Introduction

As a typical greenhouse period in Earth’s history, the Mesozoic Era was characterised by the recurrent occurrence of global hyperthermal events1. Investigating possible tipping points, global changes and high-resolution terrestrial responses to these Mesozoic hyperthermal events will deepen our understanding of past and present carbon-cycle perturbations. The Late Triassic Carnian Pluvial Episode or CPE (~234–232 Ma) represents a warming event characterised by oceanic anoxia2,3,4,5 and significant enhancement of the hydrological cycle6,7 and continental weathering8,9. These phenomena led to significant changes to life in marine and terrestrial ecosystems3,10,11,12. In particular, the CPE resulted in a burst of reef growth and spread of neopterygian fishes in the sea3, as well as the radiation of terrestrial animals and plants, including dinosaurs13,14, turtles15, crocodiles16, and various modern conifers17,18. This interval is synchronous with carbon-cycle perturbations marked by one or more pronounced global negative carbon-isotope excursions (CIEs), with amplitudes of ~2–4‰11,19,20,21,22,23,24.

The emplacement of the Wrangellia Large Igneous Province (LIP) is considered a primary trigger of the CPE, likely amplified by the release of isotopically light CO₂ from surface feedbacks into the ocean-atmosphere system, driving a warmer, more humid climate and associated negative CIEs19,25,26. The identification of multiple mercury (Hg) enrichments and increased input of unradiogenic osmium (Os) in sedimentary successions encompassing the CPE support a contribution from elevated volcanic emissions26,27,28,29,30,31. Nevertheless, the Wrangellia LIP (estimated 5 × 1018 g C)19 alone could not have released enough isotopically light carbon to account for the magnitude (~2–4‰) of the global CIEs, as suggested by mass-balance calculations and modelling12,19,23,32. Consequently, despite evidence for volcanic emissions at the time of the CPE, climate-carbon cycle feedbacks may have been critical to the CPE.

The increased global humidity is widely considered a prominent feature of the CPE, supported by the record of higher terrigenous inputs into shallow-marine sediments and a shift from playa-lake and continental sabkha environments to rivers or freshwater lakes3 (Fig. 1 and Supplementary Text 1). However, given the lack of a high-resolution chronostratigraphic framework, the correlation between enhanced humidity on land and the CPE interval remains unclear33,34. In several regions, including the basins of the Newark Supergroup in the USA, sedimentary evidence of increased humidity during the CPE has not been identified35. In addition, records of arid events have been found in the Wessex Basin (UK)36, Central European Basin (ref. 37, but see ref. 38), and in terrigenous material (derived from aeolian dust) of a deep-sea sequence in the Inuyama area (Japan)39 (Fig. 1 and Supplementary Text 1). Therefore, the manifestation of global hydrological cycle changes during the CPE is still under debate33.

Fig. 1: Palaeogeographical reconstruction of the Carnian world and locations of Large Igneous Provinces (LIPs) and the data indicating environmental changes during the Carnian Pluvial Episode41.
figure 1

A: Wrangellia LIP; B: South Taimyr complex; C: Kara Dere Basalt; D: Huglu-Pindos volcanic; E: Qingling-Dabie Orogenic Belt; F: Nada Kangri; detailed data on the records locations are listed in Supplementary Text 1.

Full size image

In this study, we report on a continuous lacustrine sequence (Dalongkou section) in the southern Junggar Basin, northwestern China, which provides a record of the CPE based on biostratigraphy and radiometric ages (radiometric methods in Supplementary Text 2) and allowed the construction of an integrated astronomically calibrated age model. The high-resolution palaeontological, sedimentological and geochemical signals from the Dalongkou section enable the detection of the CPE onset. Additionally, mercury (Hg) concentrations and isotope ratios, proxies for enhanced volcanic activity40, have been measured within the established astronomical time frame. We further employ cyclostratigraphic and organic-carbon isotopic data to explore the carbon-cycle dynamics. Finally, we investigate the global changes in hydrological cycling during the CPE using a combination of palynological and sedimentological data, as well as Earth System modelling. Together, these data allow a comprehensive overview of climate–carbon-cycle dynamics, including potential driving mechanisms for the CPE, and coeval changes to the hydrological cycle.

Results and Discussion

Age and stratigraphy

The Junggar Basin in the Xinjiang Uygur Autonomous Region, northwestern China, is part of the Kazakhstan-Junggar Plate. The Dalongkou section (43°58′30″N, 88°53′03″E), located on the southern margin of the Junggar Basin, represents a high-palaeolatitude (~60°N) lacustrine succession that developed during the Late Triassic41 (Fig. 1). The Huangshanjie and Karamay formations, over 500 metres thick, are continuously deposited without obvious hiatus. The Karamay Formation features significant alluvial-fan and fluvial components, with large sets of coarse sandstone and interbedded mudstone42. The conformably overlying Huangshanjie Formation consists largely of lacustrine deposits, dominated by dark mudstones, locally siltstone, sandstone, and interbedded limestone (Fig. 2a). Diverse plant macrofossils, spores, pollen, and vertebrate fossils in the Karamay and Huangshanjie formations attest to the lacustrine origin of these sediments42. The vertebrate Fukangichthys fauna and the palynological zonation (Aratrisporites, Dictyophyllidites-Aratrisporites, and Lycopodiacidites-Stereisporites) suggest that the boundary between the Middle Triassic and Upper Triassic lies between the uppermost Karamay Formation and the bottom of the Huangshanjie Formation42. The U-Pb radioisotopic dating on two sandstone samples from the uppermost Karamay and lower Huangshanjie layers shows a maximum depositional age of 234.8 ± 0.8 Ma to 232.9 ± 0.7 Ma (Fig. 2 and Supplementary Text 2), further constraining the lower Huangshanjie Formation to the lower–middle Carnian43. These absolute ages place the onset of the CPE (~234 Ma) between the Middle-Upper Triassic boundary (237 Ma) and the stratigraphic level of the sample with the youngest depositional age of 232.9 ± 0.7 Ma.

Fig. 2: The integrated stratigraphy calibrated to the astronomical time scale of the Huangshanjie Formation, Dalongkou section (Northwest China).
figure 2

The lithology (a) is calibrated to the astronomical time scale (ATS). Detrended magnetic susceptibility (MS) series (b) and its Continuous Wavelet Transform (c). The Gaussian filtered 405-kyr long-eccentricity cycles (red curve: 0.0019–0.003 cycles/m) and amplitude modulation of the Taner filtered short-eccentricity cycles (orange curve: 0.0075–0.0115 cycles/kyr) of the detrended MS series are composited with it (b). d δ13Corg, e clipped δ13Corg (from −25.5‰) is calibrated to the ATS. The red curve in (e) is the Gaussian filtered 405-kyr long-eccentricity cycles in clipped δ13Corg series (0.0019–0.003 cycles/kyr). f continuous wavelet transform of the clipped δ13Corg series. Abbreviations: POE, pre-onset excursion; CIE, carbon-isotope excursion; Mud, Mudstone; Sha, Shale; Silt-mud, Silty mudstone; Argi-silt, Argillaceous siltstone; Sand, Sandstone.

Full size image

Cyclostratigraphy and chemostratigraphy

Magnetic susceptibility (MS) constitutes a sensitive proxy of lithological and environmental changes (see “Methods”). Utilising rough sedimentation rate constraints derived from age constraints (~6 cm/kyr), correlation coefficients and hypothesis testing (5- to 10-cm/kyr optimal sedimentation rate) (Supplementary Text 3), significant Milanković cycles can be discerned through time-series analysis of the MS records in the Huangshanjie Formation (Fig. 2b, c and Supplementary Text 3). The spectral analysis of the MS series indicates a series of obvious sedimentary wavelengths, i.e., ca. 34 m, 13.3 m, 10.8 m, 9 m, 8 m, 3 m and 1.6 to 1.3 m (Supplementary Text 3), approximately in accord with the ratios of the 405-kyr long-eccentricity, 173-kyr obliquity, 125–95-kyr short-eccentricity, ~33-kyr obliquity and ~21–17-kyr precession cycles around 234 Ma, based on the La2004 astronomical solution44. The 405-kyr long-eccentricity and 173-kyr obliquity cycles are widely regarded as metronomes for establishment of deep-time astrochronology45; therefore, the inferred 405-kyr long-eccentricity (50–25 m) and 173-kyr obliquity (16.7–12.5 m) wavelengths in the MS series are tuned to the 405-kyr and 173-kyr synthetically made sinusoid, respectively. The spectral analysis of the two tuning schemes shows highly correlated astronomical signals corresponding to the 405-kyr long eccentricity, 173-kyr obliquity, and 131–95-kyr short eccentricity components, thereby confirming the reliability of the astronomical calibration (Supplementary Fig. 5). The U-Pb age of 232.9 ± 0.7 Ma was used to constrain tentatively the ~4-Myr floating astronomical time scale (ATS) tuned to the 405-kyr pure sinusoid.

We combined biostratigraphy (palynology and vertebrate fossils) and U-Pb dating to place the CPE in the lower part of the Huangshanjie Formation, and further constrained the onset of the CPE using carbon-isotope data. We present high-resolution bulk organic carbon-isotope (δ13Corg) records spanning 316.20 m, covering the top part of the Karamay Formation and the lower–middle part of the Huangshanjie Formation (~234–230 Ma, Supplementary Text 4). Our results of Rock-Eval, kerogen macerals and compound-specific (long-chain n-alkane) δ13C analyses (analysis methods in Supplementary Text 4) indicate that the bulk organic carbon-isotope record from the Dalongkou section dominantly reflects the atmospheric δ13C-CO2 signal, which reflects global carbon-cycle perturbations (Supplementary Text 4). The δ13Corg values exhibit relatively stable background values of –25‰ to –23‰ in the basal 38.00 m, followed by pronounced (∼6.5 ‰, 139.2 kyr) and moderate (∼3.5‰, 114.3 kyr) negative excursions at 39.50 m to 50.50 m and 60.25 m to 71.75 m, named CIE1 and CIE2, respectively (Fig. 2d,e and Supplementary Fig. 15). Additionally, we found an abrupt but relatively weak negative δ13C shift (∼1.2 ‰, 19.0 kyr) stratigraphically below the level of the CPE onset (CIE 1, Fig. 3), which is here termed the pre-onset excursion (POE). A similar feature was first reported in the Palaeocene–Eocene Thermal Maximum46 and similar phenomena were then also identified at the Triassic–Jurassic transition47 and Cretaceous hyperthermal events (Oceanic Anoxic Event 1d48; Oceanic Anoxic Event 249). However, there remains uncertainty as to whether the underlying mechanisms are consistent across different events, and whether they represent global phenomena. Given its short-lived duration (e.g., several centuries to many millennia in the PETM and ~19 kyr for the CPE POE here)50, the POE may be present in other hyperthermal events but remains enigmatic as it can only be captured in those sections with sufficiently high sedimentation rate coupled with high-resolution analysis.

Fig. 3: Magnification of high-resolution data of the carbon-isotope excursions (CIEs) and model for the carbon cycle during the Carnian Pluvial Episode.
figure 3

S1: before the Wrangellia Large Igneous Province (LIP) eruption, carbon- isotope values were stable; S2: the Wrangellia LIP began releasing isotopically light carbon, pre-onset excursion (POE) appeared, and temperatures started to rise11; S3: the Wrangellia LIP continued or weakened, carbon-isotope values were relatively stable, and temperatures probably continued to rise; S4: temperatures reached their peak and carbon isotopes showed an abrupt negative shift; S5: the burial of isotopically light carbon exceeded its emission, and carbon-isotope values recovered.

Full size image

The two main CIEs in the Junggar Basin span about 374.2 kyr (Fig. 2 and Supplementary Fig. 15), which is broadly consistent with the time span (~300 kyr) of the first two CIEs recorded in the Bagong Formation of Tibet31 and the Dunscombe Mudstone Formation of England24. However, the entire CPE with episodic CIEs is generally considered to have lasted longer than 1 Myr based on cyclostratigraphy (~1.1 Myr), biostratigraphy and magnetostratigraphy (1.6 to 1.7 Myr)3,13,51. We here focus on the POE and CIE1 in order to decipher the onset mechanisms behind the CPE.

Mechanistic origin of the CPE

Mercury (Hg) concentrations and isotopes have been widely used as a fingerprint of far-field volcanic activity in both marine and terrestrial settings during the extreme climatic events in the Phanerozoic Aeon (e.g., ref. 52). To explore the possible relationship between the CPE and enhanced subaerial LIP activity, high-resolution (~14-kyr average sampling interval throughout the section and ~2-kyr sampling interval near the onset of the CPE) Hg concentrations and isotopes were measured (Supplementary Text 5), and show one Hg/TOC spike in the Dalongkou section as synchronous with the POE, with elevated Hg occurring through the CIE1 (Fig. 3 and Supplementary Fig. 15). Interestingly, the Hg/TOC spike pre-dating the CIE1, little discussed in the literature, may be widespread in Europe, South China, Tibet and Japan27,29,30,31, and occurs at a time of decreasing 187Os/188Os ratios recorded from far-field pelagic sections derived from the Panthalassa or palaeo-Pacific Ocean (Supplementary Fig. 16)4,28,53.

Pre- and post- POE intervals exhibit near-zero odd-Hg MIF values (mass-independent fractionation signature, most Δ199Hg values ranging from 0 to –0.04‰), whereas the POE interval itself is characterised by large ranges of odd-Hg MIF values (–0.12‰ to 0.05‰; Supplementary Fig. 15). This pattern provides evidence of a mixture of several potential sources of Hg, including terrestrial input and atmospheric deposition to the lake system during the interval of mercury enrichment40. The overall lower than volcanic (<0‰) range Δ199Hg values indicate proportionally more terrestrial Hg (e.g., organic-rich soils with Δ199Hg values of ∼ –0.25‰) imported into the lacustrine system40. However, some high Δ199Hg values also appear between the lower values, indicating that atmospheric Hg2+ transport (direct atmospheric deposition or photochemcial reduction of Hg2+) probably contributed to the mercury enrichment, thereby offsetting the low values of the terrestrial input. In short, the results point to a volcanically derived Hg anomaly just prior to and during the CPE onset.

The coincidence of the POE and concomitant volcanically derived Hg enrichment (prior to the CIE) most likely represents a pulse of volcanism that released a relatively small mass of carbon within a short interval and caused a weak warming event (Supplementary Fig. 17)47,48,54. A precursor carbon release or warming event may signal the growing instability of thermally sensitive carbon sources, potentially triggering a positive feedback loop to sustain subsequent warming (Fig. 3)50,54.

Our results show that the main carbon-isotope excursion (CIE1), which represents the CPE onset, was geologically rapid, lasting ~15.8 kyr (Fig. 3). This time interval is longer than comparable phenomena of the Triassic–Jurassic transition (1–10 kyr onset)47 and Palaeocene-Eocene Thermal Maximum (at least 4 kyr onset)55, but is shorter than the Permian–Triassic mass extinction (~40 kyr onset)56, Early Jurassic T-OAE/Jenkyns Event (~65 kyr)57, Early Cretaceous OAE1a (less than 36 kyr)58 and the Late Cretaceous OAE 2 (~80–100 kyr)49. Previous geochemical modelling results suggest that the carbon released by the Wrangellia LIP alone could not have generated the magnitude of the CIE1 ( ~3–4‰)20, and potential sources of additional 13C-depleted carbon are hypothesised to derive from weathering of organic-rich sedimentary rocks, thawing permafrost, or dissociation of sub-ocean-floor methane hydrates3,12,20,23. Such potenrial additional sources of isotopically light carbon do not constitute an isolated case for the Carnian, as most carbon emission estimations from major LIPs do not match the corresponding δ13C excursions (e.g., the Triassic–Jurassic transition47 and Palaeocene-Eocene Thermal Maximum54). Palaeotemperature reconstructions suggest relatively warm climatic conditions before the CPE, with a global mean air temperature of ~23–24 °C1,59 and tropical sea-surface temperatures in eastern Tethys of ~27 °C11. Under such warm conditions (Fig. 4a), the global permafrost surface area should be much smaller than that it is today60. Thus, the total amount of 12C-enriched greenhouse gases from thawing permafrost is too limited to generate the magnitude of CIE1 (Supplementary Text 6). However, a hydrate reservoir only 29% the size of the modern one (>1.05 × 10¹⁹ g C)61 could hold enough ¹³C-depleted carbon to produce a ~3–4‰ excursion in the δ¹³C of the exogenic carbon cycle (Supplementary Text 6). Therefore, the dissociation of sub-ocean-floor methane hydrates might be an important source of the isotopically light carbon released during the CIE1 (Fig. 3). Other feedback sources, such as peat and forest, could also contribute to the supply of 13C-depleted carbon.

Fig. 4: Simulated climate states before and during the Carnian Pluvial Episode (CPE).
figure 4

a Annual mean 2-m air temperature (shading), precipitation (contour, unit by cm/year) and near-surface wind before the CPE (with pCO2 set as 568 ppmv). b Simulated annual mean precipitation anomalies across the CPE (pCO2 changes from 568 ppmv to 1758 ppmv). The contours (40 cm/year) mark the arid climate zone before the CPE (low CO2 experiment). Stippling indicates regions where the anomalies pass the 95% confidence level (Student’s t-test). Our results indicate that the climate before the CPE was relatively warm with very limited coverage of permafrost. Under CO2 forcing, the CPE climate change exhibits spatial heterogeneity in precipitation anomalies, with the expansion of the subtropical dry zone featuring reduced precipitation around the mid-latitudes. The red stars represent the location of the study section.

Full size image

The sequence of events, including evidence for volcanic activity followed by a large CIE may support a scenario whereby the emplacement of the Wrangellia LIP prior to the CPE induced net positive climate–carbon-cycle feedbacks. The rapid release (~15.8 kyr) of isotopically depleted carbon from surface reservoirs such as methane hydrates could help explain the large amplitude of CIE1. Such a sequence may be consistent with strong positive feedback and potential tipping point (i.e., irreversible changes in the Earth system) behaviour during the CPE. However, we recognise that this is a simplified scenario, and the complexity of the event likely reflects multiple contributing factors. Further investigation is needed to understand fully the interplay of various processes that could have shaped the CPE.

Climate–carbon-cycle interaction through the CPE

Our results provide terrestrial insights into the climate–carbon-cycle interaction on 405-kyr long-eccentricity timescales through the CPE. Here, we use the term climate–carbon-cycle interaction to describe the coupled feedback system in which climate variables (e.g., temperature, hydrology and weathering) influence carbon-cycle processes (e.g., soil respiration, net primary productivity and CO₂ solubility), while the carbon cycle, in turn, regulates global climate by modulating atmospheric greenhouse gas concentrations (CO₂ and CH₄)62,63. The orbitally driven changes in Earth’s hydrological processes and carbon cycles, occurring over tens of thousands to millions of years, form the underlying pattern of climate–carbon-cycle interactions62,63,64. Previous research has indicated that the climate–carbon-cycle interaction was paced by 405-kyr long-eccentricity rhythm throughout the Oligo–Miocene time interval followed by a striking phase reversal at 6 Ma, signalling a reorganisation of the climate–carbon-cycle system in response to coldhouse conditions63. During eccentricity minima, cooler and more stable climates, within a generally warm background, facilitated the expansion of continental carbon reservoirs. This expansion led to greater land storage of isotopically light carbon and a corresponding rise in marine dissolved inorganic carbon (δ¹³C-DIC), with the reverse occurring during maxima63.

Our results reveal the pronounced in-phase variation between the filtered long-eccentricity cycles and the amplitude modulation of short eccentricity in the MS series over the CPE. This relationship suggests that the higher values of the filtered eccentricity cycles are correlated with high eccentricity (corresponding to high variance of precession), and vice versa (Fig. 2b). Remarkably, the calibrated δ13Corg fluctuations are paced by 405-kyr long-eccentricity rhythm, on the scale of ± 1‰ (Fig. 2e, f). Therefore, the climate–carbon-cycle interaction exhibits in-phase behaviour between the terrestrial δ13Corg series and 405-kyr long-eccentricity cycles during the CPE (Fig. 2). This in-phase behaviour is most pronounced in the CPE interval because the 405-kyr eccentricity-related forcing signal was amplified by internal climate feedbacks (e.g., enhanced seasonal differences, Supplementary Text 7) of the carbon cycle under hyperthermal conditions62 (Supplementary Text 3).

In conclusion, our findings show that the CPE terrestrial carbon cycling, at a δ13Corg scale of ±1‰, displays an in-phase relationship with the 405-kyr long-eccentricity metronome, which appears similar to the warmhouse climate–carbon-cycle present throughout the Oligo–Miocene interval63. This result, together with previous long-term carbon-isotope records64,65, shows that such a climate–carbon-cycle interaction may have been widespread throughout the warm Mesozoic Era, including hyperthermal intervals. Therefore, this climate–carbon interaction may be the norm after the emergence of vascular plants, whereas the coldhouse climate–carbon-cycle dynamics of the earliest Pliocene may represent a more unusual situation.

The hydrological cycle during the CPE

Our palynological results show an arid and warm climatic state in the Junggar Basin during the δ13Corg fluctuations over the interval extending from the POE to the CIE2 (Supplementary Text 8 and Supplementary Fig. 15). Over this interval, the sediments exhibit a substantial decrease in hygrophytic elements (e.g., Leiotrilete and Todisporites) from averaging 22% down to ~10%. Our AWI-ESM simulations confirm a more arid and warmer climate during this interval in the Junggar Basin (Fig. 4b).

Our results show that precipitation changes during the CPE exhibited spatial heterogeneity, accompanied by a poleward shift of pre-existing precipitation zones and no evidence for global humidification. The heterogeneous pattern of precipitation changes is characterised by predominantly increased precipitation near the Equator and at high latitudes, while subtropical latitudes exhibit diminished rainfall (Fig. 4b). Moreover, mountains and extensive continental areas can block moisture transport, resulting in enhanced net evaporation within the interior of supercontinents (Pangaea)47. In contrast to previous hypotheses regarding unusual global humidification during the CPE2,7,13,26, the integrated stratigraphy and Earth system model highlight the spatial variations in global precipitation patterns, characterised by increased aridification in continental interiors and the emergence of multiple precipitation centres in low-latitude eastern continents and polar regions, including the Arctic, northeastern Tethys, northeastern Gondwana, and the Antarctic (Fig. 4b). These precipitation changes are corroborated by geological records throughout the CPE (e.g. the progradation of the largest delta system in geological history in the Boreal Ocean and the westward migration of its main depocenter66; the large-scale increase of terrigenous inputs and the extinction of carbonate platforms in the southwestern Tethys continental shelf 67,68 and South China11,69) (Fig. 1). Our modelling of annual global changes, consistent with other palaeoclimatic simulations32,70,71, is unable to replicate the wetter climate in the western and northwestern Tethys as indicated by previous sedimentological evidence (e.g., refs. 3,51). The remaining proxy–model differences may be due to unresolved local features in precipitation or the implementation of palaeotopography. Nevertheless, the newly proposed pattern appears to largely conform to established hydroclimate features observed in other past warm periods72, thereby amplifying the spatial heterogeneity in precipitation distribution.

Implication for understanding past hyperthermal events

Understanding climate–carbon-cycle interactions and tempo of past hyperthermal events is essential to accurately project the consequences of anthropogenic carbon emissions. Our findings suggest that the CPE shares key features and perhaps driving mechanisms with several other hyperthermal events including the Triassic–Jurassic hyperthermal event, Cretaceous OAEs, as well as the Palaeocene–Eocene Thermal Maximum. Although these hyperthermal events commonly lasted up to a few million years, the onset of each appears geologically rapid, lasting 1–20 kyr, and may be consistent with a net positive feedback response of the climate–carbon-cycle to elevated volcanic carbon dioxide emissions. Moreover, during the hyperthermal events, different regions may have experienced periods of extreme drought or severe flooding, reflecting the contrasting extremes of climate. Thus, despite differences in nomenclature, these events share a fundamental nature as hyperthermal episodes. These similarities suggest that understanding their commonality might hold essential information on the nature of hyperthermal events and specifically which elements in the climate and carbon cycle contribute to these anomalously warm periods.

Methods

δ13Corg and TOC

Both organic-carbon isotopes (δ¹³Corg) and TOC contents were analysed for a total of 289 samples at the Nanjing Institute of Geology and Palaeontology, Chinese Academy of Sciences. Samples were trimmed to remove visible veins and weathered surfaces and ground using a sample crush (Vibratory Disc Mill RS 200) for geochemical analyses. One gram of homogenised sample was placed in a polypropylene Falcon tube and decarbonated overnight with 3 molar HCl for 24 h. This process was repeated at least three times to ensure the complete removal of carbonates. The carbonate-free residual powder was rinsed 4 times with Milli-Q water to reach neutral pH. Samples were then dried at 60 °C in an oven. Dry samples were crushed, loaded into capsules, and then flash-combusted at 1060 °C in an Organic Elemental Analyser (FLASH, 2000) fitted with a zero blank autosampler. The carbon-isotope composition and TOC content (wt%) of generated CO2 were measured by a thermal conductivity detector and Delta V Advantage Isotope Ratio Mass Spectrometer, respectively. All isotope data are reported in per mil (‰) variation relative to the Vienna Pee Dee belemnite (VPDB) standard. Calibration of δ13C values was accomplished using a working standard B2151 (δ13Corg = −26.27‰), GBW04407 (black carbon, δ13Corg = −22.43‰) and Urea (δ13Corg = −39.79‰). The analytical uncertainty for TOC was calculated as a relative error of less than ± 5%, and the standard deviation for δ13Corg was ± 0.2‰ (1σ), both determined through replicate analyses of standard samples.

Mercury concentrations

Hg concentrations were analysed using a Direct Mercury Analyser (DMA80) at the China University of Geosciences (Wuhan). About 150 mg for siltstone/limestone samples and 100 mg for mudstone samples were used in this analysis. Results of low and high Hg were calibrated to the GBW07424 (33 ± 4 ppb Hg) and GBW07403 (590 ± 80 ppb Hg) standards, respectively. One replicate sample and a standard were analysed for every ten samples. Data quality was monitored via multiple analyses of GBW07424 and GBW07403, yielding an analytical precision (2σ) of ± 0.5% of reported Hg concentrations.

Mercury isotopes

Hg isotopes were analyzed at the State Key Laboratory of Environmental Geochemistry, Institute of Geochemistry, Chinese Academy of Sciences, Guiyang. The analysis of Hg-isotopic compositions followed the procedures described in ref.73. Hg was extracted and concentrated using the double-combustion and trapping dual-stage protocol73. The NIST SRM 997 thallium was used for mass-bias correction, the international Hg standard NIST SRM 3133 was used as an isotopic ref. 73, and the NIST RM 8610 (UM-Almaden Mono-Elemental Secondary Standard) was used as the reference material as a secondary laboratory Hg standard (National Institute of Standards and Technology, Gaithersburg, USA). Procedural blanks were negligible (<0.13 ng, n = 8) relative to the amount of Hg in samples (>20 ng). Good recovery (98 ± 4%, 2 SD) guarantees that no Hg-isotope fractionation occurred during the pre-concentration procedure. Volatile ionised Hg0 generated by SnCl2 reduction in a cold-vapour generation system was introduced into the plasma (Nu-MC–ICP–MS) with Ar as a carrier gas.

Hg-isotopic results are expressed as delta (δ) values in per mil (‰) variation relative to the bracketed NIST 3133 Hg standard, as follows74:

$${\updelta }^{xxx}{{{\rm{Hg}}}}=[{({}^{xxx}{{{\rm{H}}}}{{{\rm{g}}}}/{}^{198}{{{\rm{H}}}}{{{\rm{g}}}})}_{{{{\rm{sample}}}}}/{({}^{xxx}{{{\rm{H}}}}{{{\rm{g}}}}/{}^{198}{{{\rm{H}}}}{{{\rm{g}}}})}_{{{{\rm{standard}}}}}{-}1]\times 1000\textperthousand$$
(1)

Any Hg-isotope composition that did not follow the theoretical MDF was considered an isotopic anomaly caused by MIF. MIF values are indicated by capital delta (Δ) notation (in per mil) and predicted from δ202Hg using the following equations74

$${\Delta }^{199}{{{\rm{Hg}}}}={\updelta }^{199}{{{\rm{Hg}}}}{-}0.252\times {\updelta }^{202}{{{\rm{Hg}}}}$$
(2)

Reproducibility of Δ199Hg was better than ± 0.05‰ (2σ).

Elemental analyses

Major and trace elements were analysed for a total of 154 samples at the Nanjing Institute of Geology and Palaeontology, Chinese Academy of Sciences. An ~50 mg of sample powder was weighed and ashed at 110 °C for 3 h. Then, the sample powder was digested in 1 ml of HNO3 and 1.5 ml of HF at 190 °C for 72 h. After cooling, samples were evaporated at 115 °C to incipient dryness, then 1 mL HNO3 was added and the sample was dried again. This process was repeated at least three times to ensure the complete removal of HF. The resultant salt was re-dissolved with 1 ml HNO3 and 2 ml Milli-Q water before it was again sealed and heated in the bomb at 190 °C for 12 h. After completing digestion, these solutions were analyzed for the target elements on a quadrupole inductively coupled plasma mass spectrometer (ICP-MS) and ICP optical emission spectrometer. Two standards (SDO-1, SRM88b) were used to monitor the analytical reproducibility. Precision and accuracy for major and trace elements are estimated to be better than 5% and 1%.

Astronomical cycles

The total 1095 samples for the magnetic susceptibility (MS) experiment were cleaned to remove weathered surfaces and broken into small pieces of ~1–2 mm3, then filled with 8-cm3 non-magnetic (plastic) sample boxes. All samples were analyzed on the Kappa Bridge Magnetic Susceptibility System (sensitivity: 3 × 10–8 SI) at the State Key Laboratory of Lake Science and Environment, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences. Then, all samples were weighed by an electronic balance, with a mass accuracy of 10–4 g. The mass magnetic susceptibility (χ) was calculated by the ratio of the measured volume magnetic susceptibility (κ) to rock density. In a relatively humid environment, ferromagnetic minerals such as magnetite and maghaemite imply relatively high MS values, while in a relatively arid environment, carbonate deposits and coarse clastic rocks have relatively low MS values due to the diamagnetism of carbonate and quart75. The obtained MS series (χ) were interpolated with a main sampling interval of 0.25 m. To strengthen the orbital signals, we clipped values larger than 8.5 × 10–5 (cm3/g), and then detrended the series by removing linear trend. We analyzed the periodicity of the untuned and tuned MS series by the Multitaper Method (MTM)76 and Evolutionary Fast Fourier Transform77 spectra. The correlation coefficient (COCO) and null hypothesis testing (H0SL)78 were applied for quantitatively tracking of the sedimentation rates. In the two tuning schemes, the 50–25 m (0.02–0.04 cycles/m) wavelengths filtered via a Gaussian filter and the 16.7–12.5 m (0.06–0.08 cycles/m) wavelengths filtered via a Taner filter are tuned to 405-kyr and 173-kyr synthetically made sinusoids, respectively. The 405-kyr tuning scheme is used to establish a floating astronomical time scale, which is further constrained by the U-Pb age of 232.9 ± 0.7 Ma. The time-height mode is based on the calibration of the peaks of the filtered curve. In the time domain, the long and short eccentricity cycles were filtered by Gaussian (0.0019–0.003 cycles/kyr) and Taner (0.0075–0.0115 cycles/kyr) filters, respectively. The Hilbert transform was used to extract the amplitude modulation of the filtered short eccentricity cycles. The δ13Corg series is calibrated to the time domain based on the 405-kyr tuning scheme with the values below –25.5‰ clipped in order to highlight fluctuations forced by eccentricity, on the scale of ±1‰. Continuous wavelet transform was used to examine the periodicity of the tuned MS series and calibrated δ13Corg series (values clipped below –25.5‰). The proportion of hygrophyte was calibrated to the time domain based on the age model of the established ATS. All of the time-series analyses were conducted using Acycle 2.879.

Model simulations

To examine the spatio-temporal changes in temperature and precipitation during the CPE, we performed climate simulations using AWI-ESM80. AWI-ESM is a coupled ocean-atmosphere model widely used in palaeoclimate research. The land-sea distribution and topography is applied from the reconstructed palaeo-digital elevation model41 from 235 million years ago. There is no continental ice sheet. The ocean component of AWI-ESM employed an unstructured triangle mesh with horizontal resolution ranging from 30 ~ 270 kilometres (Supplementary Fig. 18), whereas the atmosphere component used a consistent horizontal resolution of 3.75o × 3.75o. We performed climate simulations under different atmospheric CO₂ levels, i.e., 568, 1136, 1704, and 2272 ppmv. At the end, the 568 ppmv experiment is used as an analogue of the climate prior to the CPE, while the 1704 ppmv experiment is used to mimic the relatively warmer climate during the CPE. Two criteria were used in the selection process: first, the selected CO2 levels are close to the proxy-based reconstruction of atmospheric CO2 before and during the CPE81; second, the simulated global mean surface temperatures are close to those previously reconstructed1,41. A similar approach was also applied in previous studies to reconstruct the deep-time Earth’s climate71,82. It should be noted that the simulated climate change patterns in greenhouse climates do not differ much depending on the CO2 level. Differences in CO₂ levels primarily affect the amplitude of climate changes. All other boundary conditions, such as solar constant and Earth’s orbit, were kept the same as the pre-industrial era, unless specifically stated. The 568 ppmv experiment was run for 2000 years to allow the climate system to reach a surface equilibrium state. The 1704 ppmv experiment, initialised from the first experiment, was run for 1000 years. This procedure allows the upper ocean and surface climate to achieve an equilibrium state. The last 100 years of the two experiments were used for comparison.