Comparative Analysis of Machine Learning Models for Urban Heat Island Prediction in Malang City Using Remote Sensing Data

This study investigates the Urban Heat Island (UBI) phenomenon in Malang City, Indonesia, utilizing advanced remote sensing technologies and machine learning methods to enhance Land Surface Temperature (LST) predictions. Urbanization, driven by infrastructure development outlined in Regional Regulation No.4 of 2011, has led to a significant reduction in green open spaces, exacerbating the UBI effect. We leverage Google Earth Engine (GEE) to integrate satellite data from Landsat-8 and Sentinel-2, alongside elevation data from the Shuttle Radar Topography Mission (SRTM). Employing various predictive models, including Support Vector Regression (SVR), Gradient Boosting, and XGBoost, this research introduces the Urban Density Index (UDI) and elevation as critical variables impacting UBI dynamics. The study employs Z-score normalization to standardize data ranges, significantly improving model accuracy. Results indicate that SVR outperforms other models with the lowest Root Mean Square Error (RMSE) of 0.662 and a coefficient of determination (R2) of 0.854. Strong correlations are observed between UDI (0.91) and LST, highlighting urban density as a key contributor to elevated temperatures, while elevation shows a negative correlation (-0.41), indicating cooler conditions at higher altitudes. These findings underscore the importance of integrating UDI and elevation in UBI analyses, contributing valuable insights for urban planning and environmental management in rapidly urbanizing areas.