中文摘要：

为有效利用历史资料中的相似信息，减小模式误差对ENSO这类跨季节一年际尺度预测问题的影响提高动力模式的预测水平。作者利用一种基于统计相似的模式误差订正方法，以国家气候中心简化海气耦合模式为平台建立了相应的动力-相似误差订正（DAEC）模式，并着重探讨了系统相似程度（全相似或部分相似）、误差重估周期以及相似样本个数等因素对预报效果的影响。结果表明，利用该方法可以有效地改善原有模式的预报性能，其中“全相似”比“部分相似”更能反映海气耦合系统的相似程度，从而对模式误差做出更为准确的估计，使预报误差明显减小。海洋和大气的误差重估周期对结果也有较大影响，在不同相似程度下分别存在着某种最优配置使得预报效果达到最佳。另外，在对相似样本存在状况及影响的研究中则发现在当前资料长度内整体上只存在着有限个相似样本，在此范围内随着样本取样数目的增加DAEC模式的预报性能逐渐提高。

英文摘要：

To further reduce the impact of model error on the short term climate prediction, on the basis of an analogue correction method of errors, which utilizes the analog information from the historical datasets to estimate the evolution of model errors, a dynamical-analogue error correction model for ENSO prediction based on NCCo intermediate ocean-atmosphere coupled model has been developed. The difference between this model and the NCCo model is only that an error correction sub-model is added in the ocean and atmosphere part respectively. The impact of some basic model parameters as mentioned follows on prediction results are investigated to get the optimal parameters choices： firstly, the effect of analogue degree including the part analogue and comprehensive analogue is compared, the results exhibit that in a coupled system the comprehensive analogue is much better than the part analogue for the model in this paper, because the former can really depict the analogue degree between the current initial value and its historical partners, thus leading to a well estimation of model error. Secondly, the investigation on the effect of the re-estimate period of error （RPE） denotes that RPE is also a crucial parameter to this model. Usually, there is an optimal combination between the RPE of atmosphere and ocean model under different analogue degrees to make a good prediction. Furthermore, the results in this paper also display that there are finite analogue samples in the datasets that the authors hold, and the hindcasting skill has a linear response to the analogue sample sizes due mainly to the fact that more analogue samples can supply more error information to the model thus leading to better estimation of model error and more improvement of prediction skill. Based on the above parameter choices, the initial verification in this paper shows that the hindcast skill of this model is better than that of NCCo model for the SST prediction of the tropical Pacific Ocean, which may imply its potential application to rea