The simulation analyzed in this study (referred to as GOLFO12-PISCES) has been described and compared with observations in Damien et al. (2018). It relies on a physical–biogeochemical coupled model based on the ocean model NEMO (Nucleus for European Modeling of the Ocean, version 3.6; Madec, 2016) and the biogeochemical model PISCES (Pelagic Interaction Scheme for Carbon and Ecosystem Studies; Aumont and Bopp, 2006; Aumont et al., 2015). The model grid covers the GoM and the western part of the Cayman Sea (Fig. 1) with a 1/12∘ horizontal resolution (∼ 8.4 km). This allows us to resolve scales related to the first baroclinic mode, which is of the order of 30–40 km in the GoM open waters (e.g., Chelton et al., 1998). The model is forced with realistic open-boundary conditions from the MERCATOR reanalysis GLORYS2V3, high-frequency atmospheric forcing based on an ECMWF ERA-Interim reanalysis (Brodeau et al., 2010), and freshwater and nutrient-rich discharges from rivers (Dai and Trenberth, 2002). The open-boundary conditions of biogeochemical tracers are prescribed from the World Ocean Atlas observation database (Garcia et al., 2010) for NO3, O2, Si, and PO4 and from the global configuration ORCA2 (Aumont and Bopp, 2006) for dissolved inorganic carbon (DIC), dissolved organic carbon (DOC), Alkalinity, and Fe. The other state variables are forced with very small constant values. The analysis has been performed using 5 d averaged outputs for a period of 5 years from 2002 to 2007. We refer the reader to Damien et al. (2018) for extended model and numerical setup descriptions. In this previous study, an extensive validation of the modeled properties were carried out, focusing on physical properties that are known to influence primary production and chlorophyll concentration: the mixed layer depth and the depth and slope of the nutricline. A novel aspect was to use in situ observations collected from autonomous floats and published in Green et al. (2014) and Pasqueron de Fommervault et al. (2017) to validate not only the modeled surface chlorophyll concentration but also the chlorophyll vertical profile in the GoM. Starting from the parameters suitable for global simulations (Aumont et al., 2015), a large tuning of the biogeochemical model was carried out to reproduce the vertical profile of chlorophyll correctly. The ability of GOLFO12-PISCES to reproduce the main observed features of the GoM was demonstrated, at least at a basin and seasonal scale.

Satellite observations are used to evaluate the ability of GOLFO12-PISCES to reproduce the dynamical and biological signatures associated with LCEs. Surface geostrophic velocities are derived from a 1/4∘ multi-satellite merged product of absolute dynamic topography (ADT) provided by AVISO+ (http://marine.copernicus.eu, last access: 10 April 2021). Surface chlorophyll concentrations are from the Aqua-MODIS 4 km product (Sathyendranath et al., 2012; http://marine.copernicus.eu, last access: 10 April 2021) and consist of 8 d composites from 2003 to 2015.

Figure 1 shows the 8 d averaged satellite observations of the surface chlorophyll around 29 May 2003 (panel a) and 19 October 2004 (panel b). These observations highlight the strong contrast between the eutrophic conditions in the coastal waters and the oligotrophic conditions in the open ocean, as already addressed by several studies (Martinez-Lopez and Zavala-Hidalgo, 2009; Pasqueron de Fommervault et al., 2017). Far from the coast, these figures also reveal that the surface chlorophyll varies at a scale of the order of 100 km with a distribution that tends to follow the absolute dynamic topography (ADT) contours.

LCE trajectories are reported in Fig. 2a, superimposed onto the geostrophic climatological eddy kinetic energy (EKE) field at the surface. EKE is computed from eddy velocities defined on each grid cell as the difference between the total horizontal current and its mean value over 120 d. This time window is chosen to filter the seasonal signal. EKE is concentrated in the Loop Current (LC) and on the westward pathway of the LCEs (Lipphardt et al., 2008) demonstrating that LCEs constitute the major source of EKE in the GoM open waters (Sheinbaum et al., 2016; Sturges and Leben, 2000; Hamilton, 2007; Jouanno et al., 2016).

LCE annual composites of surface geostrophic velocities (Fig. 2c) and [Chl]surf (Fig. 2d) are built from 482 different satellite images. On average, we found that RLCE is ∼ 120 km and Vmax ∼ 0.6–0.7 m s-1, in agreement with previously reported LCEs (Elliot, 1982; Cooper et al., 1990; Forristal et al., 1992; Glenn and Ebbesmeyer, 1993; Weisberg and Liu, 2017; Tenreiro et al., 2018). LCEs are associated with a negative [Chl]surf anomaly (∼ -0.07 mg m-3 in the annual average). The LCEs' influence on [Chl]surf is largest in summer (Fig. 3a) when it reaches very low values (< 0.045 mg m-3), which correspond to an anomaly of ∼ -0.08 mg m-3. This anomaly is less remarkable in winter (∼ -0.06 mg m-3; Fig. 3b) when [Chl]surf is ∼ 0.17 mg m-3 within LCEs. The high chlorophyll concentrations in the northern part of the composites (in the southern part too but in smaller proportions) are related to shelves.

A total of 11 model LCEs were detected during the 5 years of simulation. Their trajectories are reported in Fig. 2b, superimposed upon the climatological EKE field simulated at 10 m. The westward–southwestward propagation of LCEs is well reproduced (Vukovich, 2007) even though the LCE translation is almost westward in GOLFO12-PISCES. A comparison with Fig. 2a shows the ability of GOLFO12-PISCES to represent the mean and transient dynamical features of the GoM open waters (also see Garcia-Jove et al., 2016).

The robustness of the composite method arises from the number of LCEs used to build the composites:

Annual composite is built from 605 5 d averaged LCE model outputs from 10 different LCEs.

Orbital velocities of composite eddies are used to distinguish different dynamical areas within LCEs. The model annual average dynamical profile at 25 m depth (Fig. 4) reveals a typical vortex-like structure with RLCE∼107 km and Vmax∼0.53 m s-1 and suggests the following decomposition:

r<50 km is the LCE core where the eddy is approximately in solid body rotation: Vorb=a⋅r, where the coefficient a is related to the Rossby number (Ro=2a/f). The ratio a/f is estimated to be ∼-0.12 (Fig. 4). In this field, the strain is reduced to a minimum and the flow is dominated by rotation.

The composite hydrological structure of modeled LCEs is shown in Fig. 5b and c. The depression of isopycnals, associated with a depression of isotherms and isohalines, is characteristic of oceanic anticyclones. In the core of the eddies, the composite depicts a salinity maximum located between 100 and 300 m, corresponding to the signature of the Atlantic Subtropical Underwater (ASTUW) of Caribbean origin entering the GoM through the Yucatán Channel (Badan et al., 2005; Hernandez-Guerra and Joyce, 2000; Wuust, 1964). This salinity maximum is not limited to the core of the LCE but gradually erodes and shallows: 36.82 psu at 200 m in the LCE core and 36.61 psu at 150 m in the background GoM common water. Details on the fate of this salinity maximum investigated with GOLFO12 simulations can be found in Sosa-Gutiérrez et al. (2020). The ASTUW layer (salinity > 36.5 psu) is also thicker in the LCE core (∼ 190 m thick) compared to the background GoM water (∼ 120 m thick). Overall, GOLFO12-PISCES reproduces the observed hydrological structure of LCEs (Elliott, 1982; LeHenaff et al., 2012; Hamilton et al., 2018; Meunier et al., 2018b).

The annually averaged LCE composite presents a lens-shaped structure exhibiting a ∼ 50 m thick layer of weakly stratified waters located between 50 and 100 m depth (Fig. 5d). This subsurface modal water presents hydrological characteristics close to the observed background GoM waters (potential temperature ∼ 25.4 ∘C and salinity ∼ 36.3 psu; Meunier et al., 2018b) and is surrounded below and above by well-stratified layers (Meunier et al., 2018a). The upper pycnocline varies seasonally and vanishes in winter due to the deepening of the mixed layer, whereas the lower pycnocline is permanent.

The downward displacement of isopycnals is accompanied by a depletion of nutrients in the upper layer of the LCE core (Fig. 5e). This is a typical feature of mesoscale anticyclones in the ocean (McGillicuddy et al., 1998; Oschlies and Garcon, 1998). The 1 mmol m-3 iso-nitrate concentration (hereafter ZNO3, sometimes referred to as the nitracline as in Cullen and Eppley, 1981, Pasqueron de Fommervault et al., 2017, and Damien et al., 2018) is located at ∼ 70 m depth in the background GoM waters, whereas it is found much deeper in the core (ZNO3∼106 m). At depth, iso-nitrate layers and isopycnals are well correlated (Ascani et al., 2013; Omand and Mahadevan, 2015). For instance, iso-nitrate concentration of 15 mmol m-3 follows the displacements of the 1026.5 kg m-3 isopycnal. However, above 150 m, the density/nitrate relation is different inside and outside the eddies (ZNO3 is collocated with isopycnal 1024.4 kg m-3 in the LCE core and with isopycnal 1024.9 kg m-3 in the background GoM).

The large difference in stratification between the LCE core and background GoM suggests a contrasted seasonal response of the [Chl]. This is evidenced by the analysis of summer and winter composites of [Chl] vertical distribution.

In summer (Fig. 6a), [Chl]surf is ∼ 30 % lower in the LCE core (r<50 km) than in the background GoM (200 km < r<330 km). A pronounced DCM, characteristic of oligotrophic environments, is deeper in the core (∼ 97 m) than in the background GoM (∼ 69 m) with chlorophyll concentrations significantly lower in the interior (∼-25 %).

In winter, the [Chl] is maximum at the surface in all the composite domains (Fig. 6b). [Chl]surf is lower in the LCE core compared to the background GoM, but the difference is less marked (∼-6 %) than in summer. The main discrepancy is the depth of the inflection point of these profiles. It is deeper in the LCE core (∼-150 m), resulting in a more homogenized [Chl] over a deeper layer than in the background GoM (∼-120 m).

However, despite reduced surface concentration both in winter and summer, the integrated chlorophyll content, [Chl]tot, shows a distinct seasonal pattern compared to the surface (Tables in Fig. 6).

In summer, [Chl]tot is lower in the LCE core (27.58 mg m-2) compared to the background GoM (29.41 mg m-2), and Δ[Chl]tot=-1.83 mg m-2.

In winter, [Chl]tot is higher in the LCE core (44.98 mg m-2) compared to the background GoM (38.03 mg m-2), and Δ[Chl]tot=+6.95 mg m-2.

The winter increase in [Chl]tot is around 29 % in the background GoM, whereas it reaches 63 % in the LCE core, leading to [Chl]tot in the core being larger than [Chl]tot in the background GoM in winter. Meanwhile, [Chl]surf remains lower within the LCE core. The fact that the [Chl] at the surface does not reflect its depth-integrated behavior means that the peculiar variability in [Chl] within LCEs may not be fully captured by ocean color satellite measurements. This is consistent with the observations and modeling results of Pasqueron de Fommervault et al. (2017) and Damien et al. (2018) which addressed the vertical [Chl] distribution in the GoM.

[Chl]tot is strongly shaped by both the seasonal variability and the LCEs. The seasonal composites of [Chl]tot, shown in Fig. 7a, confirm the summer/winter contrast and highlight a monopole structure with a relatively homogeneous distribution of [Chl]tot within the eddy's core. In order to better characterize the spatiotemporal variability in [Chl]tot induced by LCEs, an empirical orthogonal function (EOF) analysis was performed on the normalized [Chl]tot anomaly (Fig. 7b) following the methodology of Dufois et al. (2016). It consists of decomposing the signal into orthogonal modes of variability. Here, we choose to focus on the first two most significant modes which explain 40.2 % and 9.9 % of the variability. Since they both depict a similar monopole structure in the LCE core, they were added up in a mode referred to as EOF 1+2 that is responsible for 50 % of the total [Chl]tot variance within LCEs. The third eigenmode (not shown) accounts for 6.2 % and depicts a dipole structure with opposite polarity located at the east and north of the eddy center. On average, the EOF 1+2 mode is positive in winter (from December to March) and negative the rest of the year (from April to November), with a maximum in December and January and a minimum in September. This justifies, a posteriori, the choice to consider winter and summer LCE composites.

The composite evolution of the LCE [Chl]tot along their westward journey is shown in Fig. 8a and b. It illustrates how the total chlorophyll concentration is preferentially increased in winter within the LCE core as soon as the LCEs are shed from the LC. The winter [Chl]tot within LCEs is much larger (exceeding 1 standard deviation) than the background winter [Chl]tot. In terms of integrated [Chl], the LCE-induced seasonal variability overwhelms the GoM open-water background seasonal variability.

The distinct hydrological and biogeochemical properties associated with the LCE core suggest their ability to trap and transport oceanic properties. This mechanism, known as the eddy trapping (Early et al., 2011; Lehahn et al., 2011; McGillicuddy, 2015; Gaube et al., 2017), is efficient only if the orbital velocities of the vortex are faster than the eddy propagation speed (Flierl, 1981; d'Ovidio et al., 2013). The rotational velocities of the model LCEs are ∼ 0.53 m s-1 and 1 order of magnitude larger than the propagation velocities (∼ 0.046 m s-1 on average). This suggests that LCEs might have a certain ability to trap the water masses present in their core with relatively low exchanges with the exterior.

Salinity is well-suited to investigate water masses trapped within the LCE core during their propagation toward the western GoM (Fig. 8c; Sosa-Gutierez et al., 2020): salinity distribution shows a marked subsurface maximum that is not affected by biogeochemical processes. In the western Caribbean Sea, ASTUW is characterized by high salinity (∼ 36.9 psu on average) and low standard deviation (<0.05 psu). The eastern GoM salinity field reveals that most of the ASTUW crosses the Yucatán Channel within the Loop Current. During the formation of LCEs, a significant part of ASTUW is captured in the LCE core with low alteration of its properties (Figs. 5c and 8c). Within the LCE core, the water mass is transported from the eastern to the western GoM where its salinity decreases from 36.9 to 36.7 psu. Although altered, the ASTUW signature is still clearly detectable in the GoM western boundary. The other part of ASTUW entering the GoM is found in the LCE ring. Compared to the core, the salinity in the ring is on average lower (∼ 36.8 psu in the eastern GoM) and presents a high standard deviation, pointing out that more recent ASTUW co-exists with older ASTUW that yields lower salinity maxima. As LCEs travel westward across the GoM, salinity in the LCE ring decays rapidly to reach values similar to the background GoM values (∼ 36.6 psu). This homogenization mainly arises from vertical mixing and winter mixed layer convection (Sosa-Gutierez et al., 2020). Horizontal intrusions and filamentation may also contribute to this homogenization (Meunier et al., 2020). The composites also suggest that almost no ASTUW enters the GoM apart from the LCEs. The slight increase in the background salinity from the eastern to western GoM is a consequence of the diffusion of salt from the LCEs toward the exterior.

Although LCEs undergo considerable decaying rates, their erosion is particularly strong in the ring, while the core remains better isolated from the surrounding waters (Lehahn et al., 2011; Bracco et al., 2017). Since no significant [Chl]tot seasonal variability is reported in the western Caribbean Sea (Fig. 8), the biogeochemical behavior in the LCE core then has to be driven by local processes with the low influence of the horizontal advective process from the ring or of the Caribbean waters trapped during the LCE formation. Given that the LCE core is also quite homogeneous, the following discussion relies on the analysis of the seasonal cycles of selected parameters averaged within the LCE core.

The LCEs impact the upper ocean stratification (Fig. 5d), the nutricline depth (Fig. 5e), and consequently the nutrient supply to the euphotic layer (McGillicuddy et al., 2015). The relationship between mixed layer deepening and nutrient supply is studied here by comparing the ZNO3 with the MLD (Fig. 9a, b).

In late-spring and summer (from May to September), the water column is stratified (shallow MLD), and the downward displacement of the isopycnals within the LCEs pushes nutrients below the euphotic zone (see also Figs. 5e, 6a): less nutrients are available within the LCE cores for phytoplankton growth, explaining a deeper and less intense DCM. In winter, the convective mixing, fostered both by intense buoyancy losses and strong mechanical energy input at the surface, causes a larger deepening of the mixed layer within the LCE core (∼-125 m; Fig. 9a) compared to the background (∼-85 m; Fig. 9b). This asymmetry is due to a pronounced decrease in the surface and subsurface stratification within the LCE core (Fig. 5d; see also Kouketsu et al., 2012). A quantitative diagnostic of the stratification is given by the columnar buoyancy, ∫0HN2zzdz, which measures the buoyancy loss required to mix the water column to a depth H (Herrmann et al., 2008). Figure 10a reveals significant differences in pre-winter buoyancy between the eddy core and its surroundings. Assuming that the change in buoyancy content is mainly controlled by the buoyancy flux at the surface (see Turner, 1973; Lascaratos and Nittis, 1998), it suggests that mixing the water column down to ∼-210 m depth requires smaller surface buoyancy loss in LCE cores compared to the background GoM (Fig. 10b).

However, the larger winter deepening of the mixed layer within the LCE core is not a sufficient condition to explain a larger nutrient supply. Indeed, it fosters the transport of nutrients from the nitracline toward the mixed layer because both are getting closer. Figure 10c highlights that a smaller buoyancy loss mixes down the water column to greater nutrient concentration levels in the LCE core compared to the LCEs surrounding it. This likely explains the winter increase in surface nitrate concentration within the LCEs (Fig. 9a). In addition, a diagnostic of the different contributions to [NO3] evolution is proposed in Appendix B. It shows the dominant role of vertical advection and diffusion in winter in providing nutrients to the euphotic layer in the LCE core.

So far we have assumed that the surface buoyancy fluxes are identical over the LCE core and the background GoM. However, this is not strictly the case because temperature and/or salinity features in the LCEs and background waters are different (Fig. 5b, c; see also Williams, 1988). The modeled surface buoyancy loss during the winter season is ∼18 % more intense within the LCEs. This difference is substantial and probably mainly driven by additional surface cooling applied to the warm LCE core through air–sea interaction. It contributes to enhance convection within the eddy's core and then nutrient supply toward the surface.

The primary productivity PPtot presents a clear seasonal cycle both in the LCE cores and in the background GoM with lower values in October–November, a sharp increase starting in November, a maximum in February, and a gradual decrease from March to October (Fig. 9c, d). The annual PPtot is slightly lower in the LCE core (∼142.4 mg C m-2 d-1) than in the background GoM (∼ 148.9 mg C m-2 d-1). The amplitude of the seasonal cycle is larger in the LCE core: from April to November, PPtot is on average ∼ 12 % lower in the LCE core, whereas, in winter, PPtot is ∼ 14 % higher and reaches ∼ 243.2 mg C m-2 d-1 in February. Particularly in the LCE core, the PPtot seasonal cycle is tightly correlated with vertical mixing, revealing the important role of mixing in the biogeochemistry. The relatively low standard deviation of the monthly PPtot distribution in the LCE core also supports the idea that the influence of the seasonal variability in the forcing largely overwhelms their interannual and sub-monthly variability (Fig. 9c).

The ratio of the PPNtot and PPRtot provides information about the mechanisms controlling the biomass growth (Fig. 9e, f). In winter, the PPNtot plays a leading role, reaching up to 113–147 mg C m-2 d-1, driven by the winter mixing and induced [NO3] fluxes (see Appendix B). Conversely, the PPRtot is dominant from April to October. During this period, low NO3 resources are available in the euphotic layer, and the ecosystem preferentially uses ammonium to sustain the PPtot. This seasonal pattern is characteristic of oligotrophic environments such as the GoM open waters (Wawrik et al., 2004; Linacre et al., 2015). In winter, changes in PPtot are correlated with the intensity of winter mixing in the LCE core (Fig. 9c) and the background GoM (Fig. 9d). The larger PPNtot in the eddy core is consistent with a larger supply of [NO3] and is evidence that the core of anticyclones can be preferential spots of enhanced biological production.

The pressure exerted by zooplankton grazers varies seasonally (Fig. 9c, d). It shows a similar seasonal cycle in the LCE core and in the background GoM. On average, ∼ 90 % of the total growth is consumed by grazers, reaching the highest impact in March, just one month after the peak season of the PPtot in both areas. In February the difference between the primary production and the grazing rate tends to be larger in the LCE core (GRZtot / PPtot=0.95±0.08) than in the GoM background (GRZtot / PPtot=0.965±0.13; Fig. 9c), leading to an enhanced net primary production. Considering the ecosystem from a “top-down” perspective, the grazing rate also participates then in enhancing [Chl]tot within the LCE core compared to the background.

In summer, the total primary production is higher in the background GoM waters as the regenerated production rate is higher. Since grazing is known to be a major contributor of the recycling loop in the euphotic zone (Sherr and Sherr, 2002), the lower grazing rate inside the LCE during summer (Fig. 9c, d) likely explains this lower regenerated production. In addition, the biogeochemical consumption of nitrate that fosters the production of organic matter occurs in a deeper layer within the LCE core compared to the background GoM (Fig. B1e, f). It is then more likely exported out of the euphotic layer in the form of a settling particle, leading to lower remineralization rates in the upper layers to feed regenerated production. More surprising, the new primary production exhibits similar rates in both regions, although NO3 depletion occurs deeper in the LCE core. In the absence of a strong enough vertical mixing when the mixed layer is shallow, this apparent mismatch requires an additional mechanism, vertical advection, capable of supplying NO3 to the euphotic layer (Sweeney et al., 2003; McGillicuddy et al., 2015).

The model vertical velocity in the LCEs reveals an upward pumping in their core (Fig. 11). The vertical velocity between 100 and 500 m is on average + 0.07 m d-1. This vertical transport is mainly driven by two mechanisms, eddy pumping (Falkowski et al., 1991) and eddy–wind interaction (Dewar and Flierl, 1987), but their relative importance is difficult to quantify (Gaube et al., 2014; McGillicuddy et al., 2015).

The eddy pumping mechanism is related to the decay of the rotational velocities from the moment LCEs are released from the Loop Current. In the LCE core, this decay is considered as moderate since lateral diffusivity is expected to be relatively low (Sect. 4.1). This process may however be considerable in the LCE ring where the erosion rates are important (Meunier at al., 2020).

Eddy–wind interactions are due to mesoscale modulation of the Ekman transport so that they are often qualified as eddy-Ekman pumping (He et al, 2017). Following the observation of an LCE core in quasi-solid body rotation, the horizontal vorticity varies little with the radius resulting in a negligible “non-linear” contribution of the Ekman pumping (McGillicuddy et al., 2008; Gaube et al., 2015). Assuming a small effect of the eddy SST-induced (sea surface temperature) Ekman pumping, the total Ekman pumping simplifies into its “linear” contribution, computed as WE=∇×τρ0f+ζ, where ρ0 is the surface density, f the Coriolis parameter, τ the stress at the sea surface depending on both the wind and ocean currents at the surface (Martin and Richards, 2001, their Eq. 12), and ∇× the curl operator. Considering uniform wind velocities ranging from 4.5 to 7.5 m s-1 (Nowlin and Parker, 1974; Passalacqua et al., 2016) blowing over the LCE, the curl of the stress arises from the anticyclonic surface circulation generated by the eddy. Its manifestation is a persistent horizontal divergence at surface balanced by an upward pumping in the eddy interior (see Martin and Richards, 2001; Gaube et al., 2013, 2014, for further details). With ρ0∼1023 kg m-3 and f∼6.2×10-5 s-1, we estimate WE to range from + 0.06 to 0.13 m d-1, in agreement with the modeled vertical velocity within the core. The eddy-Ekman pumping mechanism could explain a large fraction of the gradual upwelling within the eddy's core (Fig. 11) and may actively contribute to the advective vertical flux of nutrients (see Appendix B). In summer, this mechanism could explain why new primary production rates are similar in the LCE core and the background GoM waters, although the nutrient pool is located much deeper in the LCE core.

The eddy-Ekman pumping persists in the LCE core throughout its lifetime as long as there is a wind stress applied at the surface. During wintertime, we expect that both vertical mixing and eddy-Ekman pumping participate to increase the new primary production. A question then arises about the relative contribution of winter mixing to eddy-Ekman pumping in the LCE core primary production increase in winter. This issue was tackled by He et al. (2017) and Travis et al. (2020) comparing the rate of change in the mixed layer depth with the vertical velocity induced by the eddy-Ekman pumping (Eq. 4 in He et al, 2017). In the GoM, even if the wind shows larger magnitudes in winter, it is also associated with a large variability. As a consequence, the variability in Ekman pumping is also found to be large, and a robust seasonal cycle which would allow us to isolate the Ekman pumping in winter cannot be clearly identified. However, in the LCE core, we estimate the mixed layer to deepen at roughly 0.8 m d-1, which is on average about 1 order of magnitude larger than the higher bound of the estimated pumping mechanism typically occurring in winter in response to stronger wind events. This supports winter mixing as the overwhelming process for the LCE-induced primary production peak in winter.