Source term estimation using an adjoint model: a comparison of two different algorithms

The location of an unknown source and its pollutant emissions are estimated from concentration observations by means of two approaches, both making use of the adjoint of a Lagrangian particle dispersion model. In the first approach, plausible source locations are estimated by identifying areas with maximum spatial and temporal consistency among backward trajectories from each sensor. In the second approach, a variational method is used to minimise the objective function and to estimate emissions at each grid-box reached by backward trajectories: the resulting map provides information on the source location and on its uncertainty. The two methods are compared using a hierarchy of test cases, starting from a controlled field experiment up to real world operational cases. Both methods give acceptable results with regard to the source location, and the variational method gives a more accurate estimation of the emitted masses. Main sources of uncertainty are discussed.