Data assimilation for dispersion models

Abstract

The design of an effective data assimilation environment for atmospheric chem-bio dispersion models is considered. Filtering techniques are studied for the purpose of data assimilation in the case of dispersion models. Atmospheric dispersion models usually lead to large scale state space models. The linear Kalman filter theory fails to meet the requirements of such applications due to high dimensionality leading to high computational costs, strong non-linearities, non-Gaussian driving disturbances and model parameter uncertainties. Two dispersion models, increasing in complexity, are developed incorporating some of the main characteristics of the actual dispersion models used in the field. Various sampling based filtering techniques are studied for implementation in the case of these models, focusing on Ensemble filtering and Particle filtering approaches. The potential of sampling based techniques is discussed in the context of variable and high state dimensionality, and nonlinear observation models. The performance of various filters is illustrated.