Seasonal forecast of tropical climate with coupled ocean–atmosphere general circulation models: on the respective role of the atmosphere and the ocean components in the drift of the surface temperature error
Doblas-Reyes, F. J.
- Publisher: Co-Action Publishing
(issn: 1600-0870, eissn: 0280-6495)
Seasonal forecasting based on coupled general circulation models suffers from important errors. In order to provide insight into the causes of these errors in relation to ocean and atmosphere models, we carried out a comparison using a set of 10-yr ensemble hindcasts of four coupled climate models of the DEMETER project. The four models are based on two different atmosphere models and three different ocean models. This allows us to analyse the relative weight of the ocean and atmosphere components in the error of a coupled model. Using the hindcast climatologies over the years 1991 to 2000, we looked specifically at the sea surface and soil level temperature over the tropics with respect to the hindcast start date. Our results indicate that the monthly evolution of large mean deviations from the observations (> ±1 °C after 6 months) can be decomposed into two terms. One is the first month error, which results from the errors in the initial conditions plus the error introduced by the first month of coupling. It corresponds to the slowly varying component of the error, comparable to an initial shift that persists during the entire coupled experiment. The other term is the remaining time-evolving error, which is fast varying. We show that whereas the slowly varying term is strongly dependent upon the ocean and atmosphere component chosen for the coupling, the atmosphere generally controls the rapidly varying term to first order. The partition appears to be more balanced over some fractions of the Pacific warm pool and the east-equatorial coastal upwelling at certain seasons. These results, specific to seasonal forecasting and probably model dependent, can hardly be interpreted in terms of coupling mechanisms. The weak sensitivity of the mean error to the ocean component could in particular be related to the current limitations of state-of-the-art climate models.