Machine Learning in Modeling Urban Heat Islands: A Data-Driven Approach for Kuala Lumpur

This study presents a novel machine learning-based approach by integrating urban land surface indices, including the Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-up Index (NDBI), Urban Index (UI), Normalized Difference Water Index (NDWI), and Albedo, derived from high-resolution Landsat 8 data, to quantitatively assess the effectiveness of Urban Heat Island (UHI) mitigation strategies for Kuala Lumpur. Nonparametric correlation analysis was used to select the most suitable land features for developing Machine Learning (ML) models and predicting Land Surface Temperature (LST). The results showed that Kuala Lumpur’s LST had risen to 2.2 °C between 2013 and 2023, driven by urban development and the resulting UHI effect. Comparative analysis of the ML models revealed that the random forest (RF) model best estimated LST, with a Kling-Gupta Efficiency (KGE) of 0.68 and a spatial bias of ± 1.6 °C. Application of the RF model showed that an improvement in NDVI by 25% can cause a drop in LST ranging from − 1.4 to 0.1 °C, while a 25% increase in albedo can decrease the LST by -1.2 to 0.1 °C. The reduction in LST is highest in areas with high LST, indicating the possibility of mitigating the extreme UHI effect by enhancing albedo and NDVI. This study offers a data-driven alternative to costly numerical simulations, making it one of the first applications of ML for UHI modeling in a tropical megacity like Kuala Lumpur.