Predicting land surface temperature under alternative urban growth scenarios using integrated FLUS–XGBoost modelling

Abstract Land surface temperature (LST) is the radiative temperature of the Earth’s surface and is strongly influenced by urbanisation and land use change. Understanding future land–atmosphere interactions is essential for managing thermal risks in rapidly growing cities. However, few studies have integrated scenario-based land use modelling with machine learning to evaluate future urban thermal environments. This study develops and validates an integrated future land use simulation-XGBoost framework to analyse historical (2004–2024) and predict future (2034) LST under multiple urban growth scenarios in Varanasi, India. The results indicate a substantial transformation in land use, with built-up areas increasing from 31.45 km 2 to 75.11 km 2 , accompanied by a substantial rise in mean LST of 5.3 °C. A strong relationship between land use land cover and LST is observed, with vegetation providing cooling effects and exposed surfaces contributing to thermal hotspots. Scenario-based analysis indicates that the sprawl scenario produces the highest LST, while the Ecological scenario reduces temperatures by approximately 3 °C–4 °C. The findings highlight the importance of land use planning in mitigating urban heat and provide a robust framework for climate-sensitive urban development.