Abstract/Summary

In a recent study, a new method for assimilating observations has been proposed and applied to a small size nonlinear model. The assimilation is obtained by confining the analysis increment in the unstable subspace of the Observation-Analysis-Forecast (OAF) cycle system, in order to systematically eliminate the dynamically unstable components, present in the forecast error, which are responsible for error growth. Based on the same ideas, applications to more complex models and different, standard and adaptive, observation networks are in progress. Observing System Simulation Experiments (OSSE), performed with an atmospheric quasi-geostrophic model, with a restricted "land" area where vertical profiles are systematically observed, and a wider "ocean" area where a single supplementary observation is taken at each analysis time, are reviewed. The adaptive observation is assimilated either with the proposed method or, for comparison, with a 3-D VAR scheme. The performance of the dynamic assimilation is very good: a reduction of the error of almost an order of magnitude is obtained in the data void region. The same method is applied to a primitive equation ocean model, where "satellite altimetry" observations are assimilated. In this standard observational configuration, preliminary results show a less spectacular but significant improvement obtained by the introduction of the dynamical assimilation.