Sie befinden Sich nicht im Netzwerk der Universität Paderborn. Der Zugriff auf elektronische Ressourcen ist gegebenenfalls nur via VPN oder Shibboleth (DFN-AAI) möglich. mehr Informationen...
The accurate and rapid reconstruction of a pollution source represents an important but challenging problem. Several strategies have been proposed to tackle this issue among which we find the Bayesian solutions that have the interesting ability to provide a complete characterization of the source parameters through their posterior probability density function. However, these existing techniques have certain limitations such as their computational complexity, the required model assumptions, their difficulty to converge, the sensitive choice of model/algorithm parameters which clearly limit their easy use in practical scenarios. In this paper, to overcome these limitations, we propose a novel Bayesian solution based on a general and flexible population-based Monte Carlo algorithm, namely the sequential Monte Carlo sampler. Owing to its full adaptivity through the learning process, the main advantage of such an algorithm lies in its capability to be used without requiring any specific assumptions on the underlying statistical model and also without requiring from the user any difficult choices of certain parameter values. The performance of the proposed inference strategy is assessed using twin experiments in complex built-up environments.
•A statistical method is developed to solve the source term estimation problem.•This Bayesian solution can be used in many settings and does not require any specific model assumptions.•A full characterization of the uncertainties associated to the unknown quantities of interest is provided to the users.•The proposed approach is completely automatic and adaptive which therefore makes it easy to be used by non-experts.•The overall approach has been successfully validated through twin experiments in complex atmospheric environments.