Influence of the Caribbean Sea eddy field on Loop Current predictions

Previous studies have shown how the passage of eddies from the Caribbean Sea (CS) to the Gulf of Mexico (GoM) can impact the Loop Current (LC) system, in particular the detachments of LC Eddies (LCEs). Here we used numerical modeling to investigate the impact of the eddy field in the CS on LC predictions. We used a HYCOM ocean model configuration of the North Atlantic at 1/12 degrees resolution to perform two data-assimilative experiments: one in which all available observations were assimilated (Ref), and one in which all available observations were assimilated except in the CS, where climatological altimetry values were assimilated instead of actual observations, leading to dampening the mesoscale activity there (NoCarib). These experiments took place in 2015, when the LC was very active with several LCE detachments, re-attachments, and separations. Each of these experiments was used to initialize 28 60-day forecast simulations every 10 days. In terms of model Sea Surface Height (SSH), the forecasts initialized with the Ref experiment had, on average, lower errors than the forecasts initialized with the NoCarib experiment in the southeastern part of the GoM, with a peak during the 31-40 day forecast period. More importantly, the errors in predicting the date of the next LCE detachment or separation were smaller in the forecasts initialized from the more realistic Ref experiment. Finally, the forecasts initialized by the NoCarib experiment showed a much higher level of false negatives predictions, meaning that no LCE detachment was predicted whereas a detachment actually happened. Overall, 68% of LCE detachments were predicted with an error smaller than 15 days in the forecasts initialized from the more realistic Ref experiment, but only 32% in the forecasts initialized from the NoCarib experiment, stressing the importance of the CS eddy field for predicting the LC evolution. These findings have implications on the GoM predictability, highlighting the need to either run data-assimilative models covering both the GoM and the CS, or pay particular attention to accurate boundary conditions for limited-area GoM models.