An inter-comparison of three urban wind models using Oklahoma City Joint Urban 2003 wind field measurements

Three wind models are compared to near-surface time-averaged wind measurements obtained in downtown Oklahoma City during the Joint Urban 2003 Field Campaign. The models cover several levels of computational approximation and include in increasing order of computational demand: a mass-consistent empirical-diagnostic code, a Reynolds-averaged Navier-Stokes (RANS) computational fluid dynamics (CFD) model, and a Large Eddy Simulation (LES) CFD code. The models were run with identical inlet and boundary conditions using the same grid resolution; the choice of the specific computational set-up reflects demands for fast-response models, although it may be a sub-optimal choice for the more complex models. A qualitative comparison of the model-computed flow fields with the Joint Urban 2003 wind measurements shows that all three models compare favorably to the near-surface wind measurements in many locations, although there are often instances of winds being calculated poorly in specific locations. The CFD models, however, had clearly superior looking flow fields, whereas the empirical-diagnostic code produced fields that were less smoothly varying. The inter-comparison exercise was supported by point-by-point quantitative comparisons of the wind speed and wind direction and with statistical measures. The RANS-CFD code, for example, was within 50% of the measured wind speed 62% of the time as compared to 53% for the LES model and 49% for the empirical-diagnostic code. For wind direction, the RANS-CFD code was within 30 of the measured wind direction 58% of the time as compared to 50% for the LES code and 43% for the empirical diagnostic code. It is noticeable that throughout the various 10P cases examined, and under the specific computational set-up used in the simulations for fast-response needs, there is no clear superiority of one model over another. In addition, for the LES model, which in theory should provide the most realistic representation of the flow field, it appears that further to the sub-optimal computational set-up, as well as the uncertainty and natural variability persistent in the real world, has resulted in diminished performance. (C) 2011 Elsevier Ltd. All rights reserved.