Data assimilation for numerical weather prediction systems : a 25 year perspective

Sammanfattning: This thesis deals with the assimilation of meteorological observations in numerical weather prediction models. A selection of research and development contributions by the author during 25 years is included and discussed in the light of international development efforts up to the present time.The central meteorological data assimilation problem is that the degrees of freedom associated with the state of the atmosphere, as represented in a weather forecast model, is at least one order of magnitude larger than the degrees of freedom of available observational data. To compensate for this lack of observations, various constraints to reduce the degrees of freedom of the model state have to be introduced.Development efforts during the early 1970's were devoted to basic spatial interpolation techniques, used to merge observed information from one point in time with earlier observed information, represented in the state variables of a forecast model. The author contributed to the development of statistical interpolation, a technique that has become a standard data assimilation tool for operational numerical weather prediction. The development was started with simple 2-dimensional schemes and continued with a more complete 3-dimensional scheme including simplified dynamical constraints.Observations from meteorological satellites have become important for operational numerical weather prediction, and the data assimilation techniques developed by the author have proven to be able to handle also these observations efficiently.The Swedish Meteorolgical and Hydrological Institute joined the international High Resolution Limited Area Modelling (HIRLAM) project in 1985 and the author has contributed to the development of the HIRLAM data assimilation. A study on the importance of proper lateral boundary conditions for limited area data assimilation is included in this thesis. More recent development efforts by the author have been devoted to the developement of a new data assimilation system for HIRLAM based on variational techniques. It is shown in a paper, included in this thesis, how high frequency gravity-inertia oscillations can be controlled in 4-dimensional variational data assimilation by means of a weak digital filter constraint. The building of more physical and dynamical knowledge into future data assimilation systems is finally suggested in this thesis. The results of an idealized study in support of this suggestion are presented.