Multi-sensor fusion of Landsat and Sentinel-1 SAR for spatiotemporal land use/land cover change mapping and land surface temperature projection in Dhaka Industrial Area

Rapid urbanization in the Global South is intensifying urban heat islands and depleting water bodies, yet monsoon cloud cover limits optical monitoring. This study evaluates multi-sensor fusion of Landsat 8/9, Sentinel-1 SAR and the Global Human Settlement Layer for LULC change mapping and LST projection in Dhaka Industrial Area, Bangladesh (2015 to 2035), using Random Forest and Gradient Boosted Trees within Google Earth Engine. SAR fusion increased LULC overall accuracy to 91.7% and eliminated water commission errors (User's Accuracy: 83.3% to 100%), while contributing negligibly to LST regression (ΔRMSE = +0.003 °C). Built-up area is projected to expand by 76.1% and water bodies to decline by 68.2% by 2035, with mean LST rising 1.86°C. The GBT model achieved a spatially validated filtered RMSE of 1.711 °C, reducing systematic bias by 54% over the optical-only baseline, with direct implications for water body conservation and heat-work policy in rapidly urbanizing delta cities.