Model error treatment in data assimilation
The talk will describe the general issue of model error treatment in data assimilation for high-dimensional chaotic dynamics, such as those encountered in environmental predictions. A review of the diverse main approaches developed in the last decades along with their storyline, characterizations and major features will be provided. Two families of methods will be described: likelihood-based or innovation-based ones. Focus will then be given to one recent approach, belonging to the latter case, developed by the authors and whose performance will be illustrated numerically. Finally, an alternative method, where the time-uncorrelated hypothesis on model error is relaxed will also be discussed and numerically results will be shown.
Dr. Alberto Carrassi Nansen Environmental and Remote Sensing Center and University of Bergen, Norway