The thermal footprint of urbanisation in a typical mid-scale African city: heat exposure and explainable machine-learning attribution of surface-temperature drivers

Urban expansion is warming African cities, yet its drivers and the exposed population are poorly quantified for mid-sized cities. This study assessed land surface temperature (LST), surface biophysical conditions and thermal stress exposure in Akure, Nigeria, a typical mid-scale West African city, from 2014 to 2025. Annual Landsat 8 composites provided LST and multi-source remote sensing indicators of vegetation, moisture and built-up cover, and their trends were tested with non-parametric methods. A thermal stress exposure index combined LST with spatial population data. Three tree-based machine learning methods were compared under ten-fold cross-validation, and explainable machine learning attributed LST to vegetation, elevation, population and built-up density. LST increased by 0.22°C per year (p = 0.016), about 2.4°C over the study period. Vegetation vigour and surface moisture declined significantly, while built-up cover increased. The selected model explained 93% of LST variance on test data (root mean square error 0.43 °C). Vegetation vigour was the strongest predictor, about three times as influential as population, raising LST by up to 2.8°C where vegetation was sparse and lowering it by up to 2.4°C where it was dense. Thermal stress exposure increased significantly (p = 0.005). The study identifies vegetation loss as the main factor associated with surface warming in an under-studied type of city. Because the framework uses only open global data, it can be transferred to other cities to locate where greening would most reduce heat exposure, supporting global goals on health, sustainable cities and climate action.