Data assimilation in meteorology
M. Fisher (ECMWF European Centre for Medium-Range Weather Forecasts)
All current operational data assimilation systems for numerical weather prediction may be regarded as exercises in weakly-nonlinear, Gaussian estimation. In this framework, a sequential approach based on the Kalman filter seems an obvious choice. However, the large dimension of the system precludes the proper treatment and propagation of the covariance matrices, and requires severe approximations to be made.
Four-dimensional variational data assimilation (4D-Var) is currently the most widely used assimilation method for operational weather prediction.
This talk will address the relationship between 4D-Var and the Kalman filter. It will be demonstrated that 4D-Var asymptotes to the solution of the full Kalman filter as the length of the assimilation window increases.