Potential predictability of European winters from the analysis of seasonalsimulations with an AGCM

The potential predictability of European winters on the seasonal scale is i nvestigated with the cycle 5.3 version of the Laboratoire de Meteorologie D ynamique general circulation model by analyzing the link between atmospheri c low-frequency variability and oceanic temperature prescribed as boundary conditions. The word "potential'' refers to the assumption that the SST is a priori known in the experiments, and to the use of a model to evaluate th e real climate predictability. Eleven simulations of the 1971-92 winters ha ve been performed with the model in SST-forced mode. The methodology used i dentifies atmospheric clusters by Ward clustering scheme, and atmospheric v ariability modes over Europe by matrix analysis of relationships between va riables. Tropical Pacific surface temperature fluctuations play a prevailin g role in the modulation of European variability: the model preferentially simulates negative phases of the North Atlantic Oscillation during El Nino episodes, and a high pressure pattern in western Europe during La Nina ones . These two situations are associated with modulations in the structure of the North Atlantic jet and of the North Atlantic storm track, in agreement with data analyses synthesized in the literature. They confirm the prevaili ng role of interactions between different scales of the flow in the mainten ance of persistent anomalies in the North Atlantic/European area. The stron g link simulated by the model between the Pacific-North American oscillatio n and the North Atlantic Oscillation plays an important role in the propaga tion of the impact of the forcing from the tropical Pacific to the North At lantic. For some winters (1971, 1984, 1989, and 1992, the number of simulations has been increased to 30. The normality of the simulated 1984 winter suggests a weak role of the tropical Atlantic in specifying climate anomalies in Eur ope. The differences in strength of the European response between the 1971 and 1989 La Nina events are linked to differences in the Pacific/North Amer ican area. A stronger spread is found in the El Nino case (1992 winter) tha n in the two La Nina cases. The sensitivity of the response to the number o f realizations demonstrates that one has to reach about 15 simulations to o btain a significant response over Europe.