Coupled increases in pelagic fish productivity and export productivity during the Paleocene-Eocene Thermal Maximum

Abstract

Ocean productivity is sensitive to climate change, and its decline in recent decades has profoundly impacted the biosphere. Yet its long-term trajectory under global warming remains a topic lacking consensus. To address this question, we use paired ichthyolith and marine barite accumulation rates to investigate marine productivity and associated biological responses during the Paleocene-Eocene Thermal Maximum. We find a widespread increase in pelagic fish productivity, particularly in the Pacific and North Atlantic, that closely tracks export productivity in trend and spatial pattern, suggesting this increase was fueled by enhanced carbon export from the surface ocean. Model simulations support the observed trend, demonstrating that under elevated pCO₂ conditions, the effects of enhanced nutrient supply can outweigh the suppressive effects of ocean stratification on export productivity. Our findings highlight that the long-term response of marine productivity to sustained greenhouse warming may differ fundamentally from the short-term declines observed in the modern ocean.

Increased productivity and fish abundance in the open ocean during extreme warming 56 million years ago contrasted with the short-term declines observed today, highlighting that marine ecosystems may respond differently to prolonged global warming.

Introduction

Marine productivity is essential for regulating climate change and sustaining the ocean ecosystem. Contemporary remote sensing data indicate that productivity has been declining in the open ocean over the past three decades, with unevenly distributed impacts on marine fisheries^1,2. This trend is particularly pronounced in low-latitude regions, primarily attributed to weakened upwelling and enhanced ocean stratification under global warming³. Yet, when looking forward, future projections of marine productivity remain highly divergent, with climate models yielding both increases and decreases in global trends by the end of this century^1,4,5. The long-term response of marine productivity to sustained warming remains poorly constrained.

The Paleocene-Eocene Thermal Maximum (PETM; ~55.9 Ma) serves as an important analog for understanding the long-term impacts of anthropogenic climate change⁶. This event was triggered by massive and rapid injection of isotopically light carbon (i.e., 3000–10,000 Gt C within 10 kyr) into the ocean-atmosphere system, resulting in a negative carbon isotope excursion (CIE) of 3–5‰^7,8. During the PETM, the ocean experienced profound perturbations, including an increase in seawater temperature by 5 °C–8 °C, acidification, deoxygenation, and turnovers in benthic foraminifera and calcareous nannofossils^9–13. Although the estimated emission rate is about an order of magnitude lower than that of current fossil fuel burning, the cumulative carbon release is comparable in amount to the projected high-emission scenario by 2100⁶. Therefore, investigating productivity changes during the PETM offers critical insights into its long-term dynamics under future climate change.

Productivity reconstruction for the PETM remains challenging due to inconsistent results from proxy records and model simulations. For example, marine barite accumulation rate (BAR) is considered a reliable tracer of export productivity¹⁴. Since marine barite formation requires organic matter decomposition to release excess barium to create supersaturated microenvironments for barium sulfate, BAR is positively correlated with the integrated carbon export out of the euphotic zone¹⁵. The BAR records from the global open ocean show an average threefold increase in export productivity during the PETM, likely driven by enhanced terrestrial nutrient input¹⁶. This increase is supported by foraminiferal Sr/Ca and organic phosphorus records from the South Atlantic and Southern Ocean^17,18. However, evidence from benthic foraminifera and nannofossil assemblages suggests a decline of nutrient supply in the South Atlantic, Southern Ocean, and North Pacific after the onset of the PETM^19–21. In addition, Earth system models indicate that a warmer and more stratified ocean could have suppressed productivity in tropical regions²². These discrepancies reveal the intricacy of productivity changes in the PETM ocean, underscoring the need for further research.

In this study, we use ichthyolith accumulation rate (IAR) as a proxy of pelagic fish productivity for the PETM. As fossilized remains of fish teeth and dermal denticles, ichthyoliths are deposited on the seafloor through natural shedding or excretion of indigestible parts by predators²³. The majority of ichthyoliths in marine sediments originate from small pelagic species (e.g., juvenile fishes, shark taxa) - mid-trophic-level consumers - whose abundance is sensitive to marine productivity and trophic transfer efficiency^24,25. Consequently, IAR has been used for reconstructing fish productivity, and potentially marine productivity, over geological time^24,26,27.

We present IAR data from five sediment cores located in the North Pacific (ODP Site 1209), Eastern Equatorial Pacific (ODP Site 1221), North Atlantic (IODP Site U1403), South Atlantic (IODP Site U1557), and Southeast Indian Ocean (IODP Site U1514) for the PETM (Fig. 1 and Supplementary Fig. 1). These data are combined with previously published IAR data from the Eastern Equatorial Pacific (ODP Site 1220)²⁸ to provide a comprehensive global ocean dataset (Supplementary Data 1). Moreover, we generate a compilation of BAR-derived carbon export (i.e., new production P_new defined by Dymond et al.²⁹), which includes new measurements from Site U1557 as well as previously published data from refs. ^16,30,31. This compilation covers the same sites or alternative sites in proximity to the IAR sites (Fig. 1 and Methods), allowing us to explore the relationship between fish productivity and export productivity in different oceanographic settings. Finally, these proxy results are further compared with equilibrated simulations using the Earth system model cGENIE to investigate potential mechanisms driving changes in productivity in a warming ocean.

Fig. 1. Locations of ichthyolith accumulation rate (IAR) and barite accumulation rate (BAR)-derived carbon export (P_new) sites plotted against the modern chlorophyll concentration at 0.5 m water depth.

The chlorophyll concentration data are from the E.U. Copernicus Marine Service Information. Dark-blue squares and light-blue squares denote the sites of IAR from this study and previous studies, respectively. Brown circles and yellow circles denote the sites of BAR-derived P_new from this study and previous studies, respectively. Due to a lack of barite records from Sites U1403 and U1514, the nearest Sites 549 and 738 are selected for data comparison. Small symbols indicate paleo-locations of the Paleocene-Eocene Thermal Maximum (PETM) sites at 56 Ma, with their corresponding modern locations shown as large symbols.

Results and discussions

Increased IAR during the PETM

Our data show pronounced increases in IAR at the Pacific and North Atlantic sites (i.e., Sites 1221, 1209, 1220 and U1403) during the PETM, with one or two distinct peaks within the main CIE interval, whereas changes at the South Atlantic Site U1557 and Indian Ocean Site U1514 are comparatively modest (Fig. 2). The highest IAR values are found in the Eastern Equatorial Pacific, ranging from 54 ichthyoliths cm⁻² kyr⁻¹ at Site 1220 to 121 ichthyoliths cm⁻² kyr⁻¹ at Site 1221. At North Pacific Site 1209 and North Atlantic Site U1403, IAR reaches 51 ichthyoliths cm⁻² kyr⁻¹and 64 ichthyoliths cm⁻² kyr⁻¹, respectively, with peak values five to eleven times higher than the background values. While the magnitude of the increase varies across locations, sites with sufficiently high sampling resolution and reliable age models show two clear IAR peaks: one at the onset of the CIE, and the other approximately 50 kyr after the onset of the CIE. By contrast, at South Atlantic Site U1557 and Southeast Indian Ocean Site U1514, IAR is much lower ( < 15 ichthyoliths cm⁻² kyr⁻¹), with only one peak observed, probably due to low sampling resolution. Among all the study sites, IAR shows a negligible correlation with mass accumulation rate (Supplementary Fig. 2), indicating that the rise in IAR during the PETM in multiple ocean basins primarily results from an increase in the ichthyolith flux to the seafloor, rather than changes in sedimentation rate.

Fig. 2. Ichthyolith accumulation rate (IAR) and δ¹³C changes with relative time from (a-c) the Pacific, (d, f) the Atlantic, and (e) the Indian Oceans across the Paleocene-Eocene Thermal Maximum (PETM).

Dark-blue squares denote IAR data from this study. Light-blue squares denote recalculated IAR using the ichthyolith data from ref. ²⁸ and the updated dry bulk density and age model from this study. Error bars represent ±1 standard error (SE). Green circles denote bulk carbonate δ¹³C data compiled from published records^{28,52,58,60,70}. In all panels, the blue shading indicates the main carbon isotope excursion (CIE) interval (0-50 kyr), and the lighter shading indicates the recovery interval (50-250 kyr).

The increase in the ichthyolith flux to marine sediments can be attributed to increased trophic transfer efficiency and/or elevated fish productivity. Typically, tooth size and type are species-specific traits, with larger species contributing larger but fewer teeth, which are useful for tracking community changes³². The PETM ichthyolith records support the stability in fish communities, as evidenced by three key observations. First, most of the ichthyoliths in our samples are smaller than 63 μm, indicating the continued dominance of juveniles or small-bodied taxa (Supplementary Fig. 3). Second, this observation aligns with previous studies that report no significant shift in tooth type diversity in the North Pacific, Equatorial Pacific, and Atlantic across the PETM, where assemblages from multiple sites were persistently dominated by two morphotype groups - conical and triangular teeth²⁸. Third, the even distribution of rare tooth types throughout the sequence further negates the presence of a transient excursion fauna associated with the PETM²⁸. Similar resilience in fish communities has been documented in ichthyolith records from other periods of abrupt climate change, such as the Cretaceous/Paleogene mass extinction and the Eocene-Oligocene transition, where turnover remained minimal and largely decoupled from climatic perturbations^24,33. While warmer waters tend to accelerate fish growth rates and shorten generation times, potentially increasing ichthyolith flux^26,34, we observe a nonlinear correlation between IAR and ocean temperature or latitude, suggesting that changes in trophic transfer efficiency alone cannot account for the IAR increase among sites (Supplementary Fig. 4). Taken together, these lines of evidence point to a widespread increase in fish productivity during the PETM, plausibly fueled by elevated marine productivity.

Relationship between IAR and P_new

The P_new records, derived from BAR at the same or nearby IAR sites, support the increase in export productivity during the PETM and further constrain its timing and magnitude (Fig. 3). In the Eastern Equatorial Pacific (Sites 1220 and 1221), P_new is relatively high, rising to 187 g C m⁻² yr⁻¹ within the main CIE interval. Notably, the two prominent peaks of P_new at the onset and ~45 kyr after the onset of the CIE, coincide with the two peaks of IAR, suggesting that elevated carbon export from the surface ocean to the deep ocean may enhance food availability for mid-trophic-level consumers like fish and in turn, contribute to higher ichthyolith fluxes to sediments. Although the P_new values at both sites are similar, IAR at Site 1221 is approximately twice as high as that at Site 1220, possibly reflecting greater inherent fish biomass at Site 1221. Nevertheless, temporal variations in P_new and IAR show a positive correlation (Supplementary Information), suggesting that the linkage between export productivity and fish productivity is better captured by their relative trends than absolute magnitudes.

Fig. 3. Comparison of ichthyolith accumulation rate (IAR) and barite accumulation rate (BAR)-derived carbon export (P_new) from (a-c) the Pacific, (d, f) the Atlantic, and (e) the Indian Oceans across the Paleocene-Eocene Thermal Maximum (PETM).

Dark-blue squares denote IAR data from this study. Light-blue squares denote recalculated IAR using the ichthyolith data from ref. ²⁸ and the updated dry bulk density and age model from this study. Error bars represent ±1 standard error (SE). Brown circles denote P_new data from this study, and yellow circles denote recalculated P_new using the barite content from previous studies^16,30,31 and the updated dry bulk density and age models from this study. In all panels, the blue shading indicates the main carbon isotope excursion (CIE) interval (0–50 kyr), and the lighter shading indicates the recovery interval (50–250 kyr).

A similar pattern emerges in the North Pacific (Site 1209), where the low P_new ( < 0.2 g C m⁻² yr⁻¹) values are consistent with the oligotrophic conditions expected for the center of a subtropical gyre. P_new exhibits a distinct peak at the onset of the CIE, followed by a gradual decline to the pre-excursion value over the subsequent 50 kyr. This rise in P_new is concurrent with the fivefold increase in IAR, highlighting the role of food supply in supporting fish productivity. Yet, the persistence of high IAR despite the low P_new suggests that additional factors, such as more efficient trophic transfer, may also contribute to fish productivity at this site, as documented in modern nutrient-depleted oceans^35,36.

In the North Atlantic, P_new at Site 549, located near Site U1403, is relatively high and increases to 166 g C m⁻² yr⁻¹, indicating a high export carbon flux in this ocean region. The peak of P_new coincides with the second peak of IAR at Site U1403 at 50 kyr after the onset of the CIE, but it does not capture the first IAR peak at the onset. This discrepancy may involve the difference in site locations and/or the relatively poor sampling resolution for P_new, particularly prior to the onset of the CIE. Regardless, the elevated values observed in both records during the PETM are higher than expected for the North Atlantic³⁷, implying a more productive environment compared to the present-day (Fig. 1).

In the South Atlantic (Site U1557) and Southern Ocean (Site 738, near Southeast Indian Ocean Site U1514), P_new remains below 7 g C m⁻² yr⁻¹ and 30 g C m⁻² yr⁻¹, respectively. The low P_new values are in line with the low IAR observed at these locations, indicating relatively low export productivity and fish productivity in the mid- to high-latitude Southern Hemisphere oceans. Despite the lack of a statistically significant site-specific correlation, both P_new and IAR exhibit an overall increase during the PETM, similar to the trend observed in other regions.

Globally, the distribution of P_new in the PETM ocean is consistent with modern water-column measurements of carbon export across ocean basins¹⁵. Compared to the other sites, the highest export productivity at the Eastern Equatorial Pacific and North Atlantic sites may be associated with their proximity to upwelling regions, where nutrient-rich deep waters are continuously supplied to the ocean surface, as well as with additional regional nutrient input (e.g., eolian dust and volcanic ash) from nearby sources. More importantly, pooled global P_new data from all sites reveal a significant positive correlation with IAR (Pearson r = 0.55, p < 0.001; Spearman ρ = 0.61, p < 0.001; Supplementary Fig. 6), suggesting that elevated export productivity was likely a key driver in sustaining the rise in fish productivity during the PETM, with responses ranging from modest in the South Atlantic and Southeast Indian Ocean to substantial in the Eastern Equatorial Pacific and North Atlantic.

Mechanisms for elevated export productivity

The mechanisms responsible for elevated marine export productivity during the PETM remain uncertain. In the context of global warming, potential mechanisms include enhanced nutrient input from continental weathering, redox-sensitive nutrient regeneration, and changes in ocean circulation and stratification. To evaluate the relative importance of these processes during the PETM, we perform a series of transient simulations using the Earth system model cGENIE (see Methods) and compare the simulated particulate organic carbon (POC) export flux at intermediate water depth (483 m) with our proxy-based productivity reconstructions.

During the PETM, the rise in atmospheric pCO₂ represents a major greenhouse perturbation. With a doubling of atmospheric pCO₂ from 1112 to 2224 ppm, the global mean POC export flux shows an initial transient decline followed by gradual recovery and stabilization slightly above the pre-PETM level (Supplementary Fig. 7). At quasi-equilibrium, the POC export flux increases by only ~0.01 mol m⁻² yr⁻¹ (Fig. 4d), with little sensitivity to the rate of atmospheric pCO₂ increase. This response is substantially smaller than the increase in P_new inferred from marine barite records across all sites, indicating additional processes beyond CO₂ forcing alone exerted important controls on export productivity during the PETM.

Fig. 4. Simulated marine particulate organic carbon (POC) export flux at the intermediate water depth (483 m) for the Paleocene-Eocene Thermal Maximum (PETM) under transient sensitivity experiments and isolated contributions of individual processes.

a Transient simulation forced solely by a doubling of atmospheric pCO₂ (PETM_CP_NER). b Transient simulation forced by atmospheric pCO₂ doubling with temperature-dependent phosphorus weathering enabled (PETM_1×_NER). c Transient simulation forced by atmospheric pCO₂ doubling with temperature-dependent phosphorus weathering and redox-dependent phosphorus recycling enabled (PETM_1×_ER). d Isolated contribution of atmospheric pCO₂ forcing, calculated as PETM_CP_NER – CP_NER_s. e Isolated contribution of weathering-derived phosphorus input, calculated as PETM_1×_NER – PETM_CP_NER. f Isolated contribution of redox-dependent phosphorus recycling, calculated as PETM_1×_ER – PETM_1×_NER. Note that color scales vary among panels. Model configuration and experiment design are summarized in Supplementary Tables 1 and 2. Additional sensitivity experiments are presented in the Supplementary Information and Supplementary Fig. 8.

A plausible explanation for the observed productivity rise is enhanced external nutrient input. In the open ocean, nutrients (e.g., N, P, Fe) are a limiting factor for marine productivity. The greenhouse effect during the PETM likely intensified terrestrial weathering, delivering more nutrients to the ocean and fueling biological productivity^13,38. Lithium isotope evidence suggests that terrestrial weathering increased globally by 50–60% during the PETM³⁹. When we consider enhanced weathering-derived phosphorus supply without a change in internal phosphorus recycling (i.e., PETM_1×_NER; Fig. 4b), the ocean-wide average POC export flux increases from 0.7 to 1.2 mol m⁻² yr⁻¹ (Fig. 4e). Increasing weathering intensities can progressively enhance export productivity, with the 1.9-fold weathering enhancement increasing the POC flux to ~1.6 mol m⁻² yr⁻¹ (Supplementary Fig. 8), consistent with the inferred global rise (close to two- to three-fold) in export productivity relative to the pre-PETM¹⁶. Spatially, the most pronounced increases in carbon export occur in the Eastern Equatorial Pacific, Atlantic, and Tethys Ocean, which is also compatible with the spatial variability indicated by our P_new record.

In addition to terrestrial weathering, iron fertilization by eolian dust and volcanic ash could be a potential regional amplifier of productivity in nutrient-rich but iron-limited regions. Biogenic magnetite and geochemical records show that massive amounts of reactive volcanic substrates were generated by emplacement of the North Atlantic Igneous Province and circum-Caribbean volcanic eruptions and transported to ocean basins proximal to the location of volcanism (e.g., North Atlantic, Eastern Pacific) via trade winds, increasing iron availability in upwelling regions^8,40,41. Meanwhile, enhanced hydrological variability in North America during the PETM amplified regional weathering and eolian dust production, further elevating productivity in the nearby oceans^42,43. Since the model does not incorporate regional iron input, the simulated POC export flux is generally lower than the P_new estimates, particularly at Sites U1403, 1220, and 1221.

Internal phosphorus recycling may also have intensified as a result of ocean warming and deoxygenation during the PETM, potentially contributing to the rise in biological productivity and subsequently carbon export^38,44. However, our model shows that when enhanced weathering-derived phosphorus supply is combined with strengthened phosphorus recycling (i.e., PETM_1×_ER; Fig. 4c), the additional increase in the POC export flux is negligible compared to the PETM_1×_NER simulation ( ~ 0.01 mol m⁻² yr⁻¹; Fig. 4f). This limited response is primarily associated with reduced upward water transport in a warmer, more stratified ocean²². A deeper thermocline would diminish the entrainment of nutrients from deeper waters into the euphotic zone, causing broad declines in productivity. Ultimately, this nutrient-transport limitation partially offsets the productivity increase expected from enhanced nutrient weathering and recycling.

The effect of enhanced nutrient supply is projected to increase global ocean productivity during the PETM. The rise in export productivity could, in turn, increase organic carbon burial and atmospheric CO₂ sequestration, acting as negative feedback to help the Earth recover from hyperthermal conditions⁴⁵. Indeed, higher organic carbon contents have been observed at sites in the Eastern Equatorial Pacific and North Atlantic during the PETM^40,41. However, the rising temperature and ocean acidification would also, in the meantime, enhance organic matter degradation through accelerating microbial metabolic rates, leading to a low organic carbon flux to sediments, decoupled from the high POC export flux towards the mesopelagic and bathypelagic zones^21,46,47. Therefore, we emphasize that an increase in export productivity in the water column does not necessarily equate to an increase in organic matter reaching the seafloor.

Implications

The coupled increases in IAR and BAR during the PETM suggest that elevated export productivity enhanced food availability for pelagic fish communities, thereby sustaining the widespread rise in fish productivity, especially in the Equatorial Pacific and North Atlantic. Despite spatial heterogeneity in ichthyolith abundances, pelagic fish exhibited resilience to climatic perturbations. These results highlight that fertilization effects from terrestrial nutrient input, which operate on longer timescales⁴⁸, can alleviate nutrient limitations imposed by enhanced ocean stratification and support higher ocean productivity under sustained warming. Such responses may differ substantially from the suppressive effects of warming observed in the modern ocean⁴⁹, raising the possibility that declining productivity trends could be moderated or potentially reversed over longer timescales (Supplementary Fig. 7). While productivity dynamics in the PETM ocean cannot be directly applied to decadal-scale forecasts, they offer valuable insights into the long-term response of marine carbon cycling and ecosystem to ongoing climate change.

Methods

Site locations

For IAR analysis, we selected 105 samples from representative sites spanning major global ocean basins to capture a wide range of oceanographic settings during the PETM (Fig. 1). In the Pacific, Site 1209 (paleolatitude 22°N, paleolongitude 172°W, paleodepth ∼1900 m) was located on Shatsky Rise in the Northwest Pacific⁵⁰, whereas Site 1221 (paleolatitude 1°N, paleolongitude 115°W, paleodepth ∼3300 m) was located in the Equatorial Pacific⁵¹. In the Atlantic, Site U1403 (paleolatitude 45°N, paleolongitude 35°W, paleodepth ∼4374 m) was situated in the Bay of Biscay, North Atlantic⁵², and Site U1557 (paleolatitude 25°S, paleolongitude 12°W, paleodepth > 4000 m) on the Walvis Ridge, South Atlantic. In the Indian Ocean, Site U1514 (paleolatitude 59°S, paleolongitude 90°E, paleodepth ∼3000 m) was located in the Mentelle Basin⁵³. We additionally compiled published IAR data from Site 1220 (paleolatitude 1°N, paleolongitude 116°W, paleodepth ∼2900 m) in the Equatorial Pacific²⁸. These sites provide broad spatial coverage across diverse latitudinal and oceanographic settings for the IAR study.

For BAR analysis, we generated marine barite data from Site U1557 and compiled published barite records from sites geographically comparable to the IAR locations to facilitate cross-proxy comparison. As barite data are unavailable for Sites U1403 and U1514, we selected the geographically nearest sites from North Atlantic Site 549 (paleolatitude 45°N, paleolongitude 14°W, paleodepth ∼1500 m) and Southern Ocean Site 738 (paleolatitude 64°S, paleolongitude 78°E, paleodepth ∼1350 m), respectively¹⁶. Collectively, these sites provide a global perspective on export productivity changes in the PETM ocean.

Ichthyolith isolation and IAR calculation

The fish teeth isolation and quantification protocol mainly followed ref. ²³. Briefly, we disaggregated 0.3–3 g of freeze-dried sediments in deionized water and wet-sieved the samples through a 38 μm mesh. We then dried the >38 μm fraction in an oven at 30 °C and hand-picked ichthyoliths using a stereomicroscope and a fine brush, and recorded ichthyolith abundance as ichthyolith counts per gram of dry sediment (ichthyoliths g⁻¹). Due to the influence of sedimentation rates, raw ichthyolith counts cannot directly reflect temporal changes in fish productivity²⁴. Therefore, we calculated IAR (ichthyoliths cm⁻² kyr⁻¹) to reconstruct past fish productivity following Eq. (1):

$$\mathrm{IAR}=\mathrm{Ichthyolith}\times \mathrm{DBD}\times \mathrm{LSR}$$ 1

where Ichthyolith denotes the number of ichthyoliths (ichthyoliths g⁻¹); DBD denotes dry bulk density (g cm⁻³), as obtained from the IODP MAD Database (https://iodp.tamu.edu/janusweb/physprops/maddat.shtml) and LIMS Portal (https://web.iodp.tamu.edu/LORE/); and LSR denotes linear sedimentation rate (cm kyr⁻¹), calculated based on the age model described below. The product of DBD and LSR determines the mass accumulation rate (MAR, g cm⁻² kyr⁻¹). We calculated standard errors (SE) for IAR values based on Poisson counting statistics (SE = IAR/√N, where N is the number of ichthyoliths counted in each sample).

Barite extraction and BAR-derived P_new calculation

We extracted marine barite from bulk sediments using a modified sequential leaching protocol⁵⁴. The leaching steps were followed by digestion and reprecipitation steps to further remove other refractory components (e.g., rutile). We then examined the purified barite crystals under a scanning electron microscope coupled with energy-dispersive X-ray spectroscopy (SEM-EDX) to confirm their pelagic origin. Only pelagic marine barite was used to calculate BAR. Similar to IAR, we calculated BAR (mg cm⁻² kyr⁻¹) as a function of barite content (mg g⁻¹), DBD, and LSR following Eq. (2):

$$\mathrm{BAR}=\mathrm{Barite\; content}\times \mathrm{DBD}\times \mathrm{LSR}$$ 2

For sites other than Site U1557, we recalculated BAR using the published barite data^16,30,31 together with the updated age models and DBD values applied in this study to ensure inter-site comparability.

Based on the core-top calibration of marine barite and export productivity in the modern ocean, we converted BAR to P_new (g C m⁻² yr⁻¹) following Eq. (3) after ref. ²⁹:

$${\mathrm{P}}_{\mathrm{new}}={\left(\frac{{\mathrm{F}}_{\mathrm{Ba}}0.171{\mathrm{Ba}}^{2.218}{\mathrm{z}}^{0.476-0.00478\mathrm{Ba}}}{2056}\right)}^{1.504}$$ 3

where F_Ba denotes BAR (mg cm⁻² kyr⁻¹); Ba represents the seawater Ba concentration (nM kg⁻¹) at 1,700 m measured by GEOSECS from the nearest stations⁵⁵; and z denotes the water depth (m) of each site reported in the proceedings reports.

Carbon isotope analysis of bulk carbonate

For the newly drilled Site U1557, we analyzed carbon stable isotopes (δ¹³C) of bulk carbonate using a Kiel IV Automated Carbonate Device coupled with a ThermoFisher 253-plus gas Isotope Ratio Mass Spectrometry in a dual-inlet system. The freeze-dried bulk sediments were ground and homogenized with an agate mortar and pestle, and weighed ~0.3 mg for isotope analysis. The Kiel IV system digested carbonate-containing materials with phosphoric acid. All bulk carbonate δ¹³C values were reported relative to Vienna Pee Dee Belemnite (V-PDB). Repeated measurements of the in-house standard ICIER-CO-1 (δ¹³C = 0.13‰) yield a reproducibility of ±0.05‰. Due to reduced carbonate preservation under ocean acidification, some samples within the negative carbon isotope excursion interval fell below the detection limit (Supplementary Data 4).

Age model

Accurate age models and sedimentation rates are essential for IAR calculations. The ages of PETM samples are expressed as time relative to the onset of the PETM (in kyr), as defined by the base of the CIE. For Sites 1220 and 1221, we adopted the age models of ref. ⁵⁶, which were correlated to the astronomically tuned chronology established at ODP Site 690⁵⁷. For Site 1209, we adopted the astronomically tuned age model of ref. ⁵⁸. For Site U1403, we followed the age model of ref. ⁵⁹, with tie points derived from bulk δ¹³C correlations to Site 690⁵². For Site U1514, we used the preliminary age model of ref. ⁶⁰, which was based on the astrochronology developed for the Mentelle Basin⁶¹. As no published PETM age model is available for Site U1557, we constructed a preliminary age model by correlating the bulk δ¹³C record of Site U1557 to the PETM δ¹³C stratigraphy of Site 690 based on ref. ⁵⁶.

The cGENIE model setup

We used the intermediate-complexity carbon-centric Grid-Enabled Integrated Earth System model (cGENIE) to simulate the controls on marine export productivity during the PETM. The cGENIE couples a three-dimensional ocean circulation module to a two-dimensional energy-moisture balance atmosphere module⁶². Marine biogeochemical cycling is represented by phosphate-limited biological productivity and a prescribed remineralization profile for sinking particulate organic matter^45,62.

Our experiments adopted the early Eocene (55 Ma) paleogeography conditions⁶³. A pre-PETM equilibrium state (CP_NER_s) was first established with a fixed pCO₂ of 1112 ppm (4× pre-industrial levels) and a mean ocean phosphate concentration of 2.15 μmol kg⁻¹²². The model was spun for 20 kyr under a closed-system configuration, in which solutes removed through burial were instantaneously compensated by weathering input, maintaining a constant global ocean phosphate inventory.

To isolate the contributions of individual processes, we followed the sensitivity experiment framework of ref. ⁶⁴, in which atmospheric pCO₂ forcing, temperature-dependent phosphorus weathering, and redox-dependent phosphorus recycling were progressively introduced from the pre-PETM state (Supplementary Table 1). All transient simulations had a total simulation length of 50 kyr, as described below. The contribution of each forcing was quantified using pairwise differences between adjacent experiments (Supplementary Table 2).

For the PETM, atmospheric pCO₂ was doubled from 1112 to 2224 ppm over 5 kyr^22,65, followed by 45 kyr of constant pCO₂ at 2224 ppm to allow the system to approach a quasi-equilibrium state (PETM_CP_NER). Additional simulations with the same pCO₂ doubling over 3 kyr (PETM_CP_NER_3kyr) and 10 kyr (PETM_CP_NER_10kyr) were performed to test the sensitivity of POC export flux to the rate of atmospheric pCO₂ increase (Supplementary Fig. 7). In these simulations, phosphorus weathering fluxes were prescribed to balance phosphorus burial fluxes, maintaining the same ocean phosphate inventory as in the pre-PETM.

To evaluate the role of external phosphorus supply, we enabled phosphorus weathering under the same atmospheric pCO₂ doubling scenario (PETM_1×_NER). This parameterization links weathering-derived phosphorus input to temperature-dependent continental silicate and carbonate weathering⁶⁶, representing enhanced phosphorus delivery from land to the ocean associated with intensified hydrological cycling under warming. Unlike PETM_CP_NER, these experiments allow the global ocean phosphate inventory to vary, while phosphorus burial remains proportional to organic carbon burial and redox-dependent phosphorus recycling was disabled. Additional simulations with 1.3-, 1.6-, and 1.9-fold phosphorus weathering intensities (PETM_1.3×_NER, PETM_1.6×_NER, and PETM_1.9×_NER) were also conducted.

To further evaluate the role of internal phosphorus recycling, we additionally enabled redox-dependent phosphorus recycling building upon the PETM_1×_NER configuration (PETM_1×_ER). The phosphorus regeneration scheme follows the cGENIE implementation of ref. ⁶⁷ described by Hülse and Ridgwell (2025)⁴⁵, in which the regeneration ratio depends on bottom-water oxygen concentration and the C/P burial ratio of settling organic matter. This parameterization represents enhanced preferential phosphorus recycling under low-oxygen conditions, which increases phosphate return from sediments to the water column during ocean deoxygenation^11,68. Additional simulations with the same range of phosphorus weathering intensities (PETM_1.3×_ER, PETM_1.6×_ER, and PETM_1.9×_ER) were also performed.

Following the above setup, we extracted the model output of POC flux at the intermediate water depth (483 m) at the end of the 50-kyr simulation. This depth corresponds to the mesopelagic depth interval where pelagic barite accumulation is typically highest, thus providing the most appropriate comparison with BAR-derived export productivity estimates⁶⁹. Changes in export productivity are linked to primary productivity and surface-ocean dynamics; however, explicitly assessing changes in primary productivity is beyond the scope of this study. We note that, as with most Earth system models, key biogeochemical processes are parameterized and therefore subject to uncertainty in the absolute magnitude of simulated responses.

Author contributions

W.Y. conceived and supervised the study, X.Z. processed fish teeth samples and performed computations, R.Z. extracted marine barite, X.Z., R.Z., M.L., and W.Y. contributed to data analysis, W.Y. and M.-Y.T. provided samples, X.Z. and W.Y. wrote the first draft, and X.Z., R.Z., M.-Y.T., M.L., and W.Y. reviewed and edited the final version of the manuscript.

Peer review

Peer review information

Nature Communications thanks Toby Tyrrell, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This research was supported by the National Natural Science Foundation of China (92479102, 42376049), Guangdong Natural Science Foundation (2024A1515012537), and University Undergraduate Teaching Quality and Education Reform Project (SJZLGC202460) grants to WY, and the National Natural Science Foundation of China (42450195) grant to ML.

Data availability

The original data generated in this study are provided in the Supplementary Information Data file.

Code availability

The model configuration and experimental setup used in this study are described in the Methods and Supplementary Information.

Competing interests

Authors declare that they have no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77863-9.