Towards a dynamically balanced eddy-resolving ocean reanalysis: BRAN3

• An 18-year, eddy-resolving ocean reanalysis (BRAN3) is described and evaluated.
• Errors in BRAN3 are 5–28% smaller than its predecessor – BRAN2.
• The magnitudes of the increments in BRAN3 are 20–50% smaller than BRAN2.
• The improvements in BRAN3 are attributed to improved dynamical balances.

The generation and evolution of eddies in the ocean are largely due to instabilities that are unpredictable, even on short time-scales. As a result, eddy-resolving ocean reanalyses typically use data assimilation to regularly adjust the model state. In this study, we present results from a second-generation eddy-resolving ocean reanalysis that is shown to match both assimilated and with-held observations more closely than its predecessor; but involves much smaller adjustments to the model state at each assimilation. We compare version 2 and 3 of the Bluelink ReANalysis (BRAN) in the Australian region. Overall, the misfits between the model fields in BRAN3 and observations are 5–28% smaller than the misfits for BRAN2. Specifically, we show that for BRAN3 (BRAN2) the sea-level, upper ocean temperature, upper-ocean salinity, and near-surface velocity match observations to within 7.7 cm (9.7 cm), 0.68 °C (0.95 °C), 0.16 psu (0.18 psu), and 20.2 cm/s (21.3 cm/s) respectively. We also show that the increments applied to BRAN3 – the artificial adjustments applied at each assimilation step – are typically 20–50%% smaller than the equivalent adjustments in BRAN2. This leads us to conclude that the performance of BRAN3 is more dynamically consistent than BRAN2, rendering it more suitable for a range of applications, including analysis of ocean variability, extreme events, and process studies.