Fine-scale observations reveal distinct frontal phytoplankton communities

Understanding the intricate dynamics of physical and biological seascapes remains a fundamental challenge, particularly because they are coupled over a wide range of spatial and temporal scales¹. Fine-scale physical features (i.e., combined meso- and submeso-scale, 1-100 km, day to weeks), such as fronts, are widespread in the ocean^2,3,4. They create a variable and ephemeral environment for non-swimming microorganisms such as phytoplankton, essential primary producers for marine ecosystems and biogeochemical cycles^5,6.

Phytoplankton life within fine-scale fronts is subject to peculiar dynamics due to sharp density gradients^7,8,9,10 and similar timescales between phytoplankton growth and fine-scale physical forcings¹¹. Remote sensing and modeling studies suggest that phytoplankton dynamics is greatly influenced by fine-scale fronts through a combination of abiotic and biotic factors^6,11,12,13. For instance, stirring by horizontal currents deforms phytoplankton patches^14,15 while vertical velocities induce changes in phytoplankton growth rate by influencing nutrient fluxes^4,16. These two processes generate biological reactions and behavioral responses of higher trophic-level organisms with a feedback on phytoplankton¹⁷ and vice versa. A few in situ studies have demonstrated the influence of fine-scale fronts on biology, particularly their strong structuring effect on shaping phytoplankton communities^{9,18,19,20,21,22}.

However, fine-scale in situ surveys are challenging due to the difficulty of tracking in space and time the physical features involved, and the complex dynamics of both fronts and phytoplankton communities. As a result, past in situ studies primarily relied on measurements conducted across fronts^{9,18,19,20,21,22}, so that high-resolution measurements of phytoplankton properties within fronts (community composition, biomass, growth and loss rates, over diel cycles), along with coupled physical and biological data, remain scarce. Moreover, past analyses focused mostly on productive, dynamic regions dominated by microphytoplankton, where fine-scale biophysical signals are stronger, such as western boundary currents (WBC) and eastern boundary upwelling systems (EBUS)^{9,18,20,23,24,25}. By contrast, the majority of the world’s ocean is oligotrophic²⁶ and characterized by lower energetic levels^27,28 and high abundance of picophytoplankton²⁹. Past studies in the oligotrophic Mediterranean Sea showed that fronts separate communities with different taxonomic composition as well as different growth and loss rates, but limited sampling within the front prevented conclusions about the frontal community itself¹⁹. Due to these limitations, the intricate interactions between fronts and phytoplankton communities, which underpin the ecological response to front structuring, remain poorly understood. This knowledge gap significantly hinders our ability to assess the importance of fine-scale systems at the global scale.

The BioSWOT-Med cruise³⁰, conducted from 21 April to 15 May 2023 (Fig. 1), marks a key step in addressing this gap. The cruise took place in the Mediterranean Sea, characterized by high plankton diversity under oligotrophic and moderately energetic conditions^31,32, and targeted the North-Balearic Front (hereafter NBF), an area marked by hydrological and hydrodynamical contrasts influencing biological production^{33,34,35,36,37,38}. Importantly, an adaptive sampling strategy was specifically designed to provide high-resolution sampling of the fine-scale variability present within this frontal region. The cruise strategy was designed to align with the 90-day fast-sampling phase of the SWOT (Surface Water and Ocean Topography) altimetry satellite mission launched in December 2022. During this phase, the satellite revisited specific areas, including the BioSWOT-Med region, once per day for calibration and validation. The daily SWOT imagery, combined with underway in situ salinity and temperature measurements and drifter trajectories, enabled fine-scale structure targeting and continuous tracking of a fine-scale front.

Prior to the SWOT mission, satellite data for fine-scale studies were either too low in effective resolution (>150 km for classical altimetry³⁹) or hindered by cloud coverage (for ocean color and infrared sea surface temperature). SWOT overcomes these limitations by providing observations of sea surface height unaffected by clouds at high enough resolution to target 15-30 km structures^39,40,41,42. Advances in altimetry, along with the unique opportunities provided by the SWOT fast-sampling phase and adaptive sampling strategy, yielded robust Lagrangian datasets used here to address a specific question: is the phytoplankton community within frontal regions significantly distinct from those in surrounding water masses? To investigate this, we analyzed the fine-scale variability of the frontal region using two complementary Lagrangian datasets, with positions carefully selected based on SWOT and chlorophyll satellite images: (1) adaptive repeated transects across a fine-scale front of the NBF (hereafter “cross-front transect”, April 29-30) and (2) 24-hour drifting stations within the front itself and the two adjacent regions defined by distinct water masses (May 5-6-7) (Fig. 1). Our dataset includes high-resolution underway data of phytoplankton biomass and community composition characterized by flow cytometry and of hydrological properties (temperature, salinity) measured by a thermosalinograph (TSG). The unprecedented resolution of the station dataset enables a robust statistical analysis of the phytoplankton community composition within a fine-scale front.

This study demonstrates that, even in oligotrophic regions characterized by moderate energetic conditions, fine-scale fronts can host, at the surface, a distinct phytoplankton community characterized by significant shifts in the relative composition of phytoplankton functional types. Contrary to productive, energetic regions, these shifts appear to favor less abundant phytoplankton groups independently of their cell size. These results could only be obtained through an adapted sampling strategy, which is essential to representatively sample phytoplankton communities associated with weakly marked fine-scale fronts.

Definition of the frontal region

The North Balearic frontal region separated a northern cyclonic and more productive region from a southern anticyclonic and less productive region⁴³ (Fig. 1b, c, d and Fig. A2) and is defined as a recurrent salinity front between the warmer and fresher Atlantic water spreading from the Algerian Basin and the saltier and colder waters of the Liguro-Provençal basin⁴⁴. We used surface absolute salinity [g/kg] from TSG data collected during the cross-front transect (spatial resolution: 18 m at the nominal speed of 6 knots) to identify a fine-scale front within the NBF.

Using a Gaussian Mixture Model (GMM)^45,46 applied to surface salinity distribution, we identified two distinct water masses “A” (saltier) and “B” (fresher) separated by the front “F” (Fig. 2a). The boundaries of F were defined by salinity values corresponding to the lower and upper limits of the 1σ confidence intervals for the A and B Gaussians, respectively. Region F is characterized by a limited number of measurements when the strategy of cross-front transect was applied, due to the crossing of a fine-scale gradient. By contrast, with the drifting station strategy, the same amount of data was collected within the front during station F2 as at stations A2 and B2. The superimposed histograms of station data show that stations A2, F2, and B2 correctly targeted the different water masses. Data from station F2 reveal a water mass clearly distinct from those observed at the other two stations.

Frontal dynamics

The spatial distribution of A, F, and B regions aligns with fine-scale hydrodynamic features, as demonstrated by Finite Time Lyapunov Exponent (FTLE) patterns derived from altimetric products including SWOT data (Fig. 2b-c and Suppl. Fig. A2). The frontal dynamics were noticeably characterized by intensified eastward displacements (Fig. A2). Figure 2b, c and A2 both clearly illustrate the progress made by SWOT in revealing fine-scale frontal areas, as investigated here, compared to previous Lagrangian diagnostics derived from DUACS 2023 satellite products (Suppl. Fig. A1).

From the displacement of triplets of nearby drifters, we computed strain and vorticity at various locations in the flow field during stations B2 and F2. The F2 region was characterized by both positive vorticity and strong horizontal deformation, with maximal horizontal surface velocities reaching 0.8 m s⁻¹ (Fig. 3a, b). This is consistent with its identification as a frontal area and contrasts sharply with the strain/vorticity estimates in the B region. There, the coherent anticyclonic eddy (B2) was, as expected, characterized by negative relative vorticity and weak deformation, and exhibited maximum horizontal surface velocities of up to 0.6 m s⁻¹.

These meso-/submeso-scale turbulent features and their neighbors exerted a profound influence on the tracer field distribution observed at the surface (see salinity, Fig. 2a) but also the vertical structure of the water column (Fig. 3c, d and Suppl. Table. B1), as revealed by Moving Vessel Profiler (MVP) profiles of temperature, salinity, and fluorescence collected during the cross-front transect (Fig. 3c, d) and nutrient profiles collected during stations. Fluorescence data show a variable subsurface chlorophyll maximum (SCM) ranging from 20 to 80 m along the cross-front transect. In the saltier water mass A, the SCM was shallower than in water mass B. Local deepening and intensification of the SCM signal are observed, for example, around 50, 160 and 300 km along the cross-front transect associated with the front F (red arrows in Fig. 3d). Surface chlorophyll concentrations remained low throughout the section. Vertical nutrient profiles (nitrate and phosphate) measured during stations A2, F2, and B2 displayed near-zero surface concentrations, typical of the oligotrophic conditions of the Mediterranean Sea (Suppl. Table. B1). In the subsurface, water mass A was richer in nutrients than water mass B, with the front F showing intermediate concentrations.

A distinct phytoplankton community at the front is supported by statistical evidence

Automated flow cytometry uniquely enables high-resolution assessments of community structure, with sampling intervals as short as 15 minutes, by grouping cells with similar properties (size, shape, pigment content), assuming comparable functional traits between them. Thus, cytometric clusters have ecological relevance and serve as insightful proxies for phytoplankton functional types⁴⁷. In this study, we optimized the flow cytometer setup to focus on pico- and nano-phytoplankton (<10 µm), which dominate phytoplankton communities in the Mediterranean Sea⁴⁸. Seven phytoplankton groups were identified in real-time: four nanophytoplankton; RedNano (~10³ cells mL^-1), HflrNano, HsNano, HfNano (~10² cells mL^-1), two picoeukaryotes; RedPico (~10³ cells mL^-1), HflrPico (~10² cells mL^-1) and one cyanobacteria; OraPicoProk (sp. Synechococcus) (>10⁴ cells mL^-1) (Suppl. Fig. C1, C2), named according to the standardized cytometric nomenclature⁴⁹. The flow cytometer measures plankton cell size and abundance, converted here into carbon biomass [mmolC m⁻³] using group-specific relationships⁵⁰.

As expected⁵¹, phytoplankton communities were numerically dominated by the small-cell group OraPicoProk (Synechococcus) throughout the region (Suppl. Fig. C2). We designate OraPicoProk, RedPico, and RedNano as “major” groups due to their higher abundances throughout the entire cruise, and HflrNano, HfNano, HsNano, and HflrPico as “minor” groups due to their lower abundances. In terms of biomass, phytoplankton communities were dominated by HsNano (Suppl. Fig. C4), the larger nanophytoplankton.

Surface phytoplankton community composition along the cross-front transect and stations, expressed in biomass percentage, highlights the spatial heterogeneity of communities within the frontal area and their close association with physical patterns (Fig. 2c and Suppl. Fig. C5). Minor groups such as HsNano (Fig. 2c) contributed a higher biomass fraction in F than in A and B along the cross-front transect, and also in F2 compared to A2 and B2.

The cross-front transect provides a spatial overview of the frontal region but lacks the resolution to characterize frontal communities with sufficient accuracy due to limited sampling (26 samples within the front). To overcome this, in the following, we focus on stations A2, F2, and B2, where each water mass was sampled extensively and comparably (A2: 103 samples, F2: 85 samples, B2: 94 samples), allowing a more detailed phytoplankton community analysis beyond the transect data alone.

A thorough analysis of community composition reveals distinct phytoplankton group assemblages between stations (Fig. 4a). At station A2, the community was characterized by higher percentages of RedNano and HflrNano, whereas station B2 was characterized by higher percentages of OraPicoProk, RedPico and HsNano. At station F2, intermediate percentages of specific phytoplankton groups were observed between adjacent water masses, forming what we term “transient” groups. The transient group T1 includes HflrNano, which had higher percentages at station A2 compared to station B2, while the transient group T2 comprises RedPico, which had higher percentages at station B2 than at station A2 (Fig. 4b). The key finding is that the relative biomass fraction of some phytoplankton groups was higher or lower than it was in both A2 and B2 stations. We refer to these groups as “edge-positive” (E + ; HsNano, HfNano, and HflrPico) and “edge-negative” (E-; OraPicoProk and RedNano), respectively (Fig. 4b). E+ is composed of minor groups, while T1, T2 and E+ are composed of major groups (except for HflrNano). The mean absolute biomass [mmolC m⁻³] of most groups was lower at station F2 than expected under passive mixing, except for E+ (Suppl. Fig. C4), suggesting frontal conditions favored E+ over E- and T. The singularity of the phytoplankton community at the front is explored in more detail below.

We used two statistical methods to quantify the distinct features in the communities of stations A2, F2, and B2 using the PRIMER software (Fig. 5a): an analysis of similarity (ANOSIM) to quantify how different the three communities are, and a similarity percentages analysis (SIMPER) to quantify the contribution of each phytoplankton group to these differences. ANOSIM uses the R statistic, which compares dissimilarities, here calculated as Euclidean distances, between samples within a station to dissimilarities between samples from different stations. An R value close to 1 indicates high similarity among samples within a station and strong differentiation from samples at other stations.

The ANOSIM analysis (Fig. 5a, dashed black arrows) demonstrates significant statistical differences (R»0, p-value < 1‰) for each pair of stations (A2-B2, A2-F2, B2-F2), confirming the presence of three distinct community structures, each associated with stations A2, F2 and B2. As expected, pronounced differences existed between stations A2 and B2 (R = 0.70) representing the water masses separated by the front. The difference between stations F2 and A2 (R = 0.70) was greater than between stations F2 and B2 (R = 0.50). The total biomass [mmolC m⁻³] was also more similar between stations F2 and B2 than between stations F2 and A2 (Fig. C3b). On average, distances within stations were much smaller than between stations (Suppl. Table C1), with mean values corresponding to a 59 % increase from within- to between-station distances.

The SIMPER analysis (Fig. 5a, pie charts) shows that various phytoplankton groups contributed to the differences between the three communities. Differences between stations A2 and B2 were primarily driven by RedNano (E-), HflrNano (T1) and RedPico (T2), together explaining 63% of the difference. The E+ groups each contributed less than 15% of the difference between stations A2 and B2, while they dominated differences between A2/B2 and F2. More precisely, differences between stations F2 and A2 were strongly driven by two E+ groups (HsNano, HflrPico) and RedNano (E-), together explaining 65% of the difference. Differences between F2 and B2 were driven by the same two E+ groups and by OraPicoProk (E-) together explaining 52% of the difference. HfNano (E + ) also contributed more than 10% to the difference between F2 and A2 and B2.

Over the past decade, there has been a wide consensus that ocean fronts exert a major role in shaping phytoplankton community structure at the so-called fine-scales⁵². However, in situ studies of this process have been primarily confined to energetic and productive oceanic regions and have been hindered by the complex task of obtaining high-resolution observations of concomitant fine-scale features and plankton dynamics. Our work, to the best of our knowledge, is the first attempt to extend these observations to oligotrophic and moderately energetic current systems, while also addressing the resolution challenge in space and time.

A key ingredient was the design of a multi-platform sampling strategy capable of dealing with the weaker biophysical signals characteristics of these regions as well as with the high spatio-temporal frequencies of fine-scales. This sampling strategy, combining high-resolution underway measurements, drifters, and near-real-time Lagrangian analysis of multi-satellite data, including the high-resolution SWOT altimetric mission, enabled accurate positioning of cross-front transects and stations and provided an extensive biophysical dataset within an ephemeral (~ few weeks) and weakly marked front in the NBF region. This sampling strategy is designed for fine-scale biophysical measurements across and within a front and can be transposed to any other oceanic conditions for observing fine-scale biophysical dynamics.

The synoptic situation encountered during our cruise showcased the typical conditions for front-phytoplankton association, displaying two contrasted water masses separated by a meandering frontal zone, which stayed in place for the entire duration of our cruise. Water masses A, F, and B remained temporally stable and coherent in terms of salinity and biomass throughout both the cross-front transect and the drifting stations (Fig. 2 and Suppl. Fig. C3). Biomass and salinity were systematically higher in A than in B, with intermediate values in F. Spatially, both physical and biological tracers reveal that the front constituted a gradient between two distinct regions, A and B, indicating that F corresponded to a distinct zone between these two water masses (Fig. 2 and Suppl. Fig. C3).

Our results reveal that, beyond representing a physical barrier, the frontal region harbored fine-scale biological variability associated with a distinct phytoplankton community, reflecting unique ecological frontal characteristics under oligotrophic conditions. In particular, no frontal increase in total or group-specific cytometric phytoplankton biomass was observed at the surface (Suppl. Figs. C3 and C4). Instead, distinct relative biomass patterns emerged among the different pico- and nanophytoplankton cytometric groups within the front. Phytoplankton taxa responding to frontal dynamics have previously been classified as “winners” and “losers”⁵³. Here, we propose a framework where losers (E-) groups are major groups while winners (E + ) are minor groups (Fig. 5b). The E+ groups, although they did not exhibit a biomass peak, were the only groups not negatively affected by frontal conditions and can therefore be considered winners (Fig. 4b and Suppl. Fig. C4b). These groups include both pico- and nanophytoplankton, supporting a prior hypothesis that cell size is not a structuring factor within the frontal community in oligotrophic regions⁵⁴.

The analysis above only considered small phytoplankton at the surface, where high-resolution automated flow cytometry data were available. Vertical profiles of cryptophytes and microphytoplankton (diatoms, dinoflagellates), not detected by flow cytometry, confirm that even for these larger cells, biomass did not increase at the surface at station F2 relative to stations A2 and B2 (Suppl. Fig. C6a and Suppl. Note 1). Biomass increased at the DCM relative to the surface for all three stations, particularly for dinoflagellates and RedPico and HflrPico measured by conventional flow cytometry, but remained intermediary at F2 relative to A2 and B2 for most groups (Suppl. Fig. C6). The only group that displayed any increased biomass within the front was diatoms at the DCM, interestingly, a minor group with negligible abundance at A2 and B2 (Suppl. Fig. C6a). These observations suggest that in oligotrophic regions, fronts can act as biological hotspots that locally enhance the biomass of specific phytoplankton groups, such as diatoms. However, in our study, this effect was restricted to the DCM, did not result in an overall phytoplankton biomass increase, and did not extend to surface waters characterized by near-zero nutrient concentrations (Suppl. Table B1). Whether this lack of biomass increase is a typical feature of oligotrophic fronts remains to be determined, as another study found localized increases in vertically-integrated picophytoplankton biomass in a front south of Gran Canaria (Canary Islands), characterized by oligotrophic conditions⁵⁴.

Our observations contrast with in situ studies within energetic and productive regions such as the California Current Ecosystem (CCE) or the Gulf Stream, where fronts were found to be associated with a biomass peak of larger cells, which often lead to an increase in total biomass^9,20,21. Differences between oligotrophic and dynamic regions may be attributed to contrasting underlying mechanisms between the dynamic, nutrient-rich conditions of the CCE and the Gulf Stream and the less dynamic, oligotrophic conditions of the NBF involving different ecosystem structures, particularly longer trophic networks⁵⁵. Within productive and dynamic regions, the chlorophyll-a variability is dominated by large-scale, whereas in oligotrophic and less dynamic regions, the chlorophyll-a variability is dominated by small-scale, potentially due to more complex biological interactions⁵⁶. This suggests that plankton communities’ composition in oligotrophic regions is more sensitive to ecological or biological forcings. Moreover, in oligotrophic regions, fine-scale fronts are moderately to weakly energetic, resulting in limited nutrient input to the euphotic layer, contrary to more energetic regions. This promotes physiological adjustments of phytoplankton rather than biomass accumulation⁵⁷, and can explain why small phytoplankton species dominating in oligotrophic regions appear to be primarily shaped by fine-scale physical drivers rather than by nutrient inputs^54,58.

Gaining insight into the processes through which phytoplankton composition is modified at fronts, including in oligotrophic and moderately energetic regions, could potentially be a rich source of general knowledge on plankton dynamics. A conceptual framework is already in place that distinguishes passive, active, and reactive scenarios (i.e., respectively, spatial redistribution of biogeochemical properties by advection; physics affecting phytoplankton growth rates via modulations of nutrient enrichment or light availability; and more complex ecological and biogeochemical feedbacks mediated by physics)⁶. The relevance of these scenarios to BioSWOT-Med is briefly examined in sequence.

Regarding passive transport, a water mass with a distinct salinity signature was observed within the front on the western side of the cross-front transect (Suppl. Fig. C3a). FTLE patterns (Fig. 2b, c and Suppl. Fig. A2) and convergent and rapid drifter trajectories associated with strong horizontal deformation (Fig. 3a, b) suggest this water filament was trapped in the eastward frontal current. This implies that the frontal community, with a higher proportion of E + , may have originated from a distinct water source advected by the convergent flow. Another possible mechanism related to passive transport within the front is that the E+ group may possess physiological traits allowing it to use buoyancy to remain near the surface despite subduction, whereas the E- group may have been advected downward by frontal dynamics, similar to what has been suggested for Synechococcus and certain eukaryotic phytoplankton within the submesoscale front south of Gran Canaria⁵⁴.

Active processes are unlikely to have acted in isolation since no increase in nutrient concentration, nor in total, nor in E+ cytometric biomass was observed at the surface. However, various possible forms of interplay between active and reactive factors could have contributed to producing advantageous (resp. disadvantageous) conditions for E+ (resp. E-) groups at station F2, as discussed below. To conclude, the most likely underlying mechanisms are related to intensified horizontal physical forcings that shape flow heterogeneities and, consequently, passively influence biogeochemical fields and phytoplankton communities. Ecological interactions may then act secondarily, reactively shaping the community into a distinct frontal assemblage, given that distinct zooplankton communities were also observed during the cruise⁵⁹.

The considerations above imply that E+ species may be selected based on physiological traits suited better than other groups to the local frontal conditions, whether via grazing resistance or via growth optimization. As such, the higher contribution of E+ groups in the front could potentially be explained by distinct ecological niche strategies. Generalists are widely distributed across oceanic areas, while specialists are spatially confined⁶⁰. Applying this framework to our study, a hypothesis could be that E+ groups may behave as generalists with broad niches, can tolerate environmental changes and thrive under frontal conditions, whereas E- or T groups may behave as specialists adapted to the oligotrophic background and are disadvantaged within the front. As an example, the vertical circulation of fronts reduces daily light exposure compared to adjacent stable waters, favoring phytoplankton able to rapidly adjust their light-harvesting physiology⁵⁷, a capacity possibly found in species with broad ecological tolerance, such as generalists. This hypothesis aligns with studies showing that non-dominant species (e.g., E + ) are mainly shaped by stochastic dispersal (e.g., currents), whereas dominant species (e.g., E− and T) reflect physiological adaptation to local conditions and thus display strong bioregionalization^61,62). In this context, fine-scale fronts may disadvantage E− and T species whose traits are poorly suited to frontal conditions, while E+ species are less constrained by them. Additional information beyond flow cytometry, such as from genomic analyses (upcoming), is required to identify E+ species and assess their physiological traits.

Thanks to the dedicated sampling strategy guided by the first SWOT images, this work demonstrates the existence of a distinct frontal phytoplankton community associated with a fine-scale front of the North Balearic Frontal region. Our results extend our understanding of the specificity of phytoplankton communities at ephemeral fronts and on their ability to “live on the edge” by providing an investigation of front-phytoplankton association in moderately energetic and oligotrophic conditions, largely neglected in the past but characteristic of a substantial part of the world ocean. The main finding of this study is the demonstration, based on an extensive dataset, that in this type of region, fronts do not increase phytoplankton biomass but rather restructure community composition by enhancing the contribution of less abundant groups, potentially acting as refuges. This highlights an important role of the fine-scale variability in sustaining the heterogeneity of phytoplankton communities, in line with a statistical modeling approach previously used to reveal frontal phytoplankton communities under sparse sampling⁶³. Adding information on vertical processes, nutrient fluxes, growth and grazing rates, plankton species taxonomy, and a Lagrangian modeling framework⁶⁴ into the sampling strategy developed here will help disentangle passive/active/reactive scenarios in contrasted regions of the world ocean.

Cruise strategy

The BioSWOT-Med cruise³⁰ took place from April 21 to May 15, 2023, aboard the R/V L’Atalante, in the North Balearic Islands region of the northwestern Mediterranean Sea (Fig. 1). The timing, location, and strategy of the cruise were carefully designed to leverage data from the novel SWOT satellite mission to better resolve fine-scale oceanic features. Specifically, the BioSWOT-Med cruise was scheduled to match the SWOT “fast sampling phase”, characterized by high spatial resolution and a 1-day revisit period over ~150 km-wide oceanic regions, in the framework of the international SWOT AdAc (Adopt-A-Crossover, https://www.swot-adac.org/) Consortium. An adaptive multidisciplinary approach was employed by combining daily SWOT images and environmental bulletins provided daily by the SPASSO toolbox⁶⁵ (https://spasso.mio.osupytheas.fr/), and in situ measurements using a suite of instruments to capture physical, biological, and chemical properties. This strategy enabled the targeting of fine-scale features within the recurrent NBF. Continuous underway systems (here, TSG, MVP and automated flow cytometer) revealed a weak salinity gradient of 0.2-0.3 g/kg across a few kilometers associated with distinct biological patterns (Figs. 2 and 3). Surface drifters deployed along the converging front, in close spatial agreement with the front derived by SWOT maps, revealed an eastward frontal jet characterized by an average speed of 30 cm/s and local intensifications up to approximately 70 cm/s (Suppl. Fig. 3a).

Satellite products for front detection

Circulation patterns of the NBF are evidenced using three different types of satellite products. The European Seas DUACS Level-4 gridded product is used as a reference for conventional multi-mission altimetry-derived ADT over the western Mediterranean Sea (https://doi.org/10.48670/moi-00141). For this work, the 2023 version of the DUACS product is used (i.e., not including SWOT nadir) to evidence the contribution of the SWOT mission over traditional (pre-SWOT) multi-mission altimetry products. The 2-km ADT derived from SWOT observations allows for the precise identification of a fine-scale front location. Version v1.0.2 of the SWOT Level-3 product is used in this study. The SWOT_L3_LR_SSH product, derived from the L2 SWOT KaRIn low-rate ocean data products (NASA/JPL and CNES), is produced and made freely available by AVISO and DUACS teams as part of the DESMOS Science Team project. AVISO/DUACS, 2024. SWOT Level-3 KaRIn Low Rate ADT Expert (v1.0.2) [Data set]. CNES. https://doi.org/10.24400/527896/A01-2023.018). To extend high-resolution SWOT observations beyond SWOT swaths, we used a joint reconstruction of gridded ADT over the western Mediterranean Sea using all available altimetry missions, including SWOT data (hereafter referred to as the VarDyn product⁶⁶). VarDyn was chosen as the best representation of fine-scale circulation in the studied area and was used to compute Lyapunov exponents, a Lagrangian diagnostic detecting frontal areas¹¹. The Lagrangian Manifolds and Trajectories Analyzer (LAMTA) computed thousands of numerical particle trajectories advected backward-in-time for 15 days to derive Finite Time Lyapunov Exponents (FTLEs) and advected longitude both with DUACS and VarDyn velocity fields⁶⁵. Ocean color data from Sentinel-3 product distributed by EUMETSAT (OLCI Level 2 CHL Concentration – Sentinel-3A: https://navigator.eumetsat.int/product/EO:EUM:DAT:0178) is also used to identify the front on May 5.

A, F, B identification

The TSG continuously measured seawater drawn from the ship’s intake at a depth of 4.23 m below the waterline (time resolution: 6 s). High-resolution underway TSG data show strong horizontal salinity patterns between the north and south parts of the front. To accurately and objectively identify the boundaries of the frontal area, we used a Gaussian Mixture Model (GMM) on the cross-front transect dataset. GMM is a probabilistic approach for describing and classifying data. It attempts to fit data as a linear combination of multidimensional Gaussian distributions with unknown means and unknown standard deviations⁶⁷. (X) is the array of (n) samples (here, (n=)26,229), and (p(X)) is the probability distribution function (PDF) representing the entire dataset. GMM provides the PDF as a weighted sum of (K) Gaussian classes, indexed by k; that is,

Here (N({X;}{mu }_{k},{varSigma }_{k})) is the multidimensional Gaussian PDF with a vector of means ({mu }_{k}) and covariance matrix ({varSigma }_{k}), ({lambda }_{k}) represents the probability associated with class (k). In our case, we supposed (k=2), for A and B water masses. The Expectation-Maximization (EM) algorithm is employed to estimate the unknown means and unknown standard deviations of Gaussian distributions. In the E-step, the algorithm computes the probability that each data point belongs to each Gaussian component, while in the M-step, it maximizes the likelihood function by adjusting the parameters of the Gaussians. This algorithm is repeated until the convergence of the solution. We estimated boundaries of the front F as the position of isohalines corresponding to the lower and upper limits of the 1σ confidence intervals for the A and B gaussians, respectively.

Drifter data for front tracking

Drifters are Lagrangian oceanographic instruments freely advected by sea currents, designed to float at a specific depth and regularly transmit their GPS position, therefore giving a direct measurement of passive transport driven by ocean dynamics. During BioSWOT-Med, surface drifters^68,69,70 were deployed within distinct water masses, identified through satellite products and TSG transects, in order to guide the adaptive sampling strategy of the cruise. Here, we focus on the drifters deployed within the front, therefore supporting the localization of region F and station F2, and consequently the sampling of the surrounding water masses.

Figure 3a highlights the presence of the front by following the advection of drifters deployed on April 23-26 (first experimental phase, 5 westernmost drifters). During the cross-front transect (29-30 April) and stations A2 (7 May), F2 (6 May) and B2 (5 May), we released additional drifters to track the water mass advection path and keep the vessel on target (second experimental phase, 7 easternmost drifters). There is a general continuity and agreement of the drifter trajectories deployed in the two phases, covering continuous areas, providing also information on the scale of the frontal structure. From the drifter’s trajectories, we calculated vorticity, which describes the rotational tendency of fluid particles, and strain, which characterizes fluid deformation through stretching and shearing, following the method of Molinari et al. (1975)⁷¹.

Automated flow cytometry

An automated CytoSense flow cytometer (CytoBuoy b.v.) was connected to the seawater circuit of the TSG to perform scheduled automated sampling and analysis of phytoplankton at high frequency during the cross-front transect and stations, providing us with real-time, detailed information on the phytoplankton community structure, which helped us identify the frontal region. Three distinct analyses using the same instrument set-up (protocol) have been run sequentially every 15 min. The protocol named FLR8 had a trigger threshold fixed at 8 mV on the red fluorescence signal (FLR8) and analyzed a volume of 0.5 cm³. It was dedicated to the analysis of the smaller phytoplankton (picoplankton and nanophytoplankton), which are dominant in the Mediterranean Sea. We did not take into consideration the other protocol (FLR25) which is used to run to detect larger cells such as microphytoplankton for two main reasons: first, the analyses performed at the beginning of the cruise confirmed that the oligotrophic Mediterranean ecosystem is predominantly characterized by small phytoplankton (pico- and nanoplankton); second, flow cytometry has inherent limitations in accurately detecting larger cells when its occurrence is too low in the small volume analyzed^72,73.

The data were acquired by using the USB software (Cytobuoy b.v.) and were analyzed with the CytoClus 4 software (Cytobuoy b.v.). We determined and measured the abundance of 7 phytoplankton clusters named following the standardized cytometric nomenclature⁴⁹ except for those written in quotes: HsNano, “HfNano”, RedNano, “HflrNano”, RedPico, “HflrPico”, OraPicoProk (Synechococcus) (Fig. C1).

Biomass estimations

Phytoplankton abundances measured by flow cytometry were converted into terms of carbon biomass. Cell carbon content (Qc [mmolC  cell^-1]) can be estimated from cell biovolume (corresponding to the 3D space occupied by a cell) according to Eq. 1a⁷⁴, enabling measured abundances [cell m^-3] to be converted into biomass [mmolC m^-3] according to Eq. 1 d. Coefficients ({alpha }_{0}) and ({alpha }_{1}) are identical to those previously applied to a similar dataset¹³.

Assuming the cells are spherical, the cell diameter (equivalent spherical diameter, ESD [µm]) and the biovolume (BioV [µm³]) were estimated from the FWS signal according to Eq. 1b and Eq. 1c^75,76,77 using an empirical relationship calibrated with silica beads of known sizes. The log-log regression coefficients β₀ and β₁ correspond to the intercept and slope. This relationship between bead diameter and FWS signal improves when beads are grouped by size class, and regressions are computed separately⁵⁰. Here, this yields two sets of β₀ and β₁ coefficients: for smaller cells (OraPicoProk, HflrPico, RedPico: -1.0484 and 0.2641), and larger cells (HfNano, RedNano, HflrNano, HsNano: -8.5691 and 0.9553).

Statistical analysis of cytometric samples

A total of 451 cytometric samples were collected during the cross-front transect and at the stations. The front is under-represented in the cross-front transect dataset due to its small spatial footprint: 26 samples vs 79 for A and 66 for B. As a consequence, to ensure the robustness of the dataset, only the station data are used in statistical analysis. The number of samples per station included in the statistical analyses is as follows: 103 for station A2, 85 for station F2 (more than three times more than for F), and 94 for station B2.

We conducted a Mann-Whitney test to statistically evaluate differences in the biomass proportions of phytoplankton groups between stations. Specifically, we aimed to determine whether significant differences exist at station F2 compared to other stations. The Mann-Whitney test is a non-parametric statistical method used to assess whether two independent samples come from distributions with different medians. It evaluates the probability that a randomly chosen value from one sample (X) is greater than a randomly chosen value from the other sample (Y). In our case, X and Y represent the biomass proportions of a given phytoplankton group i at two stations (e.g., station x versus station y). We performed 18 tests in total: 7 phytoplankton groups × 2 station comparisons (F2 vs. A2 and F2 vs. B2). Additionally, we conducted the same analysis to compare transient (T1, T2) and edge (E + , E-) groups within each station. This resulted in 8 additional tests: 4 groups × 2 station comparisons (F2 vs. A2 and F2 vs. B2).

We also conducted two other statistical analyses using PRIMER 8 (https://www.primer-e.com/software) to investigate the community structures in A, F and B regions. For these analyses, we used normalized biomass proportions for each phytoplankton group to minimize bias due to significant differences between phytoplankton groups. These statistical analyses are based on the estimation of distances between pairs of samples and therefore require a sufficiently large number of samples within each water mass (A, F, and B), as well as comparable sample sizes among the three water masses. The key contribution of the sampling strategy implemented during BioSWOT-Med was to provide a large and well-balanced set of biological observations at stations A2, F2, and B2, thereby allowing the statistical analyses to be applied reliably.

First, ANOSIM analysis⁷⁸ was conducted to statistically test the existence of the A, F, and B communities. ANOSIM evaluates differences between categories (A2, F2, B2) by calculating the (R) statistic, which represents the scaled difference between average dissimilarity ranks within and between categories,

({{r}_{B}}_{_}) and ({{r}_{W}}_{_}) represent the average rank similarities between and within categories, respectively, calculated here using Euclidean distance matrices, with distances calculated between all pairs of samples. (N) is the total number of samples. An (R) value near zero indicates no group differences, while higher (R) values signify greater differences between categories.

Second, SIMPER analysis⁷⁹ was conducted to identify which phytoplankton groups are responsible for differences between A, F and B communities. SIMPER evaluates the relative contribution of species to dissimilarities between categories (A2, F2, B2) by calculating the percentage of contribution ({delta }_{{jk}}(i)) of ith species to Eulerian distance dissimilarity between one pair of samples, j and k,

This calculation is repeated for each pair (all combinations of j and k from A2, F2, and B2), and the contribution of the ith species to overall dissimilarity is represented as the mean of all ({delta }_{{jk}}(i)).