Influencing Factors and Predictive Modeling of the Urban Heat Island in Guangzhou, China

Constructing a predictive model for urban heat islands (UHIs) is essential for accurately assessing urban thermal environments and guiding sustainable development strategies. Previous studies typically modeled urban heat island based on the entire city, ignoring the differences of influencing factors between local areas. Therefore, this article explores the impact of influencing factors on local heat island based on Local Climate Zones (LCZ) zoning. We first prepared 21 features, and then created an urban heat island prediction model using machine learning technology it in each LCZ region. The prediction results were compared to those from Random Forest (RF), Gradient Boosting Trees (GBT), and Artificial Neural Networks (ANN). Experimental results indicate that the XGBoost model offers higher accuracy in predicting UHI, with accuracy exceeding 80%. The SHAP (SHapley Additive ExPlanations) analysis found significant elements impacting UHI formation in each zone, including impervious surface density (ISD), building density (BD), green space density (GSD), and the richness of vegetation. This study not only improves the accuracy of UHI predictions, but also provides the groundwork for future research into the dynamic planning of urban heat islands.