A CGAN-Based Method for Predictive Generation of Wind Environment Images in Stadium-Centered Urban Blocks

Abstract Creating a comfortable outdoor environment is crucial for the sustainable development of urban spaces and the quality of human habitation. The agglomeration effect of stadiums often drives surrounding urban blocks toward high-density development. However, the long interface of stadiums combined with dense surrounding buildings frequently leads to local issues such as excessive wind speeds, pollutant accumulation, and uneven ventilation, which negatively impact pedestrian comfort and natural ventilation performance of buildings. Therefore, conducting wind environment assessments in the early planning stage holds significant practical value. Traditional wind environment analysis methods, however, are computationally expensive and inefficient for evaluating and optimizing a large number of design proposals, limiting their applicability in urban design workflows.To address this challenge, this study proposes a CGAN-based deep learning method trained on 80 sets of wind environment cloud maps of high-density stadium-centered urban blocks. The block and stadium heights are encoded using luminance depth. The model’s performance is further evaluated in stadium shape. Experimental results demonstrate that the CGAN model significantly accelerates wind environment prediction while maintaining acceptable accuracy, offering a practical tool for rapid evaluation of early-stage urban planning schemes.This article is supported by the Heilongjiang Provincial Natural Science Foundation Project of the Heilongjiang Provincial Department of Science and Technology, titled "Research on Multi-scenario Conversion Mechanism of Ice and Snow Sports Buildings Driven by Digitalization", with Lei Li as the project leader and the project number being LH2024E049.There are no competing interests in this article.Guo and Cao wrote the main manuscript text and Guo prepared all figures.All authors reviewed the manuscript.