Page 149 EXTENDED KALMAN FILTER ANALYSIS OF SHORT-RANGE ATMOSPHERIC DISPERSION OF RADIONUCLIDES Bent Lauritzen and Martin Drews Risø National Laboratory, P.O. Box 49, DK-4000 Roskilde, Denmark, INTRODUCTION The emissions of 41 Ar from the Belgian BR1 nuclear research reactor are assessed by means of a recently proposed Kalman filter method for source term estimation based on off-site gamma measurements. Argon-41 is produced in the air-cooled reactor by neutron capture on atmospheric argon, and during normal operations at 700 kW thermal effect the radioactive isotope is released from the reactor’s 60-meter stack at a rate of approx. 1.5 x 10 11 Bq h −1 . In a recent double tracer experiment, a white aerosol tracer was added to the continuous emissions of 41 Ar. The gamma radiation from the decay of 41 Ar was measured at distances up to 1.5 km from the emission stack while simultaneous measurements of the wind field and the dispersion parameters were performed by means of lidar scanning of the aerosol plume. In addition, direct measurements of the 41 Ar release rate were obtained from permanent monitoring equipment in the emission stack. From a total of approx. 6 hours of measurements, four time series of synchronized one-minute radiation and meteorological data were generated (Lauritzen et al., 2003; Rojas-Palma et al., 2004). KALMAN FILTER ANALYSIS The gamma radiation data is analyzed by an Extended Kalman Filter (EKF) method, recently proposed for the reconstruction of the source term and the characteristics of the atmospheric dispersion during a nuclear accident situation (Drews et al., 2004; 2005). In this method, the time evolution of the parameters describing the radioactive plume, X t , is governed by an autoregressive stochastic system equation while the measurement data are coupled to the state vector by a static, non-linear measurement equation, viz. 1 t t t t X AX w − = + (1) ( ) t t t Y hX v = + (2) where v t and w t are uncorrelated Gaussian white noise terms. A Gaussian plume model, () h ⋅ , is employed in parametrizing the short-range downwind atmospheric dispersion and the radiation field from the plume. In the state space model (1)-(2), the state vector X t consists of the unknown plume parameters including the source term, while the remaining parameters needed to describe the plume are being externally forced. The measurement vector, Y t , in the present study comprises the gamma radiation data and the observed wind direction. The Extended Kalman Filter provides a best estimate of the plume parameters conditioned upon all previously measured data. The performance of the filter is controlled by the embedded parameters of the state space model, in particular the system and measurement error covariances, V[w t , v t ]. These parameters are here determined by a maximum likelihood method based on the measured data, making the analysis essentially free of external parameters. Two different versions of the state space model are applied in analyzing the results of the atmospheric dispersion experiment, cf. Table 1. In the first model, the state vector given by