High-resolution downscaling simulation of urban wind-thermal environments: An interpretable machine learning analysis of urban morphology

Accurate assessment of urban microclimates is essential for evaluating residents’ thermal comfort and building energy consumption, particularly when mitigating urban heat islands and optimizing urban ventilation corridors. In dense urban settings, mesoscale climate, building structures, and urban heat exchange collectively shape the local microclimate. However, obtaining high-precision microclimate data—and quantifying the influence of urban elements—remains methodologically challenging. To address this gap, we combined the Weather Research and Forecasting (WRF) model with LCZ classification to downscale simulations of the wind-thermal environment along Shanghai’s Outer Ring Road. Urban morphological parameters were extracted from multi-source data, and their importance was evaluated using interpretable machine learning.

The results reveal that average building height and frontal area index are the most critical morphological features influencing temperature and wind speed, respectively. This integrated methodology not only advances climate-adaptive urban design but also provides accurate boundary conditions for building energy-saving strategies and outdoor thermal comfort assessments.