耦合同化是年代际预测的关键环节之一，基于大气资料的耦合同化具有重要意义，且考虑跨圈层误差协方差的准-强耦合同化也是当今耦合同化要优先发展的前沿方向之一。当前国内外在年代际预测的初始化中大多仅为基于海洋资料的耦合同化，年代际预测中大气的作用国内外鲜有研究和关注，而考虑跨圈层误差协方差的准-强耦合同化正面临着方法和开发难度上的双重挑战。 本文基于降维投影四维变分（DRP-4DVar）的四维集合变分混合同化方法（4DEnVar），在耦合气候系统模式FGOALS-g2中建立了基于全球大气资料的弱耦合同化系统（WCDA）和两个考虑跨海-气误差协方差的准-强耦合同化系统（Quasi-SCDA），可以仅同化全球大气资料（Quasi-SCDA1），也可以同时同化全球大气和全球海表温度资料（Quasi-SCDA2）。基于耦合同化系统得到的最优分析值进行了年代际预测试验，综合评估了其对提高FGOALS-g2气候预测技巧的影响。 WCDA的同化分析与未同化相比，纬向风、经向风、温度、位势高度和比湿的系统偏差和距平均方根误差（RMSE）降低，相关系数升高。风场、温度场、海平面气压场、降水场、近地面气温以及全球海表温度（SST）的气候平均态更接近观测。东亚季风指数、厄尔尼诺-南方涛动（ENSO）、太平洋-北美遥相关型（PNA）、太平洋年代际振荡（PDO）、北大西洋涛动（NAO）和大西洋多年代际振荡（AMO）相关系数更高。利用WCDA产生的更接近观测且与模式相协调的最优初值，完成了多集合的年代际后报和预测试验，着重对东亚夏季风指数、中国东部夏季降水年代际转折和青藏高原中东部夏季降水年际变率后报效果的改进进行机理分析，发现分别与同化过程中北太平洋和北大西洋地区自由发展的海-气相互作用有关。 Quasi-SCDA采用物理上更为合理的两步同化法，通过在大气的控制变量中加入SST，并使得SST这一控制变量仅受到850hPa以下大气资料的影响，全球SST资料仅影响SST和大气的表面气压这两个控制变量的新方法考虑了跨海-气的误差协方差。评估结果表明，同化分析和集合平均的年代际后报和预测中距平RMSE与WCDA相比进一步降低，相关系数进一步升高。降水场、近地面气温和SST的气候态以及ENSO、PNA、PDO、NAO和AMO的变率均表现为Quasi-SCDA1优于WCDA，Quasi-SCDA2优于Quasi-SCDA1。 本文的重要意义在于探究了年代际预测中同化大气资料的重要作用，并提出了一种准-强耦合同化方法的新思路，将为耦合同化的发展和应用提供重要方向。

Coupled data assimilation (CDA) is one of the crucial aspects for decadal predictions. CDA based on atmospheric datasets is of great significance, and quasi-strongly CDA (Quasi-SCDA), which considers the cross-domain error covariances, is also an advanced direction for the development of CDA. The literature indicates that most of the initialization of decadal predictions assimilates only oceanic datasets in CDA, and there are only a few related works that focus on the role of the atmosphere. In addition, Quasi-SCDA, which takes into account cross-domain error covariances, faces several challenges in terms of both methodology and technology. In this dissertation, one weakly CDA (WCDA) scheme that assimilates global atmospheric data and two Quasi-SCDA schemes that consider the cross air-sea error covariances are established based on the Dimension-Reduced Projection Four-dimensional Variational Data Assimilation (DRP-4DVar) method, a kind of 4DEnVar method in the coupled climate system model of FGOALS-g2. One of the Quasi-SCDA schemes assimilates only global atmospheric data (Quasi-SCDA1), and the other scheme assimilates both global atmospheric and global sea surface temperature (SST) data (Quasi-SCDA2). Based on the optimal initial conditions (ICs) obtained from the CDA scheme, a set of decadal prediction experiments are conducted, and the contribution of the CDA scheme to the improvement in the climate prediction skills of FGOALS-g2 is evaluated. Compared with the un-assimilated experiment, the assimilated system bias and anomaly root mean square errors (RMSE) of zonal wind, meridional wind, temperature, geopotential height and specific humidity of WCDA are decreased, and the correlation coefficients of WCDA are increased. The climatological mean states of the assimilated wind field, temperature field, precipitation field, near surface temperature and SST of WCDA are closer to the observations. The correlation coefficients of climate variabilities such as the East Asian monsoon, El Ni?o-Southern Oscillation (ENSO), Pacific/North American teleconnection (PNA), Pacific Decadal Oscillation (PDO), North Atlantic Oscillation (NAO) and Atlantic Multidecadal Oscillation (AMO) of WCDA are also improved. The decadal hindcast experiments started from the optimal ICs generated by WCDA, which are closer to the observations and in harmony with the model are accomplished. The great improvements in the decadal variations in the East Asian summer monsoon index and the related decadal transition of East China summer precipitation anomalies as well as the interannual variations in summer precipitation over the eastern Tibetan Plateau of the hindcast results are selected to analyze the improvement mechanism. These improvements are related to the freely developed air-sea coupling over the North Pacific and North Atlantic during the assimilation procedure, respectively. The two Quasi-SCDA schemes use the new and more physically reasonable two-step assimilation method by adding SST to the atmospheric control variables. In addition, the control variable of SST is affected only by the atmospheric data below 850 hPa, while the global SST data influence only the two control variables of SST and surface pressure of the atmosphere and thus can consider the cross air-sea error covariances in Quasi-SCDA. The assimilated and decadal prediction results of the Quasi-SCDA scheme show that the anomaly RMSE are decreased and the correlation coefficients are increased compared with the assimilated and decadal prediction results of WCDA. The climatological mean states of the assimilated and decadal predictions (i.e., the precipitation field, near surface temperature and SST) and the climate variabilities (i.e., ENSO, PNA, PDO, NAO and AMO) all present the better performances of Quasi-SCDA1 than WCDA, and Quasi-SCDA2 is superior to Quasi-SCDA1. This dissertation emphasizes the importance of assimilating atmospheric datasets in decadal predictions, and a new idea of the Quasi-SCDA scheme is proposed, which will provide an important direction for the development and application of CDA.