CROP PRODUCTION AND FUTURE CLIMATE CHANGE IN
A HIGH LATITUDE REGION: A CASE STUDY FOR THE UPPER GREAT LAKES
REGION OF THE UNITED STATES

By

Perdinan

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Geography - Doctor of Philosophy

2013

ABSTRACT
CROP PRODUCTION AND FUTURE CLIMATE CHANGE IN
A HIGH LATITUDE REGION: A CASE STUDY FOR THE UPPER GREAT LAKES
REGION OF THE UNITED STATES
By
Perdinan

Agriculture is particularly susceptible to climate change, as inferred from large historical
variations in crop production in response to past climate variability. The major goal of this
dissertation is to evaluate the spatial variability of the impacts of projected future climate change
on crop production in a high latitude region. Corn and soybean production in the Upper Great
Lakes Region (UGLR) of the United States, encompassing the states of Michigan, Wisconsin
and Minnesota, serves as the case study. The CERES-Maize and CROPGRO-Soybean models
included in the Decision Support System for Agrotechnology Transfer (DSSAT) were employed
to simulate county-level crop production for current and future time slices.
A first step in the analysis was to assign individual counties to nearby climate stations, as
climate data (e.g., temperature and precipitation) are primary drivers for the DSSAT simulations.
A climate regionalization procedure was applied to group climate stations from the United States
Historical Climate Network into subgregions with similar characteristics of the annual cycle of
temperature and precipitation. Representative stations were chosen from each climate region, and
counties were assigned to the closest representative station. The regionalization ensured that the
spatial gradients of temperature and precipitation across the UGLR were captured in the analysis.
Daily solar radiation, also an important variable for the DSSAT simulation but one which
is infrequently recorded, was calculated using a mechanistic solar radiation model parameterized
at a central site within the study area for which concurrent observations of solar radiation,

temperature and precipitation were available. This model was selected after systematically
evaluating the sensitivity of simulated corn and soybean yield to different solar radiation sources.
Climate change projections for a mid century (2041-2070) time slice were derived from
eight combinations of regional climate models and global climate models released by the North
American Climate Change Assessment Program. The impact assessment revealed that by the mid
century, assuming the current CO2 level, corn and soybean yield for the northern part of the
region likely will increase due to more favorable growing season conditions than at present;
whereas, a slight decrease in yield due to shorter time to maturity may occur in the southern part
of the UGLR, which is currently a major crop production region. However, under elevated CO2
concentrations, the number of counties with a projected decrease in yield is smaller due to the
positive impacts of higher CO2 levels, particularly for soybean production.
As the DSSAT simulations do not capture the effects of pest and disease infestations or
economic factors on crop yield, a prototype interdisciplinary model for corn yield was developed
using an asymmetric production function to integrate the DSSAT simulated yields with
economic determinants (i.e., costs of pesticides, machinery, labor and fertilizer) to estimate
observed county-level yield. The interdisciplinary model was shown to be a viable alternative to
classical yield functions, and this model framework merits future study and refinement.
In general, this study suggests that more favorable growing conditions by the mid century
will benefit the northern UGLR where some areas may produce relatively high corn and soybean
yields especially under elevated CO2 concentrations. Despite a potentially small reduction in
crop yields, current crop production in the southern UGLR will remain large, partially due to the
positive impact of elevated CO2 concentrations.

ACKNOWLEDGMENTS

I came to Michigan State University to pursue my doctoral degree under a Fulbright
Presidential Scholarship. This dissertation would never have been completed without generous
support, tremendous effort, and patient guidance from my major adviser, Prof. Julie A. Winkler
who also provided financial support in the form of a research assistantship after my Fulbright
Scholarship ended. I also express my deep appreciation to my guidance committee, Prof. Jeffry
Andresen, Prof. Sharon Zhong, Prof. Kurt Thelen and Prof. Zhengfei Guan for their helpful
suggestions to improve my dissertation research.
I also thank Dr. Gopal Alagarswamy for his advice on the DSSAT simulation, and to Mr.
Wilson Ndovie and Mr. Jim Brown for their help with computational need. Ms. Melba Lacey,
Ms. Claudia Brown, and Ms. Sharon Ruggles were always there when I needed help to comply
with administrative procedures.
I thank my geography colleagues, specifically members of the climate group, for making
the Department Geography at Michigan State University a pleasant place to study. To all my
Indonesian friends, my thanks for our togetherness that made me feel at home.
Finally, behind each achievement of our lives, there is always strong support from our
beloved family. I dedicate this work to my lovely mother “Faizah Lis” and father “Rakiso”, my
beautiful wife “Melani Indah Lestari”, my sweet daughter “Ghania Aurasya Perdinan”, my funny
son “Daanish Airel Perdinan”, and all my brothers (Perdamaian and Budi Rakiso) and sisters
(Nensi Rakiso, Lisky Safrealia and Ririn Septyaningsih), who always gave me inspiration and
motivation to complete my studies. Without their support and love, I do not know how I would
pass through the many difficult phases of my study.
East Lansing, May 2013

iv

TABLE OF CONTENTS

LIST OF TABLES ..................................................................................................................... viii
LIST OF FIGURES ..................................................................................................................... ix
CHAPTER 1. Background and Research Objectives ..................................................................... 1
1.1. Background ..................................................................................................................... 1
1.2. Research Objectives........................................................................................................ 7
1.3. Study Region................................................................................................................... 8
1.4. Dissertation Structure.................................................................................................... 12
CHAPTER 2. Selection of Climate Information for Regional Climate Change Assessments using
Regionalization Techniques: a Case Study for the Upper Great Lakes Region, USA ........ 14
2.1. Introduction................................................................................................................... 14
2.2. Data Methods ................................................................................................................ 16
2.2.1. Climate Data ..................................................................................................... 16
2.2.2. Regionalization Procedures .............................................................................. 17
2.3. Results........................................................................................................................... 20
2.3.1. Climate regionalization for 1971-2000............................................................. 20
2.3.2. Differences between the climate regions .......................................................... 22
2.3.3. Changes in climate regions with time............................................................... 24
2.3.4. Assignment of political units to climatic regions and selection of representative
climate stations.................................................................................................. 28
2.4. Discussion ..................................................................................................................... 29
2.4.1. Interpretation of climate regions and boundaries.............................................. 29
2.4.2. Non-contiguous climate regions ....................................................................... 30
2.4.3. Temporal changes in regionalization patterns .................................................. 32
2.4.4. Potential contributions of climate regionalization to impact studies ................ 32
2.5. Conclusions................................................................................................................... 35
Acknowledgements.............................................................................................................. 36
CHAPTER 3. Traditional versus Modern Approaches of Estimating Daily Solar Radiation for
Input to Crop Process Models.............................................................................................. 37
3.1. Introduction................................................................................................................... 37
3.2. Materials and Methods.................................................................................................. 42
3.2.1. Climate Observations and Study Period ........................................................... 42
3.2.2. Daily Solar Radiation Estimates ....................................................................... 44
(a) Stochastic Generation................................................................................. 44
(b) Empirical Model ......................................................................................... 45
(c) Mechanistic Model...................................................................................... 46
(d) POWER ....................................................................................................... 48
(e) NARR and NARCCAP................................................................................. 49
3.2.3. Crop Yield Simulations..................................................................................... 50

v

3.2.4. Evaluation Methods .......................................................................................... 51
3.3. Results........................................................................................................................... 53
3.3.1. Solar Radiation.................................................................................................. 53
3.3.2. Crop Yields ....................................................................................................... 60
3.4. Discussion ..................................................................................................................... 67
3.5. Conclusions................................................................................................................... 76
Acknowledgements.............................................................................................................. 77
CHAPTER 4. Spatial Variability of Regional Climate Change Impacts on Crop Production
in High Latitude Regions: A Case Study of the Upper Great Lakes Region of the United
States .................................................................................................................................... 78
4.1. Introduction................................................................................................................... 78
4.2. Data and Methods ......................................................................................................... 83
4.2.1. Regional Crop Model Simulation ..................................................................... 83
(a) Historical Climate Observations ................................................................. 84
(b) Soil data ...................................................................................................... 88
(c) Cropping Management................................................................................ 89
4.2.2. Climate Change Scenarios ................................................................................ 92
4.2.3. Evaluation of CERES-Maize and CROPGRO-Soybean .................................. 95
4.2.4. Regional Impact Assessment of Projected Climate Change............................. 97
4.3. Results........................................................................................................................... 98
4.3.1. Historical Simulation ........................................................................................ 98
4.3.2. Future Climate Impacts................................................................................... 102
Planting Dates and Time to Maturity.............................................................. 102
Evapotranspiration and Crop Yield ................................................................ 106
4.3.3. Carbon Fertilization Effects............................................................................ 110
4.3.4. Cultivar Sensitivity ......................................................................................... 115
4.3.5. Coefficient of Variation .................................................................................. 117
4.3.6. Potential Expansion in Major Growing Areas ................................................ 121
4.4. Summary and Discussion............................................................................................ 124
4.4.1. Spatial Variation of the Climate Change Impact ............................................ 124
4.4.2. Future Yield Variability and the Potential Expansion in Production Region. 128
4.4.3. Limitations ...................................................................................................... 129
4.5. Conclusion .................................................................................................................. 130
CHAPTER 5. Development of Crop Yield Interdisciplinary Model for Regional Climate
Change Impact Assessments: A Case Study for the Upper Great Lakes Region of the
United States ...................................................................................................................... 133
5.1. Introduction................................................................................................................. 133
5.2. Materials and Methods................................................................................................ 136
5.2.1. Study Area ...................................................................................................... 136
5.2.2. Model Specification ........................................................................................ 137
5.2.3. Model Implementation.................................................................................... 139
5.2.4. Economic Data................................................................................................ 140
5.2.5. Observed Yield ............................................................................................... 142
5.2.6. Crop Model Simulation................................................................................... 143

vi

(a) Climate data .............................................................................................. 143
(b) Soil Data.................................................................................................... 145
(c) Cropping Scenarios................................................................................... 146
(d) DSSAT Simulated Attainable Yield ......................................................... 147
5.2.7. Model Estimation............................................................................................ 148
5.2.8. Application for a Climate Change Impact Assessment .................................. 149
5.3. Results......................................................................................................................... 150
5.4. Summary and Discussion............................................................................................ 154
5.5. Conclusion .................................................................................................................. 157
CHAPTER 6. Summary and Conclusion.................................................................................... 159
BIBLIOGRAPHY....................................................................................................................... 165

vii

LIST OF TABLES

Table 3-1. Geographic location and analysis period for the CRN stations located in the states of
Michigan and Minnesota ............................................................................................. 48
Table 3-2. Farming management for the DSSAT simulation ....................................................... 51
Table 3-3. Percent difference by month in the mean and standard deviation of daily solar
radiation for the different radiation sources compared to observed radiation at
Hancock, Wisconsin. ................................................................................................... 58
Table 3-4. Percent difference in the mean and standard deviation of simulated maize and soybean
yields using estimated versus observed daily solar radiation as input, and the
correlation between the yield time series for the study period. Two-tailed probabilities
are shown in parenthesis for paired t-test for equality of means and F-test for equality
of variances.................................................................................................................. 66
Table 3-5. Strengths and limitations of traditional and non-traditional sources of daily solar
radiation estimates. ...................................................................................................... 73
Table 4-1. Cultivar characteristic of the medium season corn cultivar used in the DSSAT
simulations................................................................................................................... 90
Table 4-2. The NARCCAP model combinations employed for the study ................................... 92
Table 5-1. Characteristics of medium season corn cultivar used in CERES-Maize simulations 146
Table 5-2. Model coefficients and elasticity analysis for the interdisciplinary model of countylevel corn yield for the UGLR parameterized using available data of 1997, 2002
and 2007 .................................................................................................................... 151
Table 5-3. Statistical comparison between observed yields (Obs) and estimated yield (Est)
obtained from DSSAT simulation and the Interdisciplinary model across the
UGLR ........................................................................................................................ 153

viii

LIST OF FIGURES

Figure 1-1. The study area within the Midwest region of the United States. [“For interpretation of
the references to color in this and all other figures, the reader is referred to the
electronic version of this dissertation.”] ...................................................................... 9
Figure 1-2. Average and standard deviation of corn (top figures) and soybean (bottom
figures) yields over the UGLR from 1942 to 2008. Data source: National
Agriculture Statistics Service - United States Department of Agriculture
(NASS-USDA, 2011) ................................................................................................ 10
Figure 1-3. Trends of corn (top) and soybean (bottom) yields in the UGLR from 1942
to 2008. ...................................................................................................................... 11
Figure 2-1. Distribution of the USHCN climate stations used in the climate regionalizaiton.
The UGLR study area is defined as the states of Minnesota, Wisconsin, and
Michigan .................................................................................................................... 17
Figure 2-2. Time periods used for the climate regionalization..................................................... 18
Figure 2-3. Determination of the number of clusters based on graphical evaluation of the
slope of the relationship between the sum of squares of within cluster
distances (SSWCD) and the number of clusters. The two arrows indicate the
7 and 21 cluster solutions........................................................................................... 21
Figure 2-4. The seven (top) and twenty one (bottom) cluster solutions. The climate regions for
the 7-cluster solution have arbitrarily been assigned numbers for reference in the text.
For the 21-cluster solution, each region is assigned a letter to help readers distinguish
between clusters and more easily identify non-contiguous clusters. ......................... 22
Figure 2-5. Deviation of mean monthly precipitation (top), maximum temperature (Tmax,
center), and minimum temperature (Tmin, bottom) for each climate region from the
average for the study area. The line symbols refer to cluster number; see Figure 2-4
for location of each cluster. Climate regions 1, 2, and 3 are displayed in the left-hand
panels, and climate regions 4, 5, 6, and 7 are shown in the right-hand panels .......... 25
Figure 2-6. Application of a seven cluster solution for three periods, 1911-1940 (top), 1941-1970
(middle), and 1971-2000 (bottom)............................................................................. 27
Figure 2-7. Assignment of counties in the study area to climate clusters..................................... 28

ix

Figure 2-8. The representative climate stations selected for each climate cluster........................ 29
Figure 3-1. Location of the Hancock WAWN station and of the Necedah CRN station used to
evaluate the quality of the solar radiation time series at Hancock. The additional
CRN climate stations in Michigan and Minnesota were used for a sensitivity analysis
of the coefficients of the mechanistic and empirical models..................................... 43
Figure 3-2. Daily averages of estimated (black line) and observed (grey line) solar radiation at
Hancock-Wisconsin for the 8-year study period for the weather generator (GEN),
empirical (EMP) and mechanistic (MEC) models, POWER, NARR, and the
NARCCAP (CRCM, ECP2, HRM3, and WRFG) regional climate models. ............ 54
Figure 3-3. Average daily bias for the solar radiation estimates. See Fig. 2 for definition of the
abbreviations. ............................................................................................................. 55
Figure 3-4. Root mean squared error (RMSE) and the correlation with observed daily
solar radiation for the solar radiation estimates. See Fig. 2 for definition of
abbreviations .............................................................................................................. 56
Figure 3-5. Evaluation of the performance of the mechanistic (left) and empirical (right) models
parameterized at Hancock when applied to four Climate Reference Network stations
in Minnesota (Sandstone and Goodrich) and Michigan (Chatham and Gaylord) based
2
on root mean square error (RMSE) and coefficient of determination (R ) ............... 59
Figure 3-6. Maize yields for the eight-year analysis period simulated using daily solar radiation
observations and estimates as input to CERES-Maize. OSR and ESR refer to
observed and estimated solar radiation, respectively................................................. 61
Figure 3-7. Soybean yields for the eight-year analysis period simulated using daily solar
radiation observations and estimates as input to CROPGRO-Soybean. OSR and ESR
refer to observed and estimated solar radiation, respectively. ................................... 62
Figure 3-8. Mean squared deviation (MSD), squared bias (SB), squared difference between
standard deviation (SDSD) and the lack of correlation weighted by the standard
deviation (LCS) for maize (left) and soybean (right) yield simulated using daily solar
radiation estimates compared to yield simulations using radiation observations.
Please note the difference in the vertical axes for the two plots. See text for definition
of SB, SDSD, LCS, and MSD. .................................................................................. 64
Figure 4-1. Geographical location of the Upper Great Lakes Region .......................................... 80
Figure 4-2. Selection of representative climate stations for the regional climate change impact
assessment on corn and soybean production in the UGLR based on an objective
climate regionalization (top) and their assignment to counties (bottom). Each station
is referenced by a four character identifier and the colors in the lower panel show the
counties assigned to the same representative climate station. ................................... 87

x

Figure 4-3. Projected changes in °C for simulated average maximum temperature (left) and
minimum temperature (middle), and percent change in precipitation (right) for the
growing season (March-October) between the future (2041-2070) and the control
(1971-2000) periods for the eight NARCCAP models. Letters of A-H are the
NARCCAP model identifiers (Table 4-2). ................................................................ 94
Figure 4-4. The median value of observed and DSSAT simulated yields for corn (left) and
soybean (right). The yield ratio was calculated by dividing simulated yield by
observed yield for each county. Available data refers to the numbers of years with
reported yield for each county. Note: The scale for median yield is different for corn
and soybean................................................................................................................ 96
Figure 4-5. Median values of simulated planting date (PDAT), time to maturity (TMAT),
cumulative seasonal evapotranspiration (ET), and grain yield (Yield) for corn (top
four panels) and soybean (bottom four panels) in the UGLR for the historical period
(1971-2000) ............................................................................................................. 101
Figure 4-6. Projected change in the median of planting date (PDAT, above) and time to
maturity (TMAT, below) between the historical (1971-2000) and the future
(2041-2070) period for corn production in the UGLR at the reference level of
CO2 (370 ppm) concentration. Letters of A-H are the NARCCAP model
identifiers (Table 4-2) .............................................................................................. 103
Figure 4-7. As Figure 4-6, but for soybean production .............................................................. 105
Figure 4-8. Change in the median of cumulative seasonal evapotranspiration (ET, above)
and crop yields (Yield, below) between the historical (1971-2000) and the future
(2041-2070) period for corn production in the UGLR at the reference level of
CO2 (370 ppm) concentration. Letters of A-H are the NARCCAP model
identifiers (Table 4-2) .............................................................................................. 107
Figure 4-9. As Figure 4-8, but for soybean production .............................................................. 109
Figure 4-10. Change in median planting date (PDAT), time to maturity (TMAT), seasonal
evapotranspiration (ET) and crop yields (Yield) between the historical (1971-2000)
and the future (2041-2070) period averaged across all climate model combinations
for corn production in the UGLR under elevated CO2 concentration of 490 ppm
(left) and 635 ppm (right) ........................................................................................ 111
Figure 4-11. As Figure 4-10, but for soybean production .......................................................... 112

xi

Figure 4-12. Percent change of projected seasonal evapotranspiration (ET) and crop yield
(Yield) for the future period (2041-2070) under elevated CO2 concentration of
490 ppm (490-370, left) and 635 ppm (635-370, right) for corn (above) and
soybean (below) with respect to projected ET and Yield for the future period
(2041-2070) under the reference level of CO2 concentration (370 ppm). ET and
Yield are obtained by averaging estimated future ET and Yield across all
NARCCAP climate models. .................................................................................. 114
Figure 4-13. Change in the median value of corn yield between the future (2041-2070) and the
historical period (1971-2000) for short season (Short) and long season cultivar
(Long) under different climate change projections and elevated CO2 of 635 ppm.
The median values of simulated yield for the historical period for each cultivar are
presented in the first row as a reference. Letters of A-H are the NARCCAP model
identifiers (Table 4-2). ............................................................................................. 116
Figure 4-14. Ratio of coefficient of variation (COV Ratio) of corn yield for the future period
(2041-2070) with respect to the historical period (1971-2000) under different climate
projections derived from NARCCAP models at the reference CO2 (370 ppm) level.
Average is the COV ratio averaged across all climate projections. Below two panels
are the average COV Ratio under elevated CO2 of 490 (left) ppm and 635 ppm
(right). Letters of A-H are the NARCCAP model identifiers (Table 4-2)............... 119
Figure 4-15. As Figure 4-14, but for soybean yield.................................................................... 120
Figure 4-16. Potential expansion in major corn production regions of the UGLR assessed
by categorizing the median simulated corn yields for the future period
(2041-2070) is equal to or higher than a given yield threshold, namely:
5000 kg/ha (first row), 6000 (second row), 7000 (third row), 8000 (fourth row),
and 9000 (fifth row). ................................................................................................ 122
Figure 4-17. As Figure 4-16, but for soybean............................................................................. 123
Figure 5-1. Location of the Upper Great Lakes Region ............................................................. 136
Figure 5-2. Components of the interdisciplinary model ............................................................. 139
Figure 5-3. The United States farm resource regions used to calculate the share of production
costs for corn. Data Source: the Economic Research Service – United States
Department of Agriculture (ERS-USDA, 2011)...................................................... 141
Figure 5-4. The representative climate stations (red dots) assigned to the UGLR counties for the
interdisciplinary model development. The colors indicate the counties assigned with
the same representative climate station. .................................................................. 144

xii

Figure 5-5. Observed yield (top), DSSAT simulated yield (middle), and the ratio of simulated to
observed yield (bottom) for the UGLR for 1997, 2002 and 2007 ........................... 148

Figure 5-6. Comparison between observed county-level yields (Obs), estimated yields obtained
from DSSAT simulation (DSSAT) and estimated yields obtained from the
interdisciplinary model (INT) for 1997, 2002 and 2007. The three maps in the last
row (Data) display county-level data employed for the model development.......... 152
Figure 5-7. Sensitivity of yield changes under projected climate change for the UGLR by the mid
century (2041-2070) to the reduction in the amount of chemical application by 30%
(second row), 50% (third row) and 70% (fourth row). A CO2 concentration of 370
ppm was assumed for the DSSAT simulations........................................................ 154

xiii

CHAPTER 1.
Background and Research Objectives
1.1.

Background
Recent studies (e.g., Gilland, 2002; Jaggard et al., 2010; Tilman et al., 2011) raise as a

serious challenge the ability of global food supply to meet the demand for food consumption in
2050, as human population is expected to increase to about nine billion. Climate change,
including changes in the mean climatic state and changes in variability, intensifies the challenge
of future world food supply (Rosenzweig and Parry, 1994; Schmidhuber and Tubiello, 2007).
Globally, climate change likely will adversely impact crop productivity in low latitude countries,
whereas high latitude countries may benefit (Cline, 2007; Parry et al., 2005). The adverse
impacts of climate change are also projected to increase the risk of hunger in developing
countries (Parry et al., 2005; Schmidhuber and Tubiello, 2007).
Given the perspective of climate change as a global threat to human livelihood (Barnett,
2010; Wilby et al., 2009), the estimated potential benefits of climate change to crop productivity
in high latitude regions is an interesting subject. Potential benefits are expected because
increased precipitation in high latitude regions (IPCC, 2007) may offset greater
evapotranspiration arising from future higher temperatures, leading to more favorable conditions
for crop production such as in the high latitude regions of North America (Fischer et al., 2005).
However, this generalization needs further exploration as the impacts of climate change on crop
productivity in high latitude regions likely is regionally and even locally specific (e.g., Brassard
and Singh, 2008; Southworth et al., 2002; Southworth et al., 2000). Regional variations in
climate change impacts are anticipated because changes in temperature and precipitation will

1

vary spatially, as illustrated by the impact of recent (1976-2006) climate trends on county-level
corn and soybean yield in Wisconsin of the United States (Kucharik and Serbin, 2008).
This study will evaluate spatial variability of future climate change impacts on countylevel corn and soybean production in counties located within the Upper Great Lakes Region of
the United States, encompassing the states of Michigan, Wisconsin and Minnesota. Corn and
soybean, the two most commonly grown crops in the region (Hatfield, 2012; Niyogi and Mishra,
2013), were chosen to represent the two major crop types, i.e., C4 (corn) and C3 (soybean),
which are expected to respond differently to climate change (Mera et al., 2006). Although
agricultural production is considered a major economic contributor for the region, the spatial
detail of potential future climate change impacts on crop production rarely has been studied for
the region.
Previous studies for the study area (Andresen et al., 2000; Southworth et al., 2002; 2000)
selected a few representative locations to evaluate the potential impact of future climate change
on corn and soybean production. Ten sites (Southworth et al., 2000) and nine sites (Southworth
et al., 2002), located across the states of Indiana, Illinois, Ohio, Michigan, and Wisconsin, were
selected to evaluate the consequences of projected climate change for a future period (20502059) relative to the baseline period (1961-1990) on corn and soybean yields. Southworth et al.
(2000) found specifically for locations in Michigan and Wisconsin, that future climate change
may have either a positive or negative impact on grain yield of long and medium season corn
varieties, depending on the location within these two states and/or climate scenarios; whereas,
the projected impact was negative for a short season corn variety for all locations and climate
scenarios. For example, simulated corn yields of long-season variety in the Michigan Thumb and
eastern Wisconsin were projected to increase up to 50% and 40%, respectively, regardless of

2

climate scenario. In contrast, grain yields of long season variety for south-central Michigan and
southwest Wisconsin were projected to change from 20% to -20% and 20% to -10%,
respectively, with the differences in sign corresponding to different climate scenarios. In these
two locations, the short season variety was projected to decrease by up to 30% and 50%.
Furthermore, Southworth et al. (2002) found that future climate change is likely to boost
soybean yields, especially for late-maturing soybean cultivars. They estimated that the yield of latematuring soybean cultivars may increase up to 120% in south-central Michigan and the
Michigan Thumb region. For mid-maturing cultivars, yields were projected to increase by 5% or
decrease by 25% in southwestern Wisconsin, depending on climate scenarios; whereas, soybean
productivity in eastern Wisconsin, south-central Michigan, and the Michigan Thumb were
projected to increase by up to 60%. The largest increase in yields for the early maturing cultivar
were expected in south-central Michigan where yields could possibly increase up to 120%.
Overall, the increase in yields for late-maturing variety was expected to be greater than that for
early maturing cultivars across the study locations (Southworth et al., 2002).
However, an assessment specifically for the UGLR gives somewhat different projections
compared to those for the wider Midwest region. Andresen et al. (2000) did not project a future
decrease in corn and soybean yields for the ULGR region; rather, they reported that yields in the
period of 2000-2099 for 13 locations within the UGLR are likely to significantly increase
compared to yields for historical records (1896-1996). This increase was attributed to the
positive impacts of CO2 enrichment on crop yield, particularly for soybean (Andresen et al.,
2000). A similar carbon fertilization effect was also found by Southworth et al. (2002).
Although, the conclusions drawn from the above studies may be appropriate for
estimating the impacts of climate change at particular locations, generalizing the conclusions

3

across the region could be problematic, thus further exploration is needed to identify areas with
decreasing or increasing crop production under future climate change and elevated atmospheric
CO2 concentrations. In addition, the limited locations employed in the previous analyses also
prevent the use of their results to explore potential latitudinal shifts of major crop production
regions to the north that is frequently highlighted by national-scale impact studies for the United
States (e.g., Thomson et al., 2005; Tubiello et al., 2002).
A major challenge of regional climate change impact assessments at the county level, the
scale chosen in this study, is data availability. Ecophysiological models, commonly known as
crop models (White and Hoogenboom, 2011), often have been employed for climate change
impact assessments in recent decades as reviewed by White et al. (2011a), require extensive data
inputs (e.g., climate data, soil information and farming practices) for the model simulations.
Unfortunately, the required data inputs are not available for each county in the UGLR.
Specifically for climate data, daily solar radiation, one of the main climate variables which is
needed for crop model simulations, is infrequently recorded compared to temperature and
precipitation (Liu and Scott, 2001).
Crop models are regulallry employed in climate change impact assessments because they
are able to simulate the non-linear interactions among plants, environment and farming practices
at a specific location (Hoogenboom et al., 2004; Meinke et al., 2001). However, crop models
generally poorly simulate the impacts of weeds and pest and disease infestations (Soussana et al.,
2010; Tubiello et al., 2002), and can not capture the contribution of economic determinants to
yield variability (Kaufmann and Snell, 1997; Vera-Diaz et al., 2008). The inclusion of economic
stressors has been recommended by the climate impact research community (Challinor et al.,
2009), recognizing the contribution of economic factors to yield variability (Cabas et al., 2010;

4

Kaufmann and Snell, 1997). Socio-economic conditions can influence farmers’ decisions in
managing their land and selecting farming practices (de Koeijer et al., 1999). Guan et al. (2006)
assert that economic factors such as capital and labor can create favorable growing conditions for
crop production.
A potential approach proposed to overcome the limitation of crop models is to develop
yield models that combine the outputs of crop models and economic determinants (i.e, ‘a hybrid
model’), such as the model proposed by Kaufmann and Snell (1997). This hybrid model, referred
by Kaufmann and Shell as a “interdisciplinary model”, offers an advantage over the direct use of
crop models by capturing, at least in part, the influence of socio-economic drivers on crop yield
fluctuations. In comparison to traditional empirical yield models (e.g., Almaraz et al., 2008;
Cabas et al., 2010; Tannura et al., 2008), the interdisciplinary model better captures non-linear
interactions between weather variables, environmental conditions and crop growth processes,
which are often oversimplified in empirical yield models (Challinor et al., 2009; Soussana et al.,
2010). A recent application is the development by Vera-Diaz et al. (2008) of an interdisciplinary
model for soybean yield that combines simulated yields from a crop simulation model,
geographic location (i.e., latitude and longitude) and economic variables (i.e., credit, transports
costs) to explain soybean yield variability in the Brazilian Amazon.
Yet, none of the earlier interdisciplinary models include economic determinants
associated with chemicals to control pest and diseases, capital, and labor. A recent study
conducted by Guan et al. (2006) proposed an empirical yield model based on an asymmetric
framework to distinguish the impacts of growth inputs (i.e., land, seed, environmental factors)
and facilitating inputs (i.e., pesticides, capital and labor) on attainable and actual yield
(vanIttersum and Rabbinge, 1997), respectively. Unfortunately, the utilization of dummy

5

variables to reflect environmental conditions (e.g., soil and climate) limits the application of the
model for climate change assessments. However, the model structure proposed by Guan et al.
(2006) can provide an alternative framework for the development of a new type of
interdisciplinary model. A crop model can be used to simulate crop yields (i.e., attainable yield)
based on agricultural growth inputs and agronomic practices (i.e. cultivar, row and spacing, and
planting dates), and economic determinants are used to adjust simulated yields in order to obtain
estimated yields (i.e., actual yield).
The recent release of future climate projections for the mid century (2041-2070) from an
ensemble of regional climate models (RCMs) driven by several global climate models (GCMs),
included in the North American Regional Climate Change Assessment Program (Mearns et al.,
2009), offers an opportunity to utilize higher resolution climate model outputs for the
development of climate change scenarios. Many studies (e.g., Frei et al., 2003; Kim et al., 2008;
Kjellstrom et al., 2010) argued that the finer resolution of RCMs offers an advantage in
simulating details of regional climate compared to the corresponding lateral boundary condition.
Furthermore, the extensive effort of downscaling outputs of several GCMs to a regional scale
using multiple RCMs, known as dynamical downscaling (Fowler et al., 2007), is rarely applied
for regional climate change impact assessments because dynamical downscaling is
computationally expensive (Fowler et al., 2007; Wilby and Wigley, 1997).
This dissertation attempts to address the above challenges to better understand the spatial
variation of future climate change impacts on crop production in high latitude regions by using
the Upper Great Lakes Region of the United States as a case study. This study offers a
methodology to prepare required data for crop model simulation at the county-level.
Additionally, alternative daily solar radiation sources are assessed to choose appropriate daily

6

solar radiation estimates as inputs for crop model simulations. Climate change scenarios derived
from the NARCCAP datasets are employed to evaluate the potential regional impacts of
projected climate change for the mid century on crop production in the study region. An
interdisciplinary model for corn yield is developed to explore the contribution of economic
determinants to explain the deviation between simulated and observed yields.
The outcomes of this research contribute to the enhancement of methods for conducting
regional climate change assessments, especially at the county scale. The impact assessment
conducted for the study region also enhances our understanding of the spatial variation of future
climate change impacts on crop production in high latitude regions. The relatively detailed
spatial analysis of the potential impacts allows potential latitudinal shifts of crop production
region under exposure of future climate change to be examined. The interdisciplinary model
provides an alternative approach for evaluating the possible consequences of combining climate
and economic exposures to crop yield variability. Finally, the outcomes of this research will help
decision makers in the region to devise agricultural management strategies for climate change
adaptation.

1.2.

Research Objectives
The main goal of this dissertation is to evaluate the spatial variability of future climate

change impacts on corn and soybean production in the Upper Great Lakes Region (UGLR) of the
United States. Specifically, the first goal is to develop an objectively-defined climate
regionalization for the region that is proposed to prepare climate data for crop model simulations
at the county scale. The second goal is to evaluate the sensitivity of simulated corn and soybean
yields to different daily solar radiation estimates in order to choose appropriate daily solar

7

radiation estimates for crop model simulations employed for climate change impact assessments.
The third goal is to identify the spatial distribution of future climate impacts on corn and soybean
production over the UGLR. The fourth goal is to develop an interdisciplinary model for corn
yield that can be employed for regional climate change assessments. This dissertation addresses
each objective in separate chapters. The objectives and methods will be explained in more detail
in each chapter.
1.3.

Study Region
The UGLR is comprised of the states of Michigan, Wisconsin and Minnesota (Safir et al.,

2008). This region is located in the United States (US) Midwest (Pryor and Barthelmie, 2013),
which is considered the ‘main heartland’ of agriculture in the US (Figure 1-1). Corn and
soybean, the crops selected for this study, are considered the two major agricultural commodities
in the region (Hatfield, 2012; Niyogi and Mishra, 2013).
The major growing areas of corn and soybean are located mostly in the southern part of
the UGLR (Safir et al., 2008). County averages of corn and soybean yield from 1942 to 2008 are
larger in the southern part of the region (Figure 1-2). In these areas, the variability of corn and
soybean yields is also highest. Corn and soybean yields in each of the three states display a
positive trend since 1942 (Figure 1-3). Corn yield in Minnesota increased about 108 kg/ha per
year; whereas, in Michigan and Wisconsin, yield increased about 91 kg/ha per year. Trends of
soybean yields for Wisconsin, Minnesota and Michigan were 31.5, 29.7, and 26.5 kg/ha per year,
respectively. Improvements in agricultural management and technology and favorable weather
over the period are considered the key factors contributing to increased corn and soybean yields
in the region (Andresen et al., 2001; Tannura et al., 2008).

8

ND

MN

SD

WI

MI

IA

NE

IL

KS

MO

IN OH
KY

2
Study Area
Midwest Region

Figure 1-1. The study area within the Midwest region of the United States. [“For interpretation of
the references to color in this and all other figures, the reader is referred to the electronic version
of this dissertation.”]

9

Corn

Yield StDev (ton/ha)
0.709 - 1.464
1.464 - 1.843
1.843 - 2.130
2.130 - 2.437
2.437 - 2.780
NA

Yield Average (ton/ha)
2.238 - 3.534
3.534 - 4.382
4.382 - 5.103
5.103 - 5.757
5.757 - 6.624
NA

Soybean

Yield Average (ton/ha)
0.841 - 1.154
1.154 - 1.444
1.445 - 1.674
1.674 - 1.889
1.889 - 2.189
NA

Yield StDev (ton/ha)
0.039 - 0.238
0.238 - 0.437
0.437 - 0.620
0.620 - 0.739
0.739 - 0.938
NA

Figure 1-2. Average and standard deviation of corn (top figures) and soybean (bottom figures)
yields over the UGLR from 1942 to 2008. Data source: National Agriculture Statistics Service United States Department of Agriculture (NASS-USDA, 2011)

10

Anomaly (ton/ha)

6.0

yMN = 0.1075x - 3.6545
R2 = 0.8751

4.0
yMI = 0.0911x - 3.0961

2.0

R2 = 0.9231

0.0

-2.0

yWI = 0.091x - 3.0935
R2 = 0.8902

-4.0

2008

2002

1996

1990

1984

1978

1972

1966

1960

1954

1948

1942
Anomaly (ton/ha)

1.5
1.0

yMI = 0.0265x - 0.8993

0.5

R2 = 0.8632

0.0
-0.5

yMN = 0.0297x - 1.009

-1.0

R2 = 0.8506

yWI = 0.0315x - 1.0724
R2 = 0.8453

-1.5

2008

2002

1996

1990

MN
Linear (MN)

1984

1978

1972

1966

1960

1954

1948

1942

MI
Linear (WI)

WI
Linear (MI)

Figure 1-3. Trends of corn (top) and soybean (bottom) yields in the UGLR from 1942 to 2008.

11

1.4.

Dissertation Structure
This dissertation is divided into six chapters. The first chapter (Chapter 1) provides a

background of the research, the research objectives, and the study region.
Chapter 2 discusses the climate regionalization of the study region. Climate stations
included in the United States Historical Climate Networks (USHCN) for the UGLR and
neighboring states were grouped using a combination of principal component analysis (PCA) and
non-hierarchical (kmeans) clustering method. Euclidian allocation and distance was employed to
create clustering boundaries and assign a specific climate group to counties, respectively.
Chapter 3 evaluates the sensitivity of crop models to six different daily solar radiation
estimations, i.e., point-based estimates (empirical, dynamical, and weather generator), and
gridded datasets (NASA-POWER, NARR, and NARCCAP). Statistical analyses, e.g., t-test,
mean absolute error (MAE), root means square error (RMSE), correlation coefficient
2

(r)/coefficient determination (R ), and mean squared deviation (MSD), were applied to assess the
performance of the alternative radiation sources and their impacts on corn and soybean yields.
Chapter 4 elaborates the spatial variation of future climate impacts on corn and soybean
production across counties in the UGLR. Representative climate stations were selected and
assigned to each county by considering the climate groups for the UGLR (Chapter 2). Soil
parameters for each county were obtained from the STATGO database. Faming practices were
set at the rainfed condition with multiple planting windows to optimally start planting crop
across the UGLR counties. Climate change projections of precipitation, maximum and minimum
temperature were derived from eight RCM-gcm combinations available from the NARCCAP
database.

12

Chapter 5 discusses the development of an interdisciplinary model for county-level corn
yield in the UGLR. This model was developed by relating the ratio of observed to simulated corn
yield (predictand) with economic data (predictors) at the county level. Observed corn yields were
detrended in order to be comparable with simulated yields. Economic data (i.e., total costs of
chemical, machinery, labor and fertilizer) for the model development were obtained from the
agricultural census for 1997, 2002, and 2007 and adjusted using a corresponding price index for
each economic variable. The model parameters were estimated using nonlinear least square
(NLS) estimator. The contribution of each economic variable to the dependent variable was
evaluated using elasticity analysis.
Chapter 6 summaries findings from Chapter 2 to Chapter 5. Implications of the research
outcomes and their contribution to enhancing knowledge about regional climate change impacts
also are discussed.

13

CHAPTER 2.
Selection of Climate Information for Regional Climate Change Assessments using
Regionalization Techniques: a Case Study for the Upper Great Lakes Region, USA

In collaboration with
Julie A. Winkler

2.1.

Introduction
The impacts of climate change, such as those for crop production, are regionally (e.g.,

Cline, 2007; Motha and Baier, 2005) and even locally (e.g., Goldblum, 2009; Kucharik and
Serbin, 2008) specific. Consequently, most climate change assessments are conducted at the
regional scale (Carter et al., 2007). A common approach in regional-scale assessment studies is
to select a modest number of locations that hopefully capture within-region spatial variations of
the current climate and potential climate change impacts (e.g., Brassard and Singh, 2008;
Southworth et al., 2002; Southworth et al., 2000). This approach is in part dictated by the need
for fine-scale climate scenarios for regional climate change assessments. As pointed out by
Winkler et al. (2011), climate scenario development, which typically consists of homogeneity
testing of historical time series of climate observations, application of models (dynamic and/or
empirical) to “downscale” coarse resolution output from global climate models to the regional or
local scale, and extensive evaluation of the downscaled variables, is time consuming and
resource intensive and often a “road block” for an impact assessment. Thus, for many assessment
studies, developing scenarios for a large number of locations within a region may not be an
option (Winkler et al., 2002).

14

When selecting locations for an assessment, previous studies typically have considered
the length of the historical time series or the perceived representativeness of climate stations
within a region (e.g., Almaraz et al., 2008; Andresen et al., 2000). A concern of the first selection
criterion is that the stations with the longest records may not adequately capture the spatial
variability of climate across the study area, whereas the second criterion requires that the size of
the subregion for which a station is representative be clarified and that the characteristics used to
evaluate representativeness be specified. These concerns illustrate the need for innovative
approaches for selecting climate stations for climate change assessments.
We propose and demonstrate that an objectively-defined climate regionalization can be
useful when selecting the number and location of representative stations with the goal of
capturing the spatial variability of a region’s climate. Furthermore, the proposed approach
provides an indication of the needed resources for climate scenario development. For the
illustrative application described below, a combination of principal component analysis (PCA)
and hierarchical and non-hierarchical clustering techniques is employed to group climate stations
from the United States Historical Climatology Network (Menne et al., 2009) in the Upper Great
Lakes region (UGLR) of the United States. The regionalization is based on mean monthly
maximum and minimum temperature and precipitation. These three climate variables were
chosen so that the regionalization would be applicable to a wide range of assessments, although
the choice of climate variables for a particular application should be dependent on the assessment
goals. The regionalization was performed separately for three different periods, 1971-2000,
1941-1970, and 1911-1940 to investigate the sensitivity of the climate groups to the choice of
time period for which the regionalization is conducted. To help visualize the spatial variability in
climate, cluster boundaries were defined using Euclidean allocation (ESRI, 2008). In addition,

15

Euclidian distance was employed to link the cluster regions and site specific information from
representative climate stations for each region with aggregated datasets (e.g., agricultural,
biological, social-economic) at the United States county level (USDA-NASS, 2007). The
regionalization and linkages to political units provide a starting point and framework for further
assessment activities, including climate scenario development, incorporation of additional
datasets and multiple aggregation levels into an assessment, and the use and development of
decision-making models.
2.2.

Data Methods

2.2.1. Climate Data
Record length and the quality of climate observations are critical considerations in a
climate regionalization. For this reason, climate observations from the United States Historical
Climatology Network (USHCN) ver. 2.0 (Menne et al., 2009) were used. Each station within this
network is generally considered to have a relatively long time series and the highest quality
temperature and precipitation observations (Menne et al., 2009) compared to other stations
within the United States Cooperative Observer Program (COOP) Network (Daly et al., 2007).
This research utilized 180 climate stations distributed over the study area (Michigan, Wisconsin
and Minnesota) and neighboring states (Figure 2-1). Stations from neighboring states were
included to minimize edge effects that can reduce the accuracy of clustering results (e.g.,
DeGaetano, 1996; DeGaetano, 2001; Fovell and Fovell, 1993; Stooksbury and Michaels, 1991).
Monthly mean maximum and minimum temperature adjusted for time of observation bias and
other inhomogeneities and total monthly precipitation were extracted from the USHCN database
for the period 1911-2000. The monthly time series were used to calculate climatic normals for
the three 30-year time periods, 1971-2000, 1941-1970, and 1911-1940. Temperature and

16

precipitation were selected for analysis as they are the two most commonly utilized variables for
climate regionalization (e.g., Bunkers et al., 1996; DeGaetano, 2001; Fovell and Fovell, 1993).

Figure 2-1. Distribution of the USHCN climate stations used in the climate regionalizaiton. The
UGLR study area is defined as the states of Minnesota, Wisconsin, and Michigan.

2.2.2. Regionalization Procedures
A combination of PCA and a two step clustering process (hierarchical and nonhierarchical clustering) was used for the climate regionalization. These techniques have been
widely applied to understand the structure of climate datasets, as summarized by Jolliffe and
Philipp (2010). The climate regionalization was initially performed for the period 1971-2000
(Figure 2-2). Separate regionalizations were conducted for two earlier periods, 1941-1970 and
1911-1940, although for comparison the number of clusters initially obtained for the 1971-2000
period was held constant for all three regionalizations. The latest period is preferable as a
reference, under the assumption it exhibits the current climatic conditions for the region.
17

Regionalization #3
Regionalization #2
Regionalization #1
1911

1941

1971

2000

Figure 2-2. Time periods used for the climate regionalization.

PCA was used to reduce data dimensionality (Kalkstein et al., 1987) and to remove
multicolleniarity among variables (Bohm et al., 2001). The 30-year monthly maximum and
minimum temperature and precipitation averages were initially standardized to zero mean and
unit variance to remove the influence of different physical units on the PCA performance
(Fovell and Fovell, 1993; e.g., Stooksbury and Michaels, 1991; Wilks, 1995). The PCA was then
applied to the correlation matrix calculated from the 36 climate variables (columns) and 180
locations (rows). A Varimax orthogonal rotation (Kaiser, 1959), which has been widely
employed in climate research (e.g., Bohm et al., 2001; Dommenget and Latif, 2002; Fovell and
Fovell, 1993; Richman, 1986), was applied to simplify interpretation. Only rotated components
with at least one loading >0.5 were retained for the cluster analysis following Guentchev and
Winkler (2010), resulting in four retained components explaining approximately 93% of the
variation.
A two-step clustering process was utilized to take advantage of the strengths of both
hierarchical and non-hierarchical clustering. Hierarchical clustering was used to provide an
estimate of the possible number of clusters in a dataset. The k-means partitioning technique, a
non-hierarchical method, was used to assign objects to clusters. Unlike traditional hierarchical
clustering, k-means analysis permits the reassignment of objects until a convergent criterion is
achieved, but requires that the initial number of clusters be specified at the beginning of the

18

clustering process (Dezfuli, 2011; Jolliffe and Philipp, 2010; Rhee et al., 2008; Stooksbury and
Michaels, 1991). For the hierarchical clustering, the scores of the rotated components were
grouped into different clusters using Euclidian distance as the similarity measurement and
various linkage functions (McQuity, median, average, single, complete, centroid, and Ward), all
of which are regularly used for climate regionalization (e.g., Bunkers et al., 1996; Fovell and
Fovell, 1993; Kalkstein et al., 1987; Stooksbury and Michaels, 1991). The component scores
were not standardized preceding the cluster analysis, as such standardization may cause the
distances between observations to be unrealistic (Johnson, 1998). The number of candidate
clusters was evaluated using the Sarle Cubic Clustering Criterion (CCC) (SAS Institute Inc.,
2004), Pseudo-F and Pseudo-t2 statistics (Fovell and Fovell, 1993), and distances between
clusters (Fovell and Fovell, 1993; Wilks, 1995). Candidate numbers of clusters were chosen by
searching for a breakpoint in the plots of these indices by merger level. Two breakpoints were
selected for each criterion to provide a range of possible clusters for input to the k-means
analysis.
The non-hierarchical k-means cluster analysis was then performed using each candidate
number of clusters obtained from the hierarchical clustering. A drift option was specified to
update the cluster seeds during the partitioning processes to allow for adjustments of the seeds
every time an observation was reassigned (SAS Institute Inc., 2004). The final number of
clusters was determined based on Beale’s Pseudo-F criterion (Johnson, 1998) and a graphical
evaluation of the sum of squares of the within cluster distances. Beale’s Pseudo-F was calculated
from the intracluster sum of squares and compared to critical values of the F-distribution to
assess whether a solution with additional clusters offers a significant improvement over a
solution with a smaller number of clusters.

19

To help depict the spatial distribution of climate groups across the study region, cluster
boundaries were defined objectively using Euclidian allocation, available in ArcGIS (ESRI,
2008). The Euclidean allocation method is preferable over other interpolation techniques when
the data being interpolated (in this case cluster membership) are nominal rather than
interval/ratio measurements. The feature class of cluster memberships for the climate stations
2

was converted to a raster grid of approximate size of 1 km . For cells without climate stations,
the Euclidean distance to the closest source cell (i.e., cell with a climate station) was calculated
and the cluster membership of the source cell was assigned to the cell without a climate station.
In effect, the “edges” of the raster cells falling in different clusters form the climate region
boundaries, although the boundaries were identified within ArcGIS by converting the “filled in”
raster grid to polygons. Euclidean distance was also used to assign cluster regions to counties.
In this case, the Euclidean distance of all surrounding climate stations to the centroid of each
county was calculated, and the county was assigned to the climate cluster of the nearest climate
station.
2.3.

Results

2.3.1. Climate regionalization for 1971-2000
The hierarchical clustering analyses resulted in a wide range of possible number of
clusters as input for the k-means non-hierarchical clustering. As shown in Table 1, the number of
clusters considered were 4-17, 21, 24, 26-28, 33, 35-37, 41-42, 45-46, 51, 59, and 67. The final
number of clusters was determined based on Beale’s Pseudo-F criterion (Johnson, 1998) and a
graphical evaluation of the first and second breakpoints of the sum of squares of the within

20

cluster distances. Beale’s Pseudo-F, suggested a 7 cluster solution, whereas either 7 or 21
clusters was suggested by the graphical evaluation (Figure 2-3).
The climate regionalization patterns are generally similar for the 7 and 21 cluster
solutions (Figure 2-4). The 21-cluster solution captures more local characteristics, as indicated
by the larger number, but smaller in size, clusters, especially in the lake-modified areas
surrounding the Great Lakes and along the study area borders. The latter may be a reflection of
“edge effects” on the analysis, in spite of the use of climate stations outside of the UGLR in the
cluster analysis. The 7-cluster solution suggests that there is greater spatial variability in regional
climate in Michigan and eastern Wisconsin where several non-contiguous clusters are evident.
Both solutions suggest a small number of large, broad climate regions extending from northwest
Minnesota to central Wisconsin. The discussion below focuses on the 7-cluster solution, given
the overall similarity of the solutions and the Beale’s Pseudo-F test result that the 21-cluster
solution is not a statistically significant improvement over the 7-cluster solution.
400
350

SSWCD

300
250
200
150
100
50
0
0

10

20

30

40

50

60

70

Number of Cluster
Figure 2-3. Determination of the number of clusters based on graphical evaluation of the slope of
the relationship between the sum of squares of within cluster distances (SSWCD) and the
number of clusters. The two arrows indicate the 7 and 21 cluster solutions.
21

5
3

5

5

7

1

7

5

2

6
4

2

6
7

7

6
7

6
7

6

B

C

D

F

E

K

E
G

K
K

B

Q
H

A
A

F
H

I
L

M

P

P

I
J

P
O

O

P

P

S

N
O

T
O R N S

Figure 2-4. The seven (top) and twenty one (bottom) cluster solutions. The climate regions for
the 7-cluster solution have arbitrarily been assigned numbers for reference in the text. For the 21cluster solution, each region is assigned a letter to help readers distinguish between clusters and
more easily identify non-contiguous clusters.

2.3.2. Differences between the climate regions
The differences between the climate regions were evaluated through a comparison of the
deviations of the mean monthly precipitation and maximum and minimum temperature for each
climate region from the average calculated across all climate regions, referred to as the UGLR

22

average. Starting with precipitation, western Minnesota (labeled Cluster 1 in Figure 2-4) is drier
throughout the year compared to the other climate regions (Figure 2-5). In contrast, mean
precipitation for Cluster 4 extending from southern Minnesota to south central Wisconsin is
above the UGLR average for almost all months of the year. Wintertime precipitation is higher
compared to other regions in non-contiguous Clusters 7 and 6, both located in Michigan and
eastern Wisconsin, whereas spring and early summer precipitation is lower than the regional
average. The seasonal variations are largest for Cluster 7. Distinct seasonality is evident in the
deviation of mean monthly precipitation for Clusters 2, 3, and 5. For Cluster 2, a relatively small
non-contiguous region in southwestern Minnesota, mean monthly precipitation is below the
UGLR average for all months except May. In contrast, a positive deviation in autumn
precipitation is evident for Cluster 5, located in lake effect precipitation-influenced regions of
northeastern Minnesota, northern Wisconsin, and the Upper Peninsula of Michigan. For Cluster
3, extending from central Minnesota to central Wisconsin, mean monthly precipitation is above
the UGLR average from June to October but below the large-scale average for the remainder of
the year.
Distinct differences are also seen between clusters in terms of mean monthly maximum
and minimum temperature. Small, non-contiguous Cluster 2 in western Minnesota has a
relatively higher mean maximum temperature compared to the other climate regions, although
minimum temperature is warmer than that of other regions only in summer. From winter
through early spring, the highest mean minimum temperatures in the study area are found in
Cluster 6 in eastern Wisconsin and Michigan. Cluster 5 in northern Wisconsin and Michigan’s
Upper Peninsula has lower maximum and minimum temperatures during most of the year
compared to the other climate regions with the exception of winter when temperatures are lower

23

in Clusters 1 and 3 found in Minnesota. In Cluster 7, a lake-modified region, mean maximum
temperature is above the UGLR average from November-February but below the large-scale
average for the remainder of the year. This pattern differs from that seen for Cluster 6, another
lake-modified region, where monthly mean maximum temperatures are either similar to or
warmer than the UGLR average throughout the year. Average minimum temperatures for Cluster
2 are above the UGLR average for all months, and are generally similar to the mean
temperatures observed in Cluster 6. Cluster 4 is unique in that mean monthly maximum and
minimum temperature are close to the UGLR averages throughout the year.

2.3.3. Changes in climate regions with time
The application of cluster analysis to the other time periods shows substantial changes in
the climate regions between the early- (1911-1940) and mid- (1941-1970) century time periods,
whereas the cluster pattern for the mid-century period is similar to that for the late-century
(1971-2000) period (Figure 2-6). Notable differences with time in the climate regions are found
in the Lower Peninsula of Michigan, which was grouped into a single contiguous cluster for the
early time period rather than the fragmented non-contiguous clusters seen for the later time
periods. Substantial differences are also found in Minnesota. With time the large climate region
seen in western Minnesota in the early time period retreats westward, whereas in northeastern
Minnesota and northern Wisconsin, the climate regions coalesce during the mid-century time
period into a fairly large region. This larger region is separated again during the late-century time
period into two broad regions, one in northeastern and western Wisconsin and the other in
northern Wisconsin.

24

30
20
10
0
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
4
2
0
−2
−4
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

−6
6

6

Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

−6

−4

−2

0

Tmax Deviation (0C)

2

4

Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

−30 −20 −10

Precipitation Deviation (mm)

30
20
10
0
−30 −20 −10

Precipitation Deviation (mm)
Tmax Deviation (0C)

Figure 2-5. Deviation of mean monthly precipitation (top), maximum temperature (Tmax,
center), and minimum temperature (Tmin, bottom) for each climate region from the average for
the study area. The line symbols refer to cluster number; see Figure 2-4 for location of each
cluster. Climate regions 1, 2, and 3 are displayed in the left-hand panels, and climate regions 4,
5, 6, and 7 are shown in the right-hand panels.

25

1
2
−6

3

26
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

−6

−2

0

2

−4

−2

0

2

Tmin Deviation (0C)

−4

Tmin Deviation (0C)

4

4

6

6

Figure 2-5. (cont’d)

4
5
6
7

1911-1940

1941-1970

1971-2000

Figure 2-6. Application of a seven cluster solution for three periods, 1911-1940 (top), 1941-1970
(middle), and 1971-2000 (bottom).

27

2.3.4. Assignment of political units to climatic regions and selection of representative climate
stations
One of our goals was to link climate observations with other variables required for
climate change impact studies that are available at different spatial scales and aggregation levels.
For this purpose, we assigned counties to climate group memberships based on the Euclidean
distance between the county centroid and the nearest climate station (Figure 2-7). This
assignment offers an alternative to link climate information from a climate station with other
datasets such as economic data which are available at the county level.

Climate Region
1

4

6

2

5

7

3

Figure 2-7. Assignment of counties in the study area to climate clusters.

Furthermore, as discussed in the introduction, a key motivation of the climate
regionalization was to assist in selecting a small number of representative climate stations for a
climate impact assessment. For this example, two climate stations were selected for each cluster
that was contiguous in space, and one station was selected in each of the separate areas for those
climate clusters that are non-contiguous, with preference given to stations that were close to the
midpoints of each region (Figure 2-8). The underlying consideration in our choice of

28

representative stations was uniform spatial coverage across the study area; however, the number
of representative stations per cluster and considerations in their selection should be influenced by
the intended application.

Figure 2-8. The representative climate stations selected for each climate cluster.

2.4.

Discussion
In the discussion below, we consider several issues related to the application of the

procedures described above; namely, considerations when interpreting climate regions and
boundaries, the interpretation of non-contiguous climate regions, fluctuations in climate regions
with time, and potential contributions of the proposed methods for climate change studies.

2.4.1. Interpretation of climate regions and boundaries
For any application, care must be taken to not over interpret objectively-defined climate
regions and boundaries. Although the combined approach of PCA and hierarchical and nonhierarchical clustering used here has been widely employed in previous regionalizations (e.g.,
Gong and Richman, 1995; Stooksbury and Michaels, 1991; Winkler, 1992), alternative
29

classification approaches exist and may result in differing results particularly if the clusters are
not well nucleated (Everitt, 1980). Additionally, several previous analyses (e.g. Alijani et al.,
2008; Briggs and Lemin Jr, 1992; DeGaetano, 1996) have combined methods for summarizing
climate observations, such as PCA, with spatial interpolation techniques to define climate
boundaries rather than delineating boundaries based on cluster membership as was used here.
Whatever the method used, the ensuing regionalization will be influenced by the initial density
of the climate stations, and users must take this into account for any application. A relatively
coarse station network was employed for the UGLR regionalization presented here. This was a
reasonable choice given that an intended initial application is an assessment of climate impacts
on corn and soybean production; both crops are broadly grown across the region. Other
applications may require that a denser network of climate observations is used. Euclidian
allocation within ArcGIS is a unique approach for delineating climate region boundaries and
provides a convenient means for visualizing climate groups. However, it does not take into
account abrupt climate gradients such as those observed in areas of complex terrain and that are
not captured by the climate station density. Alternative means of delineating cluster boundaries
should be considered in these situations.

2.4.2. Non-contiguous climate regions
Several of the UGLR climate regions shown above are not spatially contiguous. Similar
segmentation of climate regions was found in earlier studies that grouped climate stations in the
southeastern United States (Stooksbury and Michaels, 1991), the state of Maine in the United
States (Briggs and Lemin Jr, 1992), central and eastern North America (Gong and Richman,
1995), and the northeastern United States (DeGaetano, 1996). Several previous authors have
considered non-contiguous climate regions an artifact of aggregation error or observation biases,

30

under the assumption that near objects should have stronger correlations than distant objects.
Although k-mean clustering is generally considered better in retaining spatial coherency than
hierarchical clustering (Gong and Richman, 1995), Whitfield et al. (2002) proposed that k-means
clustering results are sensitive to the temporal aggregation level of the climate variables and
found that decreasing the temporal aggregation increased, particularly for precipitation, the
proportion of near climate stations that were clustered into the same climate group. Assignment
of geographically-close climate stations into different climate regions may also be due to
differences in local site characteristics (e.g., exposure) or observational biases (e.g., time of
observation bias) (Briggs and Lemin Jr, 1992). On the other hand, several authors have argued
that spatial contiguity is not a requirement for creating climatologically homogeneous regions
(e.g., Fovell and Fovell, 1993) and found that nearby climate stations may exhibit different
precipitation or temperature patterns (Yeh et al., 2000). Furthermore, regional climatic
conditions are generally controlled by three basic forcings, i.e., latitude, elevation or topography,
and distance to water bodies (e.g. DeGaetano, 1996; Rhee et al., 2008), and non-contiguous
climate regions can reflect spatial variations in these forces.
For the UGLR climate regionalization, we did not force the solution into coterminous
regions in contrast to several earlier regionalizations (e.g. Stooksbury and Michaels, 1991; Yeh

et al., 2000), but rather allowed for the segmentation of climate regions. One consideration in
this decision was our use of climate observations from the USHCN ver. 2.0 database in the
regionalization. Most of the USHCN stations are located in rural areas, and the monthly mean
maximum and minimum temperatures (but not precipitation) are adjusted for time of observation
bias (Menne et al., 2009). Although this dataset is still imperfect, its use does reduce the
contribution of observational bias to the segmentation of climate regions. The primary

31

consideration, however, was that most of the non-contiguous climate regions appeared along the
lake-modified zones of Michigan and eastern Wisconsin are likely due to the influence of the
lakes on local climate. Similar segmentation of climate groups located along lake zones was
observed by DeGaetano (1996) in a climate regionalization for the northeastern United States.

2.4.3. Temporal changes in regionalization patterns
An interesting finding was the substantial changes in the climate regionalization between
the 1911-1940 and 1941-1970 periods. Differences in the extent and location of the climate
regions were more modest between 1941-1970 and 1971-2000. These changes may in part result
from heterogeneities in the time series of temperature and precipitation observations.
Alternatively, the changes in the climate regionalization may reflect spatial differences in
climatic trends. In this case, the spatial distributions of climate groups may be useful for
detecting climate change in the region (Jacobeit, 2010) and can provide an alternative to
traditional trend analysis at individual stations (Pielke et al., 2000). Considerable further
analysis is needed to better understand the reasons behind the differences in regionalization
between the three time periods for the UGLR. We suspect, however, that these changes are
partly due to climatic trends. The UGLR lies at the northern edge of the Midwest “warming
hole” (Kunkel et al., 2006), with a cooling trend reported in southern portion of the UGLR and a
warming trend in the northern portion (Strode, 2003). These differences in trends may contribute
to the observed changes with time in the climate clusters.

2.4.4. Potential contributions of climate regionalization to impact studies
One advantage of objective regionalization techniques is that they avoid the use of
political boundaries to define climate regions such as what is done for National Weather Service

32

(NWS) climate divisions (Guttman and Quayle, 1996). Although frequently used, NWS climate
divisions may not be relevant for summarizing regional climate conditions (DeGaetano, 2001)
for climate impact studies. In fact, Guttman and Quayle (1996) acknowledge that climate
division boundaries, which are delineated based on drainage basins in the western states and crop
reporting districts that generally overlap with county boundaries in the eastern states, often do
not have a direct relation to climate. Heterogeneous climate conditions may exist within climatic
divisions that can cause difficulties when using climate data from the divisions for climate
applications (Rhee et al., 2008). Additionally, objective climate regionalization provides more
flexibility compared to pre-defined climate regions such as the NWS climate divisions, as
climate regions can be defined in terms of the climate variables of interest to a specific
application. Although, the climate variables employed for the UGLR regionalization shown here
are commonly-used mean monthly temperature and precipitation, climate regions can be defined
based on any number of derived parameters such as heat accumulation units, the frequency of
days above or below threshold temperatures, precipitation rates, or the timing of critical events
(e.g., date of spring freeze). For example, Shinker (2010) included the ratio of monthly-to-annual
precipitation to identify regions with similar annual cycles of precipitation occurrence in the
western United States.
A primary contribution of regionalization approaches to climate impact studies is that
they can provide an understanding of what climate type is presented by an individual station
(Pielke et al., 2002). Thus, they are useful in selecting the number and location of climate
stations to include in an impact assessment and help to ensure that regional climate variations
across the study area are captured. The selection of a limited number of representative climate
stations is helpful for reducing the workload preparing the inputs needed for climate change

33

assessments including the development of fine-scale climate scenarios. Additionally, a climate
regionalization can assist in the evaluation of regional climate model (RCM) simulations such
that the RCM performance is evaluated for the different climate types within a study area. The
selection of representative stations can also assist in the formulation of bias correction functions
that are commonly applied to RCM outputs (e.g., Rivington et al., 2008a) before they are used
for climate applications. Formulating bias corrections for all stations within a region is time
consuming, and, depending on the RCM grid resolution, climate observations are usually not
available for all RCM grid cells. In this situation, RCM grid cells can be related to a climate
group so that the bias correction developed for that group (i.e., for a representative station of the
group) can be applied to those cells for which climate observations do not exist.
As demonstrated above, climate regionalization can also be used to help link different
datasets that are required for climate change assessments but that have varying spatial resolutions
and aggregation levels. The linkage can be performed by overlying the climate regionalization
with the aggregated data, for example county-level socio-economic information, or by assigning
the nearest climate group to an aggregation unit based on Euclidian distance or some other
measure. Using the first approach, a particular aggregation unit may overlap two or more climate
groups. The second approach, which was chosen in this study, is useful to uniquely link the
aggregated information to a climate region. For the application above, we focused on linking
information at the United States county-level to the climate regions. Rhee et al. (2008) also
suggest the use of a county as an aggregation unit because county-level analysis is very useful
for planning and management strategies and many datasets, particularly economic data .that are
useful for decision making, are available at this level.

34

2.5.

Conclusions
This study attempted to demonstrate, employing the Upper Great Lakes region (UGLR)

of the United States as an example, the usefulness of climate regionalization for impact
assessments, specifically for the selection of representative climate stations for a region and to
assign those stations to different aggregation levels. Using a combination of principal
components analysis and clustering methods, the UGLR, encompassing Michigan, Wisconsin
and Minnesota, was grouped into seven climate regions based on average monthly precipitation
and maximum and minimum temperature. The seven cluster solution is able to distinguish spatial
variability of climatic conditions over the study area, as indicated by the different annual
variations of the climate variables for each cluster. Euclidian allocation, which is readily
available in the widely used ArcGIS software, was employed to define cluster boundaries.
Application of the regionalization procedures to three different 30-year time periods
indicates that the greatest changes in the distribution of the climate types occurred between the
early (1911-1940) and mid (1941-1970) century, and that the climate patterns were relatively
similar for the mid and late (1971-2000) century. This alteration of cluster memberships can be
employed to provide insights into regional climate change.
The climate regionalization for the late century time period was used to select
representative climate stations for future analysis and to link the climate observations with
county-level aggregated datasets. The value of selecting the representative stations based on a
climate regionalization is that the number of climate stations needed for the assessment, is
substantially reduced, but at the same time the spatial variability in climate is represented in the
analysis. The results of this study are useful for climate change studies within the ULGR region,
and the methods are applicable for other regions.

35

Acknowledgements
The funding for this study was provided by the Fulbright Presidential Scholarship and
NSF CNH grant 0909378.

36

CHAPTER 3.
Traditional versus Modern Approaches of Estimating Daily Solar Radiation for Input to
Crop Process Models

In collaboration with
Julie A. Winkler and Jeffrey A. Andresen

3.1.

Introduction
Crop simulation models are useful for estimating crop biophysical processes under

different environmental conditions, including climate variability and change, as well as
simulating the potential impacts of changes in agricultural management practices on yields.
Daily solar radiation, which is used in conjunction with precipitation and air temperature, is
considered a critical input to crop models for estimating evapotranspiration, water stress, plant
biomass production, and yield (Hoogenboom, 2000; Jones et al., 2003). However, the availability
of solar radiation observations is a major concern for crop model applications, as this variable is
infrequently measured compared to temperature and precipitation (Liu and Scott, 2001).
In the United States, solar radiation measurements are not included in the observations
taken by the two primary national in situ observing networks (i.e., the Automated Surface
Observing System (ASOS) and the Cooperative Observer Program (COOP)). Solar radiation is
observed by the recently-established Climate Reference Network (CRN). However, in addition
to the short period in operation, this network consists of only 114 stations (Leduc et al., 2009),
most of which are located outside of primary agricultural regions. Additional radiation

37

measurements are taken by specialized networks, such as those maintained by universities and
agricultural extension services, but coverage is non-uniform and quality control standards vary.
To illustrate the dearth of radiation measurements in the United States, Wilcox et al. (2007)
reported that of about 1445 meteorological stations archived in the National Solar Radiation
Database (NSRDB), only 40 stations measured solar radiation. The coarse spatial resolution of
daily solar radiation observations (when available) is also a concern as estimating radiation from
surrounding, but often distant, stations may not be appropriate (e.g., Hunt et al., 1998; Rivington
et al., 2006).
Given these challenges, a number of alternative approaches have been developed to
estimate daily solar radiation at a location. In general, daily solar radiation has been estimated
using stochastic, mechanistic, or empirical methods. Of these, stochastic methods have been
employed more frequently, especially for assessments of climate impacts on crop production
(e.g.,Apipattanavis et al., 2010; Kilsby et al., 2007) and possible adaptation options (e.g., Luo et
al., 2009; Meza and Silva, 2009). Stochastic methods include the use of weather generators such
as WGEN (Richardson and Wright, 1984) and SIMMETEO (Geng et al., 1986) to generate
multiple daily time series. The stochastically-generated series attempt to mimic the statistical
characteristics of a long-term time series of solar radiation at a particular site (Garcia and
Hoogenboom, 2005; Woli and Paz, 2012) rather than the value for a particular day. Typically,
the estimated daily solar radiation total is selected randomly from a distribution conditioned on a
sequence of wet and dry days. The representation of the day-to-day interrelationships between
solar radiation and observed values of other weather variables (e.g. temperature) is a concern,
particularly as preserving the interactions among weather variables is important for crop models
(Rivington et al., 2005).

38

In contrast, mechanistic and empirical approaches provide greater synchrony between
estimated daily solar radiation and observed daily values of other weather variables. Mechanistic
models predict site-specific daily incoming solar radiation (S) from daily extra-terrestrial
radiation (S0) at a location and other observations for that day such as sunshine duration (e.g.,
Yang et al., 2006), temperature (Bristow and Campbell, 1984; Weiss et al., 2001) and/or a
combination of temperature and precipitation (Hunt et al., 1998; Thornton et al., 2000; Thornton
and Running, 1999). S0 is calculated based on latitude, day of year, solar angle and solar
constant (Hunt et al., 1998; Liu and Scott, 2001). On the other hand, empirical models use
regression techniques to estimate solar radiation at a location from daily weather variables, such
as precipitation and temperature (e.g., Ball et al., 2004). Ideally, mechanistic and empirical
models should be parameterized for an individual location, which limits their application for
locations without solar radiation observations (Abraha and Savage, 2008). Liu and Scott (2001)
argue that these models can be applied to locations with a similar regional climate as the site for
which the models were initially developed, but defining “similar” climates can be challenging
particularly for areas with large topographic variations.
As the different methods for estimating solar radiation can introduce biases that may
significantly impact the outcomes of crop models (Nonhebel, 1994; Weiss et al., 2001), a number
of previous studies have evaluated the sensitivity of crop model simulations to generated solar
radiation from weather generators (e.g., Cooter and Dhakhwa, 1996; Garcia et al., 2008) and
mechanistic models (e.g., Abraha and Savage, 2008; Trnka et al., 2007). Cooter and Dhakhwa
(1996) found that interannual variability of crop yields is relatively sensitive to different sources
of generated solar radiation, even though the long-term average of crop yield is not. In line with
this finding, concern has been raised on utilizing generated solar radiation when forecasting

39

seasonal yields (Trnka et al., 2007). The sensitivity of crop model outputs (e.g.,
evapotranspiration and total biomass) to generated versus observed solar radiation also appears
to vary from location to location (Abraha and Savage, 2008). Although generated solar radiation
is still considered by many researchers to be a viable option for crop model applications when
observed solar radiation is unavailable (Cooter and Dhakhwa, 1996; Garcia et al., 2008), further
research is needed to explore other possible sources of daily solar radiation and evaluate
associated potential biases.
Recently available alternative sources of daily solar radiation time series include
“reanalysis” datasets that blend output from atmospheric models with observations, simulations
from regional climate models, and satellite estimations. Two potential sources of daily solar
radiation estimates for North America are the fine scale (32 km) North American Regional
Reanalysis (NARR) datasets (Mesinger et al., 2006), available from 1979 to present, and the 50
km resolution regional climate model simulations from the North American Regional Climate
Change Assessment Program (NARCCAP) (Mearns et al., 2009), available for 1979-2004.
Additionally, the National Aeronautical and Space Agency (NASA) Prediction of Worldwide
Energy Resource (POWER) database (Stackhouse, 2006), which was developed with agricultural
uses in mind, provides daily averaged values of daily solar radiation at a spatial resolution of one
degree latitude/longitude for the period 1983 to present based on satellite estimations.
Implementation of these alternative datasets for crop model applications would provide a
substantial advantage over mechanistic or empirical approaches as no further model development
or parameterization is required, and, unlike stochastic approaches, the interrelationships with
observed daily weather variables are maintained. The NARCCAP simulations have an additional
advantage in that projections of daily solar radiation for a future period (2041-2070) are also

40

available, facilitating analyses of potential climate change impacts and adaptation options.
However, in spite of the potential utility of these alternative datasets, only a few studies have
assessed the potential biases in the radiation estimates obtained from the POWER dataset and
their impact on crop simulations (e.g., Bai et al., 2011; White et al., 2008; White et al., 2011b),
and potential biases remain largely unaddressed for the radiation estimates from NARR and
NARCCAP.
The objectives of this study are to 1) compare observed solar radiation to estimated daily
solar radiation from stochastic, empirical, and mechanistic models (traditional approaches), and
the POWER, NARR, and NARCCAP datasets (modern approaches), and 2) assess the
sensitivity of simulated grain yield of maize and soybean to the different sources of daily solar
radiation inputs. The inclusion of the NARR and NARCCAP gridded datasets distinguishes this
study from recent work that investigated the impacts of solar radiation estimates obtained from
stochastic, mechanistic and empirical models (e.g., Abraha and Savage, 2008; Garcia et al.,
2008) and satellite estimates (e.g., Bai et al., 2011; White et al., 2011b) on crop model
applications. The analysis was performed for Hancock, Wisconsin, located within an agricultural
region in the Upper Great Lakes region of the United States. Maize and soybean were selected
because they are the two most widely grown crops in the region (Hatfield, 2012) and more
importantly represent the two major crop types, namely C4 (maize) and C3 (soybean) plants.
Each type has different light saturation thresholds and photosynthesis mechanisms which can
cause them to response differently to solar radiation as discussed by Garcia et al. (2008).

41

3.2.

Materials and Methods

3.2.1. Climate Observations and Study Period
Daily solar radiation observations for the evaluation of the different radiation estimates
along with the temperature and precipitation measurements used as input to the stochastic,
empirical and mechanistic models were obtained from the Wisconsin Automated Weather
Network (WAWN) station at Hancock, Wisconsin (Figure 3-1). This station was selected
because of its 1) relatively long period of record (1985 to present) and 2) geographic proximity
to a station (Necedah, Wisconsin) within the Climate Reference Network (CRN). Observations
from the CRN station were used to evaluate the quality of the solar radiation measurements at
Hancock, assuming that little difference should exist in the radiation time series given the similar
elevation and climate of the two locations. The equality of the means of the radiation time series
was assessed for the overlapping period of 1 October 2004 to 31 December 2010 using both
unpaired t-tests (assuming unequal variance) and paired t-tests. The two-sample F-test was used
to test for equality of variances, and Pearson’s correlation coefficient was calculated to measure
the association between the radiation observations at the two stations. Means and variances of
the solar radiation time series were not significantly different at the 95% probability level and the
two time series were highly (0.97) correlated. These analyses suggest that the solar radiation
observations recorded at the Hancock WAWN station are consistent with those from the CRN
reference station.
The study period was constrained by the period of overlap for the different radiation
estimates and observations from 1985 to 2000. Unfortunately, there were numerous missing
observations for the Hancock WAWN station prior to 1989 so that we further evaluated data
availability from 1990-2000. Within this period, a large number of missing observations

42

occurred at the Hancock WAWN in 1991, 1999 and 2000 with 184, 100, and 84 missing days,
respectively, when at least one climate variable (i.e., radiation, temperature, precipitation) was
missing. These years were removed, and, as a result, only 1990 and 1992-1998 were included in
the analysis for a total of eight years. Although relatively short, an eight-year period was also
used in a recent evaluation of the performance of several mechanistic models and a variation of
WGEN to estimate daily solar radiation for the southeastern United States (Woli and Paz, 2012).

0

0

0

0

0

0

0

0
0
0
0
0
0
0
0
-97° -96° -95° -94° -93 -92 -91 -90 -89 -88 -87 -86 -85° -84° -83°
-97 -96 -95 -94 -93° -92° -91° -90° -89° -88° -87° -86° -85 -84 -83
0

N

49°
49
0

48°
48

Goodridge

0
48°
48

0

0
47°
47

47
47°
0

46
46°

0
49°
49

Chatham

Sandstone

0

46°
46

0

45°
45

Hancock

0

44°
44

Gaylord

Necedah

0
45°
45
0
44°
44

43°
43

0

0
43°
43

0

0
42°
42

42
42°
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
-97° -96° -95° -94° -93° -92° -91° -90° -89° -88° -87° -86° -85° -84° -83°
-97 -96 -95 -94 -93 -92 -91 -90 -89 -88 -87 -86 -85 -84 -83

Figure 3-1. Location of the Hancock WAWN station and of the Necedah CRN station used to
evaluate the quality of the solar radiation time series at Hancock. The additional CRN climate
stations in Michigan and Minnesota were used for a sensitivity analysis of the coefficients of the
mechanistic and empirical models.

43

The limited number of missing precipitation and temperature observations that occurred
in 1990 and 1992-1998 were filled in with observations from the United States Historical
Climate Network (USHCN) Hancock station (Menne et al., 2009). The missing daily solar
radiation observations were filled in with daily averages of the non-missing observations. For the
development of the empirical and mechanistic models, measurements for the eight-year study
period were divided two groups. Data from 1990, 1992, 1994, and 1997 were used for model
calibration, and the remaining data were saved for model validation. Four additional CRN
climate stations (Goodridge, Minnesota; Sandstone, Minnesota; Chatham, Michigan; Gaylord,
Michigan) were employed to evaluate the sensitivity of the coefficients of the mechanistic and
empirical models.

3.2.2. Daily Solar Radiation Estimates
Daily solar radiation is defined here as the daily mean downward shortwave radiation
flux. Estimated daily solar radiation was obtained from six different sources, namely a stochastic
weather generator, a mechanistic model, an empirical equation, POWER, NARR, and
NARCCAP. All daily solar radiation estimates were converted to mega-joule per meter squared
-2

-1

(MJ m day ) for comparison and for input to the crop simulation model.

(a)

Stochastic Generation
Solar radiation was stochastically generated using a weather data management program

included in DSSAT called “Weatherman” (Pickering et al., 1994). As noted by Mavromatis and
Hansen (2001), Weatherman is a variant of the well-known WGEN (Richardson and Wright,
1984) weather generator. In Weatherman, all underlying model parameters are estimated on a
monthly basis. Daily values of the model parameters are calculated internally using linear

44

interpolation to preserve the monthly means. The major advance over WGEN is that
Weatherman replaces coefficients of variation with standard deviations to stabilize model
estimations when temperatures are close to zero Celsius (Mavromatis and Hansen, 2001). Solar
radiation is generated separately for wet and dry days (Mavromatis and Hansen, 2001; Soltani
and Hoogenboom, 2003). For this analysis, Weatherman was parameterized using observed
precipitation and maximum and minimum temperature data from the Hancock station for the 8year analysis period. Observed daily solar radiation was not included in the parameterization, as
our intent is to evaluate the sensitivity of crop models to generated solar radiation when observed
daily solar radiation is unavailable for a location. Using the parameterized Weatherman, we
generated daily series of precipitation, maximum and minimum temperature, and solar radiation.
However, only the generated solar radiation was used in the following analyses; daily
temperature and precipitation for input to the crop process models were obtained directly from
the Hancock observations, following (e.g., Andresen et al., 2001; Carbone et al., 2003).

(b)

Empirical Model
The regression equation developed by Ball et al. (2004) to estimate daily incoming solar

radiation at Keiser, Arkansas, United States, was recalibrated for the Hancock station. The Ball
et al. model employs a General Linear Model in the form:

Rs = β 0 + β1 X 1 + β 2 X 2 + β 3 X 3 + ... + β12 X 12
-2

Eq. 3-1
-1

where, Rs is daily solar radiation (MJ m day ); β0 and β1…12 are intercept and
regression coefficients, respectively; X1 and X5 are precipitation (mm) and precipitation squared;
0

X2 and X6 are maximum temperature ( C) and maximum temperature squared; X3 and X7 are
0

minimum temperature ( C) and minimum temperature squared; X4 and X8 are day of year and
45

day of year squared. The interaction terms are X9 (precipitation x minimum temperature), X10
(maximum temperature x minimum temperature), X11 (precipitation x maximum temperature)
and X12 (maximum temperature x day of year). Comparison of the radiation estimates when the
model developed for the calibration period (1990, 1992, 1994, 1997) was applied to the
validation period (1993, 1995, 1996, 1998) indicated that model performed similarly for both
periods (correlations between simulated and observed daily radiation of 0.87 and 0.89 for the
calibration and validation periods, respectively).
The transferability of the empirical model was assessed by applying the model to
estimate daily solar radiation at other CRN climate stations within the states of Michigan and
Wisconsin (Figure 3-1) and comparing the estimated radiation values with observations. The
time periods for which precipitation, temperature and solar radiation observations are available
vary for each location (Table 3-1). Days with missing observations were excluded from the
sensitivity analysis.

(c) Mechanistic Model
A number of mathematical models, which we label here as “mechanistic models”, have
been developed to estimate solar radiation based on the fraction of extraterrestrial radiation (S0)
reaching to the ground. This fraction is estimated as a function of the atmospheric transmissivity
(Ball et al., 2004; Donatelli et al., 2003; Spitters et al., 1986), which can be modeled based on
temperature and precipitation (Liu and Scott, 2001). Of the numerous mechanistic models
available (e.g., Bristow and Campbell, 1984; Hunt et al., 1998; Thornton et al., 2000; Thornton
and Running, 1999), those that utilize a combination of precipitation and temperature as inputs

46

generally are considered to be preferable to those using only temperature or only precipitation
(Hunt et al., 1998; Liu and Scott, 2001; Woli and Paz, 2012).
For this study, the Hunt et al. (1998) radiation model was used to estimate daily solar
radiation at the Hancock WAWN station. This model was chosen as it was developed for
Ontario, Canada, which has a relatively similar climate to our study location. Daily solar
radiation was estimated by:

S = a0 S 0 (t max − t min ) 0.5 + a1t max + a 2 P + a3 P 2 + a 4
-2

Eq.3-2

-1

where S is daily solar radiation (MJ m day ); a0, a1, a2, a3, a4 are the coefficients; , ; S0 is the
-2

-1

daily extraterrestrial solar radiation (MJ m day ), tmax and tmin are maximum and minimum
0

daily temperatures ( C); and P is daily precipitation (mm). The model uses equations previously
defined by Spitters et al. (1986) to estimate daily solar radiation at the top of atmosphere (S0):

S 0 = S sc [1 + 0.033 cos (360t d / 365)] sin β

Eq.3-3
-2 -1

-2 -1

where, S0 is extra-terrestrial irradiance (J m s ); Ssc is the solar constant 1370 J m s , the
cosine term is the yearly course of the distance between the earth and sun expressed in degrees,
td is the day since 1 January, and sin β, the sine of the elevation of the sun above the horizon, is
defined as:
sin β = sin λ sin δ + cos λ cos δ cos[15(t h − 12)]

Eq.3-4

where, λ is latitude of the site, th is hour of the day (solar time), and δ is the solar declination for
the day of year, measured in degrees and estimated by:
sin δ = − sin(23.45) cos[360(t d + 10) / 365]

Eq.3-5

47

The daily value of S0 is calculated by integrating the extra-terrestrial irradiance from sunrise till
sunset:

D = 12 + (24 / 180) arcsin(tan λ − tan δ )

Eq.3-6

where, D is day length.

The Hunt model performed equally well for both the calibration and validation periods with
correlations between simulated and observed solar radiation of 0.87 for the calibration period and
0.89 for the validation period. As for the empirical model, the mechanistic model parameterized
for Hancock was applied to estimate daily solar radiation at the other CRN stations (Table 3-1) to
assess the sensitivity and transferability of the model to other locations within the study region.

Table 3-1. Geographic location and analysis period for the CRN stations located in the states of
Michigan and Minnesota
Station
Chatham
Gaylord
Goodridge
Sandstone
b

State
MI
MI
MN
MN

Latitude
46.33
44.91
48.31
46.11

Longitude
-86.92
-84.72
-95.87
-92.99

Elevation
267 m
441 m
351 m
354 m

Period of Analysis
11/10/2004-12/31/2010
09/19/2007-12/31/2010
08/20/2003-12/31/2010
06/22/2007-12/31/2010

b

Distance
320 km
392 km
675 km
351 km

Distance from to the Hancock station.

(d) POWER
As noted above, satellite estimates of surface incoming solar radiation were obtained
from the NASA POWER database (Stackhouse, 2006), available at http://power.larc.nasa.gov/.
Daily solar radiation in this database was inferred using the Pinker and Laszlo (1992) radiative
transfer model in conjunction with water vapor amounts from the Goddard Earth Observing
System Data Assimilation System version 4 (NASA, 2011). Measurements from the
48

International Satellite Cloud Climatology and parameters from the NASA/Global Energy and
Water Cycle Experiment were used as inputs to the radiative transfer model. Please refer to the
methodological summary provided by NASA (2011) for more detailed information on the
derivation of the radiation estimates. The daily series of incoming radiation was extracted from
the POWER archive for the 1o latitude/longitude grid point nearest the Hancock station.

(e) NARR and NARCCAP
The NARR dataset was obtained from the NOAA Earth Systems Laboratory
(http://www.esrl.noaa.gov/psd/). Three-hourly values of downward solar radiation flux were
extracted for the NARR grid point closest to the Hancock WAWN station and then averaged to
estimate daily solar radiation. The daily averages were calculated based on local time, using the
3-hourly values from 0900 UTC of the current day to 0600 UTC of the next day.
The NARCCAP dataset was accessed through the Earth System Grid gateway at the National
Center for Atmospheric Research (http://www.earthsystemgrid.org/home.htm). NARCCAP
simulations from four regional climate models -- CRCM (Canadian Regional Climate Model),
ECP2 (Experimental Climate Prediction Center/Regional Spectral Model), HRM3 (Hadley

Regional Model 3), and WRFG (Weather Research & Forecasting Model) -- were included in
o

o

the analysis. All simulations employed coarse-scale (2.5 x 2.5 ) NCEP/NCAR reanalysis fields
(Kalnay et al., 1996) as lateral boundary conditions. For each model, the eight 3-hourly values
of surface downwelling shortwave radiation from 0900-0600 UTC were averaged for the grid
point closest to Hancock to obtain estimates of daily solar radiation.

49

3.2.3. Crop Yield Simulations
Decision Support System for Agrotechnology Transfer (DSSAT) version 4.5 was used to
simulate grain yields of maize and soybean at the Hancock site. DSSAT is a compilation of
simulation programs that model biophysical interactions among weather, soil, crops and farming
management (Jones et al., 2003). The programs allow a cropping system to respond dynamically
to changes in plant biology, farming management and environment. DSSAT has been widely
used around the world to explore the consequences of environmental changes and differing
farming practices on crop growth and development (e.g., Hartkamp et al., 2004; O'Neal et al.,
2005; Thorp et al., 2008; Vucetic, 2011). This simulation package also has been applied to
specifically assess the effects of generated solar radiation on crop evapotranspiration and yields
(e.g., Garcia et al., 2008). DSSAT’s capabilities and performances have been widely reviewed
(e.g., Mera et al., 2006; Southworth et al., 2002; Southworth et al., 2000).
Two modules of the DSSAT package, CROPGRO-Soybean and CERES-Maize, were
used in this study. Each module uniquely calculates the conversion of incoming radiation to plant
biomass though different photosynthetic processes. As summarized by Jones et al. (2003), the
CROPGRO module utilizes leaf-level photosynthesis, while the CERES family of models
employs radiation use efficiency. Consequently, the response of each module, as reflected in
crop yields, to different sources of solar radiation is expected to vary.
DSSAT requires that users supply cropping management options such as crop variety,
row spacing and planting date. This study generally adopted the cropping management options
utilized by Andresen et al. (2001), who investigated weather impacts on maize, soybean, and
alfalfa production in the Upper Great Lakes Region from 1895 to 1996 (Table 3-2) and common
farming practices applied in the region (Hoeft et al., 2000). Following Andresen et al., irrigation

50

was not used and soil fertility was set as non-limiting. These choices focus the investigation
only on the sensitivity of the crop model outputs to solar radiation estimates.

Table 3-2. Farming management for the DSSAT simulation

Management Rules

Soybean

Maize

Plant populations
Cultivar

20 plants per m
Generic Group 2

Planting dates*
Harvest

Automatically after May 1
At physiological maturity

2

2

8 plants per m
Short-season cultivar
0
- Base temperature: 8 C
- Thermal time for juvenile
phenological stages (P1): 200 degree
days
- Thermal time from silking to
maturity (P2): 685 degree days
Automatically after May 22
0

*Determined automatically based on soil temperature in the top of 10 cm ≥ 10 C.

Soil information for Hancock was obtained from the STATGO database published by
NRCS-USDA (Soil Survey Staff, 2010). Detailed soil physical properties needed for running
DSSAT were calculated using SBuild, a supporting package designed to create or modify soil
information to meet DSSAT requirements (Uryasev et al., 2003). Model simulations for the two
crops were run independently, and no rotation between the two crops was permitted. Simulated
grain yields at the time of maturity were saved for further analysis.

3.2.4. Evaluation Methods
Estimated daily solar radiation from the different sources was compared to observed daily
2

solar radiation using the coefficient of determination (R ) and root mean square error (RMSE),
both of which have been widely used in previous studies (e.g., Ball et al., 2004; Hunt et al.,
1998; Liu and Scott, 2001). Similar to Rivington et al. (2005), we also calculated the daily bias

51

(bi) between the daily radiation estimates (Sei) and observations (Soi), averaged over the study
period:

bi = Sei − Soi
1 n
Sei = ∑ Se ji
n j =1

Eq.3-7

and

1 n
Soi = ∑ So ji
n j =1

Eq.3-8

where i is day, j is year, and n is the number of years. Additionally, the paired t-test assuming
unequal variance was used to test for significant differences in mean daily radiation by month,
and the F-test was used for evaluating equality of variance by month.
The differences in crop yields obtained from estimated and observed daily solar radiation
as inputs were initially evaluated using scatter plots of the yields for the study period. Following
Garcia et al. (2008), further evaluation was conducted using the mean squared deviation (MSD)
as defined by Kobayashi and Salam (2000). MSD is composed of three parts, namely squared
bias (SB), squared difference between standard deviation (SDSD) and the lack of correlation
weighted by the standard deviation (LCS). These measures were explicitly developed for
comparing output from crop process models and have fewer underlying assumptions compared
to commonly-used evaluation methods (Kobayashi and Salam, 2000). SD is a traditional measure
of bias, whereas SDSD provides an indication, for this application, of how well the yield
estimates obtained from estimated daily solar radiation represent the magnitude of the annual
fluctuations of the yield estimates obtained using observed solar radiation, and LSC provides an
indication of how well the pattern of the annual fluctuations is captured. The mathematical
forms of MSD, SB, SDSD, and LCS are:

52

MSD = SB + SDSD + LCS

Eq.3-9

SB = ( x − y ) 2

Eq.3-10

SDSD = (SDx − SDy )2

Eq.3-11

LCS = 2 ( SDx) ( SDy ) (1 − r )

Eq.3-12

SDx =

1 n
∑ ( xi − x ) 2
n i =1

Eq.3-13

SDy =

1 n
∑ ( yi − y ) 2
n i =1

Eq.3-14

where x and y are the means of crop yields obtained from the generated (xi) and observed
(yi) daily solar radiation, i is 1, 2, 3 up to the length of study period (n), SDx and SDy are the
standard deviations for crop yields simulated using estimated and observed solar radiation,
respectively. The MSD analyses were supplemented with more conventional evaluation methods
including paired t-test for equality of means, F-test for equality of variances, and correlation
analysis.

3.3.

Results

3.3.1. Solar Radiation
Daily averages of observed and estimated solar radiation for the 8-year analysis period
indicate that the radiation estimations capture the annual cycle relatively well (Figure 3-2). The
large degree of day-to-day variability in the observed daily means and the estimated values from
some of the radiation sources, particularly during spring and summer, may partly be a reflection
of the short study period and the highly variable cloud cover at this time of year.

53

300
300

200
200

100
100

0
0

300
300

WRFG
WRFG

200
200

HRM3
HRM3

100
100

CRCM
CRCM

0
0

NARR
NARR

ECP2
ECP2

200
200

MEC
MEC

POWER
POWER

100
100

EMP
EMP

300
300

GEN
GEN

25
20
20
15
10
10
5
0
0
30
30
25
20
20
15
10
10
5
0
0
30
30
25
20
20
15
10
10
5
0
0
0
0

2

MJ/m /day
MJ/m2 /day

30
30

Day Year
Day ofof Year

Figure 3-2. Daily averages of estimated (black line) and observed (grey line) solar radiation at
Hancock-Wisconsin for the 8-year study period for the weather generator (GEN), empirical
(EMP) and mechanistic (MEC) models, POWER, NARR, and the NARCCAP (CRCM, ECP2,
HRM3, and WRFG) regional climate models.

Seasonal variations in daily bias are seen for the majority of the radiation estimates,
although the biases should be interpreted carefully given the short (eight year) study period over
which they were calculated. Solar radiation estimates obtained from the weather generator
display substantial positive biases in spring and fall, but negative biases during summer (Figure
3-3). In contrast, the magnitude of the average daily bias is relatively small for radiation
estimates obtained from the empirical and mechanistic models and especially for the POWER
estimates, although, in general, daily bias tends to be positive in mid to late spring and negative
in summer. Modest positive biases are seen for almost all days for the radiation estimates
obtained from NARR. For the four NARCCAP models (i.e., CRCM, ECP2, HRM3, and

54

WRFG), substantial fluctuations in daily biases are observed, with particularly large positive
biases found from early spring to late summer, similar to the pattern seen for the weather

GEN
GEN

EMP
EMP

MEC
MEC

POWER
POWER

NARR
NARR

CRCM
CRCM

ECP2
ECP2

10
10
5
5
0
0
−5
-5

HRM3

HRM3

WRFG
WRFG

300
300

200
200

100
100

0
0

300
300

200
200

100
100

200
200

100
100

0
0

10
10
5
5
0
0
−5
-5

0
0

10
10
5
5
0
0
−5
-5

300
300

2

2

Daily differences (MJ/m /day)
Daily differences (MJ/m /day)

generator.

Day of
Day ofYear
Year

Figure 3-3. Average daily bias for the solar radiation estimates. See Fig. 2 for definition of the
abbreviations.
2

Further evaluation based on RMSE and coefficient of determination (R ) suggests that
several of the daily solar radiation sources (i.e., POWER, the empirical and mechanical models,
and NARR) perform better than the commonly-used weather generator (Figure 3-4). However,
daily solar radiation from the weather generator agrees better with daily observations than
estimates obtained from the regional climate models, with the exception of the CRCM
2

simulations which have a higher R although also a higher RMSE. It is important to note that
while most of the estimation techniques considered attempt to preserve day-to-day relationships
between observed and solar radiation, the weather generator instead imitates the statistical

55

structure of a long time series of radiation (Apipattanavis et al., 2010; Castellvi and Stockle,
2001; Garcia and Hoogenboom, 2005; White et al., 2011b), and mimics the observed daily totals
only indirectly through sequences of wet and dry days.
8.50
WRFG

8.00
HRM3

7.50

2

RMSE (MJ/m2/day)
RMSE (MJ/m /day)

7.00
6.50

ECP2

6.00

GEN

CRCM

5.50

NARR

5.00
4.50
4.00

EMP

MEC

3.50
3.00

POWER

2.50
0.3

0.4

0.5

0.6
2
R2

0.7

0.8

0.9

R

Figure 3-4. Root mean squared error (RMSE) and the correlation with observed daily solar
radiation for the solar radiation estimates. See Fig. 2 for definition of abbreviations.

Mean daily solar radiation by month differs significantly from observed solar radiation
for most radiation sources, based on paired t-tests (Table 3-3). [Note that a 99 percent probability
level was used to assess significance, given the relative larger number of values and thus the
greater likelihood that small differences in the mean are statistically, but not necessarily
physically, significant.] Some exceptions are the insignificant differences found for
approximately the first half of the year for the empirical model and POWER estimates and in the
second half of the year for the mechanistic model. The largest differences in monthly mean daily

56

radiation are seen for the NARR and NARCCAP WRFG estimates, which exceed 20 percent in
almost all months, whereas the differences are smallest (<10 percent for most months) for the
mechanistic and empirical models and the POWER estimates. All four of the NARCCAP
models better capture the monthly variance compared to the other radiation sources, as seen in
the insignificant results of the F-test for equality of variance. The variance is particularly poorly
captured by the weather generator and mechanistic and empirical models.
The mechanistic model parameterized at Hancock performs well at the CRN locations in
Michigan (Chatham and Gaylord) and Minnesota (Goodridge and Sandstone), as indicated by the
2

2

relatively high R (above 0.79) and low RMSE (3.68 to 4.2 MJ/m /day), although model
performance is weaker at Chatham compared to the other locations (Figure 3-5). Model
performance not strongly associated with the distance between the station and the location where
the model was parameterized. For example, Goodridge, which is located farthest from Hancock
2

(~675 km), has lower RMSE and higher R values (i.e., the “best” performance) compared to the
2

other stations. On the other hand, the highest RMSE and lowest R values are found for
Chatham, which is located closest (~320 km) to Hancock. Validation of the empirical model
parameterized at Hancock indicates that it also performs well at the other four locations. RMSE
2

2

and R range from 4 to 4.6 MJ/m /day and 0.74 to 0.79, respectively. Model performance is
again weakest at Chatham, although the performance at Goodridge is relatively poor in contrast
to the mechanistic model which performed well at Goodridge (Figure 3-5).

57

Table 3-3. Percent difference by month in the mean and standard deviation of daily solar
radiation for the different radiation sources compared to observed radiation at Hancock,
Wisconsin.
Month
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec

GEN

EMP MEC POWER NARR CRCM ECP2 HRM3 WRFG
Difference in monthly mean daily solar radiation (percent)
-3.7
-1.6
-3.7
-4.9
39.2
19.0
8.0
21.5
22.0
6.2
-3.8
5.6
-6.5
30.8
6.1
4.7
17.6
19.9
4.1
-2.2
-1.0
-3.0
-6.6
20.4
10.1
11.3
26.7
2.0
0.9
17.0
6.2
22.6
14.3
10.1
13.3
28.0
2.9
2.4
18.9
10.1
10.8
27.9
10.6
8.9
25.7
-2.1
-1.4
5.8
6.3
-5.0
21.0
12.3
16.5
21.2
3.7
1.1
-8.4
-6.0
-3.4
19.9
12.7
11.5
21.7
0.3
4.9
-4.9
25.3
22.7
-3.6
16.5
13.8
31.5
-3.3
10.9
4.1
3.7
29.1
19.5
13.0
16.3
30.1
-2.3
5.1
16.8
14.0
8.2
33.2
20.7
20.1
28.2
2.5
33.3
21.3
21.7
54.9
11.4
32.2
20.7
43.6
3.0
0.6
24.4
21.2
13.4
-9.5
55.9
31.0
32.6
Difference in monthly standard deviation of daily solar radiation (percent)
12.5
9.7 -25.9
13.6
-12.5 12.7
8.0
-14.7
-72.8
3.3
14.4
-13.5 14.0
-21.1
-84.3 -28.5 -33.3
4.7
-1.0
-11.9
-1.3
13.5
-10.3 20.0
-81.8 -34.4 -34.7
0.9
-14.1
8.7
-4.2
-1.6
-15.5
-78.4 -37.3 -36.2
2.4
-13.4
4.0
-10.7 15.8
-5.7
-76.3 -34.4 -36.0
-1.4
1.3
-4.1
-10.3
-0.5
-77.7 -31.8 -33.1
-15.0
3.6
2.7
-6.4 11.0
-84.9 -35.1 -38.8
-3.4
-16.1
0.3
-6.9
-4.6
-7.8
-9.8
-77.0 -23.8 -30.0
-17.5
-5.7
-9.8
2.9
-11.3
-77.4 -20.5 -33.9
3.7
-16.9
-1.6 -28.5
-11.1
1.1
-8.3
7.9
0.2
-73.0
8.2
2.3 -37.9 21.7
10.2
-8.1 11.7
8.8
-78.2
-14.7
5.9
10.9
14.2
-58.4
38.5 -29.4 21.2
-21.3
-16.5
**Bolded values indicate significant differences at the 99 percent probability level based

on a paired t-test (assuming unequal variance) for equality of means or a F-test for equality of
variance.

58

4.4
4.2

RMSE (MJ/m2/day)

RMSE (MJ/m2/day)

4.6

Chatham

4
3.8

Hancock

Gaylord
Sandstone
Goodridge

3.6
0.78

0.79

0.8
R2

0.81

0.82

RMSE (MJ/m2/day)

4.8
4.6
4.4

Chatham
Goodridge

4.2
Sandstone

4
3.8

.82

3.6
0.74

Gaylord

Hancock

0.76

0.78

0.8

R2
Figure 3-5. Evaluation of the performance of the mechanistic (left) and empirical (right) models
parameterized at Hancock when applied to four Climate Reference Network stations in
Minnesota (Sandstone and Goodrich) and Michigan (Chatham and Gaylord) based on root mean
2

square error (RMSE) and coefficient of determination (R )
59

3.3.2. Crop Yields
Comparison of maize yields simulated using daily observed and estimated solar radiation
as input to CERES-Maize indicates a close agreement between the simulated yields (Figure 3-6).
Small differences exist between the radiation sources, although in general the correspondence
between the yields simulated using observed and estimated solar radiation is greatest for lower
(less than 4000 kg/ha) yields, whereas at higher yields the simulated values when using
estimated radiation are somewhat larger than those obtained using observed radiation for most
(but not all) radiation estimates.
Yield deviations are also small for soybean, although the pattern of overestimation and
underestimation varies from that observed for maize (Figure 3-7). For the three traditional
radiation sources (weather generator, empirical and mechanistic models), soybean yield obtained
using estimated radiation is larger than that obtained using observed radiation at yields >1000
kg/ha, although close agreement is observed for smaller yields. In contrast, yields simulated
using estimated radiation from NARR and three of the NARCCAP models (CRCM, HRM3,
WRFG) somewhat underestimate the yields simulated from observed radiation. Good agreement
is observed between the yield simulations using estimated and observed radiation when the
POWER and NARCCAP ECP2 sources of daily solar radiation are employed.

60

yields(ESR) (kg/ha)
yields (ESR) (kg/ha)

Figure 3-6. Maize yields for the eight-year analysis period simulated using daily solar radiation
observations and estimates as input to CERES-Maize. OSR and ESR refer to observed and
estimated solar radiation, respectively.

61

7000
7000

6000
6000

5000
5000

5000
5000

4000
4000

3000
3000

2000
2000

4000
4000

WRFG
WRFG
3000
3000

HRM3
HRM3
2000
2000

ECP2
ECP2

7000
7000
6000
6000
5000
5000
4000
4000
3000
3000
2000
2000

7000
7000

CRCM
CRCM

6000
6000

NARR
NARR

5000
5000

POWER
POWER

4000
4000

7000
7000
6000
6000
5000
5000
4000
4000
3000
3000
2000
2000

3000
3000

MEC
MEC

2000
2000

EMP
EMP

7000
7000

GEN
GEN

6000
6000

yields (OSR) (kg/ha)
yields(OSR) (kg/ha)

7000
7000
6000
6000
5000
5000
4000
4000
3000
3000
2000
2000

2500
2500
2000
2000
1500
1500

yields (OSR) (kg/ha)
yields(OSR) (kg/ha)

1000
1000
500
500

GEN
GEN

EMP
EMP

MEC
MEC

POWER
POWER

NARR
NARR

CRCM
CRCM

ECP2
ECP2

HRM3
HRM3

WRFG
WRFG

2500
2500
2000
2000

1500
1500
1000
1000
500
500
2500
2500
2000
2000

2500
2500

2000
2000

1500
1500

500
500

1000
1000

yields(ESR) (kg/ha)
yields (ESR) (kg/ha)

2500
2500

2000
2000

1500
1500

1000
1000

500
500

2500
2500

1500
1500

1000
1000

500
500

1000
1000
500
500

2000
2000

1500
1500

Figure 3-7. Soybean yields for the eight-year analysis period simulated using daily solar
radiation observations and estimates as input to CROPGRO-Soybean. OSR and ESR refer to
observed and estimated solar radiation, respectively.

For maize yield, mean squared deviation (MSD) is smallest when radiation estimates
obtained from the mechanistic model, weather generator, and NARCCAP EPC2 regional climate
model were used as input to CERES-Maize and largest for the other three NARCCAP models
(i.e., WRFG, CRCM, HRM3) (Figure 3-8). Separation of MSD into its components indicates that
the lack of correlation weighted by the standard deviation (LCS) is the largest contributor to the
MSD values for all radiation estimation techniques, and that, with the exception of the maize
yields obtained using the mechanistic model and NARR radiation estimates, squared difference

62

between standard deviation (SDSD) is the least contributor to MSD. Squared bias (SB) of the
simulated maize yields is small for all but two of the radiation estimates (i.e., NARCCAP CRCM
and WRFG). The relatively large LCS values suggest that the simulated maize yields using the
radiation estimates do not have the same pattern of variation across the 8-year study period
compared to the simulated yields obtained using observed radiation as input. In contrast, the low
SDSD values indicate that the magnitude of the yield fluctuations is similar whether estimated or
observed solar radiation serves as input to the CERES-MAIZE process model. The small SB
values for most of the radiation estimates suggest that the means are similar for the simulated
yields obtained using estimated and observed radiation.
The distribution of MSD for soybean yield across the different radiation sources is not
consistent with that seen for maize yields (Figure 3-8). The smallest MSD values are observed
for soybean yields obtained using the POWER and the NARCCAP ECP2 and CRCM radiation
estimates, followed by the empirical and mechanistic models. The largest MSD values are
observed for soybean yields simulated using solar radiation from the weather generator and
NARCCAP WRFG model. SDSD provides the least contribution to MSD for most radiation
estimates, with the exception of the mechanistic model and weather generator, indicating that the
fluctuations in soybean yield across the study period are similar whether estimated or observed
radiation estimates are used as input. However, in contrast to what was observed for maize yield,
squared bias (SB) is the largest contributor to MSD for all sources of radiation estimates except
for NARCCAP ECP2 for which the contribution from LCS is larger. This finding suggests that
the mean of the soybean yields obtained using the different sources differs from the mean
soybean yield simulated using observed daily radiation.

63

0.30

Maize

0.25
0.20
0.15
0.10
0.05

SB

0.07

LCS

WRFG

HRM3

ECP2

CRCM

SDSD

NARR

MEC

SB

POWER

EMP

GEN

0.00

MSD

SDSD
LCS
Soybean

MSD

0.06
0.05
0.04
0.03
0.02
0.01
WRFG

HRM3

ECP2

CRCM

NARR

PWR

MEC

EMP

GEN

0.00

Figure 3-8. Mean squared deviation (MSD), squared bias (SB), squared difference between
standard deviation (SDSD) and the lack of correlation weighted by the standard deviation (LCS)
for maize (left) and soybean (right) yield simulated using daily solar radiation estimates
compared to yield simulations using radiation observations. Please note the difference in the
vertical axes for the two plots. See text for definition of SB, SDSD, LCS, and MSD.

64

Paired tests indicate that, for all the radiation sources, the differences between maize
yields obtained using estimated versus observed daily solar radiation are insignificant (two-tailed
probability of 95%), in line with the interpretation above of the small values of the SD
component of the MSD statistic (Table 3-4). Percent deviations in mean maize yield range from
slightly over 1 percent for the POWER radiation estimates to approximately 7 percent for the
NARCCAP WRFG estimates. Similarly, the insignificant results for the F-test of equality of
variances for all radiation estimates agree with the interpretation of the SDSD component that
the magnitude of the yield fluctuations over the 8-year study period is similar. For all radiation
sources, the standard deviation of maize yield differs by less than 10 percent from that obtained
using observed daily solar radiation. On the other hand, correlations between the simulated
maize yields from estimated and observed radiation are high (>0.90) for all radiation sources,
even though the LCS component suggests that the yields do not have the same pattern of
variation across the study period.
For soybean, significant differences in mean yield are observed for the majority of the
radiation estimates, as was suggested by the larger relative contribution of SD to the MSD
values. Mean soybean yields obtained using estimated radiation as input are similar to those
from observed radiation for only the empirical model, POWER, and NARCCAP ECP2 radiation
sources. Deviations in mean yield are largest (over 15 percent) for the weather generator and
NARCAAP WRFG radiation sources, although the signs differed with a negative deviation
(underestimation) for NARCCAP WRFG and a positive deviation (overestimation) for the
weather generator. An interesting finding is that average simulated yield using the NARR
radiation estimates is smaller than that obtained using observed radiation, even though, as seen
above, NARR overestimates daily solar radiation. Although the difference in the standard

65

deviation of the simulated soybean yield, when compared to yields obtained using observed
radiation, is larger for the traditional radiation sources (> 15 percent) compared to the modern
radiation sources (< 10 percent, respectively), the F-tests indicate that the differences in variance
are insignificant for all sources of estimated radiation, in agreement with the small SDSD values
seen above. The high correlations suggest that the pattern of variation of the soybean yield
simulations is similar whether observed or estimated daily solar radiation is used as input.

Table 3-4. Percent difference in the mean and standard deviation of simulated maize and soybean
yields using estimated versus observed daily solar radiation as input, and the correlation between
the yield time series for the study period. Two-tailed probabilities are shown in parenthesis for
paired t-test for equality of means and F-test for equality of variances.

Radiation
Sources

Maize

Soybean

Difference Difference in
Difference in
Corre- Difference in
Correin Mean
Standard
Standard
lation Mean (%)
lation
(%)
Deviation (%)
Deviation (%)
GEN
2.63 (0.31) -2.29 (0.95)
0.99
24.37 (0.58)
0.99
15.58 (0.03)
7.23 (0.10)
EMP
4.37 (0.16) 1.18 (0.98)
0.98
15.11 (0.72)
0.99
1.54 (0.56) -8.67 (0.82)
0.99
0.99
MEC
17.56 (0.68)
9.80 (0.04)
POWER
1.12 (0.72) 9.39 (0.82)
0.99
-0.81 (0.54)
1.00 (0.98)
0.99
NARR
3.34 (0.31) 7.57 (0.85)
0.99
-6.66 (0.86)
0.98
-11.78 (0.01)
CRCM
6.09 (0.09) 4.21 (0.92)
0.98
3.21 (0.94)
0.99
-7.32 (0.01)*
-4.37 (0.16)
ECP2
4.12 (0.09) 5.32 (0.90)
0.99
2.09 (0.96)
0.99
3.33 (0.42) 1.69 (0.97)
0.97
0.98
HRM3
-12.48 (0.01)* -6.82 (0.86)
WRFG
6.73 (0.10) 9.55 (0.82)
0.98
0.98
-17.73 (0.00)* -7.28 (0.85)
Note: Bolded values and asterisks indicate significant values at 95% and 99% probability levels,
respectively.

66

3.4.

Discussion
The findings presented above suggest considerable potential for the use of non-traditional

sources of daily solar radiation estimates for agricultural applications. Although radiation
estimates obtained from traditional mechanistic and empirical models generally agreed well with
observed daily solar radiation based on a several performance measures (bias, RMSE, equality of
means), the gridded POWER satellite-based estimates performed as well, and, along with the
radiation estimates obtained from the four NARCCAP models, better captured the variance
compared to the traditional measures. The ability of the NARCCAP models to replicate observed
solar radiation was shown to be model dependent with smaller biases seen for ECP2 and CRCM
and considerably larger biases for WRFG and HRM3, although, collectively, biases were larger
for the NARCCAP models compared to the other radiation sources. Depending on the
application, daily radiation estimates obtained from NARR need to be used cautiously, as NARR
consistently overestimates daily solar radiation. One concern of the commonly-used weather
generator is the substantially underestimated variance of daily solar radiation. Afshin and Gerrit
(2003) also found that the WGEN weather generator, on which Weatherman is based, poorly
simulated the variance of daily solar radiation, and Woli and Paz (2012) warn that the statistical
properties of observed data on a daily time step may not be well represented by WGENR,
another variation of WGEN.
Several of our findings regarding the agreement of estimated and observed radiation are
2

consistent with those of previous studies. The RSME and R values found here for the POWER
radiation estimates for the grid point closest to Hancock, Wisconsin, are similar to those
reported for multiple sites in China (Bai et al., 2011) and across the continental USA (White et
al., 2011b), suggesting that the POWER gridded dataset is a promising source of daily solar

67

radiation for multiple midlatitude locations. Previous work which validated NARR fields with
ground observations for locations across North America (Markovic et al., 2009), within the
Mississippi River Basin (Kumar and Merwade, 2011), and over the Pacific Northwest (Schroeder
et al., 2009), also found that NARR tended to overestimate daily solar radiation. As discussed by
Schroeder et al. (2009) and Markovic et al. (2009), this overestimation is likely due to the
underestimation of cloud cover in the NARR atmospheric model component (i.e., the NCEP Eta
model). Tarasova et al. (2006), who evaluated the performance of the solar radiation scheme
employed in the Eta model, reported a systematic bias in solar radiation estimates due to
inaccuracies in cloud parameterization. The radiation estimates obtained from the NARCCAP
regional climate models also display positive biases for most months of the year. One possible
explanation is error in cloud cover within the NCEP global reanalysis (Kalnay et al., 1996;
Kanamitsu et al., 2002) used as the lateral boundary conditions for the NARCCAP models, and
that these errors may propagate to the regional climate model simulations. A tendency for
regional climate models to overestimate incoming solar radiation was also found by Rivington et
al. (2008b), who validated the output of the Hadley Centre’s HadRM3 for multiple sites in the
United Kingdom, although they attributed the systematic bias to the underestimation of cloud
cover within the regional model itself (Rivington et al., 2008b).
Although our findings indicate that mechanistic and empirical models continue to be
viable estimation options of daily solar radiation, concern remains regarding the transferability of
the model coefficients. The sensitivity analysis presented here indicates that the model
coefficients developed for Hancock, Wisconsin, performed well across the Upper Great Lakes
region. Unlike Hunt et al. (1998), who reported that the applicability of mechanistic model
coefficients depends on the distance from the calibration station, our results suggest that local

68

climate variations are more important than distance in determining the transferability of the
model coefficients. For example, the mechanistic and empirical models performed poorest at
Chatham, which experiences lake effect modification from both Lake Superior and Lake
Michigan (Andresen and Winkler, 2009). In line with our findings, Liu and Scott (2001)
recommended that models calibrated at a specific station can be used at other locations with a
similar climate regardless of distance, in contrast to Hunt et al. (1998) who specified an upperlimit (~390 km) from the calibration station when estimating solar radiation. The greater site
sensitivity shown here of the coefficients of the empirical model compared to the mechanistic
model suggests that the global solar radiation term as a function of latitude in the mechanistic
model provides considerable stability and transferability to mechanistic models.
The primary goal of this research was to evaluate the use of different radiation estimates
in crop process models, and our findings suggest that crop-specific differences exist. Maize
yield obtained from the CERES-Maize simulations with estimated solar radiation as the input did
not differ significantly from yield obtained using observed solar radiation as input, regardless of
the source of the estimated radiation. In contrast, significant differences were found in soybean
yield for the majority of the radiation sources. For the non-traditional sources, only the simulated
soybean yields from the POWER and NARCCAP ECP2 daily radiation estimates were in good
agreement with the yields obtained using observed radiation. It is interesting to note that
although the POWER satellite-derived estimates agreed well with observations throughout most
the growing season, the ECP2 estimates replicated the observed radiation well only during June
and July, suggesting that the timing of the biases can affect simulated soybean yield. More
difficult to explain is that simulated yield obtained from the other gridded datasets (NARR and
the remaining NARCCAP models) was significantly smaller than yield obtained using observed

69

radiation, in spite of the positive bias in the radiation estimates. This may partly be explained by
the relatively lower levels of light saturation for C3 crops. Net photosynthetic rate for C3
species increases with increasing light intensity only at low values of radiation, whereas for C4
crops it tends to increase at higher levels of light intensity (Garcia et al., 2008). Another
potential explanation is that the larger positive bias in the radiation estimates led to water stress
as increasing solar radiation can trigger greater crop evapotranspiration (Brown and Rosenberg,
1997). Higher levels of solar radiation can also increase the temperature of the plant canopy
which in turn results in greater water demand for crop transpiration (Payero and Irmak, 2006).
A limitation of this study and all previous works is the small number of crop types and
geographical locations included in the analysis. When our findings are placed in the context of
previous studies, potential differences by crop type are highlighted. For example, Garcia et al.
(2008) found for nine rainfed agricultural locations in Georgia that the substitution of
stochastically-generated solar radiation estimates for observed radiation did not have a
significant impact on simulated maize yield, similar to what we found for maize yield at
Hancock, Wisconsin. On the other hand, Garcia et al. (2008) also found that differences in
simulated yield for peanut, the representative C3 crop in their study, were insignificant. In
contrast, simulated yield for soybean, the representative C3 crop for this study, differed
significantly when stochastically-generated radiation was substituted for observed radiation.
When our findings are compared to previous analyses that employed radiation estimates obtained
from mechanistic models, we also find some differences in interpretation. Specifically,
Mavromatis and Jagtap (2005) concluded, based on Willmott’s index of agreement, that the yield
variability for a C3 crop (peanuts) at four locations in Florida displayed greater sensitivity than
that for a C4 crop (maize) to the substitution of estimated solar radiation from a mechanistic

70

model for observed radiation. Our study, in contrast, found no significant differences in
variability, based on the F test for equality of variances, for either maize (C4) or soybean (C3). A
caveat of these is that variants of the radiation estimation techniques can also influence the
interpretation. For example, Cooter and Dhakhawa (1996) found that, at least for maize, the
impact of generated solar radiation on simulated yield can vary with the type of weather
generator employed to obtain the radiation estimates, although both Garcia et al. (2008) and this
study used variants of the popular WGEN. Similarly, Trnka et al. (2007) found that different
formulations of mechanistic models had a substantial influence on wheat and barley yield.
Differences in the findings between studies may also reflect geographic influences. For
example, Bai et al. (2010) found that simulated maize yield obtained when substituting POWER
radiation estimates for observed radiation did not differ significantly for three regions in China,
similar to our results for Hancock, Wisconsin, but that this was not the case for the other two
regions included in their analysis.
Additional considerations also influence the choice of radiation estimate for an
application, as summarized in Table 3-5. Availability is a key consideration. One reason for the
current wide use of weather generators in agricultural applications is that they are already
packaged in software systems such as DSSAT, although users must input the temperature and
precipitation time series needed to parameterize the weather generator for an individual location.
The non-traditional gridded datasets (POWER, NARR, NARCCAP) are also freely available,
although some effort and computer expertise is required to download and extract the radiation
information, and these series are shorter in potential length than those obtained with weather
generators. As noted previously, an advantage of these radiation estimates over weather
generators is that they may better reflect the day-to-day relationships between radiation and other

71

variables such as temperature and precipitation. The lack of synoptic and temporal synchrony is
particularly a concern for weather generators when the intention is to incorporate the generated
values with observations of other climate parameters, for example blending observed
temperature and precipitation with generated solar radiation. Additional advantages of the
gridded datasets are that no additional model development or parameterization is usually
required and their large geographic coverage, global in the case of POWER and across North
America for NARR and NARCCAP. Mechanistic and empirical models are relatively easy to
develop/parameterize without extensive requirements of computer resources (Ball et al., 2004;
Yang et al., 2006). Considerable care needs to be taken, however, to apply these models only in
regions with a similar climate as the location for which the model was parameterized and those
with relatively stable climates (i.e. little or no temporal trends). Another concern of mechanistic
and empirical models is the possibility of negative radiation estimates, as these models are
usually formulated using multiple regression techniques. Alternative methods that consider daily
solar radiation as a zero bound variable such as gamma and semi-log regression could be used,
although we found larger biases for daily radiation estimates obtained using these methods
(results not shown). For the analyses above, the small number of negative values (0.3% for the
mechanistic model and 1.4% for the empirical model) was simply converted to zero.
Nonetheless, the frequency of negative estimates must be carefully evaluated whenever
mechanistic or empirical models are used to estimate daily solar radiation.
The development of future scenarios (also often referred to as projections) of daily solar
radiation for climate change assessments places additional constraints on the applicability of the
different sources of radiation estimates investigated here. All have their limitations. POWER is
an observation-based dataset, and scenario development is for the most part limited to a “delta

72

approach” whereby daily time series are adjusted by a mean change. This is also the case for
reanalysis datasets such as NARR. Weather generators have frequently been used to develop
climate change projections, although future changes in the variable(s) used to condition the
weather generator (e.g., precipitation) can have unanticipated effects on the other variables (e.g.,
solar radiation) being simulated (Wilby et al., 2002; Wilks, 1992). Mechanistic and empirical
models can be readily applied to estimate solar radiation for future climate conditions when
projected precipitation and temperature datasets are available, under the assumption that the
model coefficients are applicable for future climate conditions (i.e., stationarity). A specific
advantage of the NARRCAP simulations is that projections for a control and future period are
already available. However, larger biases occur when output from global climate models
(GCMs) is used in place of large-scale reanalyses as the lateral boundary conditions for the
region climate models (Olesen et al., 2007; Wilby and Harris, 2006), and a debiasing step may be
necessary (e.g., Rivington et al., 2008a; Themeβ1 et al., 2011).

Table 3-5. Strengths and limitations of traditional and non-traditional sources of daily solar
radiation estimates.

Radiation
Source
Weather
Generator

Strengths
•
•
•

Limitations

Traditional Methods
Often included within widely-used crop
•
process software systems.
Requires only temperature and
precipitation observations to
•
parameterize.
Can be used for climate change
projections, although future changes in
the variable(s) used to condition the
weather generator can have unanticipated •
effects on the other variables being
simulated (including radiation).

73

Relatively long time series of
temperature and precipitation
needed for parameterization.
Day-by-day interrelationships
between estimated radiation and
observed temperature and
precipitation are not directly
retained.
Underestimated variance of daily
solar radiation at study location*
throughout the year.

Table 3-5. (cont’d)

•

Empirical
Models

•
•

•

Mechanistic •
Models

•
•

Modest deviations in monthly mean daily
solar radiation when compared to
observations at study location.
For study location, simulated maize
(CERES-Maize) and soybean
(CROPGRO-Soybean) yield obtained
using radiation estimates did not different
significantly from simulated yield
obtained with observed radiation.
Can be easily applied to the development
of future scenarios of solar radiation,
under the assumption of stationarity of
the model with time.
Modest deviations in monthly mean daily
solar radiation compared to observations
at study location.
Coefficients of mechanistic model appear
less sensitive to local site conditions
compared to those of empirical models.
Can be easily applied to the development
of future scenarios of solar radiation,
under the assumption of stationarity of
the model with time.

•
•
•
•
•

•
•
•
•
•
•

POWER
satellitebased
estimates

•
•
•

Non-Traditional Methods
Downloadable gridded dataset.
•
Global coverage (1° latitude x 1°
longitude).
Continuously updated.

74

When used as input to
CROPGRO-Soybean (but not
CERES-Maize), resulted in
significant overestimation of
soybean yield at study location
compared to when observed
radiation served as input.
Model development required by
user.
Daily radiation observations
needed for model development.
Negative values for daily solar
radiation possible.
Regional climate variations
influence transferability of model
to other locations.
Underestimated the variance of
daily solar radiation at study
location during growing season.
Model development required by
user.
Daily radiation observations
needed for model development.
Negative values for daily solar
radiation possible.
Regional climate variations
influence transferability of model
to other locations.
Underestimated variance of daily
solar radiation at study location
throughout the year.
When used as input to
CROPGRO-Soybean (but not
CERES-Maize), resulted in
significant overestimation of
soybean yield at study location
compared to when observed
radiation served as input.
Coarser resolution (1° latitude x
1° longitude) may be insufficient
in areas with steep gradients in
cloud cover.

Table 3-5. (cont’d)
• Modest deviations in monthly mean daily
solar radiation compared to observations
at study location.
• For study location, simulated maize
(CERES-Maize) and soybean
(CROPGRO-Soybean) yield obtained
using radiation estimates did not different
significantly from simulated yield
obtained with observed radiation.
NARR
• Downloadable gridded dataset.
• Coverage includes entire North America
(32 km resolution).
• Continuously updated.

•

•

•
•

•
NARCCAP
regional
climate
models
(CRCM,
ECP2,
HRM3,
WRFG)

•
•
•
•

Downloadable gridded dataset.
Coverage includes entire North America
(~50 km resolution).
NARCCAP models were able to simulate
the variance of daily solar radiation at
study location.
Projected values of daily solar radiation
available for future period (2041-2070).

•

•
•

•

*Study location is Hancock, Wisconsin.

75

Overestimated the mean and
standard deviation of daily solar
radiation at the study location
during the cool season (OctoberFebruary).
Observation-based dataset; not
applicable for climate change
projections.
Overestimates daily solar
radiation.
When used as input to
CROPGRO-Soybean (but not
CERES-Maize), resulted in
significant underestimation of
soybean yield at study location
compared to when observed
radiation served as input.
Not applicable for climate
change projections.
Available only for specific time
slices (present-day time slice of
1971-2000 and future period of
2041-2070).
Ability to simulate daily solar
radiation varies by regional
climate model.
When used as input to
CROPGRO-Soybean (but not
CERES-Maize), resulted in
significant overestimation of
soybean yield at study location
compared to when observed
radiation served as input.
Error introduced when global
climate model (GCM) output is
used for lateral boundary
conditions may require debiasing
before application.

3.5.

Conclusions
The sensitivity of simulated maize and soybean yield at Hancock, Wisconsin, to different

sources of daily solar radiation estimates used as input the CERES-Maize and CROPGROSoybean crop process models was investigated. The radiation estimates were obtained from
traditional (stochastic generation, empirical and mechanistic models) and modern (satellite
estimation, reanalysis datasets, and regional climate model simulations) approaches.
Comparisons of the radiation estimates to observed radiation at the study location indicated that
the nature (e.g., magnitude, sign, and timing) of the biases differs considerably among the
different radiation estimates, but that, in general, the biases associated with the non-traditional
radiation sources are of similar magnitude to those of the traditional radiation sources.
Which sources of daily solar radiation estimates are then preferable as input to crop
process models? Our results indicate that the answer to this question is likely crop dependent.
The choice of radiation source did not significantly impact maize yield simulations from
CERES-Maize, whereas significant differences at the 95% probability level were found for
simulated soybean yield from CROPGRO-Soybean for all but three of the radiation sources.
Two of the three insignificant results were for non-traditional sources of radiation estimates
(POWER satellite-based estimations and the NARCCAP EPC2 regional climate simulation)
suggesting that non-traditional radiation datasets provide a viable alternative to traditional
radiation estimates as input to CROPGRO-Soybean.
Impacts of the different radiation estimates on yield simulations need to be investigated
for additional crops and geographic locations before broad recommendations regarding preferred
sources of radiation estimates for crop process models can be made. Furthermore, additional
considerations, such as data availability and model transferability need to be considered when

76

selecting a source of daily solar radiation estimates for agricultural applications, including
climate change assessments for agriculture.

Acknowledgements
This work was a part of dissertation research supported by a fellowship from the
Fulbright Presidential Scholarship. Partial support was also provided by NSF grant BCS
0909378 and USDA-NIFA grant APP #108628.

77

CHAPTER 4.
Spatial Variability of Regional Climate Change Impacts on Crop Production
in High Latitude Regions: A Case Study of the Upper Great Lakes Region
of the United States

In collaboration with
Julie A. Winkler, Jeffry Andresen, Kurt Thelen, and Sharon Zhong

4.1.

Introduction
Climate parameters, including precipitation, air temperature, and carbon dioxide (CO2),

are key factors that control crop growth and development. Numerous studies have demonstrated
that crop production has been highly sensitive to historical climate variability (e.g., Chen et al.,
2004; Goldblum, 2009; Porter and Semenov, 2005) and recent climatic trends (e.g., Almaraz et
al., 2008; Lobell and Asner, 2003; Lobell et al., 2011). Projected anthropogenic climate change
triggered by elevated atmospheric greenhouse gas concentrations is expected to substantially
impact crop production worldwide (e.g., Parry et al., 2004; Rosenzweig and Parry, 1994).
Additionally, the impacts of climate change on crop production are projected to be unevenly
distributed across the globe (Cline, 2007). In general, adverse impacts are anticipated in low
latitude regions where increased water stress and more rapid crop development and shorter time
to crop maturity due to warmer temperatures are expected to reduce yield, in contrast to high
latitude regions where a projected longer growing season and increased precipitation are
anticipated to provide more favorable growing conditions (Fischer et al., 2005; Jaggard et al.,
2010; Parry et al., 2005).

78

These generalizations are complicated, however, by longitudinal, in addition to
latitudinal, differences in projected climate change, particularly for precipitation; differences
between locations in agriculture production systems; and spatial variations in other
environmental factors important for agricultural production, such as soil fertility. Furthermore,
most previous climate impact assessments have primarily focused on changes in productivity in
current production regions (e.g., Brown and Rosenberg, 1999; Izaurralde et al., 2003; Parry et al.,
1999), and, with few exceptions (e.g., Thomson et al., 2005; Tubiello et al., 2002), have not
explicitly evaluated the potential poleward expansion of crop production.
This research focuses on the impacts of future climate change on county-level corn and
soybean production in the Upper Great Lakes Region (UGLR) of the United States, defined as
the states of Michigan, Wisconsin and Minnesota (Safir et al., 2008). This region, located
between approximately 41.5°N-49.5°N latitude (Figure 4-1) serves as an excellent case study for
evaluating potential latitudinal shifts in favorable growing regions. The southern portion of
UGLR is located along the northern edge of the agricultural heartland of the United States; and
Michigan, Wisconsin and Minnesota rank 11(12), 10(15) and 4(3) for grain corn (soybean)
production, respectively (Hatfield, 2012). In contrast, little crop production currently occurs in
the extreme northern portion of the UGLR, an area occupied primarily by forests and other
natural vegetation (Andresen et al., 2013). Additionally, several environmental factors
complicate future northward expansion of crop production, most notably the decrease in soil
fertility from the deep, rich prairie soils of southern Minnesota and Wisconsin to the thinner,
sandier spodosol soils of northeastern Minnesota and northern Wisconsin and Michigan. The
moderating influences of the Great Lakes, particularly cooler spring and summer temperatures,
also may impact the future expansion of favorable growing regions.

79

0
50°

0
-97°

0
-95°

0
-93°

0
-91°

0
-89°

0
-87°

0
-85°

0
-83°

0
50°

0
49°

0
49°

0
48°

0
48°

0
47°

0
47°

0
46°

Minnesota

0
46°

0
45°

0
45°

Wisconsin

0
44°

0
44°

Michigan

0
43°
USA

0
42°

0
43°
0
42°

0
41°

0
41°
0
-97°

0
-95°

0
-93°

0
-91°

0
-89°

0
-87°

0
-85°

0
-83°

Figure 4-1. Geographical location of the Upper Great Lakes Region

Only few previous studies have investigated the potential impacts of climate change on
agriculture in the UGLR, all of which considered a limited number of representative locations in
their analyses. In an early study, Andresen et al. (2000) employed empirically-derived climate
st

projections for the 21 century from two global climate models (GCMs) to assess future corn,
soybean, and alfalfa yield at 13 study sites across Minnesota, Wisconsin and Michigan, including
several locations where crop production is currently limited. Applying crop simulation models
to a single cultivar for each crop type, the authors reported higher simulated average yield of
st

corn, soybean, and alfalfa for the 21 century compared to a historical period (1896-1996), with
the greatest increases found at the more northern study locations. Additionally, larger yield
increases were obtained when CO2 enrichment was included in the simulations. A recent update
of the earlier study that focuses on corn and wheat production (Andresen et al., 2013) suggests a
80

more nuanced interpretation of future yield changes. Employing a smaller number of study sites
(five locations in Michigan, eastern Wisconsin and western New York), but a much larger
ensemble of climate projections, Andresen et al. (2013) found that the direction and magnitude
st

of projected yield change for the 21 century varied with crop type and location. Under current
CO2 concentrations, simulated corn yield decreased for the majority of the empirically-derived
climate projections at all but the northernmost location, whereas wheat yield increased for most
locations and climate projections. Furthermore, CO2 enrichment substantially increased
projected yield for wheat (a C3 crop) but not corn (a C4 crop).
In contrast to Andresen et al. (2013; 2000), Southworth et al. (2002; 2000) limited their
impact assessments to established crop production areas in the portion of the Great Lakes region
extending from central and southern Michigan and Wisconsin to southern Illinois, Indiana and
Ohio. Based on empirically-downscaled climate projections from a single GCM and small
number (10) locations across the study area, Southworth et al. (2000) found that the sign of the
projected change in simulated corn yield for a future period (2050-2059) relative to a baseline
period (1961-1990) varied with location and crop cultivar. Simulated yield for long-season corn
varieties increased at the Wisconsin and Michigan study sites but not at sites located farther
south, whereas the projected average yield of medium-season varieties decreased at all locations
although the decrease was less for the Wisconsin and Michigan locations. A projected decrease
in yield for short-season varieties was also found, with the exception of the study site in extreme
western Illinois. Based on nine locations within the study region, geographical and varietal
differences were also found for projected soybean yield, with large increases in yield simulated
for the Wisconsin and Michigan locations and only small increases, or even a decrease, projected

81

for the southern study locations (Southworth et al., 2002). In general, yield gains were greatest
for late maturing rather than the early maturing cultivar (Southworth et al., 2002), and CO2
fertilization had a positive influence on future yield.
These previous analyses point to considerable complexity in projected future changes in
productivity in the UGLR, but, nonetheless, all suggest the potential for increased productivity at
more northern locations. Any generalization is limited, however, by the small number of
locations employed in these earlier assessments. Thus, there is considerable need for a more
spatially-detailed assessment of future crop productivity that explicitly considers the potential for
the poleward expansion of crop productivity, specifically for the UGLR but with implications for
high latitude regions worldwide. In order to fill this gap, the major goal of this research is to
provide a county-level assessment of the spatial variability of the impacts of projected future
climate change by the mid century (2041-2070) on corn and soybean production across the
UGLR, including spatial variations in future planting dates, time to maturity, seasonal
evapotranspiration (i.e., cumulative evapotranspiration at maturity), and crop yield. Additionally,
the sensitivity of the projected changes to CO2 concentration and, for corn only, to crop cultivar
are assessed. Corn and soybean are employed in the assessment as they represent the major crop
rotation in the current UGLR agricultural production regions and are a C4 and C3 crop,
respectively. The assessment presented here also differs from previous assessments for the study
region in that dynamically-downscaled, rather than empirically-downscaled, climate projections
are utilized. An ensemble of climate projections obtained from the recently-released North

American Regional Climate Change Assessment Program (NARCCAP) (Mearns et al., 2009) is
employed.

82

4.2.

Data and Methods

4.2.1. Regional Crop Model Simulation
Regional simulation of corn and soybean production for all counties in the study region
was completed using the CERES-Maize and CROPGRO-Soybean models incorporated in the
Decision Support System for Agrotechnology Transfer (DSSAT) ver. 4.5 (Hoogenboom et al.,
2010). DSSAT is a cropping system model that integrates different computer programs (i.e.
modules) to simulate daily crop growth and development (Jones et al., 2003). Each module
specifically handles weather inputs, soil conditions, soil-plant-atmosphere interactions, plant
growth and development, and farming management. Inputs required for DSSAT simulation are
climate/weather variables (daily solar radiation, precipitation, maximum and minimum
temperature), soil information, and farming management. The DSSAT system shares common
routines to simulate fluctuations of soil parameters; whereas, crop growth processes are
simulated by individual plant growth modules (Jones et al., 2003). In the DSSAT system, corn
and soybean are grouped into cereals and legumes, respectively.
DSSAT has been used widely around the world to investigate the consequences of
environmental changes and crop management practices (e.g. Hartkamp et al., 2004; O'Neal et al.,
2005; Thorp et al., 2008; Vucetic, 2011), including climate change impact and adaptation
assessments (e.g. Brassard and Singh, 2008; Meza and Silva, 2009). DSSAT’s capabilities and
performances have also been reviewed widely (e.g. Mera et al., 2006; Southworth et al., 2002;
Southworth et al., 2000). According to White et al. (2011a), who reviewed 221 articles on
climate change impacts on crop production, the CERES family of models are regularly
implemented (~40% articles) to evaluate crop responses to climate change scenarios. Although,
CROPGRO is not as well known as the CERES family (White et al., 2011a), CROPGRO-

83

Soybean often has been used for climate change impact assessments in high latitude regions
(e.g., Alexandrov et al., 2002; Brassard and Singh, 2008). As noted above, both CERES-Maize
and CROPGRO-Soybean were employed in previous climate impact assessments for the UGLR
(e.g., Andresen et al., 2013; 2000; Southworth et al., 2002; 2000). In this research, DSSAT was
run for each growing season for 30-year historical (1971-2000) and future (2041-2070) periods.
DSSAT was originally developed for plot level simulations, although it subsequently has
been employed to estimate yield at a range of spatial scales including for large non-uniform areas
(e.g., Irmak et al., 2005; Jagtap and Jones, 2002). For this study, DSSAT was used to simulate
the spatial distribution of county-level corn and soybean yield. In this we follow the lead of,
among others, Haskett et al. (1995) who utilized a crop simulation model (SOYGRO) to simulate
county-level yield in Iowa. As pointed out by Irmak et al. (2005, 2344), an implicit assumption
is that “ the impacts of climate at the aggregate spatial scale are the same as those produced by
the crop models, which have limited inputs over space”.
County-level corn and soybean yield were simulated for 30-year historical and future
time slices. A 30-year time series has been considered sufficient by a number of authors (e.g.,
White and Hoogenboom, 2011) for climate change impact assessments. Detailed descriptions of
the inputs for the model simulations including the future climate scenarios are presented below.

(a)

Historical Climate Observations
Preparing climate data for DSSAT simulation on the county-scale poses some challenges

as climate stations are distributed unevenly over the study region. Several previous studies have
employed gridded fields of precipitation and temperature as inputs to crop model simulations at a
regional scale (e.g., Irmak et al., 2005; Jagtap and Jones, 2002; Quiring and Legates, 2008).
Although, the “regular” grid is an obvious advantage, the spatial interpolation is sensitive to the

84

density of climate stations, their geographic distribution across a region, and topographical
terrain (Hoogenboom, 2000). Gridded time series also are sensitive to inhomogeneities in the
original station data (Guentchev et al., 2010). When gridded fields of daily temperature and
precipitation are used, an issue that is of particular concern for this study is that the gridded
fields overestimate the number of days with precipitation and underestimate daily minimum
temperature (Pollyea, 2013; Quiring and Legates, 2008; West et al., 2007).
Recognizing these constraints of gridded fields, we elected to directly employ observed
time series of daily temperature and precipitation. Simply assigning a county to a nearby climate
station is complicated by differences between climate stations in the length of record, the
frequency of missing observations, and the inhomogeneities introduced into the record by
changes in instrumentation, time of observation, and station relocation. Thus, we sought to
identify a modest number of high quality observing stations that capture the spatial variations in
climate across the UGLR. To accomplish this, an objectively-defined climate regionalization
developed using k-means cluster analysis (see Chapter 2) was utilized to assist in the selection of
representative climate stations (Figure 4-2). The climate regionalization grouped monthly means
of precipitation and maximum and minimum temperature for the period of 1971-2000 from 180
climate stations within the UGLR and its neighboring states that are included in the United States
Historical Climate Network ver 2.0 (Menne et al., 2012a; Menne et al., 2009). The results of the
climate regionalization suggest that the climate stations within the study region can be
categorized into seven climate types (Figure 4-2, top) that are distinguished by differences in the
deviation of the annual cycle of temperature and precipitation from the UGLR average.
Each county was then assigned to a climate region, and based on the county-climate
memberships, we selected two or more representative climate stations for each climate group,

85

including spatially-fragmented county-climate clusters. A primary consideration in station
selection was the completeness of the daily time series. Even though the USHCN stations are
considered to be the highest quality climate stations in the United States, data completeness and
record length vary dramatically among the USHCN stations. Another consideration was to select
stations that captured the north-south and east-west extents within a climate region. Thus, more
stations were selected for larger regions compared to smaller regions.
A total of 36 stations out of 79 USHCN climate stations within the UGLR were selected
to capture the spatial climate variability of the region (Figure 4-2-bottom). Daily data
completeness over the growing season (April to October) of the historical period (1971-2000) is
more than 85% for the majority of the stations. The distribution of daily data completeness (in
brackets) at five percentage intervals by station is: 34 stations (>85%), 31 stations (>90%), 27
stations (>95%), and 12 stations (>99%). To assign a representative climate station to each
county, we identified counties and representative climate stations that fell within the same
climate group memberships, and then assigned a county to the closest representative climate
station within the climate region based on Euclidian distance between the climate station location
and the county centroid (Figure 4-2, bottom).

86

Climate Stations
Representative Stations

Climate Region
1

4

6

2

5

7

3

ARGL
MADA

GRFR
PRMN

PRDM

THBR
CLQT

IRWD
STBG

SPNR

MWES
MILA

CBGN
MSFD

PSTN

NWUM

MNSG NWBR

BWLR OCNT

ZMRT

ETWS
FDLC

GMDW
PRDC

BRHD

HART
MDLN
BRWT
OWSO
MCLN
ALGN
STHV
ANBR
CDWR
ADRN

Figure 4-2. Selection of representative climate stations for the regional climate change impact
assessment on corn and soybean production in the UGLR based on an objective climate
regionalization (top) and their assignment to counties (bottom). Each station is referenced by a
four character identifier and the colors in the lower panel show the counties assigned to the same
representative climate station.

87

Missing values of precipitation, maximum and minimum temperature for the USHCN
stations were filled in using daily climate values obtained from the Iowa Environmental Mesonet
(IEM) Network (IEM, 2013) for either the same station or a nearby station. IEM recently made
available complete daily time series for many climate stations in the Midwestern states including
Michigan, Wisconsin and Minnesota. IEM filled in missing data using an inverse distance
weighted interpolator applied to climate data from NCEP Stage IV precipitation, airport weather
data (ASOS/AWOS), and available NWS COOP reports (Drayl, 2012, personal communication).
Daily solar radiation is a required input for the crop simulations, but this variable is
infrequently measured. For this analysis, daily solar radiation was estimated using a mechanistic
radiation model proposed by Hunt et al. (1998) which requires daily maximum and minimum
temperature and precipitation as inputs. The choice of this approach was based on a systematic
evaluation of several alternative estimation procedures for daily solar radiation (see Chapter 3).
The mechanistic radiation model was calibrated using observed data from 1990-2008 for a
climate station, at Hancock, Wisconsin, and applied to the 36 representative climate stations.

(b)

Soil data
As soil characteristics vary spatially, for each county we selected the “dominant” soil

based on areal coverage. Soil data were obtained from the State Soil Geographic (STATSGO)
database published by the Natural Resources Conservation Service of the United States
Department of Agriculture (NRCS-USDA) (Soil Survey Staff, 2010). STATSGO has been
regularly employed for crop model simulations in the United States (e.g., Carbone et al., 2003;
Safir et al., 2008). Soil Data Viewer (NRCS, 2012), which is supplied to read STATSGO
database, was used to extract soil information stored in STATSGO layer by layer, with the
midpoint of each layer roughly corresponding to the depths in the generic soil requirements in

88

DSSAT. The eight soil depths employed in the DSSAT based on the extracted layers are: 5 cm
(0-10 cm), 15 (10-20), 25 (20-30), 40 (30-50), 60 (50-70), 80 (70-90), 105 (90-120), and 135
(120-150).
Soil parameters specifically extracted from STATSGO for each soil layer are soil
classification (e.g., loam, silty loam, sandy loam), slope, drainage, runoff potential, soil texture
(percent silt and clay), organic content, pH and cation exchange capacity. Soil color and fertility
factor were retrieved based on soil texture and taxonomy from the National Cooperative Soil
Survey (NCSS, 2011) and the generic soil database available in DSSAT package, respectively.
SBuild, a DSSAT supporting package (Uryasev et al., 2003), was employed to prepare soil
inputs and calculate specific soil information, such as drainage upper and lower limit, required
for the DSSAT simulations.

(c)

Cropping Management
DSSAT requires information on agronomic practices, e.g., planting densities, row

spacing, planting date, and crop cultivar. The agronomic practices were generally adopted from
Andresen et al. (2001), who applied DSSAT for thirteen locations in the UGLR. The planting
2

densities were determined at 75 cm spacing with the plant population of 6 plants/m and 20
2

plants/m for corn and soybean, respectively. These agronomic practices reflected farming
management and technology in the late 1990s (Andresen et al., 2001), which overlaps with the
historical time slice used in this study.
Selecting a crop cultivar for the DSSAT simulations of corn yield is challenging as
multiple cultivars are grown within each county. Furthermore, the relative proportion of
different cultivars varies spatially across the UGLR (Coulter et al., 2010b) and also annually as

89

farmers consider weather conditions and economic profits for planting a particular corn hybrid
(Hao et al., 2010). Given the focus of this analysis we selected, from several options, the cultivar
for the DSSAT simulations that best reproduced the spatial distribution of county-level corn
yield. As potential options we considered the modified short season cultivar employed by
Andresen et al. (2001), and the three corn cultivars (i.e., short, medium and long season) used by
Southworth et al. (2000). Based on visual comparisons and statistical indexes (i.e., spatial
correlation and the Willmott Index (Willmott, 1981)), we found that for the 1971-2000 historical
period the median values of simulated corn yield for the medium season cultivar better captured
the spatial variability of observed county-level corn yield defined as crop production divided by
planted areas. This was true even when compared to simulating the yield for multiple cultivars
and selecting the yield value that best matched the observed county yield (results not shown).
The cultivar characteristics for the medium season are presented in Table 4-1.

Table 4-1. Cultivar characteristic of the medium season corn cultivar used in the DSSAT
simulations.
0

P1

Degree days (base 8 C) from emergence to end juvenile phase

200

P2

photoperiod sensitivity (0-1)

0.3

P5

degree days (base 8 C) from silking to physiological maturity

800

G2

potential kernel number

700

G3

potential kernel growth rate (mg/d)

6.3

0

0

PHINT degree days required for a leaf tip to emerge (phyllochron interval) ( C d)
Source: Southworth et al. (2000)

38.9

In contrast to the simulations for corn, four maturity groups were preselected for the
soybean DSSAT simulations. The maturity groups were roughly assigned to each county based
on the distribution of soybean maturity groups in the United States published by the National
Soybean Research Laboratory (NSRL, 2013). These different maturity groups reflect the
90

sensitivity of soybean to photoperiods (Carbone, 1995). The cultivar characteristics for the four
maturity groups were available from the soybean cultivar file supplied by DSSAT version 4.5.
Planting dates were determined using the DSSAT automatic planting option. Planting
occurs when percentage soil water and soil temperature in the top 10 cm are at least 20% and 10
0

C, respectively, during a specified planting window. For the historical simulations, five and six

different planting windows were employed for the corn and soybean simulation, respectively.
The planting windows were determined by moving the starting date of a planting window later
st

th

st

by roughly 15 days during the period April 1 to June 10 for corn and from April 1 to June
th

st

th

th

th

30 for soybean (e.g., planting windows of April 1 -June 10 , April 16 -June 10 , and so on).
The end dates of the planting windows for both crops were obtained from the typical planting
and harvesting dates for corn and soybean in Michigan, Minnesota and Wisconsin as reported by
(NASS-USDA, 1997). For the future simulations, the starting date of the planting windows was
st

set earlier to March 1 considering the potential warming condition for the region. For both the
historical and future periods, the highest annual simulated yield for each county from the
different planting windows was retained for the analysis. Harvesting dates for both the historical
and future simulations were determined at maturity stage (i.e., R6 for corn and R8 for soybean)
following Andresen et al. (2001).
The different planting windows were used to take into account the short ‘optimum’ time
period to grow corn in the UGLR (Thelen, 2007) and to allow for replanting decisions following
adverse climate conditions (Benson, 1990; Laurer, 1997). Additionally, the multiple planting
windows in part compensate for the use of a single soil type to represent the soil conditions for a
county, as suggested by Moen et al. (1994).

91

No change in cropping practices over time was assumed. The crop simulations assumed
rainfed conditions were initiated at the first of January each season with soil water initialized at
drained upper limit. The nitrogen option within the DSSAT environment was turned off for corn
simulations, but turned on for the soybean simulations to capture nitrogen fixation as soybean is
a leguminous crop that is able to fixate nitrogen (Coulter et al., 2010a).

4.2.2. Climate Change Scenarios
Future climate change scenarios were developed based on the outputs of the NARCCAP
2

simulations, available at a 50 km resolution. Eight combinations of RCM/gcm simulations were
available for a future (2041-2070) and control (1971-2000) period (Table 4-2). The model
combinations are CRCM-ccsm, CRCM-cgcm3, HRM3-hadcm3, HRM3-gfdl, RCM3-cgcm3,
RCM3-gfdl, WRFG-ccsm, and WRFG-cgcm3. The model ID presented in Table 4-2 will be used
as a reference for the model identifier in the figures included in this Chapter.

Table 4-2. The NARCCAP model combinations employed for the study
ID Model
A CRCM-ccsm
E RCM3-cgcm3

ID
B
F

Model
CRCM-cgcm3
RCM3-gfdl

ID
C
G

Model
HRM3-gfdl
WRFG-ccsm

ID
D
H

Model
HRM3-hadcm3
WRFG-cgcm3

Daily maximum and minimum temperature and precipitation from the land grid points
nearest each of the 36 representative climate stations were extracted to calculate monthly means
for the future (2041-2070) and control (1971-2000) periods. Differences (∆T) in monthly means
of temperature and percentage changes (%∆P) in monthly precipitation were calculated between
the two periods for each RCM/gcm simulation and applied to the observed historical daily time
series of maximum and minimum temperature and precipitation for the 36 representative climate
stations. This simple approach of developing future climate change scenarios has been

92

previously applied for numerous climate change impact assessments (e.g., Mearns et al., 1997;
Tubiello et al., 2002; Wang et al., 2011).
For reference, delta values for the growing season (March-October) are shown in Figure
4-3. The projected change in seasonal mean maximum and minimum temperature and
precipitation for the region depends upon the combination of RCM/gcm (Figure 4-3). HRM3gfdl projects larger increases in temperature and greater decreases in precipitation, particularly
for the southern areas of the UGLR, than the other NARCCAP models. WRFG-cgcm3 and
WRFG-ccsm project slightly warmer and relatively wetter conditions compared to the other
RCM-gcm combinations. In general, the regional climate of the UGLR by mid century, as
derived from the majority of the NARCCAP models, is projected to be warmer than currently,
particularly in the southern UGLR, and wetter than currently, especially for the northern UGLR
(Figure 4-3).
Atmospheric CO2 concentration for the historical period (1971-2000) was set at 370 ppm
(i.e., approximate CO2 concentration at the end year of the historical period). The SRES A2
emissions scenario, the CO2 scenario used in the NARCCAP simulations, projects CO2 increases
from about 490 ppm to 635 ppm during the mid-century (2041-2070) future period (IPCC-DDC,
2011). We selected the projected CO2 values at the start and end of the mid century period (i.e.,
490 and 635 ppm) as input to the DSSAT simulations to assess the spatial variability of carbon
fertilization on corn and soybean production.

93

CRCM/cgcm3 CRCM/ccsm

A

RCM3/cgcm3 HRM3/hadcm3 HRM3/gfdl

B

C

D

RCM3/gfdl

E

WRFG/cgcm3 WRFG/ccsm

F

G

H

0

% PrecipChange
Precip Change

Temp.Change (
Temp. Change C)
<=1.0
<=1.0
1.0 - 1.5
1.0-1.5
1.5 - 2.0
1.5-2.0

2.0 - 2.5
2.0-2.5
2.5 - 3.0
2.5-3.0
3.0 - 3.5
3.0-3.5

<=-10
<=-10
-10 - -5
-10- -5
-5
-5-- 0
0-5
0- 5

3.5 - 4.0
3.5-4.0
>4.0
>4.0

55-10
- 10
10 - 15
10-15
15 - 20
15-20
20 - 25
20-25

25 - 30
25-30
>30
>30

Figure 4-3. Projected changes in °C for simulated average maximum temperature (left) and
minimum temperature (middle), and percent change in precipitation (right) for the growing
season (March-October) between the future (2041-2070) and the control (1971-2000) periods for
the eight NARCCAP models. Letters of A-H are the NARCCAP model identifiers (Table 4-2).
94

4.2.3. Evaluation of CERES-Maize and CROPGRO-Soybean
Before the NARCCAP simulations were used as input to the crop models, simulations
using observed climate time series were first compared to historical county yields to evaluate
how well the simulated yields capture the observed spatial variability across the UGLR. This
evaluation is essential background for interpreting the spatial variations in projected future yield.
Observed county-level crop yields were calculated by dividing crop production by
planted areas as reported by the National Agriculture Statistics Service, United States
Department of Agriculture (NASS-USDA, 2011). Because of advancement in farming practices
and technology, such as new cultivars and pest management which significantly contributes to
increased crop yields (Egli, 2008), temporal trends are expected in observed yields, whereas
temporal trends are not expected for simulated yields as farming practices and technology are
held constant throughout the simulation period. Thus when comparing observed and simulated
yield, any temporal trends in the observed yield should first be removed (e.g., Andresen et al.,
2001; Jagtap and Jones, 2002) . We detrended observed yields separately for each county with at
least five years of available data to the base year of 2000 using the approach proposed by
Tannura et al. (2008) that assumes a linear temporal trend in crop yield.

95

Corn

Observed Yield

Soybean

Observed Yield

DSSAT Yield

DSSAT Yield

Yield Ratio

Yield Ratio

Available Data

Available Data

Corn (kg/ha)
<=1500
1500 - 2500
2500 - 3500
3500 - 4500
4500 - 5500
5500 - 6500
6500 - 7500
>7500

Soybean (kg/ha) Yield Ratio
<=500
500 - 1000
1000 - 1500
1500 - 2000
2000 - 2500
2500 - 3000
3000 - 3500
>3500

<=0.50
0.50 - 0.75
0.75 - 1.00
1.00 - 1.25
1.25 - 1.50
1.50 - 1.75
1.75 - 2.00
>2.00

Data (Years)
5-9
10 - 14
15 - 19
20 - 24
25 - 29
30
< 5 or NA

Figure 4-4. The median value of observed and DSSAT simulated yields for corn (left) and
soybean (right). The yield ratio was calculated by dividing simulated yield by observed yield for
each county. Available data refers to the numbers of years with reported yield for each county.
Note: The scale for median yield is different for corn and soybean.
96

Figure 4-4 shows that the simulated corn and soybean yields are generally able to capture
the spatial distribution across the study area of observed yield for both crop types. Observed corn
and soybean yields gradually decrease from south to north across the UGLR. Higher yields are
specifically found in southern Minnesota and Wisconsin for both crops. This pattern is generally
captured by the DSSAT simulated yields, although some discrepancies appear. For corn,
underestimations are seen in southwestern Minnesota and northern Michigan, whereas yield is
overestimated in the northern portion of the study area, especially in areas where little crop
production currently occurs. Simulated soybean yields exceed observed yields particularly in the
southern part of the region, while underestimations appear in most northern counties. Overall, the
median value of simulated yield, averaged across all counties, is higher for corn by about 136
kg/ha (~2.3%) and lower for soybean by about 244 kg/ha (~9.5%), than the reported county
statistics. The Willmott Index of agreement and the spatial correlation of the median yields are
0.83 and 0.70 for corn, while the two indeces are both 0.65 for soybean. The Wilmott Index
ranges from 0 to 1 with 1 indicating good agreement (Willmott, 1981). The modest high values
of the Willmott Index and the small deviations in region-wide median yields indicate the
simulated county-level yields of corn and soybean capture the observed yield patterns well.

4.2.4. Regional Impact Assessment of Projected Climate Change
The potential consequences of future climate change were assessed by estimating the
median changes in planting dates, time to maturity (defined as the days from planting to full
maturity, i.e., to R6 for corn and R8 for soybean), seasonal evapotranspiration, and grain yield
for all NARCCAP climate model combinations at 370, 490 and 635 ppm. The first CO2
concentration was defined as a reference, while the other two are used to evaluate sensitivity of

97

the simulated outcomes to elevated atmospheric CO2 concentrations. We also compared the
coefficient of variation (COV) of crop yields between the future and the historical period to
evaluate potential change in future yield variability (Tubiello et al., 2002). COV was calculated
for each county by dividing the standard deviation of the crop yield for a time slice by the mean
value. High values of COV indicate high interannual yield variability, whereas small values
imply more stable annual production (Tubiello et al., 2002).
Specifically for corn, we also evaluated the sensitivity of future climate change impacts
to different corn cultivars, recognizing that a range of cultivars are grown in the UGLR. In
addition to the medium season cultivar, we compared yield changes between the future and
historical periods for short season and long season cultivars. These comparisons were limited to
simulations using only the highest CO2 concentration (635 ppm) for the future period.
Potential shifts in major growing regions were estimated by modifying, for each county,
the median value of the reported historical county-level yield by the suite of projected changes
in median corn and soybean yield from DSSAT simulations obtained using the eight NARCCAP
climate simulations and the three levels of atmospheric CO2 concentration.

4.3.

Results

4.3.1. Historical Simulation
The DSSAT simulated median values of planting date, time to maturity, cumulative
seasonal evapotranspiration and grain yield for the historical period provide a reference for
evaluating potential future changes. The spatial variations in the simulated values are briefly
summarized below.

98

The simulated median planting dates (PDAT) for the historical (1971-2000) period fall
between Julian day (JD) 120-127 (~April 30 to May 7). There is a slight latitudinal variation in
planting date (Figure 4-5), with median planting dates between JD 113 (~Apr 23) to JD 127
(~May 7) for the majority of the counties in northern Michigan and in central and northern
Wisconsin and Minnesota. Earlier simulated planting dates (before JD 106 (~April 16)) are seen
for a small number of counties in northwestern Minnesota, whereas for the majority of the
southern counties simulated planting dates for corn fall between JD 120 (~April 30) and JD 141
(~May 21). Simulated median planting dates are considerably later for a small number of
counties in the central and southern Lower Peninsula of Michigan (Figure 4-5). For soybean,
simulated median planting dates lie between JD 120 and JD 127 for the majority of counties in
Minnesota and between JD 120 and JD 141 for the Upper Peninsula of Michigan, and northern
Wisconsin. On the other hand, a wider range (JD 120 – JD 156) in simulated median soybean
planting dates is seen for counties in the Lower Peninsula of Michigan with simulated planting
dates as late as early to mid June (>JD 156) for a small number of counties in the central Lower
Peninsula.
The simulated median time to maturity (TMAT) for corn displays a marked south-tonorth gradient (Figure 4-5), with a generally shorter TMAT of almost 20 days for counties in the
southern compared to northern portion of the UGLR. Simulated median TMAT for soybean is
generally between 120 and 140 days for most counties in the UGLR. A few counties in
northeastern Minnesota and the eastern Upper Peninsula Michigan display simulated median
TMAT values of 140-160 days.
The regional distribution of the simulated median values of cumulative seasonal
evapotraspiration (ET) exhibits higher ET for counties in the south than for counties in the north

99

with only a slight difference in ET between corn and soybeans (Figure 4-5). Simulated ET for
corn is about 550 to 700 mm for the southern UGLR counties and about 400 to 600 mm for the
majority of the northern counties. Simulated ET for soybean is in the range of 500 to 700 mm for
the southern counties and about 400 to 550 mm for the northern counties. For some counties in
the northern Lower Peninsula of Michigan, simulated ET is lower than 400 mm.
A south-north gradient is also seen in the distribution of simulated median yield for corn
and soybeans. Simulated median yield in southern Michigan, Minnesota and Wisconsin exceeds
7500 kg/ha, whereas the median yields fall between 1500 – 4500 kg/ha for most of the northern
counties. For soybean, the simulated median yields are about 2500 to 3500 kg/ha for the southern
counties with several counties having median yields over 3500 kg/ha, while the northern counties
display yields of 500 – 2500 kg/ha with a few counties in the northern Minnesota and the Upper
Peninsula of Michigan having higher (2500-3000 kg/ha) and lower (<=500 kg/ha) simulated
yields, respectively.

100

Corn

PDAT

ET

TMAT

Yield

Soybean

PDAT

ET

TMAT

Yield

PDAT(JD)

TMAT (days) ET (mm)

Corn (kg/ha)

Soybean (kg/ha)

110 - 120
<=400
<=1500
<=500
106 - 113
120 - 130
400 - 450
1500 - 2500
500 - 1000
113 - 120
130 - 140
450 - 500
2500 - 3500
1000 - 1500
120 - 127
140 - 150
500 - 550
3500 - 4500
1500 - 2000
127 - 134
150 - 160
550 - 600
4500 - 5500
2000 - 2500
134 - 141
160 - 170
600 - 650
5500 - 6500
2500 - 3000
141 - 148
170 - 180
650 - 700
6500 - 7500
3000 - 3500
148 - 156
>180
>700
>7500
>3500
>156
Figure 4-5. Median values of simulated planting date (PDAT), time to maturity (TMAT),

cumulative seasonal evapotranspiration (ET), and grain yield (Yield) for corn (top four panels)
and soybean (bottom four panels) in the UGLR for the historical period (1971-2000)

101

4.3.2. Future Climate Impacts
Planting Dates and Time to Maturity
Substantial future changes in the spatial distribution of planting date (PDAT) for corn
production are suggested for the future time slice (Figure 4-6). In spite of differences between
climate scenarios, earlier planting dates are projected for most counties in the southern portion of
the study area. Earlier planting dates are projected for a larger number of southern counties for
the HRM3-gfdl, CRCM-cgcm3 and RCM3-cgcm3 simulations compared to the other
NARCCAP simulations; whereas, the projected future climate from WRFG-cgcm3 suggests later
planting dates for most southern counties except for the southeast Michigan. Later planting dates
(<20 days) are projected for most counties in the central and the northern UGLR, particularly
when corn production is simulated using climate scenarios from CRCM-ccsm, CRCM-cgcm3,
HRM3-gfdl, HRM3-hadcm3, and RCM3-gfdl. For the extreme northern UGLR, slightly earlier
planting dates (0-10 days) are projected by most climate models (particularly WRFG-ccsm and
WRFG-cgcm3), with the exception of HRM3-hadcm3 for which later planting dates are
projected (Figure 4-6). When the projected changes in PDAT are averaged across all climate
scenarios (Figure 4-6, average), earlier and later planting dates are seen for the southern and
northern counties, respectively, with the exception of three counties in northern Minnesota where
earlier planting dates are projected.
Projected changes in time to maturity (TMAT) for simulated corn production suggest, a
shorter TMAT for most UGLR counties, with the exception of counties in the Upper Peninsula
of Michigan and a few counties in northwest Minnesota. For these counties, a slightly longer
TMAT (<10 days) is projected when climate simulations from WRFG-ccsm and WRFG-cgcm3
served as input to CERES-Maize.

102

PDAT

A

B

C

D

E

F

G

H

Average

TMAT

A

B

C

D

E

F

G

H

Average

PDAT Change (days)
<=-30
-30 - -20
-20 - -10
-10 - 0

TMAT Change (days)

0 - 10
10 - 20
20 - 30
> 30

<=-50
-50 - -40
-40 - -30
-30 - -20

-20 - -10
-10 - 0
0 - 10
> 10

Figure 4-6. Projected change in the median of planting date (PDAT, above) and time to maturity
(TMAT, below) between the historical (1971-2000) and the future (2041-2070) period for corn
production in the UGLR at the reference level of CO2 (370 ppm) concentration. Letters of A-H
are the NARCCAP model identifiers (Table 4-2).

103

For soybean production, planting dates are projected to occur earlier for most counties
located in Minnesota, Wisconsin and the Upper Peninsula of Michigan based on the climate
scenarios obtained from CRCM-cgcm3, RCM3-cgcm3, RCM3-gfdl, WRFG-ccsm, and WRFGcgcm3. On the other hand, later planting dates (10-30 days) for soybean production are estimated
for counties in Wisconsin and the Lower Peninsula Michigan by CRCM-ccsm, HRM3-gfdl,
HRM3-hadcm3, RCM3-gfdl, and WRFG-ccsm. When the average change in planting dates
across all climate scenarios is considered (Figure 4-7, average), a primarily west-east gradient is
observed across the UGLR, with earlier planting dates projected in the western UGLR and later
planting dates projected in the central and eastern UGLR.
The changing pattern of time to maturity (TMAT) for soybean production, when
averaged over all climate scenarios, suggests a shorter TMAT (~-20 days) for most counties in
the UGLR, although a slightly longer (<10 days) TMAT is projected for several counties in
south-central Minnesota (Figure 4-7, average). Four RCM-gcm combinations (i.e., CRCMcgcm3, RCM3-cgcm3, WRFG-ccsm, and WRFG-cgcm3) project more counties with longer
TMAT compared to the other NARCCAP combinations, particularly in the western and southern
UGLR (Figure 4-7).

104

PDAT

A

B

C

D

E

F

G

H

Average

TMAT

A

B

C

D

E

F

G

H

Average

PDAT Change (days)
<=-30
-30 - -20
-20 - -10
-10 - 0

TMAT Change (days)

0 - 10
10 - 20
20 - 30
> 30

Figure 4-7. As Figure 4-6, but for soybean production

105

<=-50
-50 - -40
-40 - -30
-30 - -20

-20 - -10
-10 - 0
0 - 10
> 10

Evapotranspiration and Crop Yield
The spatial pattern of projected change between the future (2041-2070) and the historical
(1971-2000) period in the median value of simulated cumulative seasonal evapotranspiration
(ET) for corn production, when averaged across all climate scenarios, clearly displays a northsouth gradient with decreased ET projected for the southern UGLR and increased ET for the
northern UGLR (Figure 4-8, average). Moreover, this general pattern is seen for the majority of
NARCCAP models, although some differences exist. Substantially larger decreases in ET (<15%) are projected for counties in the Lower Peninsula of Michigan by CRCM-ccsm, CRCMcgcm3, HRM3-gfdl, and WRFG-ccsm. On the other hand, the area projected to experience
higher ET in the future is larger for the RCM3-cgcm3, RCM3-gfdl, WRFG-ccsm and WRFGcgcm3 climate scenarios.
Similar to the spatial pattern of ET change, corn yield is projected to decrease for most
southern counties and increase in the northern counties (Figure 4-8, below). A substantial portion
of the southern and central UGLR is projected to have decreased yield for the HRM3-gfdl,
CRCM-ccsm, CRCM-cgcm3, and HRM3-hadcm3 scenarios, with projected decreases as large as
-25% to -50%. Increased yield in the northern UGLR is seen for all climate scenarios, although
the area with projected yield increases is larger for WRFG-ccsm and WRFG-cgcm3 compared to
the other climate scenarios.

106

ET

A

B

C

D

E

F

G

H

Average

Yield

A

B

C

D

E

F

G

H

Average

ET Changes (%)
<=-15
-15 - -10
-10 - -5
-5 - 0

Yield Changes (%)
0-5
5 - 10
10 - 15
>15

<=-50
-50 - -25
-25 - 0
0 - 25

25 - 50
50 - 100
100 - 150
>150

Figure 4-8. Change in the median of cumulative seasonal evapotranspiration (ET, above) and
crop yields (Yield, below) between the historical (1971-2000) and the future (2041-2070) period
for corn production in the UGLR at the reference level of CO2 (370 ppm) concentration. Letters
of A-H are the NARCCAP model identifiers (Table 4-2).

107

For soybean production, small (-5% to 5%) changes in ET are projected for almost all
counties in the UGLR with no clear spatial pattern evident (Figure 4-9, average). A notably
larger area, particularly in Wisconsin and Minnesota, is projected to have increased (although
small) ET for RCM3-cgcm3, RCM3-gfdl, WRFG-ccsm, and WRFG-cgcm3 compared to the
other climate scenarios (Figure 4-9), whereas future ET simulated using future climates obtained
from CRCM-ccsm and HRM3-gfdl exhibits a slight decrease (~-10%) for the majority of
counties in the UGLR.
Future climate change is projected to benefit soybean yield for the majority of the
northern counties of the UGLR, although a slight decrease in soybean yield is seen for the
HRM3-gfdl climate scenario (Figure 4-9). The climate projections from HRM3-gfdl also result
in a notable decrease in yield (ranging from -25% to more than -50%) for most counties located
in the southern and central part of the UGLR. Moreover, a slight (~-25%) decrease in yield is
projected by RCM3-cgcm3, RCM3-gfdl and WRFG-cgcm3 for the central and southern
counties, whereas CRCM-ccsm, CRCM-cgcm3, HRM3-hadcm3, RCM3-cgcm3 and WRFGccsm project a moderate decrease (~-50%) for the southern counties, particularly those in
Minnesota and Michigan (Figure 4-9). Overall, the projected change in soybean yield averaged
across all climate projections (Figure 4-9, average) shows a distinct south- north gradient with
decreased yield in the southern part of the UGLR and increased yield in the northern part.

108

ET

A

B

C

D

E

F

G

H

Average

Yield

A

B

C

D

E

F

G

H

Average

Yield Changes (%)

ET Changes (%)
<=-15
-15 - -10
-10 - -5
-5 - 0

0-5
5 - 10
10 - 15
>15

Figure 4-9. As Figure 4-8, but for soybean production

109

<=-50
-50 - -25
-25 - 0
0 - 25

25 - 50
50 - 100
100 - 150
>150

4.3.3. Carbon Fertilization Effects
The consequences of different levels of atmospheric CO2 concentration on crop
production are assessed by evaluating the average changes in simulated PDAT, TMAT, ET and
Yield between the future (2041-2070) and the historical (1971-2000) period between elevated
CO2 concentrations of 490 ppm and 635 ppm and in comparison to the reference level (370 ppm)
CO2 level. The analysis reveals that PDAT and TMAT are relatively insensitive to elevated CO2
concentration compared to ET and Yield. This pattern is consistent for both corn (Figure 4-10)
and soybean (Figure 4-11) production. Thus, we explore below only the sensitivity of the spatial
distribution of ET and Yield to elevated CO2 concentrations.
For corn production, the north-south gradient in projected changes in ET and Yield seen
for the reference CO2 concentration also are evident for the higher CO2 concentrations.
However, the projected decrease in ET for the southern and central counties is larger, and the
projected increase in ET for the extreme northern counties is smaller, for the higher CO2
concentrations, particularly for the simulations employing CO2 levels of 635 ppm. These
projections suggest less future moisture stress under higher CO2 concentrations, potentially
benefiting corn production (Figure 4-10).
For soybean production, the higher CO2 concentrations appear to have little effect on the
projected spatial distribution of ET across the study region. In contrast,the projected increase in
soybean yield across much of the study area is larger for the elevated CO2 concentrations and the

110

number of counties in the southern Wisconsin and Michigan with a projected decrease in yield is
smaller for the higher CO2 concentrations (Figure 4-11).

PDAT

PDAT Change (days)
<=-30
-30 - -20
-20 - -10
-10 - 0

635ppm

490ppm

TMAT

TMAT Change (days)
<=-50
-50 - -40
-40 - -30
-30 - -20

635ppm

490ppm

ET

-20 - -10
-10 - 0
0 - 10
> 10

ET Change (%)
<=-15
-15 - -10
-10 - -5
-5 - 0

635ppm

490ppm

0 - 10
10 - 20
20 - 30
> 30

0-5
5 - 10
10 - 15
>15

Yield
Yield Change (%)

490ppm

<=-50
-50 - -25
-25 - 0
0 - 25

635ppm

25 - 50
50 - 100
100 - 150
>150

Figure 4-10. Change in median planting date (PDAT), time to maturity (TMAT), seasonal
evapotranspiration (ET) and crop yields (Yield) between the historical (1971-2000) and the
future (2041-2070) period averaged across all climate model combinations for corn production in
the UGLR under elevated CO2 concentration of 490 ppm (left) and 635 ppm (right)

111

PDAT

PDAT Change (days)
<=-30
-30 - -20
-20 - -10
-10 - 0

635ppm

490ppm

TMAT

TMAT Change (days)
<=-50
-50 - -40
-40 - -30
-30 - -20

635ppm

490ppm

ET

<=-15
-15 - -10
-10 - -5
-5 - 0

Yield

490ppm

-20 - -10
-10 - 0
0 - 10
> 10

ET Change (%)

635ppm

490ppm

0 - 10
10 - 20
20 - 30
> 30

0-5
5 - 10
10 - 15
>15

Yield Change (%)
<=-50
-50 - -25
-25 - 0
0 - 25

635ppm

25 - 50
50 - 100
100 - 150
>150

Figure 4-11. As Figure 4-10, but for soybean production

When the projected changes in ET and Yield for the future period (2041-2070) are
expressed as a percent difference between the projected values under the elevated CO2 level (490
or 635 ppm) and the projected values under the reference CO2 level (370 ppm), e.g. (ET490ET370/ET370))*100, the smaller projected ET with increasing CO2 concentrations is more
clearly seen for corn production Figure 4-12). The percent changes in projected ET under CO2
concentration of 490 ppm and 635 ppm with respect to projected ET under the reference CO2
level (370 ppm) ranges from -2% to -6% and -2% to -8% , respectively. The spatial pattern is
not as uniform for soybean production, although, as noted above, the number of counties

112

projected to have increased future ET is smaller for the higher CO2 concentration levels
compared to the reference CO2 level.
Corn yield in the UGLR is expected to increase moderately as the CO2 concentration
increases. At 635 ppm corn yield in the northern areas could increase up to 80% with respect to
the reference CO2 level (370 ppm). As for corn, soybean yield likely also will benefit from
higher concentrations of atmospheric CO2. All counties in the study region are projected to have
substantially higher yields as atmospheric CO2 concentration is increased to 635 ppm (Figure
4-12). Overall, soybean grown in the UGLR is expected to benefit more from increasing CO2
concentration than corn production.

113

Corn

ET(490-370)

ET(635-370)

Yield(490-370)

Yield(635-370)

Soybean

ET(490-370)

ET(635-370)

Yield(490-370)

Yield(635-370)

Yield (%)

ET(%)
<=-8
-8 - -6
-6 - -4
-4 - -2

-2 - 0
0-2
2-4
>4

<=10
10 - 20
20 - 30
30 - 40

40 - 60
60 - 80
80 - 100
>100

Figure 4-12. Percent change of projected seasonal evapotranspiration (ET) and crop yield (Yield)
for the future period (2041-2070) under elevated CO2 concentration of 490 ppm (490-370, left)
and 635 ppm (635-370, right) for corn (above) and soybean (below) with respect to projected ET
and Yield for the future period (2041-2070) under the reference level of CO2 concentration (370
ppm). ET and Yield are obtained by averaging estimated future ET and Yield across all
NARCCAP climate models.
114

4.3.4. Cultivar Sensitivity
Specifically for corn, we also evaluated the potential impacts of future climate change on
grain yield for short and long season cultivars. This evaluation was performed for only the
highest (635 ppm) CO2 concentration. The projected changes in future yield for the short cultivar
displays a south-north gradient similar to that seen earlier for the median season cultivar with
small yield reductions in the southern and central portions of the UGLR and modest increases in
the northern UGLR (Figure 4-13). This pattern is seen for the majority of the NARCCAP climate
scenarios. In contrast, the sign of the projected changes in future yield is positive across almost
the entire study area for the long season cultivar (Figure 4-13), with only a small numbers of
counties in southern Wisconsin and Michigan projected to experience a small (~-25%) yield
decrease. The projected increases in yield are substantial (50-100%) and are seen for all the
NARCCAP climate scenarios, although the HRM3-gfdl scenario is slightly less favorable for
some counties in south Minnesota, south-west Wisconsin and south-east Michigan (Figure 4-13,
Long).

115

Yield (kg/ha)

Yield_short

<=1500
1500 - 2500
2500 - 3500
3500 - 4500
4500 - 5500
5500 - 6500
6500 - 7500
>7500

Yield_long

Short

A

B

C

D

E

F

G

H

A

B

C

D

E

F

G

H

Average

Long

Average

Yield Change (%)
<=-50
-50 - -25

-25 - 0
0 - 25

25 - 50
50 - 100

100 - 150
>150

Figure 4-13. Change in the median value of corn yield between the future (2041-2070) and the
historical period (1971-2000) for short season (Short) and long season cultivar (Long) under
different climate change projections and elevated CO2 of 635 ppm. The median values of
simulated yield for the historical period for each cultivar are presented in the first row as a
reference. Letters of A-H are the NARCCAP model identifiers (Table 4-2).
116

4.3.5. Coefficient of Variation
Calculation of the coefficient of variation (COV) for the historical period (1971-2000)
indicates that annual corn yield variability is higher for the northern areas of the region than for
the major corn production regions in the south where values of COV are not more than 0.4 and
can be less than 0.1 (Figure 4-14). Future projections, holding CO2 concentrations constant at
370 ppm and employing a medium season cultivar, suggest a slight increase in the annual
variability of corn yield in the southern UGLR, particularly for southern Wisconsin and the
Lower Peninsula of Michigan, as indicated by the ratio of future COV to historical COV which
falls between 1 and 1.25. Comparison among the eight climate scenarios reveals that those
derived from CRCM-ccsm, HRM3-hadcm3 and HRM3-gfdl project COV ratios greater than 1
for a larger number of counties in the southern UGLR compared to the other remaining
scenarios. On the other hand, yield variability is projected to decrease by approximately half for
the mid century time slice for most counties in the northern UGLR (Figure 4-14). When the
COV ratio is averaged across the all climate scenarios, the number of counties in southeast
Wisconsin and southern with a slight increase in annual yield variation decreases as CO2
concentrations increase from 490 to 635 ppm. On the other hand, the number of counties in
northern Minnesota and Wisconsin and in the Upper Peninsula of Michigan with substantially
smaller projected future annual yield variability increases as CO2 levels increase (Figure 4-14).
For soybean yield, the majority of counties in the UGLR have a COV less than 0.55 for
the historical period with several counties in the Upper Peninsula of Michigan and few counties
in northwestern Minnesota possessing a relatively high COV (> 0.7 or higher) (Figure 4-15). By
the mid century, and assuming a reference CO2 level for the DSSAT simulations, the climate

117

scenarios obtained from six of NARCCAP climate models (i.e., CRCM-ccsm, CRCM-cgcm3,
HRM3-gfdl, HRM3-hadcm3, RCM3-cgcm3, RCM3-gfdl) project higher COV values for the
majority of counties in Minnesota, whereas WRFG-ccsm and WRFG-cgcm3 suggest smaller
COV values for the majority counties in the UGLR with the exception of increased annual yield
variability for southern Wisconsin and Michigan. The HRM3-gfdl scenario projects substantially
larger future yield variability in the southern UGLR. When averaged across all climate scenarios,
annual variability of soybean yield is likely to increase slightly (COV 1-1.25) for most counties
in south-central Minnesota, central Wisconsin and the eastern Lower Peninsula Michigan,
whereas a relatively higher annual yield variation (COV 1-1.50) by the mid century is projected
for southern Wisconsin and the southern Lower Peninsula of Michigan. On the other hand, a
majority of counties in the currently unproductive northern UGLR likely will have lower COV
values in the future (Figure 4-15). Elevated CO2 concentration is expected to reduce the area
experiencing increased future yield variability, although a slight increase in annual yield
variability in the southern UGLR is projected, even for the highest CO2 concentration employed
in this study (Figure 4-15).

118

Historical

A

B

C

D

E

F

G

H

Average

COV

COV Ratio

<=0.10
0.55 - 0.70
1.25 - 1.50
<=0.50
0.10 - 0.25
0.70 - 0.85
1.50 - 1.75
0.50 - 0.75
0.25 - 0.40
0.85 - 1.00
1.75 - 2.00
0.75 - 1.00
0.40 - 0.55
>1.00
1.00 - 1.25
>2.00
COV Sensitivity to Elevated CO2 concentration

490ppm

635ppm

Figure 4-14. Ratio of coefficient of variation (COV Ratio) of corn yield for the future period
(2041-2070) with respect to the historical period (1971-2000) under different climate projections
derived from NARCCAP models at the reference CO2 (370 ppm) level. Average is the COV
ratio averaged across all climate projections. Below two panels are the average COV Ratio under
elevated CO2 of 490 (left) ppm and 635 ppm (right). Letters of A-H are the NARCCAP model
identifiers (Table 4-2).
119

Historical

A

B

C

D

E

F

G

H

Average

COV Ratio

COV

<=0.10
0.55 - 0.70
1.25 - 1.50
<=0.50
0.10 - 0.25
0.70 - 0.85
1.50 - 1.75
0.50 - 0.75
0.25 - 0.40
0.85 - 1.00
1.75 - 2.00
0.75 - 1.00
0.40 - 0.55
>1.00
>2.00
1.00 - 1.25
COV Sensitivity to Elevated CO2 concentration

490ppm

635ppm

Figure 4-15. As Figure 4-14, but for soybean yield

120

4.3.6. Potential Expansion in Major Growing Areas
In order to better visualize the potential spatial changes in corn and soybean production,
several yield thresholds were selected and projected future county yield was categorized as
falling above or below the threshold. For corn the yield thresholds are 5000, 6000, 7000, 8000
and 9000 kg/ha, and for soybean the thresholds are 2000, 2500, 3000, 3500, and 4000 kg/ha.
The lower threshold categories were chosen based on current yields in the major production
regions within the UGLR; the higher thresholds were added given the projections discussed
above of future higher yield in some parts of the UGLR.
Starting with the lowest threshold (5000 kg/ha), corn production regions is anticipated to
expand northward into northern Minnesota and the Upper Peninsula of Michigan, especially
under the highest (635 ppm) CO2 concentration (Figure 4-16), but to contract in southern
Wisconsin and the southern Lower Peninsula of Michigan, at least for the reference (370 ppm)
CO2 concentration. This same pattern is seen for the 6000 kg/ha threshold. For the higher
threshold of 7000 kg/ha, and assuming the reference CO2 concentrations in the DSSAT
simulations, counties with projected yields, above the threshold are confined primarily to
southern Minnesota, similar to the spatial pattern seen for the historical simulations. A larger
number of counties, including some counties in south-central Wisconsin, east-central Michigan
and the Upper Peninsula of Michigan, have projected yields above this threshold when the
elevated CO2 levels were employed in the DSSAT simulations. Increasing the yield threshold to
8000 and 9000 kg/ha results in the projected yield for a smaller number of counties exceeding
the threshold, as expected, although the spatial pattern is more “scattered” compared to that for

121

the historical simulations where counties with yields exceeding these high thresholds are
confined to primarily southern Minnesota (Figure 4-16).

Historical

CO2-370

CO2-490

CO2-635

5000

5000

5000

5000

6000

6000

6000

6000

7000

7000

7000

7000

8000

8000

8000

8000

9000

9000

9000

9000

Yield of Threshold
Higher
Lower

NA

Figure 4-16. Potential expansion in major corn production regions of the UGLR assessed by
categorizing the median simulated corn yields for the future period (2041-2070) is equal to or
higher than a given yield threshold, namely: 5000 kg/ha (first row), 6000 (second row), 7000
(third row), 8000 (fourth row), and 9000 (fifth row).

At the lowest threshold for soybean production (i.e., 2000 kg/ha), the spatial distribution
of counties with yields above the threshold is generally similar for the historical and mid-century
periods when the lower (370 ppm) CO2 concentrations were used in the DSSAT simulations

122

(Figure 4-17). At higher CO2 levels, projected yield for a larger number of counties, particularly
in northwestern Minnesota and the northern Lower Peninsula of Michigan, exceeds the
threshold. As the yield threshold is increased to 2500 kg/ha and 3000 kg/ha, the number of
counties with projected yield above the threshold, assuming a constant CO2 concentration, is less
than that for the historical simulations, with large changes particularly seen in southern
Michigan. In contrast, the spatial extent of counties with projected yields above the threshold
expands northward for the two higher CO2 concentrations. For the historical period, no counties
have simulated yield that exceeds the two highest thresholds (3500 and 4000 kg/ha), but future
projected yield is greater than these thresholds for several, spatially scattered, counties,
particularly for the simulations employing higher (635 ppm) CO2 concentrations.

Historical

CO2-370

CO2-490

CO2-635

2000

2000

2000

2000

2500

2500

2500

2500

3000

3000

3000

3000

3500

3500

3500

3500

4000

4000

4000

4000

Yield of Threshold
Higher
Lower
Figure 4-17. As Figure 4-16, but for soybean
123

NA

4.4.

Summary and Discussion

4.4.1. Spatial Variation of the Climate Change Impact
This study was primarily purposed to evaluate the potential consequences of future
climate change on the spatial variability of corn and soybean production in the UGLR by the mid
century (2041-2070) with reference to a historical period (1971-2000) using DSSAT crop models
and climate scenarios obtained from NARCCAP simulations. The crop model simulations for the
historical period demonstrate that these models are able to capture spatial variations in planting
dates for corn and soybean in the UGLR, and the simulated planting dates fell within the range of
usual planting dates reported by NASS-USDA (NASS-USDA, 1997). The south-north gradient
of observed county-level yield also was well depicted by the DSSAT simulated yields.
The impacts of future climate change on corn and soybean production in the UGLR are
complex considering non-linear interactions between a changing climate and crop growth and
development. The inclusion of eight climate scenarios derived from the NARCCAP simulations
(i.e., RCM-gcm “combinations”) results in a range of possible climate change impacts for corn
and soybean planting date, time to maturity, evapotranspiration and grain yield.
Assuming for the DSSAT simulations a constant level of CO2 at 370 ppm, planting dates
for corn were projected to occur earlier for counties located in the southern UGLR and slightly
later for counties in the northern UGLR for most of the climate scenarios. In contrast, a west-toeast gradient was found for soybean production with earlier planting dates in the western UGLR
and later dates in the eastern UGLR. The changes in planting date may be associated with the
relative magnitude of warmer and wetter condition projected across the region and possible
future changes in ‘optimum’ planting windows. A number of planting windows were used in the
DSSAT simulations, and the highest yield from the different planting windows was selected for

124

further assessment. This criterion assumed that the highest yield reflected the ‘optimum’ planting
window (Thelen, 2007). Future climate change is projected to moderately shorten time to
maturity (TMAT) for corn production but slightly decrease TMAT for soybean production.
These findings likely reflect changes in the thermal time (i.e., growing degree days) for corn or
in photothermal period for soybean needed to complete crop growth and development.
For corn production, the projected decrease ET in the southern UGLR, in contrast to a
projected increase in the northern UGLR, may reflect a shorter future TMAT. Theoretically, ET
will increase as temperature increases, but a shorter TMAT will reduce the amount of time to
accumulate ET during the corn growing season.
For soybean production, only small changes in ET are projected with no clear spatial
pattern. These small changes likely reflect opposing changes in ET, precipitation and
temperature. For example, the HRM3-gfdl climate scenario suggests that a shorter TMAT,
warmer temperatures and drier conditions in Michigan will result in lower ET. However, for
other parts of the study area the increasing temperature and wetter conditions associated with this
climate scenario result in a slightly higher future ET, even though the crop growth period is
projected to be shorter.
A relatively moderate decrease in corn and soybean yield is estimated for the majority of
counties located in the central and southern areas; whereas counties in the Upper Peninsula of
Michigan, northern Wisconsin and Minnesota will benefit from climate change. This result
supports and provides more spatial detail regarding the potential positive impacts of climate
change on crop yields as suggested by previous studies for a small number of northern locations
in the UGLR (e.g., Andresen et al., 2013; Tubiello et al., 2002). On the other hand, the projected

125

moderate decrease in yields for the southern UGLR may occur due to a shorter TMAT resulting
from warmer conditions, even with wetter future conditions.
Elevated CO2 concentrations are expected to affect the magnitude of seasonal
evapotranspiration (ET) and grain yield. Increasing CO2 levels ppm are expected to slightly
decrease ET for corn production; whereas the impact of elevated CO2 is less for soybean
production. The reduction in ET for corn production is likely because a C4 crop is assumed to
have higher stomatal resistance than a C3 crop (Rosenzweig and Iglesias, 1998; White and
Hoogenboom, 2011) under elevated CO2. The higher stomatal resistance reduces the daily
transpiration rate, potentially leading to decreased seasonal evapotranspiration.
Elevated atmospheric CO2 concentrations are projected to benefit corn and soybean
yield. Of the two crops, soybean (a C3 crop) appears to benefit more from elevated CO2 than
corn (a C4 crop) perhaps because C3 crops have higher photosynthesis rates than C4 crops under
elevated CO2 concentrations (Rosenzweig and Iglesias, 1998). In addition, increased CO2 levels
may lead to increased radiation use efficiency for C3 crops; whereas, elevated CO2 levels have a
smaller effect on C4 crops (Stöckle and Kemanian, 2009). Therefore, soybean production is
anticipated to benefit more from carbon fertilization than corn production. This potential benefit
from carbon fertilization on crop yield also has been identified by previous studies for the UGLR
(e.g., Andresen et al., 2013; Southworth et al., 2002; Southworth et al., 2000). However, the
results presented here, especially for corn, suggest that a moderate decrease (~-25%) in yield
may still occur by mid century in the southern UGLR even under elevated CO2 concentrations.

126

For corn, we also examined potential consequences of future climate change under the
highest (635 ppm) CO2 level to different cultivars (i.e., short and long) considering that farmers
in the study region plant a mixture of different cultivars (Coulter et al., 2010b). The analysis for
the long season cultivar suggests that yield will increase for almost all counties in the study area.
Relatively high percentage changes are projected for the northern UGLR where long season
varieties are currently not widely produced. As for the short season hybrid, projected changes are
similar for those the medium season cultivar with yield increasing in the northern UGLR and a
slight yield decrease in the southern UGLR. These results strengthen previous analyses done by
Southworth et al. (2000) for a few locations in Michigan and Wisconsin which reported that long
season cultivars will benefit more from future climate change compared to short and medium
season cultivars likely due to a longer growing season in the future.
As mentioned above, an ensemble of climate scenarios derived from the NARCCAP
simulations with “combinations” of RCM and GCM models was used to evaluate the regional
impacts of climate change in the UGLR. When the driving GCM was held constant, the RCMs
were found to produce substantially different temperature and precipitation projections for the
UGLR, illustrating the importance of including multiple RCMs to dynamically downscale GCMs
in an ensemble of future climate projections. The variation of regional climate conditions
projected by different RCMs nested within the same GCM could happen because each RCM may
employ different grid structures, numerical schemes, surface boundary conditions and model
parameterizations to solve sub-grid scale processes, even though the RCMs were developed
using the same fundamental conservation laws and dynamic equations (Winkler et al., 2011).

127

4.4.2. Future Yield Variability and the Potential Expansion in Production Region
This study revealed that regional variations in projected maximum and minimum
temperature and precipitation by mid century likely will impact corn and soybean production in
the UGLR. The northern areas are projected to benefit from climate change; whereas a slight
decrease in crop yield may occur for the southern areas, especially for corn yield, regardless of
carbon fertilization. This result to some extent clarifies the regional variations associated with the
expectation of potential benefits from future climate change in higher latitude regions (Cline,
2007; Fischer et al., 2005; Jaggard et al., 2010).
Furthermore, analysis of future yield variability based on the coefficient of variation
(COV) suggests that the temporal variability of corn yield will increase slightly for the southern
UGLR, particularly in Wisconsin and the Lower Peninsula of Michigan, regardless of climate
scenario. For soybean, annual yield variability for the southern and central counties, including
the current major production regions, is also expected to increase in the future, regardless of
climate scenario and CO2 level. As most of these regions have historically low COV values
(<0.4), a slight increase may still result in “reasonable” annual yield variability (<0.6).
Nonetheless, the southern areas of the UGLR likely will be more vulnerable to yield variability
than at present.
Our analysis of the potential spatial shifts in crop production suggests that favorable
growing conditions and elevated CO2 concentrations in the future may create an opportunity for
growing corn and soybean in the northern UGLR, i.e. the Upper Peninsula of Michigan, and
northern Minnesota and Wisconsin, where historically little production has occurred (Andresen
et al., 2013; Miller et al., 2005). However, growing conditions will continue to remain favorable
within the current major crop production areas in the southern UGLR.

128

4.4.3. Limitations
As for other modeling studies, there are a number of some limitations related to the
development of the climate scenarios and to model assumptions. The future climate projections
were derived from a single emissions scenario (i.e., SRES A2) as this is the only emissions
scenario employed in the NARCCAP simulations. Although, we attempted to overcome this
limitation by using two elevated CO2 concentrations for the future time slice to capture the
potential benefits of carbon fertilization, the use of a single emissions scenario does not permit a
comparison of future yield under different emissions scenarios which may also result in different
future climate conditions.
Furthermore, future daily climate series were obtained by modifying daily observed
climate data (i.e., maximum and minimum temperature and precipitation) with corresponding
monthly changes calculated from the NARCCAP simulations. Although this approach has been
regularly applied in climate impact assessments to remove systematic bias involved in future
climate simulations (Moriondo et al., 2011), it assumes that the variability of the temperature and
precipitation daily series remain constant in time. Only the magnitude of temperature and
precipitation is adjusted to reflect future conditions.
The underlying assumptions of the crop models employed for the assessment also impose
limitations. The DSSAT models simulate non-linear interaction between abiotic factors (i.e.
climate and soil condition) and crop growth processes. The models do not simulate well pest and
disease infestations (Soussana et al., 2010) which may increase under climate change
(Diffenbaugh et al., 2008; Luck et al., 2011) and do not include exposure to other factors
affecting yield variability such as economic conditions (Kaufmann and Snell, 1997). Current
crop models also estimate the effects of carbon fertilization based on experiments conducted in

129

enclosed carbon chambers (Challinor et al., 2009). Thus, the models may overestimate the
benefits of elevated CO2 concentration (Long et al., 2006), as environmental conditions and
applied agricultural management practices may influence the crop response to elevated CO2
concentration (Rosenzweig and Tubiello, 2007). For example, low soil nutrients and irrigation
may reduce the benefit of carbon fertilization as discussed by many authors (e.g., Rosenzweig
and Tubiello, 2007; Tubiello and Ewert, 2002).
Additionally, except for planting date, farming management and technology (i.e.,
cultivar, row and spacing, and planting density) employed for the DSSAT simulations are
assumed constant for each county in the future (2041-2070) period to isolate the consequences of
climate change on crop production. This assumption neglects that farmers may use different
farming practices in the future as farming management and agricultural technology evolve with
time (Egli, 2008).

4.5.

Conclusion
Agricultural production has been identified as a key sector susceptible to global climate

change, and changes in global crop production will challenge the future global food supply. This
study provides more detailed spatial information regarding potential regional climate change
impacts on crop production in a high latitude region, which generally are expected to benefit
from future climate change. The analysis suggests that future climate change will alter planting
date, time to maturity, seasonal evapotranspiration, and grain yield of corn and soybean
production in the UGLR.
For corn production, the southern areas, i.e., the current major production areas, may
experience earlier planting dates by mid century, whereas slightly later planting dates may occur

130

in the northern areas. For soybean production, climate scenarios derived from the majority of the
NARCCAP simulations suggest earlier planting dates for most counties in Minnesota and some
counties in Wisconsin and Michigan with an west-east gradient of changing planting dates from
earlier planting dates in the western UGLR and later planting dates in the eastern UGLR.
Warmer temperatures are expected to reduce time to maturity for both crops in the majority
counties in the region, including in the major production areas in the southern UGLR.
Seasonal evapotranspiration for corn production in the southern UGLR is expected to
decrease in the future under warmer conditions, possibly because of a shorter time to maturity,
and increase in the northern UGLR. For soybean production, seasonal ET is projected to change
only slightly by the mid century. Furthermore, under the assumption of a constant level of CO2
concentration, warmer conditions likely result in slightly decreased corn and soybean yield in the
southern UGLR, partly due to a shorter time to maturity. On the other hand, favorable growing
conditions in the northern UGLR likely will increase corn and soybean yield. Elevated CO2
concentrations also are anticipated to benefit crop yield and partially alleviate potential negative
impacts of climate change in the southern counties, especially for soybean production.
Specifically for corn, we also found that a long season cultivar is expected to benefit more from
climate change than medium and short season cultivars.
It is important to note that the interpretation of future impacts varied somewhat
depending upon the NARCCAP simulations employed to derive future climate projections. For
example, warmer and drier growing season conditions for the southern UGLR projected by
HRM3-gfdl will more adversely impact corn and soybean yield compared to the relatively mild
and wetter conditions projected by the other NARCCAP simulations (e.g. WRFG-ccsm and
WRFG-cgcm3). We also found that the utilization of different RCMs to downscale GCM output

131

can introduce uncertainty into the impact assessment, in addition to the uncertainty introduced by
the choice of GCM.
More favorable growing conditions in the northern UGLR likely will create an
opportunity to grow corn and soybean with relatively high productivity more widely across the
study region. Interestingly, we did not observe a shift of major production regions to the north as
the major growing areas in the south are expected to still be productive in the future in spite of
the potential slight reduction in grain yield. Instead we observe an expansion of the crop
production in the future. However, farmers in the current major production areas should be aware
of several potential negative impacts of future climate change, including a potential increase in
annual yield variability. The potential opportunity of the ‘new’ production regions in the
northern areas of the UGLR hints that further exploration is needed. Further analysis that
employs more dynamic farming practices (Howden et al., 2007; Tubiello and Rosenzweig, 2008)
would be particularly useful.

132

CHAPTER 5.
Development of Crop Yield Interdisciplinary Model for Regional Climate Change Impact
Assessments: A Case Study for the Upper Great Lakes Region of the United States

In collaboration with
Julie Winkler and Zhengfei Guan

5.1.

Introduction
Crop production is susceptible to climate variability and change as can be inferred from

the impacts of recent climatic trends on crop yield (e.g., Kucharik and Serbin, 2008; Lobell et al.,
2011). The potential consequences of projected future climate change on regional crop
production have been extensively studied around the world as reviewed by White et al. (2011a).
These efforts have focused on the impacts of climate change on crop productivity which directly
influences future world food supply (e.g., Parry et al., 2005; Rosenzweig and Parry, 1994).
Ecophysiological models, commonly known as crop models (White and Hoogenboom,
2011), are regularly employed for climate change impact assessments as summarized by
previous reviews (e.g., Kang et al., 2009; White et al., 2011a). Crop models utilize a set of
mathematical equations derived at a field-scale level to quantify non-linear interactions between
climate, soil, agronomic practices and crop growth and development. Thus, crop models
simulate, although imperfectly, the biophysical responses of a particular crop to environmental
changes at a particular location (Hoogenboom et al., 2004; Meinke et al., 2001). These models
also consider the effects of agronomic practices (e.g., cultivar, planting dates, row and spacing)
on crop yield.

133

A challenging issue in the application of crop models for regional climate change impact
assessments is how to explicitly include the impacts of potential changes in regional farming
management associated with economic stressors. The inclusion of economic factors has been
recommended by the climate impact research and modeling community in recognition of the
significant contribution of economic factors to crop yield variability (Challinor et al., 2009).
Unfortunately, crop models currently are not designed to capture the effects of changes in
economic factors associated with farming management such as the use of machinery, labor, and
pesticides that can influence crop production.
Taking into consideration the limitations of crop models, Kaufmann and Snell (1997)
proposed an empirical model that integrates the outputs of crop models with economic
determinants to capture the consequences of ecophysiological and socio-economic drivers on
crop yield. This ‘hybrid’ model was referred to by Kaufmann and Snell as an “interdisciplinary

model”. The authors developed an interdisciplinary model for corn yield by relating climate
variables for corn phenological stages, with the phenological stages being estimated by a crop
model, representative economic variables (i.e., purchase inputs and loan rate) and technical/scale
variables (i.e. the use of machinery and agricultural acreage). A more recent application of an
interdisciplinary model is the soybean yield model developed by Vera-Diaz et al. (2008) that
combines simulated yields from a crop model, geographic location (i.e., latitude and longitude)
and economic variables (i.e., credit, transports costs) to explain soybean yield spatial variability
in the Brazilian Amazon. Geographic location was included to capture the effects of photoperiod
and environmental gradients on yield; the effects of climatic and edaphic conditions are captured
by simulated yields; and the economic variables adjust the simulated yields to reflect the effects
of economic factors on observed soybean yield.

134

This study proposes an alternative specification for interdisciplinary models that is
derived from an asymmetric production function as proposed by Guan et al. (2006). The Guan et
al. model is essentially a modification of the traditional translog production function. The
uniqueness of the Guan et al. model compared to the traditional translog production function is
that the model specification distinguishes between two types of crop yield, namely: attainable
and actual yield. This classification differentiates yield on the basis of governing factors
(Rabbinge, 1993; vanIttersum and Rabbinge, 1997). Attainable yield is governed by growthdefining factors (i.e., plant characteristics, solar radiation, and temperature) and growth-limiting
factors (i.e., nutrients and water). Actual yield fluctuates following growth-reducing factors such
as weeds, pests, and diseases, in addition to the growth-defining and the growth-limiting factors.
Guan et al. used growth inputs (i.e., land, seed, fertilizer, and water) and dummy variables (farm
and year dummy) to represent growth environment (i.e., farm and weather condition) to estimate
attainable yield and facilitating inputs (i.e., labor, capital, and pesticides) to adjust the attainable
yield to estimate actual yield. Although, a translog production function such as the Guan et al.
model offers an advantage to simulate the effects of possible changes in farming management
strategies on crop productivity, the use of dummy variables prevent the application of Guan et al.
model for climate change impact assessments. The model developed here replaces the attainable
yield component of the Guan et al. model with simulated yield, as the crop model can better
capture the effect of climate on yield compared to the use of dummy variables by Guan et al. The
interdisciplinary model is developed to estimate county-level corn yield in the Upper Great
Lakes Region (UGLR) of the United States. This region is selected as agriculture is considered
as one of key economic activities in the UGLR, and corn is the major agricultural commodity
(Hatfield, 2012; Niyogi and Mishra, 2013; Safir et al., 2008).

135

Future climate change is expected to impact regional corn production in the UGLR (e.g.,
Andresen et al., 2013; Southworth et al., 2000). However, the effects of possible changes in
economic factors have not been included in the previous assessments. Thus, we also demonstrate
the potential application of the interdisciplinary model for climate change impact assessments.
The use of the interdisciplinary model is expected to provide a better interpretation of the
potential future impacts of climate change on crop production compared to the use of crop
simulation models alone.

5.2.

Materials and Methods

5.2.1. Study Area
The Upper Great Lakes Region (UGLR), consisting of the states of Michigan, Minnesota
and Wisconsin is located in the United States Midwest (Figure 5-1) The southern and central
UGLR is part of the US Corn Belt (Aref and Pike, 1998), and the states of Michigan, Wisconsin
and Minnesota rank number 11th, 10th and 4th for corn production in the nation, respectively.
0

-97
-97°

0
-95
-95°

0
-93
-93°

0

-91
-91°

0

-89
-89°

0
0
-87 -85°
-87° -85

0

-83
-83°

0

0

49°
49
0
48°
48
0
47°
47
0
Minnesota
46°
46
0
45°
45
0
44°
44
0
43°
43
0
42°
42 1 cm = 1,319 km
0
-97°

-97

0

0

-81
-81°

49°
49
0
48°
48
0
47°
47
0
46°
46
0
45°
45
0
44°
44
0
43°
43
0
42°
42

Wisconsin
Michigan
0

0

0

0

0

0

0

-95° -93° -91° -89° -87° -85° -83° -81°
-95 -93 -91 -89 -87 -85 -83 -81

Figure 5-1. Location of the Upper Great Lakes Region

136

State Boundary
County Boundary

USA

5.2.2. Model Specification
The asymmetric framework proposed by Guan et al. (2006) distinguishes between two
broad categories of agronomic inputs. Pesticides, labor and capital, which are required to create
favorable conditions for farming activities, are classified as facilitating inputs. On the other hand,
factors that affect plant biophysical processes, such as land, seed, nutrients, and water, are
grouped as growth inputs (Guan et al., 2006). Based on these classifications, the model is
represented as (Guan et al., 2006, p. 206),
y = G(x1 , x 2 , x 3 , h; E) • F(z1 , z 2 , z 3 )

(Eq.5-1)

where, y is crop-yield; x1, x2, x3 and h are growth inputs (i.e., land, seed, fertilizer, and water);

z1, z2, and z3 are facilitating inputs (i.e., labor, capital, and pesticides); E represents the growth
environment (i.e., dummy variables to capture differences in biophysical condition and
management across farms and interannual variations in yield due to varying weather conditions);

G(·) defines attainable yield; and F(·) is a scaling function that has a value from zero to one.
For the modified model proposed here, G(·) in Eq.5-1 is replaced with estimated yield (ya)
obtained from simulations of the CERES-Maize model, a crop model including in the Decision
Support System for Agrotechnology Transfer (DSSAT).

If agronomic practices are held

constant across space and time for the CERES-Maize simulations, the farm dummy variable of
the E term is no longer needed, and if temporal variability is assumed to be smaller than spatial
variability the year dummy can also be omitted.
The scaling function F(·)can be estimated following Guan et al. (2006, p. 208):

F( z1 , z 2 , z 3 ) = exp[-(β 0 + β1z1 + β 2 z 2 + β 3 z 3 ) 2 ]

137

(Eq.5-2)

where, β0 is intercept; β1, β2, and β3 are coefficients for labor, capital, and pesticides.
This function ensures that values of the scaling factor range from zero to one. Eq.5-1 can now be
rewritten as

y = ya • exp[-(β 0 + β1z1 + β 2 z 2 + β 3 z 3 ) 2 ]

(Eq.5-3)

Further modifications can facilitate the application of this model to county-level yield. If
crop production is defined as crop yield multiplied by area, county-level economic costs can be
used instead of unit costs (costs per unit area) and Eq.5-3 can be rewritten as:.

y * area = ya * area • exp[-(β 0 + β1Z1 + β 2 Z 2 + β 3 Z 3 ) 2 ] • exp(e)

(Eq.5-4)

where, Z1, Z2, and Z3 are the county-level costs of facilitating inputs such as chemicals,
machinery, and labor; e is the error term. The term “area” on the left hand side of the equation
can be removed by applying the natural logarithm, and Eq.5-4 can be written as:

ln(y/ya) = - ( β 0 + β1Z1 + β 2 Z 2 + β 3 Z 3 ) 2 + e

(Eq.5-5)

The model predictors (i.e., the Z terms) can be extended to include additional variables
beyond the three terms (chemicals, machinery and labor) originally employed by Guan et al.
(2006), in order to incorporate additional facilitating factors that are not captured in the crop
model simulations. For this study, total cost of fertilizer (Z4) and total area of agricultural land
(Z5) in each county in the UGLR are added to Eq.5-5.

ln(y/ya) = - ( β 0 + β1Z1 + β 2 Z 2 + β 3 Z 3 + β 4 Z 4 + β 5 Z 5 ) 2 + e

(Eq.5-6)

Fertilizer cost was added to capture spatial and temporal variations in the amount of fertilizer
applied across the UGLR. Although, DSSAT allows users to include fertilizer in the crop
simulations, obtaining fertilizer information, including the type of fertilizer application (e.g.,
138

nitrogen, potassium), for all counties and years is difficult. Additionally, non-limiting nutrients
are often assumed for DSSAT simulations in order to isolate the consequences of climate
fluctuations on yield (e.g., Andresen et al., 2001). In this study, we also simulated DSSAT yields
under the assumption of non-limiting nutrients. The amount of agricultural land was added to
capture variations by county of the amount of land available or suitable for agriculture. The
inclusion of this term is based in part on Kaufmann and Snell’s (1997) assertion that crop
models have a limited ability to simulate the scale effects that the amount of land available for
agriculture may have on the choice of farming management practices.
The final form of the model, with “^” indicating estimated terms is:

ˆ
ˆ
ˆ
ˆ
ˆ
ˆ
ˆ
y = ya • exp [-(β 0 + β1Z1 + β 2 Z 2 + β 3 Z 3 + β 4 Z 4 + β 5 Z 5 ) 2 ]

(Eq.5-7)

5.2.3. Model Implementation
An overview of the data preparation and model implementation steps is shown in Figure
5.2 and discussed in more detail below.

Physical environment:
temperature, solar
radiation, precipitation,
soil information

DSSAT simulation
at the county level

Farming
management and
technology

Observed
countylevel yield

Simulated
Yield

Ratio of Observed to
Simulated Yield
Predictand
Non-linear regression
analysis to estimate
the model coefficients

Yield estimation
Figure 5-2. Components of the interdisciplinary model

139

Predictors
County level
economic costs:
chemicals,
machinery,
labor, and
fertilizer

Estimated countylevel yield

5.2.4. Economic Data
County-level economic data were obtained from Quick Stats 2.0 provided by the National
Agriculture Statistics Service, United States Department of Agriculture (NASS-USDA, 2011).
This database includes information from the agricultural censuses conducted in 1997, 2002 and
2007. We extracted from Quick Stats 2.0. total corn yield and planted area; total costs of
chemicals, machinery, labor (contract/hire), and fertilizers; and economic price indices with
based year of 1990-1992 for each economic variable, and agricultural land area per county.
County-level economic data reported by the NASS-USDA are aggregated across all
major crops within a county. The proportion of total cost for each economic variable specific to
corn was roughly estimated following the procedures used by Guan et al. (2006) to estimate the
share of farm-level labor and capital for individual crops. Itemized production costs by major
crop for the three analysis years (1997, 2002, and 2007) published by the Economic Research
Service – United States Department of Agriculture (ERS-USDA, 2011) for the three ERS-USDA
farm resource regions encompassing the UGLR (see Figure 5-3) were employed to calculate the
total cost proportion for corn production. Based on planted area, six out of the ten crops included
for the farm resource regions were considered to be major crops in the UGLR. These crops are
barley, oats, wheat, sugar beets, corn, and soybean. The total cost for each economic variable
i

(i.e., chemicals, machinery, labor and fertilizer) was estimated by multiplying unit costs by the
county-level planted area for each of the six major crops with detail procedure of the calculation
is explained below.

140

REGIONS
Heartland
Northern Crescent
Northern Great Plain

Figure 5-3. The United States farm resource regions used to calculate the share of production
costs for corn. Data Source: the Economic Research Service – United States Department of
Agriculture (ERS-USDA, 2011).

Each county within a United States farm resource region was assumed to have the same
unit cost for a particular crop (PCAkj, where k refers to crop type and j to the U.S. farm resource
region). PCAkj was multiplied by the total planted area for a particular crop in each county to
estimate the total costs for that crop by county (TPCki where i refers to county). Next, the total
costs (TPCki) for the six major crops were totaled, giving the total costs by county (TPCi). The
proportion (Ski) by county of an economic cost (i.e., chemicals, machinery, labor and fertilizer)
was calculated by dividing TPCki by TPCi. The calculations were performed separately for each
analysis year (i.e, 1997, 2002, and 2007) and for each economic variable (i.e., chemicals,
machinery, labor and fertilizer). Estimated Ski varies spatially and temporally, in response to
variations in planted area by crop and year. The estimation procedure is summarized as:

TPCki = Aik*PCAkj

(Eq.5-8)

141

where, PCAkj is the unit cost of an economic variable (i.e., chemicals, machinery, labor and
fertilizer) for crop k in NASS-region j, Aik is the total planted area of crop k for county i, and
TPCki is the total cost of an economic variable for crop k at county i.

Ski = TPCki/TPCi

(Eq.5-9)

where, Ski is the proportion of an economic variable for crop k at county i, TPCi is the total costs
of an economic variable for all major crops by county i.

5.2.5. Observed Yield
Observed county-level corn yield was standardized by dividing total corn production by
total area planted to corn. We chose to use total planted area instead of harvested area as the
denominator because, especially in years with environmental stress (i.e., extreme weather
conditions), some of the planted area is often not harvested which inflates the standardized yield
and reduces the spatial and temporal variations in yield.
Temporal trends in the standardized observed county-level yields were removed so that
the observed yield is comparable with simulated yield, as suggested by previous studies (e.g.,
Andresen et al., 2001; Jagtap and Jones, 2002). Advancement in farming practices and
technology, such as new cultivars and pest management, substantially contribute to yield
increases with time (Egli, 2008). The observed yields from 1990 through 2008 were detrended
separately by county following the procedure applied by Tannura et al. (2008). The linear trend
between observed yield as the predictand and year as the predictor was calculated and used to
adjust observed yields to a base year of 1990, chosen as it corresponds to the time of the
economic indices (see above). The detrended county-level observed yields for 1997, 2002, and
2007 were extracted for the model development.

142

5.2.6. Crop Model Simulation
The CERES-Maize model included in DSSAT version 4.5 (Hoogenboom et al., 2010)
was used to simulate corn grain yield for all UGLR counties. The CERES family has been
implemented worldwide to simulate crop responses to environmental changes as summarized by
Jones et al. (2003) and specifically to climate change as reviewed by White et al. (2011a). Data
inputs required for the model simulations are climate variables, soil information, and farming
practices. The primary required climate variables are daily solar radiation, maximum and
minimum air temperature, and precipitation. Soil data includes general physical soil properties
(e.g., soil texture, soil classification) and layer-specific soil information (e.g., soil fertility,
drainage upper and lower limit). Farming practices include agronomic practices such as crop
variety, row spacing, and planting date. The data preparation for the DSSAT simulation
generally followed the procedures described in detail in Chapter 4 and summarized below.
DSSAT simulations were performed for the period 1990 to 2008, which encompasses the three
years (1997, 2002, and 2007) with available economic data. The year of 1990 was selected as
the beginning year for the simulations considering the base year of the economic indices used to
detrend the economic variables.

(a)

Climate data
Climate data (i.e., precipitation and maximum and minimum temperature) was obtained

from 36 representative climate stations (Figure 5-4) from the United States Historical Climate
Network-USHCN (Menne et al., 2012b; Menne et al., 2009). These representative stations were
selected and assigned to individual counties with the assistance of a climate regionalization that
grouped USHCN stations based on their deviations from the regionally-averaged annual cycle of
precipitation and maximum and minimum temperature (see Chapter 2).

143

Figure 5-4. The representative climate stations (red dots) assigned to the UGLR counties for the
interdisciplinary model development. The colors indicate the counties assigned with the same
representative climate station.

One criterion in the selection of the representative stations was a small amount of missing
observations. When missing observations did occur in the time series of the representative
USHCN stations, “filled in” values were obtained from the Iowa Environmental Mesonet (IEM)
network (IEM, 2013), either for the same station or for a nearby station. . IEM recently published
temporally complete series many climate stations in the Midwest including stations located in
Michigan, Wisconsin and Minnesota. IEM filled in missing observations using an inverse
distance interpolation applied to climate data from NCEP Stage IV precipitation, airport weather
data (ASOS/AWOS), and available NWS COOP reports (Drayl, 2013, personal communication).
Daily solar radiation, an important variable for the DSSAT simulation but not recorded
by the climate stations, was estimated using the Hunt et al. (1998) mechanistic radiation model,
parameterized using concurrent observations of daily maximum and minimum temperatures,

144

precipitation and solar radiation for 1990-2008 at Hancock, Wisconsin, located roughly in the
center site of the UGLR. The mechanistic radiation model was selected to estimate solar
radiation as a systematic evaluation of various daily solar radiation sources (Chapter 3)
suggested that the mechanistic approach is a viable option for the generation of daily solar
radiation required for crop process models. The mechanistic model utilizes latitude to estimate
daily global solar radiation, and an empirical relationship derived from temperature and
precipitation is used to calculate atmospheric transmittance (Hunt et al., 1998; Liu and Scott,
2001).
Atmospheric CO2 concentrations for the DSSAT simulations were set at 355 ppm, the
CO2 level for early 1990 as reported by the Intergovernmental Panel on Climate Change-Data
Distribution Center (IPCC-DDC, 2011).

(b)

Soil Data
County-level soil data were extracted from the STATGO database published by the

Natural Resources Conservation Service of the United State Department of Agriculture (NRCSUSDA) (Soil Survey Staff, 2010) for the dominant soil type in each county based on areal
coverage. Soil parameters included soil classification, slope, drainage, runoff potential, the soil
texture, organic content, pH and cation exchange capacity. SBuild, a DSSAT supporting package
(Uryasev et al., 2003), was used to calculate specific soil information, such as drainage upper
and lower limits which are not available in STATGO, and to prepare soil inputs for the DSSAT
simulations. Soil color and fertility, which are also required by SBuild, were obtained based on
soil texture and taxonomy published by the National Cooperative Soil Survey, (NCSS, 2011) and
the generic soil database available in DSSAT.

145

(c)

Cropping Scenarios
CERES-Maize was run under rainfed conditions using a medium season corn cultivar.

The corn cultivar was selected after evaluating the performance of multiple corn cultivars
employed by previous studies within the UGLR (i.e., Andresen et al., 2001; Southworth et al.,
2000) in capturing the spatial variability of observed corn yield in (see Chapter 4). The cultivar
characteristics are presented in Table 5-1.

Table 5-1. Characteristics of medium season corn cultivar used in CERES-Maize simulations
0

P1

degree days (base 8 C) from emergence to end juvenile phase

200

P2

photoperiod sensitivity (0-1)

0.3

P5

degree days (base 8 C) from silking to physiological maturity

800

G2

potential kernel number

700

G3

potential kernel growth rate (mg/d)

6.3

0

0

PHINT degree days required for a leaf tip to emerge (phyllochron interval) ( C d)

38.9

Planting densities were determined at 75 cm spacings with plant populations of 6
2

plants/m following Andresen et al. (2001). Planting date was determined automatically using
the DSSAT automatic planting option that initiates planting when, during a specified planting
0

window, the percent soil water and soil temperature in the top 10 cm are at least 20% and 10 C,
st

respectively. The simulations were started on January 1 with soil water initialized at the drained
upper limit. Five different planting windows were defined by offsetting the starting date of the
st

th

planting window by roughly 15 days from April 1 to June 10
th

th

st

th

th

th

st

th

(i.e., April 1 -June 10 , April
th

th

16 -June 10 , May 1 -June 10 , May 16 -June 10 , and May 31 -June 10 ). The end date
of the planting window reflects the typical observed range of planting dates for corn in Michigan,
146

Minnesota and Wisconsin (NASS-USDA, 1997). Harvesting was considered to occur when corn
reached physiological maturity. The highest annual simulated yields from the different planting
windows were selected for the development of the interdisciplinary model. The DSSAT
simulations assumed no change in agronomic practices over time along with non-limiting
nutrient conditions.

(d)

DSSAT Simulated Attainable Yield
As discussed above, the DSSAT simulations provide the estimates of “attainable yield”

for the interdisciplinary model development, and attainable yield is assumed to be larger than
observed yield. When the simulated corn yields for 1997, 2002 and 2007 are compared with the
observed yields reported by NASS-USDA, a modest number (150 out of a total of 587) of
county-year combinations are seen to have observed yield greater than simulated yield (Figure
5-5). No distinct spatial or temporal variation is seen in the distribution of observed county
yields greater than observed yield. These differences could reflect a number of factors such as
the use of irrigation in some counties, higher yields on soils within a county that are not
dominant in terms of areal coverage, the use of different (e.g., long season) cultivars, uncertainty
in the county-level yield values, among others. The county-year combinations with DSSAT
simulated yield less than observed yield were removed from the development of the
interdisciplinary model.

147

Observed Yield

2007

2002

1997

DSSAT Yield

1997

2007

2002

Ratio = DSSAT/Observed

2007

2002

1997

Yield Ratio

Yield (kg/ha)
<=1500
1500 - 2500
2500 - 3500
3500 - 4500

4500 - 5500
5500 - 6500
6500 - 7500
>7500

<=0.50
0.50 - 0.75
0.75 - 1.00
1.00 - 1.25
NA

1.25 - 1.50
1.50 - 1.75
1.75 - 2.00
>2.00

Figure 5-5. Observed yield (top), DSSAT simulated yield (middle), and the ratio of simulated to
observed yield (bottom) for the UGLR for 1997, 2002 and 2007

5.2.7. Model Estimation
The nonlinear least square (NLS) estimator was employed to solve Eq.5-6 giving the
nonlinear form of the equation. Initial values of the model coefficients are needed as NLS solves
nonlinear equations iteratively until convergence is achieved (Greene, 2003). To obtain initial
estimates, the error term in Equation 5-6 was assumed to be zero and multiple regression was
applied to solve the liner form of the equation (Eq.5-10).

148

ˆ
ˆ
ˆ
ˆ
ˆ
ˆ
ln(ya/y) = β 0 + β1Z1 + β 2 Z 2 + β 3 Z 3 + β 4 Z 4 + β 5 Z 5

(Eq.5-10)

Across-section data analysis (i.e., varied in time and place) is likely to introduce
heteorocedasticity as explained by Gujarati (2003, p.401). Although, heterocedasticity does not
affect the values of the model coefficients, the standard errors and the t-statistics may be no
longer valid (Gujarati, 2003). To deal with this issue, heteroskedasticy consistent (HC)
estimators included in the sandwich package (Zeileis, 2006) of the R statistical software was
employed to obtain robust estimation of the standard errors and the t-statistics.
Given the non-linear form of the interdisciplinary model, elasticity analysis was
performed to evaluate the contribution of each economic predictor to the yield deviation assigned
as the predictand in Eq.5-6. First, the derivation of the model (Eq.5-6) with respect to each
economic predictor was obtained separately. Second, using the derived models, the elasticity
values of each economic predictor were calculated at the mean values of each economic variable
( Zx ). Below is the example of the elasticity calculation for pesticides (Z1). The standard errors
for the elasticity values were computed using the delta method (Greene, 2003, p.70) available in
R statistical software.

Z1
Z1
d(ln(y/ya))
•
= - 2 β1 • ( β 0 + β1 Z1 + β 2 Z 2 + β 3 Z 3 + β 4 Z 4 + β 5 Z 5 ) •
dZ1
ln(y/ya)
ln(y/ya)
(Eq. 5-11)

5.2.8. Application for a Climate Change Impact Assessment
The interdisciplinary model is used to evaluate the sensitivity of future corn yield to
variations in facilitating inputs in addition to projected climate change. As an example, we
evaluated the sensitivity of yield changes by the mid century (2041-2070) with respect to the

149

baseline period (1971-2000) to a reduction from the “current” investment in chemical
applications of 30%, 50% and 70%. The reduction scenario was selected to represent possible
changes in the regulation of chemical applications in the future. The values for the “current”
investment were obtained from the average values of available economic data for the three
analysis years (i.e., 1997, 2002 and 2007). Attainable corn yields for the future and the baseline
period by county were estimated from the DSSAT simulations of corn yield by county using the
projected temperature and precipitation obtained from climate scenarios developed from the
North American Regional Climate Change Assessment Program (NARCCAP) simulations (see
Chapter 4 for more information on the climate scenario development and DSSAT simulations).
Atmospheric CO2 concentration for the crop model simulations for both the future and the
baseline period was set constant at 370 ppm. For attainable yield, we used the median values of
the 30-year future and baseline periods of simulated corn yields. Specifically for the future
period, the county-level median corn yields were obtained by averaging the simulated countylevel median corn yield obtained from the DSSAT runs for all eight of the NARCCAP-derived
climate projections.

5.3.

Results
The interdisciplinary model explains about 62% of yield variability over the UGLR for

the three analysis years (Table 5-2). The model coefficients for chemicals, labor, and fertilizer
have a negative sign, whereas the coefficients for the other predictors are positive. A negative
coefficient indicates that additional chemicals, labor, or fertilizer will decrease the deviation
between simulated DSSAT yields and observed yields. In other words, observed yields will be
closer to DSSAT yields as chemical, labor or fertilizer inputs increase. The opposite occurs for
machinery and amount of agricultural land. The elasticity analysis suggests that labor contributes
150

less to the yield deviations compared to the other economic variables, and the significant test for
the model coefficients shows that only the labor term is insignificant.

Table 5-2. Model coefficients and elasticity analysis for the interdisciplinary model of countylevel corn yield for the UGLR parameterized using available data of 1997, 2002 and 2007
Model coefficients

Elasticity Analysis

Uncorrected

Corrected

Uncorrected

Corrected

Intercept

7.53E-01***

7.53E-01***

Chemical

-5.28E-08*

-5.28E-08**

-0.278*

-0.278**

Machinery

4.76E-09***

4.76E-09***

0.468***

0.468***

Labor

-7.36E-09

-7.36E-09

-0.067

-0.067

Fertilizer

-7.30E-08***

-7.30E-08***

-0.661***

-0.661***

Agricultural Land

9.21E-07***

9.21E-07***

0.224***

0.224***

2

R -fit
0.62
Note: *** significant at 1%, ** significant at 5%, *significant at 10%

The interdisciplinary model was applied to estimate the spatial distribution of corn yield
for the UGLR using the simulated DSSAT yield and economic inputs for 1997, 2002 and 2007
(Figure 5-6). Visual comparison suggests that the estimated yields obtained from the
interdisciplinary model better capture the spatial yield variability than the simulated DSSAT
yields for each of the three years (Figure 5-6), especially in the southern UGLR. Statistical
indices (Pearson’s product moment correlation coefficient, the Willmott Index of Agreement
(Willmott, 1981)) clarify that the estimated yields obtained from the interdisciplinary model
better imitate the spatial pattern across the UGLR than do the simulated DSSAT yields. Also, the
deviations in the mean and standard deviation between observed and estimated yield are smaller
for the interdisciplinary model than those for the DSSAT simulation (Table 5-3).

151

Observed

1997

2007

2002

DSSAT

2007

2002

1997

INT

2007

2002

1997

DATA

1997

2007

2002

Data for Model Development

Yield (kg/ha)
<=1500
1500 - 2500
2500 - 3500
3500 - 4500

DSSAT < Observed
DSSAT >= Observed

4500 - 5500
5500 - 6500
6500 - 7500
>7500

NA

Figure 5-6. Comparison between observed county-level yields (Obs), estimated yields obtained
from DSSAT simulation (DSSAT) and estimated yields obtained from the interdisciplinary
model (INT) for 1997, 2002 and 2007. The three maps in the last row (Data) display county-level
data employed for the model development.

152

Table 5-3. Statistical comparison between observed yields (Obs) and estimated yield (Est)
obtained from DSSAT simulation and the Interdisciplinary model across the UGLR
Variable
Willmott Index
Correlation
Mean.Obs
Mean.Est
StDev.Obs.
StDev.Est.

DSSAT simulation
1997
2002
2007
0.57
0.59
0.63
0.40
0.39
0.44
5439.7 5471.9 4573.1
7832.7 6746.1 5247.9
1533.9 1556.1 1738.1
2672.8 2081.4 2350.0

The Interdisciplinary Model
1997
2002
2007
0.77
0.74
0.74
0.62
0.61
0.61
5439.7 5471.9
4573.1
5062.2 4597.8
3696.5
1533.9 1556.1
1738.1
1908.3 1686.7
1817.0

As noted above, the interdisciplinary model was also used to explore the potential
impacts of a reduction in the amount of chemical application under projected climate change for
the mid century. Figure 5-7 shows that projected climate change under the reference CO2 level
(370 ppm) will slightly decrease corn yield for the majority UGLR counties, particularly those
located in the southern UGLR, and moderately (up to 50%) decrease corn yield in southern
Michigan. When chemical inputs are reduced, the interdisciplinary model suggests that the
number of counties in the southern UGLR with a moderate (25-50%) reduction in yield will
increase. A similar reduction is not seen in the northern UGLR (Figure 5-7).

153

Current

30%

50%

70%
Yield Change (%)
<=-50
-50 - -25

-25 - 0
0 - 25

25 - 50
50 - 100

100 - 150
>150

NA

Figure 5-7. Sensitivity of yield changes under projected climate change for the UGLR by the mid
century (2041-2070) to the reduction in the amount of chemical application by 30% (second
row), 50% (third row) and 70% (fourth row). A CO2 concentration of 370 ppm was assumed for
the DSSAT simulations.

5.4.

Summary and Discussion
The impacts of projected climate change on agricultural production have been

extensively studied during the past decade. Crop models that are designed to operate at a field
scale are commonly applied to evaluate the complexity of crop responses to future climate
change. A challenging issue associated with applications of crop models, particularly at a
regional scale, is how to incorporate economic factors that reflect farming management but are
not included in the crop model simulation. This study attempted to improve the utilization of a

154

field-based crop model called CERES-Maize incorporated in the DSSAT system to estimate
county-level corn yield for the Upper Great Lakes Region (UGLR). The model was developed
by integrating simulated DSSAT yields and county-level economic costs, namely: chemicals,
machinery, labor, and fertilizer, and the amount of agricultural land per county.
An underlying assumption for the model development is that observed yields (i.e., actual
yield) should always be smaller than simulated DSSAT yields (i.e., attainable yield). The
reasoning behind this assumption is that growth limiting factors such as pest and disease that can
reduce crop yields are not included in the DSSAT simulations. Simulated DSSAT yield primarily
is dictated by climatic conditions, i.e., solar radiation, temperature and precipitation, although
soil factors and some management practices (e.g., crop cultivar) also are important. Although
simulated yields tended to be higher than observed yields for most counties and years, there were
a few counties during the three-year model development period with observed yields higher than
simulated yields. In particular, DSSAT yields for 2007 underestimated county yield in the Lower
Peninsula Michigan. The possible reason is that the DSSAT, which was run under the
assumption of rainfed conditions, may exacerbate the effect of the 2007 drought, considered as
one of the worst drought in the last two decades in Michigan (Andresen, 2008), on yield. On the
other hand, farmers may apply irrigation to combat with the drought condition. Assumptions
made for agronomic practices used for the model simulation also may contribute to the deviation
between simulated and observed yields as uniform agronomic practices was assumed for the
entire region. Smaller simulated yields than observed yields were also found for some years by
previous studies that applied crop models to simulate maize (e.g., Safir et al., 2008), soybean
and/or wheat (e.g., Alexandrov et al., 2002; Jagtap and Jones, 2002) yield for a region.

155

The model coefficients for the interdisciplinary model were estimated based on empirical
relationships of the deviation between simulated yields and observed yields as the predictand and
county-level economic costs as the predictors. The model coefficients for chemicals, labor and
fertilizer have a negative sign which means they have a positive impact on the estimated yield,
although labor contributes less to the estimated yield than the other two economic variables. The
signs for chemicals and fertilizer are reasonable as chemicals and fertilizer are applied to create
favorable growing conditions for corn growth and development. Chemicals (i.e., pesticide) are
applied to manage pest and disease infestations that can decrease corn yields. Fertilizer is applied
to supply adequate nutrients to support crop growth and development. Consequently, application
of chemicals and fertilizer is expected to result in estimated yields (i.e., actual yield) closer to
simulated yields (i.e., attainable yield). For machinery and amount of agricultural land per
county, the signs of the coefficients are positive which means any increase in these two variables
results in larger deviations between simulated and observed yields. Machinery, likely functions
as a capital stock in the interdisciplinary model, partially captures the scale of corn production,
as does the amount of agricultural land.
The estimated corn yields from the interdisciplinary model better capture the spatial
variability of observed county-level corn yield in the UGLR compared to the simulated DSSAT
yields. For this application, the interdisciplinary model essentially “corrects”, using a scaling
factor, simulated yields so that they are closer to observed yields. The interdisciplinary model
was also applied, as an example, to investigate the sensitivity of yield changes under projected
climate change to a reduction in chemical applications. The results revealed that decreasing the
amount of chemicals applied to corn production can exacerbate the potential negative impacts of
climate change on corn yields in the southern UGLR which is currently the major production

156

region. This consequence may happen because warmer conditions in the future will shorten the
time to maturity which eventually will reduce the period for grain yield production as discussed
in Chapter 4. Furthermore, the decrease in chemical applications will increase the yield deviation
between simulated DSSAT yields (i.e., attainable yield) and actual yield, which means actual
grain yield will decrease as the chemical application decreases. This example of applying the
interdisciplinary model for climate change impact assessments suggests that the interdisciplinary
model can provide an option for extending the use of crop model simulations for regional climate
change impact assessments. Using the interdisciplinary model, we can experiment with different
scenarios of possible changes in the county-level economic costs, rather than assume, as is done
for most assessments, that the current farming management also applies to a future period
(Tubiello et al., 2002).

5.5.

Conclusion
The impacts of climate change on crop yield have been extensively evaluated. In this

study, we developed an interdisciplinary model based on an asymmetric production function that
distinguishes between attainable and actual yield to model county-level corn yield for the Upper
Great Lakes Region (UGLR). The model integrates observed (i.e., actual yield) and simulated
(i.e., attainable) yields with county-level economic costs. The model parameters were estimated
using a nonlinear least square (NLS) estimator.
This study shows that the interdisciplinary model can improve upon the utilization of
simulations from crop models alone and can be used to evaluate the sensitivity of crop yield to
possible changes in a crop production system as reflected by changing county-level economic
costs. As an example, we applied the model to assess a simple scenario of reduced chemical
applications for controlling pest and disease under a future perturbed climate. The analysis

157

revealed that reduced chemical applications may exacerbate the potential negative impacts of
climate change on corn yields, especially in the southern UGLR.

End Notes
i

When a specific agricultural input cost was not available for a resource region, the
corresponding input cost from the closest resource region or an older regional classification was
used. The older classification of farm resource regions is found in the online documentation
provided by ERS-USDA (2011b).

158

CHAPTER 6.
Summary and Conclusion

This study evaluates the potential impacts of projected climate change for the mid
century (2041-2070) on county-level corn and soybean production in the Upper Great Lakes
Region (UGLR) of the United States, encompassing the states of Michigan, Wisconsin and
Minnesota. Future climate scenarios were derived from simulations obtained from eight
“combinations” of RCMs and GCMs released by the North American Regional Climate Change
Assessment Program (NARCCAP). The CERES-Maize and CROPGRO-Soybean models
included in DSSAT were employed to simulate corn and soybean production for historical
(1971-2000) and future (2041-2070) periods. CO2 concentrations for the crop simulations
included a reference level (370 ppm) and two elevated CO2 concentrations (490 and 635 ppm).
A major challenge in the application of crop models such as DSSAT is the availability of
daily climate data (e.g., temperature, precipitation and solar radiation), due to the relatively
coarse and non-uniform spatial distribution of climate observing stations. This study proposed
the utilization of a climate regionalization procedure to group climate stations into subregions
with similar characteristics, based on deviations from the regionally-averaged annual cycle of
maximum and minimum temperature and precipitation. The climate regionalization can be
applied to help choose representative climate stations for an impact assessment and to assign
these stations to individual counties based on their geographic proximity. This approach is an
improvement on previous assessments for regional agriculture that selected a small number of
climate stations based on data availability or their location with respect to major agricultural
production areas. A concern for these earlier studies is that the selected climate stations may not

159

fully capture the spatial variability of climate conditions across the study region. Selecting
representative stations based on a climate regionalization helps to ensure that the spatial
gradients of temperature and precipitation across the study area are captured. Additionally, this
approach provides an alternative to the utilization of gridded datasets for regional crop model
simulations. Gridded data need to be used cautiously as they are extremely sensitive to the
density and inhomogeneities of the original observed climate series.
To overcome the limited availability of daily solar radiation, this study systematically
evaluated the sensitivity of simulated corn and soybean yield to different solar radiation sources.
The sources considered included point-based radiation estimates from empirical and mechanistic
models and a stochastic weather generator, and radiation estimates obtained from available
satellite, reanalysis and regional climate model gridded datasets. The comparison of different
daily solar radiation estimates provides a guideline to help select an appropriate daily solar
radiation source for a specific application. A mechanistic radiation model that estimates daily
solar radiation based on maximum and minimum temperature and precipitation as inputs was
selected for assessing the potential impact of future climate change on crop production in this
study. Comparison of simulated yields obtained from observed solar radiation and from
estimated daily solar radiation obtained from the mechanistic model showed relatively small
yield differences. Additionally, the mechanistic model maintains the daily relationships among
climate variables, is easily implemented for different locations in the UGLR, and can be readily
applied to estimate daily solar radiation for a future period using projected maximum and
minimum temperature and precipitation as inputs.
A detailed spatial analysis of the potential climate change impacts on corn and soybean
production in the UGLR provides a more rigorous exploration of the spatial distribution of the

160

potential impacts compared with previous studies that employed a limited locations to represent
the entire region. For the mid century, assuming the reference CO2 level of 370 ppm for the
DSSAT simulations, the assessment revealed that corn and soybean yield for the northern part of
the study region likely will increase due to more favorable growing season conditions than at
present; whereas a small reduction in yield due to a shorter time to maturity may occur in the
southern part which is currently the major crop production area within the UGLR. Elevated CO2
concentration is expected to benefit crop yield, particularly soybean yield.
In contrast to previous work, the climate change impact assessment conducted in this
dissertation employed an ensemble of climate change scenarios developed from simulations from
different RCM-GCM “combinations” where the RCMs were used to downscale GCM output,
with lateral boundary conditions obtained from the GCMs. This assessment illustrates that the
utilization of different RCMs and/or GCMs provides varying future climate projections and
implies that the direct use of outputs from a single or few GCMs may not capture well the
uncertainty associated with climate change impact assessments. While recognizing that the
coarse resolution of GCMs is not sufficient to capture regional climate variation, climate impact
researchers also need to be aware of the additional uncertainty introduced by the choice of RCMs
to dynamically downscale GCM outputs. The use of an ensemble of GCMs and RCMs as
implemented by NARCCAP enhances our understanding of the uncertainty introduced into
climate impact assessments by the climate scenarios.
This dissertation also contributes to an increased understanding of the potential spatial
shifts in crop production and of future temporal yield variability. A potential northward shift in
crop production has been highlighted by previous studies at the national scale for the United
States. The findings of this county-level analysis indicate that more favorable growing conditions

161

by the mid century likely will benefit crop production in the northern UGLR where some
counties may produce relatively high corn and soybean yields especially under elevated CO2
concentrations. Crop yield in the southern UGLR is likely to remain high, partly due to the
positive impacts of elevated CO2 concentrations. However, producers in these areas should be
aware that future climate change may result in somewhat increased temporal yield variability, as
indicated by a projected small increase in the coefficient of variation by the mid century.
Although this dissertation uses the UGLR as a case study for evaluating the regional
impacts of climate change on crop production, the results imply that the consequences of
projected climate change by the mid century for high latitude regions can be complex. Climate
change cannot be expected to always benefit crop production in high latitude regions because of
the complexity of crop responses to spatial variations in temperature and precipitation changes.
Regional soil variations also add to the complexity of crop responses to future climate change.
All climate impact assessments have limitations, and the one presented in this dissertation
is no exception. One limitation is that the NARCCAP simulations employed a single emissions
scenario (i.e., SRES A2), thus the uncertainty introduced by the choice of greenhouse gas
emissions scenario is not considered. Another limitation is the use of the delta method to modify
daily climate observations (i.e., maximum and minimum temperature and precipitation) by
projected monthly changes calculated from the NARCCAP simulations, as future changes in
climate variability are not considered.
The underlying assumptions of the DSSAT simulations also need to be considered. The
simulations do not include the effects of pest and disease infestations and exposure to other
factors affecting yield variability such as economic exposures. Additionally, the effects of
carbon fertilization are simulated under rainfed conditions, whereas different environmental

162

conditions and agricultural management practices such as irrigation may influence the crop
response to elevated CO2 concentration. Except for planting date, farming management and
technology (i.e., cultivar, row and spacing, and planting density) employed for DSSAT
simulations were held constant for the historical and future periods. This allows the
consequences of climate change on crop production to be isolated, but is unrealistic in that
management and technology will continue to evolve in the future with large impacts on crop
production.
A prototype interdisciplinary model that integrates simulated yields obtained from
DSSAT simulations with economic determinants was also developed to overcome some of the
limitations of the DSSAT simulations. This model was purposed to explore the contribution of
economic determinants (i.e., costs of pesticides (chemicals), machinery, labor and fertilizer) to
corn yield variability in the UGLR. The model was developed using an asymmetric production
function that distinguishes two yield types, namely: attainable and actual yield. The attainable
yield was obtained from DSSAT simulations, whereas actually yield was acquired from reported
observed yields The interdisciplinary model was shown to improve the utilization of the DSSAT
(i.e., CERES-Maize) simulations on a regional scale and offers an approach to evaluate not only
the potential impacts of climate exposures but also economic stresses on crop production.
Although, the interdisciplinary model was developed only for corn yield in the UGLR, the
procedures can be applied to other crops and regions.
Overall, this study suggests that more favorable growing conditions will increase corn
and soybean yields in the northern UGLR especially under elevated CO2 concentrations. The
positive impacts of higher CO2 levels will also counteract, or at least minimize, potential small

163

reductions in crop yields projected for the southern UGLR. Given that the United States is one
of the major players in world grain markets, potential changes in corn and soybean production
across all parts of the United States, including the UGLR, will significantly influence future
world food supply. Indeed, additional grain production from high latitude locations such as the
northern UGLR can help offset the potential adverse impacts of climate change on crop
production in low latitude regions. Hopefully, this dissertation will stimulate further research that
applies or extends the methods used here in order to devise plausible adaptation strategies that
can minimize the risks and maximize the benefits of climate change.

164

BIBLIOGRAPHY

165

BIBLIOGRAPHY

Abraha, M.G. and Savage, M.J., 2008. Comparison of estimates of daily solar radiation from air
temperature range for application in crop simulations. Agricultural and Forest
Meteorology, 148(3): 401-416.
Afshin, S. and Gerrit, H., 2003. Minimum data requirements for parameter estimation of
stochastic weather generators. Climate Research, 25(2): 109-119.
Alexandrov, V., Eitzinger, J., Cajic, V. and Oberforster, M., 2002. Potential impact of climate
change on selected agricultural crops in north-eastern Austria. Global Change Biology,
8(4): 372-389.
Alijani, B., Ghohroudi, M. and Arabi, N., 2008. Developing a climate model for Iran using GIS.
Theoretical and Applied Climatology, 92(1-2): 103-112.
Almaraz, J.J., Mabood, F., Zhou, X.M., Gregorich, E.G. and Smith, D.L., 2008. Climate change,
weather variability and corn yield at a higher latitude locale: Southwestern Quebec.
Climatic Change, 88(2): 187-197.
Andresen, J., 2008. Growing Season Weather Summary, Michigan 2007-2008 Highlights,
National Agricultural Statistics Service, Michigan Field Office, Michigan Department of
Agriculture.
Andresen, J.A. et al., 2013. Potential Future Impacts of Climate Change on Row Crop
Production in the Great Lakes Region. In: S.C. Pryor (Editor), Climate Change in The
Midwest: impacts, risks, vulnerability, and adaptation Indiana Univerrsity Press,
Bloomington, pp. 82-91.
Andresen, J.A., Alagarswamy, G., Rotz, C.A., Ritchie, J.T. and LeBaron, A.W., 2001. Weather
impacts on maize, soybean, and alfalfa production in the Great Lakes region, 1895–1996.
Agronomy Journal, 93: 1059-1070.
Andresen, J.A., Alagarswamy, G., Stead, D.F., Cheng, H.H. and Sea, W.B., 2000. Agriculture.
In: P.J. Sousounis and J.M. Bisanz (Editors), Preparing for a Changing Climate: The
Potential Consequences of Climate Variability and Change. Great Lakes Regional
Assessment, University of Michigan, Atmospheric, Oceanic and Space Sciences
Department, Ann Arbor, pp. 69-76.
Andresen, J.A. and Winkler, J.A., 2009. Weather and climate of Michigan. In: R.J. Schaetzl, J.T.
Darden and D. Brandt (Editors), Michigan Geography and Geology. Pearson Custom
Publishing, Boston, pp. 288-314.

166

Apipattanavis, S., Bert, F., Podesta, G. and Rajagopalan, B., 2010. Linking weather generators
and crop models for assessment of climate forecast outcomes. Agricultural and Forest
Meteorology, 150(2): 166-174.
Aref, S. and Pike, D.R., 1998. Midwest farmers' perceptions of crop pest infestation. Agronomy
Journal, 90(6): 819-825.
Bai, J. et al., 2011. Evaluation of NASA satellite- and model-derived weather data for simulation
of maize yield potential in China. Agronomy Journal, 102(1): 9-16.
Ball, R.A., Purcell, L.C. and Carey, S.K., 2004. Evaluation of solar radiation prediction models
in North America. Agronomy Journal, 96(2): 391-397.
Barnett, J., 2010. Adapting to climate change: three key challenges for research and policy–an
editorial essay. Wiley Interdisciplinary Reviews: Climate Change, 1(3): 314-317.
Benson, G.O., 1990. Corn Replant Decisions - a Review. Journal of Production Agriculture,
3(2): 180-184.
Bohm, R. et al., 2001. Regional temperature variability in the European Alps: 1760-1998 from
homogenized instrumental time series. International Journal of Climatology, 21(14):
1779-1801.
Brassard, J.-P. and Singh, B., 2008. Impacts of climate change and CO2 increase on agricultural
production and adaptation options for Southern Québec, Canada. Mitigation and
Adaptation Strategies for Global Change, 13(3): 241-265.
Briggs, R.D. and Lemin Jr, R.C., 1992. Delineation of climatic regions in Maine. Canadian
Journal of Forest Research, 22(6): 801-811.
Bristow, K.L. and Campbell, G.S., 1984. On the relationship between incoming solar-radiation
and daily maximum and minimum temperature. Agricultural and Forest Meteorology,
31(2): 159-166.
Brown, R.A. and Rosenberg, N.J., 1997. Sensitivity of crop yield and water use to change in a
range of climatic factors and CO2 concentrations: A simulation study applying EPIC to
the central USA. Agricultural and Forest Meteorology, 83(3-4): 171-203.
Brown, R.A. and Rosenberg, N.J., 1999. Climate change impacts on the potential productivity of
corn and winter wheat in their primary United States growing regions. Climatic Change,
41(1): 73-107.
Bunkers, M.J., Miller, J.R. and DeGaetano, A.T., 1996. Definition of climate regions in the
Northern Plains using an objective cluster modification technique. Journal of Climate,
9(1): 130-146.

167

Cabas, J., Weersink, A. and Olale, E., 2010. Crop yield response to economic, site and climatic
variables. Climatic Change, 101(3): 599-616.
Carbone, G.J., 1995. Issues of Spatial and Temporal Variability in Climate Impact Studies.
Professional Geographer, 47(1): 30-40.
Carbone, G.J., Mearns, L.O., Mavromatis, T., Sadler, E.J. and Stooksbury, D., 2003. Evaluating
CROPGRO-Soybean performance for use in climate impact studies. Agronomy Journal,
95(3): 537-544.
Carter, T.R. et al., 2007. New assessment methods and the characterisation of future conditions.
In: M.L. Parry, O.F. Canziani, J.P. Palutikof, P.J.v.d. Linden and C.E. Hanson (Editors),
Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working
Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate
Change. Cambridge University Press, Cambridge, UK, pp. 133-171.
Castellvi, F. and Stockle, C.O., 2001. Comparing the performance of WGEN and ClimGen in the
generation of temperature and solar radiation. Transactions of the Asae, 44(6): 16831687.
Challinor, A.J., Ewert, F., Arnold, S., Simelton, E. and Fraser, E., 2009. Crops and climate
change: progress, trends, and challenges in simulating impacts and informing adaptation.
Journal of Experimental Botany, 60(10): 2775-2789.
Chen, C.C., McCarl, B.A. and Schimmelpfennig, D.E., 2004. Yield variability as influenced by
climate: A statistical investigation. Climatic Change, 66(1-2): 239-261.
Cline, W.R., 2007. Global Warming and Agriculture: Impact Estimates by Country. Center for
Global Development and Peterson Institute for International Economics, Washington,
DC.
Cooter, E.J. and Dhakhwa, G.B., 1996. A solar radiation model for use in biological applications
in the South and Southeastern USA. Agricultural and Forest Meteorology, 78(1-2): 3151.
Coulter, J., Moncada, K. and Sheaffer, C., 2010a. Soybean Production. In: K.M. Moncada and
C.C. Sheaffer (Editors), Risk Management Guide for Organic Producers. University of
Minnesota, St. Paul, pp. 10.1-10.18.
Coulter, J., Sheaffer, C., Moncada, K. and Huerd, S., 2010b. Corn Production. In: K.M. Moncada
and C.C. Sheaffer (Editors), Risk Management Guide for Organic Producers. University
of Minnesota, St. Paul, pp. 9.1-9.23.
Daly, C., Gibson, W.P., Taylor, G.H., Doggett, M.K. and Smith, J.I., 2007. Observer bias in
daily precipitation measurements at united states cooperative network stations. Bulletin
of the American Meteorological Society, 88(6): 899-912.

168

de Koeijer, T.J., Wossink, G.A.A., van Ittersum, M.K., Struik, P.C. and Renkema, J.A., 1999. A
conceptual model for analysing input-output coefficients in arable farming systems: from
diagnosis towards design. Agricultural Systems, 61(1): 33-44.
DeGaetano, A.T., 1996. Delineation of mesoscale climate zones in the northeastern United States
using a novel approach to cluster analysis. Journal of Climate, 9(8): 1765-1782.
DeGaetano, A.T., 2001. Spatial grouping of United States climate stations using a hybrid
clustering approach. International Journal of Climatology, 21(7): 791-807.
Dezfuli, A.K., 2011. Spatio-temporal variability of seasonal rainfall in western equatorial Africa.
Theoretical and Applied Climatology, 104(1-2): 57-69.
Diffenbaugh, N.S., Krupke, C.H., White, M.A. and Alexander, C.E., 2008. Global warming
presents new challenges for maize pest management. Environmental Research Letters,
3(4): 44007.
Dommenget, D. and Latif, M., 2002. A cautionary note on the interpretation of EOFs. Journal of
Climate, 15(2): 216-225.
Donatelli, M., Bellocchi, G. and Fontana, F., 2003. RadEst3.00: software to estimate daily
radiation data from commonly available meteorological variables. European Journal of
Agronomy, 18(3-4): 363-367.
Egli, D.B., 2008. Comparison of corn and soybean yields in the United States: historical trends
and future prospects. Agronomy Journal, 100(3, Supplement): S-79-S-88.
ERS-USDA, 2011. Agricultural Resource Management Survey (ARMS): Resource Regions,
available at http://www.ers.usda.gov/briefing/arms/resourceregions/resourceregions.htm
(verified September 8 2011). Economic Research Service, United Stated Department of
Agriculture.
ESRI, 2008. Euclidean Allocation, ARCGIS Desktop Help. . Environmental Systems Research
Institute, Inc., Redlands.
Everitt, B., 1980. Cluster Analysis. Halsted Press, New York.
Fischer, G., Shah, M., Tubiello, F.N. and van Velhuizen, H., 2005. Socio-economic and climate
change impacts on agriculture: an integrated assessment, 1990-2080. Philosophical
Transactions of the Royal Society B-Biological Sciences, 360(1463): 2067-2083.
Fovell, R.G. and Fovell, M.Y.C., 1993. Climate zones of the conterminous United-States defined
using cluster-analysis. Journal of Climate, 6(11): 2103-2135.

169

Fowler, H.J., Blenkinsop, S. and Tebaldi, C., 2007. Linking climate change modelling to impacts
studies: recent advances in downscaling techniques for hydrological modelling.
International Journal of Climatology, 27(12): 1547-1578.
Frei, C. et al., 2003. Daily precipitation statistics in regional climate models: Evaluation and
intercomparison for the European Alps. Journal of Geophysical Research-Atmospheres,
108(D3).
Garcia, A.G.Y., Guerra, L.C. and Hoogenboom, G., 2008. Impact of generated solar radiation on
simulated crop growth and yield. Ecological Modelling, 210(3): 312-326.
Garcia, A.G.Y. and Hoogenboom, G., 2005. Evaluation of an improved daily solar radiation
generator for the southeastern USA. Climate Research, 29(2): 91-102.
Geng, S., Devries, F. and Supit, I., 1986. A simple method for generating daily rainfall data.
Agricultural and Forest Meteorology, 36(4): 363-376.
Gilland, B., 2002. World population and food supply - Can food production keep pace with
population growth in the next half-century? Food Policy, 27(1): 47-63.
Goldblum, D., 2009. Sensitivity of corn and soybean yield in Illinois to air temperature and
precipitation: the potential impact of future climate change. Physical Geography, 30(1):
27-42.
Gong, X.F. and Richman, M.B., 1995. On theapplication of cluster-analysis to growing-season
precipitation data in North-America East of the Rockies. Journal of Climate, 8(4): 897931.
Greene, W.H., 2003. Econometric analysis. Prentice Hall.
Guan, Z.F., Lansink, A.O., van Ittersum, M. and Wossink, A., 2006. Integrating agronomic
principles into production function specification: A dichotomy of growth inputs and
facilitating inputs. American Journal of Agricultural Economics, 88(1): 203-214.
Guentchev, G., Barsugli, J.J. and Eischeid, J., 2010. Homogeneity of Gridded Precipitation
Datasets for the Colorado River Basin. Journal of Applied Meteorology and Climatology,
49(12): 2404-2415.
Guentchev, G.S. and Winkler, J.A., 2010. A two-tier atmospheric circulation classification
scheme for the European-North Atlantic region. Physics and Chemistry of the Earth, 35:
341-351.
Gujarati, D.N., 2003. Basic econometrics. McGraw Hill.
Guttman, N.B. and Quayle, R.G., 1996. A historical perspective of U.S. climate divisions.
Bulletin of American Meteorology Society, 77(2): 293-303.

170

Hao, X.M., Thelen, K. and Gao, J., 2010. Earlier-maturing hybrids improve corn grain
profitability in the northern Corn Belt. Crop Management(December): CM-2010-120301-RS.
Hartkamp, A.D. et al., 2004. Regional application of a cropping systems simulation model: crop
residue retention in maize production systems of Jalisco, Mexico. Agricultural Systems,
82(2): 117-138.
Haskett, J.D., Pachepsky, Y.A. and Acock, B., 1995. Estimation of Soybean Yields at County
and State Levels Using Glycim - a Case-Study for Iowa. Agronomy Journal, 87(5): 926931.
Hatfield, J., 2012. Agriculture in the Midwest. In: U.S. National Climate Assessment Midwest
Technical Input Report. J. Winkler, J. Andresen, J. Hatfield, D. Bidwell, and D. Brown,
coordinators. Available from the Great Lakes Integrated Sciences and Assessments
(GLISA) Center, http://glisa.msu.edu/docs/NCA/MTIT_Agriculture.pdf. .
Hoeft, R.G., Nafziger, E.D., Johnson, R.R. and Aldrich, S.R., 2000. Modern Corn and Soybean
Production. MCSP Publications, Champaign, Illinois, 353 pp.
Hoogenboom, G., 2000. Contribution of agrometeorology to the simulation of crop production
and its applications. Agricultural and Forest Meteorology, 103(1-2): 137-157.
Hoogenboom, G. et al., 2010. Decision Support System for Agrotechnology Transfer (DSSAT)
Version 4.5 [CD-ROM]. University of Hawaii, Honolulu, Hawaii.
Hoogenboom, G., White, J.W. and Messina, C.D., 2004. From genome to crop: integration
through simulation modeling. Field Crops Research, 90(1): 145-163.
Howden, S.M. et al., 2007. Adapting agriculture to climate change. PNAS, 104(50): 1969119696.
Hunt, L.A., Kuchar, L. and Swanton, C.J., 1998. Estimation of solar radiation for use in crop
modelling. Agricultural and Forest Meteorology, 91(3-4): 293-300.
IEM, 2013. Daily Observations for The National Weather Service [NWS] Cooperative Observer
Program [COOP] The Iowa Environmental Mesonet, Iowa State University Department
of Agronomy, available at http://mesonet.agron.iastate.edu/request/coop/fe.phtml
(accessed 02/02/2013).
IPCC-DDC, 2011. Carbon Dioxide: Projected emissions and concentrations. Intergovermental
Panel on Climate Change, available at http://www.ipcc-data.org/ddc_co2.html (verified
04/03/2013).

171

IPCC, 2007. Summary for Policymakers. In: Climate Change 2007: The Physical Science Basis.
Contribution of Working Group I to the Fourth Assessment Report of the
Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge,
United Kingdom and New York, NY, USA.
Irmak, A., Jones, J.W. and Jagtap, S.S., 2005. Evaluation of the CROPGRO-soybean model for
assessing climate impacts on regional soybean yields. Transactions of the Asae, 48(6):
2343-2353.
Izaurralde, R.C., Rosenberg, N.J., Brown, R.A. and Thomson, A.M., 2003. Integrated assessment
of Hadley Center (HadCM2) climate-change impacts on agricultural productivity and
irrigation water supply in the conterminous United States - Part II. Regional agricultural
production in 2030 and 2095. Agricultural and Forest Meteorology, 117(1-2): 97-122.
Jacobeit, J., 2010. Classifications in climate research. Physics and Chemistry of the Earth, 35(912): 411-421.
Jaggard, K.W., Qi, A.M. and Ober, E.S., 2010. Possible changes to arable crop yields by 2050.
Philosophical Transactions of the Royal Society B-Biological Sciences, 365(1554): 28352851.
Jagtap, S.S. and Jones, J.W., 2002. Adaptation and evaluation of the CROPGRO-soybean model
to predict regional yield and production. Agriculture Ecosystems & Environment, 93(13): 73-85.
Johnson, D.E., 1998. Applied Multivariate Methods for Data Analysis. Duxbury Press, Pacific
Grove, CA, 570 pp.
Jolliffe, I.T. and Philipp, A., 2010. Some recent developments in cluster analysis. Physics and
Chemistry of the Earth, 35(9-12): 309-315.
Jones, J.W. et al., 2003. The DSSAT cropping system model. European Journal of Agronomy,
18(3-4): 235-265.
Kaiser, H.F., 1959. Computer-program for varimax rotation in factor-analysis. Educational and
Psychological Measurement, 19(3): 413-420.
Kalkstein, L.S., Tan, G. and Skindlov, J.A., 1987. An evaluation of three clustering procedures
for use in synoptic climatological classification. Journal of Climate and Applied
Meteorology, 26(6): 717–730.
Kalnay, E. et al., 1996. The NCEP/NCAR 40-year reanalysis project. Bulletin of the American
Meteorological Society, 77(3): 437-471.
Kanamitsu, M. et al., 2002. NCEP DOE AMIP-II Reanalysis (R-2). Bulletin of the American
Meteorological Society, 83(11): 1631-1643.

172

Kang, Y., Khan, S. and Ma, X., 2009. Climate change impacts on crop yield, crop water
productivity and food security - A review. Progress in Natural Science, 19(12): 16651674.
Kaufmann, R.K. and Snell, S.E., 1997. A biophysical model of corn yield: Integrating climatic
and social determinants. American Journal of Agricultural Economics, 79(1): 178-190.
Kilsby, C.G. et al., 2007. A daily weather generator for use in climate change studies.
Environmental Modelling and Software, 22(12): 1705-1719.
Kim, J., Jung, H.S., Mechoso, C.R. and Kang, H.S., 2008. Validation of a multidecadal RCM
hindcast over East Asia. Global and Planetary Change, 61(3-4): 225-241.
Kjellstrom, E. et al., 2010. Daily and monthly temperature and precipitation statistics as
performance indicators for regional climate models. Climate Research, 44(2-3): 135-150.
Kobayashi, K. and Salam, M.U., 2000. Comparing simulated and measured values using mean
squared deviation and its components. Agronomy Journal, 92(2): 345-352.
Kucharik, C.J. and Serbin, S.P., 2008. Impacts of recent climate change on Wisconsin corn and
soybean yield trends. Environmental Research Letters, 3(3): 1-10.
Kumar, S. and Merwade, V., 2011. Evaluation of NARR and CLM3.5 outputs for surface water
and energy budgets in the Mississippi River Basin. Journal of Geophysical ResearchAtmospheres, 116: 1-21.
Kunkel, K.E., Liang, X.-Z., Zhu, J. and Lin, Y., 2006. Can CGCMs simulate the twentiethcentury "warming hole" in the central United States? Journal of Climate, 19(17): 41374153.
Laurer, J.G., 1997. Corn replant/late-plant decisions in Wisconsin, available at
http://learningstore.uwex.edu/assets/pdfs/A3353.PDF (accessed 03/08/2013).
Leduc, S., Diamond, H.J. and Palecki, M.A., 2009. The United States Climate Reference
Network (USCRN) Annual Report for Fiscal Year 2009: US Climate Reference Network,
NOAA/NESDIS National Climatic Data Center, Asheville, NC.
Liu, D.L. and Scott, B.J., 2001. Estimation of solar radiation in Australia from rainfall and
temperature observations. Agricultural and Forest Meteorology, 106(1): 41-59.
Lobell, D.B. and Asner, G.P., 2003. Climate and Management Contributions to Recent Trends in
U.S. Agricultural Yields. Science, 299(5609): 1032.
Lobell, D.B., Schlenker, W. and Costa-Roberts, J., 2011. Climate Trends and Global Crop
Production Since 1980. Science, 333(6042): 616-620.

173

Long, S.P., Ainsworth, E.A., Leakey, A.D.B., Nosberger, J. and Ort, D.R., 2006. Food for
thought: Lower-than-expected crop yield stimulation with rising CO2 concentrations.
Science, 312(5782): 1918-1921.
Luck, J. et al., 2011. Climate change and diseases of food crops. Plant Pathology, 60(1): 113121.
Luo, Q., Bellotti, W., Williams, M. and Wang, E., 2009. Adaptation to climate change of wheat
growing in South Australia: Analysis of management and breeding strategies. Agriculture
Ecosystems and Environment, 129(1-3): 261-267.
Markovic, M., Jones, C.G., Winger, K. and Paquin, D., 2009. The surface radiation budget over
North America: gridded data assessment and evaluation of regional climate models.
International Journal of Climatology, 29(15): 2226-2240.
Mavromatis, T. and Hansen, J.W., 2001. Interannual variability characteristics and simulated
crop response of four stochastic weather generators. Agricultural and Forest
Meteorology, 109(4): 283-296.
Mavromatis, T. and Jagtap, S.S., 2005. Estimating solar radiation for crop modeling using
temperature data from urban and rural stations. Climate Research, 29(3): 233-243.
Mearns, L.O. et al., 2009. A regional climate change assessment program for North America.
Eos Trans. AGU, 90(36): 311.
Mearns, L.O., Rosenzweig, C. and Goldberg, R., 1997. Mean and variance change in climate
scenarios: Methods, agricultural applications, and measures of uncertainty. Climatic
Change, 35(4): 367-396.
Meinke, H. et al., 2001. Increasing profits and reducing risks in crop production using
participatory systems simulation approaches. Agricultural Systems, 70(2-3): 493-513.
Menne, M.J., C. N. Williams, J. and Vose, R.S., 2012a. United States Historical Climatology
Network (USHCN) Version 2 Serial Monthly Dataset. Carbon Dioxide Information
Analysis Center, Oak Ridge National Laboratory, Oak Ridge, Tennessee. .
Menne, M.J., C. N. Williams, J. and Vose, R.S., 2012b. United States Historical Climatology
Network Daily Temperature, Precipitation, and Snow Data. Carbon Dioxide Information
Analysis Center, Oak Ridge National Laboratory, Oak Ridge, Tennessee.
Menne, M.J., Williams, C.N. and Vose, R.S., 2009. The US historical climatology network
monthly temperature data, version 2. Bulletin of the American Meteorological Society,
90(7): 993-1007.

174

Mera, R.J., Niyogi, D., Buol, G.S., Wilkerson, G.G. and Semazzi, F.H.M., 2006. Potential
individual versus simultaneous climate change effects on soybean (C-3) and maize (C-4)
crops: An agrotechnology model based study. Global and Planetary Change, 54(1-2):
163-182.
Mesinger, F. et al., 2006. North American Regional Reanalysis. Bulletin of the American
Meteorological Society, 87(3): 343-360.
Meza, F.J. and Silva, D., 2009. Dynamic adaptation of maize and wheat production to climate
change. Climatic Change, 94(1-2): 143-156.
Miller, P., Mitchell, M. and Lopez, L., 2005. Climate change: Length of growing-season in the
US Corn Belt, 1911-2000. Physical Geography, 26(2): 85-98.
Moen, T.N., Kaiser, H.M. and Riha, S.J., 1994. Regional Yield Estimation Using a Crop
Simulation-Model - Concepts, Methods, and Validation. Agricultural Systems, 46(1): 7992.
Moriondo, M., Giannakopoulos, C. and Bindi, M., 2011. Climate change impact assessment: the
role of climate extremes in crop yield simulation. Climatic Change, 104(3-4): 679-701.
Motha, R.P. and Baier, W., 2005. Impacts of present and future climate change and climate
variability on agriculture in the temperate regions: North America. Climatic Change,
70(1-2): 137-164.
NASA, 2011. Surface meteorology and Solar Energy (SSE) Release 6.0 Methodology. Available
at
http://power.larc.nasa.gov/common/MethodologySSE6/POWER_Methodology_Content.
html (verified September 06 2011).
NASS-USDA, 1997. Usual Planting and Harvesting Dates for U.S. Field Crops National
Agricultural Statistics Service, United States Department of Agriculture.
NASS-USDA, 2011. Quick Stats 2.0. The National Agriculture Statistics Service, United States
Department of Agriculture, available at http://quickstats.nass.usda.gov/
NCSS, 2011. The NCSS Soil Characterization Database. The National Cooperative Soil Survey,
available at http://ncsslabdatamart.sc.egov.usda.gov/ (accessed September 10, 2011).
Niyogi, D. and Mishra, V., 2013. Climate-Agriculture Vulnerability Assessment for the
Midwestern United States. In: S.C. Pryor (Editor), Climate Change in The Midwest:
impacts, risks, vulnerability, and adaptation Indiana Univerrsity Press, Bloomington, pp.
69-81.
Nonhebel, S., 1994. Inaccuracies in weather data and their effects on crop growth simulation
results. I. Potential production. Climate Research, 4: 47–60.

175

NRCS, 2012. Soil Data Viewer. Natural Resource Conservation Service, available at
http://soils.usda.gov/sdv/ (verified 04/04/2013).
NSRL, 2013. Soybean Production: Planting, Growing and Harvesting Soybeans. National
Soybean Research Laboratory, available at
http://www.nsrl.uiuc.edu/aboutsoy/production02.html (accessed March 20 2013).
O'Neal, M.R., Nearing, M.A., Vining, R.C., Southworth, J. and Pfeifer, R.A., 2005. Climate
change impacts on soil erosion in Midwest United States with changes in crop
management. Catena, 61(2-3): 165-184.
Olesen, J.E. et al., 2007. Uncertainties in projected impacts of climate change on European
agriculture and terrestrial ecosystems based on scenarios from regional climate models.
Climatic Change, 81: 123-143.
Parry, M., Rosenzweig, C., Iglesias, A., Fischer, G. and Livermore, M., 1999. Climate change
and world food security: a new assessment. Global Environmental Change-Human and
Policy Dimensions, 9: S51-S67.
Parry, M., Rosenzweig, C. and Livermore, M., 2005. Climate change, global food supply and
risk of hunger. Philosophical Transactions of the Royal Society B-Biological Sciences,
360(1463): 2125-2138.
Parry, M.L., Rosenzweig, C., Iglesias, A., Livermore, M. and Fischer, G., 2004. Effects of
climate change on global food production under SRES emissions and socio-economic
scenarios. Global Environmental Change-Human and Policy Dimensions, 14(1): 53-67.
Payero, J.O. and Irmak, S., 2006. Variable upper and lower crop water stress index baselines for
corn and soybean. Irrigation Science, 25(1): 21-32.
Pickering, N.B. et al., 1994. Weatherman - a utility for managing and generating daily weather
data. Agronomy Journal, 86(2): 332-337.
Pielke, R.A. et al., 2000. Spatial representativeness of temperature measurements from a single
site. Bulletin of the American Meteorological Society, 81(4): 826-830.
Pielke, R.A. et al., 2002. Problems in evaluating regional and local trends in temperature: An
example from eastern Colorado, USA. International Journal of Climatology, 22(4): 421434.
Pinker, R.T. and Laszlo, I., 1992. Modeling surface solar irradiance for satellite applications on a
global scale. Journal of Applied Meteorology, 31(2): 194-211.
Pollyea, A., 2013. Spatial and Temporal Trends of Soil Moisture in the Great Lakes region of the
USA, 1900-2008, Michigan State University, East Lansing.

176

Porter, J.R. and Semenov, M.A., 2005. Crop responses to climatic variation. Philosophical
Transactions of the Royal Society B-Biological Sciences, 360(1463): 2021-2035.
Pryor, S.C. and Barthelmie, R.J., 2013. The Midwestern United States: Socioeconomic Context
and Physical Climate. In: S.C. Pryor (Editor), Climate Change in The Midwest: impacts,
risks, vulnerability, and adaptation Indiana Univerrsity Press, Bloomington, pp. 12-47.
Quiring, S.M. and Legates, D.R., 2008. Application of CERES-Maize for within-season
prediction of rainfed corn yields in Delaware, USA. Agricultural and Forest Meteorology,
148(6-7): 964-975.
Rabbinge, R., 1993. The Ecological Background of Food-Production. Ciba Foundation
Symposia, 177: 2-29.
Rhee, J., Im, J., Carbone, G.J. and Jensen, J.R., 2008. Delineation of climate regions using in-situ
and remotely-sensed data for the Carolinas. Remote Sensing of Environment, 112(6):
3099-3111.
Richardson, C.W. and Wright, D.A., 1984. WGEN: A Model for Generating Daily Weather
Variables. U. S. Department of Agriculture, Agricultural Research Service, Springfield,
VA, 83 pp.
Richman, M.B., 1986. Rotation of principal components. Journal of Climatology, 6(3): 293-335.
Rivington, M., Bellocchi, G., Matthews, K.B. and Buchan, K., 2005. Evaluation of three model
estimations of solar radiation at 24 UK stations. Agricultural and Forest Meteorology,
132(3-4): 228-243.
Rivington, M., Matthews, K.B., Bellocchi, G. and Buchan, K., 2006. Evaluating uncertainty
introduced to process-based simulation model estimates by alternative sources of
meteorological data. Agricultural Systems, 88(2-3): 451-471.
Rivington, M. et al., 2008a. Downscaling regional climate model estimates of daily precipitation,
temperature and solar radiation data. Climate Research, 35(3): 181-202.
Rivington, M. et al., 2008b. Evaluating regional climate model estimates against site-specific
observed data in the UK. Climatic Change, 88(2): 157-185.
Rosenzweig, C. and Iglesias, A., 1998. The use of crop models for international climate change
impact assessment. In: G.Y. Tsuji, G. Hoogenboom and P.K. Thornton (Editors),
Understanding Options for Agricultural Production. Kluwer Academic Publisher,
Dordrecht, pp. 267-292.
Rosenzweig, C. and Parry, M.L., 1994. Potential Impact of Climate-Change on World FoodSupply. Nature, 367(6459): 133-138.

177

Rosenzweig, C. and Tubiello, F.N., 2007. Adaptation and mitigation strategies in agriculture: an
analysis of potential synergies. Mitigation and Adaptation Strategies for Global Change,
12(5): 855-873.
Safir, G.R., Gage, S.H., Colunga-Garcia, M., Grace, P. and Rowshan, S., 2008. Simulation of
corn yields in the Upper Great Lakes Region of the US using a modeling framework.
Computers and Electronics in Agriculture, 60(2): 301-305.
SAS Institute Inc., 2004. SAS OnlineDoc® 9.1.3. SAS Institute Inc., Cary, NC.
Schmidhuber, J. and Tubiello, F.N., 2007. Global food security under climate change.
Proceedings of the National Academy of Sciences of the United States of America,
104(50): 19703-19708.
Schroeder, T.A., Hember, R., Coops, N.C. and Liang, S., 2009. Validation of solar radiation
surfaces from MODIS and reanalysis data over topographically complex terrain. Journal
of Applied Meteorology and Climatology, 48(12): 2441-2458.
Shinker, J.J., 2010. Visualizing spatial heterogeneity of western U.S. climate variability. Earth
Interactions, 14: 1-16.
Soil Survey Staff, 2010. U.S. General Soil Map (STATSGO2) Natural Resources Conservation
Service, United States Department of Agriculture, available at
http://soildatamart.nrcs.usda.gov (accessed 06/30/2010).
Soltani, A. and Hoogenboom, G., 2003. A statistical comparison of the stochastic weather
generators WGEN and SIMMETEO. Climate Research, 24(3): 215-230.
Soussana, J.F., Graux, A.I. and Tubiello, F.N., 2010. Improving the use of modelling for
projections of climate change impacts on crops and pastures. Journal of Experimental
Botany, 61(8): 2217-2228.
Southworth, J. et al., 2002. Changes in soybean yields in the midwestern United States as a result
of future changes in climate, climate variability, and CO2 fertilization. Climatic Change,
53(4): 447-475.
Southworth, J. et al., 2000. Consequences of future climate change and changing climate
variability on maize yields in the Midwestern United States. Agriculture Ecosystems and
Environment, 82(1-3): 139-158.
Spitters, C.J.T., Toussaint, H. and Goudriaan, J., 1986. Separating the diffuse and direct
component of global radiation and its implications for modeling canopy photosynthesis.1.
Components of incoming radiation. Agricultural and Forest Meteorology, 38(1-3): 217229.

178

Stackhouse, P., 2006. Prediction of Worldwide Energy Resources. National Auronautics and
Space Administration. Available at http://power.larc.nasa.gov (verified September 9
2011).
Stöckle, C.O. and Kemanian, A.R., 2009. Crop Radiation Capture and Use Efficiency: A
Framework for Crop Growth Analysis. In: V. Sadras and D. Calderini (Editors), Crop
physiology : applications for genetic improvement and agronomy. Academic,
Amsterdam; London.
Stooksbury, D.E. and Michaels, P.J., 1991. Cluster-analysis of southeastern United-States
climate stations. Theoretical and Applied Climatology, 44(3-4): 143-150.
Strode, P.K., 2003. Implications of climate change for North American wood warblers
(Parulidae). Global Change Biology, 9(8): 1137-1144.
Tannura, M.A., Irwin, S.H. and Good, D.L., 2008. Weather, Technology, and Corn and Soybean
Yields in the U.S. Corn Belt, Department of Agricultural and Consumer Economics,
University of Illinois, Urbana-Champaign.
Tarasova, T.A. et al., 2006. Impact of new solar radiation parameterization in the Eta model on
the simulation of summer climate over South America. Journal of Applied Meteorology
and Climatology, 45(2): 318-333.
Thelen, K., 2007. When is the best time to start planting corn?, Michigan Farm News edition
04/15/2007, available at
http://www.michiganfarmbureau.com/farmnews/transform.php?xml=20070415/corn.xml
(accessed 03/19/2013).
Themeβ1, M.J., Gobiet, A. and Leuprecht, A., 2011. Empirical-statistical downscaling and error
correction of daily precipitation from regional climate models. International Journal of
Climatology, 31(10): 1530-1544.
Thomson, A.M., Brown, R.A., Rosenberg, N.J., Izaurralde, R.C. and Benson, V., 2005. Climate
change impacts for the conterminous USA: An integrated assessment - Part 3. Dryland
production of grain and forage crops. Climatic Change, 69(1): 43-65.
Thornton, P.E., Hasenauer, H. and White, M.A., 2000. Simultaneous estimation of daily solar
radiation and humidity from observed temperature and precipitation: an application over
complex terrain in Austria. Agricultural and Forest Meteorology, 104(4): 255-271.
Thornton, P.E. and Running, S.W., 1999. An improved algorithm for estimating incident daily
solar radiation from measurements of temperature, humidity, and precipitation.
Agricultural and Forest Meteorology, 93(4): 211-228.

179

Thorp, K.R., DeJonge, K.C., Kaleita, A.L., Batchelor, W.D. and Paz, J.O., 2008. Methodology
for the use of DSSAT models for precision agriculture decision support. Computers and
Electronics in Agriculture, 64(2): 276-285.
Tilman, D., Balzer, C., Hill, J. and Befort, B.L., 2011. Global food demand and the sustainable
intensification of agriculture. Proceedings of the National Academy of Sciences of the
United States of America, 108(50): 20260-20264.
Trnka, M. et al., 2007. Effect of estimated daily global solar radiation data on the results of crop
growth models. Sensors, 7(10): 2330-2362.
Tubiello, F.N. and Ewert, F., 2002. Simulating the effects of elevated CO2 on crops: approaches
and applications for climate change. European Journal of Agronomy, 18(1-2): 57-74.
Tubiello, F.N. and Rosenzweig, C., 2008. Developing climate change impact metrics for
agriculture. Integrated Assessment, 8(1): 165–184.
Tubiello, F.N., Rosenzweig, C., Goldberg, R.A., Jagtap, S. and Jones, J.W., 2002. Effects of
climate change on US crop production: simulation results using two different GCM
scenarios. Part I: Wheat, potato, maize, and citrus. Climate Research, 20(3): 259-270.
Uryasev, O., Gijsman, A.J., Jones, J.W. and Hoogenboom, G., 2003. SBuild: Create/Edit Soil
Input Files for Evaluation and Application on Crop Simulation Models for DSSATv4,
Agricultural and Biological Engineering Department, University of Florida, Gainesville,
Florida.
USDA-NASS, 2007. Quick Stats: Census of Agriculture. United States Department of
Agriculture - National Agricultural Statistics Service.
vanIttersum, M.K. and Rabbinge, R., 1997. Concepts in production ecology for analysis and
quantification of agricultural input-output combinations. Field Crops Research, 52(3):
197-208.
Vera-Diaz, M.d.C., Kaufmann, R.K., Nepstad, D.C. and Schlesinger, P., 2008. An
interdisciplinary model of soybean yield in the Amazon Basin: the climatic, edaphic, and
economic determinants. Ecological Economics, 65(2): 420-431.
Vucetic, V., 2011. Modelling of maize production in Croatia: present and future climate. Journal
of Agricultural Science, 149: 145-157.
Wang, M., Li, Y., Ye, W., Bornman, J.F. and Yan, X., 2011. Effects of climate change on maize
production, and potential adaptation measures: a case study in Jilin Province, China.
Climate Research, 46(3): 223-242.

180

Weiss, A., Hays, C.J., Hu, Q. and Easterling, W.E., 2001. Incorporating bias error in calculating
solar irradiance: Implications for crop yield simulations. Agronomy Journal, 93(6): 13211326.
West, G.L., Steenburgh, W.J. and Cheng, W.Y.Y., 2007. Spurious grid-scale precipitation in the
North American regional reanalysis. Monthly Weather Review, 135(6): 2168-2184.
White, J.W. and Hoogenboom, G., 2011. Crop Response to Climate: Ecophysiological Models.
In: D. Lobell and M. Burke (Editors), Climate Change and Food Security: Adapting
Agriculture to a Warmer World. Advances in Global Change Research, pp. 59-83.
White, J.W., Hoogenboom, G., Kimball, B.A. and Wall, G.W., 2011a. Methodologies for
simulating impacts of climate change on crop production. Field Crops Research, 124(3):
357-368.
White, J.W., Hoogenboom, G., Stackhouse, P.W., Jr. and Hoell, J.M., 2008. Evaluation of
NASA satellite- and assimilation model-derived long-term daily temperature data over
the continental US. Agricultural and Forest Meteorology, 148(10): 1574-1584.
White, J.W., Hoogenboom, G., Wilkens, P.W., Stackhouse, P.W., Jr. and Hoel, J.M., 2011b.
Evaluation of satellite-based, modeled-derived daily solar radiation data for the
continental United States. Agronomy Journal, 103(4): 1242-1251.
Whitfield, P.H., Bodtker, K. and Cannon, A.J., 2002. Recent variations in seasonality of
temperature and precipitation in Canada, 1976-95. International Journal of Climatology,
22(13): 1617-1644.
Wilby, R.L., Dawson, C.W. and Barrow, E.M., 2002. SDSM - a decision support tool for the
assessment of regional climate change impacts. Environmental Modelling and Software,
17(2): 145-157.
Wilby, R.L. and Harris, I., 2006. A framework for assessing uncertainties in climate change
impacts: Low-flow scenarios for the River Thames, UK. Water Resources Research,
42(2).
Wilby, R.L. et al., 2009. A review of climate risk information for adaptation and development
planning. International Journal of Climatology, 29(9): 1193-1215.
Wilby, R.L. and Wigley, T.M.L., 1997. Downscaling general circulation model output: a review
of methods and limitations. Progress in Physical Geography, 21(4): 530-548.
Wilcox, S. et al., 2007. National Solar Radiation Database 1991–2005 Update: User’s Manual,
National Renewable Energy Laboratory, Golden, Colorado.
Wilks, D.S., 1992. Adapting stochastic weather generation algorithms for climate change studies.
Climatic Change, 22(1): 67-84.

181

Wilks, D.S., 1995. Statistical Methods in the Atmospheric Sciences: An Introduction.
International Geophysics Series, 59. Academic Press, Sandiego, 465 pp.
Willmott, C.J., 1981. On the validation of models. Physical Geography, 2: 184-194.
Winkler, J.A., 1992. Regional patterns of the diurnal properties of heavy hourly precipitation.
The Professional Geographer, 44(2): 127-146.
Winkler, J.A., Andresen, J.A., Guentchev, G. and Kriegel, R.D., 2002. Possible impacts of
projected temperature change on commercial fruit production in the Great Lakes region.
Journal of Great Lakes Research, 28(4): 608-625.
Winkler, J.A. et al., 2011. Climate scenario development and applications for local/regional
climate change impact assessments: An overview for the non-climate scientist. Part I:
Scenario development using downscaling methods. Geography Compass, 5(6): 275-300.
Woli, P. and Paz, J.O., 2012. Evaluation of various methods for estimating global solar radiation
in the southeastern United States. Journal of Applied Meteorology and Climatology,
51(5): 972-985.
Yang, K., Koike, T. and Ye, B.S., 2006. Improving estimation of hourly, daily, and monthly
solar radiation by importing global data sets. Agricultural and Forest Meteorology,
137(1-2): 43-55.
Yeh, H.Y., Wensel, L.C. and Turnblom, E.C., 2000. An objective approach for classifying
precipitation patterns to study climatic effects on tree growth. Forest Ecology and
Management, 139(1-3): 41-50.
Zeileis, A., 2006. Object-oriented computation of sandwich estimators. Journal of Statistical
Software, 16(9).

182