Fueling primary productivity : nutrient return pathways from the deep ocean and their dependence on the Meridional Overturning Circulation

Introduction Conclusions References


Introduction
Primary productivity in the ocean's sunlit surface incorporates dissolved nutrients into organic matter, a fraction of which sinks to the ocean interior.Once in the ocean's dark interior, the organic matter is respired back to dissolved inorganic carbon and nutrients, Figures thereby creating the pervasive vertical nutrient and carbon gradients that characterize the global ocean.Above the central pycnocline (nominally the 1026.8kg m −3 potential density surface), the total inventory of phosphate, an essential nutrient for photosynthesis, is approximately 150 Tmol (Conkright et al., 2001).Recent estimates of the sinking flux of organic phosphorus (Dunne et al., 2007), propagated to the depth of this density surface, are of order 3 Tmol yr −1 .Thus, to prevent the upper ocean from becoming entirely depleted of phosphate, the nutrient must be returned to the upper ocean on time scales of approximately 50 years.We hypothesize that the upward branch of the Meridional Overturning Circulation (MOC), which converts dense deep waters to lighter pycnocline waters, plays a central role in maintaining the primary productivity of the upper ocean by returning exported nutrients to depths where they can again become accessible to the ocean's sunlit upper reaches.On long time scales, the MOC's upward branch balances the sinking of deep waters and is driven by a combination of vertical mixing across the low latitude pycnocline and wind-driven upwelling and mixing in the Southern Ocean (Kuhlbrodt et al., 2007).However, the relative magnitude of the low latitude pathway versus the Southern Ocean pathway remains an open question, as does the implication for the cycling of nutrients.Here, we explore the MOC nutrient return pathways in a suite of ocean general circulation models with different physical parameterizations, seeking an understanding of the balance between the Southern Ocean nutrient pathway versus the low latitude pathway and the sensitivity of this balance to model physics.
A dominant role for the Southern Ocean in the MOC's upward branch has formerly been hypothesized based on tracer data (Toggweiler et al., 1991;Sarmiento et al., 2004Sarmiento et al., , 2007) ) in the context of theoretical considerations of ocean overturning (Toggweiler andSamuels, 1995, 1998;Gnanadesikan, 1999).In terms of nutrient cycling, this hypothesis implies a loop in which the global loss of organic matter to depths beneath the pycnocline is largely balanced by the upwelling of remineralized nutrients in the Antarctic Circumpolar Current (ACC).These nutrients are then advected and mixed across the ACC into the Subantarctic Mode Water (SAMW), a water mass formed on Figures the ACC's equatorward fringes that influences tracer properties throughout much of the global pycnocline (McCartney, 1982;Sarmiento et al., 2004;Hanawa and Talley, 2001).This conceptual model hinges on the upwelled nutrients being subducted with the SAMW at the time it forms, rather than being utilized and exported from it, as can be the case for other mode waters (Palter et al., 2005).Indeed, the SAMW is formed in a region with highly elevated surface nutrients (Fig. 1a), and it tends to retain its nutrients during formation and subduction.Evidence supporting the dominance of the Southern Ocean nutrient upwelling pathway has been obtained by model sensitivity studies, in which the strength of global primary productivity was critically reduced by completely stripping nutrients from the surface ocean in the SAMW formation region (Sarmiento et al., 2004;Marinov et al., 2006).
Additional support for the importance of a Southern Ocean upward MOC branch in influencing pycnocline nutrient concentrations comes from looking at quasiconservative combinations of nutrient tracers.For example, in the Subantarctic Zone where SAMW is formed, phosphate (PO 3− 4 ) and nitrate (NO − 3 ) are abundant, while silicic acid (Si(OH) 4 ) is nearly depleted.Thus, a useful tracer of the SAMW is Si*=Si(OH) 4 -NO − 3 (Sarmiento et al., 2004).The lowest Si* concentrations in the global surface ocean are found in the SAMW formation region (Fig. 1b).Sarmiento et al. (2004) argue that because Si* is approximately conserved in the ocean interior, the subsurface pool of negative Si* indicates the spreading of SAMW via advective and diffusive transport processes throughout much of the low latitude ocean (Fig. 1c).
Despite the apparent convergence of evidence from models and the Si* tracer towards a central role for a Southern Ocean nutrient return pathway from the deep ocean to the pycnocline, several important questions still require evaluation and quantification.What is the fraction of low latitude productivity sustained by the Southern Ocean nutrient return pathway and which water masses are involved?In which regions are the SAMW nutrients most important, and what are the pathways by which the SAMW arrives at these downstream locales?Here we examine these questions using a suite of ocean general circulation models (OGCMs).A second class of questions concerns Figures the sensitivity of the predominant nutrient return pathway to the models' parameterizations of subgridscale processes and the prescribed atmospheric forcing that drives the ocean circulation, particularly the Southern Ocean winds.The fraction of deep upwelling that occurs in the Southern Ocean is sensitive to the strength of the winds, as well as the turbulent mixing and transport due to mesoscale eddies, which are parameterized in the coarse GCMs typically used to simulate climate and biogeochemistry.The sensitivity of the modeled MOC to winds and subgridscale processes has been demonstrated in more than one idealized analytical model (Gnanadesikan, 1999;Samelson, 2004Samelson, , 2009)), and tested in a suite of GCMs (Gnanadesikan et al., 2007).
Here we quantify the sensitivity of the magnitude and distribution of nutrient supply to wind forcing and the parameterizations of subgridscale processes.We also consider the implication of these results for non-steady behavior by investigating the lateral and temporal scales over which the preformed nutrients subducted in the SAMW layer persist in the ocean interior before fueling downstream primary productivity.Preformed nutrients are those inorganic nutrients that enter the ocean interior without having fueled primary productivity while in the mixed layer.Therefore, preformed nutrients are a signature of inefficiency in the biological pump.A more efficient biological pump utilizes a greater portion of surface nutrients, converts them to organic material and increases the remineralized nutrient and carbon concentration deeper in the water column.Higher remineralized nutrient concentrations signify greater ocean carbon storage and lower atmospheric carbon dioxide, while a higher preformed nutrient inventory signals a smaller ocean carbon reservoir (Archer et al., 2000;Ito and Follows, 2005;Gnanadesikan and Marinov, 2008;Marinov et al., 2008a, b).Much of the SAMW nutrients enter the ocean in their preformed state, but when they fuel productivity downstream of their formation region, these inorganic nutrients are incorporated to organic matter and eventually converted to their remineralized state.Any increase in the subduction of the SAMW preformed nutrient pool would diminish the deep ocean carbon reservoir until these preformed nutrients are converted to their remineralized form.Evaluation of this potential for change in the ocean preformed nutrient reservoir Figures

Back Close
Full is especially timely given recent observations showing the intensification of Southern Hemisphere westerly wind strength (Thompson and Solomon, 2002) with implications for the stability of the ocean carbon sink (Le Qu ér é et al., 2007).Such an intensification is expected to increase the cross-ACC Ekman transport of preformed nutrients, the preformed nutrient load of the SAMW, and the local inefficiency of the biological pump due to the subducted preformed nutrients in the mode and intermediate waters.
Our model experiments shed light on the time scale that this perturbation might be expected to persist.
2 The theory, the models, and the experimental design

The theory
A simple conceptual model of the MOC consists of four branches: 1) deep water formation regions in the North Atlantic where surface waters become denser and sink; 2) flow away from these regions at depth; 3) upwelling and mixing that transports the dense waters from the ocean interior to the surface ocean; and 4) near-surface currents that close the circulation.Where and by what forcing mechanism the deep water returns to the surface (branch 3) is a central question in physical oceanography (for an extensive review see Kuhlbrodt et al., 2007), and, increasingly, motivates ocean biogeochemical and climate studies (Gnanadesikan et al., 2002;Sarmiento et al., 2007;Anderson et al., 2009).The two forcing mechanisms that have received the most attention are the downward mixing of heat and associated upwelling across the pycnocline (see Munk and Wunsch, 1998 for a detailed discussion) and the wind-driven upwelling of deep waters in the Southern Ocean and their subsequent buoyancy gain at the surface of the ocean (Toggweiler andSamuels, 1993, 1995).Gnanadesikan (1999)  cnal mixing, and wind forcing all determine where dense water returns to the ocean's surface, while the interplay of all three factors determines the depth and thickness of the pycnocline and the strength of the overturning cell.
For the modern ocean, the Gnanadesikan (1999) theory can be summarized as follows.Dense water forms and sinks in the high latitude North Atlantic and is exported southward in the deep limb of the MOC, a flux represented in the schematic as M n .All along its export pathways, the downward mixing of heat transforms the dense water to lighter densities and forces an upward volume flux represented as M u that reduces the deep southward flow as it approaches the Southern Ocean.This upward flux scales as: where K v is the coefficient of diapycnal diffusion, D is the pycnocline depth that separates dense water from light, and A is the area of the low latitude pycnocline.
In the Southern Ocean, laterally divergent Ekman transport forces the upwelling of abyssal waters, a flux represented as M s .The upwelled waters that cross the ACC into the SAMW formation region gain buoyancy along their journey, thus completing the dense-to-light transformation of the water masses formed in the Northern Hemisphere high latitudes.However, adding another layer of complexity, a volume transport due to mesoscale eddies is directed southward across the ACC, balancing some fraction of the northward Ekman transport.The eddy transport is thought to be proportional to the local slope of isopycnals, which is the source of potential energy from which the eddies grow (Gent and McWilliams, 1990).Thus, the flux of abyssal water to the light water realm in the Southern Ocean, M s , is equal to the difference between the Ekman transport M ek and the mass flux due to the eddies: where A i is a coefficient that determines the strength of the eddy mass transport, L the north-south scale over which the pycnocline shallows in the Southern Ocean, and L x is the Earth's circumference at the Drake Passage latitude (see also Karsten and Marshall, 2002).For the limiting case in which the eddy transport across the ACC is as large as the Ekman transport, the net Southern Ocean deep water return pathway becomes negligible and the upward branch of the MOC must be accomplished exclusively through low latitude diapycnal mixing; this is the picture of the MOC hypothesized by Stommel and Arons (1961).For the alternate limiting case in which global diapycnal mixing is, on average, as small as the diapycnal mixing measured in the ocean away from topography (Gregg and Sanford, 1980;Ledwell et al., 1993Ledwell et al., , 1998)), the Ekman upwelling of dense waters in the Southern Ocean and their buoyancy gain provide the primary driver of meridional overturning (Toggweiler and Samuels, 1998).We note that this conceptual model does not explicitly consider spatial variability in the strength of diapycnal mixing observed in the ocean (Heywood et al., 2002;Naveira Garabato et al., 2004), although the model framework outlined below allows for some exploration of this variability.

The models
The underlying architecture used to probe our questions is a suite of simulations using the Geophysical Fluid Dynamics Laboratory Modular Ocean Model version 3 (MOM3) (Pacanowski and Griffies, 1999).The MOM3 suite explores a range of mixing parameters and wind forcing, as described in several previous studies (Gnanadesikan, 1999;Gnanadesikan et al., 2004Gnanadesikan et al., , 2007;;Mignone et al., 2006), and summarized in Table 1.The models create variants of the MOC strength and structure by changing: a) the Southern Ocean Ekman transport, M ek , via the use of two different wind reanalysis products to force the model; b) the counterbalancing eddy transport via the manipulation of A i , the coefficient associated with the Gent and McWilliams (1990) (GM) parameterization of eddy transport, and c) the diapycnal mixing across the low latitude pycnocline via variations in K v , the coefficient of diapycnal diffusion.Here we present the results from a subset of four models from a larger MOM3 suite explored 4052 Figures

Back Close
Full in Gnanadesikan et al. (2007), chosen because they capture much of the inter-model differences of the full parameter range without being excessively cumbersome for comparison.
Two of the models are forced with wind stresses from Hellerman and Rosenstein (1983), which are nearly half as strong in the Southern Ocean as the European Centre for Medium-Range Weather Forecasts (ECMWF) climatological winds (Trenberth et al., 1989) used to force the other two models.Of these Hellerman and Rosenstein-forced models, one is run with a Low K v and Low A i , and is therefore referred to as LL; the other has a High K v and High A i , and is referred to as HH.LL-ECMWF is identical to the LL model in the ocean, but is forced with ECMWF winds.
A special case we call P2A resembles the LL-ECMWF model but has been modified in several ways intended to increase the realism of the simulation: the Drake Passage was narrowed from three grid points (1165 km) to two (777 km) to be closer to the real Drake Passage, which is 800 km wide; strong vertical mixing (5×10 −3 m 2 s −1 ) is imposed on the uppermost two grid cells (50 m) as a very crude approximation to what actually occurs in an oceanic mixed layer; and surface salinity is strongly restored to subsurface observations in four key locations at the onset of Austral winter in Southern Ocean deep water formation regions to simulate brine rejection during sea-ice formation.Without this salinity adjustment, restoring the model to observations suppresses Antarctic Bottom Water (AABW) formation, as the observations do not capture the outcropping of such waters.Finally, in P2A the coefficient of diapycnal mixing is increased in the Southern Ocean, first motivated by observations of intense ACC internal wave activity (Polzin, 1999) and later validated by observations of rapid diapycnal tracer spreading in the region (Naveira Garabato et al., 2004).Clearly the MOC strength and upward branch is controlled by the interplay of these parameter choices (Fig. 2b-c), with the clockwise part of the MOC in the HH model confined entirely to the region north of 20 • N, while the LL-ECMWF overturning is actually intensified in the Southern Hemisphere.We will explore the MOC differences among the models at greater length in Sect.A model of the biological cycling of phosphorus and carbon is coupled to the physical model, as specified by the Ocean Carbon Model Intercomparison Project-2 (OCMIP-2) protocol (Najjar et al., 2007).In this protocol, phosphorus is the only simulated nutrient and is conserved globally in all experiments.The phosphorus source/sink terms are governed by an extremely simple set of equations meant to represent the incorporation of phosphate into organic matter by photosynthesis in the euphotic zone and the sinking of this organic matter and its remineralization back to phosphate throughout the water column (Fig. 3).The simulated phosphorus appears in 2 pools: dissolved organic phosphorus (DOP) and phosphate (PO 3− 4 ).These pools are transported as passive tracers by the physical model.At depths shallower than 75 m, model PO 4 everywhere in the model, at a rate of κ=0.5 year −1 .

The experimental design
To investigate the role of the upward branch of the MOC in the supply of nutrients to the low latitude pycnocline and ultimately the euphotic zone, we created two sets of tags that are schematized in Fig. 4. The first is a simple set of dye tracers.In each tagging region (represented in black in the schematics), the dye value is set to one; in each destruction region (white in the schematics), the dye value is set to zero; in the rest of the domain (gray in the schematics), the dye tags are treated as conserved, passive tracers.The dyes are named according to their source locations and are fully described after the release of the tags, except for LL-ECMWF, which are for years 390-400.At year 190, the drift in average low latitude dye concentrations is less than 0.1% year −1 , very close to equilibrium, and almost all regions are fully tagged.After 200 years, only the water masses of the Bay of Bengal are not fully tagged (thus, the global sum of all tags can be slightly less than zero).
Outside the tagging and destruction regions (i.e. in and above the low latitude pycnocline), the dye tracers directly measure the fractions of water in and above the main thermocline that derive from each tagging region.The abyssal ocean is tagged only in low latitudes and only at potential densities greater than 27.4 kg m −3 , which is beneath the main thermocline in all models.The Northern Hemisphere High latitudes are tagged everywhere north of 40 • N, and further designated as stemming from the North Atlantic and North Pacific basins.In the Southern Ocean, the tags are divided into density classes in order to approximate the contribution of SAMW and AAIW to the low latitude pycnocline.By design, the southernmost tag is trapped in the Southern Ocean, and is used primarily to ensure that all water masses are tagged and a water mass budget can be balanced.
In the Southern Ocean, AAIW is most commonly identified as a salinity minimum along the 27.3 isopycnal.Because the SAMW is formed in large part by convective processes, it can be identified as a pycnostad, or region of low vertical density stratification.A comparison of our simulations with a well-studied salinity and density section along 100 • E (Sloyan and Kamenkovich, 2007) (Fig. 5), reveals that the models do a good job simulating the salinity and stratification minima that characterize the AAIW and SAMW, respectively.The model density structure in the Southern Ocean resembles the observations, giving confidence that our Southern Ocean water mass definitions are appropriate to address the question at hand.It is also reassuring that the SAMW dye tracer on the 26.8 isopycnal shows a similar spatial distribution to the Si* tracer from data (compare Fig. 6 with Fig. 1c).We note that the SAMW penetrates into the Northern Hemisphere in all of the models, regardless of whether there is net destruction or creation of light waters in the Southern Ocean. the models are largely seen in tropical latitudes where the HH model (Fig. 6b) has much lower concentrations of SAMW dye than the others.
The second set of tags is designed to trace nutrients from the same source regions as the dyes.In each source region, all model PO 3− 4 and DOP are tagged, or named according to their region, such that the total pools of PO 3− 4 and DOP everywhere in the model domain are the sum of the regionally-tagged pools.As an illustrative example, PO 3− 4 and DOP are endowed with a SAMW tag between the 26.5 and 27.1 isopycnals in the Southern Hemisphere, southward of where the 26.5 isopycnal intersects 200 m (as for the SAMW dye tracer, shown in Fig. 4c).
As the SAMW-tagged DOP (hereafter SAMW DOP) is transported in the flow field from its tagging region, it is respired to a separate pool of remineralized SAMW phosphate.The same is true for the SAMW phosphate that fuels primary productivity downstream of its tagging domain: the remineralization of the resultant DOP and sinking flux provides a source of remineralized SAMW phosphate.Thus, although the phosphorus maintains its regional identity until it enters a destruction region, it is marked as either coming directly from the tagging region, which we will refer to as preformed, or being created by remineralization away from the tagging region, which we refer to as remineralized.We note that, because the tags are not exclusively introduced in the mixed layer, preformed nutrients in our experiments differ from a traditional definition of preformed nutrients as those that have been cycled through the mixed layer and returned to the ocean interior without having fueled primary productivity.This distinction becomes particularly apparent in our discussion of the deep low latitude phosphate, which is tagged only in the interior ocean, and may therefore bear no resemblance to the traditional definition of preformed nutrients.Thus, throughout this work, references to regional tags and tagged nutrients and their preformed and remineralized components will appear in italics for clarity.Combining the equations for the source/sink terms of PO

Results and discussion
The simulated MOC in the different models behaves as predicted by the simple analytical model presented in Sect.2.1.In all models, the interplay of the chosen parameter values creates a pycnocline with a realistic depth and density and an MOC at 48 • N similar to the 16-19 Sv inferred from observations (Table 2) (Lumpkin and Speer, 2007;Talley, 2008;Forget, 2009).However, the upward branch of the MOC occurs at notably different locations among the models (Table 2).The LL model with its weak Southern Ocean winds and low mixing creates an MOC of 13.6 Sv, with more than half of the upward branch supplied in the Southern Ocean; the HH model creates 17.9 Sv of overurning, but its upward branch is realized entirely at low latitudes; the P2A and LL-ECMWF models yield more than 18 Sv of overturning, but with upward branches primarily in the Southern Ocean.Some observationally-driven inverse models (Lumpkin and Speer, 2007) and, more recently, the results from data-assimilating forward model used in Estimating the Circulation and Climate of the Ocean (ECCO) (Wunsch and Heimbach, 2007;Forget, 2009) have suggested that the upward branch of the MOC is accomplished primarily in the Southern Ocean (Table 2).Yet, another observationallydriven estimate suggests a more important role for low latitude upwelling and implies values of low latitude diapycnal mixing as high as in our HH model (Talley, 2008).Thus, the relative importance of these two possible upwelling branches lacks a definitive solution at present, providing a justification for including in our analysis models that cover the range of possibilities.Furthermore, the modeled MOC has a direct influence on the large-range of air-sea CO 2 fluxes estimated from coupled models (Cao et al., 2009).Radiocarbon, a useful tracer of ocean circulation, offers evidence consistent with the upwelling of deep waters in the Southern Ocean.We briefly compare the radiocarbon results from our four models with observations to offer a context with which to judge the various circulations.Radiocarbon enters the ocean only at the air-sea interface and then decays in the ocean interior with a half-life of 5730 years.Thus, the most-recently ventilated ocean waters are most enriched in radiocarbon, and the Figures oldest waters most depleted.The HH model strongly mixes depleted radiocarbon from the dense ocean interior to the surface in subtropical latitudes, causing a depleted radiocarbon bias in the near-surface layers (Fig. 7, red line).This bias is reversed in the Southern Ocean, where the HH model overestimates the radiocarbon in the surface due an excessive supply of low latitude, high radiocarbon surface waters.Not surprisingly, the LL model, which is forced with sluggish Southern Hemisphere westerly winds, has precisely the opposite bias: an underventilated deep ocean with a low radiocarbon bias, and corresponding radiocarbon trapping in the upper ocean.The P2A model with weak vertical mixing at low latitudes and high Southern Ocean winds and mixing is the most successful at accurately simulating the observed radiocarbon concentrations in the surface and deep ocean (Fig. 7, blue line).The P2A model is also the only one with an ACC transport at Drake Passage (126 Sv) in the range of observational estimates of 135±14 Sv (Cunningham et al., 2003); the other three models have a Drake Passage transport of approximately 80 Sv.More extensive model-data comparison with the MOM3 suite can be found in a number of previous papers (Gnanadesikan et al., 2002(Gnanadesikan et al., , 2004;;Matsumoto et al., 2004); all suggest a critical role for upwelling and/or mixing in the Southern Ocean in driving the MOC.Despite the differences among models, the SAMW comprises a sizeable fraction (45-68%) of the low latitude waters above the 26.8 isopycnal in all models (Table 3).As predicted in the analytical framework, the HH model has the smallest SAMW fraction, as water from depths beneath the pycnocline (with the Deep Low latitude tag) upwells directly across the pycnocline.Among the three models with smaller coefficients of vertical mixing, the differences are less stark; still, the SAMW fraction at low latitudes is reduced in the LL model relative to the LL-ECMWF and P2A models, as the latter are forced with stronger Southern Ocean westerly winds (i.e. the ECMWF winds).
The water mass fractions obtained from the dye tags and the fraction of low latitude phosphate obtained from the nutrient tags can be quite different (Table 3).For example, in the P2A model, the Deep Low latitude tag contributes only 5% of the low latitude pycnocline waters, but provides 11% of its phosphate, whereas the SAMW-tag con-Figures

Back Close
Full tributes 67% of the pycnocline waters but only 48% of the phosphate (Table 3).These differences arise because the phosphate concentration in a tagged water mass may be highly elevated above the phosphate concentration in the pycnocline.Thus, when the water mass is mixed or advected into the pycnocline it can contribute a greater percentage of the resultant phosphate concentration than the corresponding contribution of the dye fraction.Such is the case for the highly phosphate-enriched deep ocean.
Each of the models demonstrates some skill at reproducing the large-scale features of the nutricline: simulated PO 3− 4 concentrations arch upward from the deep, depleted bowls of the subtropical gyres to the shallow nutricline of the equator and the high latitudes, just as they do in the observational data (Figs.8a-b and 9a-b).For brevity, we have chosen to show the zonally-averaged sections only for the HH and P2A models in Figs. 8 and 9; though HH is quite different from the other three models, LL, LL-ECMWF, and P2A are qualitatively similar, at least as far as the zonally-averaged nutrient sections are concerned.
The Southern Ocean is notable for its elevated near-surface nutrient concentrations (Figs. 8 and 9), which result from strong vertical and lateral advection and mixing of nutrients to the surface (Sloyan and Rintoul, 2001;Naveira Garabato et al., 2004;Sarmiento et al., 2004) in the presence of weak, iron-and light-limited biological drawdown (Coale et al., 2004).The simplified biogeochemical model from the OCMIP-2 protocol does not simulate an iron cycle, nor does it evaluate the irradiance available for photosynthesis.Nonetheless, the biological model does represent the sluggish biological PO primary productivity, on the other hand, is highly dependent on model physics.This intermodel difference can clearly be seen in tropical nutrient cycling (Figs. 8  and 9).In the HH model compared to the P2A model, the spreading of the SAMW phosphate is restricted to a narrow band near its source region, largely because the high GM coefficient (A i ) creates a parameterized southward eddy transport that largely compensates the northward Ekman transport of SAMW (Figs. 8f and 9f).Similarly, this same GM transport causes a more restricted North Pacific phosphate spreading in the HH run relative to the runs with a lower GM coefficient (Figs.8c and 9c).Furthermore, the deep low latitude phosphate diffuses strongly across the low latitude pycnocline due to the choice of high diapycnal diffusion (Figs.8e and 9e).This strong diffusion of nutrients across the low-latitude pycnocline also causes the HH model to have lower concentrations of phosphate in the dense waters underlying the tropics, the highest low-latitude productivity, and the one that disagrees the most with observationally-based estimates (Gnanadesikan et al., 2002).
Winds also affect the extent to which the spreading of the SAMW transports nutrients away from the Southern Ocean (Fig. 10).In the models forced with sluggish Southern Ocean winds (LL and HH), the preformed SAMW phosphate comprises the majority of the total pycnocline nutrients only in a narrow band south of 30 • S near the tagging region (Fig. 10) and represents less than 10% of the nutrients in the Northern Hemisphere gyres.Spreading of the preformed SAMW phosphate is stronger in the P2A and LL-ECMWF models, with a noticeable contribution to total PO 3− 4 above the 26.8 isopycnal in the subtropical North Atlantic (Fig. 10, top row).This density horizon is chosen to represent the central pycnocline, but the qualitative results are not sensitive to the density horizon used, as long as it is above the 27.4 isopycnal below which the deep low latitude tag is assigned.
The importance of SAMW supply is more apparent in the spreading of its remineralized component (Fig. 10, bottom row).Whereas the preformed SAMW phosphate is largely confined to the Southern Ocean, the remineralized SAMW phosphate can propagate far downstream of its tagging region, contributing a significant phosphate Figures source even in the Northern Hemisphere subtropical gyres (Fig. 10).We estimate a time scale associated with the conversion of preformed to remineralized phosphate by dividing the total preformed SAMW phosphate reservoir by the rate of its consumption in fueling primary productivity.On average across the models, the preformed SAMW phosphate fuels primary productivity within 33 years of its creation.The time scale is longest in the P2A model (39 years), which has both a large SAMW preformed nutrient pool and a weak vertical mixing coefficient outside of the Southern Ocean.The time scale of preformed SAMW phosphate drawdown is shortest in the HH model (22 years), with its small SAMW pool and higher mixing rates.We again note that the designation of preformed nutrients is given to some nutrients that are tagged outside the mixed layer.We expect that if our tag adhered more closely to its traditional definition, downstream productivity would be fueled to an even greater degree by remineralized rather than preformed SAMW phosphate, but that the total productivity sustained by SAMW phosphate would be much the same.The nutrients tagged in the SAMW region are the most important source of nutrients sustaining low latitude (nominally 30 • S to 30 • N) productivity in every model except HH (Fig. 11).In the HH model, productivity throughout the low latitudes is sustained primarily by phosphate that upwells directly from the abyss across the low-latitude pycnocline, and secondarily by the SAMW phosphate.The low latitude productivity draws largely from the remineralized phosphate pool (Fig. 11 small contribution of the northern high latitudes can be attributed to the experimental design, which re-tags all nutrients that enter the low latitude abyss with the deep low latitude tag.A portion of the North Atlantic DOP and phosphate sinks and spreads to the deep low latitude tagging region (where the North Atlantic tag is reset to zero) without first fueling productivity.
How do nutrients subducted in the SAMW layer at densities greater than those in the subtropical euphotic zone come to influence productivity throughout the low latitudes?Originally tagged beneath the 26.5 isopycnal, the SAMW rises to shallower depths and lighter densities after spreading from its tagging region (Fig. 12).First, in the downwelling domain just the north of the SAMW tagging region, advective and diffusive fluxes of the SAMW at the depth of the 26.5 isopycnal are downward.This downward flux is caused by the tagging routine in which the SAMW tag is destroyed beneath the 27.4 isopycnal, but only diluted by mixing above this layer, so that the SAMW concentration maximum moves to densities slightly above the 26.5 isopycnal just to the north of the tagging region.Downstream of this downwelling domain, the SAMW is transported upwards.The upward transport is primarily achieved by vertical advection in tropical regions, along the eastern fringes of the Southern Hemisphere subtropical basins, and in the South Atlantic western boundary current off Brazil (Fig. 12).Once advected and mixed to shallower depths, the SAMW is transformed by surface buoyancy forcing and diapycnal mixing and may no longer carry the characteristic temperature, salinity and potential vorticity signature it had upon formation and subduction.However, our tags trace the water mass fraction and its nutrients downstream of the tagging regions even as the water mass itself is destroyed by such transformations.
The SAMW nutrients that are upwelled into the euphotic zone in the tropics and along the eastern boundaries of the subtropics are consumed and converted to organic particles that sink through the water column.These sinking particles are partially remineralized on layers with lighter densities and shallower depths than the SAMW had upon subduction (Toggweiler and Carson, 1995;Williams et al., 2006).Thus, a portion of the nutrients originally subducted in the SAMW remains available to support Figures subtropical productivity as it spreads along these lighter surfaces to the neighboring subtropical gyres.
Our model results suggest that nutrients slowly lost to depths beneath the pycnocline throughout the low latitudes return to lighter layers primarily in the Southern Ocean.This picture emerges from all models but HH and is summarized in Fig. 13.In the Southern Ocean where biological export does not efficiently compete with the advection and mixing of nutrients to the surface layers, nutrients are upwelled and mixed vertically in the ACC and then subducted in the SAMW.Therefore, much of the water masses formed in the Southern Ocean enter the ocean interior with elevated nutrient concentrations.Although much of these nutrients are subducted to depths well below the subtropical euphotic zone, their upwelling along the eastern and western fringes of the subtropics and in the tropics allows these nutrients to sustain subtropical primary productivity.
Not far from the SAMW source region, the SAMW preformed nutrients are consumed and converted to their remineralized counterparts within the pycnocline; further downstream the remineralized nutrients are re-incorporated into organic matter wherever they are mixed and advected into the sunlit surface ocean.In these downstream locales, the slow leakage of organic phosphorus past the base of the seasonallyaccessible layers is almost entirely balanced by the lateral and vertical supply of nutrients that were cycled through Southern Ocean water masses.In the HH model, the nutrient budget achieves a slightly more one-dimensional balance: a majority of the sinking flux of nutrients downward across the low latitude pycnocline is restored locally via the mixing of nutrients upwards across the low latitude pycnocline.However, even in the HH model, SAMW phosphate sustains more than a third of low latitude productivity; in the other models the percentage is between 57% and 75% (Fig. 13).Each of the other tagging regions besides SAMW and deep low latitude contributes less than 10% of the phosphate that fuels global low latitude productivity.Figures

Back Close
Full Our model framework has allowed us to identify the dominant nutrient return pathways from the dense ocean interior to depths that are accessible to the sunlit euphotic zone.Between 30 • S and 30 • N, the upper ocean nutrient reservoir is dominated by those nutrients that were returned from the deep ocean to the pycnocline in the SAMW formation/subduction region of the Southern Ocean (Fig. 13).Only in the model where the coefficient of vertical mixing is set to levels higher than those observed in the pycnocline of the open ocean does mixing directly across the low latitude nutricline provide the dominant return pathway of nutrients to the low latitude euphotic zone.The Northern Hemisphere subpolar regions are not the primary suppliers of nutrients to the low latitude euphotic zones; nor are the regions of Southern Ocean south of the SAMW formation region.These results are in line with previous modeling studies (Marinov et al., 2006) and with inferences from the distribution of tracers such as Si* (Sarmiento et al., 2004).But there are two major new insights afforded by this study.The first stems from our focus on the sensitivity of nutrient pathways to the winds, vertical mixing, and the strength of lateral eddy transport (via changes in the GM coefficient).Our results suggest that in models with high vertical mixing, high eddy transport, and low Southern Ocean winds, the dominant nutrient balance is more one-dimensional, i.e. organic nutrients lost by downward sinking at low latitudes are restored through the upward mixing of inorganic nutrients across the pycnocline at these latitudes.At the other extreme, in models with low vertical mixing, weak GM transport, and stronger Southern Ocean winds, organic nutrients lost across the low latitude pycnocline are restored to the pycnocline primarily in the Southern Ocean and depend on the full three-dimensional ocean circulation for their delivery to the low latitude euphotic zone.
The second major insight afforded by the study is an approximation of the spatial and temporal scale over which nutrients that enter the ocean interior in the SAMW formation region persist before being utilized in downstream locations.In less than 40 years Figures on average, phosphate entering the ocean interior in the SAMW region is utilized to fuel downstream productivity.In large part, the SAMW nutrients first fuel primary productivity south of 20 • S.This short spatial and temporal scale separating the subduction of SAMW nutrients and their downstream utilization helps explain previous modeling results that tested the influence of the SAMW region on atmospheric carbon concentrations.In the modeling study of Marinov et al. (2006), the SAMW nutrient source was "turned off" by forcing nutrients towards zero at the Southern Ocean surface north of the outcropping 27.3 isopycnal, the equivalent of increasing nutrient drawdown by enhanced export production.Although this forced drawdown decimated low latitude productivity by starving these regions of their nutrients, the experiment had little impact on ocean carbon storage or atmospheric carbon concentrations on millenial time scales.The reason for the lack of impact on atmospheric carbon is that the biological pump had been made much more efficient locally in the SAMW formation region, but the global efficiency was unchanged once the model returned to equilibrium (Marinov et al., 2006).In experiments without the forced nutrient drawdown, the SAMW nutrients were already being efficiently consumed downstream of the SAMW formation region on decadal time scales.
Our model experiments suggest a paradigm for returning nutrients to the surface ocean that contextualizes a number of phenomena.For example, it has been hypothesized that biological nitrogen fixation in the North Atlantic is sustained by excess phosphate (phosphate in excess of nitrate relative to their Redfield ratio) advected in the shallow limb of the MOC (Moore et al., 2009).The excess phosphate signature in the shallow MOC layers stems from the SAMW formation region, where high-phosphate waters from the Indo-Pacific Basin are mixed with the water masses advected across the ACC.Iron limits biological nitrogen fixation in the SAMW formation region and throughout the subtropical South Atlantic, so the excess phosphate signature is preserved during formation and subduction and is exported to the North Atlantic.Because nitrogen fixation in the relatively iron-rich North Atlantic appears to be phosphate-limited (Sa ñudo-Wilhelmy et al., 2001;Moore et al., 2009) excess phosphate source may exert an important control on the rate of nitrogen fixation.Such a dependence implies a central role for the export of SAMW in the shallow limb of the MOC.The framework suggested by the modeling experiments also provides a context for results from a recent paleoclimatology study (Anderson et al., 2009).Anderson et al. (2009) hypothesize that the termination of the last ice age was linked to an increase in Southern Ocean upwelling and the resultant efflux of carbon from the deep ocean to the atmosphere.The primary evidence used to support this hypothesis is the enhanced burial of biogenic opal in the Southern Ocean and equatorial Pacific, which occured contemporaneously with the increase in atmospheric CO 2 and the onset of deglaciation.The hypothesized mechanism is an increase in Southern Ocean upwelling, which brings deep carbon to the surface ocean south of the ACC with the upwelling of Circumpolar Deep Water and also advects a pulse of silicic acid across the ACC into the SAMW formation region.As the carbon outgasses south of the ACC and enters the atmosphere to warm the planet, the silicic acid flux fuels a diatom bloom in the SAMW formation region and -contemporaneously by geological standards -in the equatorial Pacific.The inference from our model experiments that Southern Ocean nutrients fuel a large portion of low latitude productivity and are consumed on decadal time scales offers a framework for understanding this paleo-record.
Having summarized the major insights of our model experiments and explored their connections with recent work, we recognize a list of challenges that await future study.First, for a clearer picture of the preformed nutrient content of the interior ocean, it would be helpful to separate nutrients that cycle through the surface mixed layer during water mass formation and those that enter water masses via interior mixing and sinking.Second, the MOM3 suite is a set of coarse-resolution models that depend on parameterizations for all mesoscale dynamics and are known to yield qualitatively and quantitatively different results when compared to eddy-resolving models (e.g.Hallberg and Gnanadesikan, 2006).Thus, work with eddy-resolving models should reveal different dynamics and possibly lead to different pathways for nutrients.Third, our Figures

Back Close
Full models are configured with one vertical mixing profile for the global ocean, which was altered only in P2A in order to simulate higher Southern Ocean mixing.Such uniformity in mixing does not capture the observed enhancement of mixing in regions of rough topography, such as tidal mixing (St. Laurent and Simmons, 2006) which is parameterized in more recent GCMs (Simmons et al., 2004).Finally, the OCMIP-2 biology is not fully prognostic because it is forced towards observations at the surface.Thus, the models cannot represent changes to biology at the surface forced by the models' differing physics, a challenge that can be tackled with fully prognostic biological models.Future efforts could also move beyond our steady state solution and address temporal variability at time scales other than seasonal, incorporate ocean-atmosphere coupling, and examine the hypothesized alterations in the nutrient return pathways under different climate scenarios.Introduction

Conclusions References
Tables Figures

Back Close
Full  • N (M n ) and 32 • S (M S ) using hydrographic data; details in Lumpkin and Speer (2007) and Talley (2008).( ) ) in excess of a seasonal climatology (PO Obs 4 ) (Louanchi and Najjar, 2000) fuels primary productivity (J Prod ), thus restoring the surface model to climatological values over a time scale of τ=30 days (Eq.1).The resulting productivity has two fates: two thirds becomes a source term for DOP (J DOP , Eq. 2), and one third fuels a parameterized sinking flux of organic phosphorus, that is shunted to 75 m (F 75 , Eq. 3).Below 75 m, the sinking flux decays as a function of depth (Eq.4).The vertical divergence of this sinking flux is a source term of PO 3− 4 (Eq.5).DOP is remineralized to PO Figures    2, as labeled at the top of each map.Note that we refer to preformed SAMW phosphate as phosphate that comes directly from the tagging region without ever fueling productivity, while remineralized SAMW phosphate derives either from DOP tagged in the SAMW source region and remineralized to phosphate or SAMW phosphate that has fueled primary productivity at least once (see Sect.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | proposed a simple model that nicely synthesizes these two mechanisms into a single framework by relating the large-scale MOC to winds, eddies, and diapycnal mixing, as schematized in Fig. 2. The underpinning of the theory is that along-isopycnal eddy transport, diapy-Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | excess of a seasonal PO 3− 4 climatology(Louanchi and Najjar, 2000) fuels primary productivity, represented by restoring the surface model PO 3− 4 to climatological values over a time scale of τ=30 days.The resulting productivity has two fates: two thirds becomes a source term for DOP and one third fuels a parameterized sinking flux of organic phosphorus that is shunted to 75 m.Below 75 m, the sinking flux decays as a function of depth and its vertical divergence is a source term of PO 3− 4 .DOP is remineralized to PO 3− in Fig. 4 and its caption.The physical model is spun up to a steady circulation before the dyes are released.All presented results are averages between year 190 and 200 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the Southern Ocean that results from iron and light limitation, because it allows productivity to remove surface nutrients only when model the Subantarctic Zone where SAMW is formed, the modeled SAMW is subducted with some of the highest preformed nutrient concentrations in the global pycnocline.This result is independent of the physical model; rather, it is set by the PO 3− 4 observations in this region.The degree to which these nutrients spread from the SAMW source region to fuel downstream Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , right column); thus, nutrients reaching the euphotic zone from any source region are utilized at least once, if not many times, before entering a new tagging region.Nutrients that cycle through the North Pacific and North Atlantic support relatively little low latitude productivity (Fig. 11).Their most important contribution is in the northern half of the subtropical North Atlantic and North Pacific.As the Northern Hemisphere nutrients are advected southward of 40 • N, they are promptly consumed and exported.When the sinking flux of the phosphorus penetrates past the 27.4 isopycnal, it is retagged as deep low latitude phosphate, which limits the number of times the northern-tagged phosphate can fuel productivity via recycling.To a lesser degree, the Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Stommel, H. and Arons, A. B.: On the abyssal circulation of the world ocean-I.Stationary planetary flow patterns on a sphere, Deep-Sea Res., 6, 140-154, 1961.Talley, L. D.: Freshwater transport estimates and the global overturning circulation: Shallow, deep and throughflow components, Prog.Oceanogr., 78, 257-303, 2008.Thompson, D. W. J. and Solomon, S.: Interpretation of recent Southern Hemisphere climate Discussion Paper | Discussion Paper | Discussion Paper | Table 1.Description of the four ocean GCMs.Model acronyms are described in the text.Differences from LL are highlighted in bold.Discussion Paper | Discussion Paper | Discussion Paper |

3
Maximum global overturning at 48 • N and 32 • S calculated from the Ocean Comprehensive Atlas (OCCA), produced at the Massachusetts Institute of Technology (MIT) by the Estimating the Circulation and Climate of the Ocean (ECCO) group.The OCCA Atlas is publicly available (http://www.ecco-group.org/)for the data-rich Argo period from 2004 to 2006 (Forget, 2009).OCCA is an ocean climatology produced by calculating the least squares fit of a global full-depthocean and sea-ice configuration of the MIT general circulation model (MITgcm) to satellite and in situ data.Discussion Paper | Discussion Paper | Discussion Paper |

Fig. 1 .Fig. 2 .
Fig. 1.Global nutrients at the ocean's surface and in the SAMW layer.(a) Surface phosphate from WOA01 (Conkright et al., 2001).Surface Si*=Si(OH) 4 -NO − 3 from WOA01.(c) Si* on the 26.8 isopycnal, at the center of the SAMW layer.In panels (a) and (b), black hatches indicate regions between 65• S and 65• N where annual maximum mixed layer depths exceed 400 m, where mixed layer depths are taken from de Boyer Mont égut et al.(2004).Panel (c) is recreated afterSarmiento et al. (2004).The white area represents the region where the annual mean surface density exceeds 1026.8 kg m −3 .

Fig. 3 .
Fig. 3.A schematic showing the governing equations of the OCMIP-2 biological model.Terms that begin with the letter J are source/sink terms (black text), while terms that begin with the letter F are flux terms (grey text).At depths shallower than 75 m, model PO 3− 4 (PO model 4

Fig. 4 .Fig. 5 .Fig. 6 .Fig. 7 .Fig. 8 .
Fig. 4. A schematic of the experimental design used to examine the nutrient return pathways to the low latitude surface ocean.Each panel represents the deployment of a regional tag in zonal cross-section.The curvy lines represent selected isopycnals, as labeled at the top of each panel.In the black area, the dye tag, phosphate, and DOP are tagged according to the following conventions: (A) the Northern tag (which was further subdivided into North Atlantic and North Pacific) set on all water north of 40 • N. (B) The Deep Low latitude tag, set for all water with potential densities greater than 27.4 between 40 • N and a southern boundary defined by the latitude where the 26.5 isopycnal intersects 200 m.(C) The SAMW tag, set between that same boundary and the 27.1 isopycnal.(D) AAIW tag, set between the 27.1 and 27.4 isopycnals southward of the same boundary.(E)Southern tag for all water denser than AAIW and southward of the boundary.In the white areas, the tag is destroyed and set to its new region.In the gray areas, all tags behave according to their conservation equation.
Fig. 10.The fraction of total phosphate above the 26.8 isopycnal comprised by preformed SAMW phosphate (top panels) and remineralized SAMW phosphate (bottom panels) in the 4 models described in Table2, as labeled at the top of each map.Note that we refer to preformed SAMW phosphate as phosphate that comes directly from the tagging region without ever fueling productivity, while remineralized SAMW phosphate derives either from DOP tagged in the SAMW source region and remineralized to phosphate or SAMW phosphate that has fueled primary productivity at least once (see Sect. 2.3).

Fig. 11 .
Fig.11.The zonally-averaged fraction of primary productivity fueled by phosphate (pre-formed+remineralized) from each of the tagging regions (left column) and by the remineralized components of each tagged phosphate between 30 • S and 30 • N (right column) for the four models described in Table1.Note the panels on the left show the full latitude range of the model ocean, while the panels on the right show only the low-latitudes.The contribution of the preformed components of tagged phosphate to sustaining low latitude productivity is the difference between the contributions of the total tagged phosphate (left) and the remineralized component of the tagged phosphate (right).Negative values reflect tagged DOP that is respired to phosphate without having fueled primary productivity before entering a destruction region.The lines refer to phosphate from the North Pacific (red solid), North Atlantic (red dotted), Deep low latitude (green solid), SAMW (black solid), AAIW (black dotted), and Southern (magenta solid) regions (refer to Fig.4for tagging protocol).

Fig. 12 .Fig. 13 .
Fig. 12. Vertical fluxes of SAMW dye (in fractional concentration per m 2 s) at the depth of the 26.5 isopycnal.Upper row for the HH model and lower row for P2A.The left hand panels are the vertical advective flux (wC, where w is the vertical velocity and C is the concentration of the SAMW dye tag).The right hand panels are the vertical diffusive flux (κ v d C/d z, where κ v is the coefficient of diapycnal diffusion).The colorbar applies to all panels, and positive numbers represent an upwards flux.The line plots are the zonal means of the maps (black for vertical advection and blue for vertical diffusion).

Table 2 .
Comparison of the pycnocline and MOC characteristics of the four models from the MOM3 suite and with estimates from observations.

Table 3 .
Percentage contribution of each tagged water mass to the upper ocean volume (above the 26.8 isopycnal) between 30 • S and 30 • N in bold lettering.Listed in parentheses are the percentage contributions of preformed tagged phosphate (first number) and the remineralized tagged phosphate (second number) to the total phosphate inventory above the 26.8 isopycnal.The columnwise sum of the preformed and remineralized contributions (both numbers in parentheses) should be close to 100% of the total phosphate reservoir.However, some columns may sum to less than 100% due to roundoff error and because part of the Bay of Bengal is only partially tagged.