Forecast errors of subsurface temperature and salinity are substantially reduced with an efficient, two-step, multiscale Ensemble Optimal Interpolation (EnOI) system, applied to a near-global eddy-resolving ocean model. A critical element of any data assimilation system is the background error covariance, which for EnOI is typically a static ensemble of anomalies from a long model run. Here, we construct two ensembles — one based on intraseasonal anomalies from a free run of the same eddy-resolving ocean model used to underpin the forecasts, and a second ensemble of climatogical anomalies calculated using a relatively coarse, 1-degree global ocean model. For each assimilation cycle, the coarse-resolution ensemble is used to “correct” the broad-scales, and the high-resolution ensemble is used to “correct” the eddy-scales. Corrections from the coarse steps are more effective at reducing systematic errors in the subsurface ocean whereas the high-resolution steps typically produce vertically coherent corrections associated with mesoscale eddies. We compare two configurations of multiscale data assimilation with different localisation radii in the coarse data assimilation step. The best performance and slowest error growth was found with localisation that was large enough to encompass neighbouring profiles in each assimilation cycle. The efficacy of the approach is demonstrated in ocean reanalyses over 2017-8 that assimilate data every 3 days. We demonstrate clear improvements in the representation of temperature and salinity at all depths around Australia. Model-observation differences are particularly improved in and below the thermocline. The corrections to the ocean state with multiscale data assimilation follow water mass structures. The increased computational cost of this multiscale approach is modest (about double the analysis step), but the performance improvement is significant, making this approach suitable for research and operational applications.

通过应用于近全球涡旋分辨海洋模型的高效、两步、多尺度集合最优插值 (EnOI) 系统，显着降低了地下温度和盐度的预测误差。任何数据同化系统的一个关键要素是背景误差协方差，对于 EnOI 而言，它通常是来自长期模型运行的异常的静态集合。在这里，我们构建了两个集合——一个基于用于支持预测的相同涡旋解析海洋模型的自由运行的季节内异常，以及使用相对粗糙的 1 度全球海洋模型计算的第二个气候异常集合。对于每个同化循环，粗分辨率系综用于“校正”宽尺度，而高分辨率系综用于“校正”涡度尺度。粗步骤的校正在减少地下海洋的系统误差方面更有效，而高分辨率步骤通常会产生与中尺度涡旋相关的垂直相干校正。我们在粗数据同化步骤中比较了具有不同定位半径的多尺度数据同化的两种配置。最好的性能和最慢的误差增长是在定位足够大以包含每个同化循环中的相邻剖面时发现的。该方法的有效性在 2017-8 年的海洋再分析中得到证明，该分析每 3 天同化一次数据。我们展示了澳大利亚各地所有深度的温度和盐度表示的明显改进。温跃层内部和以下的模型观察差异得到了特别改善。通过多尺度数据同化对海洋状态的修正遵循水团结构。这种多尺度方法增加的计算成本是适中的（大约是分析步骤的两倍），但性能提升是显着的，使得这种方法适用于研究和运营应用。