mgumms

3'

’Qfll‘fl

l

as”.

~5m¢mgsz

m; .aava

1
O:

n _
hank. .
MkmmA “mm
5.1..
1

ii:

a .71;

1 13 Al‘
1.... ....

,3!

is.

:r \Tl.
.3... 2:: .t.
L1... A

 

 

 

 

 

 

Jul... .

L .

~..l,l

£17..)

1‘
1.»

mi

.
‘Qwaéz,
i, I \«‘v

.mwr. an!
awn:
n.

. l. \‘.~‘C\l
)l?)ll§‘
.13..ch
$1.13.;
:firu.
942......

nine»

“1!!
#336

a.
(la: 4|}:
1 1.5.5..

 

 

amass b

This is to certify that the
dissertation entitled

ANALYSIS OF EFFECTS OF SOIL PROPERTIES,
TOPOGRAPHICAL VARIABLES AND MANAGEMENT
PRACTICES ON SPATIAL-TEMPORAL VARIABILITY OF
CROP YIELDS

presented by

Xuewen Huang

.9
>59
CID
Egg
353,2.
_EC
4323
2

has been accepted towards fulfillment
of the requirements for the

 

Doctoral degree in Crop and Soil Sciences

 

 

94/4.

' Major Professor’s Signature
67%// b /0/
/ v

Date

MSU is an afinnative-action, equal-opportunity employer

 

.-.g.-.-o-n-I-u-c--c-o-n-I-o-o-cg.-.—

.-.—.¢--~u-n---t-n-u;m -.

o~—--‘.-o-n-¢v-I-I---.—-._

 

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

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 KI/ProilAcc8Pres/CIRC/DateDue indd

 

ANALYSIS OF EFFECTS OF SOIL PROPERTIES, TOPOGRAPHICAL
VARIABLES AND MANAGEMENT PRACTICES ON
SPATIAL-TEMPORAL VARIABILITY OF CROP YIELDS

By

Xuewen Huang

A DISSERTATION

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

DOCTOR OF PHILOSOPHY
Department of Crop and Soil Sciences

2008

ABSTRACT

ANALYSIS OF EFFECTS OF SOIL PROPERTIES, TOPOGRAPHICAL
VARIABLES AND MANAGEMENT PRACTICES ON SPATIAL-TEMPORAL
VARIABILITY OF CROP YIELDS

By
Xuewen Huang

Crop yields are highly variable across fields and years as a result of the complex
interactions among topography, weather conditions, and management practices.
Understanding the relationship between yield and these factors is a critical component of
site-specific management systems.

The first study was to demonstrate the feasibility of mapping soil carbon using
newly developed on-the-go near infrared spectroscopy (N IRS) measurements and Landsat
Enhanced Thematic Mapper (ETM) image reflectance in a 50-ha field. Regression
coefficients between measured and predicted carbon values were equal to 0.70 and 0.46
using NIRS data and ETM imagery, respectively. When topographical attributes, such as
elevation, wetness index (WI), and slope were included into the regression model along
with NIRS and ETM data, the regression coefficients improved to 0.81 and 0.62. The
results indicated that combination of the NIRS and ETM measurements with topography is
a valuable tool for accurate total carbon mapping in glacial till soils.

The second study was to identify spatial clusters from historical yield data, and
relate the clusters to the underlying soil electrical conductivity (EC), and topographical

attributes, and to validate whether cluster groups can be used to accurately predict yield

patterns. Both EC and topographical attributes were found to be helpful in explaining yield
variability. Consistently low and high yield clusters were identified. Four to five yield
clusters can explain about 40 % of the yield variation. Areas with lower EC and slope
tended to form high yield clusters in the studied fields. The information from historical
yield classification could be helpful in delineating management zones within a field.

The third study was to analyze the effects of management practices on the
relationship between crop yield and topography, yield and precipitation with 10-yr
com-soybean-wheat rotation in Long Term Ecological Research site, located at Kellogg
Biological Station. A nonparametric spline regression was used to characterize boundary
yields, that are maximum yields, versus WI. The algorithm was also used to compare the
yield difference across the range of the WI between two treatments. Management practices
significantly interact with WI and influence crop yields. The interaction can also be
affected by summer precipitation. The yield difference between no-till and conventional
tillage was strongly influenced by WI in dry or normal year but not in wet year. The
relationship between boundary yield and W1 in most of the crops for most of the studied
years had a convex shape. No-till and low input systems tended to produce higher
maximum and average yields than conventional system at the lower WI areas. The results
suggested that it is possible to maximize yield and profit by farming site-specifically based

on landscape position.

ACKNOWLEDGMENTS

I sincerely thank Dr. Sasha Kravchenko, my major advisor, for her guidance,
support and patience. Particularly, I appreciate her for giving me the opportunity to
working on statistical consulting and encouraging me to pursue dual degree, PhD in Crop
and Soil Sciences and MS in Applied Statistics, which was a great help in my path to
career success. My appreciation is also sent to my committee members, Dr. Delbert
Mokma, Dr. Jiaguo Qi, Dr. Alvin Smucker, and Dr. Kurt Thelen, for their excellent
advice and teaching.

Dr. Thelen helps me collect near-infrared spectroscopy measurements and soil
samples. In addition, he allows me to use his 8 years yield data of Carr Farm. 1 am
indebted to you. I would like to thank Dr. Philip Robertson, Andrew Corbin and Joe
Simmons for their helpful suggestion and discussion on the data analysis and Suzanne
Sippel for preparing yield data from Kellogg Biological Station. My thanks also go to our
laboratory colleague Dr. Xinmei Hao, Senthil Subrarnanian, Wei Wang for their great
help. Especially, I thank Li Wang from Statistics Department. She taught me
nonparametric spline regression and made many graphs of spline regression lines and
confidence bands for comparing yield difference between treatments for me.

I am grateful for the financial support from Kirk and Marjorie Lawton Graduate
Student Support and Dissertation Completion Fellowship offered by the Department of
Crop and Soil Sciences and the Graduate School in my last semester.

Finally, my appreciation always goes to my parents, my wife Zongying and my
son Matthew. It is their love, understanding and supports that make my graduate study

and this dissertation happen.

iv

TABLE OF CONTENTS

LIST OF TABLES .......................................................................................................... vii
LIST OF FIGURES ......................................................................................................... ix
CHAPTER I

INTRODUCTION .............................................................................................................. 1
CHAPTER II

TOTAL CARBON MAPPING IN GLACIAL TILL SOILS USING NEAR
INFRARED SPECTROSCOPY, LANDSAT IMAGERY AND TOPOGRAPHICAL
INFORMATION

ABSTRACT ............................................................................................................ 5
INTRODUCTION .................................................................................................. 6
MATERIALS AND METHODS ....................................................................... 10
Field and Data Description ............................................................................ 10
Statistical Data Analyses ................................................................................ 15
RESULTS AND DISCUSSION .......................................................................... 16
Carbon predictions based on NIRS ................................................................ 17
Carbon predictions based on Landsat ETM ................................................... 22
Regression Validation and Total Soil Carbon Prediction .............................. 25
SUMMARY .......................................................................................................... 28
ACKNOWLEDGMENTS ................................................................................... 29
REFERENCES ..................................................................................................... 30
CHAPTER III

YIELD VARIABILITY IN UNDULATING FIELDS: RELATIONSHIP TO SOIL
ELECTRICAL CONDUCTIVITY AND TOPOGRAPHY, YIELD
CLASSIFICATION AND PREDICTION

ABSTRACT .......................................................................................................... 34
INTRODUCTION ................................................................................................ 35
MATERIALS AND METHODS ....................................................................... 40
Field and Data Description ............................................................................ 40
Correlation and Regression Analysis ............................................................. 43
Cluster Analysis ............................................................................................. 44
RESULTS AND DISCUSSION .......................................................................... 46
Descriptive Statistics and Data distribution ................................................... 46
Yield/precipitation relationship ..................................................................... 49
Spearrnan Correlation Coefficients among Yields ........................................ 52
Pearson Correlation Coefficients among Yield, EC and Topography ........... 53
Yield Classification by Cluster Groups ......................................................... 57
Factors Underlying Yield Clusters ................................................................. 60
Yield Prediction from Cluster Groups ........................................................... 63
SUMMARY .......................................................................................................... 65
REFERENCES .................................................................................................... 66

CHAPTER IV

EFFECTS OF MANAGEMENT PRACTICES ON RELATIONSHIPS BETWEEN
CORN, SOYBEAN, AND WHEAT GRAIN YIELDS WITH FIELD
TOPOGRAPHY AND WEATHER

ABSTRACT .......................................................................................................... 70

INTRODUCTION ................................................................................................ 71

MATERIALS AND METHODS ....................................................................... 76

Site Description and Data Collection ............................................................. 76

Statistical Analyses ........................................................................................ 81

RESULTS AND DISCUSSION .......................................................................... 86

Descriptive Statistics of Topographical Variables and Crop Yields ............. 86

Yield/ Precipitation Relationship among Management Practices .................. 89

Relationship between Boundary Crop Yields and Wetness Index ................ 97

Relationship between Average Crop Yields and Wetness Index ................. 108

CONCLUSION .................................................................................................. 125

ACKNOWLEDGMENTS ................................................................................. 126

REFERENCES ................................................................................................... 127
CHAPTER V

CONCLUSION .............................................................................................................. 132

APPENDICES ................................................................................................................ 134

vi

LIST OF TABLES

CHAPTER II

TOTAL CARBON MAPPING IN GLACIAL TILL SOILS USING NEAR
INFRARED SPECTROSCOPY, LANDSAT IMAGERY AND TOPOGRAPHICAL
INFORMATION

Table 2.1. Stepwise linear regression parameters (p < 0.05) used to estimate soil total
carbon via NIRS and remotely sensed ETM reflectance data with and without, soil
moisture, texture and topography .................................................................................. 18

CHAPTER III

YIELD VARIABILITY IN UNDULATING FIELDS: RELATIONSHIP TO SOIL
ELECTRICAL CONDUCTIVITY AND TOPOGRAPHY, YIELD
CLASSIFICATION AND PREDICTION

Table 3.1. Crop rotation sequences in the study sites ................................................... 40

Table 3.2. Statistical summary of raw and standardized yields from 1996 to 2003 in
two Fields ......................................................................................................................... 47

Table 3.3. Statistical summary of EC and topographical variables in two Fields ..... 49

Table 3.4. Spearman Correlation Coefficients between Yields from 1996 to 2003 in
two Fields ......................................................................................................................... 53

Table 3.5. Pearson Correlation Coefficients between EC, topographical variables
and yield from 1996 to 2003 in two Fields ..................................................................... 54

Table 3.6. Means of topographical variables and EC by cluster in Field 1 and 2 ...... 61

Table 3.7. Yield cell means by cluster in Field 1 and in Field 1 vs. the actual
standardized yields in 1997 and 2002 ............................................................................ 64

CHAPTER IV

EFFECTS OF MANAGEMENT PRACTICES ON RELATIONSHIPS BETWEEN
CORN, SOYBEAN, AND WHEAT GRAIN YIELDS WITH FIELD
TOPOGRAPHY AND WEATHER

Table 4.1. Descriptive statistics of topographical variables from 4 treatments at the
LTER ............................................................................................................................... 80

Table 4.2. Average yield and standard deviation for corn, soybean and winter wheat
from the four studied treatments along with summary statistics of the wetness index
values for the studied treatments ................................................................................... 88

vii

Table 4.3. Parameters of the linear regression model for the relationship
between standardized yield and spring and growing season daily rainfall from 4
treatments ........................................................................................................................ 90

Table 4.4. Results of multiple regressions and partial correlation coefficients
between standardized yield with spring and growing season daily rainfall from 4

treatments ........................................................................................................................ 95

Table 4.5. Comparison of r2 and mean square error between nonparametric (spline)
and parametric regression (segment) from 4 treatments in corn yield of 2002 ....... 102

Table 4.6. Model fit statistics and variance components ........................................... 110

Table 4.7. Mean yield estimates (Mg ha") among 4 treatments with and without
wetness index as a covariate ......................................................................................... 111

viii

LIST OF FIGURES
CHAPTER II
TOTAL CARBON MAPPING IN GLACIAL TILL SOILS USING NEAR
INFRARED SPECTROSCOPY, LAN DSAT IMAGERY AND TOPOGRAPHICAL
INFORMATION
Fig. 2.1. Elevation, layout of soil sample and NIRS test sites in Carr Farm ............. 11
Fig. 2.2. ETM image taken on April 14, 2004 in Carr Farm and vicinity area ......... 14

Fig. 2.3a. Plot of the observed total carbon values vs. predictions obtained from

multiple regression equation with NIRS PCs ............................................................... 21
Fig. 2.3b. Multiple regression validation: TC vs. PCs from NIRS plus

topography ....................................................................................................................... 21
Fig. 2.3c. Multiple regression validation: TC vs. PCs from ETM .............................. 23

Fig. 2.3d. Multiple regression validation: TC vs. PCs from ETM plus topography..23

Fig. 2.4. Regression equation cross-calibration ............................................................ 26
Fig. 2.5a. Carbon map predicted by PCs of N IRS and topography ........................... 27
Fig. 2.5b. Carbon map predicted by PCs of ETM image and topography ................ 28
CHAPTER III

YIELD VARIABILITY IN UNDULATIN G FIELDS: RELATIONSHIP TO SOIL
ELECTRICAL CON DUCTIVITY AND TOPOGRAPHY, YIELD
CLASSIFICATION AND PREDICTION

Fig. 3.1. Daily average precipitation from March to August in the study years ....... 42
Fig. 3.2. Standardized yields in 2 Fields from 1996 to 2003 ........................................ 48

Fig. 3.3. The standard deviation of standardized yields in the 2 Fields from 1996
to 2003 .............................................................................................................................. 48

Fig. 3.4. Standardized yields vs. July daily precipitation over 8 years (2 crops and 2
Fields combined) ............................................................................................................ 50

ix

Fig. 3.5. Standardized yield vs. total precipitation of March and April (2 crops and 2
Fields combined) ............................................................................................................ 51

Fig. 3.6. Standardized yields vs. total growing season (May to August) precipitation
over 8 years (2 crops and 2 Fields combined) .............................................................. 51

Fig. 3.7. The cubic clustering criterion and over-all R2 vs. number of cluster in
Field 1 .............................................................................................................................. 58

Fig. 3.8. Average standardized yields from 1996 to 2003 in Field 1. The length of
vertical bars is equal to the cluster standard deviation .............................................. 59

Fig. 3.9. Average standardized yields and their standard deviations represented by

vertical bars for each cluster from 1997 to 2003 in Field 2 ......................................... 60
Fig. 3.10. Elevation (a), yield clusters (b), and slope (c) in Field 1 ............................. 62
Fig. 3.11. Elevation (a), yield clusters (b), and slope (c) in Field 2 ............................. 62
Fig. 3.12. Spatial distribution of standardized yields in 1997 (a) and 2002 (c) and 6-
year yield clusters (b) in Field 1 ..................................................................................... 64
CHAPTER IV

EFFECTS OF MANAGEMENT PRACTICES ON RELATIONSHIPS BETWEEN
CORN, SOYBEAN, AND WHEAT GRAIN YIELDS WITH FIELD
TOPOGRAPHY AND WEATHER

Fig. 4.1. Layout of the 24 plots in the experimental site (4 treatments and 6
replications for each treatment), and the 4 x 4 m interpolated elevation map of the
LTER site at KBS ............................................................................................................. 77

Fig. 4.2. Example of the layout of locations of the elevation (dots) and yield
measurements (triangles) obtained in each plot of the LTER site at KBS ................. 79

Fig. 4.3. March, April, May, June, July, and August daily average precipitations
from 1996 to 2005 and 30-year average ......................................................................... 81

Fig. 4.4. Scatter plot and boundary line of corn yield vs. wetness index (WI) for T1
in 2002. Dots represent raw data. Triangles represent top 4 yield data for each
increment of WI ............................................................................................................. 83

Fig. 4.5. Histograms of wetness index (WI) for 4 treatments ....................................... 88

Fig. 4.6A. Standardized (Std) yield vs. Spring (march and April) and Summer
(June, July and August) Daily Precipitation for treatment Tl .................................. 90

Fig. 4.68. Standardized (Std) yield vs. Spring (march and April) and Summer (June,
July and August) Daily Precipitation for treatment T2 ............................................... 91

Fig. 4.6C. Standardized (Std) yield vs. Spring (march and April) and Summer
(June, July and August) Daily Precipitation for treatment T3 .................................... 91

Fig. 4.6D. Standardized (Std) yield vs. Spring (march and April) and Summer
(June, July and August) Daily Precipitation for treatment T4 .................................... 92

Fig. 4.7a. Standardized (Std) yield vs. summer (X) and spring season (Z) daily
precipitation for treatment T1 ........................................................................................ 92

Fig. 4.7b. Standardized (Std) yield vs. summer (X) and spring season (Z) daily
precipitation for treatment T2 ........................................................................................ 93

Fig. 4.7c. Standardized (Std) yield vs. summer (X) and spring season (Z) daily
precipitation for treatment T3 ........................................................................................ 93

Fig. 4.7d. Standardized (Std) yield vs. summer (X) and spring season (Z) daily
precipitation for treatment T4 ........................................................................................ 94

Fig. 4.8. Comparisons of 2-linear-line and spline regression for corn yield in 2002 vs.
WI. Dots are top 4 yield data. Dotted lines indicate 95% confidence band. Dash line

is spline line ....................................................................................................................... 99
Fig. 4.9. Boundary line of corn yield vs. WI fitted by spline regression ................... 101

Fig. 4.10. Plot of 95% confidence band of the difference of corn maximum yield
between T1 and T2 as a function of WI ....................................................................... 105

Fig. 4.11. Plot of 95% confidence band of the difference of corn maximum yield
between T1 and T3 as a function of WI ..................................................................... '..106

Fig. 4.12. Plot of 95% confidence band of the difference of corn maximum yield
between T3 and T4 as a function of WI ....................................................................... 107

Fig. 4.13. Plot of estimated corn yield regression models for the 4 treatments at
different WI values. I is for 1996, II is for 1999 .......................................................... 113

Fig. 4.13 cont’d. Plot of estimated corn yield regression models for the 4 treatments
at different WI values. 111 is for 2002, IV is for 2005 ................................................. 114

xi

Fig. 4.14. Plot of 95% confidence band of corn yield difference between T1 and T2,
where WI as covariate. I, II are for years 1996, 1999, respectively .......................... 117

Fig. 4.14 cont’d. Plot of 95% confidence band of corn yield difference between T1
and T2, where WI as covariate. III, IV are for years 2002, 2005, respectively ........ 118

Fig. 4.15. Plot of 95% confidence band of corn yield difference between T1 and T3,
where WI as covariate. I and II are for years 1996, 1999, respectively .................... 120

Fig. 4.15 cont’d. Plot of 95% confidence band of corn yield difference between T1
and T3, where WI as covariate. III and IV are for years 2002, 2005, respectively..121

Fig. 4.16. Plot of 95% confidence band of corn yield difference between T3 and T4,
where WI as covariate. I and II are for years 1996, 1999, respectively ................... 123

Fig. 4.16 cont’d. Plot of 95% confidence band of corn yield difference between T3
and T4, where WI as covariate. III and IV are for years 2002, 2005, respectively..124

Images in this dissertation are presented in color

xii

CHAPTER I

INTRODUCTION

Crop yields are highly variable across fields as a result of the complex
interactions between many factors, including topographical attributes, soil properties,
precipitation, and management practices. Understanding the relationship between yield
and these factors is a critical component of site-specific management systems. With the
adoption of electrical sensors and global positioning system (GPS) in crop production in
the last decade, tremendous data, such as yield map, densely-measured electrical
conductivity (EC) and elevation were collected. How to efficiently extract the valuable
information from those data is a challenge. In addition, characterizing spatial variation of
field conditions and exploring the relationship between yields and the underlying factors
will provide essential information for decision-making in precision farming.

Field topography has a major impact on water and also on nutrient availability for
crop production. It is the major source of not only soil but also yield variability. However,
yield/topography relationships vary substantially from year to year. These variations are
often associated with the prevailing weather conditions during the growing season of
each particular year. Weather conditions may substantially influence the apparent gains
or losses due to management and other factors on a regional as well as field scale.
Management practices, including tillage and chemical inputs, can interact with
topographical attributes to influence yield variability.

With long-term data collection of crop yield and other variables and advances in
topographical modeling and statistical methods in the recent years, exploring the

relationship between yield and topographical attributes under various management

practices and weather conditions will significantly improve our understanding of crop
production. The overall goal of my research was to analyze the effects of soil properties,
topographical variables and management practices on spatial-temporal variability of crop
yields.

The first objective was to determine the feasibility of mapping soil organic carbon
using newly developed on-the-go near infrared spectroscopy (NIRS) and Landsat ETM
imagery reflectance data for predicting soil carbon in glacial till soils with and without
additional topographical information in an undulated field. Accurate field maps of soil
total carbon are important not only for defining site-specific management zones in
precision agriculture, but also for reliable assessment of soil carbon sequestration
potential on field and watershed scales. Highly variable soil and topographical attributes
of glacial till soils can affect the quality of carbon map with a limited number of soil
samples, because soil carbon is often highly spatially variable as influenced by natural
soil variability and topography. Therefore, quick, reliable and cost-effective techniques
are needed for measuring soil carbon changes in response to changes in land management
on an agricultural field. A field study of carbon mapping was conducted in April 2004 in
Kalamazoo County, Michigan. NIRS spectra were collected in a 50-ha field. Eighty-five
soil samples were collected on the NIRS transects from 0 -10 cm depth. Landsat ETM
imagery during the period of soil sampling was obtained. Principal component regression
was used to relate NIRS and ETM reflectance data to measured soil carbon.

The second objective was to characterize relationship between yield and soil EC,
yield and topographical attributes and to identify spatial clusters from historical yield

data, and relate the clusters to the underlying soil EC, and topographical attributes.

Inexpensive and densely measured soil EC and topographical attributes were widely used
to explain yield variability. Available historical yield data could also provide usefiil
information about yield spatial variability and potential management zones, which will
assist crop production site-specifically. Data used for this study were from SO-ha Carr
Farm in Kalamazoo County, Michigan. Eight year com-soybean yield data collected
using GPS and yield monitor were available to study the spatial-temporal variability.
Detailed elevation and soil EC were also available for analyzing yield variation. Cluster
analysis on EC and topographical variables and historical yield data was used to identify
management zones within the field.

The third objective was to analyze the effects of management practices on the
relationship between crop yield and topography, yield and weather from long-term
experiment. The management effects can be altered by field topography and weather
conditions. To make a better farming decision site-specifically needs thorough analysis
and understanding of production variability and its relationship with underlying
environmental factors in multi-crop with long term data. Information on
yield/topography/weather relationships will contribute to the effort in expanding the
conservation management practices and likely increase the grower’s profit from site-
specific farming while shifting from conventional to conservation practice. Data used for
this study were from the Long Term Ecological Research site at WK. Kellogg Biological
Station. Ten-year com-soybean-wheat yield data collected using GPS and yield monitor
were available to study the effects of management on yield variability. Detailed elevation
and derived topographical variables in each plot were available to study effect of

management practices on relationship between topography and yield.

A nonparametric spline regression was used to characterize relationship between
maximum yields and wetness index (WI). The algorithm was also used to compare the
yield difference across the range of the WI between two treatments. Average yields
difference between treatments in this one-factor randomized complete block design
experiment was first studied by analysis of variance (ANOVA). However, those varied
field conditions can cause imprecision and incorrect interpretation of the experimental
results using traditional ANOVA model. Incorporating spatially intense measurements of
topographical variables into ANOVA models, called analysis of covariance (ANCOVA),
one can correct for these confounding variables and obtain more precise estimates of
yield performance of the treatment. The 2 models were compared in order to find better

model and make the yield comparisons between treatments more precise.

CHAPTER II
TOTAL CARBON MAPPING IN GLACIAL TILL SOILS USING
NEAR INFRARED SPECTROSCOPY, LANDSAT IMAGERY AND

TOPOGRAPHICAL INFORMATION

ABSTRACT

Accurate mapping of total soil carbon is important for reliable assessment of
carbon sequestration potential from small fields to regional scales. Highly variable soil
and topographical attributes, such as elevation and slope, of glacial till soils cause
difficulties in mapping soil carbon based on a limited number of soil samples. The
objectives of this study were to determine the feasibility of mapping soil organic carbon
using newly developed on-the-go near infrared spectroscopy (N IRS) measurements and
Landsat ETM image reflectance in glacial till soils with and without additional
topographical information. The studied field was about 50 ha in size and located in
Kalamazoo County, Michigan. The predominant soil is Kalamazoo loam (fine-loarny,
mixed, mesic Typic Hapludalfs). NIRS spectra were collected along 22 north-south
transects separated by a distance of 25 m and the distance between the measurements
within a transect was 5 m. The field was bare of vegetation and relatively dry during soil
sampling. Landsat ETM imagery during the period of soil sampling was obtained.
Eighty-five soil samples were collected on the NIRS transects from 0 -10 cm depth.
Principal component regression was used to relate NIRS spectra and ETM data to
measured soil carbon. Regression coefficients between measured and predicted carbon

values were equal to 0.70 and 0.46 using NIRS data and ETM imagery, respectively.

When topographical attributes, such as elevation, wetness index, and terrain slope were
included into the regression model along with NIRS and ETM data, the regression
coefficients improved to 0.81 and 0.62, respectively. The results indicated that the
combination of the NIRS and ETM measurements with topography is a valuable tool for
accurate total carbon mapping in glacial till soils. Field soil moisture and texture can help
explain carbon variation and improved its prediction for ETM imagery data, but were

found to not be a very useful addition to NIRS measurements.

INTRODUCTION

Soil carbon plays an important role as a major source of plant nutrients and water
holding capacity as an active agent in soil structure development. Accurate field maps of
soil total carbon are important not only for defining site-specific management zones in
precision agriculture (Chen, et al., 2000; Mueller and Pierce, 2003; Martin et al., 2002;
Walter et al., 2003; Chen et al., 2005) and evaluating soil quality (Brejda et al., 2001;
Homann et al., 1998), but also for reliable assessment of soil carbon sequestration
potential on field and watershed scales (Young, 2003; Lal et al., 2004; Leifeld and Kogel,
2005)

Precision farming requires detailed spatial information on soil properties to
manage crop production with increased farm profits and reduced environmental impacts.
Detailed soil carbon maps provide essential information for site-specific decision-making
in choosing sound management practices. For example, organic carbon concentrations in

the surface soil are needed to determine site-specific application rates of some crop-

production inputs, including nitrogen fertilizers and herbicides (Blackmer and White,
1998)

More recently, global climate change studies have shown that increasing carbon
storage by soils is a practical method to mitigate greenhouse gas emissions (Robertson et
al., 2000; Lal, 2004). Conservation management practices that enhance soil carbon
storage, e. g. no-till and cover cropping, can stimulate carbon sequestration (Young, 2003;
Lal et al., 2004). Growers may be able to benefit when switching their management
practices to those that store more soil carbon by getting paid for stored carbon by private
markets or government programs (Young, 2003). However, quick, reliable and cost-
effective techniques are needed for measuring soil carbon changes in response to changes
in land management on an agricultural field or a regional level.

Laser-induced breakdown spectroscopy was found to be a promising method for
estimating total soil carbon (Cremers et al., 2001; Ebinger et al., 2003). A number of
laboratory studies used near-infrared spectroscopy (N IRS) for rapid analysis of soil
carbon (Dalal and Henry, 1986; Palacios-Orueta and Ustin, 1998; Chang et al., 2001;
Martin et al., 2002; McCarty etal., 2002; Russell et al.,2003; Dematte et al., 2004; Chang
et al., 2005), nitrogen (Dalal and Henry, 1986; Russell et al., 2003), clay content
(Dematte et al., 2004; Sarensen and Dalsgaard, 2005), and soil moisture (Dalal and
Henry, 1986; Chang et al., 2005). The NIRS is an analytical technique that allows
estimating several soil properties with acceptable accuracy in a short time (Chang et al.,
2001). But the spectral measurements obtained in laboratory conditions using a limited
number of soil samples can not capture the field spatial variability of total carbon and are

less representative of soil spatial heterogeneity present on a field or farm scale.

Recently, on-the-go NIRS was developed for in-field analysis of soil properties
(Christy et al., 2003). Near-infrared spectra are dominated by weak overtones and
combinations of fundamental vibrational bands for H-C, H-N, and H-0 bonds from the
near- and mid-infrared region (Chang et al., 2001; Christy et al., 2003; Sorensen and
Dalsgard, 2005). Because organic matter in the soil mainly consists of C, H, O and N
elements soil NIRS measurements are generally most affected by soil organic matter
(Sorensen and Dalsgard, 2005). NIRS spectra were related to soil carbon in agricultural
fields of central Iowa and Kansas (Christy etal., 2003).

Remote sensing is another cost-effective and nondestructive analytical technique
to estimate soil carbon with acceptable accuracy (Frazier and Cheng, 1989; Chang ct al.,
2001; Qiang et al., 2002; Konen et al., 2003). The reflectance from remotely sensed
imagery is well related to surface soil carbon when the surface is bare, dry and flat.
Landsat Thematic Mapper band ratios were used to classify various levels of organic
matter and iron oxides (Frazier and Cheng, 1989). The intensity of aerial photography
showed a high correlation with soil organic carbon (Chen et al., 2000). The soil line
Euclidean distance technique is based on relating a pixel's Euclidean distance of the red
and near-infiared intensity values for the bottommost point on the soil line, which is a
linear relationship between the near-infrared and visible reflectance of bare soil images.
This technique was used to estimate soil surface organic matter and other properties from
aerial images acquired by a digital camera system (Fox and Sabbagh, 2002). The
regression coefficient between measured and predicted organic matter values obtained in

their study was around 0.7.

Soil carbon is often highly spatially variable as influenced by natural soil
variability and topography mainly due to soil erosion and deposition. In Michigan, the
physiography is characteristic of a mature glacial outwash plain and moraine complex.
Soils formed in materials deposited by the glacial till and outwash processes across the
landscape resulting in formation of various landforms, such as undulating hills, valleys
and plains even within small areas. Highly variable soil and topographical attributes of
glacial till soils can affect the quality of carbon map with NIRS or imagery data only
(Mueller and Pierce, 2003). However, the quality of soil carbon maps can be improved
and spatial sampling intensities can be reduced by incorporating topographical attributes
as a source of secondary information in soil carbon mapping (Bell et al., 2000; Mueller
and Pierce, 2003; Chaplot et al., 2001; Terra et al., 2004).

Currently information on soil carbon is derived through standard procedures of
soil sampling and laboratory analyses. The methods are expensive, time-consuming, and
less accurate at the on-farm field scale level. We selected a com-soybean rotated farm
located at southwestern Michigan. The clay plus silt content of the field varied from 20 to
80 %. The range of field elevation was 13 m. The field was well-represented as glacial till
soils in US. northern combelt. The objectives of this research were to demonstrate the
feasibility of mapping soil organic carbon using newly developed on-the-go near infrared
spectroscopy and Landsat ETM imagery reflectance data for predicting soil total carbon
in glacial till soils with and without additional topographical information in an undulated

field.

MATERIALS AND METHODS

Field Description and Data Collection

The studied field was about 50 ha in size and located in Kalamazoo County, MI
(42°22' N, 86°36' W). The predominant soil mapping unit is Kalamazoo loam (fine-
loarny, mixed, mesic Typic Hapludalfs) ((Austin, 1979, Kalamazoo County Soil Survey,
Michigan). A small area of Oshtemo sandy loam (coarse-loamy, mixed, mesic Typic
Hapludalfs) was scattered in the southwest portion of the field. In the top 10 cm, sand
content ranged from 19 to 77 % and clay content ranged from 2 to 22 %. The field was in
a corn/soybean rotation with nonirrigated minimum till management (J iang and Thelen,
2004). The elevation of the field ranged from 290 m to 303 m with near level (0 — 2%),

gentle (2 — 6%), and moderate (6 — 14%) slopes (Fig. 2.1).

10

 

 

 

 

 

 

  

elevation, m

soil samples
290.9 - 294.4
294.4 - 296.5
296.5 - 298.8
298.8 - 300.8
300.8 - 303.2
NIRS points

'IIIDD'

Fig. 2.1. Elevation and layout of soil samples and N IRS test sites in Carr Farm
Elevation was measured using a real-time kinetic global positioning system (RTK
GPS) with a tractor-mounted receiver on a cart. The measurements were taken every 5 m
within and 10 m between the transect (the cart passes). Topographical attributes were
derived from the elevation data. First, irregular point elevation measurements were

interpolated into a regular 15 x 15 m digital elevation model (DEM) using function

POINTINTERP with inverse distance weighting with a circular neighborhood of 20 m
under ArcInfo Grid (ESRI, 2000). With large dense datasets, the spatial structure of soil
properties becomes less important. The methods of inverse distance weighting and
kriging perform equally well in interpolation (Kravchenko, 2003). Therefore, we applied
simple inverse distance weighting procedure to interpolate elevation data.

Slope, plan and profile curvatures, flow direction, and flow accumulation were derived
from the DEM using corresponding functions in surface hydrologic analysis of ArcInfo
GRID. Detailed definitions and explanations of slope, curvature, flow direction, and flow
accumulation functions are given elsewhere (Kravchenko and Bullock, 2000; McKenzie
etal., 2000; and ArcInfo 8.1 help). Soil wetness index (WI) also called compound
topographic index (McBratney et al., 2003) was calculated as proposed by Moore et a1.

(1993):

WI = ln(As /tan ,5) [1]

where As is the upslope area derived from flow accumulation and tan fl is a tangent

function of slope at percent-rise.

Near-infrared spectroscopy measurements were taken by Veris Technologies on
April 19, 2004. The field surface condition was dry, bare, and relatively smooth. The
NIRS spectra were collected along 22 north-south transects separated by a distance of
approximately 25 m. The distance between measurements within each transect was equal
to 5 m (Fig. 2.1). A total of approximately 3,700 NIRS measurements were collected in
this field. The NIRS instrument was mounted on a tractor driven cart equipped with a
steel shank get to a 10 cm depth. A quartz tungsten bulb was mounted within the shank to

illuminate the soil while optical fibers were used to transmit reflected light to a

12

spectrometer mounted above the shank. The measurement depth was held at
approximately 7.5 cm. A spectroradiometer measured light density (reflectance) from
wavelength 0.9 to 1.7 um on every 6 pm interval. Approximately 20 spectra per second
on one GPS location were acquired and then stored on a laptop computer. The reflectance
values were converted to absorbance as a logarithm of the inverted reflectance spectra for
further analyses. The detailed description of the NIRS measurement system was reported
by Christy et a1. (2003).

Remote sensing data, Landsat 7 ETM+ imagery from path 21 and row 31, were
acquired on April 14, 2004 (Fig. 2.3), 5 days before NIRS and soil samples were taken.
ETM imagery contains band 1 (0.45 - 0.52 pm spectral range), 2 (0.52 - 0.60 pm), 3
(0.63 — 0.69 pm), 4 (0.76 — 0.90 pm), 5 (1.55 — 1.75 pm), 7 (2.09 to 2.35 um), and
panchromatic band 8 (0.52 — 0.90 um). Bands 1 to 5 and 7 have 30 m resolution and band
8 has 15 m resolution. The radiometric values on the image were converted to reflectance
using image processing software (ERDAS Imagine, 2001). The bands 1 to 5 and band 7
were merged with band 8 to obtain a 15 m resolution using the principal component

method of resolution merge under spatial enhancement of ERDAS imaging.

13

  
 
 
   

30 m
resolution

in vicinity area
of Carr Farm ~_. ..

 
   

15 m resolution of Carr Farm 30 m resolution of Carr Farm

Fig. 2.2. ETM image taken on April 14, 2004 of Carr Farm and vicinity area
Eighty-five geo-referenced soil samples were randomly collected at the 0 - 10 cm
depth immediately after the NIRS pass. Soil water content was measured using a

gravimetric method and soil texture (clay, silt, and sand contents) was measured using a

hydrometer method. Samples for total carbon analysis were air-dried at room
temperature, cleaned from visible plant residues and ground to pass a lOO-mesh sieve.
Total carbon was analyzed by a dry combustion method using a Carlo-Erba, Series-2

Carbon-Nitrogen analyzer.

Statistical Data Analyses

Principal component analysis (PCA) was performed on the NIRS absorbance
measurements on 128 wavelengths to reduce data dimension by SAS PROC PRIN COMP
(SAS Inst, 2000). Similarly, PCA was performed on the ETM reflectance measurements
on 6 bands for 30 m resolution and 7 bands for 15 m resolution.

Multiple linear regression analyses between total carbon content and principal
components (PCs) of NIRS and ETM measurements with and without topographical
attributes and soil moisture and texture variables were performed with SAS PROC REG.
Principal component regression (PCR) described by Ehsani et a]. (1999) was used to
relate the spectral data of NIRS and ETM to soil total carbon content. Stepwise
regression was employed for PCs and topographical variable selection. The 0.05 level of
significance was used for retaining the variables in the final regression model.

For five of the sampled points in the east portion of the field, elevation was not
obtained and topographical information was not available, so those points were excluded
from the regression analysis. Further, two potential outliers (extremely high carbon
content) were identified based on Cook’s distances and excluded from the dataset.
Finally, a total of 78 samples were used for the regression analysis. The regression

models were compared based on R2, adjusted R2, RMSE, and PRESS statistics. The

15

 

PRESS statistics is a leave-one-out cross-validation in regression. The model that gives
the smallest values of the PRESS is chosen as the best one.

Regression performance was tested by the jack-knifing method using ten
independent test data sets. Each test data set consisted of 10 samples randomly selected
from the main data set. The remaining data were used to perform multiple regression
analyses and the regression predictions were compared with the measured values of the
test data. Root mean square errors were calculated for each regression analysis of each
test data set.

To obtain the map of soil carbon predictions for the studied field, PCs from NIRS
and ETM and topographical attributes found to be significant in the step-wise regression
procedures were interpolated to create 15- by 15-m cell maps in ArcView Spatial
Analyst. Then the regression predictions of total carbon for each cell were calculated
from the regression models using the PCs of NIRS and ETM and topographical attributes.
The residuals from the regression models were found to be spatially correlated. Thus, the
map of the residuals was obtained using ordinary kriging and added to the regression

prediction map to obtain the final map of total carbon predictions for the studied field.

RESULTS AND DISCUSSION

Soil total carbon content in the studied field ranged from 5.5 g kg”1 to 28.9 g kg"l
with an average of 15.4 g kg". The relatively wide range of total carbon values provided
a good data set to test the performance of NIRS and remotely sensed imagery in

predicting soil carbon.

16

Total carbon was significantly positively correlated with field soil water content,
clay content, and wetness index with correlation coefficients, r, equal to 0.57, 0.37, and
0.36, respectively, and negatively correlated with sand content, elevation, and slope with
r equal to -0.75, -0.56, and -0.28, respectively (p = 0.05). The observed correlations
support the concern that relatively large variability of texture and elevation within the
field might affect the accuracy of the carbon mapping using traditional methods when

sample size is not sufficiently large.

Carbon Predictions Based on N IRS

The first 13 principal components of NIRS explained 99% of the variation in
spectral data. Thus, the 13 PCs were used as initial explanatory variables in step-wise
regression. Step-wise selected regression equations for soil total carbon versus NIRS with
and without topography, soil moisture, and texture were presented in Table 2.1. A
comparison of the regression results from NIRS PCs only, NIRS PCs plus topography,
NIRS PCs plus topography and soil moisture, NIRS PCs plus topography and texture,
and NIRS PCs plus topography and soil moisture and texture shows expected differences
between R2, adjusted R2, RMSE, and PRESS values. The lowest R2 values and the largest
RMSE and PRESS of 0.70, 0.189, and 2.75, respectively, were observed in the regression
with only NIRS PCs. The highest 1'?2 values and the lowest RMSE and PRESS of 0.88,
0.127, and 1.47, respectively, were observed in regression with NIRS PCs, topography,

soil moisture and texture.

17

Table 2.1. Stepwise linear regression parameters (p < 0.05) used to estimate soil total
carbon (TC) via NIRS and remotely sensed ETM reflectance data with and without,
soil moisture, texture and topographical variables.

 

 

Variables Regression Equation R2 Agin RMSE PRESS
NIRS
PCs TC = 1.455 + 0.022PC1 -1.133PC2 + 0.70 0.68 0.189 2.75
7.916PC4 + 7.77PC6 + 45.51PC8
PCs and elevation (Ele) TC = 11.89 + 15.53PC4 - 20.22PC5 + 0.81 0.78 0.156 2.68
45.04PC6 — 975.2PC9 + 1886.3PC10 —
935.3PC12 — 16468PC13 -0.035Ele
PCs, moisture (Moist), TC = 8.654 + 0.199PC2 - 0.11PC3 + 0.83 0.81 0.144 1.78
and elevation 0.46lPC4 - 5.494PC13 + 0.033Moist -
0.026Ele
PCs, texture, wetness TC = 0.317 + 0.182PC2 - 0.162PC3 + 0.85 0.83 0.137 1.70
index and slope 0.478PC4 - 3.777PC11 +4.165PC12 +
0.0167clay +0.41Wl+ 0.01S10pe
PCs, moisture, texture, TC = 0.104 + 0.17C2 - 0.138PC3 + 0.88 0.86 0.127 1.47
slope, and wetness 0.331PC4 + 1.023PC7 -3.66PC11 +
index (WI) 4.25PC12 + 0.022Moist + 0.009Si1t +
0.019Clay + 0.0417WI + 0.108Slope
ETM imagery
PCs from 30 m image TC =l.508 — 0.057PC1 + 0. 209PC2 0.33 0.31 0.269 6.23
PCs from 15 m image TC =1.519 - 0.071PC1 — 0.114PC2 -+ 0.46 0.43 0.249 5.25
0.197PC3 — 0.211PC4
PCs from 15 m image TC = 14.01 - 0.07PC1 - 0.0436Ele + 0.62 0.60 0.208 3.56
and elevation, and W1 0.0617WI
PCs from 15 m image, TC = 10.62 - 0.047PC1 + 0.093PC3 + 0.75 0.73 0.173 3.05
moisture, elevation and 0.04Moist - 0.034Ele + 0.042WI
wetness index
PCs from 15 m image, TC = -0.167 - 0.05PC1 + 0.23PC3 + 0.75 0.73 0.174 2.47
texture, slope, and 0.016Silt + 0.167Clay + 0.074WI +
wetness index 0. 137Slope
PCs fiom 15 m image, TC = 5.872 - 0.046PC1 + 0.113PC3 - 0.81 0.78 0.156 2.48

moisture, texture, slope,
and wetness index,
elevation

0.659PC7 + 0.029Moist - 0.009Sand -
0.017Ele + 0.066WI + 0.13Slope

 

For NIRS, elevation was the only topographical variable significant at the 0.05

probability level. By incorporating elevation the total carbon prediction was greatly

improved. The regression explained 70% of the variations in total carbon using NIRS

PCs only, whereas 81% of the variations in total carbon were explained using PCs plus

elevation. Elevation negatively contributed to carbon content. The reason is possible

attributed that soil carbon is usually transported by runoff from high elevation area to low

18

elevation area. Our observations are consistent with other studies that found that terrain
attributes can be used to improve the quality of spatial estimates of soil carbon or allow
similar levels of map quality to be achieved with a reduced soil sampling density
(Mueller and Pierce, 2003; Terra et al., 2004).

The improvement in carbon prediction when soil moisture was added to the
regression model with NIRS PCs plus elevation was relatively small even though
statistically significant. By adding soil moisture, the R2 increased from 0.81 to only 0.83.
Possible reason for low improvement due to adding soil moisture is that field soil
moisture itself was significantly related to elevation (r = -0.26). Also soil moisture was
found to be positively correlated with wetness index (r = 0.20) and negatively correlated
with terrain slope, r of -0.37 (p = 0.05). Therefore the variability in soil moisture was
already substantially explained by topography, thus not providing substantial
improvement when added to multiple regression along with these variables. Chang et al.
(2005) found that increase in moisture compared with dry soil caused a considerable
increase of the absorbance baseline of the spectra when comparing the influence of soil
moisture on laboratory NIRS measurements of soil carbon. The accuracy of the carbon
prediction was better for the air-dried soil than for moist soil. However, consistent with
our observations, the improvements in carbon prediction accuracy when including soil
moisture in the regression were minimal. Based on these results it could be recommended
that in soils similar to those of this study collecting soil moisture data might not be
practical field carbon mapping using NIR if dense topographical information is available.

Adding soil texture information, clay content, to the regression model with NIR

PCs and topography increased the R2 from 0.81 to 0.85. Clay content significantly

19

contributed to carbon content. The reason is that fine particles have greater capacity to
hold organic matter through physical and chemical stabilization (Hassink, 1997; Krull,
2003). The regression model with NIR PCs, topography, soil moisture, and soil texture
had an R2 of0.88.

These results are consistent with observations of Christy et al. (2003) who applied
on-the-go NIRS in Iowa and Kansas fields and obtained an R2 value of 0.87 using PCR.
Lab based spectral measurements with PCR could explain 78% of the variation in organic
carbon on a Canadian prairie soil (Martin et al., 2002). They also found that NIRS
predictions underestimated the lower and higher carbon values while overestimated the
mid-range values thus suggesting a possibly non-linear relation between soil carbon and
NIRS PCs. Fig. 2.3a and 3b show observed carbon values and predictions fiom
regression with and without topography. Similar to observations of McCarty et a1. (2002)
and Martin et al. (2002) the results obtained using NIRS PCs only indicate that there was
a tendency to overestimate low total carbon content values and underestimate the higher
soil carbon contents. The non-linearity was somewhat reduced in regressions with NIRS

PCs and topography (Fig. 2.3b).

20

 

 

y = 0.71x + 0.46

2.00 -

predicted C %
S

 

 

 

 

0.50 1.00 1.50 2.00 2.50
measured C %

 

 

 

Fig. 2.3a. Plot of the observed total carbon values vs. predictions obtained from
multiple regression equation with NIRS PCs, solid line represents a 1:1
relationship, dash line is regression line of predicted vs. measured.

 

2.50

 

y = 0.81x + 0.30 ’
2.00 .

predicted C %
é

 

 

 

 

0.50 1.00 1.50 2.00 2.50
measured C %

 

 

 

Fig. 2.3b. Multiple regression validation: TC vs. PCs from NIRS plus topography,

solid line is 1:] line of measured C, dash line is regression line of predicted vs.
measured.

21

Carbon Predictions Based on Landsat ETM

The regression results obtained from the Landsat ETM data were inferior to those
obtained from NIRS (Table 2.1). ETM PCs regression based on a 30 m resolution
imagery explained only 33% of the variation in total carbon producing the highest
observed RMSE and PRESS value of 0.269 and 6.23. Thus, Landsat ETM 30 m
resolution imagery is not recommended for carbon prediction in the studied soils. The
ETM PCs regression equation from 15 m resolution imagery performed much better
explaining 46% of the variations in total carbon. When combined with topographical
data, the 15 m resolution imagery explained as much as 62% of the variations in total
carbon substantially reducing RMSE and PRESS. Elevation and wetness index were the
topographical variables that were selected based on a step-wise selection procedure as
those that significantly improved carbon prediction accuracy when added to the ETM 15
m resolution imagery data (p = 0.05).

Figures 30 and 3d show Plots of observed and predicted carbon values obtained
from multiple regression models from 15 m resolution imagery ETM PCs only and ETM
PCs plus topography. The discrepancy between measured and predicted values for ETM
PCs only was substantial (Fig. 2.3c). It was obvious that the regression equation
overestimated low and underestimated high carbon values. This is contrary to Sullivan et
a1. (2005) who reported a strong tendency to underestimate at low soil carbon content and
over estimate at higher soil content from regression residual analysis using IKONOS
imagery. The ETM PCs plus topography regression performed remarkably better than
that of ETM PCs only (Fig. 2.3d). The results encourage a combined application of less

costly satellite imagery and topographical data to map soil carbon in the studied soils.

22

 

 

 

2.00-

1.50-

predicted C %

 

 

 

0.50 . f . . . . . l
0.50 1.00 1.50 2.00 2.50

 

measured C %

i

1
1

Fig. 2.3c. Multiple regression validation: TC vs. PCs from ETM, solid line is 1:1 line
of measured C, dash line is regression line of predicted vs. measured.

 

 

1.50 -

predicted C %

1.00 -

 

 

 

 

 

0.50 1.00 1.50 2.00 2.50

 

measured C %

 

Fig. 2.3d. Multiple regression validation: TC vs. PCs rom ETM plus topography,
solid line is 1:1 line of measured C, dash line is regression line of predicted vs.
measured.

23

Carbon prediction was further improved by including soil moisture with
regression explaining 75% of the variation in total carbon. The RMSE decreased from
0.208 to 0.173. Regression with texture produced results very similar to those of
regression with soil moisture, where silt and clay was significant in the equation (Table
2.1). By adding moisture and texture data together, the R2 value further increased to 0.81.
Unlike the regression with texture only, silt and clay were found to be not significant
when included in the regression model along with soil moisture. However, sand content
was included in the model with significant negative contribution to carbon content. Soil
moisture and/or texture addition to the regression improved carbon prediction more
substantially when added to the regression model with ETM PCs as compared to the
regression model with NIR. This is because remote sensed imagery is very sensitive to
the soil surface condition, such as wetness and roughness. However, water molecules
have a direct impact on near infrared absorbance via the O-H bond. The NIRS
measurements indicated that water caused differences in NIRS absorbance (Christy et al.,
2003))

The positive regression coefficient for soil moisture and wetness index in the
model suggested that high soil moisture is coincident with high soil carbon content.
Moorrnan et a1. (2004) also found that wetness index was positively correlated with soil
carbon content. They suggested that wetness index was a strong predictor of soil organic
C in glacially-derived Mollisolls of Iowa. Thus collection of soil moisture data can be
used to achieve improving carbon prediction using ETM imagery even when dense
topographical information is available. Apart from direct soil sampling, soil moisture data

can be obtained from microwave and radar images. We hypothesize that combining ETM

24

with Radar or Microwave images will produce low—cost accurate large-area soil carbon
maps in soils similar to those of this study. Further research is needed to test this
hypothesis.

Other studies support our findings in demonstrating that aerial photos and satellite
images are less costly and practical in predicting soil carbon with reasonable accuracy.
Fox and Metal (2005) employed 4-band aerial imaging to relate soil carbon with PCs of
image intensity in two Iowa fields obtaining R2 values of 0.68. In Georgia investigations
of Coast Plain soils. Chen et a1. (2000 and 2005) predicted soil organic carbon from fine
resolution 3-band aerial imaging and multi-band ATLAS sensor image using non-linear
regression with R2 values ranging from 0.87 to 0.97. Sullivan et a1. (2005) related
IKONOS imagery reflectance to soil carbon with R2 values equal to 0.38. They
speculated that in the studied soils sandy epipedon and rough surface are the primary

factors limiting the ability to obtain accurate soil carbon estimations.

Regression Validation and Total Soil Carbon Prediction

Figure 4 shows average RMSE values obtained based on 10 test data sets for four
of the studied regressions, i.e., regression with NIRS PCs only, NIRS PCs plus
topography, ETM PCs only and ETM PCs plus topography. The ETM PCs regression
equation predictions produced the highest average RMSE, equal to 0.24. The NIRS PCs
only regression equation and NIRS PCs with elevation equation had very similar RMSE
values equal to 0.18 and 0.17, respectively. The RMSE for ETM PCs regression plys
elevation and wetness index regression was only marginally higher than those of NIRS

based methods and was equal to 0.19.

25

 

0.40

 

0.35 -

   

0.30 -
NIRS ETM imagery
0.25 - ,

0.20 -

RMSE

0.15 -

0.10 -

0.05 -

 

 

 

 

 

PCs + Ele PCs PCs + Ele +
WI

 

 

 

Regression Validation

 

Fig. 2.4. Regression equation cross-calibration

The two regression equations of NIRS PCs plus topography and ETM PCs plus
topography were used to generate a field-scale soil carbon map. The predicted soil
carbon for Carr field was based on a 15 m x 15 m cell size map. The map was corrected
by adding kriged residuals to the map obtained using linear regression predictions. Fig.
2.5a shows the total carbon map predicted by the regression equation using NIRS PCs
plus elevation. The range of predicted soil total carbon was from 3.37 g kg'1 to 24.28 g
kg"1 with an average of 14.4 g kg". Fig. 2.5b shows a total soil carbon map predicted by
ETM PCs plus elevation and wetness index. The range of predicted total soil carbon was
from 4.5 to 22.86 g kg"1 with average of 14.6 g kg]. The averages of predicted carbon
from the above two equations are very similar. However, the range of carbon values from
NIRS was wider than that from the ETM imagery. The reason is that each map cell

contained three NIRS measurements, whereas there was only one measurement in ETM

26

imagery per map cell. Thus, more of the local carbon variation was detected by NIRS.
Finer resolution imagery might improve accuracy of soil carbon prediction. In addition to
the coarse resolution of the ETM imagery, another reason for lower R2 ETM predictions
was that the ETM imagery measured surface reflectance. However, soil samples were
taken from 0- to 10—cm. There could be a difference in soil carbon values between the

surface soil and the mixed 0- to 10-cm soil sample.

 

  
 
 
 
 
 
 
 
 
 
 
 
 

total C % by NIRS
- 0.33 - 0.60
- 0.60 -1.00
- 1.00 -1.40
- 1.40 - 1.80
1:] 1.80 - 2.20
E] 2.20 - 2.43

 

 

 

Fig. 2.5a. Carbon map predicted by PCs of NIRS and topography

27

 

      

total c % by ETM
- 0.45 - 0.60
- 0.60 - 1 .oo
- 1.00 -1.40
- 1.40 - 1.80
n 5:1,. 1.80- 2.20

CI 2.20 - 2.29

 

 

 

 

Fig. 2.5b. Carbon map predicted by PCs of ETM image and topography

SUMMARY

Highly variable soil properties and topographical attributes of glacial till soils
cause difficulties in mapping soil carbon when the number of soil samples is limited. The
results of this study indicate that on-the-go field NIRS is capable of quantifying the
distribution of carbon in glacial till soil on a field-scale level rapidly and efficiently.
Including topographical data can firrther improve carbon mapping accuracy and thus such
data are strongly recommended in addition to NIRS measurements for effective and
accurate carbon predictions in glacial till and other soils. ETM imagery is useful for
assessing soil carbon in a bare field especially when combined with topographical data.

Field soil moisture and texture helped explain carbon variation and improved its

28

prediction for ETM image data, but were found to not be very useful additions to NIRS
measurements. Assessment of carbon predictions using independent test data sets showed
that predictions based on NIRS with topography were only marginally better in terms of

RMSE than the predictions obtained based on ETM imagery with topography.

ACKNOWLEDGEMENTS

We would like to thank Carr farm owner Dan Klein of Plainwell of Michigan for
allowing us to use his land for this research. We would like to thank Collin Christy from
Veris Technologies, Salina of Kansas for conducting NIRS measurements and generating
NIRS PCs and the Michigan Soybean Promotion Committee for funding the NIRS data

collection.

29

 

REFERENCES

Austin, RR. 1979. Soil Survey of Kalamazoo County, Michigan. United States
Department of Agriculture, Soil Conservation Service, in cooperation with

Michigan Agricultural Experiment Station. The Soil Conservation Service:
Washington, DC.

Bell, J .C., Grigal, D.F., Bates, RC. 2000. A soil terrain model for estimating spatial
patterns of soil organic carbon. In: Wilson, J .P., Gallant, J .C. (Eds), Terrain
Analysis —Principles and Applications. Wiley, New York, pp. 295— 310.

Blackmer, A.M., White, SE. 1998. Using precision farming technology to improve
management of soil and fertilizer nitrogen. Australian Journal of Agriculture
Research 49:555-564.

Brejda, J .J ., Mausbach, M.J., Goebel, J .J . 2001. Estimating surface soil organic carbon
content at a regional scale using the National Resource Inventory. Soil Sci Soc Am
J. 65:842-849.

Chang, C.W., Laird, D.A., Hurburgh, CR. 2005. Influence of soil moisture on near-
infrared reflectance spectroscopic measurement of soil properties. Soil Science.
170:244-255.

Chang, C.W., Laird, D.A., Mausbach, M.J., Hurburgh, CR. 2001. Near-infrared
reflectance spectroscopy -— principal components regression analyses of soil
properties. Soil Sci Soc Am J. 65:480-490.

Chen, F., Kissel, D.E., West, L.T., Rickman, D., J .C. Luvall, J .C., Adkins, W. 2005.
Mapping surface soil organic carbon for crop fields with remote sensing. J. Soil and
Water Cons. 60:51-57.

Chen, F ., Kissel, D.E., West, L.T., Adkins, W. 2000. Field-scale mapping of surface soil
organic carbon using remotely sensed imagery. Soil Sci Soc Am J. 64:746-753.

Chaplot, V., Bemoux, M., Walter, C. 2001. Soil carbon storage prediction in temperate
hydromorphic soils using a morphologic index and digital elevation model. Soil
Science 166:46-60.

Christy, C.D., Drummond, P., Laird, DA. 2003. An on-the-go spectral reflectance sensor
for soil. 2003. American Society of Agricultural Engineers Meeting. page number:
031044.

Christy, C.D., Lund, E., Laird, DA. 2003. Mapping soil carbon with on-the-go near
infrared spectroscopy.
http://www.oznet.ksu.edu/ctec/Fall%20Forum%20pdf%20files/

Papers_Abstracts/ Christy_Veris.pdf

30

Crerners, D.A., Ebinger, M.H., Breshears, D.D., Unkefer, P.J., Kammerdiener, S.A.,
Ferris, M.J., Catlett, K.M., Brown, J .R. 2001. Measuring total soil carbon with
laser-induced breakdown spectroscopy (LIBS). J. Envir. Quality. 30:2202-2206.

Dalal, R.C., Henry, R.J. 1986. Simultaneous determination of moisture, organic carbon,
and total nitrogen by near infrared reflectance spectrophotometry. Soil Sci Soc Am
J. 50:120-123.

Dematte, J .A., Campos, R.C. Alves, M.C., Fiorio, P.R., Nanni, MR. 2004. Visible-NIR
reflectance: a new approach on soil evaluation. Geoderma 121195-112.

ERDAS Imagine. 2001. ERDAS imaging help, version 8.5. ERDAS, Inc. Atlanta,
Georgia.

ESRI, 2004. ArcView and Arclnfo. version 9.0. Redlands, California.

Ehsani, M.R., Upadhyaya, S.K., Slaughter, D., Shafii, S., Pelletier, M. 1999. A NIR
technique for rapid determination of soil mineral nitrogen. Precision Agriculture.
1:217-234.

Ebinger, M.H., Norfleet, M.L., Breshears, DD. 2003. Extending the applicability of
laser-induced breakdown spectroscopy for total soil carbon measurement. Soil Sci
Soc Am J. 67: 616-1619.

Fox, G.A., Sabbagh, G.J. 2002. Estimation of soil organic matter from red and near-
infi'ared remotely sensed data using a soil line Euclidean distance technique. Soil
Sci Soc Am J. 66:1922-1929.

Fox, G.A., Sabbagh, G.J., Searcy, S.W., Yang, C. 2004. An automated soil line
identification routine for remotely sensed images Soil Sci Soc Am J. 68:1326-1331.

Frazier, B.F., Cheng, Y. 1989. Remote sensing of soils in the Eastern Palouse region with
landsat thematic mapper. Remote Sensing of Environment 28:317-325.

Hassink, J. 1997. The capacity of soils to preserve organic C and N by their association
with clay and silt particles. Plant and Soil. 191:77-87.

Homarm, P.S., Sollins, P., Fiorella, M. 1998. Regional soil organic carbon storage
estimates for western Oregon by multiple approaches. Soil Sci Soc Am J. 62:789-
796.

J enkinson, D.S., Coleman, K. 1994. Calculating the annual input of organic matter to soil
from measurements of total organic carbon and radiocarbon. European Journal of
Soil Science. 45:167-74.

31

J iang, P., Thelen, K.D. 2004. Effect of Soil and Topographic Properties on Crop Yield in
a North-Central Corn—Soybean Cropping System. Agron. J. 96: 252-258.

Konen, M.E., Burras, C.L., Sandor, J .A. 2003. Organic carbon, texture, and quantitative
color measurement relationships for cultivated soils in North Central Iowa. Soil Sci
Soc Am J. 67:1823-1830.

Kravchenko, A.N. 2003. Influence of spatial structure on accuracy of interpolation
methods. Soil Sci Soc Am J. 67:1564—1571.

Krull, E.S., Baldock, J .A., Skjemstad, J .O. 2003. Importance of mechanism and
processes of the stabilization of soil organic matter for modeling carbon turnover.
Functional Plant Biology. 30:207-222.

Lal, R., Griffin, M., Apt, J ., Lave, L., Morgan, MG. 2004. Managing soil carbon.
Science. 3041393.

Lal, R. 2004. Soil carbon sequestration impacts on global climate change and food
security. Science. 304: 1623-1627.

Leifeld, J ., Kogel, KL 2005. Soil organic matter fractions as early indicators for carbon
stock changes under different land-use? Geoderma. 124:143-155.

Martin, P.D., Malley, D.F., Manning, G. 2002. Determination of soil organic carbon and
nitrogen at the field level using near-infrared spectroscopy. Canadian Journal of
Soil Science. 82:413-422.

McCarty, G.W., Reeves, J .B., Reeves, V.B., Follett, R.F., Kimble, J .M. 2002. Mid-
infrared and near-infrared diffuse reflectance spectroscopy for soil carbon
measurement. Soil Sci Soc Am J. 66:640-646.

McBratney, A.B., M.L. Mendonca, and B. Minasny. 2003. On digital soil mapping.
Geoderma. 117 :3-52.

Moorman,T.B., Cambardella, GA, James, D.E., Karlen, D.L., Kramer, LA. 2004.
Quantification of tillage and landscape effects on soil carbon in small Iowa
watersheds. Soil and Tillage Research. 78:225-236.

Morra, M.J., Hall, M.H., Freebom, LL. 1991. Carbon and nitrogen analysis of soil
fiactions using near-infrared reflectance spectroscopy. Soil Sci Soc Am J. 55:288-
291.

Mueller, T.G., Pierce, F.J. 2003. Soil carbon maps: enhancing spatial estimates with
simple terrain attributes at multiple scales. Soil Sci Soc Am J. 67:258-267.

32

Qiang, X., Zhou, C., He, C. 2002. Using remote sensing to estimate the change of carbon
storage: a case study in the estuary of Yellow River delta. International Journal of
Remote Sensing. 23:1565-1580.

Palacios-Orueta, A., Ustin, S.L. 1998. Remote sensing of soil properties in the Santa
Monica Mountains 1. Spectral analysis. Remote Sensing of the Environment.
65: 170-1 83.

Rickman, R.W., Douglas Jr., C.L., Albrecht, S.L. 2001. CQESTR: a model to estimate
carbon sequestration in agricultural soils. J. Soil Water Cons. 56:237-42.

Robertson, G.P., Paul, E.A., Harwood, RR. 2000. Greenhouse gases in intensive
agriculture: contributions of individual gases to the radiative forcing of the
atmosphere. Science. 289: 1922-1925.

Russell, CA. 2003. Sample preparation and prediction of soil organic matter properties
by near infra-red reflectance spectroscopy. Communication of Soil Science and
Plant Analysis. 34:1557-1572.

SAS Institute. 2001. SAS user's guide, Statistics, Version 8.2. SAS Institute Inc., Cary,
North Carolina.

Sarensen, L.K., Dalsgaard, S. 2005. Determination of clay and other soil properties by
near infrared spectroscopy. Soil Sci Soc Am J. 69: 159-167.

Sullivan, D.G., Shaw, J .N., Rickman, D. 2005. IKONOS imagery to estimate surface soil
property variability in two Alabama Physiographies. Soil Sci Soc Am J. 6721789-
1798.

Terra, J.A., Shaw, J .N., Reeves, D.W., Raper, R.L., van Santen, E., Mask, PL. 2004. Soil
carbon relationships with terrain attributes, electrical conductivity, and a soil survey
in a coastal plain landscape. Soil Science. 169:819-831.

Walter, C., Viscarra Rossel, R.A., McBratney, AB. 2003. Spatio-temporal simulation of
the field-scale evolution of organic carbon over the Landscape. Soil Sci Soc Am J.
67:1477-1486

Wang, S.Q., Xu, J ., Zhou, C.H., He, CF. 2002. Using remote sensing to estimate the
change of carbon storage: a case study in the estuary of Yellow River delta.

International Journal of Remote Sensing. 23:1565-1580.

Young, L.M. 2003. Carbon sequestration in agriculture: the US policy context. American
Journal of Agricultural Economics 85:1164-1 170.

33

CHAPTER III
YILED VARIABILITY IN UNDULATING FIELDS:
RELATIONSHIP TO SOIL ELECTRICAL CONDUCTIVITY AND

TOPOGRAPHY, YIELD CLASSIFICATION AND PREDICTION

ABSTRACT

Crop yields are highly variable across fields as a result of complex interactions
among yield-affecting factors including soil properties, topographical features, and
precipitation. Understanding the relationship between yield and these factors is a critical
component of site-specific management systems. The objectives of this study were (i) to
characterize relationships of yield with soil electrical conductivity (EC) used as a proxy
for soil properties, and with topographical features; (ii) to evaluate temporal yield
variability under different precipitations during a multi-year period; and (iii) to identify
spatial clusters from historical yield data, and relate the clusters to the underlying soil EC,
and topographical factors, and to validate whether cluster groups can be used to
accurately predict yield patterns. Corn (Zea mays L.) and soybean [Glycine max (L.)
Merr.] yield data in two com—soybean fields in southwestern Michigan were obtained
from 1996 through 2003 using a combine yield monitor. Measurements of soil EC and
topographic attributes, such as elevation, slope, curvature, and wetness index, were
collected in the studied fields. Correlation analyses were used to investigate the
relationship between these variables and yield. Cluster analysis using multi-year yield
data was used to detect the spatial yield classes. Both EC and topographical attributes

were found to be helpful in explaining yield variability. Yields were negatively correlated

34

with EC and terrain slope, while positively correlated with wetness index in most years.
The relationship between standardized yield and early season (March and April)
precipitation was curvilinear, while yield increased linearly with the increase of July
precipitation. Roughly 60% of the yield variation was explained by either March and
April or July precipitation. Consistently low and high yield clusters were identified. Four
to five yield clusters can explain about 40 % of the yield variation. Areas with lower EC
and slope tended to form high yield clusters in the studied fields. The results indicated
that information from historical yield classification may be helpfirl in delineating

management zones within a field and predict yield patterns on a large scale.

INTRODUCTION

Generally, crop yields are highly variable across fields as a result of complex
interactions among yield-affecting factors including topographical features, soil
properties and management practices (Timlin et al., 1998; Kravchenko and Bullock,
2000; Kravchenko et al., 2005). The quantitative characterization of spatio-temporal
variability in crop grain yields is an important component of successful precision
agriculture applications (Bakhsh et al., 2000; Machado et al., 2002). Farmers will be
better able to implement site-specific management practices when they understand the
causes of spatial and temporal variability of yield in their fields (Fleming et al., 1999;
Kaspar, et al., 2004).

A number of studies have found that topography is the major source of yield
variability (Simmons etal., 1989; Timlin et al., 1998; Kravchenko and Bullock, 2000;

Kaspar et al., 2003; Si and Farrell, 2004; Jiang and Thelen, 2004). Field topography plays

35

an important role in the hydrological response of rainfall catchment and has a major
impact on water availability for crop production (Kitchen et al., 2003). In addition,
topographical attributes and soil properties are often highly correlated with each other
because of the processes of soil development, erosion, and sedimentation (Kaspar etal.,
2004). Lark and Stafford (1997) found yield differences between areas with different
topographic conditions because certain areas were associated with different soil types,
soil texture, or seedbed quality, drainage, and compaction. However, Taylor et a1. (2003)
reported that up to 50% of the yield variation within the field was related to crop
management practices rather than underlying soil factors.

The effect of topographical attributes on crop yield is indirect, acting by
influencing soil physical and chemical properties. Within a field, areas with relatively
low elevation often have higher soil moisture, finer texture and better nutrient status
compared with high elevation areas. High crop yields are often observed in lower
elevation areas. However, if the soil water availability is a major yield limiting factor in a
landscape the elevation-yield relationships may vary depending on whether a specific
year is wet or dry. Particularly, these variations are often associated with the prevailing
weather conditions during the growing season of a given year, such as spring and summer
precipitation (Halvorson and Doll, 1991; J aynes and Colvin, 1997). The greatest effect of
topography was often observed during extreme weather conditions and at locations with
extreme t0pography, such as low depressions or high hilltops (Kravchenko and Bullock,
2000). Kaspar et a1. (2003) found that in four years with less that normal growing season

precipitation, corn yield was negatively correlated with relative elevation, slope, and

36

curvature. In two years with greater than normal precipitation, yield was positively
correlated with elevation and slope.

Negative relationships between terrain slope and crop yield were found by a
number of studies (Ciha, 1984; Kravchenko and Bullock, 2000; Jiang and Thelen, 2004).
In general, steep slope positions tend to have more severe runoff and soil erosion, as a
result of lower soil productivity.

Surface curvature measures the rate of change of a slope. It can be positive or
negative, with zero curvature indicating that surface is either flat or the convexity in one
direction is balanced by the concavity in another direction (Wilson and Gallant, 2000). A
positive curvature indicates that the surface is upwardly convex. A negative curvature
indicates that the surface is upwardly concave. It can be used to describe the water flow
as run-in or runoff processes. Negative correlations between curvature and yield were
found during years with low precipitation because areas with concave shape (negative
curvature) could provide more plant-available water than the areas with convex shape
(positive curvature) during periods of drought (Kravchenko and Bullock, 2000).

Since yearly variations in soil wetness and moisture availability often appear to be
the reason for the yield-topography relationships to vary fiom year to year, a cumulative
topographical characteristic that effectively reflects soil moisture redistribution within a
landscape would be useful. Such characteristic, called soil wetness index (WI), has been
developed by Moore et al. (198 8) who presented the relationship between topographic
attributes and the distribution of surface soil water content in a small catchment, and
later, developed a method to calculate the WI. Late, Moore et a1. (1993) have used WI to

characterize surface saturation and soil water content in landscape. Schmidt and Persson

37

(2003) calculated WI from digital elevation model (DEM) data. They found that W1 is
highly correlated with soil water content and is helpful for site-specific management.

Soil EC is not a direct measure of any specific yield-limiting factor. However, it
reflects a number of soil characteristics related to plant growth and yield, such as soil
type (Taylor et al., 2003); topsoil thickness in claypan soils (Kitchen et al., 2003); clay,
soil water, and organic carbon content (J aynes et al., 2003). Soil EC can be considered as
a cost-effectively measure of spatially variable soil characteristics that affect crop growth.
However, similar to topography, the relationships between yield and soil EC could
fluctuate due to varied precipitation. Kravchenko et a1. (2003) found that variations in the
strength and direction of the relationship between yield and soil EC were related to the
amounts of precipitation observed early in the growing season. Also, the relationship
between EC and yield can be positive or negative depending on the type of clay in a
given site or the growing conditions of a given year. For example, the relationships
between a soil EC map and yield were not consistent and varied from field to field due to
varied subsoil or soil drainage (Kitchen et al., 2003).

Inexpensive and densely measured soil EC and topographical attributes were
widely used to explain yield variability (Kravchenko and Bullock, 2000; Kitchen etal.,
2003; J aynes et al., 2003) and identify management zones (Chang et al., 2003; Taylor et
al., 2003; Godwin et al., 2003; Larson, 2001). Available historical yield data could also
provide useful information about yield spatial variability and potential management zones
(Lark and Stafford, 1997; Doberrnann et al., 2003; J aynes et al., 2003; Perez-Quezada et

al., 2003). Additionally, yield maps can be used in conjunction with soil and landscape

38

information to help growers account for yield variability and provide insight into
improving production.

Characterizing spatial variation of field conditions, including EC and topography,
and exploring the relationships between yields and the underlying factors will provide
essential information for decision-making on site-specific management. Emerging
applications of advanced statistical procedures such as cluster analysis could be used to
identify management zones within fields and predict temporal yield patterns.

Cluster analysis was widely used to reveal patterns of soil homogeneity and for
identifying relationships among soil properties and landforms (Young and Hammer,
2000). In recent years, cluster analysis of historical yields has been used to develop
nutrient management zones (Fleming et al., 1999; Taylor et al., 2003) and economic
profit management zones (Blackmore et al., 2003). Combining yield maps, aerial
imagery, topography, and past management experience, enabled growers to develop
effective variable rate fertilizer application maps (Fleming et al., 1999). Perez-Quezada
et a1. (2003) reported that yield of one crop was a poor predictor of yield of crops grown
in different years, but after standardizing and averaging the yields, areas with the same
average performance tended to be clustered together spatially.

I used 8-year yield data and detailed topographical and EC information from a 50
ha area, to understand the yield-topography and yield-EC relationships and their effect on
yield variability. With this understanding it might be possible to predict early in the
season the extent and strength of spatial continuity in crop yield distribution and provide
recommendations for producers' decisions regarding size and location of potential

management zones. The objectives of this study were, (i) to characterize relationship

39

between yield and soil electrical conductivity (EC), and topographical features; (ii) to
evaluate temporal yield variability under different precipitations during a multi-year
period; and (iii) to identify spatial clusters from historical yield data, and relate the
clusters to the underlying soil EC, and topographical factors, and to validate whether

cluster groups can be used to accurately predict yield patterns.

MATERIALS AND METHODS

Field and Data Description

The study was conducted in two adjacent fields on Carr farm in Kalamazoo
County, Michigan (42.22 N, 86.36 W). Field 1 was about 30 ha, and Field 2 was about 20
ha in size. The predominant soil type is Kalamazoo loam (fine-loamy, mixed, mesic
Typic Hapludalf), with 0 to 6% slopes (Austin, 1979, Kalamazoo County Soil Survey,
Michigan). The site was non-irri gated and a minimum tillage system was employed
(J iang and Thelen, 2004). Crop rotation sequences in these fields are shown in Table 3.1.

Table 3.]. Crop rotation sequences in the study sites (NA: data not available)

 

Year 1996 1997 1998 1999 2000 2001 2002 2003

 

Field 1 corn soybean corn soybean corn corn soybean corn

Field 2 NA corn NA corn NA corn soybean corn

 

Corn and soybean yield data were recorded using a commercial yield monitor
mounted on a combine. Latitude and longitude for yield data points were recorded
simultaneously by a GPS receiver. Each point measurement covered an area of about 2-
by 5-m. In Field 2, because of malfunction monitor, yield data in 1996, 1998, and 2000
were not available. The yield data that were more than 3 standard deviations away from

the means were considered to be outliers and excluded from further analyses. The yield

40

point measurements were interpolated to regular 10- by 10-m grid with inverse distance
weighting method using ArcView Spatial Analyst (ESRI, 2000). For better yield
comparison among different crops, the original corn and soybean yield data were

standardized according to Blackmore (2000) and Perez-Quezada et a1. (2003).

Ysi = Rsi / X:- [1]
where Y5; is the standardized yield of the it" cell in the Sm year, R5,: is the actual yield for

the 1'“ cell in the 3“ year, Y5. is average field yield in that year. The Ysi = 1 means the

yield in the cell is equal to the average yield in the field in the S“ year.

Georeferenced EC measurements were taken in each field using Veris 3100 sensor
that operates on a principle of electromagnetic induction (Lund et al., 2001). The
measurements were taken every 3- to S-m, along transects 10 m apart. Two sets of EC
measurements were collected, corresponding to depths of approximately 0- to 30-cm
(shallow EC, ECvsh) and 0- to 90-cm (deep EC, ECvdp). The irregular point
measurements of EC were interpolated to regular 10- by lO-m grid with inverse distance
weighting method using ArcView Spatial Analyst.

Elevation was measured with a real-time kinetic GPS system mounted on a cart.
The measurements were taken approximately every 5- by 10-m. Irregular point
measurements of elevation was interpolated to regular 10- by 10-m digital elevation
model (DEM) with inverse distance weighting method using ArcInfo Grid (ESRI, 2000).
The primary topographical features, slope, curvature, flow direction, and flow

accumulation, were derived from the DEM using corresponding functions in surface

41

hydrologic analysis of ArcInfo GRID. Soil wetness index (W1) was calculated according

to Moore et a1. (1993):

WI = ln(AS / tan ,6) [2]
where A s is the upslope area derived from flow accumulation and tan ,0 is a tangent

fimction of the slope when measured as percent-rise. Daily precipitation data from 1996
to 2003 were obtained fi'om an automated weather station located about 10 miles from the
study site (KBS LTER Site Weather Station, 42° 24' 33" N, 85° 22' 18" W). The monthly
(March to August) precipitation values from 1996 to 2003 (Fig. 3.1) were used to relate
yield variation to precipitation across years. I created 10- by 10-m grid for the 2 fields
using Spatial Analyst of ArcView. Each grid cell contained yield, soil EC, and
topographical variables. The data sets then were exported from ArcView for statistical

analysis.

 

 

 

1 20
lAugust
18 .. BJuly
DJune
16 . lMay
E DAprIl
E 14 .. UMarch
.6 12 1 1
Et 1°
§ 8 -»
Q
=> 6 —»
N .
'U 4 0 c, 5,351 \r‘fl
1 2 " _'I,..:. \ -. :1 . i
1 0

 

 

 

1

1996 1997 1 998 1999 2000 2001 2002 2003 30-yr
year

1

 

Fig. 3.1. Daily average precipitation from March to August in the study years

42

Data Analysis

Correlation and Regression Analysis
Descriptive statistics for raw yield, EC, and topographical variables were
calculated by SAS PROC MEANS (SAS Inst, 2000). Spearman rank correlation

coefficients were employed to measure the association between yield data among years.

Spearman correlation (rs) is a nonparametric measure of association based on the ranks

of the data values. The rs is calculated as:

n 2
62(Rm‘ - RAJ")
rs =1 — ’=1 2 [3]
n(n — 1)

 

where R3,]- is the rank of the variable SiJ observed at location i on year j, and RiJ’ the

rank of the same variable, at the same location, but on date _] ’ , n is the number of the

observations. The rs can take values from -1 to 1.

Unlike Pearson correlation analysis, the rs does not require any assumptions

about the distribution of the values and independence of data sets. Yield data collected as
time series in this studies were not totally independent. Spearman correlation analysis
was superior compared with Pearson correlation analysis and has been widely used for
temporal stability analysis (Vachaud et al., 1985; Cassel et al., 2000; Van Pelt and
Wierenga, 2001). Pearson correlation coefficients were calculated between yields and soil
EC or topographic variables. The SAS PROC CORR was used to compute Spearman

rank and Pearson correlation coefficients.

43

Relationships between yields and precipitations were analyzed using regression
techniques. Standardized yields combined from two fields were used as a dependent
variable. Monthly or seasonal precipitations were used as independent variables. Totally,
13 paired standardized yield and precipitation data were used for the simple linear and
quadratic regression analyses. The SAS PROC REG was used to fit the regression model
and compute regression coefficients.

Cluster Analysis

Cluster analysis is used to classify individuals into two or more uniquely defined
subgroups called clusters. Within a cluster, the individuals have similar characteristics
with respect to the measured variables. However, the characteristics of individuals are not
similar to those in the other clusters (Johnson, 2000). In this study, each cluster represents
a certain area of the land with certain patterns of crop yield performance over the studied
years. When the information from cluster analysis is obtained, the underlying factors for
each cluster can be analyzed and recommendation for managing the field site-specifically
to obtain highest profit can be made.

The principle of cluster analysis is to minimize within-group variability while
maximizing arnong- group variability in order to produce relatively homogeneous groups
that are distinct from one another. The analysis is based on similarity between two
observations. The similarity measures Euclidean distance d(x, y) between two p-
dirnensional observations of x and y observations. The p is the number of measured

variables on each individual.

d(x.y) = J(x-y)'(x-y) [41

 

44

There are number of methods to group individuals into clusters. But all are based on the
distance.

In this study, 8-year yield data from 2775 cells in Fieldl, and 5-year yield data
from 2133 cells in Field 2 were used in a disjoint nonhierarchical cluster analysis using
SAS PROC FASTCLUS procedure (SAS Inst, 2000). The clustering algorithm attempts
to group the yield cells into clusters with high internal homogeneity and high between-
group heterogeneity. The clustering algorithm groups each yield cell such that it belongs
to one and only one cluster. A k-means method was used where the cluster centers are
based on a least sum-of-squares estimation (SAS help, SAS Inst, 2000).

Cluster analysis was run multiple times with different numbers of clusters in order
to determine the optimum number of clusters for the studied fields. Specifically, I
considered analyses with two, three, four, five, six, seven, and eight clusters. The criteria
that I used to estimate the best number of clusters were cubic clustering criterion (CCC)
and approximate expected over-all R2 (Khattree and Dayanand, 2000). The CCC has been
developed by the SAS Institute and measures the deviation of the clusters from the
distribution of points that would be expected if the points were actually from a uniform
distribution (Sarle 1983). The CCC is used to estimate the number of clusters based on
minimizing the within cluster sum of squares. It is obtained by comparing the observed
R2 to the approximate expected R2 using an approximate variance stabilizing

transformation.

l—E(R2)

CCC=ln
[ 1—122

]*K [5]

45

where E(R2) is expected R2, and K is the variance stabilizing transformation. The higher
CCC values indicate better clustering. The R2 is the goodness-of—fit measure resulting
from regressing a variable against the cluster means, calculated as a ratio of sum square
of regression to sum square of total (Sarle 1983). This statistics increases with increasing

number of clusters. It represents the proportion of variance explained by the clusters.

RESULTS AND DISCUSSION
Descriptive Statistics and Data distribution

Corn-soybean grain yields at the two studied Fields varied substantially from year
to year within the studied period (from 1996 to 2003) (Table 3.2). In Field 1, the highest
corn yield of 10.94 Mg ha'1 occurred in 2003, and lowest yield of 6.64 Mg ha'1 in 1996.
The greatest variation of corn yield within the field was observed in 2003, followed by
1996. The lowest yield variation was in 1998. The order of soybean yield from high to
low was 2002, 1997, and 1999. The standard deviations of soybean yield followed the
same order. In Field 2, the average corn yield ranged from 10.56 to 7.58 Mg ha". The
highest yield was in 2003, followed by 1997, 1999, and 2001. Only 1 year soybean yield

data were available in this field (Table 3.2).

The average standardized yields appear to be similar in both Fields (Table 3.2, Fig.
3.2). However, Field 2 had higher standard deviation of yield than Field 1 (Table 3.2, Fig.
3.3), which could be explained by higher variability of topographic features (Table 3.3,
Figs. 10a, 100, 11a, and 110). Within-field variation in EC and topographic features for
the two Fields is compared in Table 3.3. The average EC was higher in Field 1 than in

Field 2. The average and range of elevation for the two Fields were similar. In addition,

46

Field 2 appeared to have steeper slopes with the slopes in Field 2 ranging from 0.02 to
958° while in Field 1 from 0.03 to 7.08°.. In general, steep slopes and varied elevations
tend to result in more runoff and greater variability in soil water distribution within a
field (Moore et al., 1988; Lark and Stafford, 1997; Jiang and Thelen, 2004). In dry year,
more water may be available at low elevations and flat areas, benefiting crop growth
there. However, in wet year, low elevation areas could retain excessive water, potentially
negatively affecting crop growth. This could be the reason why yields in Field 2 were
more variable than in Field 1.

Table 3.2. Statistical summary of raw and standardized yields from 1996 to 2003 in
two Fields

 

 

Yield Standardized yield
Mg ha'1
Crop Year Avg. Std. Dev. Avg. Std. Dev.
Field 1 Corn 1996 6.64 0.91 0.73 0.14
Soybean 1997 2.6 0.33 0.90 0.13
Corn 1998 10.18 0.71 1.13 0.07
Soybean 1999 2.29 0.28 0.79 0.12
Corn 2000 9.59 0.81 1.06 0.08
Corn 2001 7.79 0.75 0.86 0.10
Soybean 2002 3.29 0.49 1.14 0.15
Corn 2003 10.94 1.39 1.21 0.13
Field 2
Corn 1997 9.2 1.73 1.02 0.19
Corn 1999 8.83 2.1 0.98 0.24
Corn 2001 7.58 1.28 0.84 0.17
Soybean 2002 3.39 0.56 1.17 0.17
Corn 2003 10.56 2.03 1.17 0.19

 

47

 

 

{- Ferd?~
LEE“ 21

   

 

0.5 ‘

standardized yield

 

 

0.0 _ .....

1996 1997 1998 1999 2000 2001 2002 2003
year

Fig. 3.2. Standardized yields in 2 Fields from 1996 to 2003

 

 

 

0.4

0.3 _ a: Field 1
a Field 2

 

standard deviation of
standardized yreld

 

 

 

 

1996 1997 1998 1999 2000 2001 2002 2003

year

 

A

Fig. 3.3. The standard deviation of standardized yields in 2 Fields from 1996 to 2003

48

Table 3.3. Statistical summary of EC and topographical variables in two Fields.

 

 

Fields Variables Descriptive Statistics
N Avg. Std. Min. Max.
Dev.

Field 1 ECvsh‘, mS/m 2637 8.45 1.78 2.7 22.9
ECvdp, mS/m 2637 6.67 1.96 3.2 22.1
Slope, degree 2775 1.48 0.89 0.03 7.08
Elevation, m 2775 298.0 2.88 292.6 303.0
Wetness Index 2775 7.99 1.50 5.05 14.6
Curvature, 10'2 m 2775 0.00 0.18 -1.88 5.65

Field 2
ECvsh, mS/m 1951 5.10 2.89 0.8 18.8
ECvdp, mS/m 1951 5.98 2.75 1.1 18.9
Slope, degree 2133 1.43 1.12 0.02 9.58
Elevation, m 2133 298.6 2.93 291.4 302.6

Wetness Index 2133 7.46 1.42 4.30 14.3
Curvature,10'2m 2133 0.00 0.29 -5.62 3.13

I ECvsh: shallow electrical conductivity; ECvdp: deep electrical conductivity.

 

Yield/precipitation relationship

Analyses of the growing season monthly precipitation and average yield data
across years showed that high corn yields were often associated with normal or higher
July precipitation (Fig. 3.1 and Fig. 3.4). Lower corn yields, such as in 1996 and 2001 in
Field 1 were coincident with drier July (Table 3.2 and Fig. 3.1). Similarly, lower soybean
yields in 1997 and 1999 coincided with drier July and August. Because corn and soybean
are in reproductive period in July, crop grth and yields are very sensitive to soil water

stress. Drought can cause remarkable yield loss in these nonirrigated fields.

Figure 5 shows the relationship between 8-yr standardized yields (2 crops and 2
Fields combined) and total precipitation of March and April. The relationship was best fit
with a quadratic curve with R2 of 0.68 (p-value $0.01). The data point of 1998 located in
the far right side was considered to be an outlier and was excluded from the regression

analysis. Year 1998 had very high spring precipitation and unusually high air

49

temperature, which produced higher yield than what could be expected based on typical
temperatures. The curve demonstrates that crop yield increased with the increase of early
spring rainfall before reaching the peak at 119.9 mm precipitation. Then, as precipitation
increased further yield gradually decreased. High yield required appropriate spring
rainfall. Based on this regression equation, total rainfall in March and April should be
between 105 and 135 mm in order to reach optimal yield. The estimation is within the
range of water requirements of corn and soybean production in Midwest (Hoeft et al.,
2000). The results suggested that either low or excessive spring precipitation tends to
decrease yield by delaying planting and seed emergence (Lark and Stafford, 1997). Less
rain can cause drought stress for seed emergence and growth. On other hand, excessive
rain can saturate soil leading to undesired crop growth condition, such as lower soil
temperature, and increased erosion that will severely affects early crop growth. The
standardized yields had very little increase with the increase of growing season (May to

August) rainfall (Fig. 3.6).

 

 

 

 

 

1.6 ~
.2 1.4 - y= 0.123x + 0.704
.2 R2 = 0.67
>.
u 1.2 r
9 8
.5 o
'2 1.0 i .
é o
0.8 —
.9, .
0.6 -
0.4 r . .
0 1 2 3 4 5
July Daily Precipitation, mm

 

 

 

Fig. 3.4. Standardized yields vs. July daily precipitation over 8 years (2 crops and 2
Fields combined)

50

 

 

 

 

 

 

 

 

 

 

F

1.4 -
'c 1.2 ~ ° Data
3 —frtted
E 1.0 ~ X
5
L
5 0.0 « 1998 /
g outlier

0 6 , y = -8.46 +0.16x - 0.0007):2

R2 = 0.60
0.4 . . . . .
80 100 120 140 160 180
Total Spring Precipitation (March and April, rrm)

 

 

 

Fig. 3.5. Standardized yield vs. total precipitation of March and April (2 crops and 2
Fields combined)

 

 

 

 

 

 

 

1.5 1
1 4 J y = 0.0002x + 0.8515
' R2 = 0.0155

2
e
-- 1.2 -
> . :
'1-3 1.0 1 A —-;
g. A
a "i O
E
5 0.8 ‘ O :
(D O

0.6 ~

0.4 . r . . .

200 250 300 350 400 450 500
Toal Growing Season Precipitation
(May to August, m)

 

 

Fig. 3.6. Standardized yields vs. total growing season (May to August) precipitation
over 8 years (2 crops and 2 Fields combined)

51

Spearman Correlation Coefficients among Yields
All Spearman correlation coefficients (rs) between crop yields in two fields from
1996 to 2003 were positive and statistically significant at probability p 5001 level
(Table 3.4). The r, between corn yields ranged from 0.1 to 0.58. The highest rs occurred

between 1996 and 2001 corn yields which both were dry in July. Extremely dry weather
during the key growth period probably was the dominant yield limiting factor, while in

normal or wet year, the weather factor could be minimal in affecting yield. Thus, yields

from years with optimal weather conditions are likely to be only weakly related. The r,
between soybean yields varied from 0.20 to 0.35. The range of r, between the soybean
and the corn yields were from 0.12 to 0.40. There was no evidence that r, between the

same crops are lower than that between different crops. Significant r, suggested that yield

of corn or soybean grown at the same location in one year could serve as indicators of
yield performance for one another in other years. However, correlation with quite low
value of the correlation coefficient was statistically significant due to the very large
number of data points in the studied data sets, even though it explained only a small
portion of yield variability. Similar observation was also made by Kitchen et al. (2003).
Perez-Quezada et al. (2003) reported that yield of one crop was a poor predictor of yield
of crops grown in different years while they studied wheat—tomato—bean—sunflower

rotation.

52

 

Table 3.4. Spearman Correlation Coefficients between Yields from 1996 to 2003 in
two Fields.

 

 

Year Crop 1997 1998 1999 2000 2001 2002 2003
Field 1 Soybean Corn Soybean Corn Corn Soybean Corn
1996 Corn 0.25§ 0.41 0.53 0.45 0.58 0.34 0.23
1997 Soybean 1.00 0.15 0.35 0.12 0.15 0.20 0.22
1998 Corn 1.00 0.26 0.34 0.41 0.21 0.21
1999 Soybean 1.00 0.39 0.34 0.29 0.20
2000 Corn 1.00 0.28 0.27 0.18
2001 Corn 1.00 0.40 0.25
2002 Soybean 1 .00 0.32
Field 2 Corn Corn Corn Soybean Corn
1997 Corn 1.00 0.41 0.40 0.40 0.22
1999 Corn 1.00 0.56 0.29 0.23
2001 Corn 1.00 0.34 0.22
2002 Soybean 1 .00 0.36

 

§correlation coefficients are significant at p-value < 0.01 level.

Still, in this study, Spearman correlation coefficients had the same signs in all
year yields indicating that the yields in com-soybean rotation were somewhat spatially
stable across years with varied weather. The stable variability of crop yield over time
suggested that management zones within the field are practically possible for site-specific
farming. This information will provide opportunity for further investigation of yield
limiting factors. However, single-factor correlation analysis provides very little direct
evidence for the cause of yield variation, thus it needs to be combined with other analyses

to provide better understanding about the yield stability across years.

Pearson Correlation Coefficients among Yield, EC and Topography

Pearson correlation coefficients between soil EC, topography and yields across
years in Field 1 and 2 are shown in Table 3.5. In most of the years, EC at shallow depth
was negatively correlated with yield. Conversely, EC at large depth was positively

correlated with yield in 5 out of 8 years. In these fields, higher portion of gravel and sand

53

contents were observed in top 24 cm depth (J iang, 2002). The deeper soil contains higher
portion of clay contents. The difference in relationships between yield and shallow and
deep EC could be attributed to varied soil texture and other physical or chemical
properties from shallow to deep horizon. Clearly, shallow EC mainly represents soil
texture. However, the direction of correlation coefficients between yield and shallow or
yield and deep EC were consistent (Sudduth et al., 1996; Kitchen et al., 2003). Their
fields had relatively uniform texture in the soil profile compared with glacial tilled soil of
this study. '

Table 3.5. Pearson Correlation Coefficients between EC, topographical variables
and yield from 1996 to 2003 in two Fields.

 

 

Year Crop EcvshlI Ecvdp Ele Slope Curv WI

Field 1
1996 Corn -0.16§ 0.06 0.16 -007 -022 0.22
1997 soybean -0.12 NS 0.15 NS 0.06 -0.05
1998 Corn NS 0.14 -0.17 -0.17 -0.07 0.15
1999 soybean -0.18 0.05 0.16 NS -0.10 0.09
2000 Corn NS 0.13 NS NS NS NS
2001 Corn -0.11 0.05 0.06 -0.10 -0.13 0.14
2002 soybean -0.14 -0.09 0.15 -0.13 NS NS
2003 Corn -0.10 NS 0.17 -0.08 NS 0.09

Field 2
1997 Corn 026 NS 0.21 -0.30 -O.16 0.17
1999 Corn -0.1 1 NS NS -0.24 -0.27 0.35
2001 Corn NS 0.20 -0.12 -0.35 -0.17 0.34
2002 Soybean -0. 10 0.08 NS -0.26 NS 0.05
2003 Corn -0.19 NS 0.14 -0.25 -0.06 0.14

 

IECvsh: shallow electrical conductivity; ECvsh: deep electrical conductivity; ele:
elevation; curv: curvature; WI: wetness index. §correlation coefficients are significant at
p-value < 0.01 level. NS indicates statistically non-significant.

Kitchen et al., (2003) noted that crop type and climate can produce very different
correlation coefficients between yield and EC from year to year. They speculated that
higher EC is associated with factors such as poor drainage and high-clay-content subsoils

that restricts root growth. Kravchenko et al. (2003) found that variations in the strength

54

and direction of the relationship between yield and soil EC were related to the amounts of
precipitation observed early in the growing season. I did not find this kind of relationship
in this study.

In Field 1, in 6 out of 8 years elevation was positively correlated with yield (Table
3.5). The Field is about 2000 m from Kalamazoo River. The condition of available soil
water in this field is good because of relative low ground water table. As a result the areas
with high elevation had other favorable conditions such as high soil temperature for crop
growth. The relationship was negative in 1998 and not significant in 2000. Those
negative or zero correlation coefficients could be related to rainfall amounts. Less than
normal in August in 1998 could be responsible for negative relationship between yield
and elevation (Fig. 3.1 and Table 3.5). Low elevation areas had higher soil water storage
in dry weather. This resulted in high yield in low elevation compared with high elevation.
In 2000, the precipitations from June to August were close to 30-year average, the
correlation coefficients between yield and elevation was not significant. The elevation
effect on yield in normal precipitation year, such as 2000, was minimal.

In Field 2, the relationship between yield and elevation were positive in 1997 and
2003, negative in 2001, and not significant in 1999 and 2002 (Table 3.5). Kaspar et al.
(2001) found that in years with less than normal growing season precipitation, corn yield
was negatively correlated with relative elevation, slope, and curvature. Conversely, in
years with greater than normal precipitation, yield was positively correlated with relative
elevation and slope. In this field, higher elevation areas are relative flat located at
southern part of the field. But low elevation areas in northern field are with high slopes.

The relationship between elevation and yield were more variable and could be affected

55

by slope. There were more interactions between elevation and precipitation due to greater
topographical variations in Field 2.

Slope had a significant negative correlation with yields in 5 out of 8 years and no
relationship in rest of 3 years in Field 1 (Table 3.5). The similar result was also found in
several previous studies (Ciha, 1984; Kravchenko and Bullock, 2000; and Jiang and
Thelin, 2004). Steep slope positions tend to have more severe water erosion. As a result
of thinner surface horizon and lower soil productivity in steep slope areas, the yields
there are often low. Consistently negative correlation coefficients between slope and
yields were observed across years in Field 2 (Table 3.5). Slope range in Field 2 was 2.5
degree larger than Field 1. As a result, slope played more important role in crop yield in
Field 2.

Correlation coefficients between curvature and yields were negative in most
years. The areas with negative curvature (concave shape) could provide more plant-
available water than the areas with positive curvature (convex shape). The relationships
between curvature and yields were similar to those between slope and yields. Kravchenko
and Bullock (2000) found that the relationship between curvature and yields could
change the direction depending on drought and wet weather. But the strength of the
relationship between curvature and yields was generally weaker in this study. The
differences in yield-curvature relationships in dry or wet years were not obvious in this
study as compared with their finding.

Yields were positively correlated with soil wetness index in most years in both
fields. Previous studies have found soil wetness index to be positively associated with

soil water content (Schmidt, and Persson, 2003). High values of the wetness index are

56

often present in converging and flat terrain, while low values were typical for steep and
diverging areas (Schmidt, and Persson, 2003). This is understandably because wetness
index is an inverse function of the slope. Steep slope positions generally have low
infiltration rate and less fine particle content and less available water stored in soil
(Moore et al., 1988; Wright et al., 1990; Moore et al., 1993). As a result, yield was low in
areas with low wetness index not only caused by less available water, but also
unfavorable topographical features. The relationship between wetness index and soybean
yield on the whole was weaker than that between wetness index and com. This could be
reflecting the greater water consumption by com than by soybean (Hoeft et al., 2000),

which makes corn to be more sensitive to field soil moisture than soybean.

Yield Classification by Cluster Groups

Figure 7 shows plots of CCC and R2 versus the number of clusters in Field 1.
These plots were used to make a decision on optimum cluster number. The highest CCC
value corresponding 2 clusters was equal to 47 with R2 value equal to 0.19. This R2 was
considered to be too small to explain variation of clustering if 2 clusters were chosen. The
second highest CCC values in Field 1 and Field 2 were obtained when the numbers of
clusters were 5 and 4, respectively (the plots of CCC and R2 versus the number of
clusters in Field 2 was not shown). The corresponding R2 were 0.37 and 0.43. Therefore,
5 clusters were considered to be the optimum number of clusters in Field 1 and 4 to be
the optimum number of clusters in Field 2. Then, each of the total 2775 yield cells in
Field 1 was assigned to one of the five clusters and each of the 2133 yield cells in Field 2
was assigned to one of the four clusters. The yields in the cells within each cluster were

more similar to one another than to the cells from the other clusters. Doberrnann et al.

57

(2003) also studied the classification of yield variability in irrigated fields using cluster
analysis. They suggested that the classification is improper if less than 50% of the yield
variability is accounted for and the number of clusters is less than five. However, for an
undulating and rainfed field, such as the fields of this study, this suggestion may not be
applicable. Varied topography greatly influenced the soil water distributions within a
small field area during the studied years. The available soil water could be highly
associated with elevation and curvature. The relationship can change the direction and
strength depending on dry or wet weather conditions. As a result, it was not possible to

get higher proportional variation explained by the classification.

 

 

50 0.5

40 1 10.4

8 30 - - 0.3
N
0 n:

20 - - 0.2

10 . :ggc value . 0.1

o - . . . . . . . 0.0

 

 

0 1 2 3 4 5 6 7 8 9
number of cluster

Fig. 3.7. The cubic clustering criterion (CCC) and over-all R2 vs. number of cluster
in Field 1

Figures 8 and 9 showed average standardized yields and their standard deviations
by year for each cluster in the two fields. As expected, there were clusters with
consistently high and low yields over years for each field. In Field 1, yields in clusters III

and V were above the average yield in most of the years (Table 3.6). Clusters III and V

58

occupied about 34 and 15 % of the field. The yields in clusters I and IV were below the
average. They occupied about 10 and 11 % of the field. The yields in cluster 11 fluctuated
around the average and accounted for 30 % of the total area. In Field 2, yields were
consistently high in cluster IV, and low in cluster II (Table 3.6). They accounted for 46 %
and 10 % of the total field-area, respectively. The yields in clusters I and III were near
average. The above clustering showed that yield patterns were stable over time. The
occurrence of spatially homogenous yield areas over time encourages the application of

site-specific management techniques.

 

 

 

 

Std Yield
3

 

+1—o—ll- -A— -II|-—)(—N—)§ -V

 

 

 

0.0 -- 4 . . . . 1 1
1990 1991 1999 1999 2000 2001 2002 2003

 

Years

 

 

Fig. 3.8. Average standardized yields from 1996 to 2003 in Field 1. The length of
vertical bars is equal to the cluster standard deviation. To avoid clutter in the figure,

the bars are shown only on one side (top) of each average.

59

 

Std Yield
3

0.8 1

 

 

 

 

1997 1999 2001 2002 2003

Fig. 3.9. Average standardized yields and their standard deviations represented by
vertical bars for each cluster from 1997 to 2003 in Field 2.

Factors Underlying Yield Clusters

The homogeneity of yield clusters could be related to some underlying field
factors. Finding the factors related to the yield patterns is helpful for practical
applications in site-specific farming. I examined characteristics of topographical and EC
variables in different yield clusters. Clearly, high shallow EC values corresponded to low
yield clusters in both fields (Tables 6 and 7). Conversely, low EC values corresponded to
high yielding clusters. The results agreed with the analysis of correlation coefficients
between shallow EC and yield (Table 3.5). Other studies demonstrated that in not saline
field soil BC was mainly affected by texture (Lund et al., 1999). In this field, shallow EC
was believed to be positively associated with sand content. High EC area represents
coarser-textured soils. Correspondingly, yield in those areas were low. Deep EC did not

Show a consistent trend among clusters (result not shown).

60

Table 3.6. Means of topographical variables and EC by cluster in Field 1 and 2
(Cluster ID and V are high, II is average, and I and IV are low yield in Field 1;
Cluster IV is high, I and [H are average, and II is low yield in Field 2)

 

 

Field 1 Field 2

Cluster I II III IV V I II 1H IV
Cells 281** 818 953 308 415 332 211 577 953
"/0 Variables* 10 30 34 . 1 1 15 16 10 28 46

ele 298.0 297.8 298.5 297.3 298.5 298.2 297.2 299.1 298.7
slope 1.81 1.43 1.60 1.52 1.36 1.60 2.43 1.55 1.10
curv 0.00 0.01 -0.02 0.05 -0.03 -0.04 0.07 0.07 -0.05
wi 7.66 8.02 8.01 7.55 8.28 7.40 6.78 6.98 7.87
ECvsh 7.00 6.91 6.41 6.98 6.26 5.38 7.13 4.92 4.67

*ele: elevation (m); curv: curvature (10'2 m); wi: wetness index; ECvsh: shallow EC
(mS/m); ** numbers of field map cells occupied by each cluster.

 

Generally, the most important effect of topography on yield consisted of
governing soil water distribution within a field. The yieldztopography relationship is
complex, particular in large fields and under varying climate conditions. Average values
of elevation for 5 clusters in Field 1 and 4 clusters in Field 2 were similar (Tables 6).
Overall, the elevation itself did not play an important role on yield spatial distribution in
the studied fields. Figures 10 and 11 showed a pattern of elevation decreasing from south
to north in both fields. Yield clusters did not follow a similar pattern. High as well as low
yield clusters were scattered around the fields.

The yield clusters appeared to be associated more with slope (Figures 10 and 11).
Slope, curvature, and wetness index explained more variation of yield patterns (Table
3.6). Greater slope and positive curvature values coincided with lower yield cell clusters.
Smaller slopes and negative curvature values occurred at high yield clusters. A steeper
slope results in increasing water flow and a greater potential for erosion. Negative

curvature indicates a concave surface which leads to decreasing flow and particle

61

deposition. Higher wetness index coincided with high yield cell clusters. Abundant but

not excessive soil water benefits plant growth and increases yield.

  
  
  
  
  
  

   

elevation, m

- 292.6—294.6
- 294.6-296 I
- 2961—2975
- 297.5499 2
- 2992-3012
- >301:

a b c
Fig. 3.10. Elevation (a), yield clusters (b), and slope (c) in Field 1.

F.

slope, degree
- 0.02-0.79
- 0.79.137
- 1.37-1.99
- 1.93-2.75
- 2.75-4 .05
- 4.05-7.08

 

  

elevation, m slope, degree

-29l.4-294.l " ficoroso
2941-2960 ' Ej 0.80-1.48
IE 2960-2979 “1.43—2.30
E: 2979-2997 ,. -2.30.3 42

L- - 3.42—5.70

- 2997-3010

e
- SOLO-302.6 _,_‘l .‘ - 570-958

 

a b c
Fig. 3.11. Elevation (a), yield clusters (b), and slope (c) in Field 2.

62

Yield Prediction from Cluster Groups

To further check accuracy of cluster analysis in predicting yield patterns, a
separate independent testing of cluster prediction performance was conducted. For that
purpose, I removed standardized yields for 1997 and 2002 out of the data sets for the two
fields. The cluster analysis was then conducted using the remaining yield data. The
average of cluster yields from the rest of the years could be used as a prediction for yield
of these clusters in 1997 and 2002. Table 3.7 shows the yield cell means by cluster and
the real standardized yields in 1997 and 2002. The real yield patterns in the two fields
agreed with the results from the cluster analysis. This indicates that the high yielding
cells of 1997 and 2002 indeed fell in the clusters with high yield as defined by the yield
data from the other years and low yield fell in the clusters with the lower yield.

To facilitate comparison between actual and cluster-analysis predicted yields we
grouped the clusters into three yield classes: high, medium, and low yielding. In Field 1,
cluster 11 was labeled as a high yield class. Cluster IV was labeled as a low yield class.
Clusters I, III, and V were labeled as a medium yield cell class. The areas of high, low
and medium yield cells predicted from the cluster analysis for the 3 classes accounted for
30, 11, and 59 % of the total area of the field. In 1997 the actual high, low and medium
yield cells accounted for 30, 17, and 53 % of the total area, respectively. In 2002, the
proportions of the actual high, low and medium yield areas were 34, 16, and 50 %,
respectively. Similarly good correspondences between overall sizes of areas with high,
medium, and low yields in cluster analysis predictions and actual 1997 and 2002 yields
were found in Field 2. High, medium and low yield patterns from cluster analysis were

consistent with actual yield patterns. Figure 12 showed spatial distribution of the 3 yield

63

 

cell clusters and actual standardized yields in 1997 and 2002 in Field 1. A number of
misclassifications existed. However, large scale predictions obtained from the cluster
analysis are considered to be reliable.

Table 3.7. Yield cell means by cluster in Field 1 and in Field 1 vs. the actual
standardized yields in 1997 and 2002

 

 

 

Cluster Predicted Actual
Means Mean Means
Fieldl 1996 1998 1999 2000 2001 2003 1997 2002

Cluster
1, N = 281 0.97 0.97 0.95 0.97 0.97 0.76 0.93 0.96 0.94
11, N = 818 1.14 1.04 1.07 1.04 1.08 1.06 1.07 1.04 1.08
111, N = 953 0.97 1.01 0.94 0.99 l 1.04 0.99 0.98 0.99
IV, N = 308 0.76 0.94 0.83 0.9 0.87 0.94 0.87 0.94 0.90
V, N = 415 1.01 0.98 1.14 1.03 0.96 0.99 1.02 1.04 0.99

Field 2
Cluster
1, N = 852 0.94 0.95 1.07 0.99 1.00 1.01
II,N=823 1.19 1.12 1.05 1.12 1.07 1.12
111, N = 261 0.86 0.79 0.66 0.77 0.84 0.85
IV, N = 109 0.35 0.94 0.85 0.71 0.83 0.82

 

     

1

$1002

:1 no data
- 0.34-0.91
- 0.91-1.07

 

 

a b c
Fig. 3.12. Spatial distribution of standardized yields in 1997 (a) and 2002 (c) and 6-

year yield clusters (b) in Field 1. Data in cells with light yellow in 2002 were missing.

64

SUMMARY

All Spearman correlation coefficient were positive and significant. The highly
significant coefficient suggested that yield of one crop spatially could be a predictor of
yield of crops grown in different years. In addition, stable yield variability over time can
be helpful to delineate sub-areas within a field for site-specific management. The
relationship between standardized yield and March and April precipitation exhibited a
quadratic curve. The yield increased linearly with the increase of July precipitation.
Roughly 60% of the yield variation can be explained by either March and April or July
precipitation. In general, shallow EC and slope were negatively correlated with yield. The
relationship between yield and deep EC was inconsistent. Wetness index was positively
correlated with yield. The role of elevation on yield was inconsistent among years. These
could be explained by varied precipitation among years. Consistently low, medium, and
high yield cells were clearly identified using cluster analysis. The 4 to 5 yield clusters
explained about 40% of the yield variation. Shallow EC values corresponded to high
yield clusters in the studied fields. Elevation did not play an important role on spatial
yield distribution. However, slope, curvature and wetness index values were closely
associated with yield patterns. Greater slope and positive curvature values resulted in
increased runoff. As a result, areas with higher slope and positive curvature values
experienced lower water and nutrient availability which led to the observed decrease in
crop yield. Historical yield data can be used to predict yield patterns. Corn and soybean

grain yield predictions using cluster analysis on a large scale were found to be reliable.

65

REFERENCES

Austin, FR. 1979. Soil Survey of Kalamazoo County, Michigan. United States
Department of Agriculture, Soil Conservation Service, in cooperation with

Michigan Agricultural Experiment Station. The Soil Conservation Service:
Washington, DC.

Bakhsh, A., T.S. Colvin, D.B. J aynes, R.S. Kanwar, U.S. Tim. 2000. Using soil attributes
and GIS for interpretation of spatial variability in yield. Transactions of the ASAE.
43: 819-828.

Blackmore, S., R.J. Godwin, and S. Fountas. 2003. The Analysis of spatial and temporal
trends in yield map data over six years. Biosystems Engineering. 84:455-466.

Cassel, D.K., O. Wendroth, and DR. Nielsen. 2000. Assessing spatial variability in an
agricultural experiment station field: opportunities arising from spatial dependence.
Agron. J. 92:706-714.

Chang,J., D.E. Clay, C.G. Carlson, S.A. Clay, D.D. Malo, R. Berg, J. Kleinjan, W.
Wiebold. 2003. Different techniques to identify management zones impact nitrogen
and phosphorus sampling variability. Agron. J. 95: 6, 1550-1559.

Ciha,A.J. 1984. Slope position and grain yield of soft white winter wheat. Agron. J.
76:193-196.

Dobermann, A., J .L. Ping, V.I. Adarnchuk, G.C. Simbahan, and RB. Ferguson. 2003.
Classification of Crop Yield Variability in Irrigated Production Fields. Agron. J.
95:1105-1120.

ESRI. 2000. ArcView 3.2 and ArcInfo. 8.1. ESRI, Redlands, CA.

Fleming, K.L., D.G. Westfall, D.F., Heermann, and J .V. Stafford. 1999. Farmer
developed management zone maps for variable rate fertilizer application. In Rust-

RH (ed.); the 2"d European Conference on Precision Agriculture, Odense, Denmark,
pp 917-926.

Godwin, R.J., G.A. Wood, J .C. Taylor, S.M. Knight, and J .P. Welsh. 2003. Precision
farming of cereal crops: a review of a six year experiment to develop management

guidelines. Biosystems Engineering. 84:375-391.

Halvorson, GA, and EC. Doll. 1991. Topographic effects on spring wheat yields and
water use. Soil Sci. Soc. Am. J. 55:1680—1685.

Heiniger, R.W., R.G. McBride and DE. Clay. 2003. Using soil electrical conductivity to
improve nutrient management. Agro. J. 95:508-519.

66

Hoeft, R. G., E.D. Nafziger, R.R. Johnson, and SR. Aldrich. 2000. Modern corn and
soybean production. MCSP Publications. Savoy, IL.

J aynes, DB, and TS. Colvin. 1997. Spatiotemporal variability of corn and soybean
yield. Agron. J. 89:30—37

J aynes, D.B., T.C. Kaspar, T.S. Colvin, and DE. James. 2003. Cluster analysis of
Spatiotemporal corn yield patterns in an Iowa field. Agron J. 95:574-586.

J iang, P. 2002. Effects of properties and landscape features on a Michigan corn and
soybean cropping system. MS. Thesis. Michigan State University. East Lansing.

J iang, P., and K.D. Thelen. 2004. Effect of soil and topographic properties on crop yield
in a North-Central com—soybean cropping system. Agron J. 96:252-258.

Johnson, DE. 1998 Applied Multivariate Methods for Data Analysts. Duxbury Press.

Kaspar, T.C., T.S. Colvin, D.B. J aynes, D.L. Karlen, D.E. James, D.W. Meek, D. Pulido,
and H. Butler. 2003. Relationship between six years of corn yields and terrain
attributes. Precision Agriculture 4: 1 79-192.

Kaspar,T.C., D. J. Pulido, T. E. Fenton, T. S. Colvin, D. L. Karlen, D. B. J aynes and D.
W. Meek. 2004. Relationship of corn and soybean yield to soil and terrain
properties. Agron. J. 96:700-709.

Khattree, R., and N.N. Dayanand. 2000. Multivariate Data Reduction and
Discrimination with SAS Software. BBU Press and John Wiley Sons Inc.

Kitchen,N.R., S.T. Drummond, E.D. Lund, K.A. Sudduth, and G.W. Buchleiter. 2003.
Soil electrical conductivity and topography related to yield for three contrasting
soil—crop systems. Agron J. 95:483-495.

Kravchenko, AN, and D.G. Bullock. 2000. Correlation of corn and soybean grain yield
with topography and soil properties. Agron J. 92:75-83.

Kravchenko, A.N., K. D. Thelen, D. G. Bullock, and N. R. Miller. 2003. Relationship
among crop grain yield, topography, and soil electrical conductivity studied with
cross-correlograrns. Agron J. 95:1132-1139.

Kravchenko, A.N., G.P. Robertson, K.D. Thelen, and RR. Harwood. 2005. Management,
topographical, and weather effects on spatial variability of crop grain yields. Agron.
J. 97:514-523.

Lark, R.M., and J .V. Stafford. 1997. Classification as a first step in the interpretation of
temporal and spatial variation of crop yield. Ann. Appl. Biol. 130:111-121.

67

Larson, W.E. 2001. Variability of soil and landscape attributes within sub-field
management zones. In Fridgen, J .J ., N.P. Kitchen, K.A. Sudduth, P.C. Robert (ed.);
Proceedings of the 5th International Conference on Precision Agriculture,
Bloomington, Minnesota, USA. pp 1-16.

Lund, E.D., P.E. Colin, D. Christy, P.E. Drummond. 1999. Applying soil electrical
conductivity technology to precision agriculture. Proc. the 4th International
Conference on Precision Agriculture. ASA-CSS-SSSA,Madison,WI.

Machado, S. E. D. Bynum, Jr., T. L. Archer, R. J. Lascano, L. T. Wilson, J. Bordovsky,
E. Segarra, K. Bronson, D. M. Nesmith, and W. Xu. 2002. Spatial and Temporal
Variability of Com Growth and Grain Yield Implications for Site-Specific Farming.
Crop Science 42: 1564-1 576.

Moore, I.D., G.J. Burch, and DH. Mackenzie. 1988. Topographic effects on the
distribution of surface soil water and the location of ephemeral gullies. Transactions
ofthe ASAE. 31:1098-1107.

Moore, I.D., P.E. Gessler, G.A. Nielsen, and GA. Peterson. 1993. Soil attribute
prediction using terrain analysis. Soil Sci Soc Am J. 57:443-452.Schmidt, F., A.
Persson.2003. Comparison of DEM data capture and topographic wetness indices.
Precision Agriculture. 4: 179-192.

Perez-Quezada, J. F ., G. S. Pettygrove, and RE. Plant. 2003. Spatial—temporal analysis
of yield and soil factors in two four-crop—rotation fields in the Sacramento Valley,
California. Agron. J. 95:676-687.

Sarle, W.S. 1983. SAS Technical Report A-108: The Cubic Clustering Criterion. SAS
Institute.

SAS Institute Inc. 1999. SAS/STAT User's Guide, Version 8. Cary, NC.

Schepers, A.R., J .F . Shanahan, M.A. Liebig, J .S. Schepers, S.H. Johnson and A. Luchiari,
Jr. 2004. Appropriateness of management zones for characterizing spatial
variability of soil properties and irrigated corn yields across years. Agron. J.
96:195-203.

Si, BC, and RE. Farrell. 2004. Scale-dependent relationship between wheat yield and
topographic indices: a wavelet approach. Soil Sci. Soc. Am. J. 68:577-587.

Simmons, F .W., D.K. Cassel, and RB. Daniels. 1989. Landscape and soil property
effects on corn grain yield response to tillage. Soil Sci. Soc. Am. J. 53:534—539.

Stein, A., J. Brouwer, and J. Bouma. 1997. Methods for comparing spatial variability
patterns of millet yield and soil data. Soil Sci Soc Am J. 61:861-870.

68

Sudduth, K.A., S.T. Drummond, S.J. Birrell, and NR. Kitchen. 1996. Analysis of spatial
factors influencing crop yield. p. 129—140. In P.C. Robert et al. (ed.) Precision
agriculture. Proc. Int. Conf., 3rd, Minneapolis, MN. 23—26 June 1996. ASA, CSSA,
and SSSA, Madison, WI.

Timlin, D.J., Ya Pachepsky, V.A. Snyder, and RB. Bryant. 1998. Spatial and temporal
variability of corn grain yield on a hillslope. Soil Sci. Soc. Am. J. 62:764-773.

Taylor, J .C., G.A. Wood, R. Earl, and R.J. Godwin. 2003. Soil factors and their influence
on within-field crop variability, Part 11: Spatial analysis and determination of
management zones. Biosystems Engineering. 84:441-453.

Vachaud, GA, A. Passerat de Silans, P. Balabanis, and M. Vauclin. 1985. Temporal
stability of spatially measured soil water probability density function. Soil Sci Soc
Am J. 49:822-828.

Van Pelt, RS, and P. Wierenga. 2001. Temporal stability of spatial measured soil matric
potential probability density firnction. Soil Sci Soc Am J. 65:668-67 7.

Wilson, J .P. and J .C. Gallant. 2000. Terrain analysis. John Wiley and Sons.
Wright, R.J., D.G. Boyer, W.M. Winant, and H.D. Perry. 1990. The influence of soil
factors on yield differences among landscape positions in Appalachian cornfield.

Soil Sci. 149:375-3 82.

Young, F.J., and RD. Hammer. 2000. Defining geographic soil bodies by landscape
position, soil taxonomy, and cluster analysis. Soil Sci. Soc. Am. J. 64:989-998.

69

CHAPTER IV

EFFECTS OF MANAGEMENT PRACTICES ON RELATIONSHIPS OF
CORN, SOYBEAN, AND WHEAT GRAIN YIELDS WITH FIELD

TOPOGRAPHY AND WEATHER
ABSTRACT

Crop yields are highly variable across fields and years as a result of complex
interactions among topography, weather conditions, and management practices. The
objective of this study was to analyze the effects of management practices on the
relationship between crop yield and topography, yield and weather using 10-yr of yield
data from a long-terrn corn (Zea mays L.)—soybean [Glycine max (L.) Merr.]—wheat
(T ritz'cum aestivum L.) rotation experiment in southwest Michigan. The 4 agronomic
treatments used in this study are chisel plowed with conventional chemical inputs (T1),
no-till with conventional chemical inputs (T2), chisel plowed with low chemical input
and a winter leguminous cover crop (T3), and organic-based chisel plowed with a winter
leguminous cover crop (T4). A nonparametric (spline) regression was used to
characterize boundary line of yields, that are maximum yields, versus wetness index
(W I). The algorithm was also used to compare the yield difference across the range of the
WI between two treatments.

The results indicated that the yield differences between the management practices
varied from year to year as a function of the prevailing precipitation and depending on
the topographical position. Management practices significantly interacted with

topographical attributes as represented by WI and influenced crop yields. The interaction

70

can be affected by summer precipitation. The yield difference between no-till and
conventional tillage was strongly influenced by WI in dry or normal year but not in wet
year. The relationship between boundary line yield data and W1 in most of the crops for
most of the studied years had a convex shape. Higher boundary line yields consistently
occurred at intermediate WI levels. No-till and low input systems tended to produce
higher maximum and average yields than conventional system at the lower WI areas. The
results suggested that it is possible to maximize yield and profit by farming site-
specifically based on landscape position. The nonparametric spline regression algorithm
utilized in the study was robust and efficient in comparing the yield differences between

treatments across a range of WI values.

INTRODUCTION

Conventional management practices relying on intensive tillage and high
chemical input were questioned with limited economic return and environmental
problems (Robertson et al., 2004; Pimentel et al., 2005). The management effects can be
altered by field and weather conditions. To make a better farming decision landscape- or
site-specifically needs thorough analysis and understanding of production variability and
its relationship with underlying environmental factors in multi-crop with long term data.
Generally, crop yields are highly variable across fields and years as a result of complex
interactions among different factors, such as topography, soil properties, and
management practices (Lamb et al., 1997; Doerge, 1999; J aynes et al., 2003; Kravchenko
et al., 2005). Therefore, better understanding of yield variation can lead not only to

increase in profitability of farming but also to improve environmental quality.

71

A large number of studies assessed the interactive effects of topography and
precipitation on yield variability for major Midwest crops, such as corn, soybean, and
wheat (e.g., Simmons etal., 1989; Timlin et al., 1998; Kravchenko and Bullock, 2000;
Kaspar et al., 2003; J iang and Thelen, 2004; Si and Farrell, 2004). Topography has been
often found explaining a substantial portion of yield variability, however,
yield/topography relationships vary substantially from year to year. These variations are
often associated with the prevailing weather conditions during the growing season of
each particular year. Spring and summer precipitation often found to be of most
importance (Halvorson and Doll, 1991; J aynes and Colvin, 1997; Kravchenko and

Bullock, 2000; Kaspar et al., 2003; J aynes et al., 2003).

Weather conditions may substantially influence the apparent gains or losses due to
management and other factors on a regional as well as field scale. Lobell and Asner
(2003) investigated the relationship between weather variations and crop production by
synthesizing data on temperature, precipitation, solar radiation, and county corn and
soybean yields throughout the United States for the period 1982-98. The key
characteristics were a large area centered in the Midwest where yields were favored by
cooler, wetter years and a smaller region including the Northern Great Plains favored by
hotter, drier years. Weather had a major influence on crop production in the Upper Great
Lakes States of Michigan, Minnesota, and Wisconsin during the past century. Low
precipitation and moisture stress were chief limitations to optimal corn and soybean
yields (Andresen et al., 2001). Overall weather effects on corn yield in Missouri were
found to operate through the within-season variations in rainfall and temperature (Hu and

Buyanovsky, 2003). The growing-season distributions of rainfall and temperature for

72

hi gh-yield years were characterized by less rainfall and warmer temperature in the
planting period and a rapid increase in rainfall further during the growing season,
particularly with more rainfall and warmer temperatures during germination and
emergence. More rainfall and cooler-than-average temperatures are the key features in
the anthesis and kemel-filling periods from June through August, followed by less
rainfall and warmer temperatures during the September and early October ripening time.
Opposite variations in rainfall and temperature in the growing season correspond to low

yield (Hu and Buyanovsky, 2003).

It is generally expected that the relationships between yield and topography and its
dependence on the weather patterns may vary under different management practices.
However, there is only limited quantitative information on how management affects
yield/topography/weather relationships. For example, Simmons et al. (1989) found that
yield reductions at drought-prone landscape positions such as shoulder and backslope
were most common under discing tillage. Kravchenko et al. (2005) observed greater
variability in crop yields under organic management as compared to conventional
management practices with weather related stresses amplifying the differences in spatial

variability patterns.

Of particular interest is assessment of these relationships in most commonly used
agricultural management practices, such as conventional tillage with conventional
chemical inputs, and comparisons with conservation management practices, including no-
till, low input and organic practices. Information on yield/topography/weather

relationships will contribute to the effort in expanding the conservation management

73

 

practices and likely increase the grower’s profit from site-specific farming while shifting

from conventional to conservation practice.

One of the commonly faced difficulties in analyses of yield relationships with
weather and topography is that the traditional correlation and regression data analyses
tools fail to provide useful information because of complex interactions between
topographical effects with a multitude of other yield-affecting factors. Boundary line
analysis (BLA) introduced by Webb (1972) provides insights on individual effect of
topographical factors in limiting the yield potential. BLA subdivides the response data
into groups corresponding to the quantitative categories of the limiting factor of interest
and then isolates a subset of the highest values from the response data within each group.
This upper boundary dataset in a scatter plot then represents the maximum possible
response to that limiting factor. The data points below the boundary line represent
conditions where the other factors have limited the yield. The method assumes that there
is a significant biological response between the potential limiting factor and a response
variable, and implies a cause-and-effect relationship (Webb, 1972; Lark, 1997; Kitchen et
al., 2003). BLA was used by a number of researchers to identify yield response to a
single factor out of the many that affected the yield (Lark, 1997; Kitchen et al., 2003;

Shatar and McBratney, 2004).

The relationships between crop yields and yield affecting factors oflen follow
complex patterns that, if described by parametric models, may require prohibitively
complex models with a very large number of parameters. Nonparametric regression

provides a useful alternative (Hardle, 1990). For example, nonparametric spline

74

regression can describe a complexly shaped relationship between a response and an
independent variable and can provide predictions for response variable averages (Fan,
and Gijbels, 1996; de Boor, 2001). Moreover, the algorithm developed by Wang and
Yang (2006) for the spline regression allows efficient constructing of a 100(1-(1) % spline
confidence band around predicted response values. This algorithm can be expanded in
comparisons between two-model predictions of crop yields across the range of the
studied yield affecting factors. The advantages of this method not only test the overall
difference between two treatments or models, but also visually show where the difference
is significantly positive or negative across the range of the covariate by plotting the data.
The information gained from this method is valuable for better understanding the

relationship between yield and yield-affecting factors and for managing the factors.

The physio graphy of Michigan presents very advantageous settings for studying
yield/topography/management interactions. It is characteristic of a nature glacial outwash
plain and moraine complex. Soil materials were formed and altered by glacial till and
outwash processes across the landscape resulting in formation of various landforms, such
as undulating hills, valleys and plains, even within relatively small areas (Mokma and
Doolittle, 1993; Crum and Collins, 1995). However, those varied field conditions can
cause imprecision and incorrect interpretation of the experimental results using traditional
analysis of variance (ANOVA) model. Incorporating spatially intense measurements of
topographical variables into ANOVA models, called analysis of covariance (ANCOVA),
one can correct for these confounding variables and obtain more precise estimates of

yield performance of the treatment (Milliken and Johnson, 2002). The ANCOVA

75

 

combines regression methodology with ANOVA. As a result, it removes the influence of

the covariate and makes the treatment comparisons more precise.

The objective of this study was to analyze the effects of management practices on
the relationship between crop yield and topography, yield and weather using 10-yr of
yield monitor data from a long-term com-soybean-wheat rotation experiment in

southwest Michigan.

MATERIALS AND METHODS

Site Description and Data Collection

The data were obtained from the Long Term Ecological Research (LTER) site,
located at Kellogg Biological Station (KBS) in southwest Michigan (85°24' W, 42°24' N).
Soils are well-drained Typic Hapludalfs of the Kalamazoo (fine-loarny, mixed, mesic)
and Oshtemo (coarse-loamy, mixed, mesic) series developed on glacial outwash (Crum
and Collins, 1995). The maximum difference in elevation within the site is 8.8 m. There
is clear decreasing gradient of elevation from north to south in the site (Fig. 4.1). There is

also considerable elevation variation within many individual plots.

76

elevation, m

D 282.8 -283.9
I: 283.9 -285.4
- 285.4- 286.9 1" ””5
- 286.9- 288.4
- 288.4 - 289.9
- 289.9 -291.5

 
  
 
 
 
  

'I'ZRO

 

Fig. 4.1. Layout of the 24 plots in the experimental site (4 treatments and 6
replications for each treatment), and the 4 x 4 m interpolated elevation map of the
LTER site at KBS. The plots are labeled with treatment codes and replication
numbers, e.g., T2R1 is replication 1 that received treatment T2. Each plot is 1
hectare in size.

A one-factor randomized complete block design (RCBD) experiment with 6

replications was established at the site in 1988. Each experimental plot is about 110- by

77

90-m in size. The experiment consists of a total of 7 treatments. The 4 agronomic
treatments used in this study are chisel plowed with conventional chemical inputs (T1),
no-till with conventional chemical inputs (T2), chisel plowed with low chemical input
and a winter leguminous cover crop (T3), and organic-based chisel plowed with a winter
leguminous cover crop (T4). All treatments are managed as com—soybean—wheat
rotations. Detailed agronomic protocols can be found on the KBS LTER website
(http://lter.kbs.msu.edu/protocols/104). The conventional chemical treatments T1 and T2
receive 123 kg nitrogen ha'lfor corn and 56 kg nitrogen ha'l for wheat but not for soybean
according to Michigan State University recommendations and insecticides and herbicides
as necessary. The low input system (T3) receives only N fertilizer in the amount equal to
3/5 of that applied to T1 and no herbicides or insecticides. The organic system does not

receive any chemical inputs.

Corn, soybean, and winter wheat grain yield data were collected via yield monitors
from all six replications of each treatment. Corn was grown in 1996, 1999, 2002 and
2005, soybean was grown in 1997, 2000, 2003, and wheat was grown in 1998, 2001, and
2004. Each yield data point covered an area of about 2 x 2.5 m (Fig. 4.2). Cleaning and
processing of yield data was conducted following Drurnmond and Sudduth's (2003)
recommendations. To eliminate the border effects, only the yield data from the central 80
x 60 m portion of each plot were used in the analysis (Fig. 4.2). The measured yield data
were interpolated into a cell-based yield map on a 4 x 4 m grid using ArcView Spatial
Analyst (ESRI, 2000). For analyses that involved comparisons among the crops, the

original yields were standardized (Blackmore, 2000; Perez-Quezada et al. 2003) as

78

Ysi = Rsi /Y.. [1]

where Y s,- is the standardized yield (StdY) of the [”1 plot in the sth year, R s,- is the actual
yield for the ith plot in the Sm year, Y , is average field yield across years. For corn Y" is
4-yr average. For soybean Y“ is 3-yr average. Y ,i = 1 means the yield in the it" plot is
equal to the average yield in the field in the Sm year. Because corn and soybean have

similar growing seasons that differ fi'om wheat, only combined standardized yield for 4-

yr corn and 3-yr soybean were used to relate weather variables across years to study the

crop yield temporal variation among treatments.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I I I I
: I I ' I
I I I : I
I I I I '
I I I I I I
I I I I I I
I I III II. I
.I ' j I ‘1‘:: I
I A s
I I I , ‘ t
I I 4 D E-
. I 1 I g
' I I ‘I. 2. u
I I I «I f I.
1 ' ' ‘ 2'
I I ‘II
: I i :II‘
. . - :--
I : :I‘
' I :I
I I ‘h
I I II
I I . :I
. I I I II
' ' : I ' : :I
I ' o :I
I I I I I I; .
I ' . I F: .-
I I I I I. III
I ' I i l I 1'.
I I I I I I I
I I I .-
' . ' 1 - :-
I I I A.
I I
' . : - . - =-
1A
' I I I ‘I
I I I I.
I ' I I f
I I ' fi '4- :-
I I I I I. ':I :-
I s
I I : I :::I 1’::
I I ’ I I. I. I-Il.
I I I :‘I: §:: :-
II I- I.- ' I I g
I ' '.
I-rrv

Fig. 4.2. Example of the layout of locations of the elevation (dots) and yield

measurements (triangles) obtained in each plot of the LTER site at KBS.

79

Elevation was measured with a real-time kinetic GPS system mounted on a cart.
The measurements within the agronomic plots were taken approximately every 2.5 m
along each transect (Fig. 4.2). The distance between transects was 5 m. The elevation
measurements were converted into a cell-based terrain map on a 4 x 4 m grid using
ArcView Spatial Analyst and topographic variables including slope, curvature, flow
direction and accumulation, and wetness index (WI) were derived from the elevation data
using surface hydrologic analysis of Arclnfo GRID. WI has been proposed as an
integrative characteristic describing distribution of soil moisture within a landscape
(Moore et al., 1988 and 1993) and was found to be highly correlated with soil water
contents (Schmidt and Persson, 2003). Based on preliminary analyses of topographical
variables (data not shown) and based on. the observations of yield/topography
relationships made at a nearby site (Huang et al., 2005), I decided to use WI as the only
topographical variable in this study. The definition and equation of wetness index were
described in previous chapters. In the studied site the WI data in all treatments were
approximately normally distributed with the mean of 7 and the standard deviation of ~1 .4.
The distributions of the WI data, as well as their mean, minimum and maximum values,

of the four studied treatments were very similar (Table 4.1).

Table 4.1. Descriptive statistics of topographical variables from 4 treatments.

 

elevation, m slope, degree wetness index

 

T1 T2 T3 T4 T1 T2 T3 T4 T1 T2 T3 T4
Mean* 287.4 287.6 286.9 285.8 1.19 1.27 1.14 0.85 6.99 6.94 6.92 7.03
Std deV 2.27 1.27 1.03 0.65 0.73 1.11 0.99 0.56 1.46 1.54 1.53 1.03
Min 283.7 285.6 284.0 287.0 0.01 0.01 0.01 0.01 4.23 3.64 3.13 4.39

Max 291.1 291.4 289.4 288.6 4.85 6.47 10.0 3.24 14.4 13.1 13.8 13.3
*sample size for T1, T2, T3, and T4 were 2672, 2931, 2816, and 3210, respectively.

80

Daily precipitation values were obtained from an automated KBS weather station
located at the LTER site. The average daily precipitation values from March through
August were used to relate yield variation to weather. The spring (March, April and May)
and summer (June, July, and August) monthly daily precipitation averages for the studied

period and 30-year average are shown in Fig. 4.3.

 

 

24 T IAugust
El July
20 L B June
< E! May
_ [J April
16 - . :1 March

 

daily preclpltation by month (mm)
.2;

 

 

   

J

 

1997 1998 1999 2000 2001 2002 2003 2004 2005 mean 1
year
Fig. 4.3. March, April, May, June, July, and August daily average precipitations

 

 

 

from 1996 to 2005 and 30-year average.

Statistical Analyses
Correlation and Regression Analysis
Descriptive statistics for yield and topographical variables for each treatment were
calculated by SAS PROC MEANS procedure (SAS Inst, 2000). Relationship between
StdY and precipitations were analyzed using regression techniques. StdY was the
dependent variable. Spring and summer daily precipitations were the independent

variables. Simple linear regression was applied to analyze the 7-yr spring and summer

81

daily precipitations on StdY. Multiple regressions were employed to analyze the
combined spring and summer daily precipitations on the StdY. The SAS PROC REG was
used to fit the regression model and to compute regression coefficients. Partial correlation
coefficients between StdY and spring precipitations where the effect of summer
precipitations on the correlation was removed were calculated using SAS PROC CORR.
Similarly, partial correlation coefficients between StdY and summer precipitations were

calculated.

Boundary Line Analysis

The relationship between WI and yield was studied using BLA. The analysis starts
with subdividing the range of values of the independent variable into classes. Then the
maximum yields are identified within each class. Within literature, identification of the
maximum yield within a class has been approached in different ways. For example,
Webb (1972) and Schnug et a1. (1996) selected a single highest-yielding data point for
each class of the yield-affecting predictor variable. However, Kitchen et al. (2003)
selected the upper edge of the yield by using the data points exceeding the 95th percentile
of each class yield data. This selection method may produce only few data points in some
classes while numerous data points in other ones. In this study, I divided the WI into
approximately 50 classes in 0.2 increments and in each class I selected the top four yield
data points. This method of selecting the upper edge data appeared as a way to use the
best features of the methods reported in literature. It represented the upper edge yield in a
more reliable manner than a single maximum data point and resulted in an equal number
of data points representing each class. Example of the yield-WI data for one of the

treatments in one of the studied years is shown on Fig. 4.4.

82

 

 

 

 

 

 

13 . A umer edge data
.7“ - rawdata
fa 11
E A A
N“ 9 - .
E fit" “ ‘ A a
.E 7 - ‘1‘“ “A
'2

5 - --
:1
L-
e
U 3 -

l 'r r r l 3 l

3 5 7 9 ll 13 15
Wetness Index

 

 

 

Fig. 4.4. Scatter plot and boundary line of corn yield vs. wetness index (WI) for T1
in 2002. Dots represent raw data. Triangles represent top 4 yield data for each
increment of WI.

Parametric and nonparametric regression methods were used to fit the boundary
line yields and W1. For parametric regression, after examining scatter plot, 2-line
regression model, also called segmented linear regression, was used to quantify the
relationship between maximum yield and WI. The segmented linear regression model is

defined as:

Y=fl01+fl11X,XSA

Y=fl02+fl12X,XZA [2]
where Y is the response variable yield and X is independent variable WI; fig, and flag are

the intercepts of the segments of the line, #11 and ,0 12 are the slopes, respectively, and A

is the joint points. The procedure of SAS PROC NLIN with Marquardt method was used

83

to perform the segmented regression as described by Shuai et al. (2003). By comparing
the slopes and joint points of fitted models, the influence of WI on maximum yield can be

revealed.

Visual assessment of the boundary yield lines showed that their relationships with
WI can not be adequately described by a single regression line. It also appeared not
possible to use a single non-linear equation for describing various shapes of the
relationships observed in different treatments and years. Thus, I decided to apply
nonparametric (spline) regression following the methodology developed by Wang and

Yang (2006). The statistical model of spline regression can be expressed as:

Yi =m(Xl')+G(Xl°)el-,l=1,...,n. [3]
where (Xi, K) are paired predictor and response variables; m is unknown regression

function; 0' and e are the standard deviation of the predictor variable and the observation
errors, respectively. It is believed that the nonparametric regression is an effective tool to
extract information from complex and less structured data (Hardle, 1990; Fan, and
Gijbels, 1996; de Boor, 2001; Wang and Yang, 2006). In addition to capturing better
trend of the complicated data, by constructing confidence bands for differences between

_ 2 spline models from the two treatments, one can compare mean functions m1(x) and

mz(x) of the treatments 1 and treatment 2 and test the hypotheses:

H03m1=m2 against H13m1¢m2 [4]

84

where m, and m; are unknown regression functions for two models. In fact, the

acceptance or rejection of H 0 is according to whether prediction difference AY = 0 line

lies completely inside the confidence band. By plotting the difference between the two
predicted firnctions with 95% confidence bands provides valuable information about
where the yields between two treatments differ across covariate WI. The detailed theorem
and algorithms to fit spline regression and construct the confidence bands were described
by Wang and Yang (2006). The codes for implementing the analysis have been written in

R by Wang.

AN OVA and ANCOVA
The effect of treatments on yield was first analyzed using ANOVA model. The

statistical model with lO-yr and 3 crops is specified as:

Yijkl =p+]}+Bj +Ck +X(Ck)1+TC,-k +TXi1+TB,-j +TBCl-jk +83“ [5]

where Y is measured yield; T, C, and X are fixed treatment, crop and year nested with
crop effects, respectively; TC and H are interactions between treatment and crop,
treatment and year; B is random block effect; TB and TBC are random interaction effects;
8 is random error. As the interaction between treatment and crop, treatment and year was
significant, one-way ANOVA and ANCOVA model were used to determine the

treatment effect for a specific crop on a year. One-way ANOVA model with subsampling

is defined as:

85

ljjkzy+7}+BJ-+£ij+dijk [6]

8,-1- ~ and. ~ N(0,0'ez)
dijk ~ i.i.d. ~ N(O,0'd2)

where Y is measured yield based on subsample within a plot; [I is the overall mean; T is

the fixed treatment effect; B is the random effect of the replications, 8 and d are the

random error and subsampling random error. One-way ANCOVA model with

subsampling is defined as:

where Y is measured yield based on subsample within a plot; 11 is the overall mean; T is
the treatment effect; ,62 is the topographical covariate and its interaction with treatments;
B is the random effect of the blocks; 8 and d are the random error and subsampling

random error. The analyses of variance and covariance were performed using SAS PROC

MIXED procedure.

RESULTS AND DISCUSSION

Descriptive Statistics of Topographical Variables and Crop Yields

The descriptive statistics of topographic variables in each treatment are shown in
Table 4.1. The elevation ranged from 283.7 to 291.4 m, that is, it was approximately 8 1n
difference between highest and lowest area where 4 treatments were located. In addition,
the relative range of elevation differed largely among 4 treatments, for example, the

range for T1 was 7.4 m, whereas the range for T4 was only 2.8 m. The areas with high

86

elevation were mainly at the northwestern corner of the field, and the areas with low
elevation were at the southern part of the field (Fig. 4.1). The entire field gradually
descended from north to south. Moreover, there was up to 2 m elevation difference within
1 hectare plot. The histograms of elevation for 4 treatments differed remarkably
(Appendix 4.2). The distributions of elevation were flat in the T1 and T4, and skewed in
the T2 and T3. In addition, the range of slope varied largely from one treatment to
another. However, WI ranged from 4 to 13 for each treatment (Table 4.1). The average
values and their ranges of WI were similar among treatments. The histogram of WI
showed that the data were approximately normally distributed in each treatment (Fig. 4.5).
My previous results demonstrated that WI was the topographical variable most strongly
related to crop yields (Huang et al., 2005; Chapter III). Thus, the WI was selected as

covariate to represent topography for yield variability characteristics.

The yield variations in corn, soybean, and winter wheat within a treatment across
years are presented in Table 4.2. Average yield for each crop varied remarkably from
year to year. Particularly for corn, for example, the yield in the T1 was 8.25 Mg ha'1 in
2005 that was more than twice of the yield in 1996 (3.66 Mg ha'l) and in 1999 (3.06 Mg
ha’l). The primary reason was due to varied inter-annual weather conditions. The year-to-
year variations observed in corn yields were higher than in wheat and soybean yields. All
yield data were close to normal distribution (data not shown). As an example, histogram
of corn yields in 2002 is shown in Appendix 4.3. Detailed yield comparison and the

effect of covariate W1 on yields between treatments will be reported in later section.

87

Frequency
100 150

50

Histogram of WI in T1

'—I

—I

 

Frequency
80 120

4D

 

 

 

 

 

 

5

8

T—l
10 12

wetness index

Histogram of WI in T3

l—lfifi

4 5

10

wetness index

Histogram of WI in T2

150

Frequency
100

 

. 3i

 

I
A

l |

5 8

wetness index

Histogram of WI in T4

150 200

Frequency
50 102]

U

wetness index

Fig. 4.5. Histograms of wetness index (W I) for 4 treatments.

Table 4.2. Average yield and standard deviation for corn, soybean and winter wheat

from the four studied treatments.

 

 

Crop year Yield (Mg ha'I) in treatment:
T1 T2 T3 T4
Corn 1996 3.66 :1: 2.08 4.35 :1: 1.69 4.34 d: 1.52 2.87 :t 1.24
1999 3.06 d: 1.44 3.36 :I: 1.28 3.95 i 0.96 4.08 t 1.09
2002 6.35 i 1.79 7.08 i 1.38 7.24 d: 1.85 6.10 :1: 2.10
2005 8.25 :t 2.53 9.56 i 1.75 8.80 d: 1.94 5.80 i 1.84
Soybean
1997 1.37 d: 0.67 1.48 i 0.41 1.68 :t 0.53 1.20 d: 0.43
2000 1.67 :t 0.24 1.64 i 0.36 1.84 i: 0.30 1.95 :t 0.26
2003 1.50 :t 0.43 1.92 i 0.72 1.36 i 0.28 1.22 :h 0.27
Wheat
1998 3.03 i 1.13 1.99 :t 0.49 1.51 i 0.37 0.73 i 0.28
2001 3.62 i 0.58 3.18 :1: 0.67 3.08 i 0.42 2.27 i 0.36
2004 4.53 :t 0.92 4.59 i 0.93 4.21 :t 0.83 2.44 i: 0.61

 

88

Yield/ Precipitation Relationship among Management Practices

Since corn and soybean have similar growing seasons, StdY from 4 years of corn
and 3 years of soybean yields were combined for analyses of yield relationships with
precipitation. The three years of wheat yields were excluded from this analysis. In the
studied site precipitation in May and precipitation of all three spring months combined
was found to be not significantly related to the StdY (p<0.05) (data not shown). Thus,
only the results for early spring (March through April) precipitation are discussed further.
The StdY for the T1, T2, and T3 were significantly negatively correlated with early
spring (March and April) daily average precipitation (p < 0.01, Fig. 4.6A, 6B, and 6C).
Roughly 30 to 60% of the StdY variation can be explained by the spring precipitation.
The negative correlation of the StdY with spring precipitation indicates that with the
increase of the precipitation StdY decreased accordingly. It is likely due to unfavorable
wet spring from excessive rainfall tends to reduce the yield by lowering soil temperature,
delaying planting, and slowing seed emergence (Lark and Stafford, 1997; Hoefi et al.,
2000). Noticeably, the relationship was possibly driven by com data from 2005, where
high yields were in company with low precipitation in spring and normal precipitation in
summer (Fig. 4.3 and 6). But it did not appear that there was much correlation between
the yield and spring precipitation in the other years. The StdY in organic system did not
show significant correlation with the spring precipitation (Fig. 4.6D and Table 4.3).
Contrary to conventional and reduced input systems, the yields in organic system did not

show a remarkable high in 2005.

89

Table 4.3. Parameters of regression model for the relationship between standardized

(Std) yield and spring and growing season daily rainfall from 4 treatments.

 

 

T1 T2 T3 T4
Std Yield vs. March and April Rainfall
N 40 36 41 37
p, 037% 0.48:: -0.34b -0.04a
r2 0.31 0.59 0.39 0.01
RMSE (Mg ha") 0.31 0.24 0.24 0.26
Std Yield vs. June, July, and August Rainfall
N 40 36 41 37
B: 0.54b 0.59c 0.55bc 0.39a
r2 0.50 0.62 0.75 0.56
RMSE (Mg ha") 0.26 0.23 0.15 0.17

 

*regression slopes within a row followed by the same letter are not significantly different at the

0.01 level by the 2-side t-test.

 

 

 

 

    

 

 

 

A 2.5
yield vs. rainfall in yleld vs. rainfall in June
‘ March and Aprill July and August

2.0 . y= 0.36: +1.57 y = 0.54): ;O.40
.- ‘ r2 = 0.310 r2 = 0.50 .
'—
"E 1.5 " O o O
2 § .0.
.2 . 8
> . ' O
u 1.0 ‘ ho ' 2
a
a) 8

O o . I
O ‘ o C o
e
0.5 . 8 - e
Linear (March and April) 8
- - - - Linear (June,July, August)
0.0 r I T 1 T
0.0 1.0 2.0 3.0 4.0
Daily precipitation, n'm

 

 

Fig. 4.6A. Standardized (Std) yield vs. Spring (march and April) and Summer
(June, July and August) Daily Precipitation for treatment T1.

90

 

 

     

 

 

 

 

l B 2.5
yield vs. rainfall in yield vs. rainfall In June
1 March and Aprlll July and August
2.0 . y= -0.47x + 1.90 y = 0.59): - 0.43
0 r2 = 0.59 r2 = 0.62 0
N
'—
5 1.5 - t
o e ‘
1: .0
33 g . '
>4 ’ o
.0 1.0 ‘ g 0 o O
H
(D o o s
' o e
0.5 - ° 9 ° 3
Linear (March and April)
- - - - Linear (June, July, August)
0.0 v I v I I I V
0.0 1 .0 2.0 3.0 4.0
Daily precipitation, m

 

 

 

Fig. 4.63. Standardized (Std) yield vs. Spring (march and April) and Summer (June,
July and August) Daily Precipitation for treatment T2.

 

 

 

 

 

 

 

 

 

f
l C 2.5
\ yield vs. rainfall in yield vs. rainfall in June
I ‘ March and Aprill July and August
I 2,0 . y = -0.34x +1.63 y = 0.55): - 0.35
m ‘ r2 = 0.39El r’ = 0.75.
'—
5 1.5 .
2
.2
>4
3 1.0 -
(I)
0.5 -
Linear (March and April)
Linear (June, July, August)
0.0 V T ' I V U ‘—
0.0 1 0 2.0 3.0 4.0
Daily precipitation, mm

 

 

 

Fig. 4.6C. Standardized (Std) yield vs. Spring (march and April) and Summer
(June, July and August) Daily Precipitation for treatment T3.

91

 

 

 

 

 

 

 

 

D 2.5
yield vs. rainfall in yield vs. rainfall in June
March and Aprill July and August
2_o .1 y = -0.04x + 0.95 y = 0.39x - 0.10
r2 = 0.01 r2 = 0.56
Q
l-
5 1.5 -
3 A
2 A § A if A
>‘ 1 o - A . - ' A
B A u A A I ’ T A ‘
m
A ‘ ,m
0.5 - 3 t
Linear (March and April)
- - - - Linear (June, Jule, August)
0.0 . . . f .
0.0 1.0 2.0 3.0 4.0
Daily precipitation, mm

 

 

 

Fig. 4.6D. Standardized (Std) yield vs. Spring (march and April) and Summer
(June, July and August) Daily Precipitation for treatment T4.

d

Std Yield in T1

 

Fig. 4.7a. Standardized (Std) yield vs. summer (X) and spring season (Z) daily
precipitation (mm) for treatment T1.

92

Std Yield in T2

 

Fig. 4.7b. Standardized (Std) yield vs. summer (X) and spring season (Z) daily
precipitation (mm) for treatment T2.

J

Std Yield in T3

 

1

Fig. 4.7c. Standardized (Std) yield vs. summer (X) and spring season (Z) daily
precipitation (mm) for treatment T3.

93

.s
0|

Std Yjeld in T4
'o

   

.°
01

Fig. 4.7d. Standardized (Std) yield vs. summer (X) and spring season (Z) daily
precipitation (mm) for treatment T4.

The regression slope and the r2 value were the highest in no-till (T2) (Fig. 4.6B).
This indicates that crop yields in no-till tended to vary more with the amount of
precipitation in early spring. The no-till yields decreased at a greater rate with an increase
of the spring precipitation as compared with the tilled treatments (Table 4.3). During wet
and cold springs, tillage was often observed to improve soil aeration and soil temperature
thus creating more favorable field conditions for planting and seed emergence as
compared with no-till (Cox et al., 1990; Fortin and Pierce, 1991; Wilhelm and Wortrnarm,
2004; Kravchenko and Thelen, 2007). But in a drier year, no-till can reduce soil water
losses from evaporation (Tomer etal., 2006) thus reducing the drought stress and

benefiting crop growth as compared with tilled soil.

94

The StdY of all treatments were significantly positively correlated with the
summer (June, July, and August) precipitations (p < 0.01, Fig. 4.6 and Table 4.3).
Roughly 60 % of the StdY variability can be explained by the summer precipitation.
Similar to the spring precipitation, the relationship between StdY and summer
precipitation in no-till had steeper slope compared with tilled treatments, demonstrating
that yields in no-till had stronger response to rainfall than those in tilled treatments.
Table 4.4. Results of multiple regressions and partial correlation coefficients

between standardized (Std) yield with spring and growing season rainfall from 4
treatments.

 

Treatment Spring + Growing Partial r of Std yield vs.
season rainfall
R2 RMSE (Mg ha") Spring rainfall Growing rainfall

 

Tl 0.53 0.26 .024NS 0.56"
T2 0.77 0.18 -O.62*** 0.65m
T3 0.78 0.15 -0.31* 080*"
T4 0.78 0.12 o.72*** O.89***

 

113 indicates no significant correlation; *, **, and *** represent significant Pearson partial
correlation coefficients at p < 0.05, 0.01, and 0.001 levels.

The combined effects of spring and growing season precipitation on StdY were
analyzed using multiple linear regression (Table 4.4). Three dimensional plots of the
multiple regression results for 4 treatments are shown in Fig. 4.7. Combining early spring
and summer precipitations effects on StdY in a multiple regression further supported the
above discussed results and allowed for further insight into spring and summer
precipitation effects on yields of the studied treatments (Table 4.4). In general, the R2
values for T2, T3, and T4 were similar (about 0.78), while the R2 value for T1 was much
lower (0.53), indicating that the yields in high chemical input with conventional tillage
system were less influenced by precipitation as compared with no-till, low chemical input

or organic systems. The stronger relationship between precipitation and yields in low

95

chemical input and organic treatments likely reflects a higher response of low input
systems to additional stresses, e. g. water stress, that further enhance the stress due to
lower availability of plant nutrients that the plants might be already experiencing. Similar

results were found by Kravchenko et al. (2005).

To further assess independent contributions of the early spring and summer
precipitations to the StdY, I calculated partial correlation coefficients (rp) between the
StdY and early spring precipitation, by partialing out the influence of summer
precipitation and the rp values for the summer precipitation by partialing out the early
spring precipitation (Table 4.4). The rp values for the early spring were negative in the T2
and T3, not significantly different from 0 in T1, and positive in T4. This further
supported the previous observations of the negative effect that wet springs might have on
T2 and an extra positive effect that better water supply may have in the organic system,
while not being crucial in conventional tilled system. Independent of the early spring
precipitation, the adequate water supply in summer had positive effect on yield in all
treatments, still being more pronounced in the systems with lower chemical inputs (T3
and T4). The rp in the T4 was the strongest (0.89), followed by in the T3 and T2 (p-value

< 0.001). The rp in the T1 was the lowest (0.56, p-value < 0.01).

It is interesting to note that in 1999, the extremely dry year, the organic system
had higher corn yield than the conventional system (Table 4.2). Higher yields in organic
systems as compared to conventional were observed by other studies as well, but not for
corn. Pimentel et al. (2005) found that during the extreme drought of 1999 in

Pennsylvania, organic legume system of Rodale farming trial had significantly lower com

96

yields and higher soybean yields than the conventional system. Average soil water
content in the organic legume system was found to be 15% higher than that in the
conventional system thus explaining the higher soybean yields in the organic system
(Pimentel et al., 2005). In this study, the organic treatment was found to not have higher
soil water content but have substantially higher soil organic matter content than in the
conventional chemical treatment (Hao and Kravchenko, 2007). The management systems
with cover crops are also known for having better soil structure with stronger aggregate
stability and higher numbers of large aggregates (Kabir and Koide, 2000; Sainju et al.,
2003; Liu et al., 2005). Increased soil organic matter and better aggregation improve soil
hydraulic properties, which may have helped the plants in the T4 to better survive the

drought conditions.

Relationship between Boundary Crop Yields and Wetness Index

Data Examination of Boundary Yields and Wetness Index

An examination of scatter plot from large data sets can be very helpful in
understanding the association between variables. BLA focuses on the upper edge of the
scatter-plot data. Those data marked as triangles represents the maximum yield response
to corresponding WI as an independent variable (Fig. 4.4). The yields below the edge
represent the conditions where other factors have limited the yield. Because of limited
space, I arbitrarily presented here only the data for corn yields in the T1 in 2002. The rest
of plots had similar pattern. The relationship between boundary yields and W1 in most of
the crops for most of the studied years had a convex shape. Higher boundary yields

consistently occurred at intermediate WI levels. Corn yield increased over a range of WI

97

from 4 to 6, and then reached to a plateau. Yields were then depressed as WI values rose

further (Fig. 4.4).

Parametric versus Nonparametric Regression

First, the boundary yields and W1 were modeled using two-segrnented linear
regression. The analysis assumed that the paired yield-WI relationship is characterized by
2-linear-line with different slopes. Through comparing the slopes and joint points
between the treatments, one can gain insight on how WI affects the maximum yield. The
two lines were connected by a joint point. The regression parameters for corn yields
across 4-yr among 4 treatments were compared in Appendix 4.1. The model 12 ranged
from 0.14 to 0.61. The joint points were the highest for the T4 in 3 of 4 years, followed
by the T1, T2 and T3. This means that the highest yields in T1 and T4 tended to be closer
to the areas with lower level or footslope than in T2 and T3. In addition, the T1 yields
had steeper slope in both the first and second lines compared with the T2 and T3. This
indicated that yields in conventional system were highly variable to WI in term of field

soil water moisture compared with no-till and reduced system.

98

Corn 2002, T1 Corn 2002, T2

 

 

4
14
L

12

Yield, Mglha
8 10

 

 

 

 

 

 

 

14
l

12

Yield. MCI/ha
8 10

6
1

 

 

 

 

 

 

l T

4 6 8 10 12 4 6 8 10 12
wetness index wetness index

Fig. 4.8. Comparisons of 2-linear-line and spline regression for corn yield in 2002 vs.
WI. Dots are top 4 yield data. Dotted lines indicate 95% confidence band. Dash line
is spline line.

To test whether the segmented regression fit the boundary data better than
nonparametric spline regression, a 95% confidence band for the data was constructed
(Fig. 4.8). If the confidence band does not cover the fitted line, the parametric regression
fit should be given up. Although the fitted line is covered by the band, the parametric
regression is not guarantied a better fit compared with nonparametric regression for a

complicated data (Wang, 2006, personal communication). The 95% confidence band

99

showed that there were considerable deviations across WI values. The bands tended to be
wider at the initial and the end of WI values where relatively few data points were
available and yields were more variable. The segmented regression lines were not
entirely covered by the 95% confidence band except for the T4 (Fig. 4.8). In addition, the
regression coefficients in spline regression were higher than in segmented regression in 3
of 4 treatments (Table 4.5). The means square errors were smaller using spline regression
than using segmented regression (Table 4.5). The results indicated that spline regression
had superiority over segmented regression. Moreover, spline regression not only can be
used to predict average response, but also can be used to compare difference by
constructing confidence interval for the difference of 2 treatments. Therefore, it is more
appropriate to characterize the maximum yield response to WI using nonparametric
spline regression. I applied spline regression to fit all data and compare the yield

difference between 2 treatments.

The relationship between boundary yields and W1 varied across years among
treatments. Figure 9 shows the splines for boundary yields of corn and W1 in 4-yr among
4 treatments. The general trends of the lines were similar to 2-segmented lines. The spline
curve, however, captured the local variation across the entire range of WI. Although
maximum corn yields were the highest in the T1 and lowest in the T4 in 1996 and 2005,
no consistent higher or lower yield for a specific treatment was observed in 4-yr period.
Corn yields in the T2 and T3 were less variable from year-to-year compared with the T1
and T4. Moreover, the yields in the T1 were more variable to lower WI values compared
with the T2 and T3, that is, the yields of T1 were very low in the lowest WI values, but

with the increase of WI corn yields had greater increase rate except in 2002 (Fig. 4.9).

100

The splines for boundary soybean and wheat yields and W1 are shown in Appendix 4.4

and 4.5. The shapes and patterns of the spline were similar to that for corn yields.

Corn 1996 Corn 1999

 

 

8 10 1214

Yield. Mqlha
6

4

 

 

 

 

2

 

 

 

 

 

12

Yield, Mglha
1o

 

 

 

 

 

 

”

a i

g _

N —1
1* l l l l l l l T j
4 6 8 1o 12 4 s 8 1o 12

wetness index wetness index

Fig. 4.9. Boundary line of corn yield vs. WI fitted by spline regression.

101

Table 4.5. The r2 and mean square error (MSE) between nonparametric (spline) and

parametric regression (segment) from 4 treatments in corn yield of 2002.

 

 

Treatment MSE(Mg ha") r2
Spline Segment Spline Segment
T1 0.90 1.00 0.49 0.43
T2 0.71 0.65 0.58 0.61
T3 0.91 0.91 0.38 0.37
T4 1.17 1.21 0.35 0.33

 

Maximum Yield Difference Influenced by Wetness Index

From all the possible comparisons that could be conducted among the studied
treatments I have selected the three that appeared to be of most practical importance. The
comparison between the T1 and T2 addresses the weather and topography considerations
involved in choosing no-till as a conservation alternative to convention tillage. The
comparison between the T1 and T3 evaluates the weather and topography related
concerns involved in deciding on reducing the chemical inputs and using cover crops;
and the comparison between the T3 and T4 evaluates the effect of complete elimination
of chemical inputs. For each comparison the differences between the spline regression
predictions of the boundary yields for each treatment were calculated for each class of WI

values and examined along with 95% confidence band.

The differences and their confidence bands plotted versus WI for the 4-yr corn
yield boundary line data are shown on Fig. 4.10. The difference was statistically
significant at 0.05 probability level as can be interpreted from the 95% confidence

interval band that did not include zero. The T1 yields were significantly higher than T2

102

across most of the WI values in 3 of 4 years. In 2002, the yields between 2 treatments did
not differ statistically because the confidence intervals included zero. Noticeably, at the
lower WI values (< 5, at the summit/steep—sloped areas), the T2 yields were significantly
higher than T1 in 1999 and 2005, whereas the T2 yields equaled to T1 at WI > 10
(depression areas). The results suggested that the T2 tended to produce higher yields at
the lowest WI, and lower yields at intermediate WI levels than Tl. Speculatively, T2
probably performed better than T1 at lower WI due to reduction of erosion and
evaporation, increase in soil carbon and improved soil structure that were often observed
as a result of long-term implementation of the no-till (Reicosky et al., 1995;

VandenBygaart et al., 1999; Rhoton, 2000).

For soybean the T2 yields were lower than T1 at higher WI values in 1997. On
contrary, the T2 yields were significantly higher than T1 across the whole range of WI
values in 2000 and 2003 (Appendix 4.6). On the other hand, the T2 yields of wheat were
lower than T1 across the whole range of WI values in 1998 and 2001 (Appendix 4.7). But,
no difference between the two treatments was found in 2004. Noticeably, either soybean
or wheat yield difference between the T1 and T2 become smaller from year 1 to 3. The
results suggested that no-till had considerable yield advantage over conventional tillage
in a long-term run. The benefits from no-till were primarily due to improved soil

properties as discussed earlier.

Corn yields in the T1 were significantly higher than in T3 in 1996, 1999, and
2005 across most of the WI values (Fig. 4.11). However, the yields in the T1 were lower

than in the T3 in 2002. At lowest WI values, the T3 had clear yield advantages over T1 in

103

3 years. As in the case of T1 and T2 comparisons, this might be attributed to the
improvement in soil organic matter and soil structure due to long-term implementation of
leguminous cover crops of the T3 system, which might be particularly beneficial at the
eroded areas with lower WI. Soybean yields in the T1 were significantly higher than in
T3 in 1997 and 2003 (Appendix 4.8). But in 2000, the T3 had higher yields than T1.
Wheat yields in the T1 were consistently higher than in T3 in 3 years (Appendix 4.9). It
was noted that the wheat yield difference between the T1 and T3 became smaller from
year 1 to year 3. Pimentel et a1. (2004) observed the similar results in the Rodale
experiments in Pennsylvania. They found that nitrogen levels in the organic systems have
improved and have not been limiting the crop yields after the first 3 years. In the short
term in organic systems, there may be nitrogen shortages that may reduce crop yields
temporarily. But this disadvantage can be eliminated by raising the soil nitrogen level

through the use of leguminous cropping systems.

Corn yields in the T3 were significantly higher than T4 in 1996 and 2005, lower
than T3 in 1999, respectively (Fig. 4.12). In 2002, no yield difference was found between
the 2 treatments. Soybean yields in the T3 were significantly higher than in T4 in 2 of 3
years (Appendix 4.10). Wheat yields in the T3 were consistently higher than in T4 in 3
years (Appendix 4.11). Overall, the low input system tended to produce higher corn and
wheat yields compared with organic systems. This might be attributed to the fact that the
T3 receives nitrogen fertilizer while T4 does not. The yield difference lines for most
comparison paralleled to the horizontal line. This result means that covariate WI did not

have profound effect on comparison of yield difference between the T3 and T4.

104

Yield Difference, Mglha

Yield Difference, Mglha

Em. .93. 53 ea 3.x. 8352.8 can; em :8 End—6:8 cm 8:. Beam—5:: 105 78:8: .3 BE A.» 8 a 55032. 3. <5.

—2

-4

-4 -2

.6

 

.: IA.» 5 Sea

 

 

 

 

 

_ J _ _
.3 I .3 :— use»

 

.3 IHN :— Howe

 

 

 

 

 

 

 

 

Sansone Snox

 

 

 

\II\\|III\\ !.I\
u .66 lilf\\
\\Il\
_ _ _ _
H—Ifingnccm
\l |Il Ifl\
...k.ll -.\oo.. \l
\ I
_ _ _ _
m w m o

53:68 .snox

105

Yield Difference, Mglha

Yield Difference, Mglha

-4 —2

5 -6

-4 —2

_6

 

.3 law 3 Se...

.3 law :— See

 

 

 

 

 

 

 

.H— lad :— nccn

 

 

 

Illlll

 

 

 

 

_ _ _ _ A
m .s m w 3 3

£238 :3:

m

_ _
.2 law :— Nccm
i— a

N m

_ _ _
w 3 2

£3.53 .saox

En. 9:. 53 e». 8.x. 8.5358 use: 3. =8 Ens—.5358 o». 8:. Ban—EE— EoE 338: .3 25 Hu 8 a 2:532— en $.—

106

Yield Difference, Mglha

Yield Difference, Mglha

'4 -2 0

.6

 

 

Hm LHA 5 ~35

.Hu lad =— See

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Sansone Enex

\ \
\ I
.1
Ill‘llli||\il|\ '1 con-e I
.u...\.I.l........4...... u... [Iii/I I \
o \ i ll! .\ I l
- \\\ I I \\\ II II use-nee-sues‘ /f\\\lil\
\ \i\|il sees-seoseoses.e<ses-
\1 \\ II \II I
\ I|| I/ \I’I
\ l
_ _ _ _ _ _ _ _ _ _ _ _ _
a \
, .HuIHAm—unecm
- I ~
flak—£533 71/ .7 x.
I u I \I \ \
I use. I'llI’l\ ‘1‘
[1“ I... 8 one.
[I \\ I)\|il.-o.oy.o~e.qu.o.
...... illlill I ii \ I
.III' 'I\ \ "\I'I’ \‘I’ \II
\\I||’II {Ieeos.O.e--.-o-e \ V r
\ illll \ill I
\ I|
II a
a
|..
_ _ _ _ _ _ _ _ _ _ _ _ _ _
m a N m o no 3 m m N m w no 2

$3308 Enex

Em. ab“. :3 a. 3.x. 25392.8 525 3. =5 95.2.38 3. 02.: 5855:: «.35 .5232. ad 25 a... an a 5:532. e». g.

107

BLA is a useful diagnostic tool to identify yield potential and estimate the
magnitude of yield loss due to variation in field WI. This analysis involved a single factor.
It ignored the interaction effects between variables. Therefore, caution must be taken
when insufficient data space coverage exists, and when interactions between one or more
variables exist (Chambers et al., 1985). That problem can lead to oversimplify the reality
of yield and overestimate the plant response to an environmental factor. Shatar and
McBrataney (2004) used multivariate boundary line analysis to investigate the
relationship between yield and soil properties. They found that major yield limiting factor

within a field can vary from a location to another.

Relationship between Average Crop Yields and Wetness Index

Comparison of AN OVA with AN COVA for RCBD

From F -test of ANOVA table, the average yield for a crop among management
practices and years and their interactions were significantly different (data not shown).
Crop yields often varied across years primarily due to inter-annual variation of weather
conditions. In addition, field topography can interact with precipitation resulting in
different treatment performance from year to year. Because this was lO-yr with 3 crops
rotated in a RCBD experiment, detailed treatment comparisons for a specific crop on a

year is more meaningful to explain the effect of topography and weather on yield.

The ANOVA and ANCOVA models were compared to select better one for
treatment comparisons (Table 4.6). The model fit statistics Akaike Information Criterion
(AIC) for ANCOVA were always smaller than for ANOVA. The smaller AIC indicates a

better model fit (Littell et al., 2006). The p-values from ANOVA for corn yield in 3 of 4

108

years were greater than 0.05. Similarly, the p-values for soybean yield in 2 of 3 years
were greater than 0.05. The higher p-value indicates that the effects of treatments on yield
were not likely to be different. However, ANCOVA showed that the interaction between
treatment and covariate WI was statistically significant (p < 0.001) for all crop and years.
This indicates that there was a significant difference between the slopes of yield versus
WI among treatments. Areas within a plot with different WI values responded differently
to the applied treatments. The variance of residuals for ANCOVA was consistently
smaller than for ANOVA. The variances of block and block by treatment in most of years
were smaller in ANCOVA than in ANOVA. The smaller variance in ANCOVA can

undoubtedly improve the predictive accuracy of a model.

Table 4.7 presents the average yield comparison between ANOVA and ANCOVA
with WI at 7 (average) among 4 treatments. Corn yields in 3 of 4 years did not differ
among treatments by ANOVA. However, the results from ANCOVA showed significant
yield difference among treatments. Similar results were found for soybean. For wheat,
however, the test results between ANOVA and ANCOVA were the same. This could be
explained as when the difference between treatments was large enough the covariate was
less useful in comparing the treatment effect. Table 4.7 also shows that ANCOVA
reduced standard errors for treatment means and thereby improved the accuracy of
estimates of yields for treatment means. Similar results were found by Kravchenko et al.
(2005) when they using soil electrical conductivity as a covariate to determine the soil P
among different treatments. Kaspar et al. (2006) found that terrain covariates can help to

remove most of the spatial variability of soil carbon and to detect much smaller

109

differences between treatment means. Therefore, I applied ANCOVA model to estimate

least-squared means of crop yield and compared the treatment and W1 effects on yield.

Table 4.6. Model fit statistics and variance components.

 

 

model AIC p valuei: Residual blk blk*trt
Corn
1996 ANOVA 24458 0.108 1.81 0.25 1.02
ANCOVA 23944 <.0001 1.68 0.27 0.93
1999 ANOVA 25291 0.149 1.01 0.00 0.59
ANCOVA 24064 <.0001 0.89 0.49 0.88
2002 ANOVA 39545 0.202 2.21 0.20 1.15
ANCOVA 38877 <.0001 2.07 0.20 1.15
2005 ANOVA 44530 <.0001 3.09 0.31 0.99
ANCOVA 43640 <.0001 2.85 0.35 0.89
Soybean
1997 ANOVA 9648 0.08 0.23 0.00 0.06
ANCOVA 8964 <.0001 0.21 0.00 0.04
2000 ANOVA -755 0.075 0.05 0.00 0.04
ANCOVA -872 <.0001 0.05 0.00 0.04
2003 ANOVA 5298 0.019 0.09 0.02 0.12
ANCOVA 5239 <.0001 0.09 0.02 0.12
Wheat
1998 ANOVA 14376 <.0001 0.27 0.00 0.25
AN COVA 13668 <.0001 0.25 0.00 0.22
2001 ANOVA 7997 <.0001 0.16 0.05 0.08
ANCOVA 7584 <.0001 0.15 0.05 0.07
2004 ANOVA 27515 <.0001 0.65 0.02 0.03
ANCOVA 27456 <.0001 0.65 0.02 0.03

 

AIC: Akaike Information Criterion.

blk: block.

blk*trt: interaction of block x treatment.

I: p value of F -test for treatment effects in analysis of variance (ANOVA), and for interaction
between treatment and covariate in analysis of covariance (ANCOVA).

110

Table 4.7. Mean yield estimates (Mg ha'l) among 4 treatments with and without

wetness index (WI) as a covariate.

 

Treatment (standard error of the treatment mean)

 

Model T1 T2 T3 T4
Corn
1996 ANOVANS 3.74(O.46) 4.53(O.64) 4.34(0.46) 2.86(0.50)
ANCOVA§ 3.72(0.45)ab+ 4.52(0.62)b 4.34(O.45)b 2.86(0.49)a
1999 ANOVANS 3.16(0.31) 3.36(0.34) 3.97(0.31) 4.07(0.31)
ANCOVA 3.14(0.29)a 3.36(O.31)ab 3.99(0.29)b 4.06(O.29)b
2002 ANOVANS 6.28(0.48) 7.10(0.48) 7.24(0.48) 6.08(0.48)
ANCOVANS 6.28(0.47) 7.09(0.47) 7.27(0.47) 6.07(0.47)
2005 ANOVA 8.34(O.47)b 9.56(0.47)c 8.81(0.47)bc 5.79(0.47)a
ANCOVA 8.32(O.46)b 9.57(0.46)c 8.84(0.46)bc 5.78(0.46)a
Soybean
1997 ANOVANS 139(011) 1.49(0.13) 1.68(0.11) 1.19(o.13)
ANCOVA 1.37(0.10)a 1.49(0.11)ab 1.68(0.10)b l.19(0.11)a
2000 ANOVA” 1.68(0.09) 1.63(0.08) 1.85(0.08) 1.95(0.09)
ANCOVA 1.68(0.09)a 1.63(0.08)a 1.85(0.08)ab 1.95(0.09)b
2003 ANOVA 1.52(O.15)ab 1.91(O.15)b 1.36(0.15)a 1.22(0.15)a
ANCOVA 1.52(O.15)ab 1.91(0.15)b 1.36(O.15)a l.22(0.15)a
Wheat
1998 ANOVA 3.08(0.21)c 1.98(0.22)b 1.51(O.22)b 0.73(0.21)a
ANCOVA 3.08(0.19)c 1.99(O.21)b 1.51(0.21)b 0.73(0.19)a
2001 ANOVA 3.63(0.15)c 3.18(0.15)b 3.08(0.15)b 2.27(0.15)a
ANCOVA 3.63(0.14)c 3.18(0.14)b 3.08(O.14)b 2.26(0.l4)a
2004 ANOVA 4.55(0.09)c 4.59(0.09)c 4.22(0.09)b 2.44(0.09)a
ANCOVA 4.55(0.09)c 4.59(0.09)c 4.22(0.09)b 2.44(0.09)a
‘ws

 

indicates no significant difference between treatments at p < 0.05 level.
* means within the same row followed by the same letter are not significant different (p < 0.05).
§represents treatment mean adjusted at average value of WI (7.0).

Regression Slopes of Average Yield versus WI

Because of significant interaction between treatment and W1, unequal slope
regression lines were plotted for corn yields versus WI among treatments in 4 years (Fig.
4.13). Generally, yields were positively associated with WI. But the regression slope for
each treatment varied from year to year as a firnction of precipitation. In 1996 and 1999,
summer precipitation values were less than normal (Fig. 4.3). The regression slope of

yield versus WI for the T1 was significantly steeper than for T2. In 2005, the summer

lll

precipitation was close to normal. The T1 had steeper slope than T2 as well. In 2002,
however, summer precipitation was more than normal. The regression slopes for the T1
and T2 were not significant different from zero. The results suggested that yields in wet
year less depend on WI. But in dry and normal rainfall years, for example in 1996, 1999
and 2005, corn yields in conventional tillage had greater response to WI values compared

with no-till.

Unlike the T1 and T2, the regression slopes for the T3 and T4 were significantly
positive in wetter year of 2000. The results indicated that the yields in low input and
organic systems increased with increase with WI values regardless of dry or wet weather.
Although corn yields for high input system did not response on WI in wet year, it was
positively related with WI for low and zero input system. This was probably due to that
when the soil water is not a limiting factor in wet year, but nutrient may become major
limiting factor on yield in low and zero input systems. The areas with higher WI values
were located at depression areas coincident with flat slope and had relative abundant
nutrient. Because of high chemical input in the T1 and T2, the influence of nutrient status

due to topography was minimal compared with low or zero input in wet year.

Similar results were observed in soybean (Appendix 4.12). In 1997, the
precipitation in summer was less than normal. The regression slope for the T1 was
steeper than T2. But 2000 and 2003, which both were close to normal, the regression
slopes for the T1 was still higher than T2, but the magnitude of the difference between T1
and T2 became minimal. The regression slope for the wheat yield in the T1 showed

significant steeper slope than for the T2 in 1998 (Appendix 4.13). They were similar in

112

2001 and 2004, which were attributed as wetter years. It was noted that the relationship

between yield and W1 for T4 were more variable across years than expected.

 

 

 

  

 

 

 

 

 

 

 

 

 

 

 

 

T I
7
T1——-—T2
61- nan—T3—fl'T4 4".
e
8
1—1 5"
an,
as.
>1 2 b—
E .1.- ——‘
s 3 c—0-0——-‘—
U a
2.1..
1 i A i l l
3 5 7 9 ll 13
Wetnesslndex l
11
7
T1—-——T2 '
6-- .x'
ON can—TB—aaT4 /
as
es
0‘
:17
e- a
2.:
.2 “3
>4 2
:1
l-
O
U
1 . : . 1 . : . : . 1
3 5 7 9 ll 13
Wetnesslndex

 

 

 

Fig. 4.13. Plot of estimated corn yield regression models for the 4 treatments at
different WI values. I is for 1996, II is for 1999. Alphabetic letters at left of a line
represent the value of regression slope from low to high. The same letters indicate

no significant slope difference between treatments (p < 0.05).

113

 

 

 

 

 

 

 

 

 

 

 

 

  

 

 

 

III
10
91 T1—-—-T2 fl...
8‘_"——T3——aT4 ‘g‘..‘ .".
N .-e r
s _______,_._.—‘__ ,-
N 7.”: ..G' " b~——
:- 4‘
....' _ _ r
3.: 61" ,z‘
.2 en 5__ ’_’
>2 c
E 4--
e
U 3.,
2.1-
1 . ‘r . 5 . 3 . f . i
3 5 7 9 ll 13
Wetness Index
‘ IV
13
Tl——--T2
.- "
11.1 non—T3—aaT4 4"“
In
8
N 9"a .—
:7
.... a b ,
13.:
E w 7‘L b ".’
>2 ."’
E 5"b N”,
O ’
U
3...
1 . t . 3 . i . 3 . 5
3 5 7 9 ll 13
Wetnesslndex

 

 

 

Fig. 4.13 cont’d. Plot of estimated corn yield regression models for the 4 treatments
at different WI values. III is for 2002, IV is for 2005. Alphabetic letters at left of a
line represent the value of regression slope from low to high. The same letters

indicate no significant slope difference between treatments (p < 0.05).

114

Average Yield Difference Influenced by Wetness Index

Similar to the comparison of the maximum yield in previous section, I compared
average yields between the T1 and T2, T1 and T3, T3 and T4 using ANCOVA model.
The yield difference between treatments was shown by constructing 95% confidence
band across the increment of WI level. A horizontal line was inserted to where the model
difference is zero. Unlike nonparametric regression, where the fitted line is a curve,
ANCOVA is combination of linear regression and ANOVA. The average yield and

confidence band of the difference corresponding to WI were linear lines.

Except in 2005 year, the corn yields between the T1 and T2 did not significantly
differ, although the T2 yields tended to be higher (Fig. 4.14). The 4-yr estimated yields
across entire WI values were 5.72 Mg ha‘1 and 6.39Mg ha”l for the conventional tillage
and no-till systems, respectively. In 2005, the T2 yields were significantly higher than T
at WI < 7.5. The regression slope for the yield difference versus WI was positive with
varied values across years. Noticeable, the slope was flat across whole WI values in 2002,
which was wetter year. WI is hydrologically based index used to characterize the likely
soil moisture regime (Moore et al., 1988). Higher summer rainfall could lead to higher
soil water content. As a result, WI did not affect corn yields in a wet year but in a dry or

normal year.

A number of yield comparisons between conventional tillage and no-till were
reported (Kapusta et al., 1996; Wilhelm and Wortmann. 2004; Hoeft, et al., 2005; Grandy
et al., 2006). But little information about the effects of topographic on the yield between

the tillage methods was mentioned. I found that no-till corn tended to produce higher

115

yields than conventional tillage, particular in lower Wl values. The areas with lower WI
were often characterized by summit, backslope and shoulder slope elements (Moorman et
al., 2004). No-till offered higher potential for reduced erosion and evaporation and
improved soil properties than conventional tillage in those areas as discussed earlier.
Therefore, it is possible to maximize yield and profit by farming site-specifically based
on landscape position. Through arranging no-till practice in the areas, such as hill,
shoulder, back slope, whereas arranging tillage operation on foot slope and depression

areas, growers might get higher yield and greater return.

No significant difference of soybean yields between the T1 and T2 was observed
in 2000 (Appendix 4.14). However, the T2 yields at lower WI values were significantly
higher than T1 in 1997 and 2003. Similar to corn, no-till system produced higher soybean
yields at lower WI values compared with conventional tillage. Wheat yield in the T1
were significantly higher than T2 in 1998 and 2001 and equaled those in the T2 in 2004
(Appendix 4.15). Clearly, the magnitude of soybean and wheat yield difference became
smaller from earlier to later years. The results indicated that no-till tended to produce

higher or similar yields to conventional tillage in a long-term run.

116

 

 

Tl-TZ in 1996

 

 

Mean Yield Difference
ha
a
l
i

 

 

 

Wetness Index

 

 

 

‘E

 

T1 -T2 in 1999

""" “1’13er
------- lower

 

—.-¢-"
.-.‘o’¢-O
- ‘.-."-'_-
' .
I'D-.-

 

V

Mean Yield Difference

 

 

 

We tness Index

 

 

 

Fig. 4.14. Plot of 95% confidence band of corn yield difference between T1 and T2,
where WI as covariate. I, II are for years 1996, 1999, respectively. Arrow-head line

indicates zero yield difference.

117

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Ill
3
' Tl-TZ in 2002
8 2] ----- upper
g l . """" lower
1!: ... ________________________
a I l — .............
2 i o ’1
.2 E
> d
5 -1 .
0
z .
-2 1 ......................................................
1
-3 4 4' i
3 5 7 9 ll 13
Wetness Index
IV
3
2 : —T1-T2in2005
g . ----- umr
2": 1 ----- lower
F'- —: ..........
5 3 ........
g a 0 ’.’_=q——-I—"" :
>1 5 . ————
S -1 -
9
2 ‘ .......
.2 ‘ ...................
-3 :-" .L . 4r T 5 I = ' :
3 5 7 9 11 13
Wetness Index

 

 

 

Fig. 4.14 cont’d. Plot of 95% confidence band of corn yield difference between T1
and T2, where WI as covariate. III, IV are for years 2002, 2005, respectively.

Arrow-head line indicates zero yield difference.

No significant corn yields differences between the T1 and the T3 were found in
1996, 2005, although the yields in the T3 were slightly higher than in the T] (Fig. 4.15).

The T3 yields were significantly higher than the T1 in 1999 at lower WI values. But they

118

were opposite in 2002. In 2002, summer rainfall was more than normal. Possibly, the
areas with higher WI values in low input system had better nutrient supply which lead to
considerable higher yields compared with conventional system in wet year. No
significant difference of soybean yields between the T1 and T3 was observed in 2000 and
2003 (Appendix IV-15). The 3-yr average yields for the T1 were fairly similar to those in
the T3. However, low input system produced higher soybean yields at lower WI values
compared with conventional input system. Wheat yields in the T1 were consistently
higher than in T3 across 3 years (Appendix 4.17). The yield difference between the T1
and T3 became smaller from 1998, 2001 to 2004. As previously discussed, there may be
nitrogen shortages that may reduce crop yields temporarily, but these can be eliminated
by raising the soil nitrogen level and improving soil productivity through the use of
legume cropping systems (Kuo et al., 1997; Mader, et al., 2002; Pimentel et al., 2005;

Villamil etal., 2006).

119

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I
4
T1-T31n1996
3 -
w . ----- umer
0
§ 2 ----- lower
e . .-
S '2 1 i _ ...............................
g g 0 #_ :
s:
-~ -
_2 . --------------------------------
'3 I I 1 i l 1
3 5 7 9 ll 13
Wetness Index
II
3
Tl-T3 in 1999
3 2 d """ “PP"
g l - ----- lower
a: —1
E 33 ‘
2 :0 .................... :
.2 2 —
>u
5 -1 ~
0
2 ‘ ______________________________________________________
-2 ‘
'3 i ; t 1 , t j.
3 5 7 9 ll 13
Wetness Index

 

 

 

Fig. 4.15. Plot of 95% confidence band of corn yield difference between T1 and T3,
where WI as covariate. I and II are for years 1996, 1999, respectively. Arrow-head

line indicates zero yield difference.

120

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

III 1.
2
E
E
8::
'5
'2
.2
>1
g .
a
o _3_ . -
2 ‘ Tl-T31n2002 ~~~~~~~
-4- ————— “my """"
‘ ------ lower
-5 . . f 1 : . + =
5 7 9 11 13
Wetness Index
IV
3
1 Tl-T3in2005
2' ----- umer
E ‘ """"" lower
2 1.
6'5... - _______________________________________
'5' ‘
25,0 .
0
5:2 .
z 7 77”” '"""°"'"'"'""""""---"':::::::::....
-2.
-3 J = r : 1 ‘
3 5 7 9 11 13
Wetness Index
1

 

 

Fig. 4.15 cont’d. Plot of 95% confidence band of corn yield difference between T1

and T3, where WI as covariate. III and IV are for years 2002, 2005, respectively.

Arrow-head line indicates zero yield difference.

The T3 yields of corn were significantly higher than T4 except for 1999 (Fig.

4.16). The regression slopes of yield difference were positive in 1996, negative in 1999

and 2002, close to zero in 2005, respectively. No consistent effect of WI on corn yields

121

difference between T3 and T4 was found. Soybean yields in the T3 were significant
higher than in T4 in 1997 only (Appendix 4.18). Wheat yields in the T3 were consistently
higher than in T4 in 3-yr period (Appendix 4.19). Organic system showed yield
disadvantage over low input system in corn and wheat, but not for soybean. European
field data demonstrated that organic wheat and other cereal grain yields average from
30% to 50% lower than conventional cereal grain production (Mader, et al., 2002). The
lower yields for the organic system compared with the conventional systems appear to be

caused by lower nitrogen-nutrient inputs in the organic systems.

122

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I
4
3 .
3
5 2 - l
a l
8:
D
'5; g 0
E
o -1 <
E
.2 -‘
-3 ' ' : f . r 4
3 5 7 9 ll 13
Wetness Index
II
3
7 T3-T4in 1999
.. 2 7 ----- upper
o .
5 ~ ----------- lower
3 1 ‘ “ ~~~~~~
E" -.1 . "“ ~~~~~~~
fl \ ‘‘‘‘‘‘‘‘
'5 1:: "~- L
E £0 7
>1 ______________
5 -1 - -------------
0 --------
E 1 ..............
-2 f 7‘
.3 l I : l 'I
3 5 7 9 ll 13
Wetness Index

 

Fig. 4.16. Plot of 95% confidence band of corn yield difference between T3 and T4,
where WI as covariate. I and II are for years 1996, 1999, respectively. Arrow-head

line indicates zero yield difference.

123

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

III
3
1 ---------------------------
2 1 -
8 l g
5
1.. l ‘
N
E 44
% an 0 a --------------------- T
s: E """""""""
fl
3 -l ‘ T3-T4 in 2002
I E ----- unrer
'2 “ ------- lower
.3 . t . 4 - 5 ' 4' ‘ =
l 3 5 7 9 11 13
* Wetness Index
FIV
7
6 ‘ 'I‘3-T4 in 2005
l ----- umer
g 5 . ------- lower
é . _______________________________________
. '7 4 -
S 2 .
7; DD
'5; E 3 1
g .
a
E) 2 .4 ........................................................
z .-
0 . : . f
. 3 5 7 9 “ 13
L Wetness Index

 

 

Fig. 4.16 cont’d. Plot of 95% confidence band of corn yield difference between T3
and T4, where WI as covariate. III and IV are for years 2002, 2005, respectively.

Arrow-head line indicates zero yield difference.

124

CONCLUSION

Boundary line analysis is a useful diagnostic tool to identify yield potential and
estimated the effects of topography on the maximum yield. The nonparametric spline
regression algorithm utilized in the study was robust and efficient in describing the
complexly shaped boundary line and comparing the yield differences between treatments
across a range of WI values. The comparison of 2 treatments can be easily understood

through visually plotted confidence bands for the difference.

The yield differences between the management practices varied from year to year
as a function of the prevailing precipitation and depending on the topographical position.
Management practices significantly interacted with topographical features as represented
by WI and influenced crop yields. Further the interaction can be affected by summer
precipitation. The yield difference between no-till and conventional tillage was strongly

influenced by WI in dry or normal year but not in wet year.

The relationship between boundary line yield data and W1 in most of the crops for
most of the studied years had a convex shape. Higher boundary line yields, that are
maximum yields, consistently occurred at intermediate WI levels. No-till and low input
systems tended to produce higher maximum and average yields than conventional system
at the lowest WI where were characterized as eroded summit and shoulder areas. It is
possible to maximize yield and profit by farming landscape- or site-specifically based on
landscape position. Through arranging no-till or low input practice in the areas, such as
hill, shoulder, back slope, whereas arranging tillage operation on foot slope and

depression areas, growers might obtain higher yield and return. The magnitude of crop

125

yield difference between conventional and low input system became smaller from earlier
to later year. Nitrogen shortages may reduce crop yields temporarily, but low input
system tended to produce similar or higher yields to conventional tillage in a long-term

run.

ACKNOWLEDGMENTS

I thank Li Wang for valuable help on nonparametric spline regression. I sincerely
thank Dr. Philip Robertson, Andrew Corbin and Joe Simmons for their helpful suggestion
and discussion on the data analysis and Suzanne Sippel for preparing yield data from

Kellogg Biological Station.

126

REFREENCES

Andresen, J .A., G. Alagarswamy, C.A. Rotz, J .T. Ritchie, and A.W. LeBaron. 2001.
Weather Impacts on Maize, Soybean, and Alfalfa Production in the Great Lakes
Region, 1895-1996. Agron. J. 93:1059-1070.

Blackmore, S. 2000. The interpretation of trends from multiple yield maps. Computer and
Electrics Agri. 26:37-51.

de Boor, C. 2001. A Practical Guide to Splines. Springer-Verlag, New York.

Chambers, J .L., T.M. Hinckley, G. Cox, C.L. Metcalf, R.G. Aslin. 1985. Boundary-Line
Analysis and Models of Leaf Conductance for Four Oak-Hickory Forest Species.
Forest Sci. 31(2):437-450.

Cox, W.J., R.W. Zobel, H.M. van Es, and DJ. Otis. 1990. Tillage effects on some soil
physical and corn physiological characteristics. Agron. J. 82:806-812.

Crum, J .R., and HP. Collins. 1995. KBS soils. Available at
www.1ter.kbs.msu.edu/soil/characterization. W.K.Kellogg Biol. Stn. Long-Tenn
Ecol. Res. Project, Michigan State Univ., Hickory Comers, MI.

Doerge, T. 1999. Yield map interpretation. J. Prod. Agric. 12:54-61.

Drummond, S., and K. Sudduth. 2003. From sensors to information: Cleaning yield
monitor data. Proc. InfoAg 2003 Conf., Indianapolis, IN. 29 July—1 Aug. 2003.
Potash and Phosphate Inst., Norcross, GA.

ESRI. 2000. ArcView Spatial Analyst, Redlands, CA.

Fan, J. and I. Gijbels. 1996. Local Polynomial Modelling and Its Applications. Chapman
and Hall, London.

F ortin, M.C., and F.J. Pierce. 1991. Timing and nature of mulch retardation of corn
vegetative development. Agron. J. 83:258-263.

Grandy, A.S., T.D. Loecke, S. Parr, and GP. Robertson. 2006. Long-Tenn Trends in
Nitrous Oxide Emissions, Soil Nitrogen, and Crop Yields of Till and No-Till
Cropping Systems. J. Environ. Qual. 35: 1487—1495.

Halvorson, G.A., and EC. Doll. 1991. Topographic effects on spring wheat yields and
water use. Soil Sci. Soc. Am. J. 55:1680-1685.

Hao, X. and AN. Kravchenko. 2007. Management practice effects on surface total
carbon: Differences along a textural gradient. Agron. J. 99: 1 8-26.

127

 

Hiirdle, W. 1990. Applied Nonparametric Regression. Cambridge University Press,
Cambridge.

Hoeft, R. G., E.D. Nafziger, R.R. Johnson, and SR. Aldrich. 2000. Modern corn and
soybean production. MCSP Publications. Savoy, IL.

Hu, Q., and G. Buyanovsky. 2003. Climate Effects on Corn Yield in Missouri. J. Appl.
Meteor. 42:1626-1635.

Huang, X., A. Kravchenko, K. Thelen, N. Martin, and G. Bollero. 2005. Yield variability
in an undulating field: classification and prediction. Proceedings of the 7th

International Conference on Precision Agriculture and Other Precision Resources
Management. July 25-28, 2004. Minneapolis, MN. pp 419-431.

J aynes, DB, and TS. Colvin. 1997. Spatiotemporal variability of com and soybean
yield. Agron. J. 89:30-37.

J aynes, D.B., T.C. Kaspar, T.S. Colvin, and DE. James. 2003. Cluster analysis of
spatiotemporal corn yield patterns in an Iowa field. Agron. J. 95:574-5 86.

J iang, P., and K.D. Thelen. 2004. Effect of soil and topographic properties on crop yield
in a North-Central com—soybean cropping system. Agron. J. 96:252-258.

Kabir, Z. and RT Koide. 2000. The effect of dandelion or a cover crop on mycorrhiza
inoculum potential, soil aggregation, and yield of maize. Agric. Ecosyst. Environ.
78:167-174.

Kapusta, G., R.F. Krausz, and J .L. Matthews. 1996. Corn yield is equal in conventional,
reduced, and no tillage after 20 years. Agron. J. 88: 812-817.

Kaspar, T.C., T.S. Colvin, D.B. J aynes, D.L. Karlen, D.E. James, D.W. Meek, D. Pulido,
and H. Butler. 2003. Relationship between six years of corn yields and terrain
attributes. Precision Agriculture 4:179-192.

Kaspar, T.C., T.B. Parkin, D.B. J aynes, C.A. Cambardella, D.W. Meek, and Y.S. Jung.
2006. Examining Changes in Soil Organic Carbon with Oat and Rye Cover Crops:
Using Terrain Covariates. Soil Sci. Soc. Am. J. 70:1168—1177.

Kitchen, N.R., S.T. Drummond, E.D. Lund, K.A. Sudduth, and G.W. Buchleiter. 2003.
Soil electrical conductivity and topography related to yield for three contrasting
soil-crop systems. Agron. J. 95:483-495.

Kravchenko, A.N., and D.G. Bullock. 2000. Correlation of Corn and Soybean Grain
Yield with Topography and Soil Properties. Agron. J. 92:75-83.

128

Kravchenko, A.N., G.P. Robertson, X.Hao, and D.G. Bullock. 2006. Management
practice effects on surface total carbon: Differences in spatial variability'patterns.
Agron. J. 98: 1559-1568.

Kravchenko, A.N., G. P. Robertson, K.D. Thelen, and RR. Harwood. 2005.
Management, topographical, and weather effects on spatial variability of crop grain
yields. Agron. J. 97: 514-523.

Kravchenko, A.G., and K.D. Thelen. 2007. Effect of winter wheat crop residue on no-till
corn grth and development. Agron. J. 99:549-555.

Kravchenko, A.N., T.M. Harrigan, and BB. Bailey. 2005. Soil Electrical Conductivity as
a covariate to improve the efficiency of field experiments. Transactions of the
ASAE. 48(4):1353-1357.

Kuo, S, U.M. Sainju, and BI. J ellum. 1997. Winter cover crop effects on soil organic
carbon and carbohydrate in soil. Soil Sci. Soc. Am. J. 61: 145-152.

Lamb, J .A., R.H. Dowby, J .L. Anderson, and G.W. Rehm. 1997. Spatial and temporal
stability of corn grain yields. J. Prod. Agric. 10:410-414.

Lark, RM. 1997. An empirical method for describing the joint effects of environmental
and other variables on crop yield. Annals of Applied Biology 1312141-159.

Lark, R.M., and J .V. Stafford. 1997. Classification as a first step in the interpretation of
temporal and spatial variation of crop yield. Ann. Appl. Biol. 130:111-121.

Liu, A., B.L. Ma, and A.A. Bomke. 2005. Effects of cover crops on soil aggregate
stability, total organic carbon, and polysaccharides. Soil Sci. Soc. Am. J. 6922041-
2048.

Littell, R.C., G.A. Milliken, W.W. Stroup, R.D. Wolfinger, and O. Schabenberger. 2006.
SAS for Mixed Models, 2nd Edition. SAS Institute Inc., Cary, NC.

Lobell, DB. and GP. Asner. 2003. Climate and management contributions to recent
trends in US. agricultural yields. Science 29921032.

Mader, P., A. Fliebbach, D. Dubois, L. Gunst, P. Fried, and U. Niggli. 2002. Soil Fertility
and Biodiversity in Organic Farming. Science 296:1694-1697.

Milliken, G.A., and DE. Johnson. 2002. Analysis of Messy Data, Vol. 3: Analysis of
Covariance. Chapman & Hall/CRC, New York.

Mokma, D.L., and J .A. Doolittle. 1993. Mapping soils and soil properties in southwest
Michigan using ground-penetrating radar. Soil Survey Horizons 34: 13-22.

129

Moorman, T.B., C.A. Cambardella, D.E. James, D.L. Karlen, and LA. Kramer. 2004.
Quantification of tillage and landscape effects on soil carbon in small Iowa
watersheds. Soil & Tillage Research 782225—236.

Moore, I.D., P.E. Gessler, G.A. Nielsen, and GA. Peterson. 1993. Soil attributes
prediction using terrain Analysis. Soil Sci. Soc. Am. J. 57:443-452.

Perez-Quezada, J .F ., G.S. Pettygrove, and RE. Plant. 2003. Spatial—temporal analysis of
yield and soil factors in two four-crop—rotation fields in the Sacramento Valley,
California. Agron. J. 95:676-687.

Pimentel, D. P. Hepperly, J. Hanson, D. Douds, and R. Seidel. 2005. Environmental,
Energetic, and Economic Comparisons of Organic and Conventional Farming
Systems. BioScience 55:573-582.

Reicosky D.C., W.D. Kemper, G.W. Langdale, C.L. Douglas, and PE. Rasmussen. 1995.
Soil organic-matter changes resulting from tillage and biomass production. J. Soil
Water Cons. 50: 253-261.

Rhoton, RE. 2000. Influence of time on soil response to no-till practices. Soil Sci. Soc.
Am. J. 64: 700-709.

Robertson, G. P., J .C. Broome, E.A. Chomesky, J .R. Frankenberger, P. Johnson, M.
Lipson, J .A. Miranowski, E.D. Owens, D. Pimentel, and LA. Thrupp. 2004.
Rethinking the Vision for Environmental Research in US Agriculture. BioScience
54:61-65.

Sainju, U.M., W.F. Whitehead, and B.P. Singh. 2003. Cover crops and nitrogen
fertilization effects on soil aggregation and carbon and nitrogen pools. Can. J. Soil
Sci. 83:155-165.

SAS Institute Inc. 1999. SAS/STAT User's Guide, Version 8. Cary, NC.

Schmidt, F., and A. Persson. 2003. Comparison of DEM data capture and topographic
wetness indices. Precision Agriculture 4: 179-192.

Schnug, E., J. Heym, and F. Achwan. 1996. Establishing critical values for soil and plant
analysis by means of the Boundary Line Development System (Bolides).
Communications in Soil Science and Plant Analysis 27:2739-2748.

Shatar, T.M., and AB. McBratney. 2004. Boundary-line analysis of field-scale yield
response to soil properties. J. Agricultural Science 142:553-560.

Shuai X., Z. Zhou, and RS. Yost. 2003. Using Segmented Regression Models to Fit Soil
Nutrient and Soybean Grain Yield Changes due to Liming. J Agricultural,
Biological, and Environmental Statistics, 8(2):240—252.

130

Si, BC, and RE. Farrell. 2004. Scale-dependent relationship between wheat yield and
topographic indices: A wavelet approach. Soil Sci. Soc. Am. J. 682577-587.

Simmons, F .W., D.K. Cassel, and RB. Daniels. 1989. Landscape and soil property
effects on corn grain yield response to tillage. Soil Sci. Soc. Am. J. 53:534-539.

Timlin, D.J., Y. Pachepsky, V.A. Snyder, and RB. Bryant. 1998. Spatial and temporal
variability of corn grain yield on a hillslope. Soil Sci. Soc. Am. J. 622764-773.

Tomer, M.D., C.A. Cambardella, D.E. James, and TB. Moorman. 2006. Surface-Soil
Properties and Water Contents across Two Watersheds with Contrasting Tillage
Histories. Soil Sci. Soc. Am. J. 70:620—630.

VandenBygaart, A.J., R. Protz, A.D.Tomlin, and J .J . Miller. 1999. Tillage system effects
on near-surface soil morphology: observations from the landscape to micro-scale in
silt loam soils of southwestern Ontario. Soil Till. Res. 51:137-147.

Villamil, M.B., G.A. Bollero, R.G. Darmody, F.W. Simmons, and D.G. Bullock. 2006.
No-Till Corn/ Soybean Systems Including Winter Cover Crops: Effects on Soil
Properties. Soil Sci. Soc. Am. J. 7021936—1944.

Wang, L. and J. Yang. 2006. Simultaneous spline confidence bands for time series
prediction function. Submitted.

Webb, RA. 1972. Use of the Boundary Line in the analysis of biological data. J.
Horticultural Science 472309-319.

Wilhelm, W.W., and CS. Wortman. 2004. Tillage and rotation interactions for corn and

soybean grain yield as affected by precipitation and air temperature. Agron. J.
96:425-432.

131

CHAPTER V

CONCLUSION

Highly variable soil and topographical attributes in glacial till soils cause
difficulties in mapping soil carbon based only on a limited number of soil samples.
However, remote sense technology and statistical methods combined with inexpensive
elevation data can estimate soil carbon with less cost and reasonable accuracy. Field on-
the-go NIRS and Landsat ETM imagery determined that it is feasible to map soil carbon
of agricultural fields on glacial till soils rapidly and cost-effectively. Regression
coefficients between measured and predicted carbon values were equal to 0.70 and 0.46
using NIRS data and ETM imagery. When topographical attributes were included into
regression model along with NIRS and ETM spectra reflectance values, the regression
coefficients improved to 0.81 and 0.62. ETM imagery is useful in a large scale,
particularly for assessing soil carbon in a bare field and combined with topographical

attributes.

Both electrical conductivity and topographical attributes were found to be helpful
in explaining yield variability. Areas with lower electrical conductivity and slope tended
to form high yield clusters in the studied fields. Historical yield classification may be

used to delineate management zones within a field and predict the yield patterns.

Crop yields were correlated negatively with spring and positively with summer
daily precipitation. Roughly 60 % of the variability of the yield could be explained by the

growing season precipitation. The yields for the high chemical input system were largely

132

influenced by precipitation in drier years and less influenced in normal or wetter years as

compared with low or zero input systems.

The yield differences between the management practices varied from year to year
as a function of the prevailing precipitation and depending on the topographical position.
Management practices significantly interacted with topographical attributes as
represented by WI. The yield difference between no-till and conventional tillage was

strongly influenced by WI in drier or normal years but not in wetter years.

The relationship between boundary line yield data and W1 in most of the crops for
most of the studied years had a convex shape. Higher boundary line yields, that are
maximum yields, consistently occurred at intermediate WI levels. No-till and low input
systems tended to produce higher maximum and average yields than conventional system
at the lowest WI. It is possible to maximize yield and profit by farming site-specifically

based on landscape position.

133

APPENDICES

Appendix 4.1. Boundary lines fitted as 2-linear-line regression results for the corn
yields in 4 year from 4 treatments.

 

Corn in 1996

 

Yield in joint point,
Mg ha"

N

Joint point

Slape 1

Slope 2

R2

Yield in joint point
Mgha‘

N

Joint point

Slope 1

Slope 2

R2

T1

9.97
163

7239

1.71bc

-0.62a
0.35

T1

10.78
163

4.9%
5.10d

-0.39ab

0.43

T2 T3
7.90 7.91
164 161
6.54b 6.25a
1.63a 1.66ab
-0.44b -0.38c
0.37 0.55
Corn in 2002

T2 T3
11.05 11.87
171 167
5.23b 5.72c
2.53c 1.70b
-0.49a -0.34b
0.61 0.37

T4

5.94
140
6.26a
1.78c
-O.45b
0.42

T4

11.80
143

6.87d
1.44a

-0.36b

0.33

Corn in 1999
T1 T2 T3 T4
7.63 5.86 6.03 7.35
163 164 156 142
7.15c 5.80a 6.26b 7.50d
1.59c 0.73a 0.94b 0.77a
-0.39a -0.16c -0.200 -0.26b
0.36 0.17 0.39 0.20
Corn in 2005
T1 T2 T3 T4
14.19 13.17 12.66 11.14
163 166 154 143
5.65c 5.08b 4.98a 7.29d
4.04c 1.87b 4.79d 1.26a
-0.24b -0.20b -0.l7b -0.52a
0.46 0.24 0.14 0.23

 

N: sample size; §

estimated parameters within a row followed by the same letter are not

significantly different between treatments at the 0.01 level by the 2-side t-test.

134

Frequency

Frequency

Histogram of WI in T1

 

 

 

 

 

 

 

 

D _
L‘l _ _.
8 .. 7 ‘
8 _
l l l I
6 8 10 12
wetnesindex
Histogram oleinTa
— r—
S — _. I—I
8 .. l—
.1
V
D _.

 

 

 

 

 

 

 

 

I l I 7 I
4 6 8 10
wetness index

Frequency

Frequency

Histogram of WI in T2

 

 

 

 

 

 

 

 

'1
a-
W 4
a- H
BIT
o-
F F I l
4 5 8 10
wetnessindex
Histogram olein T4
0
87 w
@_ _ -—
g-
B-
D-

 

 

 

 

 

 

 

I I I I
8 8 10 12

wetness index

Appendix 4.2. Histograms of elevation in 4 treatments

135

yield of T1 in 2N2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

e- I
3 ~ —
c 8—
o ‘—
3
3
I: Sq Hm?“
..I _
l I I I
6 8 10 12
Yield,Mg/ha
yieldof‘l‘31n2002
%_ _'—1
‘2 8— 7 7
a v—
3
0
7 SJ i013
DJ _
I I I I
6 8 10 12
Yield,Mg/ha

150
l

100
I

Frequency

50
I

 

150
m

11]]
I

Frequency

50
I

 

yield of T2 in 2002

_

 

 

 

 

 

 

 

I
6

I I T
8 10 12

Yield, Mglha

yield of T4 In 2002

fl

F

 

 

 

 

 

 

 

rrT
81012

Yield, Mglha

Appendix 4.3. Histograms of corn yield in 4 treatments in 2002

136

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Q—
‘ _T1
Soybean 1997 - ‘ 72
In
«5 ..~ T3
- T4
q_
E M
a
2 In-
In. N
3
'>'-' °._
N
In
‘_._
O
€_fi
Q.
at _T1
---I-2
.Q _ Soybean 2000 .
" T3
' T4
1‘
z —1
Ta
2
n In-‘
7% N ..-"".--..,'I'-"
; ~-—- ,----‘--‘:.-.-..,,,, .---~-
o— M
s‘ I.
m \
‘_-_
O
._'_
. I I I l |
Q—
Q
IQ—
M
N
5 o_
a n'
E
'0‘ '0.—
3 N
>' °.
«—
ID
‘_._
O
‘_-_.

 

 

 

wetness index

Appendix 4.4. Boundary line of soybean yield vs. WI fitted by spline regression.

137

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

O _.

' . _T1
°° Wheatrn1998 _ _ T2
°. _

N m
5
on
2
- o
1: - _
E v
>.
O
oi _

O _.

co , _T1
Wheatm 2001 _ _ T2
900 T3
:3 _ "'T4
«1 co'
5
a
E
" 0
e - —
.>__

O

oi _

q 7 T1

a . —

Wheat In 2004 _ _ T2
.0. T3
0 _.
2 d
‘o:
E
1:"
E 3 —
>.
O
N _1
I l T l I
4 6 8 10 12

wetness index

Appendix 4.5. Boundary line of wheat yield vs. WI fitted by spline regression.

138

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

N _ .
g T1—T2m1997
\
a
2 v- - -

a ” ‘-’¢-\‘ ’1
§ ’§“'-:.-.’.. 00000 O ..........
o ‘ “'.'—‘='-, _ i‘ e- ._

2 " I, .. ‘
0 0,’ ‘
iE , 1

e ‘7
E
>- 0.! —
I I fi I I I I
O
P-
g T1—T2in2000
a “‘24
2 O
I
8 ~ .’
c O b‘-~ -‘ _n"..’ A
g G . ...... ..................... f
11.: ------ a
o In "
2 =51
0
>.
°.
F
' l I T l l l l
M
N
S 9"
E , .
8': w- Tl—T21n2003 .
I
C I
§°
E , ---’ .....
o .7_. ,I ......
2 . ---”:r~;;z“ , - ’.«-‘
,9 .. '-;'.----’ w -- 1
>- 0,:— ~
r l T l l l l
5 6 7 8 9 10 11

wetness index

Appendix 4.6. Plot of 95% confidence band of the difference of soybean maximum
yield between T1 and T2 as a function of WI.

139

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

o —1
W
.C
3‘ l-‘ ’\ I
_I ’ -~‘.- ‘7‘ “I
2. 1‘s---"--’ ' 7 7
3 , ......... '..—~--' ‘~. '
CN q’. ’—-—‘ ” 5-1.-
2 '," J ‘\
O I \
E r .1
DO
2 .
.9 T1—T21n1998
>6} _
I T I I I I I
(B ”T
5 .
OI Tl-T2m2001
E N—
on I
8 a \ I
’I \‘ ’I
5 F-l "\ 0“ «’7'- ' '
s -* -.. -- .~ -o.
g .......... . - ...... - ,
D OT I“- If 7‘, \
In ,-"I “a \
.13 ._ g
>- ' ,
0']-
I I I I I I I
V-u
N
.C
B n- .
E Tl-T2m2004
3 N-
c a
e , ~
_ I ‘-. I
g F I \ d- p’
o \ ’I - -"_-_’ In’
a O 1 .r . 'o.A AA‘,.aaa'a‘ssss
E as ..s"' Iss’u- ‘u’ ~~
> ‘r—nii“"’ ‘\ I’d \
“.14
I I I I I I I
5 6 7 8 9 10 11

wetness index

Appendix 4.76. Plot of 95% confidence band of the difference of wheat maximum
yield between T1 and T2 as a function of WI.

140

 

 

 

 

 

 

 

 

 

 

 

 

 

 

N—
N
5 I
\
O ,I‘-_-"\‘ I
E F- ’-" 0.0. -"
8. ’-~:'.".:"-‘~"--slu.‘.. .
c ‘1 0” +
§° 1 ~
0” ‘
E7—/’
3; Tl-T3in1997
; 6.1-
I I I I I I I
O
F.—
G
.C
\
mug-
E o
a I
8 I
c O -’
2 o. .0“ ’p‘~ ’0'... g
g K“f—J-’ -’.o.’o“ .0
I- ..... '....’ ...... \.. ("‘\
a “2d .. ’ _ I “r
g O. k’ \" "
; T1—T3m2000
O
Flu—II
I I I I I I I I
M
2
N—
‘0:
E
.. 4
8 F_ ,__..-"
C __,-" ....... 0'4
0 b---- d ’’’’’ ’ .0...—d’--b
h ........ “3900' ' .p-.. I
g °-’_—. -_ '_._*— f ‘
E
E ‘7—
; T1—T3in2003
N-
I
I I I I I I T
5 6 7 8 9 10 11

wetness index

Appendix 4.8. Plot of 95% confidence band of the difference of soybean maximum
yield between T1 and T3 as a function of WI.

141

 

 

 

 

 

 

 

 

 

 

@—
g r-I. -."".s-“--’ “-‘u}
a st" . ~--"'._'.. """ I """""
E ’0 . ...... ”—‘--P- -'- ...I
-+ ,’ \
5 N P, I
3 r
3:
D o
2
.2
>' N— .
' T1—T3ml998
I I I fi I I I
n—
G
.C
\
s ~—
“ a ’l‘s‘ -’-s_—’.“ I
I ‘--~ " --"
g F_” '... cocoa-000‘... . ..
0 I.... -‘u I"-‘.g— ....... ..
h ’0 ’ s ‘
é: ,_ .. c
'_ 0_
o I
2
.2 ‘7‘
>.
N T1—T3in2001
I
I I I I I I I
Q
a ”—
u:
3) ,\
E N— I ‘
g I s l“_”’
F— I O . as- I
g ‘N I”..' .‘, -.-e ......°‘o
\ ’4 _. \
o ’-§.J’ -’ I
'5 r
_ P_ .
a ' T1—T3m2004
N—
I

 

 

 

_-
_
"'1
—
—
—q

—

5 6 7 8 9 10 11
wetness index

Appendix 4.9. Plot of 95% confidence band of the difference of wheat maximum
yield between T1 and T3 as a function of WI.

142

 

 

 

 

 

 

 

 

 

 

 

 

 

 

a N-
5 ‘ ”
a \ I \J
E P_‘\ -'-” OOOOOOOO
g 1-—_' -------- ‘l' “
§ ° ’
I
II:
a ‘7-
2
.2 T3—T4in1997
> N-—<
I
I I I I I I I
°._I
F
«I
5 In
a: -‘
2 ° .
- “"_s‘ 1"
I: c ...... L, ~ ‘H'w ' '
2 0 ,~-_ """ L""
I- ‘
o 24 I
'u - °
3 T3—T4m2000
>' o
Q.._I
' I I I I I I I
n
2 ~-
2
8' '—
: .-.T,'I:.".Z‘-'_'.'"’77o'-..‘.‘-. *
E " -s
D 1'- °
.2
>- QI—
I I l I I I I
5 6 7 3 9 1° 11

 

wetness index

Appendix 4.10. Plot of 95% confidence band of the difference of soybean maximum
yield between T3 and T4 as a function of WI.

143

 

<- 4 T3 -—T4 in 1998

 

Yield Difference, Mglha
2
1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

O -' "I
N
'1
I I T I I I I
N n—
-% T3—T4in2001
E N—
: _ ..... tr.-‘-‘-—:.’.. . .0 ..... -—
2 F .0 -------- ....O-'-- "-“--...
g ” \\" “ 5“
o— O
D
E
.2 ‘74
>.
0'.—
I I I I I I I
v+
g n _. T3—T4in2004
3 “~‘-“ n s I
E N O. .009 \~~ ” ““, -—"
8. ”—---‘.~.. ....... . . .0 ° 0 o
l ‘ ' ___'-""\
c \ ,’ ‘\,--
G F— I x
3 a
IE
2
.2 F'—
>.
”a
I I I I I I I
5 6 7 8 9 10 11

wetness index

Appendix 4.11. Plot of 95% confidence band of the difference of wheat maximum
yield between T3 and T4 as a function of WI.

144

 

 

   
 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

r I
2.5
C
b Tl—-—T2 b
a 20
2 ' anacTB—a—TLI
m- a
2 'fl -——'— .-
.2 a 15..
> w '
a 2 f a
0 —.
.9
g» 1.0”
m
0.5 v L . f . i . C . f
3 5 7 9 ll 13
Wetnesslndex
II
2.5
o
C
a ——‘ c
.5 2.0.. '__._.—-‘:‘,:':.—-- d
2.7“ —‘-—' .ao‘—‘.“
02 g 00“.“ -—_ b
>1 M
—-—-=T——_———
a 2 “~“— 3
o 1.5”
a T]———-T2
C
(A d
van—T3—I—‘CT4
1.0 - ~ - - ‘ P
3 5 7 9 11 I3
Wetnesslndex
[I]
2.5
8 I
a 2'04 ——————-—— a
.5 ‘ ~——‘
37:: #— b
£- €01.50 — c
a: .————..—————;:::_':E-'—-~'—'—"'.—a
:8 —O— ——‘-
g. 1.0 -- Tl—-'T2
VJ
pan—T3—vnT4
0:5 fl I 1 f ' I f f fit '
3 5 7 9 ll 13
Wetnesslndex I

 

 

Appendix 4.12. Plot of estimated soybean yield regression models for the 4
treatments at different WI. Alphabetic letters at right of a line represent the value of
regression slope from low to high. The same letters indicate no significant slope
difference between treatments (p < 0.05).

145

 

 

 

 

 

 

 

 

J \

 

 

 

 

 

 

 

 

 

 

 

I
4.5
Tl——T2 °
3 35
a ' Iran-T3—vnT4
.E...
'5'“ /
332-5 b
>" _-———
E2 _I—‘—-'—-— _.-.———— b
g 1.5 ___,_..........
a
0.5 "" - . A
5 7 9 ll 13
Wetnesslnth
II
4.5
b
—a __c
g 3.5 ’-——‘ .
N _‘— —-.---""—-
.E— “’m%'—-.
u'a
3"- 2.5 ,__.—- a
s: 50 ‘—.—‘—.—¢—
-2 —-—
3
g 1.5 Tl——T2
CCC—T3—FIT4
0.5 ' - - - .
5 7 9 ll 13
Wetnesslndex
III
5.5
_—_——————_—-'—b
V 4.5 ———f “— a
g lalug—auuupfl—III—aaulllil-fl a
N
E
';7a 3.5
3.:
.— u —,.—' c
> _i—‘—-
'52 2.5 _________..
; Tl——-T2
I l.5

0.5

I

Appendix 4.13. Plot of estimated wheat yield regression mod

 

 

 

 

n—n—T3-u—T4

L 4

 

 

5 7 9 11 13
Wetness Imhx

LA4,_

els for the 4 treatments

at different WI. Alphabetic letters at right of a line represent the value of regression
slope from low to high. The same letters indicate no significant slope difference
between treatments (p < 0.05).

146

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I 2
w I Tl-T2 in 1997
g 1 1 ----- “mt ......
g _________
'5
1:
-" r
>-'
I:
a
0
E
'2 Jr 1 I P f : t J
3 5 7 9 11 13
Wetness Index
11 2
I Tl-T2 in 2000
g 1 q ---- unrel-
In
a? _ ------ lower
5 a. — ....................................
E .: 0 4____:_——_|,
o ....................................................
s g .
S
0 -1 ..
E
-2 . ‘r . i . : . : 1 :
3 5 7 9 ll 13
Wetness Imbx
111 2
. T1-T2 in 2003
1. ----- um
é """" lower
, I-I
‘5 é ________________________
'5 w 0 ‘-— i 4.
'5: E
a ..............................
a ----------------------
i '1‘
-2 I ‘ t 1 1'
3 5 7 9 ll 13
Wetness Index

 

 

 

Appendix 4.14. Plot of 95% confidence band of soybean yield difference between T1

and T2, where WI as covariate. Arrow-head line indicates zero yield difference.

147

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I 3
E
0
3
8:
’5
g t
.2
>1
I:
a
0
E
.3 ' t 1 1 : I I
3 5 7 9 ll 13
Wetness Index
11 3
w 2
C)
1
g q-I l ‘ — -----------------------------------
'5 é ...................................................
g on 0 a
s: E ‘
g '1 ‘ 11.12 in 2001
2 _2 . ----- umer
------- lower
.3 Y = ‘ v i I _:_
3 5 7 9 ll 13
Wetness Index
111 3
3 27
fl .
2 1 _
9‘5 _.
a g «
=3 .. 0
s: E ‘
5 '1 ‘
E _2 4 ----- upper
. ------- lower
‘3 v I v 1r l I r I ‘f I
3 5 7 9 ll 13
Wetness Imbx

 

 

 

Appendix 4.15. Plot of 95% confidence band of wheat yield difference between T1

and T2, where WI as covariate. Arrow-head line indicates zero yield difference.

148

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1 2
_T1-T3 in 1997
2 1 d ----- upper
E: ------- 1
g _ ower ’__
5 g ______________
“a .3... _ ,,,,,
7.». g 0 ———— __=————-—— _______ .
> —" .........................
g ____________________
0 -14
E
'2 v i t 3 1 = ' f t ;
3 5 7 9 ll 13
Wetness Indtx
11 2
2
E l ‘ ______________
9g '7 - -----------------
a
2 go o -‘ 4.
0 0000000000000000000000000000000000000000000000
: Tl-T3 in2003
8 '1 ‘ ““““ “met
2
d ------ lower
'2 i 'r - 5 : . :
3 5 7 9 ll 13
Wetness Index J
111 2
E
E 1
F: .—
5 a. _________________
g a o ,
'5 E r ....................................................
5 Tl-T3in2000
i -1 9 ---- umr
'''''' lower
-2 r i . : . : , g , =
3 5 7 9 ll 13
Wetness Index

 

 

 

Appendix 4.16. Plot of 95% confidence band of soybean yield difference between T1

and T3, where WI as covariate. Arrow-head line indicates zero yield difference.

149

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I 3 _-
w 2 " __. ..................
O _/
g 1 I ...............................
g _ 1 . ....................
'5 _é ‘ _
a .. o 3
s: 2 -
5 '1 ‘ Tl-‘I‘3 in 1998
g 2 ‘ ----- upper
- . """"" lower
.3 i Y . V i r ‘ 1
3 5 7 9 ll 13
l Wetness Index
11 3
an 2 "'
U
:1 .
0
g 1 « _ ___________________________________
5 .é ‘ ................................................... ‘
e .. o ,
'5: z -
5 -l 1 Tl-T3 in 2001
i 2 d ---- u111m
. """" lower
'3 ' + . : . i 3 g , :
3 5 7 9 ll 13
Wetness Inchx
m 3
w 2 ‘
8 .
8 l q
9‘5 - _ ____________________________________
5 s * ____________________________________________________
a .. o ,
5: E
5 -14 _T1-13in2004
O
2 ‘ ---- “liner
.2 1 """"" lower
-3 . 5 . i . 5 . t , 4,
3 5 7 9 ll 13
Wetness Incbx

 

 

 

Appendix 4.17. Plot of 95% confidence band of wheat yield difference between T1

and T3, where WI as covariate. Arrow-head line indicates zero yield difference.

150

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I 2
4
3
§ 1‘
91-: _ 1
5 _.-.
% ED 0
5:.
g T3-T4 in 1997
g '1 ' ----- unler
------- lower
.2 ' 1 s j 1 1
3 5 7 9 ll 13
Wetness Incbx
II 2
. 1‘3-T4 in 2000
g l . ----- umer
0
g ------- lower
a g _ ____________________________________
E on 0 i =——-—#
5: E ....................................................
1:
3 1.
E -
-2 - : . 1 - 4 . : . t
3 5 7 9 ll 13
Wetness Index
III 2
3
5 11
3 - --------
t — -------------------------
E 3 — .-
i
e g. 0 ......................
“s . ..............................
g T3-T4 in 2003
i '1 ' ----- upper
------- lower
.2 f . i . 3 . l 3
3 5 7 9 ll 13
Wetness Index

 

 

 

Appendix 4.18. Plot of 95% confidence band of soybean yield difference between T3

and T4, where WI as covariate. Arrow-head line indicates zero yield difference.

15]

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I 3
w 2 ‘
O _________
1: . _________________
3 - ------- #_
§ 1 ‘T r
'6 '2 0 """"""""""""""""""""""""""" t
a an
s 2 .
5 -l ' 'I‘3-T4 in 1998
o d
E 2 4 ----- umr
1 ------- lower
I -3 1 5 . i . : r : . J.
1 3 5 7 9 11 13
’ Wetness Index
11 3
2 1
E ‘ ___________
1: 1- ’ ---------------
g _, + ...................................................
a o >
.1:
a .. .
“>1 5 -1 . 13:14 in 2001
:1
3 ' ----- upper
2 '2 . ------- lower
-3 . + . s . 3 . : . t
3 5 7 9 11 13
: Wetness Inthx
Ill 3
«a 2 ‘ ff.- .....................
O ........................
1.: . ...............
g _________________ fi
81:“ l 1
S 2 . ,
7. E0 0
2 1 : T3-T4 in 2004
s . ————— u r
z I!"
-2 ‘ ------ lower
'3 1 f 1 l v i I 1 I
3 5 7 9 11 13
Wetness Index

 

 

Appendix 4.19. Plot of 95% confidence band of wheat yield difference between T3

and T4, where WI as covariate. Arrow-head line indicates zero yield difference.

152