High-resolution (10 m) rubber map for Southeast Asia, circa 2020

This dataset provides a 10 m resolution map of rubber plantations across Thailand, Myanmar, Laos, Vietnam, Cambodia, Malaysia, Indonesia, and parts of southern China (Xishuangbanna and Hainan) for around 2020. The GeoTIFF tiles contain one categorical band: 1 = rubber plantation, 0 = other tree cover, and -99 = NoData. The map was developed to support supply-chain due diligence and risk screening, including in the context of the EU Deforestation Regulation (EUDR). Known limitations include under-mapping in areas with persistent cloud cover or weak phenological signals, particularly in parts of Malaysia and Indonesia, and omission of young, diseased, or mixed-species rubber systems. The absence of mapped rubber should therefore not be interpreted as evidence that rubber is absent. Residual commission errors include confusion with other plantations, including broadleaf plantations in northern Vietnam and palm plantations in insular Southeast Asia, as well as open forest types, clearings, and other sparse tree cover. Commission errors also resulted from focal majority filtering, which caused small patches of non-rubber vegetation embedded within rubber stands to be classified as rubber. The map was developed using Sentinel-2 phenology metrics together with radar, canopy-height, and spatial-texture information. ESA WorldCover was used to restrict the classification to areas mapped as tree cover in 2020. Full methodological details, evaluation statistics, and known limitations are provided in the associated preprint. The map is also available as a Google Earth Engine asset (projects/rubber-499107/assets/rubber_SEA_2020) and can be accessed at https://code.earthengine.google.com/?asset=projects/rubber-499107/assets/rubber_SEA_2020. The datasets provided here include 1) Native 10 m rubber classification (4° × 4° tiles, named according to the latitude and longitude of the tiles' lower-left corner) 2) A 1 km overview layer giving the number of underlying 10 m pixels classified as rubber in each grid cell (0-10,000) Antje Ahrends and Sam Harrison contributed equally to this dataset.