BAYESIAN TRACKING OF THE TOXIC PLUME SPREADING IN THE EARLY STAGE OF RADIATION ACCIDENT

The article deals with the predictions of time and space evolution of pollution dispersion during the early phase of a hypotetical radiation accident. The goal is to design a proper fast algorithm which could enable more precise online estimation of radioactivity propagation on basis of recursive procedure of Bayesian filtering. Predicted trajectory of the plume of pollutants is refined online according to the values of observations incoming from terrain. The technique should be sufficiently robust to cope an expected lack of information in the same beginning of the event. A certain modification of the particle filter (PF) method is investigated here. Its robustness is illustrated on a real but atypical meteorological situation. Short time meteorological forecast entering the model is for this case in poor correspondence with the real time local meteorological measurements. Radiological measurements are assumed to be coming periodically from the Czech Early Warning Network (EWN). The respective radiological values in the real positions of EWN receptors are generated “artificially” drawing inspiration from the real local meteorological measurements.