Predators distribution is estimated from tracking data of seabirds and seals from the Prince Edward Islands between 2003 and 2014 and collected from research programs supported by the National Research Foundation and the South African Department of Environmental Affairs (Reisinger et al., 2017)¹. The animals’ trajectories we use in this work have been analyzed in detail in Reisinger et al. (2018): here we report only the information relevant for our study, which is the large-scale geographical extension of their habitat utilization, i.e., the whole smoothed spatial range of predators. In this study we consider only the tracking data of seals and penguins, because they display a more selective habitat than flying seabirds (Reisinger et al., 2018). The Subantarctic Fur Seals (Arctocephalus tropicalis) are not considered because the at-sea distribution of this colony is mostly located north of the Subantarctic Front which bound our study region [see Figure 2 in Reisinger et al. (2018)]. Kernel densities of Antarctic Fur Seals (Arctocephalus gazella), Southern Elephants Seals (Mirounga leonina), King Penguins (Aptenodytes patagonicus), Macaroni Penguins (Eudyptes chrysolophus) and Southern Rockhopper Penguins (Eudyptes chrysocome chrysocome) from the Prince Edward Islands were calculated using the “kde2d” function from the R-package “MASS” on the observed tracks (Venables and Ripley, 2002). The habitat utilization for each colony is identified with a threshold of 95% of each predator density (complete kernel utilization distributions are available for each species in Reisinger et al., 2018; Figure 2). This allows us to identify the whole spatial range used by multiple subantarctic predators including different phases and behavior of their trajectories (usually traveling and foraging).

Seamounts are defined as geomorphological features having a vertical elevation of more than 1000 m from the seafloor with limited extent across the top summit region (Menard, 1964; Staudigel et al., 2010). The seamounts’ summit location is extracted from Yesson et al. (2011). We retain here only the shallower structures having a maximal water depth of 1000 m, in order to compare them with the ocean upper layer horizontal velocity fields derived from altimetry. In the following, a seamount is defined by a disk centered in the seamount summit and with a 70 km radius. In the same way, the region under the influence of Bouvet Island is defined by a disk centered in the island (3.36°E; 54.42°S) with a 70 km radius.

Geostrophic velocities, surface chlorophyll a (Chl a) and temperature reanalysis are evaluated between 2000 and 2015 from different products distributed by the European Copernicus Marine Environment Monitoring Service².

Geostrophic currents are quantified through an altimetry multi-satellite global product (product id: SEALEVEL_GLO_ PHY_L4_REP_OBSERVATIONS_008_047) produced by SSALTO/DUACS. This product processes data from all altimeter missions (Jason-3, Sentinel-3A, HY-2A, Saral/AltiKa, CryoSat-2, Jason-2, Jason-1, T/P, ENVISAT, GFO, ERS1/2) and has temporal and spatial resolutions of, respectively, 1 day and 1/4°. Trajectories are derived by applying a Runge–Kutta fourth order scheme with a time step of 6 h. Velocity fields are linearly interpolated in both space and time.

Satellite surface Chl a concentration is analyzed with the Global Ocean Color product (product id: OCEANCOLOUR_ GLO_CHL_L4_REP_OBSERVATIONS_009_082) produced by ACRI-ST. The product evaluated has a 1/24° spatial resolution. Here, we consider monthly-mean composites in order to reduce the cloud coverage and at the same time preserve the mesoscale variability of the Chl a blooms. Similarly to previous studies, the term “bloom” refers to regions of relative elevated Chl a compared to the background concentration (Sokolov and Rintoul, 2007). The statistical significance of the Chl a enhancement is evaluated with the bootstrap method (a detailed description of the method used for the identification of Chl a blooms is provided in the section “Seamount Mass Effect: Detection of the Lagrangian Plume Stemming From the Seamounts”). Chl a values are investigated in spring and summer (from October to February) ahead of the maximum feeding activities of higher trophic levels (Laws, 1985; Moline et al., 2004; De Broyer et al., 2014).

Phytoplankton dominance is evaluated from the PHYSAT observations (Alvain et al., 2005, 2008) between 2000 and 2011 from the month of December and January, when we expect a larger diatoms dominance (Smetacek et al., 1990; Alvain et al., 2008). The PHYSAT method has already been validated at the global and regional scale in the Southern Ocean (Alvain et al., 2005, 2008, 2013; see in particular d’Ovidio et al., 2010 Supplementary Material for a validation at the latitudes used here). The method identifies probable specific dominant phytoplankton groups based on the analysis of spectral measurements computed by ocean-color sensors. These are nano-eukaryotes, Prochlorococcus, Synechococcus, diatoms, Phaeocystis-like and coccolithophores. Note that a phytoplankton group is defined as dominant when it is a major contributor to the total pigment (Alvain et al., 2005). As demonstrated in Della Penna et al. (2018), a large number of pixels of PHYSAT daily maps are flagged as not classified in this oceanic region, so that composite maps must span very large periods in order to provide meaningful spatial patterns. For this reason, and because we study large scale features, in our analysis we have used multi-year PHYSAT composites, without resolving interannual variability.

Vertical distribution of temperatures is evaluated with the global ocean reanalysis GLORYS2V4 produced by Mercator Ocean for the Global Ocean and Sea Ice Physics (product id: GLOBAL_REANALYSIS_PHY_001_025). This product is performed with NEMOv3.1 ocean model in configuration ORCA025_LIM and has horizontal spatial resolution of 1/4° with 75 vertical levels. For the analyses we used the product version averaged by month. The water masses associated to the position of the Antarctic Polar Front (PF) are identified as the northern extent of the presence of the 2°C isotherm minimum in the subsurface (0–500 m depth) (Botnikov, 1963).

The vertical profiles of Chl a concentration and backscattering measured from the Biogeochemical Argo (thereafter BGC-Argo) floats have been extracted from the CORIOLIS Global Data Assembly Center³. Quality control tests have been achieved on the Chl a and backscattering data (Schmechtig et al., 2014, 2018). Non-photochemical quenching on daytime profiles are also corrected on the Chl a data following (Xing et al., 2012). In this aim, the maximal Chl a value above the mixed layer depth is extrapolated until the surface and corrected with a specific correction factor. In order to adapt the correction procedure to the study region, the non-photochemical quenching is corrected with the calibration factor of 3.3 used by Ardyna et al. (2019) in the same region. The mixed layer depth is evaluated in each profile as the depth of a density difference of 0.03 kg/m³ in respect to the density at 10 m depth. The density is computed based on International Thermodynamic Equation of Seawater (Sérazin, 2011).

In order to identify the effects of seamounts on phytoplanktonic blooms we analyze whether possible water parcels displaying enhanced Chl a have, or have not, been in contact with the seamounts. To achieve this, we compute a backward-in-time, altimetry-based Lagrangian calculation from the bloom periods. This calculation allows us to identify the water parcels that transited over a seamounts area, defined here as a disk of 70 km radius. This value is chosen considering that altimetry data are expected to resolve features on the order of 70 km or larger (Pascual et al., 2006). The aim is to compare the water dispersion pathways computed downstream of the seamounts, in the following referred to as the Lagrangian water plume, to the Chl a plume. This approach has already been validated for the island mass effect, to reproduce the Kerguelen bloom (d’Ovidio et al., 2015; Sanial et al., 2015) and the Crozet bloom (Sanial et al., 2014) at mesoscale precision.

For each studied period the analysis is conducted in two steps:

(1) Identification of the seamounts that are located within deep iron-enriched hydrographic regions. For each seamount we verify if, in the 4 months before the bloom, it was located within the PF. This current core of the ACC is associated to enhanced dissolved iron concentrations (De Baar et al., 1995; Löscher et al., 1997; Measures and Vink, 2001; Mengelt et al., 2001; Viljoen et al., 2018) presumably upwelled from the deep layer (Sokolov and Rintoul, 2007). The PF is identified for the whole period in the vertical temperature reanalysis as described in section “Satellite Observations and Reanalysis Data.” We applied the same approach to the water parcels that have been in contact with Bouvet Island, which is also a potential source of iron to the environment (Perissinotto et al., 1992; Croot et al., 2004). The latter contact is estimated by detecting the water masses which transited over the island in the 4 months before the bloom.

(2) Identification of the Lagrangian water plume from the selected seamounts. Each water parcel of the study region is advected backward-in-time from the blooming period for 4 months. The parcels which are found to come from an area where a seamount is present are then given a timestamp with the time, in days, since they left that topographic features. The 4-month time window chosen for the integration assumes that the extension springtime bloom is pre-conditioned by the winter-time advection of iron rich waters (Mongin et al., 2009).

The enhancement of the Chl a concentration within the 100-day Lagrangian water plume is evaluated with the bootstrap method in order to obtain its statistical significance. The bootstrap is conducted on sub-sampled Chl a maps, selecting one pixel each five both in longitude and latitude directions. This is done to remove spatial correlations of the order of the typical size of a mesoscale eddy of this region (about 100 km). For the region under the influence of Bouvet, the Lagrangian water plume spreading downstream of the island is removed from the Lagrangian water plume reconstructed with the seamounts. This allows us to distinguish the plume from the seamounts in respect to the one induced by the island.

Two BGC-Argo floats (WMO 6901585 and 6901688) sampled the Shona Ridge region. They crossed the seamount chain around 49°S between the end of December 2014 and January 2015 with a time lag of a few days (Figure 2). A local increase of the depth-integrated biomass over the mixed layer downstream of seamounts (8–10°E) is recorded by the profiling floats: the Chl a content doubled in the float WMO 6901585 (Figure 2A) and increased by approximately one third in the float WMO 6801688 (Figure 2B). These observations contrast with the homogeneous low Chl a concentration (about 0.3 mg/m³) observed by satellite in January 2015 in the near surface (background images in Figures 2A,B). When looking at the Chl a content in the water column, neglecting the variability of the mixed layer depth which is deeper than 100 m upstream of the seamounts and shallower than 40 m over them, the intensification of the Chl a signature downstream of the seamounts is evident in both floats (Figures 2C,D). The vertical profiles depict an increase in Chl a and backscattering values in the first 100 m depth downstream of the seamounts (respectively, about 1 mg/m³ and 3.0 × 10^–3 m^–1) compared to the water masses upstream (respectively, about 0.6 mg/m³ and 2.0 × 10^–3 m^–1). Note that the yellow triangles in Figures 2C,E (respectively, Figures 2D,F) locate the time of the closest position of the BGC Argos trajectories with the yellow seamount in Figure 2A (respectively, Figure 2B). They symbolize the crossing of the seamount chain. The phytoplankton enhancement downstream of seamounts is further investigate with satellite observations which have larger spatial and temporal covers.

We assume here the hypothesis of a seamount mass effect on phytoplankton biomass, in analogy to the well known island mass effect. This hypothesis implies the formation of phytoplanktonic blooms induced by natural iron fertilization spreading downstream of the seamounts. We discuss the possible mechanisms of iron upwelling in section “Discussion”.

Two seamount chains are susceptible to lift iron-laden deep waters to the surface layer and consequently enhance the phytoplankton biomass in the studied region: the Shona Ridge System (SRS), which is composed by the Shona Ridge and the Meteor Rise, and the South-West Indian Ridge (SWIR) (see Figures 1B,C for their localization). Yesson et al. (2011) identified 36 seamounts shallower than 1000 meters depth associated with the SRS, with one third of them shallower than 100 m depth, and 26 seamounts shallower than 1000 m associated to the SWIR in the study region (between 0 and 30°E). When looking at Chl a images not yearly averaged, but over time periods of 1 month during spring or summer, Chl a signatures significantly above the background (around 2-fold) do appear downstream of both the seamount chains (as illustrated in Figures 3A,B, 4A).

In order to further sustain the hypothesis of a seamount mass effect, we study the morphology of the satellite-observed Chl a plumes detected in the vicinity of seamounts. In particular, we analyze whether the shape of these plumes is consistent with a dispersal pattern stemming from a seamount. In order to achieve this, we reproduce the Lagrangian plumes associated with the seamounts during the blooms event using altimetry-derived currents. The analysis consists in detecting the water parcels which transited above the seamounts in a certain time period with a Lagrangian approach. As explained in section (Seamount Mass Effect: Detection of the Lagrangian Plume Stemming From the Seamounts), for each bloom period the Lagrangian plumes are derived only from the subset of seamounts which were in contact with potentially deep iron-enriched hydrographic provinces. Here we consider two regions that have been identified in the literature for displaying higher iron concentration, the water masses downstream of Bouvet Island (Perissinotto et al., 1992; Croot et al., 2004) and the water masses located in the ACC along the jet associated with the PF (De Baar et al., 1995; Löscher et al., 1997; Measures and Vink, 2001; Mengelt et al., 2001; Viljoen et al., 2018). This hypothesis will be discussed in section “Drivers of the Seamount Mass Effect.” We illustrate the analysis region by region.

A 15-year climatology of the December Chl a concentration in the southern Agulhas Basin is shown in Figure 5, which corresponds to the time of Chl a peak in the region (Sokolov and Rintoul, 2007). A 15-year climatology of the Lagrangian water plume issued from the seamounts from the SHS and the SWIR (black triangles), computed as in section “Altimetry-Based Lagrangian Reconstruction of the Seamount Mass Effect,” is overlapped to the Chl a map (black contour). The region covered by the Lagrangian water plume is upstream of the pinnipeds and penguins ranges (white contours). This indicates that the water masses targeted by the predators’ colonies have been under the influence of the seamounts in the previous months (the Lagrangian water plume is represented with a 100-day threshold, as in the analyses of section “Altimetry-Based Lagrangian Reconstruction of the Seamount Mass Effect”). The similarity between the climatological Chl a and Lagrangian plumes is less evident compared to the monthly analyses previously considered. This is primarily due to the smoothed signal in the Chl a climatology. However, where the mean pigment distribution is more contrasted, as in the northern edge of the Lagrangian plume, the two signals appear to have qualitatively similar patterns.

In order to investigate the signal of the seamounts in the climatological phytoplankton biomass we compute the mean Chl a concentration in a 2° latitude narrow band centered on the current core of the ACC which we identify by the PF, between 2000 and 2015 (Figure 5B). Figure 5B shows that the highest values (mean Chl a about 0.7 mg/m³) are observed upstream of the study region, between 20 and 5°W and decrease eastward. Intense Chl a concentrations extend downstream of South Georgia (36°W, 54°S) over wide distances due to the intense transport (e.g., Korb et al., 2004; Sokolov and Rintoul, 2007; De Jong et al., 2012; Borrione et al., 2014; Graham et al., 2015) and to the regular iron inputs linked to flow-topography interactions (Sokolov and Rintoul, 2007). The averaged Chl a level do not significantly increase in correspondence of the interaction between the ACC and the shallow seamounts of the SRS (Figure 5B). However, we note the high standard deviation of the signal between 0 and 15°E compared to the downstream signal (respectively, σ ≈ 0.2 mg/m³ and σ ≈ and 0.1 mg/m³) indicating an important interannual variability. Compared to the SHS, the Ob and Lena seamounts close to 50°E display a lower mean Chl a concentration (Δμ ≈ 0.1 mg/m³) and a lower standard deviation (Δσ ≈ 0.1 mg/m³), indicating a lower phytoplankton biomass and a low temporal variability.

The possibility for primary production of sustaining the large trophic chain up to megafauna, is known to largely depend not only on its abundance but also on the phytoplankton composition (e.g., Moline et al., 2004). We thus analyze our study region with maps produced with the PHYSAT algorithm (Alvain et al., 2005, 2008). The latter cluster together the ocean color spectral signal into classes associated to typical phytoplanktonic types which are statistically dominant for those bio-optical anomalies.

PHYSAT patterns downstream of the SRS and SWIR seamounts (0–30°E, 50–53°S in Figure 6) mainly contain the same classes found in island-induced blooms (70–100°E, 50–53°S in Figure 6) and differ from the type classes found in the Australian-Antarctic Basin (i.e., downstream of the Kerguelen bloom about 100–130°E) (Figure 6A). According to the PHYSAT observations, during December and January between 2000 and 2011, the seamount region is dominated more than 55% of the time by diatoms (σ = 30%) and less than 1% of the time by Phaeocystis (σ = 5%). By comparison, the Kerguelen bloom (and respectively, the offshore region downstream of this bloom) is dominated 70% (respectively 16%) of the time by diatoms and 3% (respectively 80%) of the time by the small Phaeocystis (Figures 6B–D).

What is the origin of the blooms observed in the wake of seamounts in the southern Agulhas Basin? In most of the previous studies, the Chl a anomalies observed over the seamounts and in the surrounding waters were linked to the shallow seamount depths. However, the spatial and temporal variability of biological effects were rarely understood due to the complex nature of the physical processes involved there (e.g., Venrick, 1991; Pitcher et al., 2007; Pauly et al., 2017). The process presented here is framed by the limiting effect of iron on primary production (Martin, 1990) and the preponderant structuring effect of the horizontal transport on biochemical properties and ecosystems in the ACC region. The two necessary conditions identified for the fertilization of water masses in the open ocean southern Agulhas Basin are thus: (i) the presence of hydrographic regions associated, according to the literature, to potential enhanced dissolved iron content at intermediate depth; and (ii) upwelling dynamics due to the interaction of seamounts shallower than 1000 m depth with an energetic branch of the ACC.

In terms of iron content, two possibly enriched regions are identified in the southern Agulhas Basin. The first one is composed by the water masses that have transited over Bouvet Island, which is a known iron source (Perissinotto et al., 1992; Croot et al., 2004). The second one, is composed by the water masses located in the ACC, along the jet associated with the PF. Numerous in situ observations have recorded enhanced iron concentrations in the surface waters close to this ACC front, both in the Atlantic sector (De Baar et al., 1995; Löscher et al., 1997; Viljoen et al., 2018) and elsewhere (Measures and Vink, 2001; Mengelt et al., 2001). In the deeper ocean, iron measurements are less frequent (e.g., Tagliabue et al., 2012). Nevertheless, enhanced iron concentrations appear to occur in the Circumpolar Deep Water close to the PF region in correspondence of the Mid-Atlantic Ridge (Löscher et al., 1997). Furthermore, the analysis of the global GEOTRACES database indicates a sharp ferricline in the Atlantic ACC region reflecting the signature of subsurface lateral transfer of dissolved iron from the subantarctic islands (Tagliabue et al., 2014). Indeed numerous studies have concluded that the intense velocities of the PF current core result in a strong transport of nutrients downstream of South Georgia (Korb et al., 2004; Venables and Meredith, 2009; Venables et al., 2012; Borrione et al., 2014) and in a regular iron supply to the surface water due to the ACC-bathymetry interactions (Sokolov and Rintoul, 2007).

In terms of upward dynamics, the interaction of the ACC with sharp topographic features is well known to generate upwelling hotspots carrying deep waters to the surface (Viglione and Thompson, 2016; Tamsitt et al., 2017) and enhancing surface Chl a concentrations (Sokolov and Rintoul, 2007). Three different mechanisms have been identified in the past for explaining upwelling in the presence of seamounts. The first mechanism is eddy fluxes which are intensified when the ACC encounters a rough topography due to an increase in baroclinicity. The eddy activity enhanced in this way at seamounts can vigorously drive upward northern-sourced deep water along steeply sloping isopycnals in the PF region (Tamsitt et al., 2017; Rintoul, 2018). The enhanced eddy activity at seamounts is also associated to intense flow shear, which in turn preconditions an ageostrophic submesoscale circulation characterized by deep reaching vertical motion (Lévy et al., 2018; Siegelman et al., 2019). A second possible mechanism for upwelling is the bottom pressure torque created when the ACC impinges over bathymetric obstacles. This process induces upwelling upstream of a seamount and downwelling downstream of it (Hughes, 2005; Sokolov and Rintoul, 2007; Thompson and Naveira Garabato, 2014; Rintoul, 2018). This situation has been described in the Atlantic sector of the ACC region, at the South and North Scotia Ridges, the South Orkney Islands and the South-West Indian Ridge (Sokolov and Rintoul, 2007). Finally, a third mechanism which is able to connect deep water to the surface layer is the generation of internal lee waves at seamounts. These waves propagate upward and break, enhancing diapycnal mixing (Watson et al., 2013) and fluorescence signature over the seamounts (Brink, 1995; Kunze and Sanford, 1997; Eriksen, 1998). This has been observed for instance at the seamounts of the SWIR, at the northeast of the southern Agulhas Basin (Read and Pollard, 2017).

The interaction of a strong current with a topographic sharp relief is also known to produce Taylor columns, that is, semi-permanent anticyclones on the top of the topographic obstacle which can trap water for periods of months or longer (Taylor, 1923; Meredith et al., 2015). These features can have a special ecological interest because they may retain over the seamounts aging trophic webs, thus providing predictable preying hotspots (e.g., Pitcher et al., 2007; Morato et al., 2010; Cascão et al., 2017). Such circulations, more intense at high latitudes due to the enhanced Coriolis force (Owens and Hogg, 1980; Chapman and Haidvogel, 1992), have been observed in the Southern Ocean at the Maud Rise (Van Bennekom et al., 1988) and conditions conducive to their formation have been found at the Georgia Rise, South Scotia Ridge and at the Prince Edward plateau (Perissinotto and Rae, 1990; Meredith et al., 2003, 2015).

All together, the mechanisms here proposed for the Atlantic ACC region are schematized in Figure 7. The iron content increases in the surface layer at the South Georgia Island and sinks to the subsurface along the flow direction, with regular upward movements due to the interactions between the ACC and the topography. The Chl a blooms identified in the study cases (Figures 3, 4) and the important temporal variability of the climatological surface Chl a content close to the seamounts (Figure 5) are the surface expression of this topography-ACC interactions, usually hidden in the mean climatological Chl a signal when compared with subantarctic island-blooms (e.g., Graham et al., 2015; Ardyna et al., 2017). By highlighting the role of iron content of the upstream water masses, the mechanism here suggested could explain why seamounts seem to display re-enhanced phytoplankton biomass when they are close to rich systems, such as coastal margins (Van Bennekom et al., 1988; Perissinotto and Rae, 1990; Meredith et al., 2003; Graham et al., 2015; Oliveira et al., 2016; Lemos et al., 2018), or whether they are located in productive frontal zones (Djurhuus et al., 2017). On the other hand (and following Sokolov and Rintoul, 2007) our results identify processes extending at much larger spatial scales compared with the local phytoplankton enhancements described in previous studies which have focused on seamounts in other regions (e.g., Oliveira et al., 2016; Lemos et al., 2018). This larger extent is due to intense advection by the ACC, in analogy to the spread of non-consumed iron from the subantarctic islands in wintertime which preconditions the size of the islands’ blooming area (Mongin et al., 2009; d’Ovidio et al., 2015; Graham et al., 2015).

Our approach does not explain all the patterns of primary production in the ACC region. The ACC core also interacts with a group of shallow seamounts (<1000 m depth) between 10°W and 6°E (Figure 5C) which corresponds to climatological Chl a concentrations topping at 0.7 mg/m³ (Figure 5B). Differently from the analysis of the SHS and SWIR (black contour in Figure 5A) the Lagrangian plume spreading from the seamounts close to 10°W only partially explain the Chl a plume in this region (result not shown). Most of Chl a seems to be aggregated close to the current core, which, constrained by the Mid-Atlantic Ridge, shifts northward close to 10°W (Boehme et al., 2008). The specific regional circulation catalyzed by the topography may explain the Chl a distribution. On the other hand, our analysis leaves open the question concerning why seamounts of other regions do not seem to be associated to intense Chl a plumes, such as the shallow Ob and Lena seamounts (∼50°E in Figure 5). This could be explained by the mechanism schematized in Figure 7. The deep iron, primarily upwelled to the intermediate layer at the South Georgia Island (Korb et al., 2004; Borrione et al., 2014), is assumed to slowly decay along the flow direction from west to east. Following this hypothesis, it seems plausible that the seamount effect may be less intense for topography features located further downstream of the South Georgia Island (∼35°W), as the Ob and Lena seamounts (∼50°E), compared to the closer ones (as the SRS and SWIR). We note that the mechanism here proposed consider primarily external iron inputs, i.e., the upstream iron transport. However, some local processes have also been identified as sources of surface nutrient supply at seamounts. For instance, the interaction of the porewater with sandy sediment releases reduced metal species to the water column (Huettel et al., 1998) and leads to an enhanced erosion of the sediments due to the physical and chemical actions of potential benthic communities (Rhoads and Boyer, 1982). For all these reasons, more detailed studies should be done in order to unambiguously identify the mechanisms at the origin of the Chl a enhancements observed in this study.

The surface Chl a concentrations observed in satellite images downstream of seamounts are minor compared to the ones occurring downstream of islands. For instance, in the Crozet bloom Chl a concentrations exceed 2 mg/m³ (Bakker et al., 2007), four times the ones observed in our analysis. However, we stress that, to assess the total phytoplankton biomass of a given region, one must integrate Chl a concentration over the mixed layer (ML) (e.g., Ardyna et al., 2019). The ML in the seamount region is at least between 10 and 30% deeper of the ML in the Kerguelen bloom region during January and February and between 30 and 50% deeper during December (Supplementary Figure S1). When compared to the Crozet and the South Georgia blooming regions the seamount ML increases up to 60–100% (Supplementary Figure S1). Therefore, the phytoplankton biomass integrated over the ML may present less intense contrasts between the seamount open ocean and the island bloom regions than the surface signals. Furthermore, we note that a deeper ML may also induce a lower Chl a signal in the satellite images, intensifying the observed surface differences between the two regions.

The analysis of the phytoplankton composition of the seamount plumes provides further insights about the relevance of this biological production from an ecological perspective. The phytoplankton community identified in the Atlantic ACC shows a bio-optical signature similar to the ones found in diatom-dominated island blooms, such as the region downstream of Kerguelen, and different from the nearby oligotrophic waters, characterized by Phaeocystis-types (Figure 6). This nanophytoplankton is known to be rapidly consumed by the microbial food web (Detmer and Bathmann, 1997; Pakhomov et al., 2019) and therefore its occurrence is not associated with profitable foraging grounds for top predators (Cushing, 1989; Moline et al., 2004; Pakhomov et al., 2019). Conversely, diatom-dominated blooms are more likely to support large trophic webs which lead to fish and crustaceans and higher predators (Cushing, 1989; Moline et al., 2004; Pakhomov et al., 2019). Presence of a large and rich (in terms of diversity) megafauna in our study region is confirmed by biologging data (Arthur et al., 2017; Pistorius et al., 2017; Reisinger et al., 2018). The similarity between the seamounts region to the island blooms is supported also by in situ measurements showing diatom enhanced concentrations at the Atlantic PF region in the vicinity of the Islas Orcadas Rises and Mid-Atlantic Ridge (Laubscher et al., 1993; Froneman et al., 1995; Bathmann et al., 1997; Detmer and Bathmann, 1997; Smetacek et al., 1997). On the other hand, the contrasted phytoplankton community between the seamount region and the classic HNLC areas (as the one between 110 and 130°E in Figure 6) would emphasize the interest of regions downstream of topographic highs. However, the lack of in situ phytoplankton observations in this low chlorophyll region prevent us to conclude on the potential zonal differences in the circumpolar phytoplankton communities of the PF region. Observation of mid trophic levels would also be valuable for linking the primary producers to the higher predators and verifying these hypotheses. Flynn and Williams (2012) showed that the topography-flow interaction over the Macquarie Ridge (160°E) seems to induce an increased biomass of the lanternfishes, a key prey for pinnipeds and seals (e.g., De Broyer et al., 2014). Further analyses of intermediate trophic levels should be done for explaining the biological productivity inferred from land-based top predators and assessing how each colony specifically relies on seamounts production. For instance, the distribution of surface and low trophic level foragers, such as Southern Rockhopper Penguin and Macaroni Penguin (Brown and Klages, 1987; Adams and Brown, 1989; Wilson et al., 1997; Crawford et al., 2003; Whitehead et al., 2016), might be more closely related to the surface productivity than the distribution of mesopelagic foragers.

Our analyses advocate for a connected vision of the biological productivity of the ACC region. Following the ACC, the high Chl a concentration decreases from west to east downstream of the South Georgia Island around the ACC core, but it is re-enhanced after the ACC interacts with submerged topographic features, in particular the SRS (as schematized in Figure 7). A more comprehensive view may have consequences especially in conservation policies. Currently, conservation planning in the ACC region is dominated by national programs and consequently displays important discontinuities constrained by the jurisdictional framework (e.g., exclusive economic zones). National Marine Protected Areas have been established around Heard and Mcdonald Islands, Kerguelen Islands, Crozet Islands, and Prince Edward Islands in order to conserve the rich and diverse ecosystems hosted in these oceanic waters. By comparison, the high seas of the southern Agulhas Basin, which are already managed for the fishing activities (Supplementary Figure S2), are currently not considered in conservation plans. However, our observations, along with the numerous foraging hotspots of numerous predators occurring in this latitudinal band, both from Bouvet Island and other subantarctic colonies (Biuw et al., 2010; Hindell et al., 2016; Arthur et al., 2017), highlight the need to consider the subantarctic range as a continuum. Current attempts made under the Commission for the Conservation of Antarctic Marine Living Resources (CCAMLR) aim at characterizing the ecosystems structure and function in the subantarctic region and eventually identifying priority areas for conservation. Therefore, efforts shall address the continuity of this oceanic region and the biophysical processes across the high seas and the multiple economic exclusive zones.