关键词: 城市2D/3D形态, 热岛效应, 可解释机器学习, XGBoost-SHAP模型

Abstract: With the increase of high-rise buildings in major cities worldwide, exploring the effects of urban two-dimensional (2D) and three-dimensional (3D) morphology on land surface temperature (LST) has become the key to mitigating the urban thermal environment and optimizing urban planning. Using the area within the Third Ring Road of Changsha as a case, we extracted 13 urban 2D/3D morphological factors based on 2020 multi-source remote sensing data. We used Pearson correlation analysis to examine the relationship between LST and each factor, and used the XGBoost model and SHAP method to reveal their nonlinear impacts and contributions. The results showed that in 2020, high-temperature regions mainly concentrated in the building-dense central area of Changsha, while low-temperature areas predominantly located in the forest parks in the western and northeastern parts of the city, as well as along the Xiangjiang River. The normalized difference building index (NDBI), nighttime lighting (NTL) and proportion of construction land (PCL) exhibited significant positive correlations with LST, with correlation coefficients of 0.592, 0.537 and 0.446, respectively, indicating that urbanization exacerbated surface warming. In contrast, the normalized difference vegetation index (NDVI) and the sky view coefficient (SVF) showed significant negative correlation with LST, with correlation coefficients of -0.316 and -0.200, respectively, reflecting the important role of green space and open space in mitigating the urban heat island effect. NDBI, NTL, NDVI, and elevation (DEM) had the greatest influence on LST, contributing 60.9% of the total variance. These 2D/3D morphological factors exhibited complex nonlinear effects on LST. NDBI had the most significant warming effect in the range from 0 to 0.2. The warming effect of NTL tended to saturate when its intensity exceeded 40. The cooling effect of NDVI became more pronounced as it surpassed 0.5. DEM values between 50 and 150 m produced the most signifi-cant cooling effect. This study validated the effectiveness of the XGBoost-SHAP model in uncovering the nonlinear mechanisms through which urban 2D/3D morphological factors influenced LST, offering scientific insights for urban heat management and the development of green, low-carbon, and livable urbanization.

Key words: urban 2D/3D form, urban heat island effect, explainable machine learning, XGBoost-SHAP model