The rapid identification of contaminant plume sources in urban environments can greatly enhance emergency response efforts. Source identification based on downwind
concentration measurements is complicated by the presence of building obstacles that can cause flow diversion and entrainment. While high-resolution CFD simulations
are available for predicting plume evolution in complex urban
geometries, such simulations require large computational effort. We make use of an urban puff model, the Defence Science Technology Laboratory's (Dstl) Urban Dispersion
Model (UDM), which employs empirically based puff splitting techniques. UDM greatly reduces urban
dispersion simulations by combining traditional Gaussian puff modeling with empirically deduced mixing and entrainment approximations. Here we demonstrate the preliminary
reconstruction of an atmospheric release event using stochastic sampling algorithms and Bayesian inference together with the rapid UDM urban puff model based on point measurements of concentration. We consider source inversions for both a prototype isolated building (a cube) and for observations and flow conditions taken during the Joint URBAN 2003 field campaign at Oklahoma
City.