Taken as a given: Evaluating the accuracy of remotely sensed crop data in the USA. (December 2015)

• Record Type: Journal Article
• Authors: Larsen, Ashley E.; Hendrickson, Brandon T.; Dedeic, Nicholas; MacDonald, Andrew J.
• Is Part Of: Agricultural systems. Volume 141(2015)
• Page Start: 121
• Page End: 125
• Publication Date: 2015-12
• Subjects: Agriculture -- Agricultural census -- Cropland -- Geospatial data; Agricultural systems -- Periodicals; Agriculture -- Environmental aspects -- Periodicals; 338.16
• DOI: 10.1016/j.agsy.2015.10.008
• Languages: English
• ISSNs: 0308-521X

Abstract

Abstract: Quantifying the extent of agricultural land is important for addressing a large range of ecological, environmental, and economic questions. In many cases, answers focused on land use change require fine scale spatial data on the arrangement of crop types. Here we take advantage of the simultaneous availability of fine resolution, geospatial cropland data—the Cropland Data Layer, and comprehensive tabulated data—the USDA Census of Agriculture, to better understand the accuracy of geospatial data and thus, how geospatial data may be used in scientific research. We compared area estimates for cropland and major US crops (corn, soybeans, wheat and small grains) at the county level for the contiguous US in 2012 and for a subset of states in 2007. We find that accuracy of the Cropland Data Layer is high in regions dominated by a few crop types. However, elsewhere in the US accuracy is highly variable with common large areal overestimates and underestimates (+/− 50% or more). Before employing the CDL and other geospatial data for applications such as measuring fine scale changes in land use, users should be wary of the potentially high misclassification error. Highlights: Geospatial agricultural data hold great potential for scientific applications. However, large inaccuracies were observed even for major crops and crop groups. The Cropland Data Layer is accurate in regions with simple crop composition. The CDL can be highly inaccurate in regions with sparse or complex agriculture. Great care should be taken when considering the CDL for pixel level analysis.