Evaluating SU2 for Simulation of Urban Canopy Layer Flow

Investigating the impacts of urban morphology on local and microscale atmospheric conditions is a central goal of observational campaigns in urban environments. To fully assess this relationship, however, detailed information on the flow field at the time of observation is required in order to accurately trace the recent origin of the flow that reaches meteorological sensors in the urban canopy. While stationary and mobile wind observations can only represent point- or route-wise snapshots of this wind field, computational fluid dynamics can be used to estimate the three-dimensional flow field around in-situ meteorological (e.g., temperature) sensors.

To estimate the flow field at the time of a meteorological observation, we used the Computational Fluid Dynamics (CFD) model SU2 to compute a steady-state, three-dimensional wind field. SU2 is an open-source CFD code that is also available in a parallelized version, allowing for a quick computation of flow fields around arbitrary shapes. Since the computation of urban canopy layer flows is not a primary application of this tool, our goal is the evaluation of its utility in this setting. To investigate the accuracy of the simulation results, we determined the agreement between simulated and measured wind flow using high-resolution 2D wind observations recorded on the Tempe campus of the Arizona State University. The area around the sensor location was discretized using the information given by a high-resolution land use / land cover data set. A three-dimensional wind field was computed using a RANS-simulation of incompressible fluid, initialized with data from surrounding meteorological stations. In particular, we evaluated the development of an (idealized) vertical logarithmical wind profile in the undisturbed upstream area to account for realistic inflow behavior. To evaluate the distortion of the near-surface flow by surrounding buildings, the observed and temporally averaged 2D wind is compared to the 3D vectors computed for the corresponding location within the discretized model area.