A Variational data assimilation scheme and its use of meteorological observations

Sammanfattning: Meteorological data assimilation is the process by which a model simulation of the atmosphere is corrected by observations. In operational weather forecasting, this is done in near real-time, to provide accurate initial conditions for numerical forecasts. The accuracy of the forecasts depends crucially on the accuracy of the data assimilation.The increase in available computing power and the theoretical advances in data assimilation methods have recently enabled the European Centre for Medium-Range Weather Forecasts (ECMWF) to develop and introduce operationally a new data assimilation scheme (4D-Var), based on variational methods. The four papers of this thesis present recent developments that have exploited some of the most promising possibilities of 4D-Var with respect to observation usage. They have all been incorporated in operational practice at ECMWF.Variational data assimilation schemes have several advantages that make the new scheme more flexible than earlier schemes with respect to the use of observations. The introduction of 4D-Var has enabled the effective use of a wider range of meteorological observations than previously possible. This includes, as will be demonstrated in this thesis, both conventional observations and satellite measurements.In the first paper we develop a method for the use of satellite radiance observations, and investigate the aspects of 4D-Var that make the scheme particularly well suited for the assimilation of data from a growing variety of space based observing systems. In the second paper we develop a variational quality control method. The third paper investigates the flow-dependence of the 4D-Var response to time series of surface pressure observations in the vicinity of a rapidly developing storm. The final paper presents a diagnostic method for tuning of the 4D-Var system.