Physics-Informed Hybrid Model for High-Resolution Prediction of Urban Thermal–Wind Environments Under Present and Future Climate Scenarios

Abstract

Accurately and efficiently simulating large-scale urban wind–thermal environments remains a major challenge. This study proposes a physics-informed hybrid modelling approach that couples the Weather Research and Forecasting (WRF) model, which captures mesoscale meteorological processes, with an improved CNN for rapid data-driven prediction. Using high-density Shanghai as a case study, the model effectively predicts the spatio-temporal patterns of urban wind and temperature under both current and future climate scenarios. The hybrid framework effectively captures mesoscale physical mechanisms and local urban morphological effects, and demonstrates strong interpretability, accuracy, and computational efficiency. The mean prediction errors are 0.97 °C for temperature and 0.47 m/s for wind speed. Thanks to its lightweight structure, the hybrid model can be trained and deployed without substantial computational resources, achieving a speed-up of about two orders of magnitude compared with traditional physics-based simulations. Post-hoc analyses indicate that water surface ratio (WSR) and sky view factor (SVF) are the most influential morphological factors for temperature, while green coverage ratio (GCR) and building footprint ratio (BFR) have stronger impacts on wind speed, and reveal their nonlinear impacts on urban climate. Under high-emission scenarios in 2050 and 2080, the daily maximum temperature is projected to increase by about 2–4 °C and 4–7 °C, respectively, while peak wind speed decreases by about 0.3–0.6 m/s and 0.6–1.0 m/s. The expansion of hot and low-ventilation areas indicates a higher thermal risk in dense urban districts. Overall, the proposed hybrid model provides an efficient, accurate, and transferable tool for future urban climate assessment and offers valuable support for building energy simulations and urban cooling strategy design.

Keywords: Urban thermal-wind environment, WRF, Hybrid prediction model, Future climate change