Figure 2 shows the monthly mean sea surface salinity (SSS) in the 3rd and 5th months of each model period for the first member of the SP active ensemble. The SSS field varies greatly across years because of the freshwater input, the variations in wind pattern as well as the position of the LC and the Rings (Morey et al., 2003; Schiller et al., 2011; Schiller and Kourafalou, 2014; da Silva and Castelao, 2018). Winds in the GoM are typically southereasterly in summer and northerly and northeasterly during fall and winter (Morey et al., 2003) and drive the seasonal circulation over the shallow inner shelf. In the multi-year average, in February –April low-salinity waters are confined to the Louisiana-Texas (LATEX) continental shelf, and extend offshore in summer (June to August), with a major branch of freshwater spreading to the east of the Mississippi Delta, along the coast of the Mississippi-Alabama-Florida (MAFLA) and toward the West Florida shelf. The shape and extension of this branch are linked to the position of the LC. In fall the fresh SSS retracts toward the shelf, with remnants around the LCS, and in winter (November to January) only the shelf areas are occupied by low salinity waters near the surface.

In the period considered, low-salinity waters (identified by salinity less than 34 psu) were mainly limited to the coastal areas in August 2014 because the discharge in April-July was anomalously low, but extended well offshore to as far as 26°N in 2015, when the late spring/summer discharge was highest among the 3 years (Figure 1B) and the LC was in an extended position for most of the time.

During winter and spring, the El Niño Southern Oscillation (ENSO) is an important driver of interannual variability in salinity and coastal circulation (Gomez et al., 2019). ENSO indeed modulates the global precipitation patterns over the continental United States that feed the Mississippi River basin and the wind anomalies that control the alongshore currents over the LATEX shelf via the thermal wind relation (Childers et al., 1990; Schimidt et al., 2001). This modulation results in flooding conditions and wider than normal offshore spreading of low-salinity waters in winter and early spring during El Niño events, as verified in 2016 from February to April.

As mentioned, the GoM circulation in the upper 600–800 m of the water column is dominated by the anticyclonic Loop Current (LC), which is relatively warm and salty, and its detached eddies or Rings (Hamilton, 2009). The LC and the Rings can extend to 200 – 300 km in diameter (Gordon, 1967). Whenever a Ring is formed, and through its westward drift, smaller mesoscale eddies, spinning both cyclonically and anticyclonically, are also generated. Around and within the large mesoscale structures, submesoscale circulations have been observed year-round (Poje et al., 2014; D’Asaro et al., 2018) at horizontal scales from hundreds of meters to few kilometers. They emerge from the conversion of available potential energy from the mixed layer to kinetic energy and are fueled by the density gradients associated with the energetic mesoscale structures (McWilliams, 2008) and the river discharge (Barkan et al., 2017a).

Figure 3 presents the surface salinity gradient field, the surface relative vorticity ζ = ∂v/∂×−∂u/∂y normalized by the Coriolis parameter f, and the near-surface stratification, or the squared Brunt–Väisälä frequency N2=-gρ⁢d⁢ρd⁢z, where ρ is the potential density, in 2015 for one of the SP active runs. Similar patterns are detectd in other cases but with relatively weaker magnitude or less extension for all fields. For relative vorticity, in the offshore portion of the basin – offshore the shallow continental slope – submesoscale circulations in the form of small eddies are most energetic in winter when the mixed layer is deepest and submesoscale fronts prevail in summer, due to the river-induced submesoscale enhancement, as shown in previous studies (Luo et al., 2016; Barkan et al., 2017b). Over the shelf, submesoscale eddies are more commonly observed from late spring to early fall (Horner-Devine et al., 2015; Kobashi and Hetland, 2020). Focusing on stratification, very weakly stratified waters can be seen inside the LC and Rings extending to the shelves in winter, since the nearshore temperature is low throughout the shallow water column and winds strongest at the coast. On the other hand, waters are strongly stratified in summer whenever impacted by the abundant riverine output (Morey et al., 2003).

Patterns of salinity gradients, stratification and vorticity at the mesoscales remain considerably similar across ensembles, as seen in Supplementary Figure 1 for the corresponding quantities in the MR ensembles, and in SBL21 for further comparisons of predictability skills.

Figure 4 shows the monthly averaged spatial patterns of the freshwater spreading area and normalized volume, or thickness in 2015, together with the sea surface height anomaly (SSHa) contours across the different ensembles. Again, the freshwater is identified by 34 psu isohaline. Sea SSHa contours of 0.17, 0.4, and 0.6 m outline major anticyclonic features, and the 0.17 m threshold is assumed to identify the boundary of the LC and LC eddies in the GoM following previous literature (Leben and Born, 1993; Donohue et al., 2016; Hamilton et al., 2016). In winter, westward/southward wind-driven coastal currents carry low salinity water over the LATEX shelf. During the summer, two branches of low salinity water can be present and transported eastward in the northern GoM. The first branch represents the portion of pre-existing waters on the LATEX shelf transported along-shore, which is also present in the no-river case. The second branch originates mainly from the MARS and moves eastward onto the MAFLA shelf, the West Florida shelf or is advected into the Gulf interior by the intrusion of the LC and the Rings. Finally, in August parcels of low salinity waters can be seen trapped in the cyclonic recirculation between the large anticyclonic patters (LC and Rings) and they are more relevant in the SP-active case than in the other ensembles. The trapping of low salinity waters between two mesoscale anticyclonic structures has been pointed by Androulidakis et al. (2019) as a secondary mechanism of freshwater removal by the LC eddies. The SSHa patterns are overall similar among different ensembles. Quantification of the differences is presented in SBL21, where it is found that the LC has a statistically significant lower probability to extend northward of 28°N in the SP integrations with an active river configuration, compared to the other cases.

In terms of salinity, the modeled freshwater spatial distribution is modestly affected by the setup of river forcing. This is the case although Morey et al. (2003) have demonstrated that the variability in river discharge does not control, but only modulate, the seasonal salinity patterns in the GoM, with the LCS being a player at least as important. In our simulations, both the spreading area and volume are small in the no-river case, as to be expected, and maximum values are achieved in the active river simulations. The role of small differences in LCS are most evident in comparing the SP passive and active cases and their salinity distribution in August 2015. The extension of the LC, which is slightly different among the two cases, modifies the pathway of the Mississippi River water pushing it further east in the passive case.

To quantify the different behaviors in the transport pathways and transport efficiency, we calculated the time-series of freshwater spreading area and volume, shown in Figure 5. We excluded from the calculation coastal areas where the water column depth is less than 100 m given that differences could be driven by river configuration choices in the immediate surrounding of the river mouths. This arbitrary threshold excludes all areas impacted by the modified precipitation fluxes where waters may stagnate in conditions of low winds. The Mississippi River mouth represents the exception, but its freshwater is rapidly transported offshore or to the west in both passive and active cases. We varied the threshold and tested values between 50 and 200 m and results do not change qualitatively.

Interannual changes are generally larger than the differences among different ensembles. As a reminder, for each 5-months long simulation period, the integrations share the atmospheric forcing and start from similar initial conditions (1-day apart). The freshwater area coverage tends to be low in winter, except for 2016, reaches ∼2–3 × 10¹¹ m² during summer, and drops to less than 10¹¹ m² in November of 2014 and 2015. The freshwater volume follows an analogous evolution with comparable seasonality. The largest area and volume of low-salinity waters occur in the summer of 2015. Moreover, relatively lower freshwater coverage but comparable to 2015 freshwater volume is found in August 2016, implying that low salinity waters reached deeper than in other summers, because the larger fluxes occurred earlier in the year (Figure 1), when the MLD was deeper. In the 2016 summer the introduction of freshwater in the basin due to extreme precipitation events, which is illustrated by the increasing freshwater area and volume in the no-river case, is also a noticeable occurrence.

Differences among ensembles are small. The MR and SP runs have similar LCS representations and identical river forcing, and the freshwater spreading evolves similarly over time. In the active river configuration the velocity flux associated with the freshwater input contributes to a slightly greater spread compared to the passive set-up. Differences are larger whenever the freshwater discharge is high, as verified during winter/spring 2016 and more generally in the summer season.

The freshwater spreading is a near-surface process strongly affected by the upper-layer circulation dynamics. During late spring and summer the submesoscale circulations tend to further aid spreading in the SP simulations. In these seasons the prevailing submesoscale features are in the form of density fronts with horizontal advective tendency induced by confluence and convergence (Wang et al., 2021) that contribute to the transport of the riverine water. Indeed, in summer the discrepancy between SP and MR simulations is consistent among years. The lateral convergence zones in the submesoscale frontal structures that abound around the LCS (Zhong et al., 2012; D’Asaro et al., 2018) accumulate and distribute freshwater, contributing to the wider extension of riverine inputs. The summer submesoscale contribution is highlighted comparing the instantaneous distributions of surface salinity and frontogenetic tendency, F, defined as F=D⁢|∇h⁡ρ|D⁢t=Q⋅∇h⁡ρ where DD⁢t is the total derivative and represents the rate of change (D⁢aD⁢t=∂⁡a∂⁡t+u⁢∂⁡a∂⁡x+v⁢∂⁡a∂⁡y) and Q=(Q⁢1,Q⁢2)=-(∂⁡u∂⁡x⁢∂⁡ρ∂⁡x+∂⁡v∂⁡x⁢∂⁡ρ∂⁡y,∂⁡u∂⁡y⁢∂⁡ρ∂⁡x+∂⁡v∂⁡y⁢∂⁡ρ∂⁡y) (Hoskins, 1982; Capet et al., 2008) for a day in August 2015 (Figure 6). F, which is two orders of magnitude greater in the SP-active case compared to the MR-passive one, follows closely the freshwater pathways in all cases.

From late fall to early spring, on the other hand, the mixed-layer is deeper and offshore the continental shelf fronts and filaments can be more easily fragmented by submesoscale instabilities that extract energy from the mixed-layer and generate eddies that restratify the upper water column (Luo et al., 2016). The time evolution of F over waters deeper than 100 m is shown in Figure 7, where is also evident that within each given ensemble the internal variability is small even at the submesoscales. Given the conbined annual cycle of F and mixed-layer depth, the prevailing submesoscale circulations converge material at their interior (D’Asaro et al., 2018), stretching the freshwater patches in filaments less effectively than in summer. Furthermore, in these seasons low-salinity waters flow mostly along the LATEX shelve, due to the prevailing northeasterly and northerly winds and more limited river discharge. The submesoscale eddies and fragmented fronts that occupy waters with a depth greater than ∼ 100 m, have negligible impact on the freshwater spread in “normal” winters (2014 and 2015). On the other hand, in 2016, when the river runoff was very abundant and was pushed offshore by the anomalous winds, the submesoscale circulations cause a slight reduction in volume spread compared to the MR case by limiting the mixed-layer extention (Barkan et al., 2017b).

We next analyze the freshwater transport, T, highlighting the riverine water pathways. We do so by adopting the methodology introduced by Schiller et al. (2011) and applied to the GoM in subsequent works (Schiller and Kourafalou, 2014; Androulidakis et al., 2019; Brokaw et al., 2019). The freshwater transport is defined asT = V×fw×dx×dz where V is the horizontal velocity with its zonal or meridional components, dx is the corresponding space distance, dz is the vertical thickness, and fw is the freshwater anomaly calculated as fw = (S_b − S) / S_b. Here S_b is the background salinity field without the added river discharge (i.e., the mean salinity in no-river cases in this work) and S is the salinity field in each of the active or passive runs without any threshold applied. Since V and fw can be both positive and negative, positive zonal transport can be interpreted as either freshwater moving to the east (both positive) or saltier water moving westward (both negative). In the following we modify fw by setting all the negative values to zero in order to select the transport of freshwater. The difference between setting the negative fw to zero and considering freshwater anomalies of both signs is shown in Supplementary Figure 2, while Figure 2 and Supplementary Figure 3 display monthly salinity fields in one of the SP no-river and SP-active members, respectively. We stress that the mesoscale patterns are similar, but necessarily not identical among the runs (see also SBL21), and that the SP no-river ensemble includes contributions from precipitation anomalies in its freshwater budget.

The spatial patterns of freshwater transport integrated over the upper 100 m of the water column in the last month of the 5-month long runs in 2015 are shown in Figure 8 for the first member of the SP active-river ensemble. Positive values correspond to eastward and northward transport. Anticyclonic (clockwise) freshwater transport greater than 100 m³/s is detected at the periphery of large anticyclones in summer, especially near the north, east and south boundaries of the LC eddy (Figure 8, transport vector). A similar pattern is also observed in winter but with reduced transport associated with it. A small cyclonic eddy (Figure 3) with opposite transport can indeed be seen at about 94°W, 25°N during February.

Previous studies on the offshore pathways of freshwater transport in the GoM have focused on the interactions between the LCS and MARS in summer (Schiller and Kourafalou, 2014; Androulidakis et al., 2019; Brokaw et al., 2019), following the peak of riverine discharge. Brokaw et al. (2019) identified three main pathways in summer by analyzing the surface salinity transport from 2010 to 2018, two of which are captured here and are realized by the relative positions of the LC and LC eddies with respect to the river plume. According to the first pathway, the freshwater is advected eastward from the MAFLA shelf, then southward and finally exits the GoM from the Florida Straits. The second pathway is characterized by narrow freshwater bands surrounding the LC eddies or the LC extension. This recirculation pattern advects the low-salinity waters offshore to the south, then westward and back north toward its source. The third pathway describes conditions when the Rings and the LC do not reach north enough to interact with the river plumes and no detectable interactions between the LCS and the MARS are found. This condition was relevant to portions of 2010 and 2012, when the freshwater remained distributed in the northern part of the basin with decreasing concentrations moving southward and a nearly homogenous gradient, but in the period considered here this third condition is never observed or simulated. On the other hand, a west-to-east offshore transport characterizes the summer of 2015 when two large Rings occupy the central GoM. The smaller cyclonic circulation found at about 93°W and 26°N in August acts as a bridge connecting these two anticyclones and contributes to the lateral transport of nearshore freshwaters from west to east. A similar redistribution pattern is observed also in the other two summers (Supplementary Figures 4, 5).

Our analysis indicates that the influence of the LC system on the offshore transport of freshwater extends to the winter seasons, despite its surface signature being significantly reduced. As mentioned, in winter the surface river discharge is confined to the western coastal regions due to the prevailing northerly and northeasterly winds. The low-salinity waters surrounding the LC and central LC eddies mainly originate from the confluence regions of downcoast inner shelf currents and upcoast outer shelf currents between 26 and 28°N, west of 96°W (Walker et al., 1996; Zavala-Hidalgo, 2003). The offshore transport is sustained by the jet associated with the decaying eddy or a pair of eddies usually found near the coast in this season (Zavala-Hidalgo, 2003), and seen, for example, between 93° and 96°W in Figure 3. This long-distance transport mechanism of nearshore, river-influenced waters is evident in April 2015 (Figure 8) and occurs also in 2014 and 2016 (Supplementary Figures 4, 5). Another important contribution to winter transport comes from summer coastal waters transported offshore that continue to be advected by the mesoscale field.

To expand our analysis to greater depths and fully encompass the GoM deep chlorophyll maximum located at around 100 m (e.g., Pasqueron de Fommervault et al., 2017) and the whole euphotic layer, we extended the computation of the integrated freshwater transport between 100 and 200 m (Figure 9). This calculation is relevant for the primary productivity of the basin being riverine water the main contributor to its nitrogen and phosphorus balance. Below 100 m the freshwater transport follows the boundary of the Rings and LC, with a pattern similar to that of the upper layer but overall decreased magnitude. Significantly enhanced values occur also inside/around cyclonic structures that are part of the LCS, which in August 2015 can be seen at about 94°W and 25°N in February and 93°W and 26°N.

Transport patterns and overall intensity are slightly greater in active runs than in passive configurations. This is quantified in Figure 10, where the probability density function of the transport vector is plotted for the month of August in all 3 years and two SP ensembles. Differences are small but robust, and corroborate the volume differences in low salinity waters quantified in Figure 5. Indeed, if the tails of the distributions are similar once the ensemble variability is accounted for, the core indicates greater transport in the active ensemble in the 50–300 m³/s in all members and all years. In addition, the largest transport rate identified by the flattest slope is detected in 2015 for both active and passive cases, while in 2014, the overall transport rate stays at the lowest level among all 3 years.

Finally, the LCS controls not only the horizontal but also the vertical distribution of freshwater. In the passive configuration, the strength of the subduction around the LC and the Rings is reduced compared to the active runs. This is exemplified in Figure 11, where transects taken at latitudes ranging from 26 to 28°N are shown for June 2015. Here the choice of the month is due to the very strong similarity in the LCS in the two members considered. The contour of 1026 kg/m³ density is used to outline the position of different circulation features. The freshwater can penetrate to depths close to 600 m during summer, approaching the vertical scale of the LCS structures. The most intense vertical penetration is found at the periphery of the Rings. This feature, combined with the fact that meridional currents weaken with depth, contributes to a strong downwelling/subduction of freshwater into the deep ocean. In the northern portion of the basin, where the LC and the Rings do not intrude, the freshwater transport below the mixed layer is modest or null.