Application Of Ensemble Kalman Filter To Atmospheric DispersionData Assimilation

摘要

The possibility of performing data assimilation on a Monte Carlo dispersion model with the technique namedensemble Kalman filter was examined in an idealized environment: twin experiments. The experiments were performed underthe assumption that the forecast error arises from the uncertainty contained in the source term estimation. ‘True’ source termwas simulated by adding random colored noise to the first guess value, and the ensemble of the concentration forecast wasproduced by the forecast model by using different sets of source term. With the statistics calculated from the ensemble ofradioactivity concentration forecasts, the ground level dose rates were assimilated to give an analysis result. In order to improvethe efficiency and save memory, the state vector was selected dynamically according to the region and direction of the plume.Attempt for estimating the errors in the source term was also done by augmenting the state vector with the errors. By thismethod, it is possible to update the noise as well as the model state during the analysis. A simulation for a period of 24 hourswas performed in which the ground dose rate measurements were assimilated once an hour. The main conclusion is that it ispossible to improve the forecast of Monte Carlo model by ensemble Kalman filter, and the updated source term follows the‘true’ source term very well.