Examining urban agglomeration heat island with explainable AI: An enhanced consideration of anthropogenic heat emissions

Abstract

In the context of global warming and urbanization, regional economic concentration has increased anthropogenic heat emissions (AHE), posing significant threats to health and sustainability. The oversimplification of AHE in previous urban heat island studies hinders the development and implementation of AHE-reduction strategies aimed at mitigating high land surface temperature (LST). Therefore, this study reevaluates the regional heat island (RHI) effect in the Greater Bay Area (GBA) using multisource geo-big data. The analysis reveals that central RHI intensity (RHII) exceeds 3 °C, indicating a significant heat island. We constructed an integrated LightGBM model with four AHE and other classical indicators to fit LST, achieving an R2 of 0.8931. To improve the model's interpretability, we utilized SHapley Additive exPlanations (SHAP), which identified NDVI, DEM, and building AHE as significant indicators influencing LST in the GBA, each with SHAP values exceeding 0.25. Simulations of three intensity scenarios for tiered AHE reduction strategies show that a 10 % industrial AHE reduction in heavy industrial cities can cool 40 % of these areas and decrease RHII by more than 0.03 °C. This study provides actionable guidelines for targeted RHI mitigation in the GBA and provides valuable insights for evaluating RHI in other bay areas and urban agglomerations.