Paddy Rice at 10m resolution in Southeast and South Asia in 2020

Rice is the main crop and staple food for more than half the world's population, and changes in rice area have major implications for global food security. South and Southeast Asia are the most important rice producing regions in the world, encompassing eight of the world's top producing countries. Accurate identification of paddy area is critical for accurate estimation of paddy production and assessment of food security in this region. We divided South and Southeast Asia into 37 subregions to identify paddy rice, considering the strong spatial heterogeneity in climatic, topographic and water endowments.Our study proposed a progressive mapping algorithm that followed a hybrid approach deployed on the Google Earth Engine (GEE) to generate 10m paddy rice distribution data in South and Southeast Asia in 2020. First, the hybrid approach used the MODIS data at 500m to identify the potential spatial distribution of paddy rice; second, this method used the high-resolution Sentinel-1/2 data to identify the 10m distribution of paddy rice based on the potential spatial distribution.Based on the MOD09 dataset from 2000 to 2020, we extracted the NDVI, EVI and LSWI time series at 500m resolution. And then the value of NDVI, EVI, and LSWI are judged for the period 2000-2020, and if a pixel satisfies LSWI > EVI or LSWI > NDVI, the pixel is judged as potential paddy rice.Based on the spatial distribution of potential paddy rice, we analysed the polarised scattering characteristics of forests, buildings, water, aquaculture, and rice, and found that their VH scattering coefficients have significant differences. Based on this significant difference, the following measures were taken to further extract the fine distribution of paddy rice. First, we generated monthly VH and VV time series from October 2019 to October 2020 with median composites based on Sentinel-1 data after slope correction. Second, based on the spatial distribution of potential paddy fields, VH images from three consecutive months of rice inundation, transplanting and tillering periods were used for colour synthesis, and sample sets of paddy fields and non-paddy fields (forests, water bodies, towns) were obtained by visual interpretation for each sub-region. Third, we generated a set of Sentinel-2 features using percentile synthesis (percentile: 5, 15, 25, 35, 50, 65, 75, 85, 95) for LSWI, NDVI, EVI and GCVI indicators. Finally, a supervised classification method was applied region by region, using a random forest classifier (tree number =300, leaf number=5) with Sentinel-1 and Sentinel-2 features as input to identify the detailed spatial distribution of paddy rice. Based on this approach, we generated the spatial distribution of paddy rice with 10 m resolution in South and Southeast Asia in 2020 with F1 scores between 0.9 and 0.98.