Annual Crop Type Classification of the
US Great Plains for 2000 to 2011
Daniel M. Howard and Bruce K. Wylie
Abstract
The purpose of this study was to increase the spatial and
temporal availability of crop classification data. In this study,
nearly 16.2 million crop observation points were used in the
training of the US Great Plains classification tree crop type
model (
CTM
). Each observation point was further defined
by weekly Normalized Difference Vegetation Index, annual
climate, and a number of other biogeophysical environmental
characteristics. This study accounted for the most prevalent
crop types in the region, including, corn, soybeans, winter
wheat, spring wheat, cotton, sorghum, and alfalfa. Annual
CTM
crop maps of the US Great Plains were created for 2000 to
2011 at a spatial resolution of 250 meters. The
CTM
achieved
an 87 percent classification success rate on 1.8 million obser-
vation points that were withheld from model training. Product
validation was performed on greater than 15,000 county
records with a coefficient of determination of R
2
= 0.76.
Introduction
Wide-ranging, long-term, and spatially accurate cropland
classification data are a valuable source of information for
government agencies, private sector organizations, scientists,
educators, and others who use land-cover information (Bo-
ryan
et al
., 2011). Crop type information can be incorporated
into a range of environmental models to learn more about the
overall influence that agriculture has on the environment, and
conversely, the environment on agriculture (Gilmanov
et al
.,
2013). A crop type dataset could also be used by modelers
seeking to measure the effects that certain climatic, sociologi-
cal, economic, or other factors have on cropland. There is also
a clear statistical value that can be gleaned from a crop type
dataset. For example, analysts can calculate various statistics
about crop yields and acreage totals, and explore topics such
as crop rotation tendencies, land use change, and overall
trends. The vast scientific potential of accurate crop type in-
formation was the primary driver behind developing a robust
model to utilize the most consistent and best available input
data to map crops across time and space.
In this study, classification tree modeling was applied using
the best available remote sensing and biogeophysical envi-
ronmental data to map corn, soybeans, winter wheat, spring
wheat, cotton, sorghum, alfalfa, and a class called “other.” The
“other” class included a wide range of miscellaneous crop
types and provided the model with a loosely-defined class in
which to place indefinite predictions, rather than force-fitting
into a finite class. Classification tree modeling is a data mining
technique that organizes numerous observations from multiple
variables into hierarchal and more homogenous subsets for
making classification predictions (Loh, 2011). These models
are constructed based on a sample of observation points known
as training data. Training data include information about the
variable being predicted, along with applicable ancillary data
that may have a measurable influence on certain characteris-
tics of the variable being predicted
/
see5-win.html
). The method in this study was developed to take
advantage of existing and proven crop data sources through
extensive model training. The final model was developed using
just over 16.2 million training observations. There are a num-
ber of sources that provide spatial crop type data (for example,
Wardlow
et al
., 2007; Lunetta
et al
., 2010; Howard
et al
., 2012;
NASS CDL, 2013), but are often limited to a specific region or
time frame, have inadequate spatial detail, or accuracy concerns.
Additional crop type resources are required to fill these voids in
time and space. The purpose of this study was to address this re-
quirement by increasing the spatial and temporal availability of
crop classification data using the most consistent and best avail-
able source data that have the potential of being applied on lo-
cal, regional, national, or global levels. The spatial focus of this
effort was the US Great Plains and the temporal focus was 2000
to 2011. The data products derived in this study can be applied
as a standalone resource, but were not necessarily intended to
be a complete replacement for the other available crop classifica-
tion data, but rather, as a supplemental data source.
Study Area
The focus of this study was on croplands of the US Great
Plains (Figure 1). The Great Plains is an agriculturally inten-
sive region where sustainable land management is necessary
to promote the conservation of this valuable region. Overall,
the US Great Plains represents a highly valuable and diverse
region that has a significant influence on the US domestic ag-
riculture, energy, and economy. Roughly 37 percent of the US
Great Plains consists of cultivated crops and hay or pasture.
Other significant land-cover types include grasslands and
shrubs, accounting for 36 percent and 12 percent, respec-
tively. The focus of this study was centered on the 37 percent
that is cultivated crops, hay, or pasture. As spatially defined
by Olson
et al
. (2001), the Great Plains is a vast expanse
located in central North America that extends into 14 states
in the US, as well as into parts of Canada and Mexico. For the
purposes of this study, only established croplands within the
US portion of the Great Plains (US Great Plains) were consid-
ered. Established croplands were defined as areas specifically
classified as cultivated crops or hay in both 2001 (Homer
et
al
., 2007 and 2006; Fry
et al
., 2011) versions of the National
Land Cover Database (
NLCD
).
Daniel M. Howard is with Stinger Ghaffarian Technolo-
gies (SGT), Contractor to USGS EROS Center, Sioux Falls,
SD 57198, USA; Work performed under USGS Contract
G10PC00044 (
).
Bruce K. Wylie is with the USGS EROS Center, Sioux Falls,
SD 57198.
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 5, June 2014, pp. 537–549.
0099-1112/14/8006–537
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.6.537
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
June 2014
537
