Machine Learning Driven Land Surface Temperature Prediction and Urban Heat Risk Assessment in The Gambia

Abstract Urban Heat Island (UHI) effects are intensifying across rapidly urbanizing African cities, yet limited empirical research exists for West Africa. This study investigates the spatiotemporal relationship between land use/land cover (LULC) dynamics and land surface temperature (LST) in the Greater Banjul Area (GBA), The Gambia, over the period 1990–2020, and projects future heat risks for 2040. Multi-temporal Landsat and Sentinel imagery were used to derive LULC, LST, NDVI, and NDBI. Results indicate a 36.2% increase in built-up are accompanied by a rise in mean LST from 28.2°C in 1990 to 35.7°C in 2020, with UHI intensities peaking at 48°C in 2010. Highest UHI impacts are concentrated in highly urbanized zones such as Banjul and Kanifing. Machine learning models were applied to predict LST using environmental predictors; XGBoost outperformed Random Forest (RMSE = 1.97°C, R² = 0.92). Feature importance analysis confirmed LULC (82–90%) and NDVI (65–70%) as dominant predictors of LST. Projections for 2040 indicate a mean LST increase to 37.6°C, with parks (+ 1.9°C), dumpsites (+ 1.8°C), and vacant lands (+ 1.8°C) showing the strongest warming. Heat Stress Index (HSI) assessment suggests that nearly half the population will face high to extreme heat stress by 2040. By linking remote sensing and machine learning with heat stress assessment, this study demonstrates how rapid urbanization is amplifying heat risks in GBA. Study findings highlight the strong influence of urban expansion and vegetation loss on thermal environments in coastal West Africa and provide a replicable framework for UHI and heat risk assessment in other data-scarce regions.