Extended-range weather prediction depends in a crucial way on skill at forecasting the duration of a blocking event or other persistent anomaly that is under way at initial forecast time. The ability to forecast the subsequent onset of another persistent anomaly -- after the break of the current one -- has proven even more elusive.

To address this crucial problem, we study the application of advanced data assimilation methods on predicting the transitions between atmospheric weather regimes. Marshall & Molteni's (1993) three-level quasi-geostrophic model in spherical geometry has been shown to have a fairly realistic climatology and exhibit multiple regimes that bear some resemblance to those found in observations. We adopt therefore this model for our study of transitions between such regimes.

First, the model's Northern Hemisphere regimes are identified by a number of independent classification and description methods. Next, the Markov chain of transitions between the regimes so obtained is determined in the model's reduced phase space that is spanned by its leading EOFs. Finally, we adapt NASA Goddard's Phsyical-Space Statistical System (PSAS) data assimilation framework to carry out identical-twin experiments with the model.

The purpose of these experiments is to detect preferential directions of regime transitions in the model's full phase space. Synthetic observations are simulated to correspond to both conventional and satellite networks. Their effects on extended-range prediction in the model are evaluated.