Estimating Source Term Parameters through Probabilistic Bayesian inference: An Approach based on an Adaptive Multiple Importance Sampling Algorithm
Abstract
This paper presents an adaptive approach based on probabilistic Bayesian inference to estimate the parameters of an atmospheric pollution source term. After introducing the problem and assessing the computational framework, we present an Importance Sampling based algorithm called Adaptive Multiple Importance Sampling (AMIS). It performs an efficient calculation of the source parameter posterior distribution by iteratively upgrading the proposal's parameters and recycling all generations of weighted samples, thus allowing a faster convergence and reducing the number of necessary iterations. We highlight the results of the AMIS by comparing it to a MCMC estimation in a simple example.