Introduction

Geological records archive the extreme climate warming events during the Phanerozoic, characterized by conspicuous carbon isotope excursion (CIE)1. These events include the Permian-Triassic mass extinction (PTME)2, end-Triassic mass extinction (ETME)3,4, the Paleocene-Eocene Thermal Maximum (PETM)5, and other early Paleogene hyperthermal events6,7. It is crucial to understand the potential mechanisms driving extreme climates. The triggers of these hyperthermal events remain ambiguous, including eruptions of large igneous provinces (LIPs)8,9,10, methane release5, orbitally-paced climate change11, or the superposition of multiple mechanisms12. However, the eruptions of large igneous provinces (LIPs) are commonly agreed upon as the ultimate driver3,8. For instance, many compelling pieces of evidence link the PTME to the Siberian Traps LIP (STLIP)2, the Paleocene-Eocene Thermal Maximum (PETM) to the North Atlantic Igneous Province (NAIP)13, and the ETME to the Central Atlantic Magmatic Province (CAMP)4. The massive eruption(s) and emplacement of LIP magmas can release large volumes of CO2 and CH4 into the atmosphere, and these increasing atmospheric greenhouse gas concentrations are associated with global warming and carbon cycle perturbations14.

Recently, Mercury (Hg) concentrations have been widely used to track volcanic activity during geological time3. Volcanic eruptions can release large amounts of Hg and cause enrichment of Hg in the Earth-surface system, which can easily overwhelm the background detrital Hg flux15. The residence time of Hg in the atmosphere is sufficient (0.5–2 years) to ensure its wide dispersal after a major volcanic eruption16. Thus, Hg concentration has proven to be a useful proxy for volcanism in recent years, especially providing evidence of the link between LIPs and hyperthermal event intervals10,14,17. However, non-volcanic sedimentary processes can significantly influence Hg concentrations, potentially masking volcanic signals. These include environmentally controlled variations in Hg sourcing and deposition, as well as differential preservation conditions associated with weathering, oxidation, and diagenesis18. Therefore, careful evaluation is required when using Hg concentrations as a proxy for volcanic activity.

The late Paleozoic icehouse was the largest glaciation in the Phanerozoic and recorded the only transition from an icehouse to a greenhouse climate state since the occurrence of vascular plants and complex terrestrial ecosystem19. Volcanism has been considered a causal mechanism for the increasing atmospheric pCO2 and shift of the global climate, ending the LPIA7. Lu et al.20,21 suggested that volcanism was an important driver for carbon isotope perturbations and shifts in global climates in the North China Craton (NCC) during the LPIA. However, due to the lack of high-resolution data on carbon isotope and volcanic intensity, it is not possible to definitively prove the causal relationship between carbon isotope perturbation and global volcanism in the NCC during the LPIA. Furthermore, the northern margin of the NCC was a part of the southern boundary of the Central Asian Orogenic Belt (CAOB), which was regarded as a passive continental margin during the Late Paleozoic period22. Volcanism linked to regional subduction is believed to have generally intensified, as evidenced by the abundance of volcanic ash beds in the NCC23. Therefore, the relationship between carbon isotope perturbation and global volcanic processes during the LPIA in the NCC requires further study.

We investigated the Pennsylvanian to early Cisuralian Benxi, Taiyuan, Shanxi, lower Shihezi, and upper Shihezi Formations to test for a causal link between carbon isotope perturbations and volcanism. Focusing on the Huainan section in the NCC (Figs. 1 and 2), we report U-Pb zircon ages and derive high-resolution records of δ13Corg, stable carbon isotope values of individual n-alkanes (δ13Cn-alkane), Hg concentration, Hg relative to total organic carbon (Hg/TOC), Hg relative to total sulfur (Hg/TS), mercury isotopes, and Hg anomalies based on externally studentized residuals (ESRs) from the uppermost Carboniferous through the lower Permian succession. Moreover, we compiled published δ13C data from late Carboniferous to early Permian to compare with the data in this study. Our results suggest that the carbon isotope perturbations and the climate warming events at the Permo-Carboniferous may not be solely correlated with global volcanism. Other potential driving mechanisms could also contribute to these climatic changes (such as tundra conditions, methane release, or orbitally-paced climate changes), highlighting the complexity of the Earth’s climate system during LPIA.

Fig. 1: Palaeogeographic map of the Huainan section and reference locations.
figure 1

a Global paleogeographic map at the Carboniferous-Permian transition (modified from ref. 58). MOB Mongolian Block, NCB North China Block, SCB South China Block, NQ North Qiangtang Terrane, IC Indochina Block. Red circle represents the study section. Blue circles represent other marine and continental sections, respectively, for which carbon isotope data have been generated. b Present location of the Huainan area in North China Craton (NCC). c Paleogeographic map of the NCC and adjacent regions (modified from ref. 64), showing the location of the studied area. The red point represents the Huainan area.

Full size image
Fig. 2: Stratigraphic column, Sporopollen Zone, and geochemical proxies of the Huainan section in North China Craton.
figure 2

a–e Hg, δ13Corg, total organic carbon (TOC), total sulfur (TS), and Al2O3 in the Huainan section. The blue circles represent the zircon U-Pb ages from refs. 24, 30. The red circles and squares represent the zircon U-Pb ages in this study. The yellow points represent the δ13Corg data from limestone samples. B F. Benxi Formation, LSH F. Lower Shihizi Formation, Mud. Mudstone, Silt. Siltstone, Lime. Limestone, Sand. Sandstone, CS Indospora cingulata-Sinulatisporites shansiensis. Locally estimated scatterplot smoothing (LOESS) trendlines in (a–e) using Acycle software65 (0.1 smoothing, solid lines) with 1σ bootstrapped errors (dashed lines). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).

Full size image

Results

Zircon U-Pb dating

Two tuff samples were collected from the middle-upper part of the Taiyuan Formation and the bottom of the Upper Shihezi Formation, which are similar to other Late Paleozoic tuffs from the NCC20,24,25. A total of 100 zircon grains, each ranging in size between 50 and 150 μm, were extracted separately from the HN-2 and HN-3 samples collected from the Huainan section within the NCC. These zircons exhibit well-preserved euhedral morphologies and prominent oscillatory zoning under cathodoluminescence (CL) imaging, typical of magmatic crystallization. The U-Pb ages were determined by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) and are presented in Fig. 3 and Supplementary Table 1. LA-ICP–MS analyses yielded 38 concordant zircon ages from HN-2 and 52 from HN-3, both showing bimodal distributions (Fig. 3c, e). The younger populations yield weighted mean 206Pb/238U ages of 297.8 ± 2.3 Ma (MSWD = 2.9, n = 38) and 283.7 ± 2.0 Ma (MSWD = 1, n = 52), respectively (Fig. 3d, f). The weighted mean values of the younger zircon peaks are interpreted to represent the sedimentary ages of the tuffaceous layers. Such age constraints align with frequent volcanic episodes during the Pennsylvanian, supporting recurrent influxes of pyroclastic material into the sedimentary basin during this period.

Fig. 3: Zircon U-Pb Concordia diagrams, histograms of zircon concordia age distributions, and ranked 206Pb/238U age plot with weighted mean ages.
figure 3

a, c, e, g Concordia 206Pb/238U vs 207Pb/235U age plots (left panels) of laser ablation ICP-MS U-Pb analyses of zircons (data-point error ellipses are 1σ) for samples HN-1 to HN-4. b, d, f, h Corresponding histograms of zircon 206Pb/238U age distributions and the calculation of the weighted mean ages (right panels) for samples HN-1 to HN-4.

Full size image

One sandy mudstone sample and one sandstone sample are from the top of the Benxi Formation (named HN-1) and the middle-bottom of the Upper Shihezi Formation (named HN-4), respectively. HN-1 and HN-4 at the Huainan succession from the NCC respectively separated one hundred zircon crystals, with crystal sizes varying from 50 to 100 μm. The majority of zircon grains in our samples are characterized by fine grain size, sub-rounded morphology, and either faint or absent oscillatory zoning in cathodoluminescence (CL) images, consistent with a detrital origin. Detrital zircon U-Pb dating provides valuable constraints on sediment provenance and the maximum age of sediment deposition26. Since individual zircon crystals can predate the host sediment, the youngest zircon ages define the maximum depositional age (MDA), whereas the actual depositional age (TDA) must be younger than any constituent zircon grain27. Several statistical approaches have been developed to estimate MDA, including the youngest single zircon (YSG), graphical peak age (YPP), and youngest grain clusters based on overlapping age uncertainties at 1σ or 2σ (YGC 1σ and YGC 2σ), as well as the youngest detrital zircon population (YDZ)28. In this study, we use the YGC 1σ method, which considers the youngest coherent group of grains with overlapping 1σ uncertainties, to derive robust MDAs from the U-Pb dataset. We first used the YGC 1σ method to recalculate the weighted average age by using 9 reasonable youngest grain dates in HN-1 and 5 reasonable youngest grain dates in HN-4 (Fig. 3a, g). The youngest weighted 206Pb/238U ages are 307.1 ± 4.3 Ma from HN-1 and 276.4 ± 5.2 Ma from HN-4 (Fig. 3b, h). Therefore, the youngest weighted 206Pb/238U ages from HN-1 and HN-4 are their maximum depositional age.

δ13Corg and δ13Cn-alkane

Results for δ13Corg and δ13Cn-alkane values are shown in Supplementary Table 2 and Supplementary Table 3. δ13Corg values vary from −8.81‰ to −31.25‰ (mean average = −24.33‰). Except for the outlier (δ13Corg is −8.81‰), the δ13Corg values from siltstone and mudstone samples vary from −21.77‰ to −26.21‰ (mean average = −23.65‰), then there is an obvious negative carbon isotope value in the upper Benxi Formation (−25.16‰; CIE-I), and a significant negative excursion during the upper Taiyuan Formation (−23.25‰ to −25.16‰; CIE-II). The δ13Corg values from limestone samples vary from −22.76‰ to −31.25‰ (mean average = −27.8‰). Stable carbon isotope values of individual n-alkanes (δ13Cn-alkane) range from −25.1‰ to −32.4‰ with an average value of −28.7‰ (Supplementary Table 3). δ13Cn-alkane values of n-C21, n-C23, n-C25, n-C27, and n-C29 range from −27.73 to −28.9‰, −27.53 to −28.93‰, −27.15 to −28.94‰, −28.35 to −29.98‰, and −27.31 to 28.65‰ during the CIE-II, respectively. Thus, the magnitude of the CIE-II for the δ13Cn-alkane record is 1.16‰, 1.40‰, 1.79‰, 1.63‰, and 1.34‰, respectively.

TOC, TS, Al2O3, mercury, and mercury isotopes

Results for TOC, TS, Al2O3, mercury, and mercury isotope values are shown in Supplementary Table 2, Supplementary Table 3, and Fig. 2. TOC values vary from 0.03 to 24.33% (mean average = 1.23%). TS values vary from 0.014 to 9.30% (mean average = 0.65%), and Al2O3 values vary from 0.137 to 37.33% (mean average = 18.04%). Hg concentrations vary from 2.4 to 235 ppb (mean average = 38.8 ppb). Mercury is generally hosted by organic matter, sulfide, and clay minerals in continental strata9,10,14,17,29. Hg/TOC ratios vary from 0.18 ppb/% to 1375.81 ppb/% (mean average = 80.65 ppb/%). Hg/TS ratios vary from 4.81 ppb/% to 3454.55 ppb/% (mean average = 358.59 ppb/%). Hg/Al2O3 ratios vary from 0.07 ppb/% to 38.89 ppb/% (mean average = 9.29 ppb/%). Δ199Hg ranges from −0.16 to 0.18‰ with an average value of −0.018‰. δ202Hg ranges from −2.45 to −0.66‰ with an average value of −1.30‰.

Discussion

The age of the Permo-Carboniferous strata in NCC is only broadly constrained based on biostratigraphy, with large uncertainty25, and the Benxi, Taiyuan, and Shanxi Formations are diachronous across the NCC20, making a precise chronostratigraphic framework for the Pennsylvanian continental facies of the NCC difficult. The Permo-Carboniferous boundary in Huainan in the southern NCC is uncertain. Yang et al.24,30 constrained the Permo-Carboniferous boundary to the middle part of the Taiyuan Formation in the Yongcheng Basin and reconstructed a high-resolution chronostratigraphic framework in Southern NCC.

In this study, HN-2 and HN-3 are magmatic zircon, and HN-1 and HN-4 are detrital zircon. The sample HN-1 was collected at the top of the Benxi Formation; the youngest weighted 206Pb/238U age is 307.1 ± 4.3 Ma, indicating the maximum depositional age (MDA) for this sample. Therefore, the age of the top of the Benxi Formation in the study area is younger than 307.1 ± 4.3 Ma. Yang et al.24 reported a high-precision age of 301.13 ± 0.2 Ma from the bottom of the Taiyuan Formation, which supports the interpretation based on HN-1. Sample HN-2 was collected at the middle-upper part of the Taiyuan Formation, with a weighted mean 206Pb/238U age of 297.8 ± 2.3 Ma, which is consistent with a tuff zircon U-Pb age between 299.32 ± 0.12 to 295.65 ± 0.08 Ma collected from the middle to upper Taiyuan Formation in the Yongcheng Basin24. Therefore, the stratigraphic ages in the study area are relatively consistent with those in the Yongcheng Basin. Sample HN-3 was collected at the bottom of the Upper Shihezi Formation; the youngest weighted 206Pb/238U age is 283.7 ± 2.0 Ma. The sample HN-4 was collected at the middle-bottom of the Upper Shihezi Formation; the youngest weighted 206Pb/238U age is 276.4 ± 5.2 Ma, which means the age of the middle-bottom part of the Upper Shihezi Formation in the study area is younger than 276.4 ± 5.2 Ma. Overall, we constrain the studied succession of the Benxi, Taiyuan, Shanxi, Lower Shihezi, and Upper Shihezi Formations to the late Gzhelian-Kungurian.

The new radiometric ages provided here from the Huainan section help to develop a precise chronostratigraphic framework for the Pennsylvanian continental facies of the NCC, which can augment established biostratigraphy and previous limited radiometric dates25. Notably, the 206Pb/238U age obtained in this study is consistent with the age from the Yongcheng Basin. Therefore, this age model tightly constrains the Benxi, Taiyuan, Shanxi, Lower Shihezi, and Upper Shihezi Formations from the late Gzhelian to Kungurian. We constructed a high-resolution carbon isotope stratigraphy using organic carbon isotopes derived from mudstone and siltstone samples in the NCC. These data show a significant negative excursion during the late Gzhelian (CIE-I), consistent with records from the Orogrande Basin, Naqing, and Southwest Japan (Fig. 4)31,32,33. This isotopic excursion likely reflects a global perturbation in the carbon cycle. The δ13Corg values from limestone samples in the NCC are more depleted compared to those from mudstone and siltstone samples (Fig. 5). This difference may be attributed to variations in organic matter sources, diagenetic alteration, depositional environment, and biological productivity34. However, it is worth noting that the trend of δ13Corg observed in limestone samples aligns with that in mudstone and siltstone samples (Fig. 5a), implying a common underlying mechanism influencing δ13Corg variations in the studied area. In addition, compound-specific medium-chain and long-chain n-alkanes (C23–C27) δ13C analyses during the late Gzhelian also reveal a synchronous negative excursion (Fig. 6a, b), further supporting the occurrence of a characteristic negative carbon isotope event in this interval.

Fig. 4: Comparison of global trends of δ13C during the Permo–Carboniferous.
figure 4

a Huainan section in North China Craton in this study; b Orogrande Basin32; c, d Naqing section in South China31; e Naqing section in South China55; f SW Japan33; g SW Kansas44. The gray bands indicate negative carbon isotope excursions; Locally estimated scatterplot smoothing (LOESS) trendlines in (a–g) using Acycle software65 (0.1 smoothing, solid lines) with 1σ bootstrapped errors (dashed lines).

Full size image
Fig. 5: Bulk organic carbon isotope records and lithology-specific δ13Corg distributions in the Huainan section.
figure 5

a Trends of δ13Corg from limestone samples and mudstone and siltstone samples in the Huainan section; b, c Organic carbon isotopic composition histograms for mudstone and siltstone samples and limestone samples. Locally estimated scatterplot smoothing (LOESS) trendlines using Acycle software65 (0.1 smoothing, solid lines) with 1σ bootstrapped errors (dashed lines).

Full size image
Fig. 6: Comparison of geochemical records from the studied section with global climatic and volcanic proxies across the Permo–Carboniferous transition.
figure 6

a–d δ13Corg, δ13Cn-alkane, Hg, Hg/TS, and Hg-TS residuals records in this study. e Atmospheric partial pressure of CO2 (pCO2) from ref. 37. f Sequential variation of conodont oxygen isotope (δ18O) from the Naqing Section of South China38. g Sea level change from ref. 36. h Timing of volcanic activity66,67,68,69,70,71,72. i Spatial and temporal distribution of the main record of the late Paleozoic deglaciation for Paran´a Basin, Southern Africa, India, SE Asia, Eastern Africa, and Western Australia19,39.

Full size image

An interglacial global warming event has been recognized in the late Gzhelian, coinciding with a characteristic negative carbon isotope excursion35. This interval is marked by a short-term rise in sea level and a pCO2 peak in the late Gzhelian (Fig. 6e, g)36,37. Moreover, conodont oxygen isotope (δ18O) values from South China suggest a ~1‰ decrease in the late Gzhelian, indicating a warming trend in tropical sea surface temperatures (Fig. 6f)38. This warming event also correlates with a spatial and temporal distribution of the main records of the late Paleozoic deglaciation, including significant deglaciation records in the Paraná Basin (Brazil), Vissers in Karoo Basin (South Africa), Eastern Australia, and Western Australia19,39. In addition, due to the paleoclimate aridification generated by the Hercynian tectonic activity from Moscovian to Gzhelian, the paleotropical forests in Europe, North America, and Central Asia underwent a gradual collapse40. However, the paleotropical forest of the NCC persisted and expanded because of the stable geological setting41. A major floral turnover occurred in the NCC, possibly triggered by a glacioeustatic rise in sea level42,43. The coincidence of deglaciation, increased CO2 content, higher temperature, sea-level high stand, and the floral changeover event in the NCC defines a warm interval in the late Gzhelian, which existed between major late Paleozoic glacial advance/maximum events (Fig. 6).

Glacial expansion occurred from the Asselian to the Sakmarian18,39, however, the early Asselian is characterized by elevated δ13C values with a negative peak that has been called the early Asselian climate warming event24,25. The onset of this event is seen in the study section (Fig. 4a) from the Gzhelian-Asselian transition and reaches the peak with an obviously negative carbon isotope excursion in the early Asselian (CIE-II), which also occurred in the δ13Ccarb records of Orogrande Basin (USA)32, Naqing (South China)31, Southwest Japan33, and Southwest Kansas (USA)44, suggesting may be a global carbon perturbation during the early Asselian (Fig. 4). Furthermore, compound-specific n-alkane (C21–C29) δ13C analyses of Huainan section also showed distinct negative excursions during the early Asselian, similar in magnitude to the δ13Corg from the same stratigraphic interval (Fig. 6a, b). The n-alkanes in the C21–C29 range originate from both terrestrial higher plants (C25–C29) and aquatic macrophytes or peat-forming vegetation (C21–C23)45,46,47,48. The parallel variations in δ13Corg and δ13Cn-alkane during the early Asselian, exhibiting a synchronous shift in both timing and trend with coeval marine carbon cycle disruptions (Fig. 4), provide robust evidence for a global-scale carbon cycle perturbation. The short-term rise in sea level was also shown in the early Asselian (Fig. 6g)36. Yang et al.24 and Lv et al.25 demonstrated that the land surface paleotemperature, calculated using the chemical index of alteration (CIA), shows an increase in temperature in the early Asselian. In addition, δ18O values from South China suggest that tropical sea surface temperature increased, which is consistent with land surface paleotemperature in early Asselian (Fig. 6f)25,38. A continuous increase of CO2 also corresponds to the early Asselian climate warming event (Fig. 6e)37. The early Asselian climate warming event is correlated with a spatial and temporal distribution of the main record of the late Paleozoic deglaciation and thus reflects significant deglaciation in the southern Paraná Basin, Kalahari Basin of Namibia, and Western Australia49,50. Thus, the shifts in δ18O and δ13Corg, atmospheric pCO2, sea-level variations, and late Paleozoic deglaciation distribution records strongly suggest a warm interval during the early Asselian. Lv et al.25 suggested that the release of CO2 from volcanism was a crucial driver of climate warming in the early Asselian.

After the negative carbon isotope excursion during the early Asselian, δ13Corg and δ13Cn-alkane values exhibit increased variability, with no clear long-term trend in the study section (Fig. 6a, b). This scattered distribution suggests that, unlike the well-defined negative excursions of CIE-I and CIE-II, post-excursion δ13C fluctuations may reflect a more complex interplay of regional carbon cycle dynamics, sedimentary processes, and potential mixing of organic carbon sources. Additionally, while climatic proxies such as atmospheric pCO2, δ18O, and sea-level changes continue to fluctuate, their relationship with δ13C variations becomes less distinct (Fig. 6). These observations imply that the early Asselian carbon cycle perturbation may be followed by a period of more localized or transient changes in the carbon reservoir, requiring further investigation into potential controlling mechanisms. Although these warming events are not evident in all sections, this may be attributed to the low temporal resolution of section data (Supplementary Fig. 1).

Large igneous province eruptions are widely considered to contribute to large-scale CO2 release, potentially leading to global warming, ocean acidification, carbon cycle perturbations, and ecosystem disturbances29. The substantial increase in greenhouse gases is triggered by both the extensive eruptive emissions as well as the emplacement of LIP magma intruding into organic-rich rocks29. Terrestrial organic carbon isotopes are affected by the composition of CO2 in the atmosphere. Volcanism releases comparatively isotopically heavy CO2, and where magma intrudes into organic-rich sediments, massive volumes of 13C-depleted CO2 (−22‰) and CH4 (−60‰) can be released, which would increase atmospheric pCO2, while decreasing δ13C values in the atmosphere51.

The impact of such volcanic activity on the environment can be further traced through Hg concentrations, which serve as a reliable proxy for volcanic eruptions52. The uptake of Hg by ancient sediments partly depends on the availability of specific host phases. Generally, Hg is hosted by organic matter, sulfide, and clay minerals in sedimentary rocks10,14,17. Thus, to accurately characterize long-term variation in Hg fluxes, it is essential to normalize Hg concentrations to those dominant hosts53,54. However, variations in host phase concentrations can introduce significant biases when using Hg enrichments to infer volcanic activity. Low host phase availability may artificially amplify Hg/host ratios, leading to overestimation of volcanic Hg input, whereas excessive host phase concentrations can dilute the Hg signal, potentially masking volcanic episodes18. Therefore, to accurately assess long-term Hg flux variations and distinguish true volcanic signals from sedimentary effects, it is crucial to evaluate host phase abundance and apply appropriate normalization approaches. Beyond total Hg concentrations, mercury isotopes-particularly Δ199Hg-offer additional insights into Hg cycling, as they are primarily associated with photoreduction (e.g., photochemical reduction of aqueous Hg (II))17. Photoreduction processes of volcanic Hg (Δ199Hg ≈ 0‰) in the Earth’s surface environment result in positive Δ199Hg in marine systems (e.g., seawater and marine sediments) and negative Δ199Hg in terrestrial systems (e.g., soil and vegetation)17. Although in nearshore settings, an intermediate Δ199Hg value (i.e., near-zero) could also arise from the mixing of multiple Hg sources17. This isotopic framework still provides a valuable tool for identifying the origin of Hg in different depositional settings.

In this study, Hg concentration shows high correlation with TS (r = 0.74, n = 166), a stable correlation with TOC (r = 0.42, n = 164), and a minor correlation with Al (r = 0.085, n = 166) (Fig. 7). Moreover, mercury-sulfide complexes commonly show stronger stability constants than Hg-organic complexes, resulting in preferential uptake of Hg by reaction with hydrogen sulfide in euxinic settings54. The TS-TOC plot can discriminate the dominant Hg host in sedimentary rocks, in which the thresholds of TS = 1.0% and TS/TOC = 0.35 can separate most samples with an organic host from those with a dominant sulfide Hg host54. In our study area, most of the samples fall within the organic matter-host field (Fig. 7d). However, the sliding correlation coefficients between Hg and TOC, TS, and Al2O3 show sulfide may be the dominant Hg host during the late Gzhelian to Sakmarian, and sulfide and organic matter may be the major Hg host between Artinskian to Kungurian (Fig. 8a). The sliding correlation coefficients between Hg and TOC, TS, and Al can help to show how dominant Hg host phases vary stratigraphically, enabling a more robust identification of Hg anomalies potentially linked to exogenic volcanic input. Despite this, linear regression alone is insufficient for precisely quantifying the magnitude or statistical significance of potential anomalies relative to overall dataset variability18. A recent study has shown that externally studentized residuals (ESRs) can effectively identify Hg anomalies by quantifying deviations from expected trends based on regression models18. In standard statistical analysis, data points with externally studentized residuals exceeding 3 are typically considered outliers based on the “3-sigma rule.” Persistent positive residuals after normalization to lithogenic proxies (e.g., Al, Ti) may indicate true mercury enrichment events unrelated to host lithology variability, such as volcanic activity. Further details of this method are given in ref. 18.

Fig. 7: Mercury host phase identification using TOC, TS, Al, and TOC-TS relationships.
figure 7

a Cross plot of mercury Hg versus Total organic carbon (TOC). b Cross plot of mercury Hg versus Total sulfur (TS). c Cross plot of mercury Hg versus aluminum (Al). d TOC-TS. In (d), the thresholds of TS/TOC = 0.35 and TS = 1.0% separate most samples with a dominant organic Hg host (faint yellow field) from those with a dominant sulfide Hg host (gray field) (thresholds and base map from ref. 54).

Full size image
Fig. 8: Mercury, host proxies, and isotope records in the Huainan section.
figure 8

a The depth sliding correlation coefficients between Hg and TOC, TS, and Al2O3. The dashed and solid lines denote correlations required for significance at the 90 and 95% levels, respectively, as estimated using a Student’s t-test. T = Sliding window. b–e Hg, Hg/TOC, Hg/TS, Hg/Al2O3, and Hg-TS residual records in this study. f, g Δ199Hg and δ202Hg in this study.

Full size image

At the Huainan section, Hg concentration shows high correlation with TS (R2 = 0.66, i.e., r = 0.81, p < 0.01) during the late Gzhelian to Sakmarian. The volcanic intensity curve inferred from the Hg/TS ratios is shown in Fig. 8. The calculations of ESRs need to fulfill the following preconditions: geochemical data (TOC, TS, and Al2O3) exhibit at least moderate and significant coefficients of determination with Hg (R2 > 0.3, i.e., r > 0.547, p < 0.01)18. There are no obvious peaks of Hg/TS during the Gzhelian and Hg anomalies based on ESRs (~0; Supplementary Table 2), which means that maybe there is no record of volcanism in the NCC (Supplementary Fig. 2). The warming event in the late Gzhelian is associated with the CIE-I, showing an increase in organic sources. One potential source of the isotopically light C is volcanism; however, the evidence from Hg/TS and Hg anomalies based on ESRs may exclude this possibility. The other potential sources of the isotopically light C may be from the tundra, intensified rift-related tectonic activity7,51, methane release5, orbitally-paced climate change11, or the superposition of multiple mechanisms5,12. Further study is required in the study area during the late Gzhelian. A slight Hg/TS peak and Hg anomalies based on ESRs (~2.5; Supplementary Table 2) were recorded in the study area before CIE-II (Supplementary Fig. 2). Although the values do not exceed the conventional threshold of 3 for highly significant anomalies, it still represents a statistically elevated Hg level relative to the expected background. This suggests a potential exogenic Hg contribution (volcanism; VP-I). The negative Δ199Hg during VP-I suggests inputs primarily from terrestrial plants and soils, whereas the positive excursion of Δ199Hg (−0.14 to −0.10‰, Fig. 8f) aligns with an increasing contribution of seawater-derived Hg. This trend indicates an enhanced influx of volcanically sourced Hg, likely facilitated through atmospheric deposition. Furthermore, Lv et al.25 suggested that the early Asselian warming event was likely driven by global volcanic activity. Similarly, Huang et al.55 proposed that orbital configurations may have amplified the climatic feedback of volcanism, intensifying this short-term warming interval. Therefore, we propose that the early Asselian warming event was predominantly driven by volcanic activity, with the Skagerrak Large Igneous Province being the most plausible source. A Hg anomaly, indicated by elevated ESR values (>3), was recorded in the study area following CIE-II (Supplementary Fig. 2; VP-II). Although the Hg/TS ratio does not exhibit a distinct peak, this may result from TS masking the Hg signal. The negative Δ199Hg values (Fig. 8f) suggest inputs mainly from terrestrial plants and soils, potentially influenced by continental chemical weathering, soil erosion, combustion of organic-rich sediments, and wildfire activity3,14,56.

Hg exhibits a moderate correlation with TOC (r = 0.45, p < 0.05) and TS (r = 0.42, p < 0.05) during the Artinskian to Kungurian (Supplementary Fig. 3). There is a peak in Hg/TS and Hg/TOC records during the late Artinskian, with near-zero Δ199Hg indicating strong inputs of volcanic Hg lacking MIF (VP-III). The absence of anomalous signals in δ13Corg and δ13Cn-alkane records during the early Kungurian suggests that the observed volcanic activity was likely regional in scale and did not induce a significant carbon cycle perturbation.

In summary, our data records do not fully support a direct correlation between volcanic activity and carbon perturbations during the LPIA. The late Gzhelian carbon isotope excursion (CIE-I) lacks coeval Hg anomalies or isotopic signals, making a major volcanic trigger unlikely. Instead, it could be influenced by various mechanisms, such as methane release, permafrost degradation, orbitally-paced climate changes, or the simultaneous occurrence of multiple mechanisms, which warrant consideration. In contrast, the early Asselian excursion (CIE-II) is preceded by moderate Hg enrichment and a slight ∆199Hg shift, suggesting a possible pulse of volcanic Hg input. The decoupling between volcanism and carbon isotope shifts, particularly during VP-III, indicates that volcanic activity alone cannot explain the full complexity of carbon cycle dynamics during the LPIA. While local ecosystem variability may also play a role, our current dataset lacks independent paleoecological evidence to evaluate its contribution. Further multiproxy studies are needed to better constrain the relative importance of these drivers.

Conclusion

In this study, we present high-resolution records of Hg concentration, TOC, TS, and δ13Corg, supplemented by Hg isotope and δ13Cn-alkane analyses at a targeted resolution, from the latest Gzhelian to Kungurian stages in the NCC. Our findings reveal two global δ13C negative excursions (CIE-I and CIE-II) that coincide with shifts in atmospheric CO2 levels, conodont oxygen isotopes, sea-level changes, and deglaciation events, providing new insights into the interplay between volcanic activity and the carbon cycle during the LPIA. Notably, volcanic proxies such as Hg/TOC, Hg/TS, and ESR values show no significant anomalies associated with CIE-I, suggesting that volcanism was likely not the primary driver of this event. In contrast, moderate volcanic signals preceding CIE-II, including slight Hg enrichment and a positive ∆199Hg shift, indicate a possible volcanic contribution. The Hg anomaly in the late Artinskian lacks corresponding carbon isotope shifts, implying a localized volcanic event without global carbon cycle impact. The variable relationship between volcanism and carbon isotope shifts suggests that volcanic activity may not be the sole driver of these climatic changes, highlighting the need for more detailed investigation into the mechanisms behind the late Paleozoic ice age (LPIA) climate transitions.

Method

Geological setting

During the Permo-Carboniferous, the NCC was located between several significant tectonic and oceanic domains, bordered by the Inner Mongolia Uplift (IMU) in the north, the North Qinling Orogenic Belt (NQB) in the south, the Paleo-Tethys Ocean to the west, and the Panthalassic Ocean (PTO) to the east25,57. Paleogeographic reconstructions place the NCC along the northeastern edge of the Paleo-Tethys Ocean, at low paleolatitudes ranging roughly from 5° to 15°N57,58. The eastern segment of the Paleo-Asian Ocean was subducting below its eastern side (the current northern margin).

During the Late Carboniferous, terrigenous siliciclastics derived from tectonic uplifts progressively filled the cratonic basins within the NCC, a process associated with compressional regimes documented in previous studies59. The Middle-Late Carboniferous witnessed a prominent northward marine transgression, resulting in a paleogeographic zonation from northern uplands to southern shallow-marine settings22. Moreover, widespread pyroclastic horizons within Carboniferous coal seams reflect recurrent volcanic activity linked to ongoing plate subduction along the NCC margin60. The Huainan Coalfield is located in the southeast of NCC, the North Qinling Belt (NQB) to the south (Fig. 1b, c).

Samples

We collected 100 mudstone samples, 28 limestone samples, and 38 siltstone samples from the Benxi Formation to the upper Shihezi Formation from the Huainan section. Sampling locations are shown in Fig. 2. The samples were crushed using an agate ball and sieved through a <200 mesh and divided for total organic carbon (TOC) analysis, total sulfide (TS) analysis, δ13Corg analysis, major elements analysis, and mercury concentration analysis. We also selected 18 mudstone samples for mercury isotope analysis and compound-specific isotope analysis. Two tuff samples were collected from the middle-upper part of the Taiyuan Formation (named HN-2) and the bottom of the Upper Shihezi Formation (named HN-3), respectively. One sandy mudstone sample and one sandstone sample are from the top of the Benxi Formation (named HN-1) and the middle-bottom part of the Upper Shihezi Formation (named HN-4), respectively.

Major element analysis

Aliquots of each sample were prepared for different analytical procedures. The X-ray fluorescence spectrometry (PW4400/40) with a relative analysis error of ±5% was used to determine the content of A12O3. The major elements analyses were detected in the Petroleum Geology Research and Laboratory Center, RIPED.

TOC and organic carbon isotope analysis

Samples for TOC (total organic carbon) were first treated with phosphoric acid to remove inorganic carbon, and then the TOC values were measured using a carbon-sulfur analyzer (CS580-A) with the lower detection limit of 100 mg/g and an absolute analytical error of ± 0.2%. The δ13Corg was then measured using an Isoprime 100 IRMS (Elementar, UK) coupled with an Elementar Vario (Elementar, UK), and calibrated with USGS 40 (δ13CVPDB = −26.39 ± 0.04‰) and USGS41a (δ13CVPDB = +36.55 ± 0.08‰) reference materials. For every 12 samples, a laboratory reference-Gelatin from bovine skin (δ13CVPDB = 14.7 ± 0.2‰) was inserted for calibration and to monitor stability. The isotope results were analyzed as the ratio of the heavier isotope to the lighter isotope (13C/12C) and expressed as “δ” in parts per 1000 or per mille (‰) relative to internationally defined standards (Vienna Pee Dee Belemnite, VPDB). The measurement errors were less than ±0.2‰ for δ13C values. The δ13Corg and TOC analyses were conducted in the Petroleum Geology Research and Laboratory Center, RIPED.

Mercury concentration and isotope analysis

Hg concentrations were analyzed using an Atomic Fluorescence Spectrophotometry (AFS-2202E), at Petroleum Geology Research and Laboratory Center, RIPED. About 150 mg for siltstone samples and 100 mg for mudstone samples were used in this analysis. Certified reference materials (CRMs), GBW07312 and GBW07404, were simultaneously analyzed to evaluate Hg determination accuracy. The accuracy of Hg for CRMs is within 2.50–8.43% of the reported values. The relative standard deviation (RSD) of duplicate samples was within 5%.

Hg isotope analyses were measured using multi-collector inductively coupled plasma mass spectrometry (MC-ICPMS, Nu Plasma 3D) at School of Earth System Science, Tianjin University based on published methods61. Prior to measurement, sample trapping solutions were diluted to low nanogram-per-gram levels (typically 0.5–2 ng/g) to ensure a concentration match within ±10% of both the bracketing primary standard (NIST SRM 3133) and the secondary standards (NIST SRMs 3177 or 8610). All solutions were prepared in an identical acid matrix to maintain consistency. Instrumental mass bias was corrected using a dual approach: internal normalization based on the measured 205Tl/203Tl ratio using NIST SRM 997, and external standard-sample bracketing relative to NIST SRM 3133. Five Hg isotopes (198Hg, 199Hg, 200Hg, 201Hg, and 202Hg) and two Tl isotopes (203Tl, 205Tl) were measured. Mercury isotope compositions are expressed in δ notation, as defined by the following equations62:

$${{\rm{\delta }}}^{{\rm{X}}}{{\rm{Hg}}}=\left[{\left({\scriptstyle{{{\rm{X}}}}\atop} \!{{\rm{Hg}}}{/}^{198}{{\rm{Hg}}}\right)}_{{\rm{sample}}}/{\left({\scriptstyle{{{\rm{X}}}}\atop} \!{{\rm{Hg}}}{/}^{198}{{\rm{Hg}}}\right)}_{{\rm{standard}}}-1\right] \times 1000$$

where XHg is 199Hg, 200Hg, 201Hg, or 202Hg, and “std” represents the NIST SRM 3133 standard. Mass-independent fractionation (MIF) is expressed in capital delta notation (Δ, ‰), representing the deviation of δ199Hg, δ200Hg, and δ201Hg from values predicted by mass-dependent fractionation (MDF) based on δ202Hg, using the MDF law62:

$${\Delta }^{199}{{\rm{Hg}}}={{{\rm{\delta }}}}^{199}{{\rm{Hg}}}-0.2520\times {{{\rm{\delta }}}}^{202}{{\rm{Hg}}}$$
$${\Delta }^{200}{{\rm{Hg}}}\approx {{{\rm{\delta }}}}^{200}{{\rm{Hg}}}-0.5024\times {{{\rm{\delta }}}}^{202}{{\rm{Hg}}}$$
$${\Delta }^{201}{{\rm{Hg}}}={{{\rm{\delta }}}}^{201}{{\rm{Hg}}}-0.7520\times {{{\rm{\delta }}}}^{202}{{\rm{Hg}}}$$

The analytical precision for δ202Hg and mass-independent fractionation (MIF) values (Δ199Hg and Δ200Hg) was typically better than ±0.10‰ (2SD), based on repeated measurements of international standards (NIST SRM 8610) and representative sediment samples. These values were adopted as the analytical uncertainty for all Hg isotope data. All isotope data are presented in Supplementary Table 3.

Compound-specific isotope analysis

The analyses for n-alkane values were carried out using the Soxhlet lipid extraction method by adding dichloromethane (DCM) and methanol (MeOH) as solvents (9:1 ratio) for 24 h. After obtaining the total lipid extracts (TLEs), 5α-androstane was added as an internal standard. The samples were concentrated using rotary evaporation and fractionated using aminopropyl columns to remove polar compounds, followed by silica gel column chromatography for the isolation of n-alkanes, which were eluted using 3 mL of hexane. Subsequently, purified n-alkanes were analyzed using a GC-2010 gas chromatograph coupled to an Isoprime 100 isotope ratio mass spectrometer (IRMS). The GC effluent was oxidized in a combustion furnace (~850–1000 °C) to convert n-alkanes into CO2, which was then transported by helium carrier gas into the IRMS for δ13C measurement. The compound-specific isotope analysis were detected in the Beijing Craton Technology Co., Ltd.

U-Pb zircon geochronology

Zircon crystals were extracted using standard mineral separation procedures, including crushing, grinding, sieving, and subsequent separation via heavy liquids and magnetic techniques. Well-preserved, euhedral grains exhibiting clear oscillatory zoning under cathodoluminescence (CL) imaging were selected for U-Pb geochronology. The initial separation was conducted by Langfang Chenshuo Rock Mineral Separation Technology Service. U-Pb isotopic analyses were performed by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) at Wuhan Sample Solution Analytical Technology Co., Ltd. Isotopic data were processed using Data Cal and Isoplot 3.0 software, and Concordia diagrams were constructed to constrain zircon ages.