The Cropland Data Layers (CDL) are high-resolution geo-referenced data products made available by the U.S. Department of Agriculture. However, the CDL lacks in its ability to be employed as a tool to identify the impact of the gradually evolving drivers of land use change, e.g., climate change, due to its limited historical depth. We implement a robust, phenology-based satellite image classification algorithm to identify historical cropland allocation in eastern South Dakota and North Dakota predating the initial CDL by 13 and 22 years, respectively. Five major land covers, i.e., corn, soybeans, wheat, alfalfa and grass (including native grass, hay and pasture) are identified using archived Landsat-5 surface reflectance data, while achieving CDL-like accuracy. The long-term rate of grassland loss during 1985–2011 is found to be significantly lower (26,781 hectares or 1.5% annually) relative to the near-term rate of grassland loss during 2006-’11 (84,545 hectares or 5.2% annually). We find similar discrepancy in regional corn expansion rates. Our value-added raster data provide opportunities for improved identification of land use drivers, whereas relying solely on the CDL’s restricted historical extent may lead to biased land use change estimates and misguide policy.
19 December 2017 in American Journal of Agricultural Economics
Perception biases documented in the literature often pertain to subject matters that are difficult to observe or measure such as one’s ability. We study perception biases with respect to a concrete indicator that can be objectively measured: land use changes in a local area. We examine four hypotheses about land use change perceptions and test them with farm survey data complemented by satellite data. We discover systematic biases in farmers’ perceptions about local land use changes that are consistent with motivated beliefs, and also evidence that links perceptions with intended future land conversions. Alternative explanations and policy implications are discussed.