Continental scale land cover classification using MODIS surface reflectance products

The objective of this study is to develop land cover classification algorithm for Global and/or continental scale by using multi-temporal MODIS land reflectance products. Time-domain co-occurrence matrix is introduced as a classification feature that represents time-series signature of land covers. And the non-parametric minimum distance classification using Euclidian distance or cosine distance is conducted. As results, Surface Reflectance 8-Day L3 product and Nadir BRDF-Adjusted Reflectance product showed similar classification accuracy of 91%-93% for IGBP-17 land cover categories. Furthermore, it is confirmed that the accuracy is 14%-17% higher than that of MODIS land cover product for the target classification area.