Methods of Data Assimilation

摘要

An overview of ocean forecasting techniques amalgamating numerical models, observations and data assimilation methods is presented. The basics of data assimilation as an application of estimation theory or control theory is described and the corresponding statistical and numerical methods are introduced. Classical approaches like Kaiman filter or optimal interpolation are explained as well as state of the art methods like reduced rank filters and smoother approaches. Problems and challenges of coastal ocean forecasting are identified, which are associated with the specific variables of interest for coastal applications, such as: complex physics complicating the assimilation of data; characteristic time scales; vigorous adjustment process arising in sequential data assimilation, when models are restarted; specific data and observational platforms in coastal ocean and maximising the outcome of synergies between different data types; model and observation error specification; coupling coastal and deep ocean models and seamless transition between coastal and open-ocean scales. Illustrations of some of the above challenges and their treatment in the area of the German Bight are given by describing a pre-operational HF radar data assimilation system using three WERA stations, as well as an assimilation system using FerryBox surface temperate and salinity measurements.