POD-Kriging surrogate modeling for rapid prediction of the impact of façade protruding rib geometries on urban canyon wind flow

Understanding the impact of building geometries, e.g., façade protrusions such as ribs, is crucial for accurate urban wind flow predictions. While computational fluid dynamics (CFD) with large-eddy simulations (LES) provides reliable wind-field predictions in urban canyons with façade-protruding elements, they are too time-consuming and resource-intensive for rapid assessments. To address this challenge, this study develops a reduced-order surrogate model using the Proper Orthogonal Decomposition (POD) and Kriging surrogate model to predict wind fields in typical urban canyons with various rib-like horizontal protrusions under perpendicular approaching wind. Based on CFD data from LES simulations, the model is trained to predict the flow fields inside the canyon using two geometric parameters, depth d and separation s of protrusion ribs. First, the POD method is employed to decompose the flow fields from LES simulations into a POD mode basis. Then, the Kriging surrogate model is trained to learn the relation between the POD mode coefficients and the rib's geometric parameters. A series of evaluations of the POD-Kriging surrogate model's performance is conducted, and the results show good agreement with LES results, achieving a mean absolute error of 0.012 m/s, which is superior to Reynolds-averaged Navier-Stokes (RANS) CFD simulations. Discrepancies are confined to high-gradient regions, while bulk flow predictions are highly reliable. This approach is 7 × 10³ times faster than LES and 6 × 10² times faster than RANS, significantly enhancing the efficiency of predicting wind fields. © 2026 Elsevier Ltd.