The cycling of nitrogen (N) controls the fertility of the global ocean1,2 and may influence the ocean’s role in climate3. The ultimate source of marine N to the ocean is N2 fixation, whereby specialized plankton, diazotrophs, transform inert N2 gas into bioavailable ammonium. This process occurs dominantly in the warm, sunlit surface of the (sub)tropical ocean. The ultimate sink for marine N is denitrification, the bacterial reduction of oxidized N species to N2 gas in oxygen-deficient waters and sediments. Considerable uncertainty exists regarding the magnitude of the global N source and sink terms, including whether these are balanced on timescales of ocean mixing4,5,6,7. An incomplete understanding of the controls on N2 fixation, particularly at the regional scale, likely contributes to the apparent discrepancies in estimates of oceanic N loss and gain.

A stoichiometric phosphorus (P) excess relative to N, termed ‘excess P’, is a condition proposed to favour N2 fixation6. Much of the global surface ocean hosts excess P8, which ultimately derives from denitrification in oxygen-deficient zones and sediments6,9. The South Atlantic receives excess P both remotely and regionally, in the latter case following seasonal N loss in the northern Benguela upwelling system (NBUS)6,10,11, yet this basin appears to support little N2 fixation. Direct and geochemical estimates of South Atlantic N2 fixation rates, although notably sparse, range from 1.8 to 8.4 Tg N.a−1, with disagreement as to the regional distribution6,12,13,14,15,16,17 (Table 1). These low rates have been attributed to iron limitation of diazotrophs9,14,17,18,19 as the South Atlantic receives at least a hundred-times less aeolian iron than the North Atlantic (3 ± 2 versus 896 ± 14 nmol iron.m−2.d−1)20 where N2 fixation is considerably higher, with estimates ranging from 24 to 47 Tg N.a−1 6,9,1216,21 (Table S1). The excess P that goes largely unused in the South Atlantic traverses the basin in its equatorward-flowing surface layer, ultimately fuelling N2 fixation in the (sub)tropical North Atlantic following its entrainment into iron-replete, P-limited surface waters9,14,22. Nevertheless, some N2 fixation likely occurs in the South Atlantic, yet its magnitude and distribution, and the controls thereon, remain poorly characterized.

Table 1 Existing areal N2 fixation rate estimates (Tg N.a−1) for the South Atlantic.

The estimation of N2 fixation has often relied on ocean models, most of which exploit regional variations in the nitrate-to-phosphate (NO3:PO43−) ratio6,12,13 to derive rates from inverse circulation fields. Variations in N:P are typically quantified by the parameters N* and P*6,23,24, defined here as [NO3] − 15.5 × [PO43−]23 and [PO43−]−[NO3]/15.56, respectively, where 15.5:1 is the mean N:P ratio of nutrients supplied to the tropical South Atlantic thermocline (SI.1). The conceptual basis for the estimation is that diazotrophy pairs the production of newly-fixed N with the assimilation of P. N2 fixation thus removes P but not N from surface waters and leads to a sinking organic matter (OM) flux with an N:P ratio that is higher than that of the net supply of nitrate and phosphate to the surface6,23,25. However, the N:P ratio of plankton biomass, and thus of sinking OM, varies regionally26, complicating the calculation of N2 fixation rates based on nutrient stoichiometry27. Recent modelling efforts have attempted to address this shortcoming by including variable, yet still prescribed, OM N:P ratios13.

Nitrate isotope ratio measurements provide an important alternative source of information for diagnosing N2 fixation14,28,29. N2 fixation introduces nitrate to the thermocline with a δ15N of −2 to 0‰28,30,31,32, distinct from that of deep-ocean nitrate (≥5‰33; δ15N = [(15N/14N)sample/(15N/14N)ref − 1] × 103, in ‰ versus N2 in the air). Negative excursions in source-to-thermocline nitrate δ15N can thus be used to diagnose and quantify N2 fixation28. The N isotope ratios have some major advantages over N:P ratios in this regard. In particular, the δ15N of newly-fixed N is relatively well-constrained30,31, whereas the N:P ratios of both diazotrophs and non-N2-fixing plankton are uncertain and potentially dynamic25,26,27.

Here we analyse nitrate isotope ratios and nutrient stoichiometries for the tropical South Atlantic Ocean based on new and existing measurements14,34,35,36 (see ‘Methods’). These data reveal substantial N2 fixation in the Angola Gyre to the east of the basin and none in the west, in contrast to current notions that N2 fixation occurs predominantly in the western tropical South Atlantic6,13 (Fig. 1). We hypothesize that the confluence of remote and regional sources of excess P and iron supplied proximately from the ocean margin determines the incidence of N2 fixation in the Angola Gyre. Our findings have implications for regional variations in N2 fixation and its coupling to N loss throughout the global ocean37,38.

Fig. 1: Surface P* across the South Atlantic Ocean.
figure 1

Station locations (coloured symbols) overlaid on an annual climatology of the South Atlantic surface P* concentration (=[PO43−]–[NO3]/15.5; SI.1). As indicated in the legend, circles represent stations sampled during the CoFeMUG cruise in November/December 200714,34, triangles show stations sampled during the MET131 cruise in October 2016, and diamonds indicate stations sampled during the CLIVAR/GO-SHIP A16N cruise in September/October 201314,49. Red symbols show Angola Gyre stations and blue symbols represent western basin stations. The red cell indicates the approximate position of the Angola Gyre, and the white arrows show the circulation of thermocline waters in the tropical South Atlantic Ocean. The black box encloses P* values calculated from the CoFeMUG data set, while the other P* data are from WOA188 (SI.1). The dotted light-grey contours show the distribution of recent modelled water column-integrated N2 fixation rates for the South Atlantic(1, 25, 50 and 100 mmol N.m−2.a−1)13. The South Atlantic Tropical Gyre is defined by the sSEC, SEUC and AC, where sSEC is south South Equatorial Current, SEUC is South Equatorial Undercurrent, and AC is Angola Current. Additionally, nSEC is north South Equatorial Current, EUC is Equatorial Undercurrent, eSEC is equatorial South Equatorial Current, cSEC is central South Equatorial Current, SECC is South Equatorial Countercurrent, GCUC is Gabon-Congo Undercurrent, ABFZ is Angola-Benguela Frontal Zone (indicated by the parallel wavy lines), NBUS is northern Benguela upwelling system, BC is Benguela Current, BrC is Brazil Current, and STsG is Subtropical subgyre.

Results and discussion

Properties of the Angola Gyre

The Angola Gyre, a permanent upwelling feature embedded within the larger-scale cyclonic South Atlantic Tropical Gyre39,40,41, was encountered east of 7.5°W along a zonal transect of the Atlantic at 6-15°S (Fig. 1; CoFeMUG14,34 and MET131 cruises; see ‘Methods’). The gyre is characterized by a shoaling of the thermocline, defined as the waters between potential density anomalies (σθ; in kg.m−3) 26.2 and 27.0, overlying an oxygen minimum zone where apparent oxygen utilization (AOU = O2saturated − O2observed) is >240 µM40,41,42 (Fig. 2a, b, f, g). High rates of primary productivity fuelled by persistent nutrient upwelling drive elevated OM export from the Angola Gyre mixed layer, which is remineralized in the aphotic zone below the surface43,44. A subsurface iron plume emanating from the nearby low-oxygen marginal sediments penetrates the Angola Gyre (dFe ~0.8–2 nM; Fig. 2c). Intense remineralization and repeated cycles of supply and export retain this bioavailable iron within Angola Gyre waters34. Local remineralization also yields a nitrate concentration maximum (>40 µM) beneath the thermocline (σθ > 27.0), contrasting with the lower concentration of sub-thermocline nitrate (<35 µM) and AOU (<180 µM) in the tropical western basin (west of 7.5°W; Figs. 2b, g and 3a).

Fig. 2: Zonal trends in physical and biogeochemical properties across the tropical South Atlantic.
figure 2

Gridded section plots across the CoFeMUG (30°W to 12°E; ad) and MET131 (6°S to 15.1°S; fh) transects of a, f conservative temperature in °C, where the red contour denotes the 12 °C isotherm, the shoaling of which between 7.5°W and 12.2°E marks the approximate zonal extent of the Angola Gyre40; b, g apparent oxygen utilization (AOU) in μM; c dissolved iron concentrations in nM34 overlaid by N* contours (black) in μM; and d, h nitrate δ15N in ‰ versus N2 in the air (CoFeMUG data from ref. 14). e, i shows a recent modelled water column-integrated N2 fixation rate along 11°S (e) and 11°E (i) in mmol N.m−2.a−113, with the western basin values in blue and the Angola Gyre values in red. Black (grey) contours on a, b, d, f, g, h (e and h) are potential density surfaces indicating tropical water masses, where TSW is Tropical Surface Water, TW is Thermocline Water, SAMW is Subantarctic Mode Water and IW is Intermediate Water (SI.2). Small grey circles in c, d and h indicate the sampling resolution.

Fig. 3: Characteristics of tropical South Atlantic nitrate.
figure 3

Density profiles of a nitrate concentrations in µM; b nitrate δ15N in ‰ versus N2 in air; c N* (= [NO3] – 15.5 × [PO43−]; SI.1) in µM; and d the depth-specific concentration of newly-fixed nitrate in μM in the Angola Gyre. In ac, warm (cool) coloured profiles represent stations from the Angola Gyre (tropical western basin). Round markers show data from the CoFeMUG transect while triangle markers represent data from the MET131 transect. The legend provides station longitude. The grey shading on ad indicates the density range of TW. In c, the dashed (solid) black line indicates the mean gridded N* for the Angola Gyre (western basin) stations; only CoFeMUG data are shown here because of the higher depth resolution of this sample set. d shows the concentration of newly-fixed nitrate in the Angola Gyre, computed by multiplying the depth-specific fraction of newly fixed nitrate (fdepth_specific; equation S3b) by the corresponding nitrate concentration at each depth. To calculate fdepth_specific, a nitrate δ18O-based correction is applied to the δ15N data to remove the phytoplankton nitrate assimilation signal from Tropical Surface Water (SI.7). Light grey data show results from scenario 1 while dark grey data show scenario 2 (see text for details). The mean of each scenario is indicated by the bold line of corresponding colour.

Nitrate δ15N across the transect is consistently high in Tropical Surface Water (σθ < 26.2) due to isotopic fractionation during nitrate assimilation by phytoplankton, which raises nitrate δ15N (and δ18O45,46; which, in ‰ versus VSMOW = [(18O/16O)sample/(18O/16O)ref − 1] × 103; SI.3) to >12‰. The mean nitrate δ15N in the underlying thermocline is 5.7 ± 0.7‰ in the Angola Gyre, reaching as low as 4.6‰, a salient decrease from the western tropical basin where the average δ15N of thermocline nitrate is 6.3 ± 0.1‰ (Figs. 2d, h and 3b). Subantarctic Mode Water (SAMW; σθ = 27.0–27.25), recently formed in the Southern Ocean, underlies the thermocline of the tropical South Atlantic and is the ultimate source of nutrients to its surface waters47. In the western tropical basin, SAMW nitrate δ15N is 6.2 ± 0.1‰, indistinguishable from the δ15N of SAMW nitrate beneath the Angola Gyre (6.4 ± 0.1‰; t(150) = −12.05, p = 0.01) (Figs. 2d, h and 3b).

Mean SAMW N* is 0.0 ± 0.1 μM across the basin, increasing in the Angola Gyre thermocline to 0.7 ± 0.9 μM, coincident with the source-to-thermocline decline in nitrate δ15N. N* decreases into the western tropical basin thermocline, to −0.4 ± 0.5 μM (Fig. 3c and SI.1 and 4).

Zonal trends in the tropical South Atlantic: evidence for an exogenous source of N to the Angola Gyre

Thermocline nitrate δ15N in the western tropical South Atlantic basin is similar to that of the underlying source (6.3 ± 0.1‰ versus 6.2 ± 0.1‰) while the δ15N of nitrate in the Angola Gyre thermocline is distinctly lower, by 0.7–1.8‰ (Figs. 2d, h and 3b). Since surface-water nitrate is generally completely consumed, and given that nitrification in the ocean interior typically competes with no other process—thus negating the impact of isotopic fractionation during remineralization in the subsurface—remineralized OM produced from Tropical Surface Water will return nitrate to the thermocline with a δ15N that is indistinguishable from the underlying SAMW source29,48,49. The fact that the δ15N of thermocline nitrate in the Angola Gyre is lower than SAMW nitrate δ15N signals an exogenous source of N to surface waters, the consumption and subsequent remineralization of which yields low-δ15N nitrate in the Angola Gyre thermocline28,32.

Zonal and vertical changes in nutrient stoichiometry corroborate the notion of an exogenous source of N to the Angola Gyre. The SAMW-to-thermocline N* increase in the gyre is consistent with the addition of N in stoichiometric excess of P (i.e., relative to the N:P ratio of the regional nutrient supply, 15.5:1) while the SAMW-to-thermocline N* decrease in the western tropical basin suggests that sinking OM may be lower in N:P than the regional nutrient supply (SI.4).

Despite the existing data pointing to complete nitrate consumption in the Angola Gyre surface, instances of incomplete nitrate assimilation could occur in this perennially-upwelling feature, and these could cause the loss of high-δ15N nitrate (by advection), with low-\({{{{{\rm{\delta }}}}}}\)15N OM retained and remineralized in the gyre thermocline29,50. However, the regeneration of OM deriving from partial nitrate assimilation, which would have an N:P of ~15.5:1, could not yield the high N* observed in the thermocline. Additionally, mixed-layer nitrate was measurable at only three of 19 stations at the time of our sampling, reinforcing the view that nitrate is seldom unconsumed in Angola Gyre surface waters in spring, and likewise in autumn44. The δ15N of thermocline nitrate could arguably be lowered by the lateral export of high-δ15N dissolved organic N from the gyre surface. However, high-δ15N dissolved organic N export cannot account for the regional increase in thermocline N*. We thus rule out partial nitrate assimilation and dissolved organic N export as explanations for the low \({{{{{\rm{\delta }}}}}}\)15N of Angola Gyre thermocline nitrate.

Atmospheric deposition can introduce exogenous N to the surface ocean with a δ15N < 0‰51 and an N:P ratio >1000:152, potentially explaining both the low-δ15N nitrate and high N* of the Angola Gyre thermocline. However, estimates of the N deposition flux to the South Atlantic are orders of magnitude too low to account for our observations52 (SI.5). We thus conclude that the exogenous N in the Angola Gyre thermocline derives from N2 fixation.

Origin of the N2 fixation signal in the Angola Gyre thermocline

The low δ15N of Angola Gyre thermocline nitrate could originate, at least partly, in the tropical North Atlantic basin where N2 fixation rates are elevated14,17,28,49, and be transported into the gyre by equatorial feeder currents (Fig. 1). We address this possibility by estimating the concentration-weighted influx of nitrate δ15N by the three Angola Gyre feeder currents: the Equatorial Undercurrent, the South Equatorial Undercurrent, and the South Equatorial Countercurrent40. We estimate the δ15N of the Equatorial Undercurrent, South Equatorial Undercurrent, and South Equatorial Countercurrent end-members to be 5.7‰, 6.0‰, and 6.3‰, respectively (‘Methods’; SI.6). The relative contribution of each current to Angola Gyre thermocline nitrate based on transport volumes42 yields a mean flux-weighted nitrate-δ15N of 6.1 ± 0.1‰. While lower than the δ15N of underlying SAMW (6.4‰), this end-member is not low enough to decrease Angola Gyre thermocline nitrate δ15N to 5.7‰ (nor to the observed minimum of 4.6‰). Thus, some amount of the low-δ15N signal must derive from local N2 fixation.

Our conclusion that N2 fixation occurs in the Angola Gyre is further supported by the slightly lower δ15N of nitrate at the depth of the AOU maximum compared to that of core SAMW (6.2‰ versus 6.4‰; Fig. 2b, g, d, h). Given that the provenance of the high AOU signal is in situ primary production40, OM remineralized at the AOU maximum is lower in δ15N than SAMW nitrate, consistent with its low δ15N deriving from N2 fixation in Angola Gyre surface waters. Additional support comes from direct rate measurements made during the CoFeMUG cruise53 and other rate data from near the Angola Gyre15,54,55, which indicate N2 fixation rates ranging from undetectable to ~170 μmol N.m−2.d−1. Regional observations of diazotrophs (Trichodesmium spp. being the most abundant) and diatoms hosting N2-fixing symbionts (Rhizosolenia and Chaetoceros spp.) further substantiate the occurrence of N2 fixation in the Angola Gyre53,54,56,57.

Quantifying the N2 fixation rate in the Angola Gyre

Using our nitrate δ15N data, we estimate the annual rate of N2 fixation in the Angola Gyre, considering two scenarios: (1) the low-δ15N of thermocline nitrate is generated solely from in situ N2 fixation and superimposed on SAMW nitrate, and (2) local N2 fixation contributes the fraction of low-δ15N thermocline nitrate that is not supplied by the equatorial feeder currents. For both scenarios, we estimate the fraction (f) of newly fixed nitrate in the Angola Gyre as:

$$f=\frac{{{{{{{\rm{\delta }}}}}}}^{15}{{{{{{\rm{N}}}}}}}_{{{{{{\rm{AGmean}}}}}}}-{{{{{{\rm{\delta }}}}}}}^{15}{{{{{{\rm{N}}}}}}}_{{{{{{\rm{scenario}}}}}}\_{{{{{\rm{mean}}}}}}}}{{{{{{{\rm{\delta }}}}}}}^{15}{{{{{{\rm{N}}}}}}}_{{{{{{{\rm{N}}}}}}}_{2}{{{{{\rm{fix}}}}}}}-{{{{{{\rm{\delta }}}}}}}^{15}{{{{{{\rm{N}}}}}}}_{{{{{{\rm{scenario}}}}}}\_{{{{{\rm{mean}}}}}}}}$$

δ15NAGmean is the mean concentration-weighted δ15N of Angola Gyre nitrate for the upper water column (σθ ≤ 27.0), estimated to be 5.7 ± 0.4‰ (5.5–5.9‰; 5th−95th percentiles; Methods), δ15NN2fix is the δ15N of nitrate regenerated from diazotrophic OM, −1‰30,31, and δ15Nscenario_mean is the scenario-specific mean concentration-weighted δ15N of nitrate supplied to the Angola Gyre thermocline (6.4 ± 0.1‰ and 6.1 ± 0.1‰ for scenarios 1 and 2, respectively).

For scenario 1, f = 9% (6–12%; 5th−95th percentiles) and for scenario 2, f = 6% (2–9%; 5th−95th percentiles). On average, therefore, 6–9% (associated with scenarios 2 and 1, respectively) of the nitrate above SAMW originates from local N2 fixation. This mass-balance method neglects particulate organic N, which is reasonable given that this pool contributes negligibly to total N, accounting for <1% of the fixed N reservoir above σθ = 27.0. Our approach is validated by the concentration-weighted δ15N that it yields for nitrate above σθ = 27.0 in the western tropical basin of 6.3 ± 0.1‰ (6.2–6.4‰; 5th−95th percentiles), indistinguishable from that of SAMW (6.2 ± 0.1‰). This correspondence further confirms that N2 fixation is negligible in the tropical waters west of the Angola Gyre.

To visualize the depth distribution of newly-fixed nitrate in the Angola Gyre, we calculate fdepth_specific (i.e., the fraction of the nitrate pool at each depth that is newly fixed) for each sample above σθ = 27.0, after removing the signal of phytoplankton assimilation from nitrate δ15N using the coincident nitrate δ18O data (‘Methods’; SI.7). This exercise reveals that a substantial fraction of Tropical Surface Water nitrate is newly-fixed, which is not apparent from the nitrate δ15N profiles because fractionation during nitrate assimilation overprints the isotopic signal of newly nitrified N (Fig. 3b, d). It is not appropriate, however, to integrate these depth-specific estimates to yield a measure of the newly-fixed N inventory. This is due to the likelihood of ‘double-counting’—that is, low-δ15N nitrate assimilated in the surface is returned to the thermocline when the resulting OM is exported and remineralized, such that at the scale of the water column, correction of shallow nitrate-δ15N for nitrate assimilation is unnecessary28. Instead, the mass-balance approach represented by Eq. 1 is most appropriate for estimating the water column burden of newly fixed N.

Multiplying f (Eq. 1) by the mean Angola Gyre nitrate concentration for σθ ≤ 27.0 indicates that local N2 fixation supplies 313–641 mmol N.m−2 to the thermocline nitrate reservoir. The residence time of thermocline waters in the Angola Gyre is 4.4–8.5 years40 and its areal extent is 2.6 × 1012 m2. These constraints yield an N2 fixation rate of 2.8-5.4 Tg N.a−1 (mean of 4.1 ± 2.8 Tg N.a−1) for scenario 1 and 1.4-2.6 Tg N.a−1 (mean of 2.0 ± 3.0 Tg N.a−1) for scenario 2, considering both extrema for residence time (‘Methods’; SI.8). Our estimates are equivalent to daily rates of 102–390 μmol N.m−2.d−1, comparable to incubation-based N2 fixation rates measured near the Angola Gyre15,53,54,55 (SI.9).

The Angola Gyre is a hotspot for N2 fixation in the South Atlantic

Our N2 fixation rate estimates imply that the Angola Gyre accounts for 28–108% of the highly uncertain mean South Atlantic N2 fixation rate of 5.0 ± 3.3 Tg N.a−1 (Table 1). Since the latter derives from few observations that are sparsely distributed across the basin, it would be unwise to conclude that the Angola Gyre makes a disproportionate contribution to South Atlantic N2 fixation. Nonetheless, since the Angola Gyre occupies only ~10% of the South Atlantic by area, our data indicate that it is a hotspot for N2 fixation. Conditions particular to the Angola Gyre must render this feature favourable for N2 fixation.

The phosphate-bearing, nitrate-deplete surface waters of the South Atlantic receive a limited supply of aeolian iron20, a condition proposed to limit N2 fixation across the basin9,13,14,17,38. The non-aeolian iron supply to the Angola Gyre must thus be substantial. The Congo-shelf-zone near the Angola Gyre has been reported to supply a substantive quantity of iron to the marginal ocean, equivalent to 40 ± 15% of the entire South Atlantic aeolian iron flux58, and the elevated iron concentrations observed in the vicinity of the Angola Gyre (Fig. 2c) are evidence of this sedimentary iron supply. We hypothesize that the subsurface iron plume, which is trapped by the retentive regional circulation of the Angola Gyre through repeated cycles of upward supply, export, and remineralization34, relieves diazotrophs in the overlying surface waters of iron limitation.

We thus conclude that N2 fixation occurs in the Angola Gyre because of the overlapping biogeography of excess P and bioavailable iron. While the iron supply to the Angola Gyre is local, originating from the nearby low-oxygen margin sediments, the excess P derives from both regional and remote N loss. The apparent importance of margin-derived iron echoes a growing body of work showing that sedimentary iron sources are important to the global ocean iron budget59,60,61,62. The notion that both iron and excess P together limit N2 fixation at the regional scale in the South Atlantic is compatible with previous work showing negligible rates of N2 fixation in the western tropical basin where measurements were made in P-bearing but iron-limited waters9,17,22.

The possibility of additional N2 fixation hotspots

We hypothesize that N2 fixation hotspots are inherent to ocean regions characterized by biogeochemical conditions analogous to those encountered in the Angola Gyre, possibly enhanced by retentive circulation features (SI.10). In the subtropical South Atlantic where the Brazil Current recirculation forms a subgyre adjacent to the continental shelf41 (Fig. 1), an N2 fixation hotspot should result as waters bearing a remote (i.e., Southern Ocean-derived) and regional (i.e., NBUS-derived) supply of excess P6 (Fig. 1) encounter a local supply of sediment-derived iron62. Observations of diazotroph blooms63 and low-δ15N thermocline nitrate64 in the Brazil subgyre validate this prediction. In addition, the cyclonic Guinea Dome in the eastern tropical North Atlantic receives excess P from local upwelling of P-rich subsurface waters8 and iron from dust deposition, possibly augmented by an iron flux from the nearby low-oxygen margin sediments60. Direct N2 fixation rate measurements15,16, the presence of diazotrophs16,65, and low-δ15N thermocline nitrate66 in the vicinity of the Guinea Dome are consistent with this region also being an N2 fixation hotspot.

The surface waters above or just offshore of the ocean’s major suboxic zones would, in general, appear well-suited to host hotspots of N2 fixation. Denitrification in both the shallow subsurface water column and the adjacent sediments generates an N deficit (i.e., a P excess)6. At the same time, the adjacency of these zones to low-oxygen margin sediments, coupled with water column suboxia overlying some of the sediments, supplies iron to the water column59,67,68. These regions are also characterized by upwelling, which would transport both the excess P and the iron into the surface mixed layer. In these surface waters or just offshore, phytoplankton assimilation leaves nitrate-free but P- and iron-bearing surface waters6,67,68,69, a seemingly ideal recipe for N2 fixation.

However, the existing data are ambiguous as to the rates of N2 fixation above and nearby the suboxic zones. Observations from the Arabian Sea support elevated N2 fixation70,71,72 while incubation-based measurements in the eastern tropical Pacific suggest negligible to extremely high rates38,73,74,75,76,77,78. Notably, incubation experiments suggest that N2 fixation occurs in the Costa Rica Dome75, a retentive feature embedded within the eastern tropical Pacific suboxic zone that is characterized by similar hydrography to the Angola Gyre. In the water column of all the suboxic zones, a decrease in nitrate δ15N is observed above the very high nitrate δ15N associated with denitrification79,80,81,82; this upward δ15N decline has been interpreted as evidence of local N2 fixation79. Additionally coupled nitrate δ15N and δ18O measurements from the eastern tropical North Pacific indicated a lower nitrate δ15N than is expected from nitrate δ18O in the upper portion of the suboxic zone, also consistent with a role for N2 fixation80. However, N cycling processes occurring within suboxic waters (e.g., nitrite oxidation) provide an alternative explanation for the observed δ15N-δ18O decoupling80,81,82,83,84, complicating the use of the nitrate isotopes to evaluate N2 fixation above and near the major suboxic zones. Importantly, the Angola Gyre is not complicated by the same suboxic zone N cycling processes, allowing us to pinpoint the role of N2 fixation in this system. Our findings for the Angola Gyre thus suggest that the nitrate δ15N decline in the shallow subsurface of the suboxic zones may have a similar origin. The nitrate isotope data of the major ocean suboxic zones should be revisited with this new perspective.

Implications of regional N2 fixation hotspots

N2 fixation has been proposed to be spatially and/or quantitatively coupled to N loss at the basin scale6,10,85. Our estimate of N2 fixation in the Angola Gyre is notably comparable to N loss from the NBUS (1.4–2.5 Tg N.a−1)11,86. In this regard, the generation and advection of excess-P waters from the NBUS have been hypothesized to fuel N2 fixation in the vicinity of the Angola Gyre10, prompting the suggestion that N sources and sinks in the south-eastern Atlantic are coupled. However, surface P* is perennially elevated across the (sub)tropical South Atlantic, including in Angola Gyre and Brazil subgyre surface waters6,9 (Fig. 1). Accordingly, it is not clear that the additional flux of excess P from the NBUS, by itself, stimulates N2 fixation in the Angola Gyre. Rather, N2 fixation would likely occur in the Angola Gyre regardless of whether it receives excess P from the NBUS, provided that the iron supply is sustained. That is, the greatest influence of the NBUS on N2 fixation in the Angola Gyre may be through its augmentation of the iron supply. In summary, our findings argue that a P-driven coupling of N2 fixation to denitrification is contingent on a supply of iron, which is most easily achieved along coastal suboxic zones59.

Because the South Atlantic hosts a basin-wide surface P excess, a future increase in the iron supply to the south-eastern Atlantic due to expanding low-oxygen margins34,62,87 and/or a climate change-driven increase in aeolian deposition88 could enhance N2 fixation in and beyond the Angola Gyre. Consequently, the meridional flux of excess P across the Atlantic could decrease, engendering a shift in the dominance of N2 fixation from the North to the South Atlantic.

Lessons for estimating basin-scale N2 fixation rates

Our observations contradict diagnoses of the regional distribution of N2 fixation in the South Atlantic computed from nutrient climatologies and velocity fields from ocean general circulation models6,13 (e.g., P* convergence). This inconsistency highlights the current inadequacy of nutrient fields to deliver regional representations of N2 fixation. Modelled distributions suggest that N2 fixation predominantly occurs in the western and central tropical South Atlantic at a rate of 7.5-8.3 Tg N.a−1, yet the nitrate-δ15N data from these longitudes bear no evidence of the remineralization of low-δ15N, diazotroph-derived OM (Figs. 2d and 3b). Thus, while the model diagnoses offer a powerful framework from which to estimate basin-scale N2 fixation rates and conceptualize the global drivers thereof6, some of the fundamental assumptions inherent to this approach can lead to erroneous diagnoses, potentially underpinning the discrepancy between our observations and existing model solutions. First, coarse resolution (2–4°) general circulation models used to simulate current velocities and transports typically resolve only gross circulation features. Although this limitation is widely acknowledged6,12,13, the implications for estimates of N2 fixation rates and distributions may be important, particularly at the regional scale. In particular, such general circulation models are unable to reliably simulate ocean-margin environments12,37,61. Second, the available South Atlantic nutrient climatologies include few data, particularly in the central basin, such that the choice of data set can considerably alter the modelled regional distribution of N2 fixation6. Third, diagnoses of N2 fixation from inverse models are highly sensitive to the N:P ratios of plankton biomass and exported OM. These properties are difficult to measure and are thought to be dynamic and sensitive to environmental conditions25,26,27,89. Elemental ratios of OM are thus typically parameterized6,13 and may not accurately capture the plasticity of OM stoichiometry. Fourth, to parameterize dissolved OM in models, its production and degradation rates also have to be prescribed. However, few measurements of such rates exist—particularly with respect to dissolved organic phosphorus, which diazotrophs can assimilate when ambient phosphate concentrations are low90,91—resulting in modelled dissolved OM dynamics that are poorly constrained and potentially erroneous92. Finally, the analytical limit of quantification for [PO43] and [NO3] is such that the propagated error on P* is ≥0.1 μM93 (SI.11). Interpretations of ≤0.1 μM differences in P* may thus inaccurately diagnose the rate and/or distribution of N2 fixation94.

Pending enhanced observational coverage and improved model simulations of margin environments, convergent N2 fixation distributions for the South Atlantic and other dynamic ocean-margin regions will remain elusive. Our work highlights the utility of nitrate isotope ratios for identifying and estimating N2 fixation, and underscores the importance of regional observations, which are necessary for understanding the regional controls on N2 fixation that ultimately determine its global distribution.

Materials and methods

Sample and data provenance

The CoFeMUG nutrient and nitrate isotope data were first published in refs. 34,14, respectively. Samples from the CoFeMUG cruise were collected onboard the R/V Knorr in November and December 2007 (Fig. 1). Seawater samples were measured for nitrate isotopes at the Woods Hole Oceanographic Institution using the denitrifier method95,96. Nitrate concentration and isotope data were first published from the CLIVAR A16N cruise in refs. 49,66. The samples were collected onboard the NOAA R/V Ronald H. Brown between August and October 2013 as part of the GO-SHIP and CLIVAR program. The shipboard-ADCP data from this cruise are provided in Firing & Hummon (2010)97. The MET131 cruise data have not been previously published. This cruise was undertaken onboard the R/V Meteor in October 2016, with nine stations sampled between 6°S–15°S and 7°E–12°E. Twenty-four 12 L Niskin bottles attached to a Sea-Bird rosette with conductivity–temperature–depth (CTD) and oxygen sensors were remotely fired over the upper 1000 m to collect seawater samples for nutrient and nitrate isotope analysis. Samples were collected in thoroughly rinsed 30 ml HDPE bottles, filtered (0.4 μm), and immediately frozen at −20 °C until analysis.

Analysis of MET131 cruise nutrient concentrations and nitrate isotopes

Nitrate and phosphate concentrations ([NO3], [PO43]) and the dual isotopes of nitrate were analysed at the Helmholtz–Zentrum Hereon in Germany. The concentrations of PO43−, nitrite ([NO2]) and nitrate plus nitrite ([NO3] + [NO2]) were measured colourimetrically using an automated continuous flow system (AA3, Seal Analytical, Germany) with a detection limit of 0.01 μM98; [NO3] was then determined by subtraction. The dual isotopes of nitrate, following nitrite removal99, were measured using the denitrifier method95,96 whereby nitrate is quantitatively converted to nitrous oxide gas (N2O) by denitrifying bacteria that lack an N2O reductase enzyme. The N and O isotope ratios of the N2O were measured using a Thermo Delta Plus XP isotope ratio mass spectrometer in-line with a GasBench II. The international reference materials, IAEA-N3100 and USGS-34101, were included in each run to enable post-run calibration of the N2O measurements to N2 in the air (δ15N) and VSMOW (δ18O). The standard deviation of duplicate measurements of δ15N and δ18O was <0.2‰ and <0.4‰, respectively.

Data gridding and interpolation

The CoFeMUG and MET131 data were gridded prior to the generation of section plots and the calculation of mean water-column values (Figs. 2 and 3c, d). The z-grid (pressure) bins focused on the upper 1000 m. Python 3.6 Xarray linear method was used as the interpolation scheme.

Nitrate δ15N transported into the Angola Gyre

Cruise A16N sampled both the Equatorial Undercurrent (EUC) and South Equatorial Undercurrent (SEUC) at 0.5°S and 3.6°S, respectively, while the CoFeMUG cruise sampled the South Equatorial Countercurrent (SECC) at 27.5–30°E (Fig. 1 and SI.6). The mean [NO3]-weighted δ15N for the EUC, SEUC, and SECC were estimated to be 5.7‰, 6.0‰ and 6.3‰, respectively. To appropriately weight the nitrate flux of each feeder current, two transport volume scenarios were considered due to the uncertainty in the contribution of the EUC to the Angola Gyre42. Scenario A considered the net transport across the northern limb of the Angola Gyre, of 8 Sv, to be apportioned between the SEUC (59%) and the SECC (41%), with no contribution from the EUC42. Scenario B considered the net transport across the eastern limb of the Angola Gyre, of 11 Sv, to be apportioned between the SEUC (43%), the SECC (30%), and the EUC (27%)42. The mean flux-weighted nitrate δ15N resulting from both transport scenarios was 6.1 ± 0.1‰; this value is used in scenario 2 throughout the study.

Assimilation-corrected, depth-specific fraction of newly fixed nitrate

At the base of the euphotic zone, co-occurring nitrate assimilation and nitrification of low-δ15N OM deriving from N2 fixation raises the δ15N of nitrate, potentially yielding an underestimation of the depth-specific fraction of newly-fixed nitrate inferred from the δ15N data (SI.7). During nitrate assimilation, nitrate δ18O and δ15N rise in a 1:1 ratio as nitrate consumption proceeds45,46 while during nitrification, nitrate δ18O and δ15N become decoupled48,102,103. The nitrate δ18O data thus reveal (1) the depth at which nitrate assimilation begins to alter to isotopic composition of the nitrate pool and (2) the magnitude of the nitrate assimilation signal, two insights that are not apparent from the nitrate δ15N data. Assuming that isotopic fractionation associated with nitrate assimilation similarly alters nitrate δ15N and δ18O as nitrate consumption proceeds45,46, we quantified the nitrate assimilation signal at each depth in the water column and subtracted it from the nitrate δ15N data, beginning at the depth where nitrate δ18O starts to rise above its mean deep-ocean value. Following the removal of the nitrate assimilation signal, the fraction of newly fixed nitrate at each depth was calculated.

Statistical analyses

To estimate uncertainty on the mean δ15N of Angola Gyre and western basin nitrate (σθ ≤ 27.0), we calculated the 95% confidence interval associated with a t-distribution (for n < 30). The same method was adopted to estimate uncertainty on f (Eq. 1) for scenarios 1 and 2. To characterize the uncertainty associated with the Angola Gyre N2 fixation rates, we ran a Monte Carlo simulation of 100k steps assuming a normal probability distribution. We report the standard deviation of the output as the error associated with the rates.