Detecting forest-to-oil-palm conversion using sentinel time series for EUDR compliance: A case study in Uganda

Monitoring forest-to-oil-palm conversion for the EU Deforestation Regulation (EUDR) requires evidence not only of forest loss, but also of its timing and the subsequent establishment of oil palm. Global forest-change products provide valuable annual information, but their ability to detect relevant changes, avoid false alarms, and identify subsequent oil-palm establishment requires evaluation in local forest–plantation mosaics. To address this gap, we developed and evaluated a transparent spatio-temporal workflow using freely available Earth observation data, with Buvuma Island, Uganda, as a case study. The workflow estimates a proxy change year from Sentinel-2 Normalized Difference Water Index (NDWI) dynamics and uses a Random Forest classifier to distinguish stable forest from forest-to-oil-palm conversion based on pre- and post-change Sentinel-1/2 time-series features and topographic data. Using 100 labelled locations for model development and internal validation, the workflow achieved 89.8% agreement at 734 held-out reference locations. Post-change optical features ranked prominently, Sentinel-1, provided complementary structural information, and most errors occurred along forest–plantation boundaries. At the same locations, an application-specific comparison with Global Forest Change (GFC) and Tropical Moist Forests (TMF) products identified frequent disagreements in baseline forest status, change occurrence or conversion timing. These comparisons are conditional on the purposively assembled reference sample and do not constitute representative map-wide validation. The workflow combines local transition detection with commodity-specific interpretation to support targeted EUDR screening, with adaptation to other landscapes requiring local reference data and validation.