Data assimilation for dispersion models

The design of an effective data assimilation environment for dispersion models is studied. These models are usxially described by partial differential equations which lead to large scale state space models. The linear Kalman filter theory fails to meet the requirements of this application due to high dimensionality, strong non-linearities, non-Gaussian driving distxirbances and model parameter uncertainties. Application of Kaiman filter to these large scale models is computationally expensive and real time estimation is not possible with the present resources. Various Monte Carlo filtering techniques are studied for implementation in the case of dispersion models, with a particular focus on Ensemble filtering and particle filtering approaches. The filters are compared with the full Kalman filter estimates on a one dimensional spherical diffusion model for illustrative purposes.