Statistical adjustment of simulated climate: example of seasonal rainfall of tropical America.

This study presents the skill of the seasonal rainfall of tropical America from an ensemble of three 34-year general circulation model (ECHAM 4) simul ations forced with observed sea surface temperature between 1961 and 1994. The skill gives a first idea of the amount of potential predictability if t he sea surface temperatures are perfectly known some time in advance. We us e statistical post-processing based on the leading modes (extracted from Si ngular Value Decomposition of the covariance matrix between observed and si mulated rainfall fields) to improve the raw skill obtained by simple compar ison between observations and simulations. It is shown that 36-55 % of the observed seasonal variability is explained by the simulations on a regional basis. Skill is greatest for Brazilian Nordeste (March-May), but also for northern South America or the Caribbean basin in June-September or northern Amazonia in September-November for example. (C) 2000 Academie des sciences / Editions scientifiques et medicales Elsevier SAS.