. and J .T. Ritchie. 1998. Simulation of atrizine leaching in relation to water-
table management using the CERES model. Journal of Environmental
Management 52: 241-258.

Grant, RC. 1997. Resources and Research: Redefining Water Quality Management
From a Watershed Perspective. The Environmental Professional. 19: 62 - 66.

Great Lakes Governors Task Force. 1985. Water Diversion and Great Lakes Institutions.
Final Report and Recommendations: Great Lakes Governors Task Force on Water
Diversion and Great Lakes Institutions.

Helalia, AM. 1993. The relation between soil infiltration and effective porosity in
different soils. Agricultural Water Management 24: 39-47.

Hendrick, R.L. and G.H. Comer. 1970. Space variation of precipitation and implications
for raingage network design. J. Hydrol. 10: 151-163.

Hill, R.W. 1991. Irrigation Scheduling. pp. 491 - 509. In J. Hanks and J .T. Ritchie (ed.)
Modeling Plant and Soil Systems. Series: Agronomy Monograph number 31 .
ASA-CSSA-SSSA. Madison, WI.

Hubbard, K. G. 1994. Spatial variability of daily weather variables in the high plains of
the USA. Journal of Climate and Applied Meteorology. 68: 29-41.

Hudlow, MD. 1988. Technological developments in real-time operational hydrologic
forecasting in the United States. J. Hydrol. 102:69-62.

Hudlow, MD, J .A. Smith, M.L.. Walton, and RC. Shedd. 1991. NEXRAD: New era in

hydrometeorology. In: Hydrological Applications of Weather Radar, edited by I
Cluckie and C. Collier, pp. 602 -— 612, Ellis Horwood, New York.

185

Huff, F .A. 1967. Time distribution of rainfall in heavy storms. Water Resources
Research. 3: 1007-1019.

Huff. FA. and W.L. Shipp. 1969. Spatial correlations of storm, monthly, and seasonal
precipitation. J. Appl. Meteor. 8: 542-550.

Hsu, K., et al. 1996. Precipitation estimation from remotely sensed information using
artificial neural networks. Journal of Applied Meteorology. 36: 1176-1190.

International Benchmark Sites Network for Agrotechnology Transfer (IBSNAT). 1986.
Decision Support System for Agrotechnology Transfer (DSSAT). Documentation
for IBSNAT Crop Model Input and Output Files. Version 1.0. IBSNAT
Technical Report 5. IBSNAT. Honolulu, HI.

International Joint Commission. 1985. Great Lakes Diversions and Consurnptive Uses.
A report to the Governments of the United States and Canada under the 1977
Reference.

Kerkides, P., H. Michalopoulou, G. Papaioannou, and R. Pollatou. 1996. Water balance
estimates over Greece. Agricultural Water Management. 32:85-104.

Keulen, H. van. 1982. Crop production under semi-arid conditions, as determined by
moisture availability. In: Penning de Vries FWT, van Laar HH (eds).
Simulation of plant growth and crop production. Pudoc, Wageningen. p. 87.

Kitchen, M. and PM. Jackson. 1993. Weather radar performance at long range:
Simulated and observed. .1. Appl. Meteorol. 32:975-985.

Klemes, V. 1983. Conceptualization and Scale in Hydrology. Journal of Hydrology 65:
l — 23.

Knisel, W.G. and R.W. Baird. 1969. Runoff volume prediction using daily climatic data.
Water Resources Research. 5(1):84-94.

Knox, J .W., Weatherhead, E.K., and RI 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., Weatherhead, E.K., and RI Bradley. 1997. Mapping the total volumetric
irrigation water requirements in England and Wales. Agricultural Water
Management 33: 1-18.

Kunkel, KB 1990. Operational Soil Moisture Estimation for the Midwestern United
States. Journal of Applied Meteorology. 29:1158-1166.

I86

Lafolie, F. L. Bruckler, A.M deCockborne, and C. Laboucarie. 1997. Modeling the
water transport and nitrogen dynamics in irrigated salad crops. Irrigation
Science. 17 : 95-104.

Lal, H., G. Hoogenboom, J-P. Calixte, J. W. Jones, and F. H. Beinroth. 1993. Using
crop simulation models and GIS for regional productivity analysis. Transactions
ofthe ASAE. 36: 175-184.

Lorente, J ., and A. Redano. 1990. Rainfall rate distribution in a local scale: The case of
Barcelona city. Theoretical and Applied Climatology. 41: 23-32.

Lyons, S. W. 1990. Spatial and temporal variability of monthly precipitation in Texas.
Monthly Weather Review. 118: 2634-2648.

Mack, AR. and W.S. Ferguson. 1968. A Moisture Stress Index For Wheat By Means Of
A Modulated Soil Moisture Budget. Can. J. Plant. Sci. 48: 535-543.

Madsen, H.B. and K. A. Holst. 1990. Mapping of Irrigation Need Based on
Computerized Soil and Climatic Data. Agricultural Water Management 17: 291-
407.

Natural Resources Conservation Service, USDA. 1994. State Soil Geographic
(STATSGO) Data Base. Miscellaneous Publication Number 1492.

Neff, E. 1977. How much rain does a raingage gage? J. Hydrology. 35: 213-220.

Niemczynowicz, J. 1987. Storm Tracking Using Rain Gauge Data. Journal of
Hydrolog) 93: 135-152.

Peralta, R. C ., M. A. Hegazy, and G. R. Musharrafieh. 1994. Preventing pesticide
contamination of groundwater while maximizing irrigated crop yield. Water

Resource Research. 30: 3138-3193.

Richardson, CW. and Ritchie, J .T. 1973. Soil Water Balance for Small Watersheds.
Transactions of the ASAE. 16:72-77.

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

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

Ritchie, J .T. 1998. Soil water balance and plant stress. In G,Y, Tsuji et al (eds.);
Understanding Options for Agricultural Production, 41-54.

187

Shanholtz, V0. and TM. Younos. 1994. A soil water balancemodel for no-tillage and
conventional till systems. Agricultural Water Management 26: 155-168.

Shanna, M. L., G.A. Gander, and C. G. Hunt. 1980. Spatial variability of infiltration in a
watershed. J. Hydrology. 45: 1001-1022.

Shih, SF. 1982. Rainfall variation analysis and optimization of gauging sytems. Water
Resources Research. 18(4): 1269- 1277.

Singh, R. and J. Singh. 1996. Irrigation planning in cotton through simulation modeling.
Irrigation Science. 10: 85-98.

Smith, J .A., DJ. Seo, M.L. Baeck, and MD. Hudlow. 1996. An intercomparison study
of NEXRAD precipitation estimates. Water Resources Research 32 (7): 203 5-
2045.

Stewart, J.B., C.J. Watts. J .C. Rodriguez, H.A.R. De Bruin, A.R. van den Berg, J.
Garatuza—Payan. 1999. Use of satellite data to estimate radiation and evaporation
for northwest Mexico. Agricultural Water Management 38: 181-193.

Stockle, CO. and LG. James. 1989. Analysis of deficit irrigation strategies for corn
using crop growth simulation. Irrigation Science. 10: 85-98.

Stol, P. T. 1972. The relative efficiency of the density of rain-gauge networks. J.
Hydrology. 15: 193-208.

Stooksbury, D. E., and P. J. Michaels. 1991. Cluster analysis of southeastern US.
climate stations. Theoretical and Applied Climatology. 44: 143-150.

Sumner, G. N. 1983. The use of correlation linkages in the assessment of daily rainfall
patterns. J. Hydrology. 66: 169-182.

Swaney, D.P. , J .W. Jones, W.G. Bogess, G.G. Wilkerson, and J .W. Mishoe. 1983.
Real-time Irrigation Decision Analysis Using Simulation. Transactions of the
ASAE. ? () 562-568.

Sweat, M.J. and R. van Til. 1987. Michigan: Water Supply and Use. US. Geological
Survey Water-Supply Paper 2350. In: National Water Summary 1987, pp. 305 —
312.

Thomas, A. 1992. Agricultural water balance of Yunnan Province, PR China:
agroclimatic zoning with a Geographical Information System. Agricultural Water
Management 21: 249-263.

188

Thomthwaite, C. W., and J. R. Mather. 1955. The water budget and its use in irrigation.
Yearbook of A griculture. US. Department of Agriculture. 346-358.

Ulaby, F .T., P.C. Dubois, and J. van Zyl. 1996. Radar mapping of surface soil moisture.
Journal of Hydrology, 184: 57-84.

United Nations Population Fund, 1993. State of world population report.

United States Department of Agriculture, Natural Resources Conservation Service. 1994,
State Soil Geographic (STATSGO) Data Base. Miscellaneous Publication
Number 1492.

Victor, U. S., N. N. Srivatsava, and B. V. Ramana Rao. 1988. Quantification of crop
yields under rainfed conditions using a simple soil water balance model.
Agricultural and Forest Meteorology. 39: 73-80.

Williams, J .R., C.A. Jones, and RT. Dyke. 1990. The EPIC model. EPIC-
Erosion/Prodcutivity Impact Calculator: 1. Model Documentation USDA-ARS,
Tech. Bull. No. 1768, 3-92.

Yates, D.N. 1996. WatBal: An Integrated Water Balance Model for Climate Impact

Assessment of River Basin Runoff. Water Resources Development, 12: No. 2.
121-139.

189

TAT UNIV. LIBRRRIES

i1 11 1111 11

3018 29522

nICHIan s
llllllllllllll
3129