Spatial heterogeneity of urban morphology and land surface temperature based on an interpretable GWRF– SHAP machine learning model
作者
Yuanzhen Song
出处
期刊:Journal of physics [IOP Publishing] 日期:2025-07-01卷期号:3060 (1): 012040-012040被引量:1
标识
DOI:10.1088/1742-6596/3060/1/012040
摘要
Abstract As the urban heat island (UHI) effect becomes increasingly severe, urban morphology has been widely recognized as a practical approach to mitigating this phenomenon. However, existing studies have primarily relied on global linear models, with limited consideration given to urban morphological impacts' nonlinear characteristics and spatial heterogeneity. This study developed an analytical framework integrating a geographically weighted random forest (GWRF) model and SHapley additive explanations to systematically uncover urban morphology's nonlinear effects and spatial heterogeneity on land surface temperature (LST). First, we obtained LST data for the central urban area of Tianjin from 2013 to 2023 using the Google Earth engine platform. We analyzed spatiotemporal trends using the Mann–Kendall test and Sen's slope estimator. Subsequently, we integrated the GWRF model with SHAP values to quantify the nonlinear influences and spatial heterogeneity of urban morphological variables. The results indicate that: (1) From 2013 to 2023, LST in Tianjin’s central urban area exhibited a fluctuating upward trend, with annual dispersion decreasing year by year; (2) Significant warming areas were primarily located in the southwestern part of the central city, whereas cooling zones were mainly concentrated in the eastern and northern regions; (3) The GWRF model significantly outperformed traditional linear models in terms of goodness-of-fit, identifying NDVI as the primary cooling factor and revealing the spatial heterogeneity of morphological influences. This study provides theoretical and empirical support for formulating context-specific strategies to regulate land surface temperature.