Comparisons of JU2003 observations with four diagnostic urban wind flow and Lagrangian particle dispersion models

Abstract Urban wind flow and dispersion models are needed that can satisfactorily account for the effects of the three-dimensional (3D) building geometries but which run much faster than Computational Fluid Dynamics (CFD) models. With sufficient speed, the models can be used for rapid response and for applications where many simulations must be performed in a short time period. To satisfy this need, several diagnostic wind flow models have been developed for urban areas, where mass-consistent principles are used in combination with local wind observations to solve for the mean wind flow on domains of size ranging from a few hundred meters to several kilometers on a side, within which detailed 3-D building geometries are defined. Simple assumptions about vortex flow structures are parameterized near buildings and in street canyons. The wind flow results are used as inputs to a Lagrangian particle dispersion model (LPDM), where the needed turbulent velocities and time scales are parameterized using standard boundary layer profile formulas combined with special relations around buildings. As part of a collaborative study, with the intent of advancing each model, the developers of four of these models have run their models for two tracer releases (one daytime and one nighttime) during the Joint Urban 2003 (JU2003) field experiment. The four models are: QUIC by Los Alamos National Laboratory (LANL), 3DWF by the Army Research Laboratory, the urban Lagrangian model by the Israel Institute for Biological Research (IIBR), and Microswift/Spray (MSS) by Aria Technologies and SAIC. The comparison uses nearly identical domains and grid systems, and all models use the same input wind profile. The simulated patterns of wind fields and tracer contours are in good qualitative agreement. For wind speed near the surface, the mean model biases are less than about 20% and RMS errors are about 1–2 m s −1 . For tracer concentrations, the four models give similar quantitative results, where the mean relative biases suggest that the individual models can be sometimes as much as a factor of two high or low, and where the scatter suggests that, for all models, about 30 or 40% of the simulations are within a factor of two of observations. In most cases, the observed plume is broader than the simulated plume, and the models are biased toward slight underestimation of the dispersion of the plumes to the tall rooftops.