Mapping Land Degradation in a Tropical Watershed through Machine Learning: Implications for Sustainable Land Management in West Papua, Indonesia

Land degradation poses a serious threat to ecosystem productivity and sustainable development, particularly in tropical regions undergoing rapid land-use change. This study aims to address this challenge by mapping the severity of land degradation in the Prafi Watershed, West Papua, Indonesia (a hotspot of deforestation, agricultural expansion, and urbanization). Using the Random Forest (RF) machine learning algorithm on the Google Earth Engine (GEE) platform, the analysis integrated climatic, topographic, biophysical, and anthropogenic variables to classify degradation into five levels: very low, low, moderate, high, and very high. The model was trained and validated with a 70:30 data split, achieving high accuracy (overall accuracy = 97%, Kappa = 0.95). Results indicate that 70% of the watershed experiences low-level degradation, while high to very high degradation (13%) is concentrated in sloping areas recently converted to agriculture. This study demonstrates that RF combined with cloud-based geospatial big data enables precise, scalable land degradation assessment in data-scarce tropical environments. It further reveals how natural and human-driven factors jointly shape degradation patterns. The resulting spatial framework provides actionable evidence for watershed conservation and sustainable land-use planning in West Papua and comparable regions. Given the urgency of land restoration in the tropics, this approach highlights the critical role of machine learning and open geodata in strengthening evidence-based environmental governance and supporting national sustainability commitments.