POD Galerkin reduced-order model with ensemble Kalman filtering for urban street canyon flow predictions

In the context of climate change and growing urban populations, fast and accurate prediction tools for urban atmospheric flows are essential to help urban planners address increasingly critical issues such as air quality and thermal comfort. However, the turbulent, three-dimensional, and unsteady nature of these flows makes their numerical simulation computationally expensive. This study develops a reduced-order model (ROM) for isothermal urban street canyon flows, based on Proper Orthogonal Decomposition (POD) and Galerkin projection, whose inherent instability is mitigated through data assimilation (ensemble Kalman filtering) using sparse velocity measurements. This hybrid framework enables long-term prediction of the multi-scale turbulent flow at reduced computational cost, while keeping estimation errors stable and bounded. The reduction capabilities of POD are first examined, highlighting its limitations in reconstructing the flow within the canyon. The influence of the different parameters of the method on the predictive power of the model is then analysed. With a sufficient number of observation probes, the optimised reduced-order model achieves promising performance in predicting the instantaneous velocity field, especially the streamwise component. Key flow statistics, such as the Reynolds shear stress, are also well reproduced.