Abstract
Urban heat island (UHI) effect poses significant environmental and health risks, particularly in rapidly urbanizing cities. Green spaces are known to mitigate UHI through shading and evapotranspiration, but their distribution is often suboptimal. This study integrates machine learning (ML) and satellite imagery to optimize urban green space distribution for UHI mitigation in Mumbai, India. Landsat 8 and Sentinel-2 data were used to derive land surface temperature (LST) and green space indices (NDVI, NDWI) for 2020–2023. A Random Forest model predicted LST with high accuracy (R² = 0.92, RMSE = 1.2°C), identifying built-up density and vegetation cover as key predictors. A genetic algorithm was then employed to optimize green space allocation, minimizing mean LST subject to land availability constraints. Results show that strategic placement of green spaces can reduce average LST by up to 2.5°C compared to current distribution. Optimized scenarios prioritize high-density residential and industrial zones, where UHI intensity is highest. Spatial equity analysis indicates that optimized distributions also reduce thermal disparities between socioeconomic groups. This framework demonstrates the potential of combining ML with satellite data for evidence-based urban planning, offering a scalable approach for heat-resilient cities.