Exploring the non-linear impacts of environmental and anthropogenic factors on land surface temperature in southwestern Bangladesh using explainable AI

Land surface temperature (LST) in southwestern Bangladesh has changed rapidly in recent years due to both environmental conditions and human activities. This study examines how LST varies over space and time in the Jashore–Jhenaidah–Chuadanga region and identifies the main factors controlling these changes between 2015 and 2025. A total of 4600 spatial sample points were created by combining Landsat-derived LST with vegetation, moisture, built-up, air pollution, population, and building-related indicators. Three machine-learning models, Random Forest Extreme Gradient Boosting, and Gradient Boosting Machine, were trained using 80% of the data and tested with the remaining 20%. The results show a clear north–south warming trend across the study area over time. Among the three models, RF performed best, with R 2 values increasing from 0.79 in 2015 to 0.93 in 2025 and RMSE remaining below 1.0 °C. To understand how different factors influence LST, Partial Dependence Plots and SHapley Additive exPlanations were applied. In 2015, vegetation and surface moisture played the strongest cooling roles, while built-up land increased surface temperature. By 2025, air pollution (PM 2.5 ) and building density became the main warming factors, while building height emerged as the strongest cooling factor. Elevation showed a moderate but consistent influence throughout the study period. These findings show that LST in southwestern Bangladesh is controlled by a combination of land cover, urban form, and human activity, with their relative importance changing over time. The results can help urban planners identify priority areas for green space protection, building design, and pollution control in rapidly growing regions.