The Urban Heat Island (UHI) effect, characterized by elevated urban temperatures compared to surroundings, poses serious challenges to human health, energy demand, and urban sustainability. Remote sensing has become a key tool to map and monitor UHI via land surface temperature (LST) and land cover/land use variables. In recent years, machine learning (ML) techniques—including regression, random forests, neural networks, support vector machines—have been increasingly integrated with remote sensed data to improve prediction of UHI dynamics. This review synthesizes studies from 2015 to 2020 on methods, data sources, ML models, and predictors used in forecasting or modeling UHI. Key findings include: (i) remote sensing sensors such as Landsat, MODIS, Sentinel, and derived indices (NDVI, NDBI, albedo, sky view factor) are widely used; (ii) ML models often outperform classical linear regression when data is sufficient; (iii) spatial and temporal resolution matters critically for predictive accuracy; (iv) major predictors are impervious surface, vegetation fraction, built-up density, albedo, surface moisture. Gaps include lack of consistent cross-city or cross-climatic comparative studies, limited use of deep learning (especially convolutional neural networks) for spatial-temporal forecasting of UHI, and issues of generalization across sensors or regions. The review concludes with recommendations: more multi-sensor data fusion; standardization in ground-truthing; integrating climatic, morphological, socio-economic features; development of ML models with transferability; and focus on predictive forecasting (not just retrospective modeling) for urban planning and mitigation.
