Machine learning for urban wind simulation: A comprehensive review

Urban wind simulation plays an important role in pedestrian comfort assessment, wind safety evaluation, air quality analysis, and broader urban environmental planning. Conventional computational fluid dynamics (CFD) approaches, although physically grounded, remain computationally demanding and highly reliant on mesh quality, which makes them impractical for large-scale or iterative urban design workflows. This study reviews recent developments in data-driven, physics-informed, and hybrid machine learning methods for urban wind prediction, with attention to their modeling assumptions, validation strategies, and generalization behavior. The literature is organized along the urban wind simulation pipeline to provide a structured overview of current surrogate modeling approaches and their practical limitations. The review highlights open challenges related to robustness, benchmarking, and uncertainty quantification, and discusses directions for improving the reliability and scalability of ML-based urban wind modeling.