Estimating Global Cropland Extent with Multi-year MODIS Data

• DOI: 10.3390/rs2071844
• Publication Date: 2010-07-22T06:23:06Z
• Authors: Kyle Pittman; Matthew C. Hansen; Inbal Becker-Reshef; Peter V. Potapov; Christopher O. Ju...
• Keywords: Moderate-resolution imaging spectroradiometer; Land Cover; Agricultural land; Data set

This study examines the suitability of 250 m MODIS (MODerate Resolution Imaging Spectroradiometer) data for mapping global cropland extent. A set 39 multi-year metrics incorporating four land bands, NDVI (Normalized Difference Vegetation Index) and thermal was employed to depict phenology over period. Sub-pixel training datasets were used generate a classification tree models using bagging methodology, resulting in per-pixel probability layer. product subsequently thresholded create discrete cropland/non-cropland indicator map from USDA-FAS (Foreign Agricultural Service) Production, Supply Distribution (PSD) database describing per-country acreage production field crops. Five cover products, which attempted croplands context multiclass classifications, perform regional evaluations extent map. The layer further examined with reference principle food crops: corn, soybeans, wheat rice. Overall results indicate that best depicts regions intensive broadleaf crop (corn soybean), both correspondence existing maps associated high matching thresholds. Probability thresholds wheat-growing lower, while areas rice had lowest confidence. Regions absent agricultural intensification, such as Africa, are poorly characterized regardless type. reflect value generic agriculture regions, but little sensitivity low intensification. Variability accuracies between dominated by different types also points desirability crop-specific approach rather than attempting aggregate.