Does the US Cropland Data Layer Provide an Accurate Benchmark for Land-Use Change Estimates?

作者：Reitsma Kurtis D; Clay David E^*; Clay Sharon A; Dunn Barry H; Reese Cheryl

来源：Agronomy Journal, 2016, 108(1): 266-272.

DOI：10.2134/agronj2015.0288

Even though the cropland data layer (CDL) has been used in policy discussions it has not been independently validated using publically available information. The projects objective was to conduct an independent validation of the CDL. South Dakota was selected as a model system because it is located in a climate transition zone, with row crop production being the dominant practice in eastern South Dakota and the grazing of grassland being the dominant practice in western South Dakota. High resolution imagery was used to determine land-uses (cropland, grassland, non-agricultural, habitat, and water) at 14,400 points in 2006 and 2012. Based on comparisons between the CDL and ground collected data, a confusion table was constructed and the CDL user (% false positive = 100-user accuracy) and producer (% false negatives = 100 - producer accuracy) accuracies determined. The % false positives and % false negatives are oft en referred to as Type I error and Type II error. In 2006, the CDL cropland producer accuracy (% of ground collected sites that were correctly identified) ranged from 89.2% in the east central to 42.6% in the Northwest, whereas the CDL grassland producer accuracy ranged from 95.2% in the Northwest to 38.9% in the Southeast. Similar results were reported for 2012. Grassland CDL producer and user accuracies were highest when grasslands were the dominant practice and cropland producer and user accuracies were highest when croplands were the dominant land-use. These results suggest that inherent CDL errors introduce uncertainty into land-use change calculations.

• 出版日期2016-2