资源环境科技发展态势分析平台

In this study, a simple ocean data assimilation scheme was applied to initialize a set of seasonal hindcasts, which started from each January, April, July and October during 1982-2010 with six ensemble members. In the scheme, sea surface temperature (SST) was the only observed information used to estimate the ocean initial states. Predictions of SST, 2-m temperature (T2m) and precipitation (Prate) over land were assessed and compared with hindcasts from the North American Multimodel Ensemble (NMME) project which were all based on sophisticated ocean initialization schemes with subsurface observations assimilated. The skill comparison indicated that, for all variables evaluated, the prediction skill by the simple ocean initialization procedure was well within the range of skills from individual NMME models. The result suggests that even though sophisticated initialization schemes have the potential to best capture the seasonal climate predictability, most present-day capabilities of seasonal predictions can also be accomplished by utilizing SST only. Further, significant seasonal dependence of prediction skill was also identified in hindcasts by the simple initialization scheme. Specifically, the ENSO SST predictions were featured by the significant "spring barrier" problem. It is argued that the ENSO prediction skill seasonality could be due to "double dip" contribution from the seasonality of climatic feedbacks in the tropical Pacific. In particular, these feedbacks not only play a role during the forecast leading to the largest errors in spring, but also cause the largest errors in the SST-derived ocean initial conditions in spring. Complementary aspects of seasonal forecasts based on simple initialization scheme, and use of low-resolution models are also discussed.