Seasonal to decadal climate forecasting

The atmosphere and the ocean are subject to many dynamical instabilities, which limit the time during which their behaviour can be deterministically forecasted. At longer timescales, the atmosphere can be predicted at best using statistical methods, as a response to external forcing linked to sea- and land- surface anomalies. Climate being defined as the mean of atmospheric states, it appears that it can be predicted up to a few months in advance, which is the characteristic time of the so-called slow components of the climate system. Forecasting in sometimes be extended to longer time ranges, especially when the coupled ocean-atmosphere system exhibits internal variability modes, with characteristic times of a few years. Seasonal climate forecasting is most often based upon Monte-Carlo simulations, where the various realisations correspond to slightly different initial conditions. The present sate-of-the-art Europe (ECMWF) and/or in the USA (M) allows to forecast such major phenomena, as El Nino, up to six months advance. Finally, some parameters may exhibit predictability at still longer time-ranges (inter-annual to decadal), but only certain regions. The example of electricity production is used to underline the potentially large economical benefit of seasonal (C) 2002 Academie des sciences / Editions scientifiques et medicales Elsevier SAS.