Introduction

A holistic understanding of the carbon-climate dynamics of past warming events has important implications for CO2-induced anthropogenic climate change. The Paleocene-Eocene Thermal Maximum (PETM; ~56 Ma1,2,3) represents the largest disruption of the global carbon cycle in the Cenozoic4, which led to 5–6 °C global warming5,6, ocean acidification7, ocean deoxygenation8,9,10,11, and intensified tropical cyclones12,13. The prominent 3–6‰ negative carbon isotope excursion (CIE) registered in both terrestrial and marine sections is consistent with major emissions ( ~ 2000 to >13,000 Pg C) of 13C-depleted carbon to the atmosphere and/or ocean and on a timescale of a few to no more than ~ 20 kyr14,15. Recent work suggests that the North Atlantic Igneous Province (NAIP) and associated CO2 emissions may have triggered the PETM16,17,18, followed by carbon sequestration through organic carbon burial19 and silicate weathering20. The PETM was proceeded by a transient warming accompanied by a smaller CIE21—known as the pre-onset excursion (POE) and which is recorded in terrestrial records from the Wyoming Bighorn Basin21 together with only a few shallow marine sections (Atlantic coastal plain, southwest Pacific Ocean, the North Sea and the Pyrenean foreland basins)22,23,24,25,26. The POE is a short-lived warming event that occurred about 38 kyr to >100 kyr27 prior to the PETM onset with an estimated duration of no more than a few centuries22 to millennia27. As an environmental precursor to the PETM, the POE is absent in deep-sea sedimentary records because its short duration may have limited its preservation to surface and shallow water records22. Resolving a global POE signal could be further complicated by bioturbation, sediment mixing, and chemical burndown of deep-sea carbonates7,28, which could only be understood by studying shallow marine and terrestrial sections. The POE warming may represent an early warning signal on the instability of carbon reservoirs and set the stage for a climatic threshold crossing occurred during the PETM. Previous studies suggest that the PETM is modulated by astronomical forcing29,30,31, and linked with the POE via repeated, catastrophic CO2 release27, such as methane hydrate dissociation22, either as a direct response of the warming or via positive feedback mechanisms. Furthermore, the close timing between the initial stage of the NAIP and the POE suggests that volcanism and magmatism may also serve as a viable trigger32. However, the global extent of the POE, its relationship with the PETM and exact mechanisms that triggered the POE—whether from methane hydrate release, volcanic activity, or orbital drivers—remain debated.

Here we report ultra-high-resolution biogeochemical records from a recently discovered coastal shallow marine section in the eastern Tethys that span both the POE and the PETM (Fig. 1). The Kuzigongsu section (39°45’10” N, 75°17’29” E) is located in the western Xinjiang Uygur Autonomous Region of China, which was covered by the Turan Sea—an arm of the Tethys Ocean during the early Paleogene (Fig. 1). The eastern Tethys was a restricted shallow-water carbonate platform environment33, and a critical site for the formation of warm and saline intermediate water and the burial of organic matter34. Abundant calcareous nannofossils35 and well-preserved organic matter and oyster shells (Fig. S1) allow for an integrated sedimentological, biogeochemical, isotopic, organic geochemical, and global carbon cycle modeling approach to unravel the paleoenvironmental evolution of the eastern Tethys during the POE and PETM, thus filling a critical spatial data gap and advancing knowledge on forcing and recovery mechanisms of ancient hyperthermals.

Fig. 1: Maps showing the paleogeography, location and outcrop image of the study site at the Kuzigongsu section.
figure 1

(a), paleogeographic map of the study area during the early Paleogene106. The base map was created from https://www.odsn.de using the reconstruction data from ref. 107 and edited using Adobe Illustrator, (b), present location of the study site generated using Map data ©2019 Google, and (c), a photo of the outcrop. Panel (a) also shows other shallow water Paleocene-Eocene Thermal Maximum (PETM) and pre-onset excursion (POE) records in Aktumsuk108 (1), Kheu River and Guru-Fatima109 (2-3), West Siberian Sea110 (4), southern Tibet (Tingri and Gamba)111,112, Tarim Basin35, Denmark (E-8X, 22/10a-4, Grane, and Fur)18,64,113,114, Svalbard14, Arctic (Lomonosov Ridge)115, and Mid-Atlantic Coastal Plain Sites (Ancora, Wilson Lake, Clayton, and Millville located in the New Jersey, and South Dover Bridge or SDB and Cambridge-Dorchester Airport located in the Salisbury Embayment in Maryland)116,117,118,119,120,121.

Full size image

Results and discussion

Astronomically tuned high-resolution PETM and POE records from the understudied eastern Tethys

The PETM has been identified at the Kuzigongsu section through calcareous nannofossil biostratigraphy (the NP9/NP10 boundary) and carbon isotope stratigraphy35. It occurs at 19.9 m (on a depth scale of 0 to 48 meters in Fig. 2) and corresponds to a ~ 6–8‰ negative carbon isotope excursion (CIE)—among the largest CIEs observed in shallow marine sites4. The CIE magnitude is ~ 6.3‰ in carbonate35, ~ 6.0‰ in organic matter, and somewhat amplified in long-chain n-alkanes ( ~ 7.8‰), which is likely a result of an enhanced hydrological cycle36 and elevated pCO222,37. The primary δ13Ccarb signal is likely well preserved, based on: (1) the strong covariation between δ13Ccarb and δ13Corg (r2 = 0.75, p < 0.001; Fig. S3); (2) the fact that most δ13Ccarb and δ18Ocarb data plot within the area of primary carbonates38 (Fig. S3), and (3) the presence of only a weak correlation (r2 = 0.18, p < 0.001, Fig. S3) between δ13Ccarb and Mn/Sr, as a strong correlation would indicate diagenetic alteration39.

Fig. 2: Characteristics of the Paleocene-Eocene Thermal Maximum (PETM) and pre-onset excursion (POE) records at the Kuzigongsu section.
figure 2

(a, b), δ13Ccarb and δ18Ocarb from Wang et al. (2022)35. Note the two novel sea surface temperature estimates based on oyster fossil ∆47 and \({{{{\rm{TEX}}}}}_{86}^{{{{\rm{H}}}}}\) at 29.8 m depth. (c), wt.% CaCO3. (d), Mercury to total organic carbon content (Hg/TOC) ratio. (e), Hg concentration. (f), astronomically tuned age model based on magnetic susceptibility (MS) across the POE and PETM. The color bar represents spectral power and the green band represents the bandwidth of the precession cycles. Numbers in (f) indicate precession cycles assignments.

Full size image

The POE is found at ~ 8.4 m below the PETM onset within lower nannofossil Zone NP9a35 and occurs in a 1.2-meter-thick interval (10.3 to 11.5 m) characterized by a –1 to –2.5‰ CIE (Fig. 2). Specifically, we observed CIEs of –2.5‰ in carbonate and –2.1‰ in organic matter, but in contrast to the PETM, only ~ –1‰ in long-chain n-alkanes. The relatively smaller recorded magnitude in the n-alkane record is likely due to the lack of data at 10.8 m depth where δ13Ccarb and δ13Corg values reach their minima (Fig. 2).

Power spectrum analysis of the detrended magnetic susceptibility (MS) data series shows significant peaks in wavelength at 0.8, 1.2, 1.9, 3, 5, 6.5, and 9.8 m (see Methods, SI and Figs. S3–S5), with the filtered 1.2–1.9 m cycles interpreted as precession signal with an assumed 21 kyr duration and the filtered 5 to 9.8 m cycles as short eccentricity ( ~ 100 kyr). Spectral analysis revealed sedimentation rates averaging between 6.0 and 8.3 cm kyr–1 (Fig. S5) and suggests that the durations of the PETM and the POE at our study site are ~ 127 kyr and ~ 21 kyr, respectively (age model option 1; see SI and Supplementary Data 2 for details). The PETM and POE are separated by ~ 144 kyr ( ± 21 kyr). The estimated PETM duration of 127 kyr is shorter than inferred from the deep sea sites (e.g., ~ 170 kyr from Röhl et al.2 and Zeebe and Lourens3), likely due to incomplete preservation of the entire PETM at Kuzigongsu with a change in lithology that truncates the recovery phase. The POE onset duration of ~ 7.0 kyr (age model option 1) is similar to, but slightly longer than the 2 to 5.5 kyr estimated by Bowen et al.21. An alternative age model (Option 2) that accounts for the significant drop in wt.% CaCO3 and a likely truncation assumes the filtered 6–10 m cycles represent ~ 20 kyr precessional signal. This age model option provides a duration of ~ 39 kyr for the PETM, ~ 4 kyr for the POE and ~ 54 kyr between the PETM and POE, which suggests the study site only preserves the PETM onset and the plateau, rather than the recovery (see SI for more discussion). However, due to the uncertainty in the astronomically tuned age model, we assume that the POE onset duration ranges from 500 to 7000 years to cover the full range of reported values in the literature22,27.

Paleoenvironment of the eastern Tethys during the POE and PETM

We use a multi-proxy approach to reconstruct the paleoenvironmental evolution of the eastern Tethys during the POE and PETM (Supplementary Data 1). Our records (Figs. 2, 3) include C/N ratios as indicators of organic matter source, weight percent (wt.%) CaCO3 as a proxy for ocean acidification and detrital dilution, trace element geochemistry for marine nutrient and chemical weathering proxies, organic biomarkers as proxies for marine microbial communities, and mercury content as a possible indicator of the NAIP activity. Together, our new data suggest that the shallow eastern Tethys experienced profound environmental changes, including extreme warmth, eutrophication, and biological turnover. Furthermore, the moderately high sedimentation rates (optimal sedimentation rate fluctuates between ~ 6–8 cm kyr–1; Fig. S3, S4) at this shallow site (estimated water depth is ~ 30–50 m based on microfacies analysis and foraminifera indicators33) yield highly expanded records that provide unique details on the relationship between the PETM and the POE. Such details are generally obscured in deep-sea sites because of lower sedimentation rates, dissolution, and bioturbation7.

Fig. 3: Proxy-based reconstruction of environmental changes across the Paleocene-Eocene Thermal Maximum (PETM) and pre-onset excursion (POE) at the Kuzigongsu section, eastern Tethys. Relative age from the onset of the PETM is based on an astronomically tuned age model described in Methods and Supplementary Information.
figure 3

(a), δ13Corg from bulk organic matter and δ13Cn-alkanes from long-chain n-alkanes (nC25 in blue circles, nC27 in green squares, and nC31 in purple triangles); (b), Total organic carbon content (TOC); (c), organic carbon to nitrogen ratio (C/N); (d), Mn* (pink circles)63 and V/Al ratio (orange circles) as redox proxies; (e), Biomarker indices based on C29 hopane ββ/(αβ + βα + αβ) (blue circles), Crenarchaeol/(Crenarchaeol+isoGDGT-0) or Cren/(Cren+isoGDGT-0) (dark blue diamond), and C29 2-Methylhopane index (2-MeHop) (red squares).

Full size image

The section is characterized by a rapid decrease in wt.% CaCO3 from >80 wt.% to near 0 wt.% at ~ 9 m—a shift which precedes the POE and PETM and may be attributed to significant reduction of carbonate production, detrital dilution, and/or shallow ocean acidification7,40. The sharp decrease (−2.5‰) in oxygen isotopes of marine carbonate values (δ18Ocarb) is consistent with an abrupt and significant warming during the POE and PETM (Fig. 2b), though the magnitude is likely an artifact of diagenetic overprinting. Additionally, a portion of the δ18Ocarb decrease may represent a decline in local salinity as the δ18Osw as epeiric sites can be strongly influenced by freshwater input from surrounding continents5. Clumped isotope data from a well-preserved oyster specimen (at 29.8 m; Fig. S1) indicate that the eastern Tethys surface water temperature was around 32.5 ± 1.5 °C (1σ) at the recovery phase of the PETM (Fig. 2). This estimate is similar to our independent temperature estimate of 30.6 ± 4.5 °C (1σ) based on the \({{{{\rm{TEX}}}}}_{86}^{{{{\rm{H}}}}}\) proxy41 for the sample at the same depth. However, the thermal maturity is relatively high for this section and the cyclized isoGDGTs abundance is low, preventing us from obtaining a high-resolution and precise \({{{{\rm{TEX}}}}}_{86}^{{{{\rm{H}}}}}\) temperature record at the site (Fig. 2).

In the organic matter fraction, peak TOC and C/N ratios coincide with the lowest δ13Corg values during the PETM, suggesting increased terrestrial organic matter input at the study site, a likely consequence of intensified continental weathering and/or higher terrestrial primary production42. The inferred increase in terrestrial weathering is supported by the higher values of Ti/Al and K/Al ratios11. Elevated C29 hopane ββ/(αβ + βα + αβ) ratios (average = 0.3) during the PETM indicate increased input of fresh organic matter either due to higher primary productivity or increased flux of fresh terrestrial organic matter into the basin (Fig. 3e). Lower C29 hopane ββ/(αβ + βα + αβ) ratios (average = 0.1) in the pre- and post-PETM samples suggest relatively low primary production in the surface waters with background input of reworked and more mature organic matter from the surrounding continents42. Similarly, Crenarchaeol/(Crenarchaeol+isoGDGT-0) ratios range from 0.1 to 1.0, with a significant decrease during the POE and PETM. Crenarchaeol (with four cyclopentane rings and one cyclohexane ring) is considered as a biomarker for Thaumarchaeota43. The lower Cren/(Cren+isoGDGT-0) ratios during the POE and PETM therefore likely reflect a reduction in marine Thaumarchaeota, which may be attributed to warmer surface ocean temperature and lower dissolved oxygen concentration44. The occurrence of 2-methylhopanes (2-MeHop) in the PETM interval indicates a transient perturbation of surface ocean characteristics (Fig. 3). The C29 2-MeHop Index, calculated as 100 × (C29 2-MeHop)/(C29 2-MeHop + C29 Hop)45, ranges from ~ 0–38% with two prominent peaks, at 20.9 m and 28.5 m respectively, corresponding to the peak values of TOC and C/N ratios. Several studies reported that the occurrence of 2-MeHop in the sedimentary record can be viewed as indicators of stress responses to the capacity of microbial respiration under hypoxia46, nitrogen fixation47, increased productivity48, and changes in pH49, corroborating the interpretations of elevated primary productivity discussed above. Furthermore, the anomalously high C29 2-MeHop Index during the PETM may be attributed to marine nitrogen cycle perturbation as a result of biogeochemical changes. This is similar to observations of other major carbon cycle perturbations of the Phanerozoic, such as the end-Permian mass extinction event50, the end-Triassic extinction event51, and the Mesozoic Oceanic Anoxic Events52,53.

Ocean deoxygenation may have been enhanced by increased primary productivity from elevated nutrient input due to enhanced terrestrial weathering. This suggestion is supported by negative Mn* values (Eq. 1) from the POE to the PETM (Fig. 3), which are associated with more reducing conditions due to significant redox-related changes in the solubility of Fe and Mn54.

$${{{\rm{Mn}}}}*\,={{\mathrm{log}}}[({{{{\rm{Mn}}}}}_{{{{\rm{sample}}}}}/{{{{\rm{Mn}}}}}_{{{{\rm{shales}}}}})/({{{{\rm{Fe}}}}}_{{{{\rm{sample}}}}}/{{{{\rm{Fe}}}}}_{{{{\rm{shales}}}}})]$$
(1)

The values used for the Mnshales and Feshales are 600 and 46,150 ppm, respectively55. Furthermore, the inferred surface ocean deoxygenation is consistent with elevated V/Al ratios over the same interval (Fig. 3) because V ions ( + 4 and +5 valence) are closely coupled with the redox cycle of Mn56. Widespread deoxygenation is well documented in many ocean basins across the globe during the PETM8,57, including the North Sea58, the Arctic Ocean59, the Atlantic and Caribbean60,61, and the northwestern Tethyan margins62. However, no significant changes in these redox indicators were observed across the POE63, suggesting relatively stable redox conditions in the eastern Tethys at this time.

Mercury content (or Hg concentration normalized as a ratio to organic carbon content—Hg/TOC) has been used as a signal of NAIP activity by several previous studies18,64. Our site exhibits two prominent Hg/TOC peaks that preceded the onset of the POE ( ~ 11 kyr) and the PETM ( ~ 26 kyr) (Fig. 2), supporting a possible link between Hg source and the 13C-depleted carbon source. However, because of the overall low Hg concentrations at the study site, establishing a direct link between the NAIP and the Hg peaks is not straightforward. Low Hg is likely due to dilution by carbonate and detrital input, the long distance of the site relative to Hg source, and/or Hg transport via oceanic waters rather than global atmospheric transport18. Increased Hg concentrations across the POE and PETM compared to background values suggest that multiple possible sources and processes may have been at play in addition to the NAIP activity. For example, variations in Hg concentrations in the sedimentary records can be caused by changes in river runoff, weathering, transport of terrestrial materials, primary productivity, source of organic matter, and post-depositional processes (e.g., diagenesis and dissolution)18, which could become more important at the study site because of its restricted carbonate platform setting65. Deoxygenation and changes in organic matter preservation and transport cannot fully account for the excess Hg as shown by the steeper Hg gradient to TOC within the PETM and POE interval at our site (Fig. S6). Moreover, Hg fluxes associated with wildfire (e.g., Arctic region66, northeastern US margin67, and England68) may have been far less than the Hg fluxes associated with a large igneous province event69, and therefore cannot provide sufficient Hg into the study site. Principal component analysis (PCA) suggests that Hg is most closely related to C/N ratios (higher C/N ratio indicates more terrestrial organic source) and δ13Corg during the PETM, which reflect changes in source of organic matter and 13C-depleted CO2 emissions (Fig. S6). The C/N ratio exhibits no significant change across the POE, suggesting the increase in Hg and Hg/TOC ratio is unrelated to changes in source of organic matter. On the other hand, C/N ratio shows a large increase across the PETM, which indicates that changes in source of organic matter may have contributed to the increased Hg concentrations. These potential processes do not preclude volcanic involvement, however, especially via more complex pathways than simple atmospheric loading and deposition70. Despite these potential complex sources of Hg, we cannot completely exclude direct and indirect involvement of the NAIP in driving the Hg changes in the study section71. For example, the NAIP was active as early as 62 Ma71, and its peak activity may have encompassed both the POE and the PETM72,73,74,75. A negative shift in 187Os/188Os ratios has been observed prior to the PETM in several sites globally32,71,76,77, lending support to the occurrence of large igneous province activity prior to the PETM. Furthermore, hydrothermal vent complexes in the northeast Atlantic region78,79 further support that the NAIP activity can at least partially explain the observed Hg records.

Thermogenic CO2 emissions associated with NAIP activity during the POE

The PETM carbon emission history has been extensively modeled in the past, with estimated carbon emission rates ranging from 0.3 to 1.7 Pg C yr-1 for a CIE onset duration from ~ 3000 to ~ 20,000 years and cumulative amount of carbon added ranging from ~ 2500 to ~ 13,000 Pg C14,15,21,79,80,81. Because the carbon emission history preceding the PETM has not been systematically quantified in an Earth system model and very little is yet known about the CO2 source during this time22, we then focused our model analysis on the POE (Table S1 and Fig. 4). Our new high-resolution geochemical data, together with an orbitally tuned astronomical age model, provide a unique opportunity to assess the effects of CO2 emissions during the POE.

Fig. 4: Data assimilation results from our cGENIE Earth system modeling based on the pH-δ13CDIC double inversion of four scenarios based on different assumptions of POE onset duration.
figure 4

(a–d), δ13Csource values of the diagnosed carbon source for the four age models (see age model interpretation in the main text). (e–h), Model-diagnosed rates of CO2 emission for the four age models. (i–l), Cumulative amount of CO2 emitted for the four age models. The gray shaded area represents 1500 years.

Full size image

We quantify carbon emissions over the POE using a data assimilation approach that considers paired δ13CDIC-pH variation across the POE within an Earth system model of intermediate complexity cGENIE, following the approach detailed in Gutjahr et al.15. In this, changes with time in annual global mean surface ocean pH (derived from δ11B proxy data from the Mid-Atlantic Coastal Plain with a change of ~ −0.1 to −0.3 pH units22) constrain the emission rate of CO2 to the atmosphere. Similarly, the change with time in observed δ13C of annual global mean surface ocean DIC (δ13CDIC) (reconstructed by applying an anomaly derived from the δ13C data of a global compilation; Fig. S7) refines the δ13C value of the (pH-constrained) CO2 emissions. The novelty of this approach is that it offers a unique solution of the mean δ13Csource without having to make a specific assumption about the carbon source (e.g., compare with Cui et al.14; see Methods and SI for detailed model results and sensitivity tests). To account for the uncertainty in the POE onset duration, we place our records on four different age models, including age model options 1 and 2 from this study, an age model from the Bighorn Basin based on Bowen et al.21 and an assumed age of 500 years based on Babila et al.22 (a summary of our model results and sensitivity analyses for the POE is listed in Table S1). (Fig. 4).

The flux-weighted δ13Csource values across the entire emission duration vary between −30.8 and −44.5‰ for the four age models used in our simulations with the minimum change in pH suggested by Babila et al.22, consistent with a thermogenic CO2 source82 (−30 to −65‰; Fig. 4a–d and Table S1). Longer POE duration (e.g., Age 1 associated with ~ 7000 year POE onset) necessitates lower flux-weighted δ13Csource values (−44.5‰) over the entire emission interval (Fig. 4a) at slower emission rate ( ~ 0.2 Pg C yr−1). We note that the δ13Csource values become progressively lower from the POE onset, likely resulting from a faster rate of change toward its minimum values in the δ13C forcing. This may represent a shift from thermogenic methane to biogenic methane ( ~ −34 to <−70‰83,84) emissions during the development of the POE. The average carbon emission rate over the entire emission period ranges from 0.2 to 1.3 Pg C yr−1 (Fig. 4e–h), comparable to those estimated for the PETM from sill-degassed CO2 and thermogenic methane (0.2 to 0.5 Pg C yr−1 from Jones et al.16; 0.6 Pg C yr−1 from Frieling et al.79). Larger magnitude of pH changes (e.g., ΔpH = ~ 0.2 to 0.3) yield overall larger average peak CO2 emission flux (2.9 Pg C yr−1) and higher average δ13Csource values (−19.5‰) (Table S1), still consistent with largely thermogenic methane source. The pH change for the POE has been documented at only a single location, using a novel approach to measuring boron isotopes (δ11B), and therefore has a high degree of uncertainty22. Considering the smaller magnitude of δ13C excursion, the smaller degree of warming, its shorter duration, and the minor ecological responses, the changes in pH during the POE are unlikely to exceed those during the PETM (ΔpH = ~ 0.3)85. A higher average carbon emission rate is associated with shorter POE onset duration (Fig. 4e), which represents a combined impact of the imposed ΔpH forcing and age models used (Fig. S7).

The cumulative CO2 emission during the POE ranges from ~ 1030 to 1765 Pg C (Fig. 4i–l), with peak pCO2 reaching ~ 1180 to ~ 1220 ppm—a rise of ~ 350 to ~ 390 ppm above ~ 830 ppm (Fig. S8). The modeled cumulative carbon emitted during the POE falls within the range of the 400 to 1600 Pg C suggested by Babila et al.22 using similar ΔpH. However, if the actual ΔpH was at the lower end (lower than ~ 0.1), it is more likely that the carbon source was primarily biogenic methane. Associated with the diagnosed carbon emissions is a modelled global sea surface temperature rise (ΔT) of ~ 1.1 to ~ 1.3 °C (Fig. S8). Although the paleotemperature history of the POE is currently poorly known, existing Mg/Ca ratios of planktonic foraminifera from the mid-Atlantic coastal plain suggest that the surface ocean temperature increase was ~ 2 °C with an uncertainty of ± 1°C due to salinity variations22, consistent with our modeled temperature changes within uncertainty. The POE warming may also help explain the observed increase in warm-water coccolithophore taxa in the eastern Tethys35.

Thermogenic CO2 related to the NAIP activities may have been the dominant carbon sources during the POE via contact metamorphism by intrusive activity through hydrothermal vent complexes78. It should be noted that mantle convection models suggest that a peak NAIP carbon emission flux at ~ 0.5 Pg C yr−1 could occur between 1 and 20 kyr16, comparable to those simulated in our inversion experiments, despite the geochronology of the NAIP continental flood basalt sequences being not very well constrained71. It is also important to note that a caveat of cGENIE in interpreting our results is the lack of terrestrial biosphere and potential changes in orbital forcing, which could impact the climate responses and lead to uncertainties in carbon emission estimates. Although this study provides a range of estimates on the carbon source and emission flux during the POE, more precise δ11B-based global surface pH records, detailed history of the sill intrusion of the NAIP, sea surface temperature records from across different latitudes, and better-constrained geochronology of the NAIP activity are clearly needed to reduce the uncertainty of the estimated thermogenic carbon emission fluxes from the NAIP.

The evolution of mean core-top carbonate (CaCO3) with time in the model exhibits a smaller magnitude of δ13C decrease for simulations with bioturbation turned on compared to those without bioturbation (Fig. S9). Similarly, core-top wt.% CaCO3 also exhibits smaller degree of dissolution for experiments with bioturbation on (Fig. S9). Longer experiment duration allows for a larger CIE magnitude regardless of whether bioturbation is on. This is due to the combined effects of bioturbation and dissolution as a result of the cumulative carbon emission (Fig. S9 and Fig. 4i–l), supported by a comparable Eocene hyperthermal event86. These experiments support the inference that short POE onset duration (less than millennial timescale) and bioturbation are the main causes of the lack of POE signal in the deep-sea sedimentary records.

In conclusion, we report astronomically tuned, ultrahigh-resolution PETM and POE stratigraphic records from a recently discovered site in the eastern Tethys. Geochemical proxies based on carbonate, bulk organic matter, and biomarkers suggest that the eastern Tethys experienced profound carbon cycle perturbations during the POE and PETM. Our integrated stratigraphic data and Earth system modeling together suggest that the millennial time-scale POE may be attributed to mainly thermogenic CO2 emission associated with sill intrusion prior to the main eruption phase of the NAIP, with contributions from amplifying feedbacks such as biogenic methane release. Furthermore, our findings predict substantial carbon fluxes driving the POE (averaging 0.2 to 1.3 Pg C yr-1), which could be tested by refined geochronological investigations of potential sources such as the NAIP. The POE may have set the stage for the ecosystem threshold crossing and the extreme carbon cycle disruption occurred during the PETM.

Methods

Cyclostratigraphy and astronomically tuned age model based on magnetic susceptibility measurements and time series analysis

A total of 480 samples at 10 cm intervals spanning both the POE and the PETM weighing 4 to 8 grams were measured for bulk mass-normalized magnetic susceptibility (MS or χ) using KLY-4S Kappabridge after being crushed in a copper rock hammer and placed in a 2 × 2 × 2 cm3 cubic plastic holder. The MS measurements were conducted at the Paleomagnetism and Environmental Magnetism Laboratory at the China University of Geoscience (Beijing). Measurements were made at room temperature with an applied field amplitude of 200 A/m and frequency of 976 Hz. Each measurement is corrected for the contribution of the plastic sample holder. Each sample was measured three times, with the average value corrected by mass to obtain χ in units of m3 kg-1. Relative standard deviations between the three runs were smaller than 0.5%.

Time-series analysis was conducted using MS data with the open-source software Acycle V2.487 because MS measures the magnetic mineral concentration, and is considered as a proxy for detrital fluxes from land to the ocean88. The MS data series was first detrended by subtracting a 40 m “loess” trend (locally estimated scatterplot smoothing, a non-parametric method for a series of data smoothing with a default window size of 35%) to remove non-periodic or high-amplitude long-term trends following the procedures described in Li et al.89. The multi-taper method (MTM)90 with 2π tapers was used to estimate the spectrum for the detrended MS series and confidence levels (mean, 90%, 95%, and 99%) were provided to test against robust first-order autoregressive model AR(1) red noise in order to reveal the MS series’ dominant wavelength. The evolutionary power spectra were calculated with “Evolutionary Spectral Analysis” function in Acycle with a sliding window of 10 m and a step of 0.1 m to identify any secular trend in dominant frequencies, which may be attributed to variations in sedimentation rates. The time scale optimization (TimeOpt; Meyers91) and correlation coefficient (COCO; Li et al.87) methods were used to identify the optimal sedimentation rate using Acycle’s “COCO” and “TimeOpt” functions, which use 2000 Monte Carlo statistical simulations to test the null hypothesis of no orbital forcing. The evolutionary versions of COCO and TimeOpt functions (i.e., eCOCO and eTimeOpt) were used to track changes in sedimentation rates. In addition, the “Spectral Moments” function was used to estimate variable sedimentation rates based on a periodogram with two spectral moments: evolutionary mean frequency (μf) and evolutionary bandwidth (B) (Fig. S3, S4). Subsequently, “Dynamic Filtering” function was used to apply dynamic filtering and isolate interpreted precession cycles from the MS data series. Since the power of long-term cycles (i.e., short eccentricity cycles) may have muted the manifestation of precession cycles in the evolutive harmonic analysis (EHA), we remove the > 4 m cycles that may be associated with eccentricity cycles to reveal precession-related cycles as the most prominent signal in the EHA spectrogram (Fig. S5). The significant power of the interpreted precession cycles in the EHA spectrogram allows us to effectively isolate this signal from EHA (Fig. S5). We then use the precession cycles to construct an astrochronological timescale for the study interval. Analyses of TimeOpt and COCO indicate alternation of optimal sedimentation rates (i.e., 6.0 cm kyr-1 and 8.3 cm kyr-1) (Fig. S3). Spectral Moments, eTimeOpt and eCOCO together suggest the estimated sedimentation rate ranges from 4.2 to 10.6 cm kyr-1 with increased sedimentation rate during the PETM body (Fig. S3–S5).

Stable carbon isotopes of bulk organic matter and wt.% CaCO3

HCl-treated carbonate-free powders were measured for total organic carbon (TOC) and total nitrogen (TN) concentrations on a Vario EL-III elemental analyzer, and the δ13Corg analyses were made using a thermo DELTA plus XL mass spectrometer at State Key Laboratory of Organic Geochemistry, Guangzhou Institute of Geochemistry, Chinese Academy of Sciences. Three reference materials were used to monitor the measurement of carbon isotopic ratio of bulk organic carbon, which included black carbon (−22.43‰), Urea#1 (−34.13‰), and Urea#2 (−8.02‰). Precision based on repeated measurement of these three standards were 0.12‰, 0.08‰, and 0.09‰, respectively. δ13Corg values were reported in VPDB and analytical precision was better than ±0.1‰ based on replicate analyses of the standards processed with each batch of samples. Weight percent (wt.%) CaCO3 was measured using a modified acid soluble weight-loss method92.

Carbonate clumped isotope geochemistry

The carbonate clumped isotope thermometer is based on the thermodynamic stability of C–O bonds at varying temperature, in which “clumping” of the rare, heavy isotopes of carbon and oxygen (13C and 18O) occurs more frequently at lower temperatures93. The excess occurrence of the 13C18O16O isotopologue of CO2 relative to a stochastic distribution of the heavy isotopes among all CO2 molecules is referred to as the mass 47 anomaly and notated as ∆47, in which \({\Delta }_{47}=\left(\frac{{}^{47}R}{{}^{47}R^{*}}\right)\times 1000\) where 47R = [13C16O18O + 12C17O18O + 13C17O2]/[12C16O2] and * denotes a stochastic distribution of isotopes. Clumped isotope thermometry presents a significant innovation over oxygen isotope-based thermometry because the temperature estimate is independent of the bulk isotopic composition, and thus requires no assumptions about δ18Ocarb or δ18Owater. This mineral formation temperature can be used to calculate δ18O of ancient waters when paired with δ18Ocarb values of the same sample, which is measured concurrently with ∆47.

Carbonate clumped isotope measurements of one Eocene fossil oyster (Crassostrea sp.) and one modern oyster specimen (Crassostrea hongkongensis) collected from northern South China Sea (21°42'7.89” N, 111°55'44.61” E) in 2022 were made at the Pennsylvania State University in April 2022 (see SI). Approximately 8 mg of pure carbonate powder was digested in a 105% phosphoric acid common acid bath at 90 °C to yield CO2. Evolved CO2 was passed through a Protium Isotope Batch Extraction (IBEX) carbonate preparation line to purify the sample gas. The gas is passed through a cryogenic trap to separate CO2 from water, a silver wool-packed borosilicate column to trap sulfides, and a gas chromatography column packed with Poropak to separate CO2 from other compounds with a He carrier gas. The purified CO2 gas is once more frozen into a cryogenic trap before being frozen into a microvolume, and passed through a polished nickel capillary to the MAT 253 Plus bellows. Purified CO2 sample gas was analyzed on a Thermo MAT253 Plus dual inlet IRMS relative to an Oztech working gas.

∆47 values versus the working gas were projected to the Intercarb-Carbon Dioxide Equilibrium Scale94 (I-CDES) using a carbonate standard-based empirical transfer function. ETH 1, 2, 3, and 4 were measured to build the reference frame and for interlaboratory comparison, and IAEA-C2 and Carrara Marble were treated as unknowns. Individual replicates were averaged to create final sample ∆47 values and reported with a 95% confidence interval. Temperatures were calculated using the T-∆47 calibration of Anderson et al.95. The average measured ∆47 value for the oyster fossil is 0.573 ± 0.011 (2σ), while the ∆47 value for the modern oyster specimen is 0.604 ± 0.028 (2σ). The calculated sea surface temperature in the eastern Tethys based on early Eocene oyster fossil is 32.5 ± 3.9 °C (2σ). The calculated modern sea surface temperature based on modern oyster specimen is 21.6 ± 8.7 °C (2σ), falling in the range of the observed average annual sea surface temperature (24.1 ± 5.6 °C) in northern South China Sea in 2022.

Biomarker and stable carbon isotopes of long-chain n-alkanes

Around 11 grams of dried and powdered sample were extracted for their biomarker content using a microwave system (Milestone Ethos EX) and using 20 ml of a dichloromethane and methanol mixture (9:1). The total lipid extract was separated using silica flash chromatography and elution with hexane:DCM (9:1) for the apolar and DCM:MeOH (2:1) for the polar fraction. The apolar fractions were characterized on a Thermo Scientific ISQ single quadrupole mass spectrometer (MS) coupled to a gas chromatograph (GC). Compounds were separated using a fused silica column (50 m × 0.32 mm) with a ZB1 stationary phase and helium as the carrier gas. The GC was programmed for: injection at 70 °C (1 min hold), ramp to 130 °C at 20 °C/min, followed by a ramp to 300 °C at 4 °C/min (20 min hold). The MS continuously scanned between m/z 650-50. The apolar fractions were subsequently analyzed using an Isoprime 100 combustion isotope ratio mass spectrometer, coupled to an Agilent GC, to determine the δ13C of the long-chain n-alkanes. We used the same type of column and temperature program as used for the GC-MS analyses. Samples were measured in duplicate on the GC-C-IRMS, and the average is reported here. An in-house CO2 reference gas was used to calculate compound specific δ13C values relative to Vienna Pee Dee Belemnite (VPDB). δ13C values of the C29 n-alkane are not reported here due to possible co-elution with other lipids. All biomarker and stable carbon isotopes of long-chain n-alkane analyses were performed at the University of Bristol.

Methods for GDGTs

Polar fractions were filtered through a 0.45 µm filter at the university of Bristol. The filtered polar fractions were redissolved in hexane: isopropanol (99:1) and analyzed using a high-pressure liquid chromatography atmospheric pressure chemical ionization mass spectrometer for their GDGT distribution. We used two ultra-high performance liquid chromatography silica columns to separate compounds, following Hopmans et al. (2016)96, and analyses were performed in selective ion monitoring (SIM) mode.

The thermal maturity of the organic matter in this section was estimated using the hopane isomerisation index: C29 ββ/(αβ + βα+ ββ)97. The results indicate that the thermal maturity changes across the section, but the C29 ββ/(αβ + βα+ββ) ratio is consistently below 0.4 (Fig. 3). This is indicative for an elevated thermal maturity, but well below the oil window. Although this level of thermal maturity will not affect apolar compounds like hopanoids or n-alkanes, it is likely to impact more labile biomarkers such as glycerol dialkyl glycerol tetraethers (GDGTs)98. We determined the GDGT distribution in all samples. As expected with this level of thermal maturity, GDGT concentrations were low and, in most samples, branched (br)GDGTs were absent, as were isoprenoidal (iso)GDGTs containing cyclopentane rings. However, a few samples did have isoprenoidal (iso)GDGTs with cyclopentane rings. This includes the sample at depth 29.8 m that hosts the well-preserved oyster shell fossil. This sample has a TEX86 value of 0.76, which results in an SST of 30.6 ± 4.5 °C using the \({{{{\rm{TEX}}}}}_{86}^{{{{\rm{H}}}}}\) calibration99. Although we treat this estimate cautious as thermal maturity might have impacted the GDGTs distribution, this TEX86-based SST is consistent with the clumped isotope data from well-preserved oyster shell fossils from the same sample, adding confidence that we are able to constrain the SSTs at this site during the recovery phase of the PETM.

Earth system modeling

The carbon-centric Grid Enabled Integrated Earth system model (cGENIE) is an intermediate complexity climate model that couples a 3D ocean (36 × 36 grid, 16 levels) with a 2D atmosphere that has the capability to track biogeochemical cycling of elements, stable carbon isotopes, marine sediments, and continental weathering15,100. Bathymetry, paleogeography, planetary albedo, and wind fields are configured for the late Paleocene-early Eocene with the same initial and boundary conditions as Gutjahr et al.15. For example, the δ13C value of late Paleocene-early Eocene atmospheric CO2 (δ13CCO2) is set as ~ ‒5‰, and the atmospheric pCO2 is set as ~ 830 ppmv. The moderately high pCO2 allows for a small buildup of sea ice (0.5%) in the northern polar regions. We then run a number of ‘double inversion’ experiments in which δ13C of surface ocean dissolved inorganic carbon (δ13CDIC) and surface ocean pH22 are used as the two data assimilation constraints for the POE. The δ13CDIC forcing is based on the high-temporal-resolution δ13Ccarb data from the shallow Tethys Kuzigongsu section using astronomically tuned age models. For our inversion experiments, the model was first spun up for 20 kyr to establish the basic ocean circulation and climatic state under published late Paleocene-early Eocene boundary conditions, including paleogeography and paleobathymetry101,102. This is followed by an open-system spin-up of 200 kyr to allow the long-term δ13C cycle to reach balance. A range of inversion experiments were carried out (Table S1; Fig. S7–9). Although uncertainty exists for pre-PETM δ11B, the surface ocean pH at the end of the open-system spinup is 7.75, same as those used in Gutjahr et al. (2017)15, which is adapted as the initial surface ocean pH forcing in the “double inversion” experiment.

First, the “double-inversion” modeling takes the observed pH data, which constrains the flux and magnitude of CO2 emissions, and the observed δ13C values of the dissolved inorganic carbon of the surface ocean, which simultaneously determines the source of the emitted carbon by computing the δ13C values of the carbon source. At each model time step, a pulse of CO2 is emitted to the atmosphere at a given rate if the δ13C value is lower than the previous time step, and the modeled surface DIC δ13C values and the observed δ13C values at the Kuzigongsu section are compared. If the current modeled surface DIC δ13C value is higher than the data value, the δ13C value of the emitted CO2 is assigned a value of –100‰. In contrast, if the current modeled surface DIC δ13C value is lower than the data value, the δ13C value of the emitted CO2 is assigned a value of 0‰. δ13C values of the emitted CO2 between –100‰ and 0‰ can be achieved by binning the emission fluxes in time and averaging flux-weighted δ13C values. Justification for the choice of these end-member δ13C values of the emitted CO2 is provided in Gutjahr et al.15. During the experiments, cGENIE continually adjusts the rate and δ13C value of emitted CO2 into the atmosphere in order to simultaneously reproduce the two proxy records as a function of time. In these experiments, we assume that the POE onset occurred as a linear decline in both δ13C and pH simultaneously (Fig. 4a, b). We use the same “double-inversion” methodology in both the main experiments and the sensitivity experiments, both starting from the same open-system spin-up state (Table S1).

Sensitivity experiments and analyses

We carried out sensitivity experiments to explore the importance of the duration of the POE onset ( ~ 7000, ~ 1600, ~ 850, and ~ 500 years based on age model option 1, age model option 2, Bowen et al. (2016), and Babila et al. (2022), respectively) using a global compilation of marine carbonate δ13Ccarb records (Table S1; Fig. S7). We also tested the effect of larger pH decrease (i.e., ‒0.24 and ‒0.32 pH unit) in combination with each of the four assumed age model (Table S1). Additionally, we test the role of bioturbation on the carbon isotope excursion magnitude of core-top carbonates (Fig. S9).