Wetland mapping in East Asia by two-stage object-based Random Forest and hierarchical decision tree algorithms on Sentinel-1/2 images

Accurate information on wetland extent in East Asia is essential to assess progress towards Sustainable Development Goals (SDGs) and the use of wetland resources, where wetlands benefit a quarter of the world's population and millions of wild birds in over three global migratory corridors. In this study, using 122,128 Sentinel-1 and 89,752 Sentinel-2 images acquired in 2021 available on the Google Earth Engine platform, we developed a novel two-stage classification for continental-scale wetland mapping and generated the first and up-to-date 10-m resolution wetland map of East Asia. Such a two-stage classification method, which integrates automatic sample generation and spatiotemporal features, combined an initial object-based random forest classifier with a subsequent hierarchical decision tree for secondary waterbody types. The resulting comprehensive map with 3 broad categories and 12 sub-categories in East Asia, named EA_Wetlands, achieved over 88% overall accuracy. According to EA_Wetlands, the total wetland area in this region is 481,802.49 km2, mainly distributed in Northeast China and the Qinghai-Tibet Plateau (41.02%). Of all wetlands in East Asia, about 68.26% are inland wetlands. The highest proportion (29.67%) is identified for inland marsh among 12 sub-categories. Among five countries, China has >88% of the total wetland resources in East Asia, followed by Mongolia (3.57%). South Korea has the largest ratio between wetland and national land areas (10.43%). EA_Wetlands, as the first 10-m resolution wetland data product, will have great applications and benefit wetland conservation and policy management. It will be a critical support for evaluating the implementation of wetland-related international conventions at country and continental scale in East Asia.