年代际时间尺度的气候异常及其预测直接关系国民经济的可持续发展，但由于长时间尺度观测资料的匮乏，造成对影响年代际变率的物理机制的认知非常有限，年代际预测也远没有达到需求。已有研究表明，年代际变率（如AMO、THC、PDO等）主要是由耦合模态引起，具有一定的可预报性，对它的预报强烈依赖于耦合系统的初值。那么意味着初值的好坏一定程度上决定了预报的成败，而耦合模式的模式偏差也会导致模式很难预测出年代际气候转型。因此本研究旨在发展“耦合”初始化（同化）方法为年代际预测提供初值。该方法的创新点是利用大气-海洋耦合样本来构造背景协方差，同化前后保证了大气-海洋变量间的动力协调性，维持了海气交换界面的通量平衡。进一步地，利用该方法将海表风场观测同化入耦合模式中，基于热带大洋海表风场和海流的高相关性，通过同化风场可显著提高模式流场的精度，从而抑制由于流场模拟不足造成的模式偏差，有助于提高年代际预报技巧。

Decadal climate signals, and their predictions are directly related to the national economy sustainable development, but due to the lack of long time-scale observational data, how the physical mechanisms affect the decadal variability is less understood, and the decadal forecast is also not satisfied by the social demand. Previous studies show that the decadal variability (e.g., AMO, THC, PDO) is mainly caused by the coupled air-sea modes, and can be predicted. This also means that the decadal prediction is strongly dependent on the initial condition of the coupled system. Thus the good or bad initial condition can determine the success or failure of the forecast, and the model bias of the coupled model also can make the model is difficult to predict decadal climate transition. Therefore, this study aims to develop a coupled initialization (assimilation) method on providing the initial condition for the decadal prediction. The innovation of the method is the use of the atmosphere-ocean coupled ensemble members to construct the background covariance, which can ensure the dynamically consistent between the atmosphere-ocean variables during the assimilation cycle, and also can maintain the flux balance at the air-sea exchange interface. Further, based on the high correlations between the sea surface wind and the ocean currents over the tropical oceans, the accuracy of the ocean currents can be significantly improved through using the coupled assimilation method on assimilating the sea surface wind observations into the coupled model, and the model biases induced by the inaccurate simulated ocean currents also can be alleviated. All these advancements introduced by the coupled data assimilation method can help to improve the skill of the decadal prediction.