Bayesian inversion of inflow direction and speed in urban dispersion simulations

Abstract

Computational fluid dynamics (CFD) is a powerful tool to investigate airflow and pollutant dispersion in urban areas and can provide critical information for urban planning and public health assessment. Although CFD approaches can reproduce detailed airflow and dispersion results in complex areas, it is extremely sensitive to inflow parameters. The conventional way to set inflow conditions is to measure the mean wind direction at a single upwind detector and adopt it as the inflow direction in the CFD model. In this paper, a novel method is proposed to provide an alternative way to obtain the inflow parameters in urban dispersion simulations by using multiple near-field wind detectors within urban areas and a probabilistic inverse method, the Bayesian inference. A field release experiment in a real built-up area is adopted as the test case. Simulation results of both wind and concentration are compared between the proposed and conventional methods. The proposed method provides the probability distributions of inflow parameters, which well reproduce the near-field airflow field in the urban area. With respect to the dispersion simulation, the statistical analysis shows better performance of the proposed method in the point estimates of concentrations than the conventional method. In addition, the proposed method yields the probability distributions of predicted concentrations, which offers a rigorous quantification of uncertainties in urban dispersion simulations.