Past large-scale fires in urban areas have stressed the need to develop means of assessing the risks posed by smoke plumes to the population and the environment. One of the challenges is to quickly inform the authorities on the areas impacted by the plume and the pollutant concentration levels to which people may have been exposed. We present in this work an inverse Bayesian method, based on Markov Chain Monte Carlo (MCMC) principle, developed to retrieve the source term of a large-scale fire by assimilation of in-situ pollutant concentration measurements. The inverse method was used to characterise the source of a large warehouse fire near Paris in 2021 and the associated uncertainties. Focusing on model error, a sensitivity analysis is performed on the dry deposition velocity used in the forward modelling.