Main

The Great Atlantic Sargassum Belt (GASB)1 acts as a floating ecosystem for marine life, providing essential food and shelter for important species, such as tuna, marlin, turtles and birds2,3. However, at highest abundances, its stranding events burden coastal ecosystems and impair the well-being of coastal communities. During major GASB years, tons of Sargassum wash ashore, affect community health and tourism, and require costly management and removal efforts. Decomposing Sargassum emits hydrogen sulfide gas, thereby posing widespread health risks that are exacerbated by plastic debris caught in the algae as it travels to the Caribbean4,5. Thus, the GASB has strong and complex effects on wildlife and coastal populations.

Historical reports place large quantities of Sargassum in the Gulf of Mexico and the Sargasso Sea. In 2010, exceptionally strong westerly winds in the North Atlantic are hypothesized to have caused the export of Sargassum from the Sargasso Sea to the tropical Atlantic, where it has since been seasonally aggregated by the intertropical convergence zone (ITCZ)6. Since then, a noticeable increase in Sargassum blooms has been observed in the equatorial Atlantic, leading to more frequent stranding events across Caribbean islands7. The magnitude of Sargassum blooms is modulated interannually by various physical oceanic processes, with record-high biomasses observed in 2015, 2018, 2021 and 20221,5,7,8. There is a general consensus that shifts in nutrient availability govern the recent surge in Sargassum blooms1,9,10. However, the origin and dynamics of the nutrient sources that drive the episodic nature of the GASB and the relative importance of the different nutrients are yet to be resolved1,9,10,11,12.

The tropical North Atlantic is widely recognized as conducive to dinitrogen (N2) fixation13,14, both with regard to its warm, sunlit, well-stratified surface layer15,16 and the supply of nutrients. Equatorial Atlantic upwelling and the northwestward flow of the South Equatorial Current carry excess P into the tropical North Atlantic and the Caribbean region (Fig. 1). Whereas aeolian dust fluxes are limited in the southern equatorial Atlantic, high fluxes of dust from the Sahara and Sahel region supply abundant iron to the surface and subsurface waters of the (sub)tropical North Atlantic, even reaching the far western Atlantic Ocean and Caribbean Sea17,18. In the context of high iron availability, the supply of excess P appears to drive much of the N2 fixation in the Caribbean region today19,20,21,22,23, and geochemical evidence suggests that excess P supply has similarly controlled N2 fixation in this region over the past 160,000 years24.

Fig. 1: Coral core and water sample locations in the wider Caribbean region.
figure 1

a,The coral core locations are indicated by large triangles with black outlines, and water samples are denoted by large circles with black outlines. The small circles correspond to previous measurements taken from ref. 40. The colour scale for all symbols shows the nitrate δ15N (expressed in ‰ with respect to air) at 200 m depth. Coral core sample locations are Hog Reef, Bermuda (Bm: 32.4469° N, 64.8252° W), Cayo Santa María, Cuba (Cu: 22.6677° N, 79.0997° W), Puerto Morelos National Marine Park, Mexico (Mx: 20.8900° N, 86.8100° W), Turneffe Atoll, Belize (Bz: 17.4027° N, 87.8905° W), Caye d’Olbian, Martinique (Mq: 14.4669° N, 61.0164° W), Little Tobago Island, Trinidad and Tobago (Tb: 11.2986° N, 60.5083° W) and Cahuita, Costa Rica (Cr: 9.7514° N, 82.8192° W). b, Water-column excess P (in µM; expressed as P*, which equals PO43− − NO3−/16) at 200 m depth, based on the GLODAPv2.2022 dataset76. Newly measured nitrate δ15N and P* values from an east–west transect at 13° N were obtained on the sailing yacht Eugen Seibold during the expedition of December–March 2022/2023, and data from north–south transects were obtained on the GO-SHIP A20 (16 March to 16 April 2021) and A22 (20 April to 16 May 2021) sections. New data are indicated as large circles with black outlines, and these are plotted on the same scale as the global dataset. Both panels include the prevailing surface currents (black arrows) and the extent of the GASB (brown shaded area). Basemaps created with Ocean Data View v.5.6.3 (https://odv.awi.de).

Full size image

The growth of Sargassum in the Caribbean region requires the supply of the macronutrients N and P1,9,10,25. To maintain high growth rates, Sargassum utilizes P, for example, to produce ribosomal RNA26,27. In general, photosynthetic autotrophs in the low-latitude ocean rely largely on one of three N sources: (1) nitrate transported from the subsurface, (2) N that is recycled (largely as ammonium) from heterotrophic metabolism at the surface and (3) autotrophic ‘N2 fixation’, the conversion of N2 to ammonium. The 15N-to-14N isotope ratios of Sargassum in the Sargasso Sea are remarkably similar to those of prokaryotic phytoplankton that rely predominantly on recycled N28. However, the 15N-to-14N ratio of newly fixed N falls within the same range (Extended Data Fig. 1), and Sargassum is known to maintain an association with epiphytic N2-fixing bacteria, which transfer newly fixed N directly to Sargassum, possibly in exchange for P12,29,30. Thus, Sargassum probably acquires N through both recycled N and epiphytic N2 fixation, with the latter representing a potential competitive advantage over non-fixing phytoplankton in P-bearing, N-poor environments (that is, in the presence of excess P, that is, positive P*, where P* = PO43− − NO3−/16)19,31. Such a special strategy is particularly pertinent to seaweeds such as Sargassum, which have a higher N:P ratio than phytoplankton9,10. Thus, the mutualistic relationship of Sargassum with epiphytic N2-fixing cyanobacteria may allow Sargassum to occupy the same niche as common phytoplanktonic N2 fixers found across the Atlantic Ocean. However, little information is known about ongoing variations in N2 fixation in the region, its relationship to other nutrients or its relevance to recent GASB blooms.

In this Article we use the N isotopic composition of coral-bound organic matter to reconstruct recent changes in Caribbean N2 fixation. We measured the 15N-to-14N ratio of coral-bound organic matter (expressed as CB-δ15N = [(15N/14N)sample/(15N/14N)air − 1] × 1,000‰) from continuous coral cores spanning the period 1900–2021, with subannual to annual resolution from six locations across the Caribbean Sea (Cuba, Belize, Martinique, Mexico, Tobago and Costa Rica). We interpret these data in the context of tropical Atlantic pycnocline P* values, which both influence and are altered by N2 fixation (Fig. 1). We find a close coupling between reconstructed N2 fixation, the supply of excess P to the tropical North Atlantic/Caribbean, and Sargassum blooms over recent decades.

Coral-bound nitrogen isotopes as a tracer of N2 fixation

The production and remineralization of plankton organic matter generally results in an oceanic dissolved N:P ratio of ~16:1, also known as the Redfield ratio32. Deviations from the stoichiometric Redfield ratio can be driven by net inputs and losses of N through N2 fixation and denitrification, respectively19,33. N2 fixation increases the bioavailability of N relative to P (N:P > 16)15,34, whereas sedimentary and water-column denitrification reduce the bioavailability of N relative to P (N:P < 16). As such, denitrification generates excess P, which in turn favours diazotrophic growth when sufficient iron is available, ultimately replacing lost N19. This response of N2 fixation to N losses due to denitrification is important in maintaining the marine global N inventory and couples it to P on adequately long timescales15,24,35,36, whereas spatial aspects of this coupling are influenced by iron availability20,21,22.

The stable isotopes of N record changes in the global marine N cycle37,38. N2 fixation introduces fixed N into the ocean with a δ15N value of around −1‰ (ref. 39), which is lower than that of mean global pycnocline nitrate (~6.2‰ (ref. 40)), thereby enabling the identification41, rate estimation20 and reconstruction38 of N2 fixation. The isotopic impact of denitrification depends on whether it occurs in the water column or in seafloor sediments19,42. Water-column denitrification occurs in the oxygen-deficient zones of the ocean and selectively removes 14N-nitrate, thereby increasing the δ15N of the ocean nitrate pool relative to the δ15N of newly fixed N39,40. By contrast, sedimentary denitrification consumes most of the nitrate diffusing into sediment pore waters, minimizing the escape of 15N-enriched residual nitrate into the overlying water column43. As a result, sedimentary denitrification lowers the N:P ratio of ocean waters with little isotopic effect44.

The CB-δ15N of symbiont-bearing corals is sensitive to the δ15N of the fixed N supplied to the typically nutrient-poor, oligotrophic reef environments in which they live45,46. In areas with high N2-fixation rates, such as the western tropical North Atlantic, newly fixed N lowers the δ15N of the fixed N pool41,42, which can be tracked via CB-δ15N (Fig. 1a)47. CB-δ15N is largely protected from post-depositional alteration, making it a reliable tool for assessing processes in the marine N cycle48, even in fossil Palaeozoic corals49,50.

The CB-δ15N values of our Caribbean records range from 1.87‰ to 6.68‰ (Extended Data Fig. 2). The lowest average values obtained are from Cuba (2.85 ± 0.41‰), and the highest average values are found in Costa Rica (5.09 ± 0.55‰). The average CB-δ15N from Cuba, Martinique (3.01 ± 0.25‰), Belize (3.33 ± 0.36‰) and Mexico (3.67 ± 0.23‰) agree with the measured mean Caribbean nitrate δ15N (3.28 ± 0.76‰). The average CB-δ15N values at Tobago (4.60 ± 0.31‰) and Costa Rica are above the mean Caribbean nitrate δ15N. The relatively high CB-δ15N at Tobago is consistent with its southeastward location, in the path of the Brazil Current that carries higher δ15N nitrate and organic N northwestwards from the equatorial Atlantic into the Caribbean (Fig. 1a), with much of the tropical North Atlantic N2 fixation occurring downstream of this20. The high CB-δ15N from the Costa Rican margin is probably due to local coastal processes, including possible anthropogenic influences (Supplementary Discussion). The CB-δ15N time series from Belize, Cuba, Martinique and Mexico may also be influenced by local processes during some periods, but they show remarkable similarities in their multidecadal variability when normalized (Extended Data Fig. 3). We analysed the underlying shared natural variability of those four records by normalizing and smoothing them with a Gaussian filter (Methods). The normalized records are then combined into a master chronology to evaluate the variability of N2 fixation in the Caribbean region from 1900 to 2021 (Fig. 2).

Fig. 2: CB-δ15N master record and comparison to the AMO.
figure 2

a, The normalized master CB-δ15N record (z scores) with the propagated error (±1s.d., shaded region) is based on four coral cores (Belize, Cuba, Martinique and Mexico from 1900 to 2021) from the wider Caribbean region and is closely related to Caribbean nitrate δ15N. Higher normalized CB-δ15N values correspond to lower rates of N2 fixation, whereas lower normalized CB-δ15N values are indicative of higher N2 fixation rates. b, Compared with a is the AMO index derived from Trenberth and Shea52 (black line) and the locally estimated scatterplot smoothing (LOESS) of depth-integrated P* values of the equatorial North Atlantic (−10 to 30° N, 20 to 60° W) from the GLODAPv2.2022 dataset for the period of 1981–2020 (orange line). The mean LOESS-smoothed P* values show congruent variability with the AMO. Positive AMO phases (red bars) represent a warmer North Atlantic compared with the South Atlantic, whereas negative AMO phases (blue bars) represent a colder North Atlantic compared with the South Atlantic. The positive AMO states correspond to a northward displacement of the ITCZ; negative AMO states indicate a southward displacement of the ITCZ.

Full size image

Multidecadal modulation of N2 fixation

The normalized master CB-δ15N record shows multidecadal variability, with dominant 16-, 32- and 64-year cycles that are characteristic of the Atlantic Multidecadal Oscillation (AMO) (Fig. 2 and Extended Data Fig. 4). The AMO index represents multidecadal sea surface temperature (SST) variability in the North Atlantic51,52, whereas the Atlantic Meridional Mode (AMM) index traces the higher-frequency mode of SST variability53,54. Atlantic SST variability is linked to changes in atmospheric and oceanic circulation, the position of the ITCZ and the corresponding location and strength of surface winds over the Atlantic55. Multidecadal variability has been reconstructed with SST proxies in coral cores56,57, speleothems58 and sediment cores59 throughout the Caribbean, demonstrating a connection between basin-scale variability and local climate conditions, but its potential influence on N2 fixation has not been investigated.

Anomalous cold North Atlantic and warm South Atlantic SSTs, which correspond, respectively, to a negative AMO/AMM index and a strengthened Hadley cell in the Southern Hemisphere, result in a southward displacement of the ITCZ55,60,61. Under these conditions, trade winds (easterlies) are maximal and enhance equatorial upwelling, the strength of the South Equatorial Current and the Caribbean Current (Extended Data Fig. 5)17,18,24. By contrast, positive AMO/AMM phases are characterized by a northward displacement of the ITCZ62, weaker easterlies, reduced equatorial upwelling and an enhanced North Equatorial Counter Current (NECC)63.

Upwelled waters in the equatorial North Atlantic tend to be depleted in nitrate compared with phosphate due to their origins in the Southern Ocean and Indo-Pacific19, and thus contain higher excess P than Caribbean water masses19,33 (Fig. 1 and Extended Data Fig. 6). Measurements of excess P over the past decades are spatially scarce and temporally discontinuous. Nevertheless, a compilation of the available data from multiple oceanographic cruises shows that the concentration of excess P in the equatorial North Atlantic is closely coupled with multidecadal variability, with negative AMO states corresponding to enhanced supply of excess P (Fig. 2b).

Variations in our master CB-δ15N record show asignificant positive correlation with the AMO index for the period between 1900 and 1972 (adjusted r2 = 0.33, P < 0.01) and a higher correlation from 1972 to 2021 (adjusted r2 = 0.60, P < 0.01), when the AMO and AMM are in phase (Fig. 2 and Extended Data Fig. 7), consistent with an anthropogenically forced AMO-like signal in the modern era64. Negative CB-δ15N anomalies, which are indicative of enhanced N2 fixation, align with negative AMO phases when a southward displacement of the ITCZ leads to enhanced equatorial upwelling. Strengthened easterlies and the Caribbean Current, in turn, would allow upwelled waters to penetrate into the Caribbean Basin more effectively. The correlation between our CB-δ15N and the AMO indicates that the supply of excess P has controlled N2 fixation in the Caribbean over the past 120 years.

A remarkable aspect of our findings is their consistency with previous findings regarding Caribbean N2-fixation changes on the vastly different timescales of the Earth’s orbital cycles24. As with AMO, precession drives cycles in equatorial Atlantic upwelling65, and the phases of more vigorous upwelling are associated with maxima in N2 fixation24. From the perspective of the ~22,000-year precession cycle, Caribbean N2 fixation is currently near maximal rates, with AMO further modulating N2 fixation.

Linking excess phosphorus supply, N2 fixation and Sargassum blooms

The biomass of floating Sargassum has been continuously mapped and quantified since 2000 (Fig. 3a,b)66. Changes in macro- and micro-nutrient availability have been suggested as driving recent increases in Sargassum biomass1,9,10. However, there is disagreement as to the sources and mechanisms behind the post-2011 GASB. Proposals have included nutrient inputs from atmospheric deposition, rivers and oceanic changes1,9.

Fig. 3: Seasonal CB-δ15N values from Martinique compared with Sargassum biomass and environmental parameters.
figure 3

a, CB-δ15N values (versus air) of the coral core collected in Martinique from 2000 to 2021 compared with the biomass of Sargassum spp (plotted in a logarithmic scale). b, Inverted AMM index based on Chiang and Vimont53 for the 2000–2024 period compared with the biomass of Sargassum spp (plotted in a logarithmic scale). More negative AMM states correspond to a southward displacement of the ITCZ, stronger easterlies, enhanced equatorial upwelling and a weakened NECC, which result in greater supply of excess P to the Caribbean. c, Area-weighted Atlantic (10 to 20° N, 20 to 60° W) total dust deposition and total black carbon (BC) deposition calculated using the ECHAM5/MESSy Atmospheric Chemistry (EMAC) model. d, Monthly discharge for the Amazon River measured at the Óbidos gauge station (1.92° S, 55.67° W) and for the Orinoco River measured at the Ciudad Bolivar gauge station (8.15° N, 63.54° W) obtained from SO-HYBAM material-transport datasets. In all panels, the vertical grey bars mark the Sargassum spp. peak seasons (April to September) since 2011.

Full size image

Our high-resolution CB-δ15N record of N2 fixation from Martinique shows a positive correlation with the reconstructed log-transformed Sargassum biomass from 2011 to 2021 (adjusted r2 = 0.51, P < 0.01). In addition, the first two extreme Sargassum blooms in 2015 and 2018 coincide with the highest rates of N2 fixation of the past 120 years (Figs. 2 and 3a). The interannual coupling between N2 fixation and Sargassum blooms appears to be modulated by the AMM (Fig. 3a,b). The AMM is correlated with changes in log-transformed Sargassum biomass (adjusted r2 = 0.30, P < 0.01) and N2 fixation (adjusted r2 = 0.32, P < 0.01) since the first appearance of the GASB in 2011. The negative phase of the AMM is associated with negative SST anomalies near 10° N, corresponding to a southward displacement of the ITCZ that leads to unusually strong trade winds near 5° N53,67. The AMM typically reaches a minimum in the boreal spring when the GASB starts to bloom. During negative AMM states, the seasonal development of the NECC along 5–10° N in the second half of the year connects Sargassum to the shallower mixed layer around the Guinea Dome during late autumn54,67,68, potentially also explaining the emergence of long Sargassum bands during the second half of the year (Extended Data Fig. 8).

The evidence that an increased supply of excess P (negative AMM) both enhances N2 fixation and encourages Sargassum blooms indicates a mechanism linking the two. The average N retained in Sargassum biomass (0.91 µmol N m−2) is much smaller than the surface particulate N pool suspended in the mixed layer (~30 mmol N m−2), such that the supply of recycled N to the surface mixed layer is far in excess of that required for Sargassum growth. This implies that Sargassum is not a major competitor for recycled N in surface waters. Accordingly, the increased supply of recycled N that accompanies periods of rapid N2 fixation cannot explain Sargassum blooms. Rather, the explosive growth of Sargassum is probably driven by the N2 fixation of its epiphytes, which can channel their newly fixed N directly to their host. That is, Sargassum appears to occupy the same niche as other diazotrophs in the tropical North Atlantic12,29,30, acting de facto as an N2 fixer itself. This perspective also explains why Sargassum accumulations are much stronger in the equatorial North Atlantic compared with previous blooms in the Sargasso Sea. N2 fixation in the modern Sargasso Sea is currently occurring at a very limited rate20,41, potentially due to a lower excess P supply19. Thus, Sargassum, with its limited capacity to compete with phytoplankton for recycled N, was unable to reach biomass accumulations in the Sargasso Sea that are comparable to the GASB.

High rates of N2 fixation in the Caribbean region are also evident during negative phases of the AMO before 2011 (Figs. 2 and 3b), which indicates that Sargassum was only governed by these dynamics once it reached the equatorial North Atlantic. Before this import, there is no evidence of comparable Sargassum blooms in the area. Thus, we suggest that the arrival of Sargassum in the equatorial North Atlantic triggered the GASB, by making it more proximal to equatorial upwelling, giving Sargassum greater access to the excess P supply.

Other environmental parameters and Sargassum

Other proposed drivers of Sargassum since 2011 include Amazon–Orinoco discharge, black carbon deposition, Saharan dust, atmospheric N and warmer SSTs1,9,11. However, their variability is largely seasonal, with negligible interannual fluctuations, and thus cannot explain the observed interannual biomass trends and amplitudes.

There is no interannual relationship between Sargassum biomass and Amazon (adjusted r2 = 0.06, P > 0.05) or Orinoco (adjusted r2 = 0.01, P > 0.10) water/nutrient discharge9 (Fig. 3d), perhaps because these discharges are restricted to coastal regions whereas most of the Sargassum biomass is found in open waters. During the Sargassum blooms in 2011 and 2012, satellite observations indicate that Sargassum accumulated near the river outflows and was then exported to the equatorial region and the Caribbean7. Thus, excess P from the Amazonas may have stimulated N2 fixation and Sargassum in 2011 and 201269, analogous to the blooms of the diatom Richelia intracellularis—with N2-fixing endosymbionts—found in the North Brazil margin70. Atmospheric N deposition is highest near continents71, but it is four orders of magnitude lower than naturally occurring open-ocean N2 fixation20,72 and appears not to be a viable driver of Sargassum blooms (Extended Data Fig. 9). Previous studies have attributed the variability in N:P ratios in Sargassum biomass over the past decades to anthropogenic N or P inputs9,10,73. However, Sargassum N:P ratios align closely with the N:P ratio of upwelled water, with a positive offset (Extended Data Fig. 10). Furthermore, the higher N:P ratios in Sargassum9,10 suggest that its relationship with N2-fixing symbionts is central to fulfilling its N requirements, especially in environments of high excess P.

We find only a weak correlation of Sargassum biomass with the modelled area-weighted deposition of wet dust (adjusted r2 = 0.04, P > 0.05), dry dust (adjusted r2 = 0.04, P < 0.01), wet black carbon (adjusted r2 = 0.08, P > 0.05) or dry black carbon (adjusted r2 = 0.14, P < 0.01). Although dry dust and dry black carbon deposition are correlated to Sargassum biomass, the relationship is negative, inconsistent with an important role for nutrients from these sources74. Regarding the potential role of SST and sea surface salinity (SSS)11, in situ experiments show optimal Sargassum growth at high temperatures (from ~23 to 28 °C) and salinities above 34 PSU (practical salinity units)1,75. In the equatorial North Atlantic, SST and SSS are in this range and show no correlation with Sargassum biomass (SST: adjusted r2 = 0.06, P < 0.01; SSS: adjusted r2 = 0.07, P < 0.05) (Supplementary Fig. 1).

Using an Akaike information criterion (AIC), we find that CB-δ15N and AMM can explain 56% of Sargassum biomass from 2011 to 2021 (Supplementary Table 1). Adding any alternative nutrient sources did not improve the AIC, underscoring excess-P-driven N2 fixation as the dominant driver of Sargassum blooms.

Implications for future Sargassum blooms

In the tropical Atlantic, wind-driven equatorial upwelling and northward transport of excess P, in the context of a high aeolian iron supply, enhances N2 fixation. The resulting increase in the supply of both P and N has allowed Sargassum to expand since 2011, when the macroalgae were imported from the Sargasso Sea. Since then, negative AMM states have aligned with periods of high Sargassum biomass. Thus, the AMM can be used to better predict the annual extent of Sargassum blooms in the future, supporting efforts to mitigate the impacts of Sargassum blooms on Caribbean reef ecosystems and coastal communities.

Methods

Coral samples

Coral drill cores were collected between 1997 and 2021 from living colonies of massive corals at multiple locations across the Caribbean by different research groups. An overview of the metadata is given in Supplementary Table 2, and includes coral species, location, depth at which the coral core was sampled, sampling resolution, age model method and the length of the record. Cores were sliced into longitudinal slabs and rinsed with deionized water and sent to the Martínez-García Laboratory at the Max Planck Institute for Chemistry (MPIC) in Mainz for analysis.

Analysis of coral-bound nitrogen isotopes

The CB-δ15N measurements were performed in the Martínez-García Laboratory (MPIC). We used the persulfate oxidation–denitrifier method41,77, applied to corals by Wang et al.46,78, with modifications described by Moretti et al.79.

A drilling path was drawn based on ultraviolet scans and subsequent pairs of high- and low-density bands on previously bleached and rinsed coral slabs. Sample material was then carefully extracted perpendicular to the main growth axis using a millimetre drill bit attached to a Dremel hand tool. The milled material was vacuumed and the powder was split into fine (5–63 µm) and coarse (>63 µm) size fractions. Aliquots of fine powder and coarse powder were further used for the analysis of oxygen isotope (δ18O) and δ15N analyses, respectively.

Coarse powder (6 ± 1 mg) was weighed into a 4 ml VWR borosilicate glass vial, which was filled with sodium hypochlorite (4.25 ml) and left on a shaker at 120 revolutions per min for at least 24 h. The sodium hypochlorite was removed the next day with a pre-combusted glass pipette attached to a vacuum line set at 500 mbar. Samples were then rinsed three times with Milli-Q water (4 ml; 18.2 MΩ cm−1 at 25 °C) and left to dry at 60 °C overnight.

Once fully dried, the coarse powder (6 ± 0.2 mg) was weighed inside an in-house clean room to minimize contamination. Thereafter, skeletal-bound organic matter was released by dissolving the final amount of material with 4 N hydrochloric acid (45 µl). Concurrently, a persulfate oxidative reactant solution was prepared inside the clean room using 6.25 N sodium hydroxide (a 4 ml spike) to reach high pH. Persulfate oxidative reactant solution (1 ml) was added to each dissolved sample and at least ten empty cleaned vials (blanks), and the batch of vials was placed in a custom-built sample rack that was tightly sealed with a polytetrafluoroethylene sheet before being autoclaved at 121 °C for 65 min.

A 1 ml volume of concentrated denitrifying bacteria (Pseudomonas chlororaphis) was injected into growth medium (800 ml) and left for 4–6 d to grow in the dark at room temperature on a shaking rack. Once the bacteria had grown sufficiently, the medium was transferred to autoclaved polyethylene bottles and centrifuged at ~8,800 × g for 10 min. The supernatant was then discarded and the remaining bacteria pellet was resuspended in a buffered (pH 6.3) resuspension medium. From this, 3 ml aliquots were pipetted into separate muffled glass vials (20 ml), each of which was capped with a septum and tightly sealed before being placed upside-down on a needle rack with a small additional needle for venting. The needle rack supplies a continuous flow of N2 for at least 3 h to replace the internal atmosphere and dissolved gases with pure N2. The bacteria vials were removed from the rack, and the oxidized sample (~0.5 ml) was injected into each bacteria vial. Once injected, the bacteria vials were kept in the dark for 2–3 h to ensure the quantitative transformation of nitrate (NO3−) to nitrous oxide (N2O) before being frozen at −21 °C.

On the day of analysis, the bacteria were thawed, lysed with several drops of 10 N sodium hydroxide and placed on a mass spectrometer for isotope analysis. The δ15N value of the N2O was determined using a custom-built inlet system automated for extraction and purification coupled to a Thermo MAT253 Plus stable isotope ratio mass spectrometer80,81. Long-term precision was determined by analysing internal coral standards with each sample batch, which yielded an average carbonate standard reproducibility of ±0.2‰.

Coral oxygen isotopes

Oxygen isotopes were measured on coral cores from Martinique, Belize and Costa Rica (Supplementary Fig. 2). For each run, 55 coral carbonate samples of 100–200 µg were analysed for δ18O in the inorganic stable isotope laboratory at the MPIC in Mainz. One International Atomic Energy Agency carbonate standard (IAEA-603) (n = 10) and one Virje University Internal Carbonate Standard (VICS) (n = 11) were used to calibrate the analyses to the Vienna PeeDee Belemnite (VPDB) scale. Samples were measured using an isotope ratio mass spectrometer (Delta V Advantage, Thermo Scientific) which is connected to a GasBench II unit (Thermo Scientific). Each sample was placed in a 12 ml Exetainer vial (part no. 9RK8W; Labco). Samples and standards were then put into a hot block heated to 70 °C. First, the vials are flushed with helium to remove atmospheric CO2. Then, >99% H3PO4 (5–10 drops) was added and the sample was left to dissolve for 1.5 h. Finally, the sample was transferred in helium carrier gas to the GasBench II unit, where water and contaminant gases were removed before subsequent isotope analysis using the isotope ratio mass spectrometer. The average analytical precision, based on the reproducibility of IAEA-603, was 0.11‰ (1s.d., n = 42) for oxygen isotopes and 0.09‰ (1s.d., n = 42) for carbon isotopes.

Age model

The δ18O data for samples from Martinique, Belize and Costa Rica were calibrated against their respective OIv2SST dataset (taken from https://climexp.knmi.nl/start.cgi). The highest δ18O values were anchored to the lowest SST, which translates to February at Caye d’Olbian, Martinique and at Turneffe Atoll, Belize, and January at Cahuita, Costa Rica, whereas the lowest δ18O values were anchored to the highest SST, which served as the basis of our age model. The δ18O and SST yield a negative correlation with SST of −0.15‰ per °C (r2 = 0.65) for Martinique, −0.18‰ per °C (r2 = 0.56) for Belize and −0.28‰ per °C (r2 = 0.46) for Costa Rica (Supplementary Fig. 3). Age models for annually resolved records had already been established for samples from Bermuda47, Cuba82 and Tobago83, whereas X-ray density bands were used for the coral core from Mexico.

Analysis of seawater nutrient concentrations

Water samples were collected between December 2022 and March 2023 using a rosette water sampler equipped with five five-litre bottles according to the protocol detailed in Schiebel et al.84. The sampling was conducted along an east–west transect across the Atlantic at 13° N and across the Caribbean Sea at 11° N during a cruise aboard the research sailing yacht Eugen Seibold (https://www.mpic.de/4224334/sy-eugen-seibold). All water samples were frozen on collection and kept frozen at −21 °C until analysis.

Analyses for NO3−, nitrite (NO2−) and phosphate (PO43−) were conducted at the MPIC. Concentrations of NO3− and NO2− were first determined according to Braman and Hendrix85 using a nitrogen oxides analyser (T200, Teledyne API) with a detection limit of 0.01 µM and a precision (±1s.d., n = 55) of 0.55%. The PO43− concentrations were determined using a continuous flow autoanalyser (QuAAtro, Seal Analytical) with a detection limit of 0.01 µM and a precision (±1s.d., n = 55) of 0.5%. For the GO-SHIP A20 (EXPOCODE: 325020210316) and A22 (EXPOCODE: 325020210420) samples, the concentration data were generated as part of the GO-SHIP programme and accessed via the CCHDO Hydrographic Data Office (US San Diego Library Digital Collections; CCHDO Hydrographic Data Archive, 2023, https://doi.org/10.6075/J0CCHAM8).

Analysis of seawater nitrogen isotopes

The δ15N values of NO3− + NO2− and NO3−-only were measured using the denitrifier method77,80 in the Martínez-García Laboratory (MPIC; for the Seibold samples) and at Princeton University (for the GO-SHIP A20 and A22 samples), following the protocols of ref. 81. NO3− + NO2− or NO3− (2–20 nmol N depending on concentration) was quantitatively converted to N2O gas by a strain of denitrifying bacteria (Pseudomonas aureofaciens) that lacks active N2O reductase enzymes. The δ15N of N2O was determined using the previously described purpose-built inlet system coupled to the Thermo MAT253 Plus stable isotope ratio mass spectrometer77,80,81.

Excess phosphorus data

Excess P, expressed as P* = PO43− − NO3−/16 (ref. 19), was calculated from nutrient measurements compiled in the GLODAPv2.2022 (Global Ocean Data Analysis Project version 2.2022) dataset between 1983 and 2020. These data were based on cruises that have reliably measured NO3− and PO43− in the North Atlantic (−10 to 30° N, 20 to 60° W). To avoid seasonal biases, only years with sufficient data throughout the year and covering the whole latitudinal range were included (Supplementary Fig. 4). Values were depth-integrated until reaching the pycnocline. The pycnocline depth was calculated based on the POAMA/PEODAS analysis z20 dataset taken from the Climate Explorer website (https://climexp.knmi.nl).

Sargassum biomass estimation

The methods for estimating Sargassum biomass from satellite observations since 2000 have been detailed in Wang et al.86 and Hu and et al.87. Briefly, satellite images were analysed to examine the image features (that is, spatial anomalies), and these image features were delineated using a computer deep-learning model (Hu et al.87). The spectral shapes of these delineated image features, relative to the surrounding seawater, were examined to determine the presence of Sargassum. The amount of Sargassum within each image pixel was first estimated as a percentage cover, and then converted to wet biomass using a field-determined calibration constant86. Finally, many satellite images within a month were used to remove data gaps (due to clouds and other factors) and to calculate the average biomass at a given location.

Dust and black carbon model

The EMAC (ECHAM5/MESSy2 Atmospheric Chemistry)88 was used to calculate wet and dry depositions of mineral dust and black carbon over the equatorial North Atlantic (0 to 20° N, 20 to 60° W) for the period 2003–2019. The EMAC model describes tropospheric and middle atmosphere processes, and their interactions with the land and oceans. For this work, we used the DDEP (Dry DEPosition) submodel89 to estimate dry deposition, whereas the SCAV (SCAVenging) submodel was used to simulate wet deposition90. A detailed description and evaluation regarding the EMAC configuration and the submodels used in this study can be found in Holanda et al.91, in which the black carbon calculations are also evaluated against observations. A detailed evaluation of the model performances in reproducing dust transport can be found in Abdelkader et al.92.

Caribbean CB-δ15N stack

A master CB-δ15N record was constructed using the records within the mean Caribbean nitrate δ15N (Extended Data Fig. 3), which includes Cuba, Martinique, Belize and Mexico. This ensured that the variability within each record was probably driven by natural processes and reduced the inclusion of local anthropogenic effects, for example, terrestrial runoff. Nevertheless, it is worth indicating that the record from Mexico showed a disagreement with the other selected records around the 1980s; this coincides with a period of very sudden and rapid development in the area that may have temporarily affected the record. The master record was constructed from normalized CB-δ15N time series (normalized CB-δ15N = [(CB-δ15N − average CB-δ15N)/(standard deviation of CB-δ15N)]) and then smoothed with a Gaussian filter using the SciPy package (v.1.11.2).

Cuba—the dataset with the longest continuous record (1900–2015)—was selected as the base time series, to which the other three records were sequentially integrated. Where overlaps occurred, data points were averaged to mitigate any abrupt transitions. This averaging was weighted by the confidence levels of the original data that were provided by each source. For example, data points with a confidence level of 95% will contribute more to the average than those with a confidence level of 75%. When calculating the overall confidence level, error propagation was applied by considering the individual CB-δ15N record confidence levels and then taking the square root of the sum of the squared individual errors. The final composite (master CB-δ15N) time series was analysed to identify significant temporal trends and anomalies.

Statistical analysis

CB-δ15N values were imported to a Python3 Jupyter Notebook (v.5.7.4) using the Pandas software library. Data were plotted with Seaborn/Matplotlib and reprocessed for wavelet analysis according to the waipy script (https://github.com/mabelcalim/waipy). The continuous wavelet transform significance test was based on Torrence and Compo93, and cross wavelet analysis was based on Maraun and Kurths94. LOESS of isopycnic P* values was conducted in the Python3 Jupyter Notebook (v.5.7.4) with >1,000 bootstraps to provide a better characterization of the uncertainty in the estimates. Linear and multiple regressions and AIC analyses to understand the relationship between abiotic environmental conditions and Sargassum blooms were all conducted using RStudio (v.4.3.0). Correlations are expressed as adjusted r-squares. Unlike r-squared, adjusted r-squared increases only if the new predictor enhances the model more than would be expected by chance. It can also decrease if a predictor improves the model by less than expected by chance. Standard deviations are given as ±1s.d.

Ethics and inclusion statement

This study was conducted ensuring fairness, respect, care and honesty throughout the research process. Local collaborators were engaged as equal partners in the design, execution and interpretation of the study, with co-authorship offered in recognition of substantive contributions. Fieldwork was carried out with the appropriate research permits and in close collaboration with local institutions and stakeholders. Training and mentorship opportunities were provided to early-career scientists and students, with particular emphasis on capacity building in the regions where data were collected. All data and findings will be shared transparently with local partners and relevant authorities to support both scientific knowledge and local conservation efforts. We are committed to equitable knowledge exchange, avoiding exploitation of local resources or communities and ensuring that the benefits of this research extend to the regions in which it was conducted.