Measuring Cropland Change: A Cautionary Tale

Abstract

Remotely sensed imagery has been used for decades to quantify the area, rates, and types of land use and land cover change (LULCC) from local to global scales. Common platforms used to capture data include satellite- and aerial-based sensors and cameras. Inseparable from those data are errors—misclassified pixels in sensor-based data or incorrect observations in aerial photographs—created by spectral confusion on the part of the sensor or by misclassifying the raw data during interpretation. Such errors, if not sufficiently explained or taken into consideration when reporting LULCC results, could lead to dubious conclusions. For this article, four commonly used and readily available data sets were analyzed to determine the amount of cropland change in South Dakota from circa 2006 to 2012. Two of the four data sets (the National Land Cover Dataset and the Cropland Data Layer) were created from satellite-based sensors, one was created from photographic data (from the National Agricultural Imaging Program), and the fourth was a survey-based assessment of agricultural land use (Census of Agriculture). Depending on which classes were used or how the data were manipulated prior to calculating areal totals, South Dakota had between 13 million and 19 million acres of cropland in 2012—a difference the size of the state of New Jersey—with rates of change from 2006 to 2012 ranging from 0.28 percent to 22.65 percent.