Scholarlink

<p>Mitigating urban heat island effects is crucial for sustainable urban development, yet the spatial nonlinear and threshold effects of urban factors on land surface temperature (LST) remain poorly understood. To address this, an interpretable spatial machine-learning framework is developed by integrating geographically weighted machine learning models with Shapley Additive exPlanations. Using Shenzhen, China, as a case study, the proposed framework is validated. The effects of three categories of urban factors are examined, including urban morphology, land cover, and human and social activities. Results show that under geographically weighted coupling, the eXtreme Gradient Boosting model outperforms both multiple linear regression and random forest models. Results indicate that the impacts of urban factors on LST vary across time, and the human and social activities exert the greatest impact on diurnal LST across Shenzhen. Threshold analysis indicates that building heights above 13 m increase cooling effects during daytime. Building coverage ratios exceeding 0.2 exacerbate warming effects, highlighting the need for green infrastructure mitigation. High surface albedo areas experiencing daytime warming may benefit from the reduced use of heat-retaining material. The pervious surface fraction exceeding 0.3 would enhance cooling throughout the diurnal cycle. Furthermore, population density and gross domestic product significantly amplify warming effects beyond thresholds of 5497 persons/km<sup>2</sup> and 551.7 billion CNY/km<sup>2</sup>, respectively. These findings highlight the importance of threshold-based urban planning strategies for balancing daytime cooling and nighttime heat retention, providing valuable insights for sustainable urban development.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Mitigating urban thermal effects by analyzing nonlinear and threshold influences of urban factors using interpretable spatial machine-learning

  • 期刊论文

  • 发表日期：

    2026年2月4日

查看全文
  • Yonghang Xie,
  • Ali Hashemizadeh,
  • Yutian Lei,
  • Cheng Fan,
  • Hironori Watanabe

摘要

Mitigating urban heat island effects is crucial for sustainable urban development, yet the spatial nonlinear and threshold effects of urban factors on land surface temperature (LST) remain poorly understood. To address this, an interpretable spatial machine-learning framework is developed by integrating geographically weighted machine learning models with Shapley Additive exPlanations. Using Shenzhen, China, as a case study, the proposed framework is validated. The effects of three categories of urban factors are examined, including urban morphology, land cover, and human and social activities. Results show that under geographically weighted coupling, the eXtreme Gradient Boosting model outperforms both multiple linear regression and random forest models. Results indicate that the impacts of urban factors on LST vary across time, and the human and social activities exert the greatest impact on diurnal LST across Shenzhen. Threshold analysis indicates that building heights above 13 m increase cooling effects during daytime. Building coverage ratios exceeding 0.2 exacerbate warming effects, highlighting the need for green infrastructure mitigation. High surface albedo areas experiencing daytime warming may benefit from the reduced use of heat-retaining material. The pervious surface fraction exceeding 0.3 would enhance cooling throughout the diurnal cycle. Furthermore, population density and gross domestic product significantly amplify warming effects beyond thresholds of 5497 persons/km2 and 551.7 billion CNY/km2, respectively. These findings highlight the importance of threshold-based urban planning strategies for balancing daytime cooling and nighttime heat retention, providing valuable insights for sustainable urban development.