Bayesian source identification of urban-scale air pollution from point and field concentration measurements

Air pollution poses a major threat to health, environment, and global climate. Characterizing the emission parameters responsible for air contamination can help formulate appropriate response plans. We propose an advanced methodology that uses Markov Chain Monte Carlo (MCMC) sampling within a Bayesian inference framework to invert for emission parameters of air contamination in an urban environment. We also use a high-resolution Lagrangian dispersion model to provide microscale wind computations as well as pollution concentration values in the presence of urban features with high complexity. Buildings and land use features were all integrated in a realistic urban setup that represents the region of King Abdullah University of Science and Technology, KSA. Boundary meteorological conditions acquired from a Weather Research and Forecasting (WRF) model simulation were employed to obtain the mesoscale wind field. We design numerical experiments to infer two common types of reference observations, a pollutant concentration distribution and point-wise discrete concentration values. The local L2 norm and global Wasserstein distance are investigated to quantify the discrepancies between the observations and the model predictions. The results of the conducted numerical experiments demonstrate the advantages of using the global optimal transport metric. They also emphasize the sensitivity of the inverted solution to the available observations. The proposed framework is proven to efficiently provide robust estimates of the emission parameters. 

Air pollution dispersion modeling Bayesian Inference Optimal transport Source term estimation Urban environment