A Lagrangian study of the contribution of the Canary coastal upwelling to the open North Atlantic nitrogen budget

. The Canary Current System (CanCS) is a major Eastern Boundary Upwelling System (EBUS), known for its high nearshore productivity and for sustaining a large ﬁshery. It is also an important, but not well quantiﬁed, source of nitrogen to the adjacent oligotrophic subtropical gyre of the North Atlantic. Here, we use a Lagrangian modelling approach to quantify this offshore transport and investigate its timescales, reach, and contribution to the fueling of productivity in the offshore regions. In our 5 Lagrangian model, we release nearly 10 million particles off the Northwest African coast and then track all those that enter the nearshore region and upwell along the coast between 14 ◦ N and 35 ◦ N. We then follow them as they are transported offshore, also tracking the biogeochemical transformations, permitting us to construct biogeochemical budgets along the offshore moving particles. The three-dimensional velocity ﬁeld as well as the biogeochemical tracers and ﬂuxes are taken from an eddy-resolving conﬁguration of the Regional Ocean Modeling System (ROMS). Our Lagrangian model analysis reveals a very intense offshore 10 transport of nitrogen, with about 20-40% in the form of organic nitrogen. The transport varies greatly along the coast. Even though the central CanCS (21 ◦ N-28 ◦ N) transports the largest amount of water offshore, its offshore transport of nitrogen is somewhat smaller than that in the southern CanCS (14 ◦ N-21 ◦ N), primarily because of the higher nitrogen-content of the upwelling waters there. small ± Our results highlight the importance of CanCS upwelling key of to the open North Atlantic and stress the need for improving the representation of EBUS in global coarse resolution models.

around Cape Blanc (21 • N) (Barton, 1987;Hernandez-Guerra et al., 2005). The CUC is a nearshore surface jet associated with the coastal upwelling front, which flows equatorward along the northwest African coast (Pelegri et al., 2006). The Cape Verde frontal zone, between Cape Blanc and the Cape Verde archipelago, is dominated by a permanent cyclonic circulation with a poleward boundary current, the Mauritanian current (MC), that extends at depth beyond Cape Blanc as a slope undercurrent typical of eastern boundary upwelling systems (Barton et al., 1989;. Upwelling is permanent along 5 most of the Moroccan coast (21-35 • N), albeit with weaker intensity and stronger seasonality north of 26 • N (Cropper et al., 2014). South of Cape Blanc, upwelling is present essentially in late fall and winter (October-March). Upwelled waters have different nutrient contents depending on their respective sources. North of Cape Blanc, the relatively nutrient-impoverished North Atlantic Central Waters (NACW) feed most of the upwelling . In contrast, waters upwelling south of Cape Blanc have a higher nutrient content as they are fed by the nutrient-richer South Atlantic Central Waters (SACW) brought 10 by the MC (Schutte et al., 2016;Kounta et al., 2018;Glessmer et al., 2009;Peña-Izquierdo et al., 2015).
Intense mesoscale structures, including eddies and filaments develop in different parts of the CanCS and contribute to the offshore transport of the coastal waters (Lovecchio et al., 2018). Àlvarez-Salgado et al. (2007) estimate that persistent coastal filaments at the West African coast and Iberia export carbon at a rate 2.5 -4.5 times higher than Ekman transport. Pelegri et al. (2005a) found that in the Canary Basin coastal filaments and cyclonic eddies cause localized offshore export of nutrients and 15 organic carbon. The role of upwelling filaments in the shelf to open ocean transport of organic matter depends also on eddyfilament interactions, which frequently occur on ocean margins (e.g. Barton et al., 1998;Brink and Cowles, 1991). Oceanic eddies may entrain filament waters with higher content of organic matter, enhancing the shelf-ocean exchange (Aristegui et al., 1997). Filaments may return to the continental shelf part of the water upwelled and expelled from the coast by means of eddy-associated circulation. This recirculation decreases the impact of filaments on the offshore transport of organic matter 20 (e.g. Basterretxea and Aristegui, 2000).
The quantification of the magnitude and type of lateral export is challenging due to the complex biogeochemical and physical dynamics of the offshore transport. Much of it is being driven by mesoscale processes, whose appropriate in-situ sampling goes beyond the abilities of current observing systems, thus requiring the use of high-resolution coupled physical/biogeochemical models. Yet, the number of high-resolution model-based studies that addressed and quantified the coastal-open ocean exchange 25 in the CanCS remain limited (e.g., Fischer and Karakas, 2009;Lachkar and Gruber, 2011;Pastor et al., 2013;Auger et al., 2016;Lovecchio et al., 2017;Lovecchio et al., 2018). Such studies have stressed the importance of eddies and coastal filaments in the offshore transport of coastal upwelling. Lovecchio et al. (2017), for example, examined the export of organic carbon from the CanCS using a coupled physical-biogeochemical ocean model with a telescopic grid that covers the whole Atlantic ocean while maintaining a high resolution along the coast of northwest Africa. They demonstrated that about a third of the organic 30 carbon produced along the northwest African coast is transported offshore, and some of it well beyond 1500 km from the coast, contributing substantially to the net community production there. In a follow-up study, Lovecchio et al. (2018) showed that much of this transport is driven first by filaments (in the first 100 km from the coast), and then later taken over by westward propagating mesoscale eddies. However, these authors focused on the transport of organic carbon only and so did not estimate the contribution of the Canary upwelling to the offshore export of nutrients into the open ocean and the implications this might 35 have for the biogeochemistry of the North Atlantic Ocean. Furthermore, the Eulerian approach used by Lovecchio et al. (2017Lovecchio et al. ( , 2018 does not allow for the identification of waters that upwell at the coast and the tracking of their trajectories after they enter the open ocean. Here, we aim to close this gap, and investigate the contribution of the upwelling waters to the nitrogen budget of the open North Atlantic, thereby considering the transport of both organic and inorganic forms of nitrogen. To this end, we use a 5 Lagrangian approach to quantify the offshore reach, the spatial structure and the dominant timescales of the offshore transport of upwelled waters by tracking all open ocean waters that upwell along the coastal region of the CanCS. Lagrangian approaches have a long and rich history in the study of dynamical systems, especially in the atmosphere (Lin et al., 2012), but also in the ocean (van Sebille et al., 2018), as they present a number of advantages relative to the much more commonly used Eulerian approach, especially in the context of offshore transport., By taking the moving particle as a frame of reference, the Lagrangian 10 tracking of water masses is much better suited to analyze the connectivity between the coastal and the open ocean regions.
Furthermore, the Lagrangian method can be used to derive conditional statistics where subsets of particles that fulfill certain criteria are analyzed. This is useful for instance here to restrict the analysis of offshore transport to upwelling particles only.
By not only tracking waters through the appropriate seeding of Lagrangian particles, but also tabulating the biogeochemical transformations along the pathways, we can also establish Lagrangian budgets. This permits us to develop a holistic perspective 15 of the pathways taken by nutrients and carbon along their journey from the ocean's interior through upwelling and then their offshore transport to their ultimate fate of being exported to depth again (Frischknecht et al., 2018). Such an analysis of the origin, transformation and fate is impossible to undertake when using an Eulerian point of view. Thus, our Lagrangian approach permits us to take deep insights into the working of the CanCS system and its connection to the open ocean.
Previous studies have used the Lagrangian approach to study different aspects of the CanCS. For instance, Brochier et 20 al. (2011) conducted a ROMS-based Lagrangian experiment to study the transport of ichthyoplankton (fish eggs and larvae) due to filaments between the West African coast and the Canary Islands. Mason et al. (2012) used a Lagrangian approach to characterize the source waters of upwelling in the CanCS between 31 • N and 35 • N. Yet, these studies were limited to specific regions of the CanCS and did only partially sample coastal upwelling there. Here we substantially expand on these previous efforts by sampling and tracking all open ocean waters that upwell along the West African coast between 14 • N and 35 • N, 25 and quantifying the offshore export of water, nutrients and organic matter. We also investigate the kinetics and the structure of this offshore transport and explore the role of water recirculation and capes in enhancing both coastal upwelling and offshore export. Finally, we examine the contribution of the CanCS upwelling to the Open North Atlantic nitrogen budget. 30 We use a CanCS configuration of the Regional Ocean Modeling System (ROMS)-AGRIF (http://www.croco-ocean.org/) similar to that used by Lachkar et al. (2016). ROMS solves the primitive equations and has a free-surface and a terrain-following vertical coordinates (Shchepetkin and McWilliams, 2005). We use a rotated-split third-order upstream biased operator for the advection of momentum and material properties (Marchesiello et al., 2009). The non-local K-profile parameterization (KPP) scheme is used to represent the subgrid vertical mixing (Large et al., 1994). The biogeochemical model is a nutrientphytoplankton-zooplankton-detritus (NPZD) model based on nitrogen (Gruber et al., 2006). It uses a system of ordinary differential equations representing the time-evolution of the following state variables: nitrate (NO − 3 ), ammonium (NH + 4 ), phytoplankton, zooplankton, two pools of detritus and a dynamic chlorophyll-to-carbon ratio. The two classes of detritus represent, 5 respectively, fast-sinking large organic matter particles and slow-sinking small particles. The small particles can coagulate with phytoplankton to form large detritus. In the water column, small and large detritus are remineralized to ammonium at rates of 0.03 and 0.01 day −1 , respectively. Sinking particulate organic matter that reaches the seafloor is also remineralized to ammonium at a rate of 0.003 day −1 (Gruber et al., 2006).

Models and configuration
The model domain covers the region from 10 • N to 42 • N in latitude and from 30 • W to 6 • W in longitude with a grid 10 resolution of 1/20 • . This corresponds to a mesh size of about 5 km, which is sufficient for fully resolving mesoscale processes.
The vertical grid consists of 32 layers with enhanced resolution near the surface. The bathymetry is derived from the ETOPO2 file provided by the National Geophysical Data Center (Smith and Sandwell, 1997) and a minimum bathymetry of 50is set.
We use a monthly climatological forcing based on the Comprehensive Ocean‚ Atmosphere Data Set (COADS) (da Silva et al., 1994) for surface heat and freshwater fluxes. Surface temperature and salinity are restored to COADS observations using 15 kinematic heat and freshwater flux corrections following Barnier et al. (1995). Wind stress is derived from the QuikSCATbased Scatterometer Climatology of Ocean Winds (Risien and Chelton, 2008). The initial and lateral boundary conditions for temperature, salinity and nitrate are derived from the World Ocean Atlas (WOA) 2009. Other ecological tracers are initialized uniformly to arbitrary low values. Currents at the boundaries are derived from temperature and salinity data using geostrophy together with Ekman transport in the upper 40 m. The model starts from rest and is spun up for 9 years. We then run it for 3 20 more years, on which we carry out the analysis, and store these outputs at a daily frequency.
The model was evaluated by confronting its outputs to a wide array of in-situ and satellite-based observations (see supplementary material section for full details). We find that despite some local biases and discrepancies with observations, the model generally shows good skill in reproducing the large-scale features of the circulation and the productivity of the Canary current region (Figs S1, S2). More importantly, it reproduces the strength and structure of the Canary coastal upwelling and 25 captures the observed vertical structure of nitrate and organic matter (Figs S3 to S17). Overall, this CanCS-only ROMS setup has similar strengths and weaknesses as the telescopic grid setup employed by Lovecchio et al. (2017Lovecchio et al. ( , 2018. We will discuss the potential impact of the model limitations on our results in the discussion section.

Lagrangian experiment
The Lagrangian particle tracking experiment is performed offline with ARIANE (Blanke and Raynaud, 1997;http://stockage.univ-30 brest.fr/ grima/Ariane/) using ROMS daily output. ARIANE tracks water particles based on the velocity output of the model.
In this experiment, ARIANE runs based on the model output of zonal and meridional velocities (ARIANE internally computes vertical velocities from the continuity equation). ARIANE analytically computes streamlines of particle trajectories across grid-walls by assuming a steady-state flow. The data points for successive days are used as piecewise steady flow where the  velocity is assumed to be static for each day and not interpolated between different days. The velocity at a given point inside the regular cells of the model is computed by linearly interpolating the velocity at opposite faces of a cell. ARIANE's computation of particle trajectories doesn't account for the subgrid vertical mixing in the model.
With our Lagrangian experiment we aim to study trajectories of open ocean water masses that enter the coastal region and upwell between 14 • N and 35 • N (see sample trajectories in Figure 1b). This region extends in north-south direction a total of  (Fig. 1a, Fig S20). To this end, we carry out our Lagrangian experiment in three steps: particle release (quantitative ARIANE experiment), particle tracking (qualitative ARIANE experiment) and upwelling identification (see the schematic in Fig S18).
In the quantitative experiment, a coastal upwelling region is set up between 14 • N and 35 • N. This region is bounded by 10 the particle release strip (both entry and exit) to the west, the coast to the east and particle exit (but not entry) sections at the northern and southern ends of the upwelling coastal region (Figure 1a). The particle release strip is set at the limit of the continental shelf defined at 50 km westward of the 200 m isobath along the coast, which ensures sampling coastal upwelling occurring at the shelf break ( Fig. 1b). This results in a coastal upwelling region with a width that varies from below 60 km near 27 • N where the continental shelf is narrowest to over 160 km near 25 • N where the shelf is widest. . Particle release is 15 based on the onshore water flow volume into the coastal region, where each particle is tagged with the corresponding initial transport. We limit the maximum transport a particle is associated with to 0.01 Sv, which limits the maximum volume assigned to one particle to 0.864 km 3 . Through a daily particle release for one year at each cell along the particle release strip, a total of 9,888,387 particles are released. By the end of the quantitative experiment, for each particle, we have its location, time of release and associated water volume.

20
A qualitative ARIANE experiment is then run on the initial conditions set by the quantitative experiment to track the particle trajectories and their along-path nitrogen concentrations. Particles are tracked for 720 days since their release until they leave the simulation's domain. Along each particle's trajectory, its latitude, longitude and depth are saved daily along with biogeochemical tracers. Note that around 4.2%, 0.37% and 0% of the particles released in the southern, central and northern subregions, respectively, leave the domain southward or northward within one year of their release. Assuming the ocean to be 25 an incompressible fluid and its velocity field to be non-divergent, the volume transport is conserved. We quantify the amount of nutrients carried by each particle to a given location as the product of its associated volume and the concentration of the tracer associated with the particle when it reaches that location.
Subsequently, tracked particles that enter the coastal region and cross upwards the 70 m depth, the average depth of the euphotic zone in the Canary coastal region, are identified as upwelled particles. Since only particles that enter the coastal area 30 and upwell there are followed, our experiment disregards wind stress curl-driven upwelling that occurs in the open ocean and can locally be important, particularly in the southern subregion (Lovecchio et al., 2017). From all tracked particles, 352,873 (3.57%) upwell, which form the basis for our analyses (see additional details in Fig S18).
Previous studies have shown subregional differences within the CanCS in circulation patterns, mesoscale activity, seasonality of upwelling, biology and sub-surface nutrient concentration Pelegri and Benazzouz, 2015;  Our Lagrangian analysis reveals an annual upwelling of offshore-derived waters of nearly 80,000 km 3 (see Table 1). This is about 25 km 3 of water per year per kilometer of coastline, or 0.8 m 3 s −1 per meter of coastline. The strongest upwelling occurs in the central CanCS, with a particularly strong peak around Cape Bojador (27 • N) (Fig. 2a). This subregion alone is responsible for more than half of the total upwelling (40,705 km 3 yr −1 ). The northern subregion contributes about 25,000 km 3 per year to the upwelling, while the southern subregion has the smallest upwelling (13,000 km 3 yr −1 ).  Table 1. The annual number of upwelling particles and their associated water and nitrogen, the median distance of upwelling to the coast and the net transport of water to 400 km and 1000 km from coast (as percent of upwelling volume) in the three subregions as well as the entire CanCS region in our experiment.
Most of the upwelling stems from deeper waters, i.e., waters that enter our analysis region below 70 m. 85% of the tracked particles and 88% of the associated water volume follow that path (termed 'deep source'). The remaining upwelling stems from waters that enter the region within the euphotic zone and then are transported below 70 m before upwelling (here termed 'shallow source') ( Figure 1b). This fractional distribution varies little between subregions. But the offshore distribution of the upwelling differs strongly between the three CanCS subregions ( Fig S22). These differences are reflected in the median 10 upwelling distances being 64.1km, 37.1km and 41.1 km in the southern, central and northern subregions, respectively ( Table   1).
Most of this variability has to do with the variability of offshore Ekman transport and curl-driven upwelling between the three subregions. For instance, the central subregion has strong year-round coastal upwelling whereas upwelling tends to be strong only in the summer season in the northern subregion (Pelegri and Benazzouz, 2015). Similarly, the Ekman-driven upwelling 15 in the southern subregion is restricted to the winter and spring (Pelegri and Benazzouz, 2015;Capet et al., 2017). There are distinct differences also with regard to the wind curl-driven upwelling between the three subregions. Indeed, the wind stress curl is predominantly downwelling-favorable in the northern and central subregions and upwelling-favorable in the southern subregion. This enhances upwelling in the southern region, particularly in the open ocean, but its effect is only partially sampled in the present study given our focus on the coastal region. Other inter-regional variations may also stem from the design of 20 the experiment. For instance, our experiment identifies limited coastal upwelling south of Cape Blanc and between Capes Barbas and Bojador because their bathymetry is shallower than the 70 m upwelling depth criterion used here. Furthermore, the upwelling strip is relatively narrow in the northern subregion because of the narrower shelf, thus limiting the offshore spread of coastal upwelling there ( Figure S21).
Finally, the upwelling patterns of water and nitrogen have a few important differences. The southern subregion has the 25 lowest upwelling water volume yet the strongest upwelling flux of nitrogen (Fig. 2). This is primarily due to the shallower Over the whole CanCS, the water offshore transit times (the median time for particles to traverse a given offshore distance interval) are longer in the nearshore region and decrease as we go further away from the coast (Fig 3). Large differences in the offshore transport timescales exist between the three CanCS subregions. In the first 200 km from the coast, the offshore transport is fastest for particles upwelled in the central subregion and slowest for those in the northern subregion (Fig 3;   Fig. S25). Indeed, in the central subregion, 80% of the upwelling particles reach 200 km in three months and 50% reach that 10 distance in two months only. In contrast, less than 30% of particles upwelling in the northern subregion reach 200 km offshore in three months and it takes more than four months for half of them to reach that distance. At larger distances from the coast (beyond 400 km), the offshore transport becomes fastest for particles that upwelled in the southern subregion with nearly 80% of particles reaching 1200 km in two years, while only 70% and 30% of particles upwelling in the central and northern subregions, reach that distance in two years. This can also be seen in the water residence times being shortest for the central subregion particles up to 400 km and for the southern subregion particles beyond 400 km (Table S1).

Net offshore transport
Integrated over the whole analysis domain from 10 • N to 41.5 • N, the CanCS exports over 70'000 km 3 yr −1 of water toward the open North Atlantic (Figure 4a). The maximum offshore transport is reached at close to 150 km from the shore at around the 5 edge of the coastal upwelling area most sampled upwelled water (nitrogen). It is around 150km because at closer distances to the coast (<100-150km), upwelled volume is only partially sampled, while further distances (>100-150km) are never reached by a proportion of particles because of recirculation retaining them close to the coast and alongshore transport exporting some particles out of the model domain. Beyond this maximum, the transport decreases gradually as the number of particles reaching farther offshore distances within the two-year integration period declines. But even at a distance of 1200 km from the coast, 10 the offshore transport still amounts to 40'000 km 3 yr −1 .
There are substantial differences between the different subregions. The central subregion of the Canary system is responsible for over half of the entire net offshore transport of water from the CanCS at any distance from the coast (Figure 4a). The offshore transport associated with the northern subregion upwelling is initially very strong, but decreases sharply thereafter, so that beyond 800 km, it becomes the weakest of all three regions. Reduced coastal upwelling and low offshore transport 15 efficiency (i.e., the ratio of offshore transport volume at a given distance to upwelling volume) explain this pattern (Table 1, Figure 4a). The southern subregion is associated with the smallest offshore transport of water up to 800 km offshore due to its small coastal upwelling volume. b a Figure 4. Net offshore transport of (a) water and (b) nitrogen as a function of the distance to the coast in the three subregions (left axis) as well as the entire CanCS (right axis) in our experiment.
Around 150 km from the coast, the nitrogen transport reaches values as large as 500 Gmol yr −1 at 200 km, decreasing exponentially further offshore, although with a relatively long decay length scale, such that at a distance of 1200 km, the offshore export of nitrogen still amounts to 100 Gmol yr −1 .
Although offshore water transport was minimal for the southern CanCS particles, this subregion is responsible for the greatest offshore export of nitrogen (nearly half of the offshore export of nitrogen by the whole CanCS region at this distance).

5
This is a direct consequence of the large upwelling flux of nitrogen (Table 1 and Fig. 4b). This pattern reverses further offshore.
Beyond 200 km from the coast, the offshore transport of nitrogen associated with the central subregion upwelling exceeds that originating from both the southern and northern subregions. At all distances past 600km, the central subregion contributes over half of the nitrogen offshore export by the entire CanCS. At these distances, the nitrogen transport of waters stemming from the central subregion is at least twice as large as that from the southern subregion, and four times as large as that from the northern 10 subregion.
The magnitude of the offshore transport of nitrogen at any distance from the coast depends both on the volume of offshore transport and how efficiently nitrogen is stripped from the waters by biological productivity and the resulting organic nitrogen exported to depth. It is thus instructive to assess the specific forms of nitrogen being transported offshore.

Nitrogen allocation 15
In the nearshore 50 km, phytoplankton is very efficient in taking up the inorganic nitrogen that is being upwelled, and fixing it into organic forms of nitrogen (Fig 5a). This results in nearly 100% of the offshore transported nitrogen to be in the form of organic nitrogen. But as additional nitrogen is being supplied from below, the fraction of the upwelled nitrogen that gets consumed decreases rapidly with increasing offshore distance. Furthermore, some of the fixed organic nitrogen is being lost through sinking, so that beyond the nearshore 50 km region, inorganic nitrogen in the form of nitrate dominates the nitrogen 20 pool at all distances from the coast (Fig 5a). Beyond 300 km, ammonia is the second largest pool. Overall, organic nitrogen contributes only 30% to the total nitrogen pool. Within the organic nitrogen pool, small detritus contributes the most to the offshore transport, while the contribution of phyto-and zooplankton is much smaller and that of the large detritus particles essentially negligible.
Given its dominance in terms of the total offshore transport, the southern subdomain is also the main region determining the 25 whole CanCS pattern of nitrogen allocation (Fig 5b). In this southern subdomain, the fraction of organic nitrogen is particularly low, being only 20% at the peak of the offshore transport. With increasing distance, the fraction increases to 30%. This indicates a much further offshore extension of the conversion of inorganic nutrients to organic matter in this domain. The central domain has a nitrogen allocation pattern that is similar to that of the whole CanCS (Fig 5c), while the transport in the northern subdomain is not only weak, but also the least dominated by the offshore transport of nitrate (Fig 5d). nature, making this an important part of the offshore transport. In contrast, the very small contribution by the large detritus is largely a consequence of its rapid export to depth (see also Gruber et al., 2006). Upwelled particles that are transported offshore are also subject to vertical circulation that distributes them vertically (Fig   6, S27). Subduction may potentially slow down offshore transport since velocities deeper in the water column are smaller.

5
However, subduction can also increase the efficiency of offshore transport by minimizing the depletion of nitrogen in surface waters.
For the northern and central subregions, upwelling particles are subject to moderate subduction at around 100 to 200 km from the coast (Fig 6). In the southern subregion, however, upwelling particles are subject to very little subduction while a strong open ocean upwelling maintains them near the surface. Similarly to the water volume transport, subduction of nitrogen 10 is strongest in the northern subregion and weakest in the southern subregion (Fig. S24). The stronger downward advective transport occurring in the central and northern subregions is due to a strong negative wind stress curl in these subregions, which is absent in the southern subregion (Lovecchio et al., 2017). Furthermore, persistent filaments associated with prominent capes in the central and northern subregions may contribute to enhanced subduction of upwelling water and nutrients there (Lovecchio et al., 2018). The offshore transport of organic nitrogen is smaller in magnitude than that of inorganic nitrogen and is mostly limited to the near surface in all subregions (FIg 7, Fig. S24 and Fig. S28). In contrast, the transport of inorganic nitrogen shows a subsurface secondary maximum at between 50 and 100 m in addition to the surface maximum (Fig 7, Fig. S24). Our analysis highlights the two opposing effects of subduction in slowing offshore transport speeds but also reducing biological uptake (and nitrogen loss due to sinking) and hence helping maintain high nitrogen content by upwelled water.

5
The convergence of the Canary coastal current flowing from the north and the Mauritanian current flowing from the south leads to a strong offshore transport of nitrogen around cape Blanc (21 • N) ( Fig. 7 and Fig. S29). The nitrogen (both organic and inorganic) channelled through the confluence of the two currents originates predominantly from waters upwelling in the central and southern subregions, with a small contribution from particles upwelling in the northern subregion. This is consistent with the finding of Lovecchio et al. (2017) who -using a Eulerian approach -also found the Cape Verde Front to be a major channel and Capes Dra (28.5 • N -29.5 • N) and Ghir (30 • N -31 • N) in the northern subregion ( Fig. 1a and Table S2). Previous studies of the Canary upwelling mostly stressed the role of filaments in locally enhancing offshore transport regardless of the source of upwelling. Similarly, we find an enhancement of offshore transport of upwelled particles in the first 200km from the coast within the latitudinal range of all capes except Capes Verde and Juby (Figure 8). Transport at latitudes of capes Blanc, Barbas, Dra and Ghir is larger than at the non-cape latitudes as far as 1000 km from the coast. We further separately consider the 5 enhancement of offshore transport associated with (i) enhanced local upwelling around the capes and (ii) increased export of non-locally upwelled water at each cape, i.e., the export of waters that upwelled far away from the capes but is then transported along the coast toward the cape, from where it is exported toward the open ocean.
The offshore transport of water upwelled around Cape Bojador in the central subregion and Capes Dra and Ghir in the northern subregion is 20% to 30% larger than that originating from non-cape areas (per coastal length) ( Figure S30). Increased 10 offshore transport of nitrogen due to increased local upwelling is seen only in Capes Dra and Ghir out of the seven capes examined (Figure 8). Cape Bojador does not show enhanced local nitrogen upwelling (although a slight enhancement is shown in its offshore export).
Cape Blanc, Cape Barbas and Cape Ghir are highly efficient in exporting non-locally upwelled waters to the open ocean ( Fig.   S31). Cape Blanc's enhancement of offshore export of nitrogen by non-local upwelling compared to the non-cape part of the 15 central subregion is 350% -400% in the first 400 km then constantly increases further offshore to reach a 935% enhancement at 1200 km. Similarly, Cape Barbas shows an enhancement of non-local upwelling of about 200 % for all distances from the coast while Cape Ghir shows a 250% enhancement from 100 km to 800 km offshore (Figure 8). Non-locally upwelled waters that are transported offshore around major capes represent more than 30% of the total offshore transport at all distances from the coast ( Fig. 4 and Fig. 8). Indeed, most of the water and nitrogen exported offshore around major capes is non-locally upwelled 20 (Table 2). Each cape also sources more of its export from non-local upwelling than any of the non-cape coastal areas. Non-local upwelling accounts for over 75% of the nitrogen export in all capes except Cape Verde and Cape Bojador while it accounts for a minority of the source of non-cape parts of the coast except in the central subregion ( Table 2).
The analysis of the source waters for particles exported around the major capes reveals a strong alongshore transport of upwelled particles that connects upwelling between different coastal regions (Fig. 9). This is consistent with previous studies 25 that found strong meridional alongshore advection of nutrients (Carr and Kearns, 2003;Pelegri et al., 2006;Meunier et al., 2010;Troupin et al., 2012;Pastor et al., 2013;Pelegri and Benazzouz, 2015;Auger et al., 2016;Lovecchio et al., 2017). The ability of capes to facilitate offshore export of coastal water through their interaction with alongshore advection has also been previously documented (Meunier et al., 2010;Troupin et al., 2012). With the exception of Cape Blanc and Cape Barbas, most capes export offshore water that mostly first upwells north of their latitude (Fig. 9). In the central and northern subregions, 30 this is primarily due to the southward offshore flow by the Canary current. In the southern subregion, both the Canary and Mauritanian currents are important sources for coastal upwelling, particularly during winter (Klenz et al., 2018). This is visible in the northern source of upwelling that leaves the coast at all capes in this subregion and the southern source of upwelling that leaves the coast at Cape Blanc and Cape Barbas (Figure 9). We conclude that capes such as Cape Blanc and Cape Barbas  Figure 8. Enhancement of net offshore transport of nitrogen by capes. (a,b,c) Transport within the latitudinal span of each cape or non-cape area. Transport occurring within each latitudinal span is considered at any distance from the coast irrespective of the location of upwelling or coastal export. (d,e,f) Transport associated with locally upwelled water in each cape or non-cape coast. (g,h,i) Transport associated with non-locally upwelled water at each cape or non-cape area. Note that the transport is normalized by coastal length (divided by the length of the coast at the respective cape/non-cape).
act to concentrate and export non-locally upwelled waters, and that this effect is much more important than the occurrence of enhanced local upwelling at the capes.  Figure 9. Latitude of upwelling of nitrogen exported at each cape or non-cape coast.

Role of recirculation
Particles that are transported offshore past a certain distance can return back to that same distance or never cross it again. We term the transport associated with the latter "direct transport". Particles that cross the same distance multiple times can end up 5 further offshore (an odd number of crossings) or closer to the coast (an even number of crossings). We term them "indirect transport" and "net recirculation", respectively (Figure 10a).
It is worth noting that only direct and indirect transport contribute to the net offshore transport of water, while net recirculation does not (although it may slightly contribute to transport of nitrogen) ( Figure 10). The sum of direct and indirect transports adds up to give the net transport of water to each distance while the sum of all three represents the total volume of water that has 10 reached each distance at any point during the experiment. In the nearshore region, the meandering of the Canary Current as well as the coastal upwelling cell can cause upwelling particles to recirculate closer to the coast (Mittelstaedt and Hamann, 1981; Cape/Non-cape Local N (Mmol N km −1 ) non-local N (Mmol N km −1 ) Local N (%) Local Water (%)  Table 2. First two columns show daily contribution of local and non-local upwelling to nitrogen that leaves the coastal region at each cape (in Mmol N km −1 ) per km of coastal length. The last two columns show the contribution of local upwelling (in %) to total nitrogen and water exported at each cape or non-cape coast. Mittelstaedt, 1983;Estrade et al., 2008). But the leading cause of recirculation in the open ocean is the ubiquitous presence of mesoscale structures, particularly eddies (Mason et al., 2011;Mason et al., 2012;Barton and Aristegui, 2004;Schutte et al., 2016;Kounta et al., 2018;Ndoye et al., 2017;Capet et al., 2017). Therefore, contrasting the direct and indirect components of the transport can be used to gauge the relative importance of the mesoscale eddies and current meandering in the offshore transport.

5
Examining the three components of the offshore transport in the three subregions reveals that the direct and indirect components virtually explain all the transport of water and nitrogen at all latitudes and distances from the coast. Yet, important differences exist between the three subregions in terms of the relative importance of these two forms of transport (Fig. 10).
For the southern subregion particles, direct transport contributes by a large share to the total transport of water and nitrogen at all distances except in the first 200 km offshore where the indirect transport dominates. In contrast, for particles upwelled 10 in the central and northern subregions, the direct transport has a greater share to transport only past 700 km offshore, while the indirect transport dominates closer to the coast. The relative importance of indirect transport is particularly strong between 100 km and 400 km in these two subregions. For instance, at 200 km the indirect transport contributes by up to 65% and 80% to the total offshore transport of water in the central and northern subregions, respectively. For the transport of nitrogen, the share of the indirect transport at the same distance is 60% and 85% in the two subregions. 15 The importance of the indirect transport of upwelled waters in the central and northern subregions can be linked to the prominent role played by mesoscale eddies there. Indeed, eddies with length scale of 100 km to 300 km are known to be   indirect transport (blue) and net recirculation (red) to the total net offshore transport of water (solid) and nitrogen (dashed) at each distance to the coast in the entire CanCS region (b) as well as the southern (c), central (d) and northern (e) subregions. Note that the net recirculation has no contribution to net offshore water transport but contributes slightly to the net offshore transport of nitrogen.
important at these latitudes (Mittelstaedt, 1991). These include a recurrent cyclonic eddy south of Cape Juby and the cyclonic and anticyclonic eddies entrained by the Canary Archipelago, forming the so called Canary Eddy Corridor (CEC), which is located at 22 • − 29 • (Arístegui et al., 1994;Piedeleu et al., 2009;Sangrà et al. 2009). This region of long-lived westwardpropagating eddies is known to contribute strongly to the offshore transport of organic matter and carbon (Sangrà et al., 2009).
The prominence of water recirculation in the central and northern subregions -materialized by the dominance of the indirect transport -up to 600-800 km offshore is consistent with the longer crossshore transit times characterizing these latitudes (Fig Table 3. Nitrogen sources and sinks for the NATR and NASE provinces (in mmol N m −2 yr −1 ). Sources considered are the offshore transport by the Canary upwelling, the net meridional transfer of nitrogen through Ekman divergence, N2-fixation and atmospheric deposition. The only sink considered is new production (NP). Atmospheric deposition (Duce et al., 2008) 9.5 ± 2.5 14.5 ± 2.5 Net meridional transfer (Williams and Follows, 1998) 30 (NO3 only) 30 (NO3 only) * estimates are based on maps provided by the cited paper ** Contribution of upwelling by each subregion is calculated by adding the net amount of nitrogen upwelling particles carry to each province in the top 100 m by the end of the experiment. That is, whenever a particle enters a province within the top 100m, the nitrogen it carries into the province is added to the particle's source subregion's contribution to the province. On the contrary, when a particle from a given subregion leaves a province within the top 100m, the nitrogen it carries with it when it leaves the province is subtracted from the contribution of the subregion of the particle to the province.
The strong export of nutrients and organic matter from the CanCS to the oligotrophic open ocean fuels new production (NP) and contributes to heterotrophy there. Here we quantify this contribution in the North Atlantic Tropical Gyral Province (NATR) and the North Atlantic Subtropical Gyral East (NASE) provinces as defined by Longhurst et al. (2007) (see Fig. S20). The Lagrangian approach allows for the isolating of the contribution of the upwelling particles to the transport of nitrogen into the 5 NATR and NASE provinces, adjacent to the CanCS (see Fig. S20). We calculate the net offshore transport of nitrogen in the top 100 m at the precise boundaries between each CanCS subregion and each Longhurst province.
New Production (NP) is estimated from the Net Primary Production (NPP) and the available estimates of the f-ratio (f-ratio= N P N P P ) in the literature. Since new production and export production (EP) are numerically the same in a system in steadystate (Dunne et al. 2006), we also use estimates of the e-ratio (e-ratio= EP N P P ). For NPP, we use the Vertically Generalized 10 Production Model (VGPM) estimates, based on sea color data from 1997 to 2017, as well as a carbon-based productivity model (CbPM) by Westberry et al. (2008) to derive NPP for each province. We also use an in-situ estimate of NPP for the NATR made by Tilstone et al. (2009) based on Carbon-14 uptake (Table 3). We use estimates of the f-ratio from Laws et al.
(2000) (see their Plate 3) as well as previous estimates of the eratio from Henson et al. (2011) and Siegel et al. (2014). The derived f-ratio values are typically low and vary from 4% in the NATR according to Henson et al. (2011) to 15% in the NASE 15 following Laws et al. (2000) ( Table 3).
The southern, central and northern coastal upwelling subregions in our study have total upwellings of 282.3, 200.9 and 79.2 Gigamol N yr −1 , respectively. A total of 563 Gmol N yr −1 , 358 Gmol N yr −1 and 32 Gmol N yr −1 reach the upper 100ṁ of the NATR and the NASE, respectively. When normalized by the area of each province, the CanCS upwelling source corresponds to 44 mmol N m −2 yr −1 and 7 mmol N m −2 yr −1 for the NATR and NASE, respectively (Table 3). Most of the upwelled 20 nitrogen exported to The NATR originates from the central (46.5%) and southern subregions (43%), whereas the nitrogen upwelled in the northern subregion dominates (76.7%) the nitrogen supplied to the NASE. Hence, the nitrogen transported (in the top 100m) from the Canary coastal upwelling to the NATR and NASE provinces represents 27 -79 % and 2 -5 % of their total New production, respectively (Table 3).
Finally, the contribution of the Canary upwelling also appears significant in comparison to other major sources of new nitro-25 gen to the oligotrophic open North Atlantic Ocean such as N 2 fixation (Moore et al, 2009;Luo et al., 2012;Fernandez-Castro et al., 2015), atmospheric deposition (Duce et al., 2008) and net meridional transfer through Ekman divergence (Williams and Follows, 1998). Indeed, the upwelling source seems to exceed contributions from N 2 fixation, the atmospheric deposition and meridional transfer in the NATR province. In the NASE province, the upwelling contribution is larger than the N 2 fixation but weaker than the atmospheric deposition and the meridional transfer (Table 3).  Table S4). Our lower offshore transport estimates may at least partially result from our Lagrangian approach that focuses on coastally upwelled water only, disregarding non-upwelling waters and open ocean upwelling. Furthermore, restricting the analysis to upwelling particles in our study limits offshore flux to near-surface waters as most upwelled waters remain at very shallow depth (see section in supplementary material for more detail).
Previous model-based studies of the Benguela and California coastal upwelling systems suggest other EBUS may also sig-5 nificantly contribute to the nitrogen budget of adjacent open oceans (Table S5). For instance, Gutknecht et al. (2013) found significant net offshore transport of nitrogen in the Benguela current system (BenCS), with organic nitrogen exceeding inorganic nitrogen at far offshore distances. They estimated the contribution of the BenCS in the top 50m, to amount to 100 ± 40 mmol N m −2 yr −1 to the adjacent South Atlantic Subtropical Gyral Province according to the classification by Longhurst (2007) (Gutknecht et al., 2013). This contribution, however, is likely overestimated since results in the highly productive Walvis  (Table S5). The efficiency of offshore transport of organic matter and dissolved inorganic nitrogen in the top 100 m between 100 km and 500 km are almost identical between their study of the CalCS and the present study of the CanCS.

8 Implications and caveats
Our Lagrangian approach identified waters upwelling along the Northwest African coast and characterized the kinetics, structure and timescale of their offshore transport. Using a Lagrangian approach allowed us to assess the reach and fate of coastally upwelled nitrogen far from the coast as well as study its biogeochemical transformation along particle trajectories. Our estimate of offshore export of organic nitrogen/carbon by upwelling is lower than the estimate by Lovecchio et al. (2017), at all 20 distances from the coast (Table S3). These differences can in particular stem from the differences in approaches (Lagrangian vs. Eulerian) and the studied parts of the CanCS (their budget is computed for the area between 9.5 • N and 32 • N) that differ between the two studies. We found the offshore transport to be generally slowest in the first few hundreds of km from the coast and to increase in speed with increasing offshore distances (Fig 3). This is particularly true for waters upwelling in the northern and the southern subregions of the CanCS. The existence of an inverse relationship between upwelling water recirculation and 25 their cross-shore transit times (Fig. S32) suggests the slowdown of offshore transport in the first 300-500km to result from a larger role played by eddies that act to weaken the net offshore transport there. Previous studies have highlighted the role of eddies in contributing to organic carbon and nutrient offshore transport in EBUS (Nagai et al., 2015;Lovecchio et al., 2018). The findings of the present study reveal that enhanced water recirculation driven by intense eddy activity may reduce the efficiency of the offshore transport of water.

30
In the first 200km, the offshore transport is fastest and the water recirculation is weakest for particles that upwelled in the central CanCS subregion (Fig 3, Fig. S32). We show that this region is characterized by the presence of major filamentgenerating capes such as Cape Blanc and Cape Barbas that contribute strongly to offshore transport of water and nutrients.
They act as concentrators that collect coastal water and nutrients and channel it to the open ocean as most of the water exported (84%) at the capes upwells non-locally. Lovecchio et al. (2018) have highlighted the role of persistent filaments in mediating more than 80% of the offshore transport of organic carbon in the first 100km from the coast. Here, we show that key capes along the CanCS represent hotspots of offshore transport of water and nitrogen, thanks to the intense filaments that develop around them. Our findings also highlight the importance of the alongshore transport as most of the waters exported at capes 5 have upwelled at latitudes far away from the location of the capes. This is consistent with the concept of the three-dimensional biological pump highlighted in previous studies, according to which strong offshore and alongshore lateral transports decouple the locations of upwelling and of biological export (Plattner et al., 2005;Lovecchio et al. 2017Lovecchio et al. , 2018. As the along-trajectories nitrogen-depletion rates depend on the depth of the transport, the subduction of upwelling waters to deeper layers occurring mostly between 100 and 200km offshore tends to affect the efficiency of the offshore transport of Finally, our study confirms the strong alongshore variations of the CanCS in terms of offshore transport reach and efficiency. 15 Our results show that the offshore transport of nitrogen off the CanCS depends on many important factors, including the intensity of upwelling, the efficiency of the offshore transport, controlled by the kinetics of the transport and the nitrogen depletion rates. In particular, factors such as eddies, filaments, and alongshore transport all contribute to modulate the magnitude and reach of the water volume and nutrient content transported offshore. The complexity of these factors and their interactions explain the complexity of the offshore transport and its spatial diversity. 20 Yet, the study has several caveats that stem from the limitations of the experimental design or are inherent to the tools. The physical model underestimates EKE in parts of the domain, including in the southernmost part. This may be in part due to the monthly climatology wind forcing we use, potentially leading to an underestimation of vertical mixing and an incomplete representation of the complex upwelling dynamics. However, our analysis shows that the model successfully captures much of the spatial and seasonal variations in upper ocean mixing ( Fig S1) and reproduces upwelling distributions and pathways 25 that are not inconsistent with previous findings. On the other hand, our biogeochemical model is based on nitrogen and has no representation of other potentially limiting nutrients such as iron and phosphate. Furthermore, the model lacks a representation of nitrogen fixation. Yet, previous studies indicate nitrogen to be the main limiting nutrient (e.g., Moore et al., 2013) and nitrogen fixation to be very low (Luo et al., 2012;Moreira-Coello et al., 2017) in the CanCS upwelling region. Also, the model only has one class of phytoplankton. However, the specific nature of the phytoplankton community appears to be of secondary 30 importance as our model reproduces quite well the mean vertical profiles of POC in different parts of the domain (Fig S15).
Therefore, we believe these limitations in the biogeochemical component of the model should not affect the study's main conclusions.
A potentially more serious limitation stems from our Ariane-based Lagrangian analysis. First, ARIANE includes resolved transport but does not explicitly take into account subgrid mixing in the computation of trajectories. This is a common problem 35 in most Lagrangian trajectory based studies. We think including subgrid mixing should increase the vertical dispersion of upwelled water and nitrogen. While this may reduce the offshore transport efficiency (weaker offshore velocities at depth), it should also reduce along-trajectory nitrogen depletion. A compensation (even partial) between these two processes may lead to an overall small change in the estimated total supply of upwelled nitrogen to the open North Atlantic ocean. Yet in contrast to water volume, tracing the transport of nitrogen is somewhat more difficult given the chemical transformations between 5 inorganic and organic nitrogen. Yet, as has been shown also by Frischknecht et al. (2018), the Lagrangian method permits a lot of new insight into the offshore transport of nitrogen, since total nitrogen, i.e., the sum of inorganic and organic nitrogen is conserved except for the part that is sinking. The component we lose through sinking does not affect our conclusions, since this component is lost to the ocean interior, from where it will not find its way back into the waters that are transported offshore.
More importantly is our lack of consideration of the vertical mixing. We have good evidence that this component is relatively 10 small. If the supply of nitrogen from surrounding waters to upwelling waters due to mixing were large enough to cancel the loss due to organic matter sinking, there would be no such a sharp decline in total nitrogen as a function of distance to the coast relative to that for water volume (Fig 4). Moreover, this decline is larger for water particles originating from the southern subregion that are transported at the shallowest depths and that are subject to the strongest sinking loss (Fig S23). This suggests that although the potential changes in nitrogen due to subgrid mixing can locally be important, they are unlikely to affect the 15 large-scale transport estimates in a significant way. variations driven by the variability of atmospheric forcing and those caused by changes in the analysis depth.
It is worth noting that only coastal waters that originate from the open ocean are sampled as particles are released at the western section of the coastal region only. However, we have found that only a very small amount of water, 1% and 3% of the total volume tracked in this study, enters the upwelling coastal region along the coast from the northern and southern boundaries, respectively. Furthermore, given the high rate of recirculation near the coast, a vast proportion of the particles that 30 may have entered the coast alongshore, once upwelled and exported offshore, are likely to return to the coast from the open ocean, in which case they would be sampled in our particle release. Therefore, discarding particles entering the coastal ocean from the southern and northern boundaries of the upwelling strip is likely to cause only a limited error in our quantification of the offshore transport.
Due to a larger upwelling volume and a faster offshore transport in the nearshore region (up to 400 km offshore), the central CanCS subregion is responsible for the largest net offshore transport of water at any distance from the coast. Conversely, the southern CanCS is associated with the smallest offshore transport up to 800 km offshore due to a smaller volume of coastal upwelling. Yet, the southern subregion exports the highest amount of nitrogen offshore at 200 km because of a large upwelling 5 flux of nitrogen associated with a shallower thermocline and higher subsurface nitrogen concentration. At larger distances from the coast, the offshore transport of nitrogen associated with the central subregion exceeds that originating from the southern subregion because of a lower nutrient depletion caused by a more frequent subduction of upwelled water into deeper layers where phytoplankton growth is light-limited. The analysis of nitrogen allocation along trajectories reveals that the offshore transport of organic nitrogen is generally less than half of that of inorganic nitrogen and is mostly limited to the near surface in 10 all subregions.
The pattern of offshore transport is characterized by the presence of latitudinally alternating offshore-onshore corridors indicating a strong contribution of mesoscale eddies and filaments to the mean transport. Major capes along the CanCS that are favorable to the formation of persistent filaments are associated with an enhanced offshore export of water and nitrogen.
This results primarily from an enhancement of local export of water and nitrogen non-locally upwelled (by up to a factor 4 15 to 9 for C. Blanc) relative to non-cape areas. The offshore transport of nitrogen by non-local upwelling exported by all capes constitutes more than 1 3 of the total offshore transport of upwelling waters at all distances from the coast. All capes source the majority of water and nitrogen they export from non-local upwelling.
Finally, we found the supply of nitrogen by the Canary upwelling to the NATR and the NASE provinces, to amount to 43.7 mmol N m −2 yr −1 and 7.3 mmol N m −2 yr −1 , respectively. This represents 53±26% and 3.5±1.5% of the total potential new 20 production in the two provinces, respectively The contribution of the Canary upwelling appears significant in comparison to other major sources of new nitrogen to the open North Atlantic Ocean. This emphasizes the importance of the CanCS upwelling in particular and EBUS in general as a key source of nutrients to the open ocean and the offshore transport as a mechanism of supply of these nutrients to the adjacent oligotrophic gyres. Our work thus stresses the need for improving their representation in global coarse resolution models, as has been indicated by previous works (e.g., Large and Danabasoglu, 2006). Institute of Technology Zurich (ETH Zurich). The authors are also grateful to Michael Mehari for producing the Taylor diagrams presented in this work. We thank two reviewers for their constructive comments that helped to improve and focus our paper. The authors are grateful to Brochier, T., Mason, E., Moyano, M., Berraho, A., Colas, F., Sangrà, P., Hernández-León, S., Ettahiri, O., and Lett, C.: Ichthyoplankton transport from the African coast to the Canary Islands, Journal of Marine Systems, 87(2), 109-122, 2011.