The USDA Foreign Agricultural Service Production, Supply and Distribution database was used to threshold the global cropland probability product into a discrete cropland/non-cropland map. The thresholds were calculated and applied on a per-country basis with the exception of the European Union where only aggregate area statistics were available. To calculate the threshold, the PSD acreage values for production field crops (barley, corn, cotton, oats, rice, rye, sorghum, soybeans and wheat) were averaged over the 2000-2008 timeframe. Then the probability threshold that most closely matched the PSD area was calculated, with all pixels with a probability matching or exceeding the threshold classified as "Cropland" and all pixels below that threshold classified as "Non-Cropland".
In this image areas classified as "Cropland" are shown in red with the cropland probability shown in a beige-to-green gradient for areas classified as "Not Cropland."
Once the area-matching threshold was calculated and applied the resulting discrete cropland map was compared with five satellite-derived global landcover products (GLC2000, IGBP Discover, MOD12Q1, Ramankutty Year 2000 Agricultural Lands, and UMD 1km) to determine spatial accuracy. The five datasets were each reclassified to a yes/no cropland product and a composite map was made with a value range from 0 (where none of the 5 called a pixel cropland) to 5 (where all 5 products called a pixel cropland). The composite map is shown below where pixels in red indicated all 5 products agree, yellow means 4 agree, green 3 agree, cyan 2 agree, blue 1 agreed, black 0.
High-resolution examples of how the MODIS Cropland Extent compares to this composite validation set can be seen for the North China Plain, the Indo-Gangetic Plain, the United States Corn Belt, and Argentina.
Discrete cropland/non-cropland data can be downloaded either by MODIS tile or global mosaic here.
These data may be used for valid scientific or educational purposes as long as proper citations are used. We ask that you credit the Global Cropland Extent data as follows:
Pittman, K., Hansen, M.C., Becker-Reshef, I., Potapov, P.V., Justice, C.O. (2010) Estimating Global Cropland Extent with Multi-year MODIS Data. Remote Sensing, in press.