Evaluating surface-only data assimilation for subsurface state estimation in the Gulf Stream

The Gulf Stream and its associated eddies play a vital role in the transport of energy, momentum, and biogeochemical tracers across the northwest Atlantic [1] as well as regulating the European Climate [2]. In modeling investigations of the Gulf Stream, submesoscale-permitting resolutions (up to 1/50°, i.e., kilometer-scale), are reported to increase realism in current separation, path, and vertical penetration [3] when compared to mesoscale-resolving resolutions. However, running long-term basin-scale simulations at sub-mesoscale-permitting resolutions remains impractical with current computational resources. Here, data assimilation is a valuable tool to help bridge this resolution gap by using observations to constrain ocean models. In large-scale western boundary current regions like the Gulf Stream, the most widely available data are satellite observations, which provide information only at the sea surface. While assimilating these surface-only datasets can directly correct surface features (e.g., front location and intensity), how to best use this data to constrain subsurface conditions remains an area of active research for both physics [4] and biogeochemistry [5].
This study evaluates the extent to which the assimilation of sea-surface observations alone can accurately reconstruct subsurface frontal structures and vertical profiles. Specifically, we investigate the degree of fidelity with which surface-only data can constrain the seasonal subsurface temperature and density gradients, originally resolved in a submesoscale-permitting simulation (1/50°), when assimilated into a coarser, mesoscale-resolving configuration (1/16°) and 59 vertical levels. Data assimilation is performed using the DART system [6] in an ensemble Kalman filter framework. The DART system is coupled with MITgcm [7], which is configured with hydrostatic primitive equations at a horizontal resolution of 1/16°. Starting in January 2017, the model is integrated with hourly ERA5 atmospheric forcing, with initial and lateral boundary conditions derived from GLORYS12V1 reanalysis. Pseudo-observations of sea level anomaly (SLA), sea surface temperature (SST), and sea surface salinity (SSS) are sampled from a reference dataset. We evaluate three assimilation scenarios (SLA-only, SLA+SST, and SLA+SST+SSS), including vertical profiles of velocity and potential temperature, depth of the mixed layer, and surface biases in SSH/SLA and SST. Results are compared with a submesoscale-permitting simulation (1/50°) and the GLORYS12V1 reanalysis. The results from this study will provide a benchmark for more advanced data-assimilation techniques, including techniques utilizing machine learning algorithms.

References
[1] D. Kang et al. (2016), Journal of Physical Oceanography, 16(4), 1189–1207.
[2] J. B. Palter (2015), Annual review of marine science, 7(1), 113–137.
[3] E. P. Chassignet and X. Xu (2017), Journal of Physical Oceanography, 47, 1999–2021.
[4] Z. Chen et al. (2022), Frontiers in Marine Science, 9.
[5] B. Wang et al. (2021), Ocean Science, 17(4), 1141–1156.
[6] J. L. Anderson et al. (2009), Bulletin of the American Meteorological Society, 90(9), 1283–1296, 2009.
[7] J. Marshall et al. (1997), Journal of Geophysical Research: Oceans, 102(C3), 5733–5752.