Loop Current Eddy Thor Separation

• The Loop Current Eddy Thor separation event is characterized by the detachment of an anticyclonic eddy driven by coherent deep–surface interactions.
• Forecast models employ NCODA data assimilation and ensemble methods to quantify deep eddy initialization errors affecting surface ring shedding.
• Accurate deep-ocean observations are critical for constraining mesoscale dynamics and improving forecast skill in the Gulf of Mexico.

A Loop Current Eddy Thor separation event refers to the physical and forecasted phenomena involving the partitioning of a large anticyclonic Loop Current Eddy (LCE), specifically “Thor,” from the main Loop Current in the Gulf of Mexico. The process exemplifies the dynamical interactions between the deep and upper ocean, with separation mechanisms and forecast skill critically influenced by the state of deep mesoscale eddies. LCE Thor serves as a benchmark case for contemporary data-assimilative ensemble ocean forecasting systems, where the ability to predict eddy separation is tightly coupled to the representation and initialization of deep cyclonic and anticyclonic features (Cooke et al., 12 Nov 2025).

1. Physical Characterization of Loop Current Eddy Separation

The Loop Current (LC) is a high-velocity, warm-water current that intrudes from the Caribbean into the Gulf of Mexico. It periodically sheds large anticyclonic eddies (LCEs) into the western Gulf—a process known as LCE separation or “ring shedding.” During the “Thor” event, the detachment was governed not simply by surface-layer instabilities, but by vertically coherent interactions between deep and surface fields. Deep mesoscale eddies, diagnosed using the stream function η_ref at approximately 2000 m, play a pivotal role. Anticyclonic and cyclonic features in the deep layer co-evolve with the surface front, enabling the “neck pinch-off” and full ring shedding when their intensities and phasing are favorable (Cooke et al., 12 Nov 2025).

2. Modeling Framework and Ensemble Initialization

Forecasts of the Thor separation event were conducted using a coupled system consisting of the Navy Coastal Ocean Model (NCOM) v4.0 (Δx ≈ Δy = 3 km, 49 hybrid levels) and the COAMPS atmospheric component. Data assimilation was implemented via the Navy Coupled Ocean Data Assimilation (NCODA) 3D-Var framework, which uses a cost function to integrate observational and background states:

$$J(x) = \tfrac{1}{2}(x-x_b)^T B^{-1}(x-x_b) + \tfrac{1}{2}[H(x)-y]^T R^{-1}[H(x)-y]$$

with variables as defined in (Cooke et al., 12 Nov 2025). Perturbations for the 32-member ensemble were generated by the ensemble-transform (ET) method using 24-hour forecast error covariances, explicitly sampling the uncertainty in upper- and deep-ocean states.

3. Data Assimilation and Observational Constraints

The NCODA system assimilates satellite-derived sea-surface height (SSH) and sea-surface temperature (SST), as well as temperature and salinity profiles from the upper 1000 m of the water column. No velocity observations were directly assimilated. Deep-ocean observations—such as those from CPIES (current- and pressure-recording inverted echo sounders)—were reserved solely for verification and were not used to constrain the deep initial state. The deep (>1000 m) field in forecasts is thus controlled primarily by dynamical adjustment from upper-ocean increments and initial perturbations, rather than direct assimilation (Cooke et al., 12 Nov 2025).

4. Deep Eddy Detection and Diagnostic Metrics

Deep mesoscale eddies are diagnosed by local extrema in η_ref, defined as

$$\eta_\text{ref}(x, y) = \frac{P_\text{ref}}{\rho_b g}$$

where $P_\text{ref}$ is the bottom pressure anomaly and $\rho_b$ is the bottom density. Anticyclones correspond to local maxima in η_ref (positive anomaly), cyclones to local minima (negative anomaly). Eddy intensity is quantified by peak $|\eta_\text{ref}|$, and lateral extent by the radius of closed η_ref contours. Root-mean-square error (RMSE) is used as the quantitative metric for forecast skill, calculated for both SSH (surface) and η_ref (deep) at various forecast lead times.

5. Evolution of the Thor Event: Deep–Surface Coupling

The timeline of the Thor separation event is partitioned into phases according to forecast analysis:

Phase	Key Deep Features	Surface Manifestation
Pre-forecast	Cyclone–anticyclone dipole on Mississippi Fan	LC in extended configuration
Weeks 1–4	Homogenized deep fields among ensemble members	Minimal surface RMSE spread
Weeks 5–6	Best members: strong anticyclone, intensifying deep cyclone; worst: weak/misplaced eddies	Divergence in forecast SSH RMSE; onset of ring pinch-off
Weeks 10–12	Deep cyclone locked to bathymetric constraints in best members	Successful LCE detachment in best members; misplacement or connection in worst

Best-performing members (lowest initial deep RMSE) exhibited enhanced skill in forecasting the timing and geolocation of the Thor detached eddy, in contrast to worst-performing members, which showed substantial surface errors (SSH RMSE up to 5 cm vs. 1.5–2 cm in best cases by week 11) and misplaced rings (Cooke et al., 12 Nov 2025).

6. Mechanistic Link: Deep Eddy Modulation of Surface Separation

Deep mesoscale eddies via η_ref maxima (anticyclones) and minima (cyclones) modify the upper ocean by exerting vertical vortex-stretching and constraining isopycnal tilting. This local intensification of the upper baroclinic front, particularly when a deep cyclone forms beneath the LC “neck,” enables a robust and spatially accurate ring separation. Best ensemble members, characterized by realistic deep features, reproduced the observed neck pinching and detachment events, while those with weaker or misaligned deep eddies failed to replicate the ring-shedding chronology and geography (Cooke et al., 12 Nov 2025).

The correlation between initial deep RMSE and subsequent surface SSH forecast RMSE was linear (slope ≈ 1.2 cm SSH per 1 Pa η_ref error, with r ≈ 0.8 at six-week lead), quantitatively linking deep-initial-condition quality to forecast skill during LCE separation epochs.

7. Significance and Implications for Data Assimilation

The Thor event exemplifies the necessity of accurate deep-ocean initialization in ocean forecasting frameworks reliant predominantly on upper-ocean observations. As deep fields influence the timing and topology of surface eddy events on 5–10 week timescales, the assimilation of deep in situ observations (e.g., from CPIES or Argo profiling floats reaching ≥2000 m) is critical for constraining low-mode uncertainty and enhancing LCE separation forecast reliability. The findings motivate future efforts to routinely incorporate deep observational data streams, particularly in strongly baroclinic and topographically intricate regimes like the Gulf of Mexico LC system (Cooke et al., 12 Nov 2025).