A new prototype coupled ocean-atmosphere Ensemble Kalman Filter reanalysis product, the Community Earth System Model using the Data Assimilation Research Testbed (CESM-DART), is studied by comparing its tropical climate variability to other reanalysis products, available observations, and a free-running version of the model. The results reveal that CESM-DART produces fields that are comparable in overall performance with those of four other uncoupled and coupled reanalyses. The clearest signature of improvements in CESM-DART is in the analysis of the Madden-Julian Oscillation (MJO) and other tropical atmospheric waves. MJO energy is enhanced over the free running CESM as well as compared to the other products, suggesting the importance of the surface flux coupling at the ocean-atmosphere interface in organizing convective activity. In addition, high-frequency Kelvin waves in CESM-DART are reduced in amplitude compared to the free-running CESM run and the other products, again supportive of the oceanic coupling playing a role in this improvement. CESM-DART also exhibits a relatively low bias in the mean tropical precipitation field and mean sensible heat flux field. Conclusive evidence of the importance of coupling on data assimilation performance will require additional detailed direct comparisons with identically formulated, uncoupled data assimilation runs.

A reliability budget is used to diagnose potential sources of error (departure from observations) in the CESM-DART product. In areas with sufficient observations, the mean bias in zonal wind was generally very low compared to the spread due to ensemble variance, which did not exhibit patterns associated with Northern Hemisphere jet streams but did have regional enhancement over the Maritime Continent. However, the Residual term was often the largest contributor to the budget, which is problematic, suggesting improper observational error statistics and inadequately represented ensemble variance statistics. The departure and residual exhibit significant seasonal variability, with a strong peak in boreal winter months, indicating the model’s deficiencies during the energetic Northern Hemisphere winter. Ocean temperature contained large error in areas with eddy production indicating inadequate ensemble variance due to poor model resolution. Periods when the the Madden-Julian Oscillation (MJO) was active exhibited lower error, especially in the western equatorial Pacific during MJO phases with reduced convection. In contrast, during MJO phases with enhanced convection in that region, the ensemble variance is increased yet the error is comparable to non-MJO conditions, suggesting a controlling effect of the convection parameterization. Further studies evaluating the impact of the coupled assimilation procedure on the reliability budget will be illuminating.