Large-to submesoscale surface circulation and its implications on biogeochemical / biological horizontal distributions during the OUTPACE cruise ( southwest Pacific )

The patterns of the large-scale, mesoand submesoscale surface circulation on biogeochemical and biological distributions are examined in the western tropical South Pacific (WTSP) in the context of the OUTPACE cruise (February–April 2015). Multi-disciplinary original in situ observations were achieved along a zonal transect through the WTSP and their analysis was coupled with satellite data. The use of Lagrangian diagnostics allows for the identification of water mass pathways, mesoscale structures, and submesoscale features such as fronts. In particular, we confirmed the existence of a global wind-driven southward circulation of surface waters in the entire WTSP, using a new high-resolution altimetry-derived product, validated by in situ drifters, that includes cyclogeostrophy and Ekman components with geostrophy. The mesoscale activity is shown to be responsible for counter-intuitive water mass trajectories in two subregions: (i) the Coral Sea, with surface exchanges between the North Vanuatu Jet and the North Caledonian Jet, and (ii) around 170W, with an eastward pathway, whereas a westward general direction dominates. Fronts and small-scale features, detected with finite-size Lyapunov exponents (FSLEs), are correlated with 25 % of surface tracer gradients, which reveals the significance of such structures in the generation of submesoscale surface gradients. Additionally, two high-frequency sampling transects of biogeochemical parameters and microorganism abundances demonstrate the influence of fronts in controlling the spatial distribution of bacteria and phytoplankton, and as a consequence the microbial community structure. All circulation scales play an important role that has to be taken into account not only when analysing the data from OUTPACE but also, more generally, for understanding the global distribution of biogeochemical components.

samples. Because the Prochlorococcus population cannot be uniquely distinguished in the SYBR stained surface samples, bacteria were determined as the difference between the total cell numbers of the SYBR stained sample and Prochlorococcus enumerated in unstained samples. Cytograms were analyzed using FCS Express 6 Flow Cytometry Software (De Novo Software, CA, US). These data are used to investigate the small-scale distribution of the different community groups and its relation with the concomitant dynamics at submesoscale.

Satellite data
Several satellite datasets were exploited during the campaign to guide the cruise through an adaptive sampling strategy using the SPASSO software package (http://www.mio.univ-amu.fr/SPASSO/) following the same approach of previous cruises such as LATEX (Doglioli, 2013;Petrenko et al., 2017) and KEOPS2 (d'Ovidio et al., 2015). SPASSO was also used after the cruise in order to extend the spatial and temporal vision of the in situ observations. 10 Four different altimetry-derived velocity products were tested in this study to choose the product that best represents the surface circulation during the cruise. First the daily Ssalto/Duacs product (Ducet et al., 2000), from AVISO (Archiving, Validation and Interpretation of Satellite Oceanographic 3) data base, for the period from 1 January 2004 to 31 December 2015, was used to extract daily delayed-time maps of absolute geostrophic velocities (1/4 • x 1/4 • on a Mercator grid, since 15 April 15 2014). Three other altimetry products were specifically produced, for the first time, at 1/8 • resolution for the WTSP region by Ssalto/Duacs and CLS (Collecte Localisation Satellites), with support from CNES (Centre National d'Études Spatiales), from June 2014 to June 2015. In particular, they provided daily maps of absolute geostrophic velocities, daily maps of the sum of absolute geostrophic velocities and Ekman components, and the same product as the latter that also includes a cyclogeostrophy correction (Penven et al., 2014). Ekman surface currents refer to the wind-induced circulation relative to 15 m and are com- 20 puted from ECMWF ERA INTERIM windstress with an Ekman model fitted onto drifting buoys (Rio et al., 2014).
A preliminary comparison between Lagrangian numerical particle trajectories (see Section 2.3.1), computed with each of the above products, and the observed trajectories of the different floats launched during OUTPACE (section 2.1) allowed us to identify the product including geostrophic and Ekman components with the cyclogeostrophy correction (hereafter total altimetry-derived velocity field) as the most accurate in situ surface currents (see figure A1 for a comparison between float 25 trajectories and numerical particle derived streamfunction).
Daily near-real-time maps of sea surface temperature (SST) and ocean color were also specifically produced for the WTSP from December 2014 to May 2015 by CLS with support from CNES. They are constructed with a simple weighted data average over the 5 previous days (giving more weight to the most recent data), and have a 1/50 • resolution (2 km at the Equator) in latitude 30 and longitude. The temperature product corresponds to maps of SST deduced from a combination of several intercalibrated infrared sensors (AQUA/MODIS, TERRA/MODIS, METOP-A/AVHRR, METOP-B/AVHRR). The ocean color product corresponds to maps of chlorophyll concentration issued from the Suomi/NPP/VIIRS sensor (http://npp.gsfc.nasa.gov/viirs.html).
These satellite data are compared with in situ data from the underway survey. A correlation of 0.8 between in situ mea-5 Biogeosciences Discuss., https://doi.org /10.5194/bg-2017-456 Manuscript under review for journal Biogeosciences Discussion started: 3 November 2017 c Author(s) 2017. CC BY 4.0 License. surements and co-located satellite data validates the satellite-derived SST and chlorophyll concentration. A supplementary correlation with the daily High-Resolution SST blended from NCDC/NOAA (Reynolds et al., 2007) and in situ SST showed a similar correlation. These results corroborate the accuracy of the CLS products in our region of interest.

Lagrangian diagnostics
2.3.1 Surface water mass pathways detection 5 To investigate the water mass movements at large and meso-scale, we used the Lagrangian diagnostic tool Ariane that can trace water mass movements from the trajectories of numerical particles that enter and exit a predefined domain (Blanke and Raynaud, 1997;Blanke et al., 1999). In this study the numerical particle trajectories are computed with altimetry-derived surface currents from the products listed above (see section 2.2). As many Lagrangian particles as desired can be integrated in two different ways: backward in time to assess the origins of the water masses or forward in time to investigate their fate. Addi-10 tionally, this Lagrangian tool allows for the computation of two different diagnostics : i) qualitative diagnostics that compute typically few particles with a steady recording of the positions along their trajectories; ii) quantitative diagnostics that compute thousands of particles with statistics available for initial and final positions, and with the diagnostic of the main pathways. In this study we use 3 different configurations of the Ariane tool depending on the objectives.
First, to identify the altimeter product that best fit the observed trajectories of the floats launched during OUTPACE, a compar- 15 ison was done with the trajectories of numerical particles (typically considering release of several hundred of particles). They were initially positioned around the launching position of the floats with a resolution of 1-2 km. The particles were advected forward in time for 96 (LDA), 78 (LDB) and 70 (LDC) days, corresponding to the time lapse between the launch day of the floats and the last available day of satellite data. These qualitative experiments allows for the comparison of successive positions (every 6 hours) of numerical particles computed with the 4 different products and those of the floats. Thus the choice of 20 the best surface velocity product relied on the best fit between observed and numerical trajectories. The total altimetry-derived velocity field (i.e. the product including geostrophy, cyclogeostrophy and Ekman components) will be used for all diagnostic computations, as it best fits in situ data (Fig.A1).
To study the large scale circulation in the WTSP, quantitative experiments were performed to find the main pathways of the waters entering and exiting the box contouring the WTSP (Fig. 2). Particles were launched along each section of the box 25 (North, East, South, West) and advected forward with the total altimetry-derived surface currents. We simulated ten years of particle trajectories by repeating ten times the available dataset. We compare the results of this simulation with the ten years integration of available geostrophic AVISO surface currents over this time period. The comparison between these simulations ensures that the use of a one year time period looped several times does not significantly modify statistical outputs. It is clearly more interesting to use CLS data instead of commonly used AVISO geostrophic surface currents because they include the wind 30 effect and cyclogeostrophy with higher resolution, as well as better represents in situ data.
Another objective is to identify the mesoscale trajectories of surface water masses sampled at each of the LD stations. Backward and forward quantitative experiments were performed to identify the main pathways of the surface water masses arriving 6 Biogeosciences Discuss., https://doi.org /10.5194/bg-2017-456 Manuscript under review for journal Biogeosciences Discussion started: 3 November 2017 c Author(s) 2017. CC BY 4.0 License. and leaving each LD station using total altimetry-derived surface currents (Figure 4). Almost one million numerical particles were initially distributed along a square box surrounding the position of the LD station. The calculations were stopped whenever the particles return to the initial box (hereafter called meanders) or were intercepted on one of the four remote sections located around the LD station. The boxes size are tuned in order to minimize meanders and the loss of particles in the domain.
Percentage of both quantities are reported in Table A1. Forward computation times are identical to the qualitative diagnostics.

5
For backward experiments, the particles were advected for 183 (LDA), 201 (LDB) and 209 (LDC) days corresponding to the maximum time lapse allowed by CLS satellite data availability.

Eddies and filaments identification
To set up the mesoscale context during the OUTPACE cruise, we used the Lagrangian Averaged Vorticity Deviation method Haller et al., 2016) that allows identification of coherent structures from altimetry-derived 10 surface velocity fields (code available at https://github.com/Hadjighasem/Lagrangian-Averaged-Vorticity-Deviation-LAVD).
The detected features are able to trap water masses for a certain period (defined by the integration time) and transport them along their route. In this study we computed the detection with the total altimetry-derived velocity field and chose an 8 day time integration with respect to the duration of LD stations. Indeed this time interval provides a confirmation or rebuttal of the assumption that LD stations have been performed in a coherent structure, as targeted during the cruise. This Lagrangian 15 diagnostic is also compared with a hybrid Lagrangian and Eulerian approach combining the calculation of the Okubo-Weiss (OW) parameter and a retention parameter (RP), computed with the same velocity field. The OW parameter identifies structures such as eddies by separating the flow into a vorticity-dominated region and a strain-dominated region (Okubo, 1970;Weiss, 1981). The RP identifies the number of days a fluid parcel remains trapped within a structure core, defined by a negative OW parameter. Both parameter calculations are detailed in d' Ovidio et al. (2013). As for the LAVD detection method, the RP allows 20 for the identification of potentially trapping coherent structures.
Submesoscale flow features in two-dimensional data are evaluated with altimetry-derived finite size Lyapunov exponents (FSLE), computed with the algorithm of d 'Ovidio et al. (2004). This Lagrangian diagnostic detects frontal zone on which passive elements of the flow should theoretically align. Here we used the total altimetry-derived velocity field to compute 25 the algorithm. The main parameter values for the algorithm are described in de Verneil et al. (2017b). The OUTPACE cruise occurred in the relatively open ocean and far enough from islands to trust the FSLE diagnostic calculated from altimetry.
We compare the horizontal positions of the fronts, detected with FSLE, with surface gradients measured both with the TSG (temperature and salinity) and the underway survey (chlorophyll). Indeed these high frequency samplings provide access to submesoscale gradients. A point by point correlation is calculated whenever a strong gradient of density (or chlorophyll) corre-30 sponds or not to a high FSLE value (i.e. > 0.05 day −1 ) indicative of a front. Sensitivity tests have been performed to choose the thresholds on density (chlorophyll) gradients to ensure the stability of the correlations calculated. These tests ended with the selection of gradient larger than 0.1 kg m −3 km −1 and a 0.2 mg m −3 km −1 thresholds for density and chlorophyll gradients, respectively.

Large scale wind-driven pathways
The geostrophic large scale mean circulation and directions of the main currents in the WTSP are well established from the literature. However, the trajectories and pathways of surface waters may change and be more complex when the effect of the wind is added and resolution is increased especially in the context of inter-annual ENSO (El Niño Southern Oscillation) vari-5 ability. We decided to use a Lagrangian integration of numerical particles to simulate the transport of surface fluid parcels at the scale of the WTSP region using altimetry-derived ocean currents including the wind effect. Figure 2 shows the transport calculated from the sum of almost 13 million numerical particles advected with the total surface altimetry-derived flow for 10 years (see Section 2.3.1). This figure highlights the westward transport of the SEC in the northwestern part of the domain, but also the eastward transport at 10 • S due to the output of the South Equatorial Counter Current (SECC) from the Solomon Sea 10 (Ganachaud et al., 2014). Both these pathways follow the well-known circulation of the SEC and SECC in this region. Additionally, a clear surface meridional transport is noticeable from 10 • S to 25 • S. In the very southeastern part of the domain some surface waters seem to recirculate to the east from 170 • W, which is probably an indicator of the gyre circulation or an artefact of mesoscale dynamics due to the short averaging period considered by the present study. The meridional transport does not correspond to any surface current but is mainly due to the south-easterly trade winds. This meridional component appears 15 due to the addition of the Ekman component in altimetry surface velocities (Fig.B1, Supplementary Material). The large scale transport of surface waters in the OUTPACE area is thus a combination of the transport by general well-known surface currents and wind-driven circulations. Surface waters globally travel the WTSP from northeast to southwest. This meridional transport seems to separate the gyre surface waters that will recirculate towards the east near 175 • W and waters that will continue to be advected in the WTSP towards the Coral Sea. At the scale of the WTSP, the individual transport calculated from each initial 20 section (see Section 2.3.1) reveals that 80% of the surface waters crossed during the OUTPACE cruise originate from the «North» section, 8-15% from the «East» section and very few from the «South» and «West» sections (Fig. C1). We can thus identify a general wind-driven surface transport in the WTSP as follows: the surface waters enter the WTSP from the northeast with the SEC and are gradually advected to the south with a part (east of 170 • W) that directly recirculates within the gyre and another part (west of 170 • W) that follows a southwestern propagation through the different WTSP archipelagos. From a 25 biogeochemical point of view, two types of waters can be differentiated: the relatively oligotrophic but richer Melanesian waters (from 160 • E to 170 • W) and the ultra-oligotrophic gyre waters (west of 170 • W) (Fumenia, personal communication).The wind-driven surface transport highlighted here could participate in the biogeochemical variations between western and eastern waters. Indeed the path through the Melanesian area may enrich these waters due to the contact with multiple islands whereas waters that directly recirculate within the gyre keep their ultra-oligotrophic characteristics. This pathway induces complex re-30 circulation through the Melanesian archipelago, not visible on the large scale transport.
The WTSP circulation is also strongly impacted by ENSO conditions, responsible for SST variability on inter-annual to decadal timescales (Sarmiento and Gruber, 2006). A negative Southern Oscillation Index (SOI), El Niño phase, is characterized The mean wind velocity measured during OUTPACE is shown to be close to mean velocities during El Niño (data not shown) and Moutin et al. (2017) clearly showed that the OUTPACE cruise took place during an El Niño phase but they determined that climatological effects, upon the results of the cruise, were minimized. Otherwise, as this region is characterized by an intense mesoscale circulation, we can expect that it participates in the biogeochemical variations in the region. Thus we investigate the 5 mesoscale feature trajectories on the entire WTSP and in particular the mesoscale circulation around three biogeochemically different locations: i) LDA located in the Coral Sea, at the end of the surface waters' journey across the WTSP; ii) LDB, in the Melanesian waters, west of the transition zone with gyre waters; iii) LDC, in the gyre waters.

Mesoscale activity and trajectories of surface waters
The major goal in this section is to identify whether mesoscale activity, and in particular trapping features, actively participate 10 in the transport of different water mass properties across the WTSP ocean. The LAVD method is chosen in order to track the coherent structures for a time period of 8 days (see Section 2.3.2). Figure 3 shows the total altimetry-derived velocity field and the mesoscale structures identified for the first day of each LD station (LDA, LDB and LDC). It reveals that many mesoscale structures are found in the entire WTSP with no specific region with higher abundances of these features. A comparison with the eulerian Okubo-Weiss (OW) parameter method shows a good agreement with the LAVD detection method. Indeed, a mean 15 OW parameter value of -0.24 day −2 is calculated inside the contour of coherent structures detected with the LAVD method.
It indicates that wherever a coherent structure is identified, the OW parameter also identifies a mesoscale feature. Most of the mesoscale structures detected with the LAVD method are identified as retention areas, with the RP, ensuring the trapping characteristics of these features (Fig. C2). A tracking of these coherent structures highlights that they all show a general zonal propagation as expected from the mean transport in this area. If the major part propagates westward, the meridional band be- Vanuatu, in relatively highly productive waters (Fig.5). Far to the west, LDB lies in the eastward eddy-propagation band inside a phytoplankton bloom whereas LDC is located in nutrient-poor waters in the South Pacific (SP) gyre (Fig.5). Figure 4 shows the streamfunctions calculated from numerical particle advection using total altimetry-derived surface velocities (see Section 2.3.1). They represent the origin (Fig. 4 top) and the fate (Fig.4 bottom) of each LD stations' waters, respectively the backward 5 and forward computations. One would expect LDA surface waters to come from the East as it is located on the path of the westward NCJ. However they seem to have multiple origins: i) easterly, directly from the NCJ transport; ii) northerly, directly from waters that have circulated between the Vanuatu islands before heading south to LDA; and iii) from a meridional tortuous recirculation path (∼162 • E) within the Coral Sea. Both i) and ii) origins agree with the integrated transport entering the Coral Sea induced by the complex topography (Kessler and Cravatte, 2013). The meridional recirculation determined here suggests 10 that eddy-eddy interactions might be responsible for the emergence of complex paths between the NVJ and the NCJ. Indeed it matches the area described by Rousselet et al. (2016) as the region of exchange between NCJ and NVJ waters through eddy trapping and transport. We identify a probable water mass mixing area, in the Coral Sea center, through complex mesoscale stirring. This stirring may also create surface gradients as depicted by Maes et al. (2013). After LDA sampling, an intense signature of the NCJ is detected at the surface. Another portion of surface waters directly crash on New Caledonia's northern coast 15 but this pathway might not be so relevant due to the satellite's lack of resolution near coastal areas. The westward circulation scheme is consistent with previous studies focusing on the NCJ (Ganachaud et al., 2008;Gasparin et al., 2011) and the results reported by Barbot et al. (2017).
In the eastern WTSP, LDB surface waters seem to follow the same general path: they flow from northeast towards southwest before they reach a group of islands and then recirculate to the east towards LDB. After LDB sampling, they continue their way 20 to the east before heading back to the south or to the south-west. The eastward propagation of mesoscale structures, mentioned above, is sustained by the eastward circulation of surface waters around the LDB site (Fig. 4). de Verneil et al. (2017b) also pointed out a possible eastward transport to explain the origin of the bloom sampled at LDB. Indeed the surface waters might be iron-enriched through contact with the islands and thus create favorable conditions for a phytoplankton bloom. At this site, adjacent to the nutrient-poor SP gyre, the biological dynamics could be specially enhanced by the mesoscale eastward transport 25 of essential chemical supplies for phytoplankton development. If this eastward transport is revealed to be quasi-permanent, it could be associated with recurrent bloom events in this area but this assumption requires further analyses to be generalized.
Further west, as one would expect, the waters sampled during LDC travelled from the east and flow to the west after LDC (Fig.4). We notice a recirculation area east of LDC where the waters seemed to follow a looping trajectory before reaching LDC. Both lagrangian methods (LAVD detection and advection of numerical particles) agree that a mesoscale structure trapped 30 water masses east of LDC (from 164 • W) and transported them until the sampling region of LDC. Indeed, the high rates of meanders (70% and 44% for backward and forward integration, respectively) (Table A1), also suggest the waters were still trapped in the structure core during LDC sampling. This observation is in good agreement with de Verneil et al. (2017a)

Fine scale distribution of tracers
The surface tracers' distribution is mostly driven by the wind-induced circulation but also by the transport through mesoscale activity. In a more ephemeral way, tracers can be dispersed following small-scale perturbations such as frontal features. Here we try to detect and quantify the influence of such features on the density and chlorophyll surface gradients using both in situ and satellite observations. 5 As described in Section 2.2 and shown by Figures 5a and 5b, the comparison between in situ observations and satellitederived data results in reasonable correlations. As both datasets are comparable, SST and chlorophyll concentrations from remote sensing are then used to investigate horizontal gradients in the WTSP (Fig. 5a and Fig. 5b). To assess the spatial scale of submesoscale gradients (typical of 1-10 km) in terms of ocean dynamics, we use a lagrangian methodology based on the calculation of FSLE (see Section 2.3.2) that allow for fronts detection. Figure 5c shows regions where fronts are frequently 10 generated during the time period of the cruise. We notice that the gyre is a region less suitable for fronts to occur and persist. We also identify east of the Fiji islands a zonal band at 18 • S where almost no fronts occur during the OUTPACE cruise. Southeast of LDB a mesoscale structure, that is also identified on Figure 3, creates a frontal barrier that lasted for more than 30 days. It seems that this structure matches with strong surface gradients in chlorophyll and in SST, consequently separating colder and relatively chlorophyll-rich waters to the south from warmer but chlorophyll-poor waters to the north of the front. Overall, as 15 shown by Figure 5, the most frequent and long-lived fronts seem to help in structuring the spatial distribution of tracers such as SST and chlorophyll concentration by creating physical barriers, isolating areas with different biogeochemical characteristics.
To try to quantify the influence of frontogenesis on the structuring effect of surface tracers, we decided to compare the surface sharp gradients measured by the TSG or with the underway fluorimeter with the presence of a front detected from satellite products. 20 The strong surface density gradients (as defined in Section 2.3.2) represent 9% of the data measured by the TSG during the OUTPACE cruise. The comparison with FSLE reveals that 25% of the strong surface density gradients match with a physical front. Through a bootstrapping re-sampling, the same method is applied to the 91% of TSG data identified as non-gradient, and demonstrates that only 14 ± 1% of homogeneous density areas match with a front. These latter results also exhibit that an FSLE existence does not necessarily create a gradient but probably needs pre-existing tracer gradients and a lifetime longer 25 than few days. The same calculations have also been performed for temperature and salinity gradients independently and show similar results. The relatively better correlation between density gradients and FSLEs than between no-density gradients and FSLEs, attests that gradients are not randomly distributed with regard to FSLE structure and proves that FSLE detection can be a good candidate to explain the presence of in situ surface gradients. Despite the effort to increase the altimetry resolution to 1/8 • , it is still not enough to fully resolve the submesoscale gradients. Consequently the correlation between surface gradients 30 and FSLE can not increase much higher than the 25% calculated here. This result converges with Hernández-Carrasco et al. with FSLE and the gradient sampled with the TSG, and consequently the method applied here may not identify the match.
Therefore the methodology could be improved by focusing on FSLE values around a certain radius from the position of the gradient to eliminate the uncertainty caused by the absolute point by point comparison. Another way to improve the method would be to only take into account cross-front gradients that are more likely to be induced by a physical front than alongfront gradients. However, considering that the SST distribution is also governed by other processes than advection, such as 5 the diurnal cycle, for example, the 25% correlation is large enough to sustain the idea that the submesoscale circulation can participate actively in the spatial structuring of surface tracers such as SSS, SST or density.
The same approach was performed with a reactive tracer, the chlorophyll concentration sampled with high frequency. It shows that 35% of strong surface chlorophyll gradients, representative of 1% of the entire underway sampling, agree with the presence of FSLE. Re-sampling over the 99% of non-gradient areas, indicates that 28 ± 14% of homogeneous chlorophyll 10 areas match with an FSLE. The high percentage of FSLEs matching with a «no chlorophyll-gradient» area gives little confidence on the fact that 35% of chlorophyll gradients were actually caused by the presence of a physical barrier. As chlorophyll concentration is driven by many biological processes, it may be more accurate to associate gradients of phytoplankton abundances, responsible for chlorophyll gradients, with small-scale features. Hereinafter, using two case studies of plankton high frequency sampling, we propose to compare microbial abundances with frontal features.

Physical barriers' influences on phytoplankton community
To test the hypothesis of Bonnet et al. (2015), that pointed out the use of FSLE to explain some correlations between Trichodesmium spp. abundances and gradients, we measured the abundances of microbial groups of plankton in samples collected during two high-frequency sampling transects ( Fig. 6 and 7). LDB high frequency sampling crosses from North to South the bloom patch described in de Verneil et al. (2017b). The spatial distribution of organisms presents relatively high concentration 20 of bacteria at the center of the bloom but decreases when exiting this feature (Fig. 6). The tip of an FSLE barrier is visible near the center of the transect. This barrier coincides with a salinity and temperature gradient (data not shown) associated with a sharp decrease in surface chlorophyll concentration and in PPE, HNA and LNA (bacteria in general) abundances (Fig.   6b). There is a sharp increase of DIP turnover time when exiting the patch in the south, indicating that Phosphorus is quickly consumed inside the bloom. The front thus seems to create a barrier for organisms which grow and accumulate on one side of 25 the front as demonstrated by relatively high surface chlorophyll (peak at 0.8 µg l −1 ) and low Phosphorus. According to Mann and Lazier (2013) phytoplankton growth may be stimulated in aggregates where they can easily take up nutrients released by bacterial decomposition of organic matter. This phenomenon may also support the persistency of the bloom during LDB.
Indeed the bloom may be sustained in time by submesoscale features creating an aggregation of micro-organisms that can benefit from each other. Aggregates can thus influence the community structure by creating favorable growing conditions for 30 a species at the expense of others. A good example is LDA high-frequency sampling (Fig. 7a). Indeed we can notice a region of high abundance of bacteria at 165 • E 15' that is to be bounded by two FSLE barriers. This trend is confirmed by figure 7b which shows an increase of the abundance of Prochlorococcus, bacteria, HNA and LNA associated with a spike of FSLE values at 45 km. At the same time the abundance of PPE seems to decrease where bacterial abundances are the highest indicating Biogeosciences Discuss., https://doi.org /10.5194/bg-2017-456 Manuscript under review for journal Biogeosciences Discussion started: 3 November 2017 c Author(s) 2017. CC BY 4.0 License. that this group may not find an advantage in these features. Another FSLE peak at 80 km was characterized by the decrease of Prochlorococcus, bacteria, HNA and LNA while the PPE abundance increases. The surface chlorophyll follows the same pattern as that of Prochlorococcus, bacteria, HNA and LNA with a relative increase of chlorophyll concentration ( 0.3 µg l −1 ) within the region bounded by the FSLEs. Another peak in chlorophyll concentration is noticeable around 30 km ( 0.4 µg l −1 ) but may be associated with other organisms than those analysed with cytometry. The physical fronts thus not only structure 5 the spatial distribution of organisms by creating barriers but seems to create border regions influencing the community structure, abundances and diversity. Indeed modelled fine-scale structures have already been shown to delimit niches of different phytoplankton types but also to modify phytoplankton assemblages and diversity (d 'Ovidio et al., 2010;Lévy et al., 2014a).
Consistent with these previous numerical results, we therefore find out with in situ observations that microbial growth seems to benefit from the conditions engendered by frontal structures. It remains important to determine how these two study cases 10 could be generalized for the organisms implied in the N 2 fixation cycle.

Conclusions
We document here the surface circulation at different spatial scales (from 1000 km to 10 km) and its influence on horizontal dispersal of biogeochemical components in the WTSP during the OUTPACE cruise. This study is conducted thanks to the combined use of value-added high-resolution altimetry products, in situ observations and Lagrangian numerical simulations. 15 The total altimetry-derived velocity field, combining geostrophy and wind components, revealed a wind-driven meridional pathway of surface waters in the WTSP. This surface trajectory can be linked to the biogeochemical differences between Melanesian waters and gyre waters: a part of the water masses directly recirculates into the gyre whereas the other part is driven across the multiple islands of the WTSP.
The mesoscale activity is confirmed to be intense and mostly westward except in the region of transition between Melanesian 20 waters and gyre waters (180 • W -170 • W). Most of these mesoscale structures demonstrated ability to trap waters, however no obvious biogeochemical variations were linked to eddy entrapment. We identify two areas where the mesoscale circulation might have a strong influence on water mass transport. First, the central Coral Sea appears as a region of exchange between distinct NVJ and NCJ waters through eddy transport. Second, the band between 180 • W and 170 • W could emerge as a recurrent bloom formation area due to the simultaneous effect of N 2 fixation, well-known to sustain summertime blooms in the WTSP, 25 and the eastward mesoscale transport of island-enriched waters.
Associating the surface small-scale gradients with Lagrangian diagnostics of frontal features, we showed a correlation of at least 25% highlighting the role of submesoscale activity in governing the horizontal dispersal of surface tracers. The smallscale features also participated in the horizontal distribution and community structure of phytoplankton patches sampled during two original high-frequency sampling of surface phytoplankton abundances. Maes, C., Gourdeau, L., Couvelard, X., and Ganachaud, A.: Figure A1.