Using MODIS time series and random forests classification for mapping land use in South-East Asia

Broad scale and continuous land-use/cover mapping is important for research in the context of global and climate change. We have therefore developed a method based on MODIS time-series and Random Forest classification to map forested, non-forested and plantation areas in South-East Asia. Our approach is optimized for regions with frequent cloud cover and scarce reference data. Results show that the method performs acceptable in terms of accuracy under given conditions. Furthermore, various validation strategies were compared with the result that pixel-by-pixel validation procedures with manual assignment must be handled carefully when applied to large scale land-use/cover maps.