Machine Learning Insight into the Cooling Intensity of Urban Blue-Green Spaces During Heatwaves

Abstract

Urban blue-green spaces are essential land cover types that play a critical role in mitigating urban heat stress. However, the cooling performance of these spaces during heatwave events is influenced by a complex interaction of topographical features and landscape configurations. This study examines the spatial variation in cooling intensity (CI) induced by blue-green spaces within the central urban area of Hue City, Vietnam. Land surface temperature in the study area was derived from Landsat 8 satellite imagery captured between 1 May and 30 September 2024, a period marked by record-high heatwaves. The analysis employs an extreme gradient boosting machine (XGBoost version 1.6.2) to quantitatively assess the relationship between CI and the contributing factors for various urban blocks. The XGBoost model demonstrates high predictive accuracy, shown by a coefficient of determination of 0.97. Notably, approximately 92% of predictions yield errors between −1 °C and +1 °C. Interpretation using SHapley Additive exPlanations helps identify primary influencing factors governing the CI. The presented framework presents a robust data-driven approach for evaluating the effectiveness of blue-green spaces in mitigating thermal stress in Hue City. These findings provide practical recommendations for urban planners aiming to enhance thermal comfort in the study area.