LI Ke, LIANG Manchun, SU Guofeng
Models are needed to quickly predict the atmospheric dispersion of radioactive material released in a nuclear accident. However, the uncertainties in the source term, meteorological data, and other conditions reduce the dispersion model prediction reliability. Data assimilation (DA) is usually introduced to improve the model predictions. The paper presents a DA method based on a Gaussian puff model to improve the predictions using some observed data. The method modifies the puff parameters to approximate the observed data in an iterative search. The four model parameters modified using particle swarm optimization in this study are the release rate, release height, wind direction, and mean wind speed. The method is applicable to mesoscale atmospheric dispersion with uniform and stable conditions over a flat area. Twin experiments are used to verify this DA method. The correlation coefficient between the experimental group and the control group at the observation points is 0.99. The source estimation in the non-steady condition is also tested with the correlation coefficient of 0.68, slightly better than the ensemble Kalman filter method. The method converges rapidly with good model predictions; thus, this method is useful for data assimilation of atmospheric dispersion.