Urban environments introduce strong spatial heterogeneity into atmospheric flow, heat transfer, and pollutant transport, yet many of these processes occur at scales that are not explicitly resolved in regional atmospheric models. Buildings, surface morphology, thermal heterogeneity, and turbulent inflow can substantially modify local flow and transport within the atmospheric roughness sublayer, creating an important gap between regional-scale atmospheric information and neighborhood- and building-scale conditions.

This talk presents our efforts to address this gap through multi-fidelity surrogate modeling (MFSM). The work examines how urban morphology and atmospheric inflow conditions influence turbulent flow, wind loading, thermal transport, and environmental exposure across individual-building to neighborhood scales. Boundary-layer wind-tunnel experiments are used to validate CFD models across representative built-environment configurations, which are then used to characterize normalized flow and transport features. These relationships can be further examined under different atmospheric, climate, and terrain conditions using observational data. Large ensembles of computationally efficient models, supported by selected high-fidelity simulations and experimental data, are used to retain important local physics while extending the analysis across different morphological and atmospheric conditions. Data-driven and machine-learning approaches are also explored to improve the scalability and generalization of the surrogate models.

The resulting framework can support questions of when neighborhood-scale representation is necessary, what level of modeling fidelity is justified, and how limited observations can be used effectively in heterogeneous urban environments. Applications include urban microclimate and heat exposure, wind engineering, air quality and environmental transport, and other problems where local built-environment effects are not resolved by larger-scale atmospheric information. The talk will also discuss opportunities for connecting these local-scale modeling capabilities with regional atmospheric observations and models.