Scholarlink

<p>Urban Heat Islands (UHIs) and rising surface temperatures are emerging as critical challenges in the context of rapid urbanization and climate change. This study presents a machine learning-based framework for forecasting urban heat patterns and analyzing microclimate dynamics, using Delhi, India, as a representative case of a densely populated and fast-growing urban environment. Long Short-Term Memory (LSTM) and Prophet models were employed to forecast temperature trends from 2003 to 2021, with LSTM outperforming Prophet in all metrics: Mean Absolute Error (MAE) of 0.92 vs. 1.89, Root Mean Squared Error (RMSE) of 1.23 vs. 2.38, and R<sup>2</sup> of 0.98 vs. 0.91, respectively. Microclimate monitoring at three urban locations revealed significant thermal stress, with globe temperatures (Tg) reaching up to 43.3&#xa0;°C and Predicted Percentage of Dissatisfied (PPD) exceeding 96% in high-density areas. Thermal comfort indicators such as WBGT, PMV, and wind speed demonstrated strong correlations with surface morphology and vegetation cover. SHAP (SHapley Additive exPlanations) analysis confirmed the dominance of annual climatic trends (+ 3.56 SHAP value) over weekly variations in forecasting accuracy.</p><p>While the study is geographically situated in India, the methodology, predictive framework, and interpretive tools offer high scalability and applicability to other global cities facing similar climatic stress. The results provide critical insights for urban planners, policymakers, and climate resilience practitioners worldwide. The integrated use of AI-driven forecasting and empirical microclimate data serves as a replicable model to support evidence-based urban heat mitigation and adaptive urban design strategies in both the Global South and Global North cities.</p> Graphical abstract <p></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

Smart cities, hot cities: ML-based forecasting of urban heat patterns

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月17日

查看全文
  • Sagar Tomar,
  • Kishor Sitaram Kulkarni

摘要

Urban Heat Islands (UHIs) and rising surface temperatures are emerging as critical challenges in the context of rapid urbanization and climate change. This study presents a machine learning-based framework for forecasting urban heat patterns and analyzing microclimate dynamics, using Delhi, India, as a representative case of a densely populated and fast-growing urban environment. Long Short-Term Memory (LSTM) and Prophet models were employed to forecast temperature trends from 2003 to 2021, with LSTM outperforming Prophet in all metrics: Mean Absolute Error (MAE) of 0.92 vs. 1.89, Root Mean Squared Error (RMSE) of 1.23 vs. 2.38, and R2 of 0.98 vs. 0.91, respectively. Microclimate monitoring at three urban locations revealed significant thermal stress, with globe temperatures (Tg) reaching up to 43.3 °C and Predicted Percentage of Dissatisfied (PPD) exceeding 96% in high-density areas. Thermal comfort indicators such as WBGT, PMV, and wind speed demonstrated strong correlations with surface morphology and vegetation cover. SHAP (SHapley Additive exPlanations) analysis confirmed the dominance of annual climatic trends (+ 3.56 SHAP value) over weekly variations in forecasting accuracy.

While the study is geographically situated in India, the methodology, predictive framework, and interpretive tools offer high scalability and applicability to other global cities facing similar climatic stress. The results provide critical insights for urban planners, policymakers, and climate resilience practitioners worldwide. The integrated use of AI-driven forecasting and empirical microclimate data serves as a replicable model to support evidence-based urban heat mitigation and adaptive urban design strategies in both the Global South and Global North cities.

Graphical abstract