Abstract The performance of two common approaches to data assimilation, an Ensemble Adjustment Kalman Filter (EAKF) and a 4-dimensional variational (4D-Var) method, is quantified in a popular community ocean model, the Regional Ocean Modeling Systems (ROMS). Two distinct circulation environments are considered: the California Current System (CCS), which is an eastern boundary upwelling regime, and the Indian Ocean (IO) characterized by an equatorial waveguide subject to the energetic seasonal reversals of the Indian and Asian Monsoons. In the case of the CCS, experiments were performed using synthetic observations, so-called Observing System Simulation Experiments (OSSEs). An extensive suite of CCS OSSEs were conducted to explore the performance of both data assimilation approaches to system configuration. For the EAKF, this includes the method for generating the seed ensemble, ensemble size, localization scales, and the length of the assimilation window. In the case of 4D-Var, the influence of assimilation window length, and the formulation of the background error covariance were explored. The performance of the EAKF was found to be influenced most by the size of the ensemble and by the method used to generate the initial seed ensemble where centering of the ensemble was found to yield improvement. For 4D-Var, the assimilation window length is by far the most critical factor, with an increase in system performance as the window length is extended. In general, the EAKF and 4D-Var systems converge to similar solutions over time, which are independent of the starting point. The EAKF employs a First-Guess at Appropriate Time (FGAT) strategy, and some experiments indicate that short FGAT windows can be problematic due to the introduction of frequent initialization shocks. While the EAKF generally out-performs 4D-Var in the OSSEs, analysis of the innovations from the two systems through time indicates that they track each other closely. Additional Observing System Experiments (OSEs) were performed in the CCS and IO configurations of ROMS using real ocean observations. In this case, the comparison of the EAKF and 4D-Var state estimates with independent observations indicates that the EAKF and 4D-Var state estimates diverge over time, although the 4D-Var estimates are somewhat better by some measures. The relative performance of the EAKF and 4D-Var systems is similar across the wide-range circulation regimes that characterize the CCS and IO, suggesting that the results presented here are a robust indicator of expected performance in other regions of the world ocean.

中文翻译：

---

集合调整卡尔曼滤波器和 4D-Var 海洋数据同化系统的区域和盆地尺度应用

摘要 两种常用的数据同化方法，即集合调整卡尔曼滤波器 (EAKF) 和 4 维变分 (4D-Var) 方法的性能在流行的社区海洋模型区域海洋建模系统 (ROMS) 中进行了量化。考虑了两种不同的环流环境：加利福尼亚洋流系统 (CCS)，它是东部边界上升流制度，以及印度洋 (IO)，其特征是赤道波导受印度和亚洲季风的高能季节性逆转影响。在 CCS 的情况下，实验是使用合成观测进行的，即所谓的观测系统模拟实验 (OSSE)。进行了一套广泛的 CCS OSSE，以探索两种数据同化方法对系统配置的性能。对于 EAKF，这包括生成种子集合的方法、集合大小、定位尺度和同化窗口的长度。在 4D-Var 的情况下，探讨了同化窗口长度的影响，以及背景误差协方差的公式。发现 EAKF 的性能受集合大小和用于生成初始种子集合的方法的影响最大，其中发现集合的居中会产生改进。对于 4D-Var，同化窗口长度是迄今为止最关键的因素，随着窗口长度的延长，系统性能会提高。一般来说，EAKF 和 4D-Var 系统随着时间的推移会收敛到类似的解决方案，这与起点无关。EAKF 采用在适当时间先猜测 (FGAT) 策略，一些实验表明，由于频繁的初始化冲击的引入，短 FGAT 窗口可能会出现问题。虽然 EAKF 在 OSSE 中的表现通常优于 4D-Var，但随着时间的推移对这两个系统的创新进行分析表明它们彼此密切跟踪。使用真实海洋观测在 ROMS 的 CCS 和 IO 配置中进行了额外的观测系统实验 (OSE)。在这种情况下，EAKF 和 4D-Var 状态估计与独立观察的比较表明 EAKF 和 4D-Var 状态估计随着时间的推移而发散，尽管 4D-Var 估计在某些方面更好一些。EAKF 和 4D-Var 系统的相对性能在表征 CCS 和 IO 的宽范围循环制度中相似，