Systemic underestimation of the mid-deep current in the South China Sea by multiple ocean reanalysis products

In the South China Sea (SCS), the mid-deep current is a vital part of the three-dimensional (3D) circulation, exhibiting critical effects on the water mass renewal, material cycling, and energy balance. However, the paucity of observational data for the currents in the mid-deep layer of seas presents doubts pertaining to the reliability of ocean reanalysis products. In this study, the performances of three mainstream ocean reanalysis products, Ocean Reanalysis System 5 (ORAS5), global ocean reanalysis and simulation (GLORYS), and simple ocean data assimilation (SODA), were evaluated regarding simulating the mid-deep currents in the SCS, using the mid-depth velocity data derived from Argo trajectories and year-long mooring observations. The underestimation of the eddy kinetic energy in this layer by the ocean reanalysis products yielded a weaker intensity and variability of the model velocity field compared to the observed values. The majority (>70 %) of simulated velocities at a depth of 1000 m were lower than the Argo-derived values. With reference to the mooring data, velocity underestimation occurred in 80 % of the study area at a depth >1200 m. Even for the best-performing product, GLORYS, the well-modeled area proportion was <25 %. Further analysis of the seasonal circulation patterns and Luzon Strait volume transport indicated that: 1) all products underestimated the volume transport in the Luzon Strait relative to the results of previous studies; 2) GLORYS demonstrated optimal performance in capturing the eddy structures and seasonal responses; 3) ORAS5 achieved a relatively uniform error distribution at different depths; and 4) SODA produced a weak seasonal response and lacked mesoscale features. Overall, the current study elucidates the modeling of the mid-deep current in the SCS and enhances the data assimilation capabilities of the existing ocean reanalysis products. • Mid-deep South China Sea circulation is evaluated with multi-source data. • Less than a quarter of the area is well-simulated by models. • Reanalysis data greatly underestimates the intensity of circulation.