University of Southern California Dissertations and Theses

Precision agriculture and GIS: evaluating the use of yield maps combined with LiDAR data

(USC Thesis Other)

Precision agriculture and GIS: evaluating the use of yield maps combined with LiDAR data

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

PRECISION AGRICULTURE AND GIS:

EVALUATING THE USE OF YIELD MAPS COMBINED WITH LIDAR DATA

by

Bennett Charles Gaumitz

A Thesis Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(GEOGRAPHIC INFORMATION SCIENCE AND TECHNOLOGY)

August 2016

Copyright 2016 Bennett Charles Gaumitz

ii

DEDICATION
I dedicate this document to my parents without whose unwavering support and faith this
document could never have been completed.

iii

ACKNOWLEDGMENTS
I am grateful to my mentor Dr. Richard Vanden Heuvel, who inspired me to create the topic that
is presented here. I also want to thank my advisor Dr. John Wilson who stayed committed to
helping me throughout my work on my thesis.

iv

TABLE OF CONTENTS
DEDICATION ii
ACKNOWLEDGMENTS iii
LIST OF FIGURES vii
LIST OF TABLES xii
LIST OF ABBREVIATIONS xiii
ABSTRACT xiv
CHAPTER 1: INTRODUCTION 1
1.1 Digital Soil Maps 1
1.2 Management Zones 3
1.3 Motivation 4
1.4 Thesis Organization 5
CHAPTER 2: BACKGROUND AND LITERATURE REVIEW 6
2.1 Precision Agriculture 7
2.1.1 Soil Sampling 8
2.1.2 Fertilizer Recommendations 11
2.1.3 Yield Assessment 12
2.1.4 Use of a Soil Survey 13
2.2 Digital Soil Maps 14
2.2.1 Attributes related to digital soil mapping 15
2.2.2 Digital soil maps in practice 16
2.3 Light Detection and Ranging (LiDAR) 19

v

CHAPTER 3: METHODOLOGY 21
3.1 Overview of Fields in Study Area 23
3.1.1 Dob Along G 23
3.1.2 Emmert 24
3.1.3 Harstad 25
3.1.4 Merkt’s 27
3.1.5 Pribnows 28
3.1.6 Stefoneks 29
3.2.1 Yield Table 32
3.2.2 Yield Interpolation 33
3.2.3 Composite Yield 34
3.2.4 Validation and Comparison of Yield Interpolations 35
3.3 Constructing DEMs from LiDAR 36
3.3.1 Creation of DEM 36
3.3.2 DEM Derived Attributes 38
3.4 Final Analysis Steps 39
3.4.1 Differences between single year and multiple-year yields 39
3.4.2 Impact of slope on yield differences 39
CHAPTER 4: RESULTS 40
4.1 Yield Interpolation Evaluation 40
4.2 Yield Interpolation Results 42
4.2.1 Yield Interpolation 43

vi

4.2.2 Yield Difference Rasters 49
4.2.3 Yield Summary 55
4.3 Yield and Slope Comparison Results 56
CHAPTER 5: DISCUSSION AND CONCLUSION 62
5.1 Yield Interpolation Results 62
5.2 Slope and Yield Comparison 63
5.3 Final Thoughts 65
REFERENCES 66

vii

LIST OF FIGURES
Figure 1: Current NRCS map of soil types in Iowa ........................................................................ 2
Figure 2: Example of fertilizer management zones based on imagery, electrical conductivity,
and yield .......................................................................................................................................... 3
Figure 3: Map illustrating the creation of management zones based on phosphorous levels.
Eight areas representing differences in phosphorous levels are simplified into three zones .......... 4
Figure 4: Example of precision agriculture process from beginning to end. As shown, this
process is centered on fertilizer use. ............................................................................................... 8
Figure 5: Examples showing the diamond pattern and systematic unaligned grid methods to
sample soil. ..................................................................................................................................... 9
Figure 6: Example of soil sampling methods. This figure shows the use of multiple samples
at each grid point to create a composite sample ........................................................................... 10
Figure 7: Example of a yield map showing six ranges of values that are normalized by
bushels per acre ............................................................................................................................. 13
Figure 8: Comparison of (a) SoLIM-derived soil map; and (b) a conventional soil map in
Lubrect, Montana .......................................................................................................................... 15
Figure 9: Example of a regional DSM as used in Germany ......................................................... 17
Figure 10: Demonstration of Management Zone Analyst software that shows the inputs (top)
and examples of 2, 4, and 6 created zones from these inputs ....................................................... 18
Figure 11: Comparison of the slope gradient calculated with (a) 1-m LiDAR derived DEM;
b) 5-m LiDAR derived DEM and (c) 10-m USGS derived DEM ................................................ 20
Figure 12: Overview of the study area. Individual fields from left to right are Stefoneks,
Merkt’s, Harstaad, Dob Along G, Pribnows, and Emmerts.......................................................... 21

viii

Figure 13: Aerial overview of the study area. The area is characterized by numerous rivers
and lakes, in addition to farmland and woodlands. ....................................................................... 22
Figure 14: (a) Photograph of Dob Along G taken in 2015 and (b) an aerial view of the field
with the boundary in blue ............................................................................................................. 23
Figure 15: Example of a cleaned yield maps generated using Yield Editor for Dob Along G.
The classification breaks have blue representing poor yield, green moderate yield, and red
high yield (in bushels per acre). .................................................................................................... 24
Figure 16: (a) Aerial view of Emmert with the boundary in blue and (b) a photograph of the
field taken in 2015 ........................................................................................................................ 24
Figure 17: Example of a cleaned yield points generated using Yield Editor for Emmert. The
classification breaks have blue representing poor yield, green moderate yield, and red high
yield (in bushels per acre). ............................................................................................................ 25
Figure 18: (a) Aerial view of Harstad with the boundary in blue and (b) a photograph of the
field taken in 2015 ........................................................................................................................ 26
Figure 19: Example of a cleaned yield maps generated using Yield Editor for Harstad. The
classification breaks have blue representing poor yield, green moderate yield, and red high
yield (in bushels per acre). ............................................................................................................ 26
Figure 20: Aerial view of Merkt’s with the field boundary in blue .............................................. 27
Figure 21: Example of a cleaned yield maps generated using Yield Editor for Merkt’s. The
classification breaks have blue representing poor yield, green moderate yield, and red high
yield (in bushels per acre). ............................................................................................................ 28
Figure 22: (a) Aerial view of Pribnows with the field boundary highlighted in blue and (b) a
photograph taken of the field in Spring 2014. .............................................................................. 28

ix

Figure 23: Example of a cleaned yield maps generated using Yield Editor for Pribnows. The
classification breaks have blue representing poor yield, green moderate yield, and red high
yield (in bushels per acre). ............................................................................................................ 29
Figure 24: (a) Photograph of Stefoneks taken in Spring 2014 aned (b) aerial imagry with
the field boundary in blue ............................................................................................................. 30
Figure 25: Example of a cleaned yield maps generated using Yield Editor for Stefoneks. The
classification breaks have blue representing poor yield, green moderate yield, and red high
yield (in bushels per acre). ............................................................................................................ 31
Figure 26: (a) Terrain in 25 m classes; (b) a 5 m DEM raster of Dob Along G study area. ......... 37
Figure 27: Slope raster based on the 5 m DEM with class breaks that emphasize the
variations in slope. ........................................................................................................................ 38
Figure 28: Yield Interpolations for Dob Along G from 2012 to 2014 and a composite yield
for all three (in bushels per acre). ................................................................................................. 44
Figure 29: Yield Interpolations for Emmert from 2012 to 2014 and a composite yield for all
three (in bushels per acre). ............................................................................................................ 45
Figure 30: Yield Interpolations for Harstad from 2012 to 2014 and a composite yield for all
three (in bushels per acre). ............................................................................................................ 46
Figure 31: Yield Interpolations for Merkt’s from 2012 to 2014 and a composite yield for all
three (in bushels per acre). ............................................................................................................ 47
Figure 32: Yield Interpolations for Pribnows from 2012 to 2014 and a composite yield for
all three (in bushels per acre). ....................................................................................................... 48
Figure 33: Yield Interpolations for Stefoneks from 2011 to 2013 and a composite yield for
all three (in bushels per acre). ....................................................................................................... 49

x

Figure 34: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Dob Along G. It has five categories with a count of the number
of cells within the category. .......................................................................................................... 50
Figure 35: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Emmert. It has five categories with a count of the number of cells
within the category. ....................................................................................................................... 51
Figure 36: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Harstad. It has five categories with a count of the number of cells
within the category. ....................................................................................................................... 52
Figure 37: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Merkt’s. It has five categories with a count of the number of cells
within the category. ....................................................................................................................... 53
Figure 39: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Pribnows. It has five categories with a count of the number of cells
within the category. ....................................................................................................................... 54
Figure 40: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Stefoneks. It has five categories with a count of the number of cells
within the category. ....................................................................................................................... 55
Figure 41: DEM and slope class map for Dob Along G created from LiDAR ............................ 57
Figure 42: DEM and classified slope map for Emmert created from LiDAR .............................. 57
Figure 43: DEM and slope class map for Harstad created from LiDAR ...................................... 58
Figure 44: DEM and slope class map for Merkt’s created from LiDAR...................................... 58
Figure 45: DEM and slope class map for Pribnows created from LiDAR ................................... 59

xi

Figure 46: DEM and slope class map for Stefoneks created from LiDAR .................................. 59

xii

LIST OF TABLES
Table 1: Example of fertilizer recommendations based on phosphorous and potassium on
rice farms in Indonesia .................................................................................................................. 11
Table 2 Raw Yield Data from Dob Along G recorded in 2012 .................................................... 32
Table 3: Column Data Names and Units for Import ..................................................................... 33
Table 4: Interpolation evaluation showing the actual and predicted means and standard
deviation (SD) of the yields at 20 location in each year and the composite. Possible
Interpolation failures are highlighted in yellow. ........................................................................... 41
Table 5: Example of comparison between actual and predicted yield values for Harstad
2014 with the significant differences highlighted in yellow ......................................................... 42
Table 6: Difference of Means Statistical Test Results. Possible failures of interpolation are
highlighted in yellow .................................................................................................................... 43
Table 7: Mean and coefficient of variation for all six fields......................................................... 56
Table 8: Slope and Yield Comparison .......................................................................................... 60

xiii

LIST OF ABBREVIATIONS
DEM Digital Elevation Model
DSM Digital Soil Map
GIS Geographic Information System
IDW Inverse Distance Weighted interpolation method
LiDAR Light Detection and Ranging
MZA Management Zone Analyst
NRCS Natural Resources Conservation Service
TIN Triangular Irregular Network
USDA U.S. Department of Agriculture
USGS U.S. Geological Survey

xiv

ABSTRACT
Precision agriculture in practice utilizes GIS far less effectively than it should. My work at a soil
consulting company has shown that part of the problem is that the literature does not show an
effective way of analyzing soil through GIS that is both scientific and able to be used by those
associated with agriculture. The thesis aimed to answer two questions: (1) Are there significant
differences between a multiple year composite yield and a single year and if so, are these
significant enough to have an impact on normal operations? (2) Are the areas where such
differences occur related to slope? Six fields were used for this study: Dob Along G, Emmert,
Harstad, Merkt’s, Pribnows, and Stefoneks located near New Richmond, Wisconsin. Three years
of yield data were used for each field and slope data was created using LiDAR from St. Croix
County. These yield data were interpolated using standard industry practices. A single year was
compared to composite years to determine what differences, if any, exist between them. Each
year and the composite had actual and predicted values compared using the difference of means
statistical test to validate the success of the interpolation. A DEM was created from LiDAR and
this was used to create a slope map of each field. This slope map was used to divide the yield
points by five slope classes: 0 -1°, 1-2°, 2-3°, 3-5°, and >5°. The mean yield and variation was
then compared for each class to determine any patterns associated with slope values. The results
show that where there is significant variation between single years of yield data the composite
will fail. The difference between the composite and a single year is useful in identifying which
fields are causing the composite to fail and eliminating them. Slope did not consistently correlate
to changes in yield or variation. Dob Along G, Emmert, and Harstad showed no correlation,
while Merkt’s, Pribnows, and Stefoneks showed decreasing yield and increasing variation as
slope increased.

1

CHAPTER 1: INTRODUCTION
Precision Agriculture is a practice in agricultural that examines and manages a section of farm
rather than its whole. It uses various tools, especially rigorous soil testing, to provide a detailed
understanding of the soil on a single farm or a section of that farm. This allows for variable-rate
fertilizer recommendations, based on soil testing, to improve particular areas while using the
least amount of fertilizer. The use of a Geographic Information System (GIS) within this process
is under-utilized. Primarily it’s use is reactive, as a final test to understand the effects of the
previous year’s fertilizer recommendation. The yield data taken from farmers is made into a map
that demonstrates the overall performance of a specific farm. If a farm is under-performing this
indicates a need to revise future recommendations. However, since yield maps are not used to
make recommendations, these are generally just stored for possible later use.
1.1 Digital Soil Maps
There has been some effort to create digital soil maps (DSMs) that combine attributes like soil
type, weather pattern, and chemical composition in a GIS to demonstrate the overall fertility of
an agricultural region or to predict its production. The most notable example is the regional soil
map created by the Natural Resources Conservation Service (NRCS) that describes the general
soil types within a specific area, e.g. the Iowa Regions Map in Figure 1.
There has been substantial research put into the creation of predictive DSMs that have
established methodologies and data needs for their creation (Zhu et al. 2001; McBratney, Santos,
and Minasny 2003). Models like Scorpan (McBratney, Santos, and Minasny 2003) use a variety
of attributes from numerous sources to provide a detailed and complete regional DSM and
complicated techniques to ensure the accuracy of the final output. The NRCS uses traditional
techniques to construct a similar map that relies on expert knowledge and surveying methods.

2

Figure 1: Current NRCS map of soil types in Iowa
(http://www.nrcs.usda.gov/Internet/FSE_DOCUMENTS/nrcs142p2_006361.pdf)

3

Similar methods have been used in the creation of digital methods by capturing and using expert
knowledge (Zhu et al. 2001). These approaches are used to create a comprehensive map at the
regional level that is ineffective in modern precision agriculture which uses information at the
sub-farm level to make recommendations.
Comprehensive regional DSMs fall short of this goal in two ways: (1) the detail is
intended for a region, so it is not precise enough to be used for even a single farm, let alone a
section of that farm; and (2) the attributes used in the creation of these maps are too numerous,
creating extraneous information that can make such a map confusing or cumbersome for direct
precision application.
1.2 Management Zones
In order to change this, there have been numerous attempts to create functional management
zones for farmers (Fleming et al. 2000; Fleming, Heermann, and Westfall 2004; Fridgen et al.
2004). These maps are designed to delineate zones based on soil attributes to better organize a
farmer’s field for fertilizer application. Figure 2 offers an example of a three-zone management
map based on imagery, electrical conductivity, and yield (Fridgen et al. 2004). Figure 3 shows an
additional example of management zones based on phosphorous levels.
Figure 2: Example of fertilizer management zones based on imagery, electrical
conductivity, and yield. Source: Fridgen et al. (2004).

4

Some of these authors have expressed concerns about the use of these kinds of maps
because of changes that can occur due to topographic effects and climate changes that can affect
the productivity of a field over time (Schepers et al. 2004). However, these techniques are not
generally used in practice, due to the difficulty of obtaining and updating the attributes used to
create these zones.

Figure 3: Map illustrating the creation of management zones based on phosphorous levels.
Eight areas representing differences in phosphorous levels are simplified into three zones
(http://www.innovativegis.com/basis/pfprimer/Appendix_A/Appendix_A.htm)
1.3 Motivation
Consultants and farmers have direct access to critical details for the creation of management
zones from soil sampling. This means that a predictive map does not have to show all of the soil

5

information, but only requires the attributes necessary to contribute to the soil management
process. Rather than creating something new that can already be extrapolated from soil sampling,
additional attributes can be used to supplement these data. These attributes must correlate with
necessary information and provide enough detail to develop a predictive map. Researchers have
documented DEMs and imagery related to information like chemical composition and yield (e.g.
Varvel, Schlemmer, and Schepers, 2004; Dobos et al. 2000).
My research aims to utilize techniques related to yield maps and Light Detection and
Ranging (LiDAR) to explore their potential for use in precision agriculture. The emphasis will be
on the accuracy of single- and multiple-year yield maps and if the slope generated from LiDAR
can explain the differences between these maps. The underlying assumption is that in some cases
a single year yield map is sufficiently accurate to characterize the yield of a field and that slope is
an indicator of where single year maps fail due to higher variability from year to year and less
accuracy overall.
1.4 Thesis Organization
The remainder of this thesis consists of four chapters. The second chapter explores background
information on precision agriculture, DSMs, yield maps, and LiDAR data in an effort to explain
the importance of these supplemental data and their use. The third chapter describes the
techniques that were used to demonstrate the effectiveness of the yield maps and LiDAR data for
documenting the fertility needs on the study farm fields. The fourth chapter presents the results
obtained with this methodology and compares the results with individual years of yield data. The
fifth and final chapter notes the conclusions that can be drawn from these results and discusses
their implications for precision agriculture.

6

CHAPTER 2: BACKGROUND AND LITERATURE REVIEW
Digital soil mapping in general and precision agriculture in particular have been studied for more
than 50 years. A variety of approaches have been proposed for transferring traditional soil
mapping to digital methods. Traditional soil mapping methods were created to form a general
overview of regions and so were inherently generalized and therefore inappropriate for use in
precision agriculture, which has begun to dominate the agricultural industry.
This change in agricultural methods requires new mapping designs that are more specific
and precise in character. Traditional soil maps use discrete areas to mark general soil zones, yet
in reality this is not the case. Methods, such as kriging, have been used to change these into
continuous classes by allowing the general soil lines to become fuzzy (McBratney, De Gruijter,
and Brus 1992). This process also allows for some prediction of classes as these intermediary
soil types bridge the gap between the general soil types already known. Similarly, Bragato
(2004) has proposed using fuzzy modeling techniques on traditional soil maps to produce soil
zones that are not separated by simple lines but fade into one another. This characterization of
the soil more closely resembles the real-world and bridges the gap between the traditional and
the digital.
The utilization of other resources that relate to attributes characterizing soil can enhance
the quality of smaller scale maps. Satellite information was found in some ways to eliminate the
need for traditional soil laboratory work to characterize the general composition of soil
regionally (Nanni and Demattê 2006). Dobos et al. (2000) utilized a DEM to enhance regional
satellite information because satellite data alone have too few details to characterize soil type.
These techniques are a good start to moving from traditional to digital methods, but precision
agriculture requires far more detail.

7

The use of management zones offers a new way of looking at a digital soil map. These
zones are created within a single farm and are characterized by a large amount of testing that
allows for different management techniques (fertilizer, irrigation, etc.) depending on the needs of
a specific zone. Some approaches have used a variety of attributes obtained through digital and
physical means, including soil brightness, electoral conductivity and yield, to delineate these
zones (Schepers et al. 2004). These kinds of approaches require changes to zone management
depending on season and climatic variation.
This chapter covers the related information and literature for this thesis. Section 2.1 will
give an overview of precision agriculture, Section 2.2 will detail digital soil maps including
attributes used in their creation and their use in practice, and Section 2.3 will detail the use of
Light Detection and Ranging (LiDAR) in creating DEMs. These sections will highlight precision
agriculture as the subject of this thesis, overview the process of creating DSMs as a way of
developing which attributes would work with this thesis, and discuss the application of LiDAR
to show how it might be used to create some of these attributes.
2.1 Precision Agriculture
Precision Agriculture is designed to increase the efficiency of a total farm by dividing that farm
into sections and applying different fertilizer and management practices to ensure the best yield
at the lowest cost. The idea is to create greater efficiency between fertilizer use and the possible
benefits. This is especially important for managing the environmental impact of agriculture,
limiting fertilizer to the least needed diminishes fertilizer run-off and the detrimental
environmental effects associated with mining and applying fertilizer (Schieffer and Dillon 2013).
Schieffer and Dillon (2013) have suggested using governmental policy in the form of taxes or
credits to minimize fertilizer use. In order to maximize the benefit-to-cost ratio,

8

recommendations, analysis, and experimental fertilizer are used in combination to reduce cost
while still allowing for the maximum growth of crops in a given area (McBratney et al. 2005).
Generally, this is accomplished by grid sampling a farm and using laboratory testing to
determine the chemical composition of each grid cell individually. These grids can vary in size
depending on the desired level of precision. The tests performed on each grid sample can vary
but typically look at the levels of potassium, phosphorous, organic matter, zinc, and the acidity
of the soil. Once these values are determined an agronomist makes recommendations based on
these values for a whole field, management zone, or individual grid cell. This can depend on the
size of the field, management practices, and fertilizer type. Finally, yield values are collected to
determine the effectiveness of the recommendation and assess the quality of the farm after each
season. This process is illustrated below in Figure 4.
2.1.1 Soil Sampling
Soil sampling is a major source for determining the type and quality of soil information. It is
important that this information be accurate; however, over-sampling is cost prohibitive with
limited practical benefits (Fortunati, Banff, and Pasturenzi 1994). The method of sample
Soil sampling
•Spring, summer, and fall periods
•Several years between samplings
Laboratory Testing
•Chemical Composition
•Variable-rate recommendations
Fertilizer Application
•Pre-plant and post-plant
•Uniform or variable-rate
•Based on farm divisions
Assessment
•Grower Feedback
•Yield Analysis
•Plant Response
Figure 4: Example of precision agriculture process from beginning to end. As
shown, this process is centered on fertilizer use.

9

collection can be based on a grid superimposed on the landscape, soil type, management zone, or
farm unit.
Of these, gridded soil sampling is the most common as it requires the least amount of
guess work and can be part of a streamlined sampling process, e.g. no expert knowledge is
needed to acquire large numbers of samples. Sample locations can consist or random samples
within a grid, a diamond patterned, or a systematic unaligned grid (Wollenhaupt and Wolkowski
1994). These techniques are illustrated in Figure 5. The diamond patterned methods shift the
sampling points by connecting four grid sampling points in a diamond shape, while the
systematic unaligned grid uses a random number table to generate the x, y coordinates for the
sample location within each grid. Both of these methods allow for a more randomized approach
to sampling that does not require composite samples. The grid approach offers the most
variability but commonly accepted practice is a 2.5-acre square grid. Figure 6 shows an example
of composite sampling based on such a grid size.
Figure 5: Examples showing the diamond pattern and systematic unaligned grid methods
to sample soil. Adapted from Wollenhaupt and Wolkowski (1994).

10

The type of soil sampling methods used can greatly affect the results because, regardless
of the number of samples taken, it still provides only a snapshot of the soil. However, this
snapshot diminishes as the number of samples per unit area increases, so that smaller grid sizes
give more specific results.

Figure 6: Example of soil sampling methods. This figure shows the use of multiple samples
at each grid point to create a composite sample
(http://cropwatch.unl.edu/ssm/soilsampling)
The quality is also affected by when and where the sample is taken within the overall
area. Fertilizer is applied in bands, to help reduce fertilizer use (Beyrer and Below 2014); if this
is the case, the characteristics of samples taken within a band or outside of one will vary greatly.
Finally, the time at which a sample is taken can greatly affect the result; for example, if a sample

11

is taken before any new fertilizer has been applied the results will be different than when it is
taken after the fertilizer was applied.
2.1.2 Fertilizer Recommendations
Fertilizer recommendations are based on laboratory testing of the soil samples. These are
performed or managed by an agronomist who interprets the results. Each sample receives its own
recommendation based on the tests results. Different fertilizers are used to manage deficiencies
in key composition values. For example, potassium chloride (potash) is used to add potassium to
the soil. The most important of these values are potassium, phosphorous, and pH content;
however, other values can be used for a more in-depth assessment, e.g. zinc, magnesium,
nitrogen, or calcium.
These values are rated, e.g. very low to very high, which correlate with different levels of
fertilizer use. These recommendations can be based on expert knowledge, scientific analysis, or
both. An example of a recommendation based on chemical values is shown in Table 1. The
values in the table show that differences in nutrient levels are interpreted as high, medium, or
low, and that the recommendations are based on what is already added to a field, e.g. straw. This
process is usually only performed once every several years, as the expense can be greater than
the return if performed more frequently.
Table 1: Example of fertilizer recommendations based on phosphorous and potassium on
rice farms in Indonesia
Nutrient P
(mg P2O5*100g
-1
)
K
(mg K2O*100g
-1
)
Fertilizer
(kg*ha
-1
)
Sp-36 KCl
Without
straw
Straw
Added
Low <20 <10 100 100 50
Medium 20-40 10-20 75 50 0
High >40 >20 50 50 0
Adapted from Sulaeman et al. (2012)

12

Fertilizer rates can be applied as a single or variable rate. A variable rate takes the within-
field variability into account rather than just applying a uniform fertilizer to the field. A variable
rate program is not always used for all fertilizer applications. For example, nitrogen is generally
applied at a uniform (single) rate, due to the limited availability of nitrogen tests.
2.1.3 Yield Assessment
Yield is usually assessed through the use of yield monitors. These monitors record the amount of
crop taken at a specific point when a combine harvests the crop. Based on the mass flow, speed,
moisture content and GPS position, yield is calculated for a specific location and is then used to
generate a yield map for the whole field. These points are only precise to the width of the
combine and its speed. A faster speed can cause incorrect readings to occur which is dependent
on the type of combine that was used. This can be a powerful tool for assessing the results of
fertilizer changes or to detect problem areas in a particular field; however, it does have some
limitations.
The size of the harvester combine makes a difference in accuracy as a smaller, research
combine covers a smaller area and is therefore more precise than a regular sized combine. The
yield monitor itself requires routine calibration and monitoring by the combine driver to ensure
the best results. Even in the most ideal circumstances error can be up to 10 percent of yield. If
the necessary steps are not followed or the driver is inattentive the error generated can make the
yield map useless. This is seen as large gaps in the yield data or inaccurate readings of yield.
Since most yield monitors are used by farmers who are not technicians, the precision of yield
monitoring can vary greatly depending on the field, harvester, and/or farmer.
The yield map is generated from the thousands of points captured with the yield monitor.
The number of points varies greatly depending on the size of the field and harvester combine.

13

Generally, this is normalized as bushels per acre and three or more classes are given colors
ranging from red to green to indicate bad or good yields, respectively in comparison to the
overall yield strength of the field. Figure 7 shows an example of a yield map normalized in this
way with six value ranges.
Figure 7: Example of a yield map showing six ranges of values that are normalized by
bushels per acre (http://www.ptprecisionag.com/products__services)
2.1.4 Use of a Soil Survey
In practice, the soil survey is used as a way to provide a general overview of a farm by showing
the types of soil located there. The soil survey that is typically used was created by the USDA in
the 1970s. These maps show discrete classifications of soils by region. Each soil classification

14

comes with a description of its appearance, structure, and depth. These soil surveys were
developed before precision agriculture became popular and, therefore, are insufficient for use
with it. The survey can give a general indication of what soils can be found in the area but are
not specific or precise enough to depict or differentiate the conditions found in a single field.
This is why they are used only to provide a general overview and not included in variable-rate
recommendations.
2.2 Digital Soil Maps
More comprehensive approaches than the soil survey have been proposed that take more factors
into account to create complete digital soil maps. A prime example of this process is the Scorpan
method that utilizes information on soil, climate, organisms, topography, age, and location to
create a complete soil map of a desired region (McBratney, Mendonça Santos, and Minasny
2003). This approach considers a large number of attributes and uses complex geostatistical
techniques to create a final map.
Another example is SoLIM which attempts to eliminate the use of traditional soil maps
through the inference of soil types from the environmental factors that create them and the
similarity of soils near one another (Zhu et al. 2001). Both Scorpan and SoLIM produce a
regional soil map that is based solely on mathematical principles and digital information (e.g.
Figure 8). However, these methods are very complicated and require many attributes to complete
the map.
Fleming et al. (2000) describe an approach that is applicable to precision agriculture,
specifically, variable rate fertilizer recommendations based on grid sampling. Their approach
describes the use of aerial photographs, topography, and farmer knowledge to create site-specific
management zones where different rates of fertilizer could be applied. These methods were

15

found to be effective in determining zones but could not entirely eliminate the grid sampling
method for determining the fertilizer rate required to maximize production. This approach uses
far fewer attributes but also illustrates that soil sampling is currently unavoidable.

Figure 8: Comparison of (a) SoLIM-derived soil map; and (b) a conventional soil map in
Lubrect, Montana. Source: Zhu et al. (2001)
Scull et al. (2003) reviewed predictive mapping techniques, including geostatistical,
statistical, decision tree analysis, and expert systems. However, the authors noted that despite
these advances, more work needs to be conducted to bridge the gap between GIS methods and
their use in soil science. This observation illustrates that a variety of tools are available, but that
they are not being fully utilized.
2.2.1 Attributes related to digital soil mapping
In order to understand how a digital map works, it is important to consider the attributes that
form the components of that map. Fuzzy K or c-means clustering is a technique that allows for
the more accurate depiction of soil classes. De Bruin and Stein (1998), for example, used this
approach to describe the soil landscape by using attributes derived from a DEM, including slope,
aspect, and curvature.

16

The DEM itself has considerations that need to be addressed to ensure its effective use.
Thompson, Bell, and Butler (2001) considered different levels of information and combinations
from other sources. They found that important information was lost as the resolution was moved
from 10 to 30 m and that more information is lost as different sources of elevation data are used
together. Smith et al. (2006) also explored the effects of neighborhood size and resolution on the
accuracy of the soil map, and found that a more accurate DEM does not necessarily produce
more accurate soil maps because the quality of the soil map will also depend on the type of
landscape (flat or steep) and a correctly determined neighborhood size.
Yield can play an important role in determining the quality of soil. High yield indicates
that the soil is producing the optimal number of bushels. Kasper et al. (2003) showed that yield is
far more effective when combined from multiple years of yield data, due to climate variability.
This was especially helpful in dry years, but may not be as helpful in wet years.
2.2.2 Digital soil maps in practice
There are relatively few examples of potential map creation strategies and their use in
agricultural practices. Behrens and Scholten (2006) show how DSM techniques have been
applied to soil maps in Germany, including the Scorpan process discussed above. They
demonstrate that it is easier to apply these techniques in smaller countries and that an effective
regional map can be created, as shown in Figure 9.
Precision agriculture has seen most of its practical use in the creation of management
zones. For example, Mann, Schumann, and Obreza (2011) delineated citrus productivity zones.
While this does not relate directly to the crops in this project it shows that zones can be created
by looking at related attributes, i.e. yield, canopy volumes, and soil color. Most importantly,
Fridgen et al. (2004) have created a software platform called the Management Zone Analyst

17

(MZA) that automatically creates zones for agricultural fields. It uses a model that combines
electrical conductivity, slope, and elevation with the fuzzy c-means clustering algorithm (Figure
10). This that it is possible to define management zones that can be created to manage specific
areas of a farm. However, methods of determining electrical conductivity are expensive, time-
consuming, and not fully developed.

Figure 9: Example of a regional DSM as used in Germany. Source: Behrens and Scholten
(2006)

18

Figure 10: Demonstration of Management Zone Analyst software that shows the inputs
(top) and examples of 2, 4, and 6 created zones from these inputs. Source: Fridgen et al.
(2004)

19

2.3 Light Detection and Ranging (LiDAR)
LiDAR is an active remote sensing application that sends laser pulses from a satellite or aircraft
and receives that light back giving a range based on the time it takes the pulse to return. This can
be done either by profiling or scanning the ground. Profiling sends just a single pulse in one
direction at a time, whereas scanning sweeps the laser back and forth to cover large amounts of
area. Pulses can be sent very quickly, up to 100,000 per second in some cases, generating large
datasets (Hopkinson and Chasmer 2008). These datasets are commonly used to create high
resolution digital elevation models (DEMs).
LiDAR point clouds refer to the raw points that represent the ranges from each pulse.
Visually these points are not very useful because of the large number of points over a relatively
small area. This is why these points are used to create a DEM that visually depicts differences or
changes in elevation for a chosen area. Esri (2013) presents a guide to do this that consists of
converting the LiDAR points into a multipoint and finally into a terrain dataset. The accuracy of
the dataset is dependent on the distance between the LiDAR points and the choices made in its
creation. This process is further explained in Chapter 3.
LiDAR derived DEMs have been shown to be more accurate than those derived from
topographic maps, even those provided by the U.S. Geological Survey (USGS) (Shi et al. 2012).
A comparison of the slope gradient produced by Shi et al. (2012) of a 1-m and a 5-m LiDAR
compared with a 10 m USGS DEM is shown below in Figure 11. Although the limited
availability of LiDAR can make it impossible or expensive to use in certain regions, it is
becoming more widely available and a better alternative to traditional DEMs.
This chapter demonstrated the use and benefits of precision agriculture; however, digital
soil maps were shown to be ineffective if used with it. The use of yield in DSMs demonstrate

20

their usefulness in predicting the output of a farm. LiDAR was shown to be a possibility for
creating a sufficiently accurate DEM for precision agriculture. Chapter 3 uses yield interpolation
techniques to analyze the effectiveness of composites (combinations of multiple years of data) to
predict past yield and examine the process of creating a DEM from LiDAR to assess its
applicability in precision agriculture. Finally, a derivative slope map was created from the above
DEM and analyzed in relation to yield to determine if a relationship exists.

Figure 11: Comparison of the slope gradient calculated with (a) 1-m LiDAR derived DEM;
b) 5-m LiDAR derived DEM and (c) 10-m USGS derived DEM. Source: Shi et al. (2012)

21

CHAPTER 3: METHODOLOGY
The data used in this thesis are based in a rural section of St. Croix County, Wisconsin. It is
located to the south of New Richmond, west of Hudson, and east of Glenwood City. Figure 12
illustrates the location of the farms used in this study. It follows County Road G and is mostly
characterized as hilly with some flat areas. Figure 13 shows an aerial view of the terrain. From
the photograph it is easy to see the variation in terrain with the numerous lakes, rivers, and
vegetation.
Figure 12: Overview of the study area. Individual fields from left to right are Stefoneks,
Merkt’s, Harstaad, Dob Along G, Pribnows, and Emmerts.
Paul Schottler, a farmer with various fields in this area, provided the yield data for six of
his fields. These fields are discussed in detail in Section 3.1. The observations for these fields in
Section 3.1 were provided by an agronomist. This included three years’ worth of information for
CR - G
·
New Richmond

22

each field as well as an aerial view of each individual field along with a photo taken from each
site in Spring 2015. The data provided were unprocessed, containing a table with attribute
information for each yield point that will be discussed below in Section 3.2.
Figure 13: Aerial overview of the study area. The area is characterized by numerous rivers
and lakes, in addition to farmland and woodlands.
LiDAR data for the study area was obtained from the Planning and Land Use Department
of the St. Croix County Government Center. This data consists of only the unprocessed point
cloud and was used to derive associated elevation data. LiDAR data is discussed in further detail
below in Section 3.3.

23

3.1 Overview of Fields in Study Area
3.1.1 Dob Along G
Dob Along G is a 312-acre field that is characterized by a gently sloping and mostly flat
landscape. Figure 14 shows an aerial view of the field with an outline of the border and a
photograph taken of the field in Spring 2015. There is a north-south oriented crest that lines up
roughly with the household on the north end of the property. It is bordered by County Rd. G to
the north, 160 St. to the east, 140 St. to the south, and a separate field on the west. Figure 15
shows an example of the yield data for this field. This yield data is the composite with a thin
lower section containing moderate yields and the large upper area showing high yields with
moderate to low yields along the edges. There are also thin strips with missing data.

Figure 14: (a) Photograph of Dob Along G taken in 2015 and (b) an aerial view of the
field with the boundary in blue

24

Figure 15: Example of a cleaned yield maps generated using Yield Editor for Dob Along G.
The classification breaks have blue representing poor yield, green moderate yield, and red
high yield (in bushels per acre).
3.1.2 Emmert
Emmert is a 68-acre field that is characterized as a mostly flat field with some gently sloping
area. This field has limited soil variability throughout. Figure 16 shows an aerial view of the field
with an outline of the border and a photograph taken of the field in Spring 2015. This photo
shows the field edge and a knoll located at the north-central end of the field. It is bordered by
190 St. to the east and otherwise surrounded by separate fields on its other sides.
Figure 16: (a) Aerial view of Emmert with the boundary in blue and (b) a photograph
of the field taken in 2015

25

Figure 17 shows an example of the yield data for this field. This yield data is the
composite with a somewhat rectangular shape with little differences in the yield except for some
lower yields around the edges.

Figure 17: Example of a cleaned yield points generated using Yield Editor for Emmert. The
classification breaks have blue representing poor yield, green moderate yield, and red high
yield (in bushels per acre).
3.1.3 Harstad
Harstad is a 135-acre field that is characterized by a large amount of variability in soil with many
hills and small flat areas. Figure 18 shows an aerial view of the field with an outline of the border
and a photograph taken of the field in Spring 2015. There is an east-west crest that travels from
the southeastern side to the house located in the southcentral section of the field. This crest
marks a boundary line between a mostly flat area in the south and an extremely hilly area in the
north. The remainder of the field is moderately hilly with no flat areas and high variability in soil
type as can be seen by the color difference. It is bordered by a forested area to the north,
residential land to the east, 130 Ave. to the south, and separate fields and forests to the west.
Figure 19 is an example of cleaned yield data for this farm.

26

Figure 19: Example of a cleaned yield maps generated using Yield Editor for Harstad. The
classification breaks have blue representing poor yield, green moderate yield, and red high
yield (in bushels per acre).
Figure 18: (a) Aerial view of Harstad with the boundary in blue and (b) a photograph of
the field taken in 2015

27

3.1.4 Merkt’s
Merkt’s is a 90-acre field that is characterized by moderate soil variability with a few hills and
ridge lines. Figure 20 shows an aerial overview of the field with its boundary in blue. It is
bordered by 160
th
Ave to the north, residential areas to the west, and trees on the other sides.
Access was limited so no pictures were obtained for this field. Figure 21 is an example of a
cleaned yield for the farm.

Figure 20: Aerial view of Merkt’s with the field boundary in blue

28

3.1.5 Pribnows
Pribnows is a 360-acre field with limited variability, a gently sloping landscape and crest just
west of the building on the south side. Figure 22 shows an aerial view and photograph of the
field taken in Spring 2014. It is bounded by 160
th
St to the east, 140
th
to the south, County Rd G
to the North. Additionally, there are fields and trees to the north where the field moves away
from the County Road.
Figure 21: Example of a cleaned yield maps generated using Yield Editor for Merkt’s.
The classification breaks have blue representing poor yield, green moderate yield, and
red high yield (in bushels per acre).
Figure 22: (a) Aerial view of Pribnows with the field boundary highlighted
in blue and (b) a photograph taken of the field in Spring 2014.

29

Figure 23 is an example of cleaned yield data for the farm.
3.1.6 Stefoneks
Stefoneks is a 240-acre field that is separated in two by 85
th
St. It is best characterized by a
highly variable landscape that show large differences between the two halves. It is obvious that
this field has only one name because of management purposes and not because it represents a
homogenous field. The east side of the field features a ridge line in the center of the field with
gentle slopes flowing from there to the east side. The west side of the field is a continuation of
the ridge line with flat areas to the north of the ridge and gentle slopes flowing to the west. On
the far west there is a series of knolls. There are also variations in the soil color at the southwest
end. Figure 24 shows an aerial view of the field and a photograph of it taken in Spring 2014. The
field is bounded on the south by 140
th
Ave., farm fields to the east, trees in the northwest,
another field to the northeast, and residential areas with trees in the southwest corner. Figure 25
is an example of cleaned yield data for the farm.
Figure 23: Example of a cleaned yield maps generated using Yield Editor for Pribnows.
The classification breaks have blue representing poor yield, green moderate yield, and red
high yield (in bushels per acre).

30

Figure 24: (a) Photograph of Stefoneks taken in Spring 2014 and (b) aerial imagry with the
field boundary in blue

31

3.2 Working with Yield Data
The fields obtained from Nick Schottler came in the form of shape files. In order to correctly
process this data, the table had to be exported into a spreadsheet and manipulated there. Once the
data was imported into Excel the columns needed to process yield in the USDA Yield Editor
were placed in the correct order. This allows for errors in the data to be removed by this editor.
Once this data has been processed in the editor it is ready to be imported back into ArcGIS. The
editor already properly links each data point to the proper UTM projection, in the case zone 15N.
Finally, the yield data is interpolated using the accepted practice which relies on the inverse
distance weighted (IDW) with distance squared interpolation method. This final process shows
the result that a farmer will look at to evaluate his or her yield.

Figure 25: Example of a cleaned yield maps generated using Yield Editor for Stefoneks.
The classification breaks have blue representing poor yield, green moderate yield, and red
high yield (in bushels per acre).

32

3.2.1 Yield Table
Yield data was acquired in the form of thousands of georeferenced points that vary in quantity
based on the size of the field and calibration of the yield monitor. For example, Dob Along G has
over 4,600 unprocessed yield points to consider. This data comes in the order that the yield
monitor provides it. An example of a truncated list of these data points is provided in Table 2 for
Dob Along G.
Table 2 Raw Yield Data from Dob Along G recorded in 2012
FID Product Swath
Width (ft)
Elevation (ft) Time (date) Yield
Volume
(bu/ac)
Speed
(mph)
Crop Flow
(V)
15599 CORN 30.1 1040.8 2012-10-16 184.89 5.02 3398.20
15600 CORN 30.1 1040.8 2012-10-16 185.20 4.97 3368.53
15601 CORN 30.1 1040.7 2012-10-16 154.57 5.99 3388.79
15602 CORN 30.1 1040.8 2012-10-16 186.33 4.97 3388.79
15603 CORN 30.1 1040.7 2012-10-16 186.38 5.00 3409.08
15604 CORN 30.1 1040.6 2012-10-16 185.18 5.02 3401.58
15605 CORN 30.1 1040.6 2012-10-16 187.30 4.94 3389.34
15606 CORN 30.1 1040.6 2012-10-16 186.12 4.99 3402.00
15607 CORN 30.1 1040.5 2012-10-16 166.26 5.02 3054.05
15608 CORN 30.1 1040.4 2012-10-16 168.98 4.99 3088.60
15609 CORN 30.1 1040.3 2012-10-16 169.70 4.99 3101.86
15610 CORN 30.1 1040.3 2012-10-16 171.09 4.97 3113.11
15611 CORN 30.1 1040.3 2012-10-16 171.13 4.99 3129.84
15612 CORN 30.1 1040.2 2012-10-16 169.58 4.99 3099.84
15613 CORN 30.1 1040.1 2012-10-16 167.48 5.02 3076.52

As can be seen from the table there are instances where units for a column are not given.
These must be determined based on a reasonable suspicion of what units correspond to the
number in that column based on common practice and magnitude. For example, a typical swath
width on a combine is 30 ft so the column labeled swath width is clearly in feet. This table also
shows data that can help remove erroneous points. For example, if the speed suddenly varied
from its relatively stable 5 mph this would be an indication that the yield monitor was picking up
routine shifts in the combine that were not meant to be monitored.

33

Once these fields have been exported from the shape file into a table the columns need to
be made to match the format used by the Yield Editor which is shown below in Table 3. The data
must appear in this exact order with the correct units in order for the software to correctly
process the data. The Yield Editor than removes points based on user selections and manual
edits. The Yield Editor will automatically check the data from the table to determine automatic
limits on various elements including but not limited to flow delay, velocity (maximum and
minimum), minimum swath, and yield (both minimum and maximum). The listed filter
selections above were used to clean the data. Attributes related to yield are chosen based on the
user crop choice and position based on the user input UTM zone.
The Yield Editor was next used to remove erroneous points by looking for excessive low
or high yield values and other anomalies associated with the data. This cleaned yield data was
then by imported back into ArcGIS to complete the yield analysis.
Table 3: Column Data Names and Units for Import
Column Name Units
Longitude decimal degrees
Latitude decimal degrees
Flow pounds per second
Time seconds
Logged Interval seconds
Distance inches
Swath inches
Moisture percent wet
Adapted from Yield Editor 2.0 Manual
3.2.2 Yield Interpolation
The yields were interpolated using the IDW technique with distance squared because this is the
most commonly used technique to visualize the thousands of yield points available from a yield
monitor. This was done with the IDW tool using a radius of 5 m and the smooth neighborhood

34

function. This tool finds whatever points lie within this radius and uses these to compute the
value of a cell. The smooth neighborhood function makes sure that points closer to the center of
the circle carry greater weight than those further away. The class ranges for the interpolation
were broken down into six evenly distributed classes between 0 and 300. The upper limit was
chosen because it represents an optimal maximum yield to stop at. The interpolation used a
smooth nearest neighbor function to generate the raster because this is the standard approach in
yield mapping.
3.2.3 Composite Yield
The composite yield is generally created by normalizing yield between zero and one through
dividing the yield by the maximum yield a particular farm can achieve, which is determined by
the analyst. Once normalized any number of years can be combined together, but typically three
are considered a good number of years. This process was not strictly adhered to for this project
because it inherently loses data in the normalization process and limits the ability to look at
actual yields.
The composite yield was created by combining multiple years of yield data into one yield
interpolation. The process was the same as for a single field (as discussed above), but before the
yield undergoes error checking, all of the uncleaned yield data is combined into one table. In this
way no data was lost through any intermediate normalization. This also means that the Yield
Editor may remove different points from the composite than for each year. This is because the
editor uses a statistical technique to determine which points are removed. Differences can and
did occur from one year to the next and the composite.
The reasoning behind using so many yield points is that any errors that exist in a single
year will be diminished by using multiple years of data. This thesis examines these differences to

35

look at the efficacy of this claim. Figures 15, 17, 19, 21, 23, and 25 above show the cleaned yield
editor results for each of the six fields as viewed from within the editor.
3.2.4 Validation and Comparison of Yield Interpolations
In order to validate the yield interpolations, 20 approximately equal sized grids were generated
on each field and a control point was randomly selected within each grid. These control points
were then removed and the interpolation run on the remaining points. The interpolated raster
values for the location of these points were then compared to the actual values of these points
from the original yield data.
The difference of means statistical test was used to evaluate the significance of the
differences for each year and the composite for all fields. The standard error and degrees of
freedom were calculated using Equations (1) and (2) below. Finally, the test statistics was
calculated using Equation (3) and used to find the p-value in the statistical chart. Where SE is the
standard error, DF is the degrees of freedom, s1 and s2 are the standard deviation for the actual
and predicted values respectively, n1 and n2 are the size of the sample for the actual and predicted
values respectively, t is the test statistic, x1 and x2 are the mean of the actual and predicted values
respectively, and d is the hypothesized difference between the two means. Values were
considered significant if below 0.05 which is a standard threshold value for use with scientific
data.
SE = sqrt[(s1
2
/n1) + (s2
2
/n2)] (1)
DF = (s1
2
/n1 + s2
2
/n2)
2
/ {[ (s1
2
/ n1)
2
/ (n1 - 1)] + [ (s2
2
/ n2)
2
/ (n2 - 1)]} (2)
t = [(x1 - x2) - d] / SE (3)
Once these data were validated using the above method some comparisons can then be
made by looking at the changes in mean and variability between interpolations from all years of

36

the fields. This can help identify years where drastic changes could affect the composite and
demonstrate where improper classification values could damage a yield interpolation.
3.3 Constructing DEMs from LiDAR
The LiDAR from the St. Croix Government Center came in the form of point clouds saved as
.LAS files. This LiDAR information was obtained by AYRES Associates in the spring of 2014.
AYRES is an engineering and architectural company that has geospatial and GIS related services
to clients. The provided processed DEM has a nominal point spacing of one meter with a vertical
tolerance of 0.61. However, the raw point cloud was used in order to fully explore the process of
DEM creation. The data was separated into cells of points that correspond to latitude and
longitude. Each yield dataset had one to four cells of LiDAR data associated with it depending
on its size and location. A DEM of each field was created with these point cloud data.
3.3.1 Creation of DEM
In order to create a DEM from LiDAR, the grid cells associated with a field are extracted and
transformed into multipoint vector data. This simply transforms the raw LiDAR points into a
usable format. A terrain dataset was created using this multipoint data with an average point
spacing of one meter. One meter was chosen because it represents the most accurate dataset that
can be created with the LiDAR available. The terrain dataset is a triangular irregular network
(TIN) with different levels of detail called pyramids. These pyramids are for visualization
purposes only as the most detailed level is used to create the DEM. Basically the terrain dataset
is an intermediate step before the creation of the final DEM. The spacing between LiDAR points
is dependent on the data provided. In this case the data provided from St. Croix County was
described in its metadata as having a nominal point spacing of one meter. The example terrain

37

for Dob Along G is shown in Figure 26. It can be changed to greater levels of tessellation to
increase efficiency if less accurate data is required.
Once the terrain has been created it is visibly inspected for errors and transformed into a
raster. This raster is the DEM and is a digital representation of the elevation in the area. The
difference between the DEM and the terrain dataset is that the terrain is constructed with
triangles while the DEM is constructed with squares. The DEM allows for easier visualization,
manipulation, and classification of the data. This DEM is shown in Figure 26b. As can be seen
from the elevation dataset there is some noise associated with the data. For example, there is a
small area included in the DEM that differs from the terrain located in the northwest of Dob
Along G. Man-made structures can have a direct impact on the quality of the DEM as well.

Figure 26: (a) Terrain in 25 m classes; (b) a 5 m DEM raster of Dob Along G study area.
m
m
m
m
m
m
m

38

3.3.2 DEM Derived Attributes
The DEM can be used to construct derivative maps of various other associated attributes; for
example, slope, aspect, and elevation contours. This project used DEM data to create slope maps
that were used for visual analysis. The slope map better characterized how the elevation varied
across each of the study fields.
The slope map was created directly from the DEM as the rate of change in percent. The
class breaks used in these derivative slope maps are very important because a small change in
slope could have a great impact on yield. Also, if not correctly symbolized the slope map will be
mostly monotone on such a flat surface. An example of this output is shown in Figure 22. For the
purposes of this thesis five slope classes were chosen: 0°, > 0°- 1°, > 1° - 2°, > 2° - 3°, > 3° - 5°,
and > 5°. These ranges emphasize small changes in slope.
Figure 27: Slope raster based on the 5 m DEM with class breaks that emphasize the
variations in slope.

39

3.4 Final Analysis Steps
All of the methods and data described in Sections 3.1 through 3.3 were next used to answer two
questions: (1) Are there significant differences between a multiple year composite yield and a
single year and if so, are these significant enough to have an impact on normal operations? (2)
Are the areas where such differences occur related to differences in slope?
3.4.1 Differences between single year and multiple-year yields
The yields in each of the six fields were interpolated using the above methods four times, once
for three single years and once for these years combined. These results were compared against 20
randomly selected control points to demonstrate the efficacy of this approach and the full results
were then used to compare the single year and multiple-year yields. Finally, yield difference
rasters were created highlighting the differences between the single and multiple-year
interpolations to show the areas where the most significant differences were located. This
workflow was implemented to answer the first question.
3.4.2 Impact of slope on yield differences
For the second question, a DEM was created for each farm using the LiDAR procedure discussed
above. This DEM was in turn used to create a degree slope map with five categories of slope.
The first category containing only 0° slope values was removed as no field had members
included in this category. This final slope map was then compared to the yield difference rasters
created above to identify whether the increased yield differences were associated with those parts
of the field with higher slopes.
The results are reported in Chapter 4 and their significance is discussed in Chapter 5.

40

CHAPTER 4: RESULTS
The results are presented for each of the six farms in three steps (i.e. sections). The first
compares predicted and measured yields to assess the efficacy of IDW as an interpolation
method for yields on the six fields. The second compares a single year to a composite year yield
in order to show if a single year offers similar guidance to the composite yield output. The third
compares the difference between the composite and single year yield to the slope value to
determine which slope values, if any, accompanied the largest differences in yield values.
4.1 Yield Interpolation Evaluation
The predicted values from the interpolation were compared with the actual values from 20 yield
points within each field. These points were chosen randomly within grid squares in each field
excluding a 10 m buffer inside the field boundary. This approach generated a spatially stratified
set of test points for each field and also minimized any edge effects associated with the field
boundary.
Table 4 shows the yield mean for each year in each field along with the composite. The
difference between each pair of means is also shown. This difference was used as to judge the
efficacy of the interpolation. The composite yields for four of the six fields (highlighted in
yellow in Table 4 below) shows how IDW struggled to predict yields using three years of data
that was passed through and filtered by the Yield Editor, thereby contradicting the usual pattern
in which larger volumes of data improves the efficacy of interpolation.
Table 5 shows the yield values for Harstad in 2014 to demonstrate how small differences
in the mean can mask substantial yield volatility for individual locations (as illustrated by the
actual and predicted yields for the grid cells shaded in yellow). Similar results occurred for all
six fields in 2014 although Harstad shows the most significant differences and although this

41

Table 4: Interpolation evaluation showing the actual and predicted means and standard
deviation (SD) of the yields at 20 location in each year and the composite. Possible
Interpolation failures are highlighted in yellow.
FIELD YEAR SD ACTUAL PREDICTED SD DIFFERENCE
DOB ALONG G 2012 26 190 189 20 1
2013 40 176 168 27 7
2014 46 159 153 20 6
All 31 202 197 21 5
EMMERT 2012 19 194 189 10 5
2013 34 88 86 16 2
2014 63 117 128 39 10
All 19 194 116 32 79
HARSTAD 2012 45 159 164 26 5
2013 42 79 80 30 1
2014 51 132 122 22 10
All 42 79 121 31 41
MERKT'S 2012 37 159 157 20 2
2013 27 54 61 13 7
2014 51 100 98 30 2
All 27 54 98 26 43
PRIBNOWS 2012 20 190 187 13 3
2013 27 88 89 24 2
2014 32 172 170 18 2
All 27 88 148 18 60
STEFONEKS 2011 20 197 194 11 3
2012 21 173 178 10 5
2013 66 152 151 48 1
All 21 173 171 31 2

would not impact single field evaluations, it could greatly impact precision agriculture
recommendations.
Finally, as a check to see if these results are significant, a difference of means statistical
test was used to generate a p-value and evaluate whether or not the reported differences were
significantly different from zero. This p-value indicates that a result is significant if it is less than
0.05. Table 6 shows a summary table for these values with statistically significant results (i.e.
differences) highlighted in yellow. As can be seen, yields for four of the six composites show
statistical significance. This result occurred because the composite is not simply a combination

42

Table 5: Example of comparison between actual and predicted yield values for Harstad
2014 with the significant differences highlighted in yellow
ID Actual Predicted
1 145 167
2 23 66
3 161 126
4 89 137
5 187 119
6 154 143
7 41 117
8 186 140
9 148 101
10 124 129
11 176 116
12 74 120
13 105 121
14 189 138
15 90 75
16 142 123
17 123 112
18 179 129
19 105 119
20 202 136

of the three cleaned years but a combination of the unclean points that were then cleaned by the
Yield Editor together. If one year is drastically different from the other two, then the composite
fails to predict yield values. Normally, drastically different years would be removed by the
analyst before a composite is made.
4.2 Yield Interpolation Results
The similarities and differences between the interpolated individual year and the composite
yields are presented in three parts below. The first part shows all of the interpolated yields for
each year and the composite in each field. The second part shows yield difference rasters for the
composite and each year.

43

Table 6: Difference of Means Statistical Test Results. Possible failures of interpolation are
highlighted in yellow.
Field Year P Value Significant
Dob Along G 2012 0.899332 No
2013 0.500536 No
2014 0.572421 No
All 0.540056 No
Emmert 2012 0.272875 No
2013 0.806777 No
2014 0.537597 No
All 1.21E-10 Yes
Harstad 2012 0.689998 No
2013 0.958383 No
2014 0.405363 No
All 0.001136 Yes
Merkt's 2012 0.814679 No
2013 0.320121 No
2014 0.91044 No
All 8.78E-06 Yes
Pribnows 2012 0.524044 No
2013 0.809683 No
2014 0.809867 No
All 1.38E-09 Yes
Stefoneks 2012 0.55353 No
2013 0.360466 No
2014 0.942276 No
All 0.801536 No

The third part reports the mean and coefficient of variation for each year and the composite and
also includes a discussion of yield between years and fields as a way to characterize the strength
of the yield for a particular field and the changes that can occur within it.
4.2.1 Yield Interpolation
The interpolated yield maps for these fields reproduced in Figures 28-33 were constructed using
yield data acquired from 2012 to 2014 (except for Stefoneks which includes data for 2011-2013).

44

Figure 28: Yield Interpolations for Dob Along G from 2012 to 2014 and a composite yield
for all three (in bushels per acre).

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

45

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

Figure 29: Yield Interpolations for Emmert from 2012 to 2014 and a composite yield for all
three (in bushels per acre).

46

(bu/ac) (bu/ac)
(bu/ac)
(bu/ac)
Figure 30: Yield Interpolations for Harstad from 2012 to 2014 and a composite yield for
all three (in bushels per acre).

47

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

Figure 31: Yield Interpolations for Merkt’s from 2012 to 2014 and a composite yield
for all three (in bushels per acre).

48

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

Figure 32: Yield Interpolations for Pribnows from 2012 to 2014 and a composite yield
for all three (in bushels per acre).

49

Figure 33: Yield Interpolations for Stefoneks from 2011 to 2013 and a composite yield for
all three (in bushels per acre).
4.2.2 Yield Difference Rasters
The single year interpolations were then subtracted from the composite to create three difference
yield difference rasters for each of the six fields. These yield difference rasters show how the
estimated yield for each year in each raster cell differed from the composite. Figures 34 - 39
show the three yield difference rasters created using this method.
Dob Along G and Stefoneks have the most consistent results with a composite that did
not fail the significance test. These maps also show the most variation in negative and positive
differences that are generally centered around the middle. The remainder of the fields show the
differences where the composite failed the significance test. Generally, these fields show
negative differences for the first year, positive results for the second year, and mixed results for
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

bu/ac
bu/ac
bu/ac
bu/ac
bu/ac
bu/ac

50

the final year. The first year consistently performed the worst for these four fields and the
composite would likely perform better if this particular year was removed.
Figure 34: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Dob Along G. It has five categories with a count of the number of
cells within the category.
(bu/ac)
(bu/ac)
(bu/ac)

51

Figure 35: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Emmert. It has five categories with a count of the number of cells
within the category.
(bu/ac)
(bu/ac)
(bu/ac)

52

(bu/ac)
(bu/ac)
(bu/ac)
Figure 36: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Harstad. It has five categories with a count of the number of cells
within the category.

53

(bu/ac)
(bu/ac)
(bu/ac)
Figure 37: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Merkt’s. It has five categories with a count of the number of cells
within the category.

54

Figure 38: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Pribnows. It has five categories with a count of the number of cells
within the category.

(bu/ac)
(bu/ac) (bu/ac)

55

Figure 39: Maps showing the difference between the single year and the composite yields
(in bushels per acre) for Stefoneks. It has five categories with a count of the number of cells
within the category.
4.2.3 Yield Summary
Table 7 shows the summary metrics for the six fields included in this study. The mean and the
coefficient of variation are reported in bushels per acre. The mean gives an idea of the overall
yield capability of a field, this is useful in accessing the general quality of the field, and the
coefficient of variation gives an idea of the variability of yield within a field.
This table shows that all of the fields produce low yields. A good yield from a field
would be 300 to 400 bu/acre. It is clear that Dob Along G and Stefoneks show the most
consistent yield across the three years. The yield of the other four farms plummeted in 2013 and
Merkt’s and Harstad also recorded low yield in 2014. The lowest coefficient of variation was
(bu/ac)
(bu/ac)
(bu/ac)

56

Table 7: Mean and coefficient of variation for all six fields
Field Metric 2012 2013 2014 2012-2014
Dob Along G Mean 179 156 151 183
Coefficient of
Variation
.164 .251 .234 .186
Emmert Mean 185 86 130 124
Coefficient of
Variation
.083 .237 .294 .214
Harstad Mean 161 79 118 112
Coefficient of
Variation
.179 .407 .279 .251
Merkt’s Mean 151 61 96 91
Coefficient of
Variation
.183 .357 .361 .274
Pribnows Mean 181 84 155 136
Coefficient of
Variation
.100 .290 .236 .203
Stefoneks* Mean 188 173 149 165
Coefficient of
Variation
.115 .121 .338 .206
* Stefoneks results are for 2011-2013
reported for all six fields in the first year of record and generally, relatively large variations were
associated with low yields. These results suggest that the fields which perform poorly also
display great variation in yield which is most likely due to either landscape variation across the
field and/or low fertility soils. The slope results help to clarify whether one or both of these
explanations are applicable.
4.3 Yield and Slope Comparison Results
The DEM and slope classes for all six fields are summarized in Figures 35 through 40. The
ranges of elevation found in each DEM differ depending on the range and variability of elevation
found in each field. The slope, however, is displayed with a series of consistent classes to allow
for comparison between fields. There is also a count of the number of cells in each slope
category.

57

Figure 40: DEM and slope class map for Dob Along G created from LiDAR

Figure 41: DEM and classified slope map for Emmert created from LiDAR

>0 -1°
>1 – 2°
>2 – 3°
>3 – 5°
>5°
>0 -1°
>1 – 2°
>2 – 3°
>3 – 5°
>5°
m

m

58

Figure 42: DEM and slope class map for Harstad created from LiDAR

Figure 43: DEM and slope class map for Merkt’s created from LiDAR

>0 -1°
>1 – 2°
>2 – 3°
>3 – 5°
>5°
>0 -1°
>1 – 2°
>2 – 3°
>3 – 5°
>5°
m

m
m

m

59

Figure 44: DEM and slope class map for Pribnows created from LiDAR

Figure 45: DEM and slope class map for Stefoneks created from LiDAR
>0 -1°
>1 – 2°
>2 – 3°
>3 – 5°
>5°
>0 -1°
>1 – 2°
>2 – 3°
>3 – 5°
>5°
m

m
m

m

60

The mean and coefficient of variation for yield is summarized by slope class for each
field of each year and the composite in Table 8. The number of cells belonging to each class is
shown as well.
Table 8: Slope and Yield Comparison
Field Slope
class
No. of
Cells
Metric 2012 2013 2014 2012-
2014
Dob Along G 0-1 123,229 Mean 186 160 148 189
CV .146 .253 .251 .175
1-2 233,696 Mean 181 159 152 185
CV .160 .247 .232 .182
2-3 143,345 Mean 175 153 153 178
CV .173 .246 .221 .189
3-5 40,353 Mean 171 146 150 173
CV .169 .245 .224 .191
> 5 4,015 Mean 155 132 140 151
CV .149 .282 .248 .197
Emmert 0-1 18,295 Mean 186 81 121 119
CV .091 .256 .363 .255
1-2 45,471 Mean 187 85 130 124
CV .081 .239 .302 .222
2-3 36,219 Mean 185 .87 134 126
CV .074 .223 .27 .199
3-5 16,216 Mean 181 89 134 126
CV .076 .225 .241 .171
> 5 1,922 Mean 164 86 126 119
CV .101 .246 .206 .177
Harstad 0-1 26,231 Mean 167 86 118 116
CV .151 .442 .267 .248
1-2 50,413 Mean 163 78 119 113
CV .174 .421 .284 .256
2-3 49,149 Mean 164 77 119 112
CV .168 .377 .281 .244
3-5 64,466 Mean 162 78 118 112
CV .169 .391 .281 .248
> 5 45,638 Mean 152 79 116 110
CV .216 .413 .272 .255
M er kt ’ s 0-1 14,451 Mean 163 61 108 100
CV .121 .310 .338 .241
1-2 25,196 Mean 161 61 102 96
CV .134 .341 .341 .256
2-3 24,515 Mean 159 62 96 93
CV .146 .345 .335 .255
3-5 43,084 Mean 153 62 95 91

61

CV .165 .348 .343 .253
> 5 49,501 Mean 138 60 91 86
CV .222 .389 .390 .305
Pribnows 0-1 168,781 Mean 184 85 158 138
CV .085 .252 .228 .187
1-2 248,935 Mean 184 86 158 138
CV .091 .272 .230 .196
2-3 142,485 Mean 181 85 154 135
CV .097 .304 .233 .203
3-5 61,229 Mean 171 76 146 125
CV .122 .375 .258 .235
> 5 8,792 Mean 151 64 133 111
CV .169 .439 .300 .287
Stefoneks 0-1 65,961 Mean 194 178 152 169
CV .104 .109 .325 .199
1-2 113,919 Mean 176 176 152 168
CV .111 .111 .325 .198
2-3 89,727 Mean 189 174 151 166
CV .109 .111 .330 .205
3-5 113,919 Mean 185 170 149 163
CV .112 .121 .343 .206
> 5 62,282 Mean 177 164 140 155
CV .129 .143 .359 .209

Dob Along G shows yield generally decreasing with slope but with no consistent pattern
in terms of variability. Emmert has few steep slopes and no evident pattern in yield or variability.
Harstad has a large number of cells spread out over all the slope classes but yields and variability
behaved similarly across all slope classes. Merkt’s shows decreasing yield corresponding with
increasing variability as the slope increases. The same is true for Pribnows and Stefoneks.

62

CHAPTER 5: DISCUSSION AND CONCLUSION
This thesis project tackled an exceedingly difficult topic because both natural and human factors
can affect the yield and its variability through space and/or time. Most precision agriculture
applications fuse two or more data sources and types (elevation, soil attribute(s), yield, etc.) and
use interpolation to integrate these disparate datasets across space and/or time. The efficacy of
this type of approach for interpolating yields and trying to clarify the role of slope, if any, was
taken up for the work at hand.
5.1 Yield Interpolation Results
The yield interpolation provided some interesting results. The validation results showed that the
composite gave accurate representations of yield only when all years used had similar averages
and variability; for this reason, Dob Along G and Stefoneks have accurate composites, while the
remaining fields each have a year that served as an outlier compared to the others.
The difference maps created by subtracting the composite from each year were a good
way to identify these outliers. Normally, when creating these composites, an agronomist or
researcher will look at each year and remove a year’s results if the interpolation looks drastically
different. This could not be done here because only three years of yield data were available. The
difference map is a viable way to objectively remove outlier years without resorting to a
subjective test. All that would need to be decided is how much a field can be different from the
composite in order for it to be removed. Just as importantly, the presence of variability might be
used to justify the use of maps from the previous year and to avoid the use of composite yield
maps altogether.
The yield interpolation itself can provide a great description of the field. Higher yield
values come from those fields with more uniform soil. More variability is seen where the yield is

63

lower but this variability does not necessarily correlate with slope. This is probably due to a
more complicated relationship between yield and slope than was assumed in this thesis project. It
should be noted that the fields used here are hilly with poor soils so many of the judgments and
descriptions can only be extended to similar types of fields. Another area with flat or nearly flat
landscapes and better soils may have vastly different characteristics and relationships.
Overall, this section answers the first question: Are there significant differences between
a multiple year composite yield and a single year and if so, are these significant enough to have
an impact on normal operations? The answer is it depends on the farm in question. In some
cases, like Dob Along G and Stefoneks, the yield data is consistent enough that one year could
be used instead. However, where the composite fails there is no way to avoid using the previous
year’s result to guide the next year’s strategy.
The alternative answer to the two aforementioned strategies may be the removal of
problem years while using one of the remaining years as a “best fit” year. However, there are
many factors that can create the so-called “problem” years, from equipment failures to
environmental and climate factors. If the former, these years can and should be eliminated; if the
latter, it would be questionable to remove these years if the purpose of the interpolation is to
characterize the field. Likely, this will have to be determined on a case-by-case basis where more
complicated fields are paid more attention than those that are simpler.
5.2 Slope and Yield Comparison
The results for the slope and yield comparison are equally interesting. Here it can generally be
seen that there was little to no change in variation or yield between slope classes. This
demonstrates that for slope values between 0 - 5° there was little to no effect on yield.

64

Considering this grade includes slopes that would normally be considered as special
conditions for an agricultural field it is safe to say that slope on its own does not significantly
affect yield. It also can be said that slope cannot be used to represent landscape changes without
additional data to supplement it. One suggestion for complementary data would be one or more
soil properties within a field since these might help to better characterize the landscape.
The two fields that should some signs of changes at the > 5° slope class, Merkt’s and
Pribnows, displayed some of the highest variability in yields. This could indicate frequent slope
changes across the field, although Stefoneks also has frequent changes in slope. This result could
mean that for some fields steep slopes could indicate drop offs in yield. It is, however, a mystery
as to why these fields in particular show this kind of result when the yields in three other fields
were relatively stable.
Overall, this section answers the second question: Are the areas where such yield
differences between the composite and a single year occur related to slope differences? The
answer is a resounding no. There were four fields where this was true and only half of these
indicated any changes in yield as slope increases and these changes only occur at very steep
slopes. This result might have occurred for many reasons. Most likely, the slope by itself could
not account for complicated landscapes where a variety of factors play significant roles.
As discussed above in Section 5.1, supplementary information could help to clarify these
relationships, but how much and what form requires further research. One important reason that
this could have failed is because this data was taken from a working farm and not from one used
for research. This is important because a farmer can see the slope on his field and decide to
account for problems associated with it by adding more manure or fertilizer in specific areas. The

65

purpose of this thesis was to look at the results from actual farms and this may have greatly
impacted the ability to answer the second question.
5.3 Final Thoughts
While the results of this thesis are mixed as far as the expected result is concerned, some
interesting results were discovered throughout the work at hand. The potential for misuse of
composite yield maps is probably the most significant. Without proper scrutiny a yield map can
give a very false impression of the field. Moreover, the general practice so far has been to
subjectively remove years that look like outliers (i.e. years that look significantly different than
the others). This thesis offers an approach to objectively determine which years could be
removed. This should be used with caution, however, as sometimes years that differ from the
others may provide valuable yield information.
Finally, slope is not a sole factor for determining problems with yield. The landscape as a
whole needs to be characterized in more detail by using more data to supplement the data
picture, and there may be some value at looking at steep slopes and determining if there is a
special relationship between yield and slopes that are both steep and highly variable.

66

REFERENCES
Behrens, T. and T. Scholten. 2006. "Digital soil mapping in Germany—a review". Journal of
Plant Nutrition and Soil Science 169, no. 3: 434-443.
Beyrer, T. A. and F. E. Below. 2014. "Improving Corn Yield Potential with Banded Phosphorus
Fertilizer." Paper presented at the international annual meeting of the SSA, CSSA, and
SSSA; Long beach, California, November 2 - 5.
Bragato, G. 2004. "Fuzzy continuous classification and spatial interpolation in conventional soil
survey for soil mapping of the lower Piave plain. Geoderma 118: 1-16.
De Bruin, S. and A. Stein. 1998. "Soil-landscape modelling using fuzzy c-means clustering of
attribute data derived from a Digital Elevation Model (DEM)." Geoderma 83, no. 1: 17-33.
Dobos, E., E. Micheli, M. F. Baumgardner, L. Biehl, and T. Helt. 2000. "Use of combined digital
elevation model and satellite radiometric data for regional soil mapping." Geoderma 97, no.
3: 367-391.
Esri. 2013. "Using LIDAR in ArcGIS." http://desktop.arcgis.com/en/arcmap/10.3/manage-
data/las-dataset/using-lidar-in-arcgis.htm. Esri (accessed 12-8-2014).
Fleming, K. L., D. G. Westfall, D. W. Wiens, and M. C. Brodahl. 2000. "Evaluating farmer
defined management zone maps for variable rate fertilizer application." Precision
Agriculture 2, no. 2: 201-215.
Fleming, K. L., D. F. Heermann, and D. G. Westfall. 2004. "Evaluating Soil Color with Farmer
Input and Apparent Soil Electrical Conductivity for Management Zone Delineation."
Agronomy Journal 96, no. 6: 1581-1587.
Fortunati, G. U., C. Banff, and M. Pasturenzi. 1994. "Soil Sampling." Fresenius' Journal of
Analtytical Chemistry 348: 86-100.
Franzen, D., F. Casey, and N. Derby. 2008. "Yield Mapping and Use of Yield Map Data." NDSU
Extension Service Site-Specific Farming 3, no. 1M (12-2008): 12-8-2014 (accessed 12-8-
2014). Yield Mapping and Use of Yield Map Data.
Fridgen, J. J., N. R. Kitchen, K. A. Sudduth, S.T. Drummond, W. J. Wiebold, and C. W. Fraisse.
2004. "Management zone analyst (MZA)." Agronomy Journal 96, no. 1: 100-108.
Hopkinson, C. and L. Chasmer. 2008. "LiDAR." In Encyclopedia of Geographic Information
Science, edited by Karen Kemp, 257-258. Thousand Oaks: SAGE Publications, Inc.
Kaspar, T. C., S. C. Thomas, D. B. Jaynes, 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, no. 1: 87-101.
Mann, K. K., A. W. Schumann, and T. A. Obreza. 2011. "Delineating productivity zones in a
citrus grove using citrus production, tree growth and temporally stable soil data.” Precision
Agriculture 12, no. 4: 457-472.
McBratney, A. B., J. J. De Gruijter, and D. J. Brus. 1992. "Spacial prediction and mapping of
continuous soil classes." Geoderma 54, no. 1: 39-64.
McBratney, A. B., M. L. Mendonça Santos, and B. Minasny. 2003. "On digital soil mapping."
Geoderma 117, no. 1: 3-52.
McBratney, A., B. Whelan, T. Ancev, and J. Bouma. 2005. "Future directions of precision
agriculture." Precision Agriculture 6, no. 1: 7-23.
Nanni, M. R. and J. A. M. Demattê. 2006. "Spectral reflectance methodology in comparison to
traditional soil analysis." Soil Science Society of America Journal 70: 393-407.

67

Ping, J. L. and A. Dobermann. 2005. "Processing of Yield Map Data.” Precision Agriculture 6:
193–212.
Schepers, A. R., J. F. Shanahan, M. A. Liebig, J. S. Schepers, S. H. Johnson, and A. Luchiari.
2004. "Appropriateness of management zones for characterizing spatial variability of soil
properties and irrigated corn yields across years." Agronomy Journal 96: 195-203.
Schieffer, J. and C. Dillon. 2013. "Precision agriculture and agro-environmental policy." In
Precision Agriculture '13, edited by John V. Stafford, 755-760. Wageningen, The
Netherlands: Wageningen Academic Publishers.
Scull, P., J. Franklin, O. A. Chadwick, and D. McArthur. 2003. "Predictive soil mapping: a
review.". Progress in Physical Geography 27: 171-197.
Shi, X., L. Girod, R. Long, R. DeKett, J. Philippe, and T. Burke. 2012. "A comparison of
LiDAR-based DEMs and USGS-sourced DEMs in terrain analysis for knowledge-based
digital soil mapping." Geoderma 170: 217-226.
Smith, M. P, A. Zhu, J. E. Burt, and C. Stiles. 2006. "The effects of DEM resolution and
neighborhood size on digital soil survey." Geoderma 137: 58-69.
Sulaeman, Y., D. Nursyamsi, L. R. Widowati, and M. S. Husnain. 2012. “Phosphorus and
Potassium Decision Support System: Bridging Soil Database and Fertilizer Application.”
http://www.agnet.org/library.php?func=view&id=20140306163320. Food and Fertilizer
Education Center (accessed: 12-10-2014).
Thompson, J. A., J. C. Bell, and C. A. Butler. 2001. "Digital elevation model resolution: effects
on terrain attribute calculation and quantitative soil-landscape modeling." Geoderma 100:
67-89.
Varvel, G., M. Schlemmer, and J. Schepers. 1999. “Relationship Between Spectral Data from an
Aerial Image and Soil Organic Matter and Phosphorus Levels.” Precision Agriculture 1:
291-300.
Wollenhaupt, N. C. and R. P. Wolkowski. 1994. "Grid Soil Sampling." Better Crips 78: 6-9.
Zhu, A.-X., B. Hudson, J. Burt, K. Lubich, and D. Simonson. 2001. "Soil mapping using GIS,
expert knowledge, and fuzzy logic.". Soil Science Society of America Journal 65: 1463-
1472.

Abstract (if available)

Abstract Precision agriculture in practice utilizes GIS far less effectively than it should. My work at a soil consulting company has shown that part of the problem is that the literature does not show an effective way of analyzing soil through GIS that is both scientific and able to be used by those associated with agriculture. The thesis aimed to answer two questions: (1) Are there significant differences between a multiple year composite yield and a single year and if so, are these significant enough to have an impact on normal operations? (2) Are the areas where such differences occur related to slope? Six fields were used for this study: Dob Along G, Emmert, Harstad, Merkt’s, Pribnows, and Stefoneks located near New Richmond, Wisconsin. Three years of yield data were used for each field and slope data was created using LiDAR from St. Croix County. These yield data were interpolated using standard industry practices. A single year was compared to composite years to determine what differences, if any, exist between them. Each year and the composite had actual and predicted values compared using the difference of means statistical test to validate the success of the interpolation. A DEM was created from LiDAR and this was used to create a slope map of each field. This slope map was used to divide the yield points by five slope classes: 0 -1°, 1-2°, 2-3°, 3-5°, and >5°. The mean yield and variation was then compared for each class to determine any patterns associated with slope values. The results show that where there is significant variation between single years of yield data the composite will fail. The difference between the composite and a single year is useful in identifying which fields are causing the composite to fail and eliminating them. Slope did not consistently correlate to changes in yield or variation. Dob Along G, Emmert, and Harstad showed no correlation, while Merkt’s, Pribnows, and Stefoneks showed decreasing yield and increasing variation as slope increased.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Beach morphodynamic change detection using LiDAR during El Niño periods in Southern California

PDF

Satellite derived bathymetry in the Canadian Archipelago using multispectral and LiDAR space-based remote sensing

PDF

Using GIS to identify potential dynamic marine protected areas: a case study using shortfin mako shark tagging data in New Zealand

PDF

Finding the green in greenspace: an examination of geospatial measures of greenspace for use in exposure studies

PDF

Modeling prehistoric paths in Bronze Age Northeast England

PDF

Questioning the cause of calamity: using remotely sensed data to assess successive fire events

PDF

Exploring remote sensing and geographic information systems technologies to understand vegetation changes in response to land management practices at Finke Gorge National Park, Australia Between ...

PDF

Geological modeling in GIS for petroleum reservoir characterization and engineering: a 3D GIS-assisted geostatistics approach

PDF

Comparing Landsat7 ETM+ and NAIP imagery for precision agriculture application in small scale farming: a case study in the south eastern part of Pittsylvania County, VA

PDF

Surface representations of rainfall at small extents: a study of rainfall mapping based on volunteered geographic information in Kona, Hawaii

PDF

Spatial and temporal patterns of long-term temperature change in Southern California from 1935 to 2014

PDF

Predicting the presence of historic and prehistoric campsites in Virginia’s Chesapeake Bay counties

PDF

Evaluating the utility of a geographic information systems-based mobility model in search and rescue operations

PDF

Classifying ash cloud attributes of Eyjafjallajökull volcano, Iceland, using satellite remote sensing

PDF

Applying least cost path analysis to search and rescue data: a case study in Yosemite National Park

PDF

Use of remotely sensed imagery to map sudden oak death (Phytophthora ramorum) in the Santa Cruz Mountains

PDF

The role of precision in spatial narratives: using a modified discourse quality index to measure the quality of deliberative spatial data

PDF

A Maxent-based model for identifying local-scale tree species richness patch boundaries in the Lake Tahoe Basin of California and Nevada

PDF

Selection of bridge location over the Merrimack River in southern New Hampshire: a comparison of site suitability assessments

PDF

Utilizing advanced spatial collection and monitoring technologies: surveying topographical datasets with unmanned aerial systems

Asset Metadata

Creator Gaumitz, Bennett Charles (author)

Core Title Precision agriculture and GIS: evaluating the use of yield maps combined with LiDAR data

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 07/11/2016

Defense Date 05/22/2016

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag composite yield. agriculture,geographic information systems,GIS,LiDAR,light detection and ranging,OAI-PMH Harvest,precision agriculture,slope,yield

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Wilson, John P. (committee chair), Kemp, Karen (committee member), Lee, Su Jin (committee member)

Creator Email gaumitz@usc.edu,gaumitz@uwalumni.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-267022

Unique identifier UC11281064

Identifier etd-GaumitzBen-4535.pdf (filename),usctheses-c40-267022 (legacy record id)

Legacy Identifier etd-GaumitzBen-4535.pdf

Dmrecord 267022

Document Type Thesis

Format application/pdf (imt)

Rights Gaumitz, Bennett Charles

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA