The Coastal and Regional Ocean Community (CROCO) model is an evolution of the three-dimensional ocean ROMS model that uses the Boussinesq approximation and hydrostatic vertical momentum balance to resolve the primitive equations on a rotating frame (Shchepetkin and McWilliams, 2005).

The SouthWest indiAn subtropical Gyre (SWAG12) grid configuration spans a geographical area from 6.8°W to 68.0°E, and from 47.5°S to 6.0°S, with a horizontal resolution of 1/12° and 75 terrain-following vertical levels on new Sigma coordinates (thetaS=5; thetaB=2; Hc=100). Surface forcings are provided by a bulk formulation using the daily ERA-5 atmospheric reanalysis (with a resolution of 1/4°; Hersbach et al., 2020) with a relative wind approach (Renault et al., 2017). PISCES is a biogeochemical module that has been implemented into CROCO with plankton compartments represented by two phytoplankton types (nanophytoplankton and diatoms) and two zooplankton types (microzooplankton and mesozooplankton) [see details in Aumont et al. (2015)].

Monthly outputs from the 1/4° GLORYS2v4 ocean reanalysis were used to force the boundaries of SWAG122 from 1993 to 2016. GLORYS2v4 is the non-assimilated version of GLORYS, including biogeochemical variables from the PISCES model. The initial condition in January of 1993 was also obtained from the 1/4° GLORYS2v4 ocean reanalysis. A simulation was conducted for 27 years (1991-2017), which included a spin-up of seven years (1991-1997). Three-day averages of model parameters are recorded to calculate eddy statistics in the MC.

This coupled model shows good agreement with physical and biogeochemical observations (see Supplementary Material , Supplementary Figures S1 - S7 ). In particular, the eddy kinetic energy is well represented with high values in the eastern part of the channel, along the African coast ( Supplementary Figure S1 ), which is characteristic of the rings distribution (Halo et al., 2014a). The cross-shore gradient of surface chlorophyll-a (Chl-a) is well captured with higher concentration found over the shelf, consistent with satellite observations ( Supplementary Figure S5 ). The maximum Chlorophyll-a concentration is found in the subsurface, representing the DCM, and is largely dominated by nanophytoplankton as observed by Barlow et al. (2014) ( Supplementary Figure S8 ). The south-to-north gradient of physical and biogeochemical properties along 41°E is also consistent with observations ( Supplementary Figures S6 - S8 ). The seasonal evolution of chlorophyll-a ( Supplementary Figures S9 , S13 ) is coherent with literature, with a bloom of nanophytoplankton peaking in August (Lévy et al., 2007; Tew-Kai and Marsac, 2009). This winter boreal bloom is concomitant with higher nutrient concentrations, especially nitrate and phosphate ( Supplementary Figures S9 , S11 ). Note that river inputs are not considered in the coupled model; Thus, the increase in nutrients over the shelf and the southern MC ( Supplementary Figures S11 , S12 ) is likely due to wind-driven cool-water upwelling (Lamont et al., 2014; Malauene et al., 2014) and the deepening of the mixed layer depth (MLD) induced by seasonal wind intensification and negative surface heat flux (Chiswell et al., 2015). Furthermore, there is a strong contrast with the open ocean over the Sofala Bank and Delagoa Bight with low SST and high SSS ( Supplementary Figure S10 ), typical of upwelling conditions driven by eddy interaction with the shelf, thereby enhancing productivity (José et al., 2014; Lamont et al., 2014; Huggett and Kyewalyanga, 2017). Indeed, the Sofala Bank is one of the most productive regions off the Mozambican coast (Sá et al., 2013), as reproduced by the model ( Supplementary Figures S13 , S14 ). Diatom concentrations are, on average, very weak and are confined to the Sofala Bank and Delagoa Bight during the boreal spring and summer, with Chl-a concentrations representing about 1/3 of the total Chl-a ( Supplementary Figure S13 ). Over the Mascarene Basin, the model represents a weak increase in diatoms and nutrient concentrations (nitrate and phosphate, more pronounced in boreal winter, Supplementary Figures S9 – S12 ), coherent with the upwelling of nutrient-rich waters, which increases chlorophyll concentration and secondary production levels downstream (Gallienne and Smythe-Wright, 2005). Consequently, the intrusion of the SEC and the NEMC into the MC is likely to transport different water masses - in terms of biogeochemical composition - depending on the seasonal cycle (Schouten et al., 2003) and to increase the amplitude of the seasonal variation of chlorophyll-a in the Mozambique Channel (Omta et al., 2009).

To quantify the average properties of mesoscale eddies, we detect eddies in the 20-year run using the py-eddy-tracker method (Mason et al., 2014), an open-access tool based on the method of Chelton et al. (2011), which seeks the outermost closed contours of the SSH. Closed contours of the SSH are identified at 2 cm intervals. Non-eddy features are filtered by applying a battery of tests (e.g., shape test, size test, SSH extreme value test – see Supplementary Material , section 4 and Supplementary Figure S15 , for details). The py-eddy tracker has been successfully used to study the eddy properties from numerical simulations (Mason et al., 2017, 2019; Smith and Fortin, 2022) and is used by CMEMS on satellite observations (Mason et al., 2019). This method is robust and exhibits low sensitivity to the parameters involved in the closed contour tests. The eddy detection was performed over the MC, [30-50°E; 10-30°S] on the 20 years of simulation (1998-2017): for every single eddy, its position, radius, and amplitude are recorded. From the eddy detection method, the radius is estimated according to their maximum speed (speed radius) or the closed contour area (effective radius). In this study, we focus our analysis on the speed criteria that presumably delimit the eddy core, i.e., the solid body rotation of the eddy.

Eddies were tracked by comparing the eddy center position at consecutive time steps over the entire run. Cyclone and anticyclone tracks are treated separately. The distances between all identified positions between two-time steps are calculated using the ellipse method of Chelton et al. (2011). If multiple vortices fall within the ellipse, in this case, the vortex is assigned according to the minimum of a set of dimensionless similarity parameters, S (Penven et al., 2005), computed for each t+1 candidate, as follows:

Where Δd is the eddy separation distance at times t and t+1, Δa is the eddy area variation (based on L_spd ), and ΔA is the amplitude variation. Characteristic values for eddy separation distance, area, and amplitude are given by d₀ , a₀ , and A₀ , where d₀ = 25km, a₀ = π60²km², and A₀ = 2cm, respectively. From the eddy trajectories, the eddy lifetime can be estimated. We select long-lived eddies for trajectories longer than 30 days (Chelton et al., 2011; Halo et al., 2014a). From there, the velocity, direction of propagation, and evolution of biogeochemical properties along eddy trajectories can be evaluated.

The eddy detection and tracking analyses were also applied to the AVISO dataset, and the eddy characteristics from the model and the observation are in good agreement ( Supplementary Figures S16 - S19 ). According to the good agreement of both the water properties and the mesoscale activity characteristics, we are confident with the statistical analysis described below.

To assess the vertical and radial structure of the primary eddy characteristics, a composite analysis was conducted regardless of the eddy lifetime. The composite analysis consisted of computing an average eddy over all detected eddies. The computed normalized radius R corresponds to the boundary of the eddy core, the speed radius. To account for the region surrounding the eddy core, the composite is performed over 3R. The R-normalized horizontal coordinates were used for each eddy, and the physical and biogeochemical properties were averaged across all eddies in this normalized coordinate system, where the eddy center was at 0 and the eddy core’s edge was at ±1R. The eddy core is limited to ±1R (the solid body rotation area); The region surrounding the eddy is assumed to be far from the eddy core limit and we arbitrarily choose ±2R as a minimum limit; Between the eddy core limit and the surrounding eddy limit, the region is defined as the edge of the eddy.

To spatially evaluate the local influence of eddies, we calculate Individual eddy property anomalies, by computing the difference between the eddy core ( R ∈ [ − 1 ; + 1 ] ) and the area surrounding the outermost eddy limit ( R ≥ | ( 2.7 ) | ), averaged over 100m depth. The area surrounding can be linked to the non-eddy structure. An eddy anomaly indicates the relative local impact of eddies.

Finally, the north-to-south vertical impact of eddies was diagnosed and compared to the 20-year climatology to better appreciate the spatial impact of eddies according to latitudinal and vertical gradients.

Over the 20-year simulation, we identified 19561 cyclonic eddies and 14769 anticyclonic eddies over the MC (from 13.5°S to 25°S), i.e., 978 cyclonic and 738 anticyclonic observations per year. One-third of identified anticyclonic eddies are rings [using the radius criteria ≥ 100km, as in Halo et al. (2014a)], with a total of 4400 entities detected. Mean radius and speed are 60km and 6.4km/day for cyclonic eddies, 58km and 6.5km/day for anticyclonic eddies (with R<100km), and 127km and 10.5km/day for rings, consistently with prior observations (Halo et al., 2014a; Schouten et al., 2003).

Cyclonic and anticyclonic eddies are heterogeneously distributed in the MC ( Figures 1A, B ) in agreement with the surface eddy kinetic energy map ( Supplementary Material ). Cyclonic eddies are fairly ubiquitous in the channel, but long-lived ones are mainly found in the eastern part of the channel, in agreement with Halo et al. (2014a). Another site of high concentration for long-lasting cyclonic eddies is located in the northwestern region of Madagascar, at the northern entrance of the channel, as mentioned by Halo et al. (2014a). Cyclonic eddies are slightly more abundant during the first half of the year ( Figure 1C ). We tracked 571 trajectories of cyclonic long-lived eddies, with various patterns: cyclonic eddies along the African coast traveled southward with an offshore component. In contrast, northern cyclonic eddies traveled westward with a small southward component. Cyclonic eddies forming in the eastern part of the channel from 15°S to 22.5°S traveled southwestward (Halo et al., 2014a). Cyclonic eddies can also form over the biologically productive African shelf (Leal et al., 2009; Paula et al., 1998), in Sofala Bank (Malauene et al., 2018), and more predominantly in the Delagoa Bight (Cossa et al., 2016). Anticyclonic eddies are distributed throughout the channel, except over the African shelf of Sofala Bank where the formation of short-lived cyclonic eddies dominates. Notably, persistent anticyclonic eddies, such as rings, are concentrated within the ‘eddy corridor’ along the Mozambique shelf edge. This phenomenon follows the underlying topography, spanning approximately 100 to 500 kilometers offshore. These anticyclonic eddies travel southward toward the Agulhas system, as documented by Malauene et al. (2018); Schouten et al. (2003); Quartly et al. (2013), and Halo et al. (2014a).

We found 455 and 278 tracks of long-lived anticyclonic eddies and rings, i.e., 23 and 14 tracks per year, respectively. Restricted to a region spanning 1/2° around 17°S we found ~5 ring tracks per year, consistent with the literature (de Ruijter et al., 2002; Halo et al., 2014a; Schouten et al., 2003). Most anticyclonic eddies and rings occur in the first quarter of the year (39% in January-February-March).

While the most limiting nutrients in the western Indian Ocean are generally nitrate and phosphate, with silicate being more accessible (Barlow et al., 2007; Kyewalyanga et al., 2007; Leal et al., 2009; Sá et al., 2013), nitrate is recognized as the primary factor limiting phytoplankton production in the MC (Koné et al., 2009). Therefore, our study focuses only on nitrate and Chl-a in nanophytoplankton and diatoms. All analyses related to the other nutrients and zooplankton are provided in the Supplementary Material ( Supplementary Figure S20 ).

Figure 2 illustrates the distinct cross-shore properties of cyclonic eddies, anticyclonic eddies, and rings. Cyclonic eddies display an upward doming signature of isotherms, isohalines, and isolines of nitrate, all reaching about 25m. Symmetrically, anticyclonic eddies exhibit a downward bend of about 25 meters, characterized by warm and fresh waters, slightly depleted in nitrate in the core. The downwelling in rings is even more pronounced with isolines deepening to ~50m with overexpressed gradients. It is important to note that the surrounding (for R> ± 2) physical and biogeochemical properties of cyclonic eddies, anticyclonic eddies, and rings exhibit differences. This disparity is mostly due to the spatial distribution of each eddy type associated with large-scale water masses gradients (see Supplementary Figures S3 , S5 , S10 ). Note also that the vertical profile of salinity for all composites is typical of the region ( Supplementary Figure S4 ) with a vertical inverse gradient: saltier waters are at the surface contrasting with fresher waters at depth.

Regarding the effect on phytoplankton, nitrate doming in the core of cyclonic eddies leads to a doubling of Chl-a in diatoms at 75m depth and a slight decrease in Chl-a in nanophytoplankton, compared to the surrounding areas. Such a shift in the ecosystem assemblage (an increase in diatoms and a decrease in nanophytoplankton) is likely driven by vertical nutrient input from eddy-pumping which favors diatom growth. There is also a slight surface signature of diatoms driven by the vertical mixing of the top of the DCM within the MLD. In anticyclonic eddies, the core shows a decrease in diatom Chl-a and no significant effect on nanophytoplankton Chl-a compared to the surrounding areas, however at the edge of the eddy there is a slight increase within the MLD (between ±1R and ±2R). This phenomenon is even more pronounced in the rings, where nitrate concentrations is higher in the surrounding areas especially at the eastern edge; this is associated with local increase of Chl-a in nanophytoplankton from the subsurface to the surface. Chl-a in diatoms is higher at the eastern edge of the rings due to the distribution of eddies near the African coast, where the eddy-topography interactions lead to strong nutrient enrichment (Marsac et al., 2014).

As expected, on average, cyclonic eddies positively influence primary production, likely through vertical advection of nutrients at the base of the euphotic layer, in the subsurface at the DCM (Chenillat et al., 2016). In the DCM of cyclonic eddies, a stronger signature over diatoms occurs, which is note the case over nanophytoplankton. This highlights the role of cyclonic eddies in modifying planktonic community composition compared to the non-eddy environment (Pereira et al., 2024). On average, anticyclonic eddies slightly promote nanophytoplankton primary production at the edge, and rings drive a strong response of nanophytoplankton from the edge to the core: enriched shelf water is entrained around these anticyclonic structures and can be transported along isopycnals downward to the core and below the MLD, without surface signature. Both anticyclonic eddies and rings appear to be unfavorable for diatom growth due to limited vertical advection of nutrients. This statistical confirmation highlights the significance of eddy-topography interactions in influencing the biological production in the MC, as described by José et al. (2014), especially for rings. Cyclonic eddies could also be affected by eddy-topography interactions in terms of nutrient enrichment and primary production, but the main dynamics that is highlighted by this statistical analysis is the local vertical input from eddy-pumping. It should be noted that an additional impact of nutrient input from rivers is to be expected. However, this has not been considered in the context of the numerical experiment, and thus falls outside the scope of this study.

To better describe and disentangle the effect of eddies on biogeochemical properties, our analysis focuses on their regional influence. To do this, for each long-lived eddy, we compute the concentration anomalies of nitrate, Chl-a in nanophytoplankton, and Chl-a in diatoms between the eddy core and eddy surroundings, averaged over 0-100 m depth, reduced to the surrounding values to obtain a percentage ( Figure 3 ).

Over the Sofala Bank and the Delagoa Bight, a significant increase in the concentration of chlorophyll-a and nitrate (and all other nutrients, Supplementary Figure S21 ) is observed in the cyclonic eddy cores when compared to the surrounding eddy regions. This finding is consistent with Lamont et al. (2014), which suggests that eddy-topography interactions play a role in the upwelling of deep, nutrient-rich waters favoring primary production (Leal et al., 2009). A Similar positive trend is observed over the central portion of the MC, from 15°S to 22.5°S, in open ocean waters, distant from potential eddy-topography interactions. It is noteworthy that this increase in the northern MC and over the shelf is considerably more pronounced for diatoms than for nanophytoplankton, indicating a potential shift in ecosystem composition compared to non-eddy areas, as suggested earlier. Such planktonic modulation can be explained by eddy pumping, whereby deeper, nutrient-rich waters are brought upward into the euphotic zone, favoring larger diatoms-like phytoplankton (Landry et al., 2008; Chenillat et al., 2016; Pereira et al., 2024). It is established that diatoms exhibit a rapid response to nutrient injections in the euphotic zone (Brown et al., 2008).

In the Comoros basin, cyclonic eddies cores also demonstrate positive trends of nitrate and Chl-a in diatoms (also observed for phosphate, silicate, and iron, and not consistently for Chl-a in nanophytoplankton, zooplankton, and ammonium, Supplementary Figure S21 ). This is likely the result of the intrusion of significant quantities of upwelled nutrient-rich water from the NEMC, which interacts with eddies forming around Comoros (see Figure 1 ; Huang et al., 2024).

At the southern tip of Madagascar, the cores of cyclonic eddies present positive trends in nitrate and Chl-a. This region is known for its coastal upwelling (Machu et al., 2002) where cyclonic eddies trap and efficiently transport enriched upwelled water and associated ecosystem (Chenillat et al., 2016) to the southwest.

In the offshore region of the central MC, and mostly along the shelf, anticyclonic structures present a pronounced negative trend, which is consistent with low primary production in their core (Tew-Kai and Marsac, 2009). However, some positive trends can be observed confined along the entire shelf limit, which is particularly relevant for rings. These local positive trends concern mainly nitrate and Chl-a, mostly for nanophytoplankton (and all nutrients except silicate) off Angoche, south of Sofala bank, and in the southern region, at the onset of the Agulhas Current. Such enrichment in anticyclonic eddies has already been observed (Rodriguez et al., 2003; Waite et al., 2007), particularly in those interacting with the shelf, with material entrainment at their edges (Crawford et al., 2005; Moore et al., 2007). This highlights the suppressor effect of anticyclonic eddies on diatoms: anticyclonic eddies promote the trapping of coastal waters without substantial enrichment from deep water, thereby enabling the ecosystem to self-maintain through regenerated primary production. This limits the growth of diatoms-like phytoplankton and favors production by nanophytoplankton (Chenillat et al., 2015).

In the southern MC, a mixture of positive and negative signals is observed, with a majority of negative trends for cyclonic eddies and positive trends for anticyclonic eddies for all considered biogeochemical variables. This highlights more complex dynamics with several distinct processes at play. These include Ekman pumping in cyclonic eddies, lateral transport of material within eddies, and eddy-eddy interactions that interplay with the seasonal fluctuations of different key processes, such as variability of the MLD, the euphotic depth, wind patterns, and biological dynamics.

In addition, we observed a significant latitudinal gradient in the water mass properties of the eddy cores along the MC ( Figure 4 and Supplementary Figure S21 ). The climatology shows a southward increase in nutrients and Chl-a (Sá et al., 2013), along with a decrease in temperature and an increase in salinity, which induces a deepening of the DCM. A comparison between the north-to-south climatology and the eddies reveals a significant effect of eddies on the vertical redistribution of oceanic variables, characterized by shallower nutriclines and DCM, particularly in the region north of 25°S, strictly in the MC. The isolines of temperature, nutrients, the DCM, and the MLD are observed to be shallower in the north-to-south cyclonic eddies than in anticyclonic eddies and rings. This finding is consistent with the results presented in Figure 2 . In comparison to the overall north-to-south average, the concentration of chlorophyll-a is higher within the cores of cyclonic eddies than within anticyclonic eddies throughout the entire MC. The DCM of nanophytoplankton is present throughout the MC for all eddies and rings. In contrast, the DCM of diatoms is deeper and exhibits two distinct hotspots: one around 15°S and the other around 20°S, corresponding to the Angoche and the middle of Sofala Bank, respectively. These hotspots are also visible in anticyclonic eddies but are more pronounced in cyclonic eddies. The northern hotspot is also evident in the concentration of Chl-a in nanophytoplankton within cyclonic eddies. This phenomenon can be attributed to the high eddy generation occurring in northern Madagascar ( Figures 1 , 3 ) (José et al., 2016) combining the effect of strong nutrient vertical input in eddies favoring diatoms in deeper waters contrasting with lateral advection and trapping of coastal water as they move southward along the slope. This enrichment could be further enhanced by eddy/eddy interactions and offshore export of coastal water masses by filaments. These findings are consistent with previous studies (e.g., José et al., 2014; Roberts et al., 2014) and recent in situ observations of a filament situated between a dipole formed by a cyclonic eddy and a ring during the RESILIENCE cruise (doi:10.17600/18001917). From the northernmost point to 26°S, the DCM of nanophytoplankton remains high for all eddy types, compared to the climatology. This can be attributed to the retention of water masses and ability of eddies to support a phytoplanktonic production based on remineralization, below the MLD (Chenillat et al., 2015) as they move southward in the MC. With regards to the southern hotspot, elevated subsurface Chl-a concentrations around 20°S and 26°S are attributable to the coastal increase of plankton biomass over the shelf. Our findings indicate that anticyclonic eddies and rings exert limited control over diatom production. This may be due to the physical, biogeochemical, and ecological characteristics of the background water masses trapped by these entities, favoring regenerated production by nanophytoplankton.