Precise mitigation strategies for urban heat island effect in Hong Kong's new towns using automated machine learning
Yiyan Li, Hongsheng Zhang, Yinyi Lin, Jing Ling, Huiyuan Xue, Peizhuo Guo
Abstract
• Integrated multi-source data for comprehensive quantification of the built environment. • shown the spatial distribution of built environment's impact on LST. • Dominant factors of urban heat islands show spatial heterogeneity in Hong Kong New Towns. • Urban woodland is the most influential factor on land surface temperature. New town developments aim to enhance the spatial layout and quality of living environments in metropolitan areas. These areas are vulnerable to the urban heat island (UHI) effect owing to high-density development and poor long-term planning. However, few studies have investigated the spatial distribution of the influence of built environments on UHI, limiting the ability of urban planners to develop targeted mitigation strategies. To address this gap, we analyzed the complex spatial relationship between land surface temperature (LST), an important indicator of UHI, and the built environment in all new towns in Hong Kong. We employed remote sensing images, street view images, geographical information science (GIS) data, and land-use data with automated machine learning to model the LST-built environmental relationship at various spatial scales, using Shapley Additive Explanations (SHAP) to interpret the model to show the spatial distribution of the influence of the built environment on UHI. Our best model was the Extreme Gradient Boosting Machine model with a 210-m grid R-squared value of 0.79. We found that 1) the land-use feature class had the most significant influence on LST, and 2) there was spatial heterogeneity among the major contributors to UHI. The refined UHI attribution analysis method proposed in this study enables precise modeling for smart and sustainable city planning. (211 words)