<![CDATA[Land-cover change is a primary driver of environmental degradation and local climate alteration, yet up-to-date, spatially explicit information remains scarce in rapidly developing regions such as North Sumatra, Indonesia. This study aims to develop an accurate, reproducible workflow for environmental and climate monitoring by integrating cloud-based spatial data processing with machine-learning classification. Multitemporal Sentinel-2 surface-reflectance imagery (2019 and 2024) and Landsat-derived land-surface temperature were processed on the Google Earth Engine (GEE) platform. Spectral bands were combined with vegetation, water, and built-up indices (NDVI, NDWI, NDBI, EVI, SAVI, BSI) and topographic data, after which a Random Forest (RF) classifier distinguished six land-cover classes from stratified training samples. The RF model achieved an overall accuracy of 92.4% and a kappa coefficient of 0.906, outperforming Classification and Regression Tree (86.1%) and Support Vector Machine (88.7%) baselines. NDVI, the near-infrared band, and NDWI were the most influential predictors. Between 2019 and 2024 the built-up area expanded by 18.7% while forest and cropland contracted, and these changes coincided with a measurable rise in land-surface temperature, confirming a strong inverse relationship between vegetation cover and surface heating. The results demonstrate that the GEE–RF approach provides a low-cost, scalable, and replicable basis for routine environmental and climate monitoring to support evidence-based regional planning.]]>
Copyrights © 2026
