Estimating a CBRN atmospheric release in a complex environment using Gaussian processes

In this paper, we present a new methodology for the estimation and the prediction of the concentration of pollutant in a complex environment. We take benefit of a semi-parametric formulation of the problem to perform a faster and more efficient estimation of the pollutant cloud. In a first part, we present how we use the Gaussian process to model the interactions between position and time given the observations. Then, we introduce the expansion as a function of the observations through the time, and we construct an estimator of the time of release from it within change-point detection framework. Then, we use this time estimate to obtain the position (or more likely, a confidence region of the position) of the source. Several simulations are provided in a complex city scenario that demonstrate the accuracy of the proposed technique.