Modelling urban heat island effects: a global analysis of 216 cities using machine learning techniques

Abstract Urban areas globally have become home to over half of the world's population, leading to the intensification of the urban heat island (UHI) effect, where cities experience higher temperatures than their rural counterparts. The current study develops a new model predicting UHI intensity for 216 cities across all climate zones for both the Global North and Global South using machine learning techniques, focusing on the years 2019 to 2023. Utilising a novel dataset, integrating climate, economic, population, and land use data from 216 cities worldwide, the model, trained using Support Vector Regression (SVR), demonstrates a mean absolute error (MAE) of 0.86 °C. Results reveal that wind speed significantly mitigates UHI intensity, while cities in temperate climates exhibit more pronounced UHI effects compared to those located within tropical climbs. Additionally, results show the crucial role of coastal proximity in reducing UHI intensity and find no significant differences in UHI intensity between cities in the Global North and Global South. Findings offer important empirical actionable insights alongside a robust tool for urban planners and policymakers to measure, map, and monitor the UHI effect, contributing to the development of more liveable and sustainable urban environments.