Introduction

The Late Paleozoic Ice Age (LPIA; main phase at ca. 330–254.5 Ma)1,2 archives the longest Phanerozoic icehouse and documents the only known climatic transition from a persistent icehouse to a greenhouse on Earth since the emergence of higher plants and complex terrestrial ecosystems3,4,5,6,7. The LPIA is also the only geologic period where atmospheric CO₂ concentrations (pCO₂) most closely approximate those of the Quaternary glaciations8,9,10, and thus is an important deep-time analogue for investigating present and future global environmental changes and has been highlighted for research for decades.

However, the driving mechanisms of the evolution of the LPIA are controversial, which hinders a comprehensive understanding of this event. Some studies have suggested that the atmospheric pCO2 values during the LPIA were strongly coupled with surface temperatures and global glacial records11,12 and therefore attributed the evolution of the LPIA to carbon (C) cycle anomalies caused by atmospheric pCO2 changes3,4,13,14. The mainstream view is that the terrestrial C cycle was affected by the expansion of terrestrial higher plants, silicate weathering15, and changes in atmospheric pCO2 caused by the Pangea Orogeny16. Nonetheless, the reconstructed atmospheric pCO2 values do not ideally match the glacial record7. For instance, atmospheric pCO2 exhibits an ambiguous relationship to the timing of onset, demise, and peak of the LPIA7. During 307–293.5 Ma, glaciers expanded continuously while pCO₂ fluctuated rather than decreasing steadily17. Near the peak of this icehouse period (ca. 299–290 Ma), pCO₂ oscillated frequently both above and below the ~560 ppmv CO₂ glaciation threshold11,16,17. As such, there appears to be some decoupling between glaciation and atmospheric pCO218.

Therefore, the evolution of the LPIA cannot be entirely attributed to atmospheric pCO2, and other factors might be involved. Volcanic activity is widely recognized to have played a key role in the evolution of the LPIA, based on the timing of volcanic eruptions and climate–ice sheet models11. In particular, during the Early Permian, large-scale volcanism (e.g., the Tarim large igneous province [LIP]) likely led to an increase in surface temperatures that marked the end of the Late Paleozoic glacial maximum12,19. However, in terms of volcanic–climatic effects, besides causing warming, volcanism might also have injected large quantities of sulfur (S; e.g., ash, SO2, and H2S) into the stratosphere or troposphere, which were oxidized and reacted with the hydroxyl radical and H2O to form sulfate aerosols20. This would induce climatic cooling by increasing the planetary albedo, which could possibly initiate glacial events. For instance, the Late Permian Emeishan LIP eruptions released ∼1017g sulfur gases into the atmosphere, contributing to climate cooling during the Capitanian–Wuchiapingian transition (ca. 260 Ma)21 and thereby potentially triggering the P4 glaciation of the LPIA22,23. Therefore, consideration of the radiative forcing effects of sulfate is essential to fully understand paleoclimate change and glacial events. However, the connection between LIP-driven C–S cycling and climatic changes and discrete glacial evolution during the LPIA remains poorly understood, thereby hampering understanding of its role in the evolution of the LPIA.

In this context, we investigated high-resolution continental stratum (the Lucaogou Formation, Junggar Basin) deposited from ~294 to ~285 Ma across the Sakmarian–Artinskian transition (ca. 290 Ma) in the mid- to high-latitude region (~45 °N) of northern Pangea24,25,26 (Fig. 1; Supplementary Note 1). This stratum is characterized by organic-rich deposits, substantial carbon sequestration26,27, and abundant volcanic materials (e.g., ash, volcaniclastic debris, and tuff)27,28, with its volcanic records integral to the broader Tarim LIP29,30,31. The interval represents a key period, coinciding with multiple critical events: a climate transition from a long-lived icehouse (340–290 Ma)12 to a postglacial greenhouse state8,32,33,34, the end of the Late Paleozoic glacial maximum12,19, and dramatic volcanic–tectonic activities (e.g., the eruption of the Tarim LIP35,36 and the closure of the Ural Seaway37). These temporal and spatial associations thereby offer an ideal archive for investigating Early Permian volcanism (or LIPs) and associated climatic–environmental perturbations. Here, we examined the links between volcanism, Sakmarian–Artinskian climate change, and the P1–P2 glacial history22 using sediment Hg concentrations and C–S isotope stratigraphy, to refine understanding of major geological events during the LPIA and obtain a deep-time perspective on current and future global warming.

Fig. 1: Geological setting of the Junggar Basin and sampling locations as well as stratigraphic framework of the Lucaogou Formation.
figure 1

a The location of the Junggar Basin and the distribution of some typical volcanoes, including Tarim LIP, Qiangtang, Panjal, and Choiyoi, are shown in the Early-Middle Permian paleogeography (modified from refs. 89,90.). b Regional structural framework of the Junggar Basin26, location of Well J10025 (orange star), and distribution of key geological units (including the Turpan - Hami Basin, Santanghu Basin, and major mountain ranges) are shown. The maximum known extent of Permian lake deposits is delineated by orange dashed lines26. c Chronological framework, stratigraphy23,24, lithology, and sample depth (yellow rhombuses) of the Lucaogou Formation in Well J10025 are presented. JJZG and QZJ denote the Jingjingzigou and Quanzijie Formations.

Full size image

Results and discussion

Terrestrial response to the transition from Sakmarian icehouse to Artinskian greenhouse

We investigate if the Lucaogou Formation of the Junggar Basin preserves a terrestrial record of the Early Permian climatic transition, as observed in many strata globally, such as South Africa’s Karoo Basin38, the central United States and Russia39, and the Yangtze Platform of South China40. This can be realized by correlations of carbon isotope excursions (CIEs), which are widely used for global correlations of major geological events, such as extinctions, anoxic events, and climate change, based on the principle that carbon isotopes record carbon cycle perturbations generated by geological events41,42.

In this study, the isotopic compositions of organic C (δ13Corg) and inorganic C (δ13Ccarb) in the Lucaogou Formation of the Junggar Basin are –31.08‰ to –16.52‰ (average = –27.49‰) and –1.92‰ to 12.71‰ (average = 7.22‰), respectively (Supplementary Note 2). Compared with the uniform values in the Lower Member, the δ13Corg and δ13Ccarb values in the Upper Member are more variable, exhibiting synchronous but generally opposing trends, and can be divided into intervals I, II, and III (Fig. 2a, b). These CIEs, including positive inorganic CIEs (pCIEs) and negative organic CIEs (nCIEs), can be employed to reflect environmental and climatic changes, as they are not caused by diagenesis, organic matter degradation, enrichment, sources (and mixing), thermal evolution, and kerogen type (Supplementary Note 3.1). Based on chronostratigraphic constraints24, the CIEs and δ13C trends in the Lucaogou Formation of the Junggar Basin are consistent with global marine and terrestrial strata record during the Sakmarian–Artinskian, such as those in the Yuzhou section in North China43, marine–coastal strata in the United States and Russia39, Zhongdi section in South China40 and the Dal’ny Tulkas section in southern Russia44 (Supplementary Fig. 4). Thus, the CIEs in the Lucaogou Formation of the Junggar Basin reflect Artinskian carbon cycle perturbations, which may have been closely related to changes in atmospheric pCO2 during the P2 stage of the LPIA11,45(Supplementary Note 3.1).

Fig. 2: Change trends and differences of carbon-oxygen isotopes and CIA in the upper and lower members of the lucaogou formation (Early Permian, Junggar Basin), compared with those in global marine strata across the Artinskian warming event.
figure 2

a, b Stratigraphic changes in δ13Corg and δ13Ccarb values (pCIEs and nCIEs refer to positive inorganic and negative organic carbon isotope excursions, respectively). c, d Stratigraphic changes in CIA and LSTs and δ18O values (note that the dashed circles represent data <–10‰; these are used primarily to help show the overall trend in oxygen isotopes rather than denoting the absolute value of the original fluid signal, with their trend indicated by the dashed line). e CIA change in the eastern Australia during the P1–P2 glaciations2. f Changes in oxygen isotopic compositions of low-Mg calcite from low- and high-latitude fossil shells46,47. g Early Permian global atmospheric pCO2 variations12.

Full size image

The chemical index of alteration (CIA) and land surface temperature (LSTs) were further used to examine paleoclimatic changes (Supplementary Note 3.2) to assess whether the climatic changes recorded by the carbon cycle in the Lucaogou Formation of the Junggar Basin are consistent with the global record. Results show that the CIA values of the Lucaogou Formation increase from 45–55 in the Lower Member to 55–85 in the Upper Member, and the corresponding LSTs also increase from 0–5 °C to 5–20 °C (Fig. 2c), which is consistent with the Sakmarian–Artinskian warming recorded by the increase in CIA values in the Karoo Basin of South Africa38 and eastern Australia (Fig. 2e)2. Given the sensitivity of carbonate oxygen isotopic compositions to temperature44, the trend is also indicative of a temperature change. The δ18O values of the Lucaogou Formation decrease by > 5‰ from the Lower to Upper Member (Fig. 2d), which is consistent with the Sakmarian–Artinskian warming recorded by the continuous decrease in δ18O values of marine low-Mg calcite fossil shells from high-and low-latitude regions of the Northern Hemisphere (Fig. 2f)46,47. Two cooling events (II-1 and −3) were further identified in interval II, which exhibits the largest carbon isotope fluctuations, corresponding to negative CIA excursions (Fig. 2c), positive δ18O excursions (Fig. 2d), and a decrease in LSTs of ~6 °C (Fig. 2e). This is also evidenced by other paleoclimatic indicators, such as C values, Fe/Mn, (Al + Fe)/(Mg + Ca), Rb/Sr and Sr/Cu ratios, which also show synergistic changes with LSTs (CIA) and δ18O (Supplementary Fig. 2 and Supplementary Note 3.2). The two short-term cooling events were superimposed on a warming trend.

The climatic changes and discrete glacial–nonglacial cycles recorded in the eastern Australia (e.g., C1–C3 and P1–P4) are the most detailed Late Paleozoic glacial-stratigraphic framework currently available2,22 and thus serve as an ideal analogue for global comparisons of the LPIA. The sedimentary and geochemical responses to the Late Paleozoic glacial events have also been obtained for several other basins (e.g., in south Africa, Brazil, and Antarctica)2. Both the Lucaogou Formation and the formation in eastern Australia exhibit an increasing trend in CIA values, with lower values in the P1 stage, a gradual increase in the nonglacial interval of P1–P2 stages, and negative excursions in the initial P2 stage (Fig. 2e). This indicates that the Lucaogou Formation of the Junggar Basin likely records the climatic fluctuations of the P1–P2 discrete glacial–nonglacial events (Supplementary Fig. 6), which may have been related to Early Permian atmospheric pCO2 changes, as the LSTs curve from the Lucaogou Formation is similar to the Sakmarian–Artinskian atmospheric pCO2 curve (Fig. 2c, g)12.

In summary, geochemical data from the Lucaogou Formation in the Junggar Basin document Artinskian carbon cycle perturbations and the second-order cooling events within the Artinskian warming stages. This represents a identified terrestrial record of the Early Permian climate evolution in the Northern Hemisphere, with implications of global significance.

Early Permian volcanism and Hg–C–S cycles

In the Lucaogou Formation of this study, the sediment mercury (Hg) concentrations, total organic carbon (TOC), total sulfur (TS) contents, and TS/TOC ratios in the nCIEs intervals are elevated and associated with negative δ34Sbulk (bulk sulfur isotope) excursions (nSIEs; amplitude > 10‰; Fig. 3a–c), collectively indicative of coupled Hg–C–S cycling and associated climate change. Sediment Hg enrichment and concentration anomalies are generally associated with volcanism and are reliable fingerprints of past volcanic activity48,49,50, and thus sediment Hg concentrations were used to assess whether volcanism was the main driver of the Sakmarian–Artinskian C–S cycle fluctuations (Supplementary Note 3.3).

Fig. 3: Variations in volcanic activity, Hg–C–S geochemical data (concentration and isotopes), and other related parameters in the Sakmarian–Artinskian Lucaogou Formation of the Junggar Basin.
figure 3

a Covariation of sediment Hg concentrations and Hg/TOC ratios. The peak values in Hg/TOC corresponds to that in Hg concentrations. Five volcanic events of different intensities (VP-1, −2, −3, −4, and −5) in two stages (LIP-Ⅰ and -Ⅱ) were identified (light gray shaded area). b Covariation of TOC contents and δ13Corg values. negative δ13Corg excursions (nCIEs) occur synchronously with increasing TOC contents. c Covariation of TS/TOC ratios and δ34Sbulk (bulk sulfur isotope) values. Negative δ34Sbulk excursions (nSIEs) occur synchronously with increasing TS/TOC values. d Hg/TOC ratios in isochronous strata of the North China Platform and Δ199Hg values, which together record volcanic activity43,57. e Covariation between a nutrient element (P) and geochemical proxy of primary productivity (P/Ti). f Variations in redox state (Th/U). g Covariation of land surface temperatures (LSTs) and δ18O values, and their relationship to volcanic activity.

Full size image

The background Hg concentrations of the Lucaogou Formation are 0–20 ppb, and there are five anomalies with high Hg concentrations in two stages (LIP-Ⅰ and -Ⅱ; Fig. 3a), with values of 50.1–351.7 ppb (average = 132.7 ppb) (Supplementary Data 1 and Note 2). The adsorbed form of Hg in the Lucaogou Formation is mainly onto organic matter (Supplementary Note 3.3), and thus Hg/TOC ratios were used in this study to trace volcanism and its intensity (Fig. 3a). The background Hg/TOC ratios are 0–5 ppb/% and the anomalies represent by high Hg/TOC values of 13.9–46.9 ppb/% (average = 28.0 ppb/%). The five Hg/TOC anomalies further include two peaks in the P1 glacial (LIP-Ⅰ; average Hg/TOC = 14.0 ppb/%, represented by VP-1 and −2) and three larger peaks in the P2 glacial (LIP-Ⅱ; average Hg/TOC = 37.4 ppb/%, represented by VP-3, −4, and −5) (Fig. 3a). Compared with published global Hg concentration data affected by volcanic activity (mainly at the Permian–Triassic [P–T] and Triassic–Jurassic [T–J] transitions, and in the Early Jurassic), with Hg concentration peaks of 50–400 ppb51,52,53,54, the data for the Lucaogou Formation of this study are similar. Typically, volcanic eruptions lead to an increase in Hg loading in the atmosphere–land–ocean system and modify Hg–TOC systematics49,51,55, which is also consistent with the synchronous increase in Hg concentrations and TOC contents in the Lucaogou Formation (Fig. 3a, b). Meanwhile, The Hg/TOC ratios peaks are synchronous with nCIEs (Fig. 3a, b), implying that the Hg enrichment in the Lucaogou Formation was the result of volcanism and organic-matter-rich sediments were affected by thermocatalysis during the ascent and intrusion of magma54,56, thereby inducing nCIEs.

Based on eruption timing and spatial distributions29,30,31, the two volcanic stages (LIP-Ⅰ and LIP-Ⅱ) recorded in the Lower Permian Lucaogou Formation of this study, which are genetically linked to the Tarim LIP35,36, align with Hg/TOC ratios and Δ199Hg anomalies in coeval North China Platform strata (Fig. 3d)43,57. This fine correlation indicates that Tarim LIP volcanism extended beyond the Tarim and Junggar basins (Fig. 1a), reaching wider regions (e.g., the North China Craton), thereby driving widespread nCIEs across these areas43,57. This coincided with Artinskian carbon cycle perturbations in global marine–terrestrial strata. Overall, these findings suggest that large-scale volcanism was likely the primary driver of Sakmarian–Artinskian carbon cycle perturbations and warming.

In addition to the volcanically-driven Hg–C cycle as discussed above, volcanism also controlled the sulfur cycle during the deposition of the Lucaogou Formation by regulating changes in lake sulfate concentrations, as evidenced by the coupled elevated TS/TOC ratios and nSIEs in the volcanic layers (Fig. 3c and Supplementary Note 3.4). The large amount of sulfur (e.g., H2S and SO2) released by volcanic eruptions reacted with hydroxyl radical in the atmosphere and was rapidly converted into H2SO420. The H2SO4 condensed into droplets on pre-existing particles or nucleated to produce new particles, which formed sulfate aerosols20 and eventually settled to the Earth surface system as acid rain and increased the sulfate concentration in the Lucaogou Formation (i.e., elevated TS/TOC ratios; Fig. 3c). This resulted in the weakening of microbial sulfate reduction (MSR) and the sulfate reservoir effect58, which caused the nSIEs (Fig. 3c) in association with the high Hg concentrations due to volcanism (Fig. 3a). Therefore, the nSIEs in the Lucaogou Formation record perturbations of the sulfur cycle caused by the Early Permian volcanism (e.g., Tarim LIP). Given the global significance of the carbon isotope stratigraphy of the Lucaogou Formation as discussed above, the correlated changes in the C–S cycle suggest that volcanic activity was likely also an additional driver of Early Permian global sulfur cycle perturbations.

In summary, Early Permian large-scale volcanism (e.g., the Tarim LIP) elevated atmospheric CO2 and SO2 concentrations, driving synchronized fluctuations in the Hg–C–S cycles.

Volcanically forced climate change

Based on the above results and discussion, a model of the C–S cycle is established to explain the short- and long-term climatic effects across the Artinskian warming triggered by volcanism represented by the Lucaogou Formation of the Junggar Basin (Fig. 4).

Fig. 4: Effects of volcanic C-S cycle on Early Permian climate change.
figure 4

Early Permian volcanism (e.g., Tarim LIP) that drove C–S cycling and associated long-term climatic warming and short-term climatic cooling are recorded in the Early Permian large-scale paleo-Juggar lake, Northern Pangea (modified from ref. 91 and references therein). VSRF, UV and OH refer to volcanic sulfate radiative forcing, ultraviolet and hydroxyl radical, respectively. CW, Pri-pro, nCIEs and nSIEs refer to chemical weathering, primary productivity, negative organic carbon and bulk sulfur isotope excursions, respectively.

Full size image

Firstly, the volcanic sulfate radiative forcing (VSRF) caused short-term climatic cooling and may contribute to intermittent glacial periods (Fig. 4). Volcanic activity drives synergistic changes in the Hg–C–S cycle, with a consistent response across multiple sedimentary basins, especially in the carbon cycle (Fig. 5a–d). The timing of the LIP eruption recorded in the Lucaogou Formation coincides with the initiation of the P2 discrete glacial (Fig. 5a, b), and the rapid decrease in LST values and increase in oxygen isotopes (Fig. 3g), as well as the synchronized nSIEs (Fig. 3c), suggest a rapid decrease in ambient temperatures at the onset of the eruption, rather than a sustained warming, which is mainly attributed to the VSRF effect. Based on the depositional duration (over ca. 3 Myr) and sedimentation rate (8–10 cm kyr-1) of the Lucaogou Formation during the greenhouse period26, two volcanic cooling events during the LIP-Ⅱ with temperature drops of ~6.6 °C and ~9.3 °C are roughly estimated to have cumulative durations of ~93.5 and ~167.5 kyr, respectively (Fig. 3a, g). Because the volcanism resulted in a marked increase in atmospheric sulfate concentration, the sulfate aerosols reflected ultraviolet rays and scattered solar radiation back to space, which cooled Earth’s surface20,59. Subsequent acid rain resulted in the high TS concentrations and TS/TOC ratios, and nSIEs in the Lucaogou Formation (Fig. 3c).

Fig. 5: Coupling of volcanism and its C–S radiative forcing effects with climate change and glacial evolution during the LPIA.
figure 5

a Early Permian spatiotemporal framework, showing timing constraints on P1–P2 glaciations in East Australia2 and volcanic and magmatic activities during 300–280 Ma (or P1–P2 interval). EAR (East Australian Rift67); T1–T3 (first, second and third phases of Tarim LIP36); SK (Skagerrak–Centred LIP71); PA (Panjal Traps LIP65); QT (Qiantang Basalts66). Orange43,57 and blue dots (this study) represent volcanic peaks tracked by Hg/TOC during the corresponding periods. (b–d) Early Permian Hg–C–S cycle, synchronous with the Tarim LIP and other volcanism. b Sharp increase in Hg/TOC ratios accompanied by negative Δ199Hg excursion in volcanic interlayers of the Junggar Basin and North China Plateau. c Variable magnitude nCIEs in global marine–terrestrial strata during the Artinskian warming, coinciding with LIP eruption phases. d Sharp increase in TS/TOC ratios with nSIEs in the volcanic interlayers of Junggar Basin. e Histogram of volcanics by type, binned at 10 m.y. intervals. “Other volcanics” refers to volcanic products excluding pyroclastics or ignimbrites as coded in original sources72. f Silicic LIPs record (red), modified from Bryan (2007)78, representing silicic volcanic provinces with documented volumes >105 km3. Mean tropical sea-surface temperature (green line) has been detrended and smoothed using a 50 Ma window with 10 Ma steps73. g Radiative forcing from volcanic sulfate and CO2 (ca. 360–260 Ma) versus glacial deposits per stage (note: the glacial deposition data herein excludes the updates from ref. 2). Comparison of negative radiative forcing attributable to pCO2 (green line), volcanic sulfate aerosol (red line), and total radiative forcing (orange line) relative to 560 ppmv CO2 (dashed line)72.

Full size image

However, due to the weaker volcanic intensity (low Hg peaks: ~50 ppb) coupled with generally lower temperatures (0–5 °C) during the LIP-Ⅰ, the cooling effect of VSRF was relatively insignificant, and thus no notable fluctuations in CIA and LSTs values occurred within this interval (Fig. 3g). In addition, the volcanism also enhanced the primary productivity (high P/Ti ratios; Fig. 3e) and organic carbon burial (high TOC contents; Fig. 3b) by rapidly injecting large amounts of nutrients (e.g., elevated P contents; Fig. 3e) into the Earth’s surface (or lake) system, further contributing to nCIEs, increasing the consumption and sequestration of atmospheric CO2, and amplifying cooling effects.

In contrast to the sulfur cycle, the volcanic carbon cycle causes long-term climatic warming (Fig. 4) and likely drives a shift from glacial to nonglacial conditions. During the transition from icehouse to greenhouse conditions, silicate weathering in the Lucaogou Formation intensified gradually (Fig. 2c), and organic carbon was buried (Fig. 3b), which is consistent with the global organic carbon enrichment in Lower Permian sedimentary rocks60. Theoretically, these processes drawdown and consume large amounts of CO2 from the atmosphere–ocean system, which reduces atmospheric pCO2 and cools the climate, which is inconsistent with the persistent climatic warming of the Early Permian. This suggests that there were other exogenous factors compensating for the reduction of atmospheric pCO2 and led to warming. This was most likely due to volcanism. Based on evidence from the Lucaogou Formation of this study, the volcanism during the glacial stage (i.e., during deposition of the Lower Member of the formation) was small in scale and of low intensity (sediment Hg concentration ~ 50 ppb), which caused small perturbations in the carbon cycle (nCIEs < 1.5‰; Fig. 3a, b) and almost no climate change (Fig. 3g). Volcanism during the greenhouse stage (i.e., during deposition of the Upper Member) was much larger in scale and intensity (sediment Hg concentrations up to ~350 ppb), which caused carbon cycle perturbations (nCIEs: ~6‰; Fig. 3a, b) and had dramatic climatic effects (Fig. 3g).

To further quantify the volcanic–climatic effects, the carbon input fluxes were roughly estimated based on the cGENIE model (a complex Earth system model61). Using a sedimentation rate of 8–10 cm kyr-1 as a baseline26, the duration of the multi-stage volcanism during the glacial (Lower Member) and greenhouse (Upper Member) periods of the Lucaogou Formation were calculated to be 104.5 and 291.5 kyr, respectively (with corresponding nCIEs of ~1.5‰ and ~6.5‰, respectively; Fig. 3a, b). If the input carbon source is only mantle-derived C (d13C = –6‰), then during the glacial period, the total carbon input and maximum annual carbon flux into the paleo-Junggar Lake from volcanism were 2–6 × 10³ Pg C and 0.1–0.2 Pg C yr⁻¹, respectively. During the greenhouse period, these values were 0.5–1.0 × 10⁵ Pg C and 3–4 Pg C yr⁻¹, respectively. The highest CO₂ emissions during the Artinskian period are roughly half those of the Siberian LIPs (~2.0 × 105 Pg C)35. This difference arises mainly because, relative to the peak eruption rates (8.0 km3 yr-1) and volumes (3–4 × 106 km3) of Siberian LIPs, the Tarim LIP is relatively smaller in scale, with corresponding values of ~0.06 km3 yr-1 and 0.15–0.30 × 106 km3, respectively35,36. However, The highest flux of volcanic input (0.5–1.0 × 105 Pg C and 3–4 Pg C yr⁻¹) clearly exceed both the total organic carbon burial flux (104 Pg C) and maximum carbon burial rate (4 × 10-3 Pg C yr–1) of the Lucaogou Formation26. This implies that volcanic carbon sources released by Tarim LIP may have mixed with other isotopically depleted carbon sources (e.g., via thermocatalytic release of terrestrial or rock-bound organic carbon or microbial methane release from hydrate reservoirs)61,62, yielding an input δ13C of –12‰ to –6‰61 and thereby triggering the nCIEs in the Lucaogou Formation.

For the volcanically release of SO2, we present a rough estimation based on magma volume. The Emeishan LIP has been extensively investigated, which has a total magma volume exceeding 3.0 × 105 km3 and at least 1500 Gt of SO2 was released35,63. For comparison, the total magma volume of the Tarim LIP is estimated to ~1.5–3.0 × 105 km3[36,64, and thus it may have released more than 750 Gt of SO2. Volcanism at the periphery of the Tarim LIP could have resulted in the released SO2 flux being >1500 Gt with twice the total eruptive area of the LIP at ~6 × 105 km229,30,31. These eruptions, combined with other large global volcanic events (e.g., the Panjal Traps65, Qiangtang dike66; Fig. 5a) in the Early Permian, may have caused intermittent volcanic winters and the discrete glacial events. Therefore, the C–S radiative forcing revealed in this study provides an explanatory perspective on the evolution of the LPIA.

Linkage between volcanism and the evolution of discrete glaciations during the LPIA

The LPIA is composed of eight discrete glacial–nonglacial intervals (Carboniferous C1–C4; Permian P1–P4), where intervals preserving glacial activity and cold-climate indicators alternated with those lacking such evidence2,22. Notably, these cyclically repeated glaciations are generally correlated with widespread intraplate magmatic events65,66,67,68, including the multipulsed (327–250 Ma) European–northwest African magmatic province (EUNWA; and its initiation as the 300 Ma Skagerrak Centered LIP (SLIP)), and the 260 Ma Emeishan LIP of China (Fig. 5a)69,70,71. However, how volcanism shaped the evolution of discrete glaciations and the LPIA remains unclear sufficiently.

The volcanic C–S radiative forcing mechanism elucidated herein offers a perspective on the evolution of LPIA, and the VSRF may have contributed to initiate and sustain the icehouse conditions. From the perspective of LPIA evolution, during the peak of the volcanic eruption (ca. 300–290 Ma), the frequency of glacial deposit, average ignimbrite flux, and anomalies in tropical mean sea-surface temperature all reached their maxima (Fig. 5e, f)72,73. In contrast, during periods when the frequency and intensity of eruptions declined (e.g., after 290 Ma and before 320 Ma), both sea-surface temperature anomalies and glacial deposit volumes diminished (Fig. 5e, f). All these patterns thus reflect cooling effect potentially linked to the VSRF mechanism, as supported by the C–S radiative forcing model (Fig. 5g)72. Volcanic sulfate forcing maintains effective pCO2 at a level near or below the 560 ppmv glaciation threshold from ca. 360 to 257 Ma, corresponding exactly to the interval recording glacial deposits of the LPIA (Fig. 5g). The interval of maximum frequency of glacial deposits (ca. 298–295 Ma) falls within maximum (negative) radiative forcing (–1.6 Wm–2) induced by sulfate aerosols (Fig. 5g)72. Thus, sulfate aerosols likely counteracted Milankovitch-scale and longer-period pCO₂ maxima, sustaining long-term icehouse conditions from the Late Carboniferous into the Permian72.

Beyond VSRF, volcanism induced sulfate anomalies and elevated carbon burial rates, which also contribute to low pCO2 levels. Firstly, volcanic materials are rich in nutrient (e.g., Si, Fe, and P) that can enhance marine and terrestrial primary productivity by stimulating biological proliferation. This nutrient-driven process efficiently sequesters atmospheric CO₂. Secondly, volcanism elevates sulfate concentrations, increasing the acidity in marine–terrestrial environments and promoting dissolution of reactive Fe74. The Fe fertilization effect further boosts primary production and carbon sequestration. Notably, elevated P contents, P/Ti ratios (Fig. 3e) and TOC contents (Fig. 3b) in the volcanic layers, alongside high Fe contents in the Permian Lucaogou Formation, offer solid evidence for these processes.

Conversely, as volcanism ceased, nutrient supply to the lake declined. This led to a decrease in primary productivity (P/Ti; Fig. 3e) and TOC contents (Fig. 3b), while promoting the accumulation and emission of greenhouse gases (e.g., CO₂) that drive climate warming. Meanwhile, rising temperatures will exacerbate the oxidation of the terrestrial–marine environment and accelerate the CH4 cycling, an ecosystem process highly temperature-dependent75,76. These patterns are also evident in the Lucaogou Formation. From icehouse (Lower Member) to greenhouse period (Upper Member), the Lucaogou Formation shows a decline in average TOC contents (4.6% to 3.6%; Supplementary Data 1) and thermal maturity (vitrinite reflectance, Ro = 0.84 to 0.65)77, alongside shifts toward more oxic depositional environments (Th/U ratios = 1.5 to 2.1; Supplementary Data 1) and intensified methanogenesis25. This indicates a decrease in carbon sequestration and an increase in carbon emissions from icehouse to greenhouse period in the paleo-Junggar Lake, aligning with the sharp decline in Early Permian global terrestrial burial fluxes of organic carbon (peaking at ~300 Ma)60. Collectively, these changes in the carbon cycle intensified climate warming and favored the shift from icehouse to greenhouse conditions.

In conclusion, while this study offers a perspective on the cooling effects caused by LIP eruptions, it should be noted that not all LIPs have the capacity to affect climate change. Instead, silicic supereruptions—particularly those yielding felsic to intermediate magma compositions—are the primary drivers of sulfur injection into the stratosphere72,78. The cooling effects are harder to document than long-term warming effects, as the former are easily overprinted or superimposed by the latter. Exploring the links between volcanism, magmatism, and glacial evolution thus remains a key challenge, requiring more high-precision data and research. Additionally, glacial evolution is likely driven by multiple interacting factors, including tectonic activity and tectonic belt drift34,79, orogeny16, silicate and ophiolite weathering12,15, and terrestrial carbon cycles (e.g., vegetation-induced carbon sequestration)3,14. These factors cannot be ignored, and their integrated effects merit further exploration.

Our findings also have implications for the modern geoengineering initiative to mitigate global warming, one of whose potential measures is to counteract warming by injecting sulfate aerosols into the stratosphere80, a process similar to the VSRF of this study. Based on our results, sulfate aerosols do exert short-term cooling effects but cannot prevent long-term warming and may adversely affect the environment (e.g., acidification, altered redox state; Fig. 3c, f). Thus, in a comprehensive perspective, carbon emission reductions, rather than VSRF, represent a more effective long-term strategy for slowing global warming and sustaining a habitable Earth.

Conclusions

Geochemical proxies from the Lucaogou Formation in the Junggar Basin record Artinskian warming events and second-order climatic fluctuations (especially cooling) across the discrete P1–P2 glacial intervals of the LPIA. These changes are associated with the Tarim-LIP-driven multi-stage volcanism, which substantially perturbed the Hg–C–S cycles. The volcanic sulfur cycle elevated atmospheric sulfate concentrations, cooling the climate and potentially triggering the discrete P2 glaciation. However, the VSRF effect cannot prevent the persistent Artinskian warming, while the volcanic carbon cycle reinforced long-term warming.

Based on the Sakmarian–Artinskian record in the Junggar Basin, the dual radiative forcing effect of the LIPs-driven C–S cycling needs to be fully considered when explaining the short- and long-term climatic changes, and glacial–nonglacial cycles. Incorporating negative VSRF effect into atmospheric pCO2 models improves consistency with the glacial evolution of the LPIA. The VSRF effect likely represents a long-overlooked yet crucial mechanism for initiating and sustaining glacial evolution of the LPIA, and thereby triggers a series of carbon cycle-related events, which likely aid in maintaining low atmospheric pCO2 levels during the LPIA. However, owing to the scarcity of high-precision data and paleoclimate records, further investigation into volcanic carbon and sulfur cycling during the LPIA is required to systematically bridge the knowledge gap. This might be general for the critical events throughout geological history.

Methods

Sampling

The samples were collected from the Lucaogou Formation in well J10025 in the Jimusar Sag, Junggar Basin (Fig. 1b, c). Ninety-three samples from the Lower and Upper Members were used for sedimentological and geochemical investigations. The samples were made into thin-sections for petrographic observations and powdered to 200 mesh for geochemical analysis. The total organic C, and total N and S (TOC, TN, and TS, respectively) contents, major and trace element concentrations, and (in)organic C and O isotopic compositions were determined at the State Key Laboratory of Critical Earth Material Cycling and Mineral Deposits, Nanjing University, Nanjing, China. Sulfur isotope analysis was undertaken at the International Center for Isotope Effects Research, Nanjing University and sedimentary mercury (Hg) concentrations were determined at the State Key Laboratory of Geological Processes and Mineral Resources, China University of Geosciences, Wuhan, China.

Element mapping

Element distributions in smooth rock surfaces and thin-sections were semi-quantitatively determined with an X-ray fluorescence (XRF) spectrometer element scanner. The instrument consists of a 50 μm micro-focusing source, convergent X-ray lens, and Mo target, and was operated at a maximum voltage of 50 kV, maximum current of 1 mA, continuously adjustable power, and scanning interval of 13–20 μm. The relative content of each element from Al to Ba was obtained in counts per second (cps).

Mercury concentrations

We weighed 80–150 mg of dried sample powder (80–120 mg for mudstone or shale and 120–150 mg for limestone) into a Ni capsule for analysis with a LECO® AMA254 Hg analyzer (the Ni capsule was first oven-dried at a high temperature to minimize the Hg blank). To ensure the data were accurate, a standard and duplicate sample were analyzed after every 13 samples. The coal standard 502–685 yielded a Hg content of 41 ± 8 ppb, with a relative standard deviation (RSD) of less than ±5%.

Total organic carbon, total nitrogen, and total sulfur contents

The TOC, TN, and TS contents were determined with a Vario Macro CHNS elemental analyzer (ECS 4024 CHNSO; Costech). Firstly, 10% HCl was reacted with ~2 g of sample powder for 48 h to completely remove carbonates, and then the solution was centrifuged at 3000 rpm for 10 min and the solid residue was washed with deionized water. This procedure was repeated 7–8 times until the solution pH reached 7. Finally, the residual solid was dried at 60 °C, and 30 mg of the sample was wrapped in Sn foil for analysis. The accuracy of the data is better than ±0.1%.

Major- and trace- elements analyses

The powdered samples were dried in an oven to remove volatiles, and then ~1 g of powdered sample was mixed with ~11 g of lithium tetraborate, melted at 1050 °C, and cooled to form glass disks for analysis. The samples were analyzed for major elements using a Thermo Scientific ARL 9900 XRF spectrometer. The voltage and current were 40 kV and 70 mA, respectively, and the scan time for each element was 20 s. Based on the measured values for the standards (BHVO-2 and BCR-2), the relative error was ±1% when the element abundance was >1.0 wt.% and ±10% when the element abundance was <1.0 wt.%. An Element XR inductively coupled plasma–mass spectrometer (ICP–MS) was used to determine the trace element contents. An aliquot of each sample (50.0 ± 0.2 mg) was dissolved in a mixture of HF and HNO3, and digested in high-pressure Teflon bombs. Rhodium was used as an internal standard and the data accuracy was <5%.

Organic carbon isotopes

Organic C isotopes (δ13Corg) were determined with a Finnigan MAT 253 mass spectrometer. The powdered sample was acidified and air-dried, mixed with CuO powder (mass ratio of 1:8) in a quartz tube that was vacuum-sealed, and heated to 850 °C. After cooling in a cryogenic trap to separate H2O, CO2 was directly introduced into the inlet system of the mass spectrometer. The standard was GBW04407 (δ13CV-PDB = –22.43‰ ± 0.07‰), and the results are reported relative to Vienna Peedee Belemnite (V-PDB) with an accuracy of ±0.1‰.

Inorganic carbon and oxygen isotopes

The C and O isotopic compositions of whole-rock samples were determined with a Thermo Finnigan Delta V Plus continuous flow isotope ratio mass spectrometer (IRMS). An aliquot of 80–120 μg of sample powder was dissolved in orthophosphoric acid at 70 °C for >12 h and connected to the continuous flow sample preparation device (Gas Bench II) of the IRMS. V-PDB was used as the standard. All isotope ratios are reported in per mil (‰), and the C and O isotope data accuracy is better than ±0.1‰ and ±0.5‰, respectively.

Sulfur isotopes

For the whole-rock sulfur isotopes analyses. Firstly, 10% HCl was reacted with dried sample powders (<200 mesh) at 25 °C to completely remove carbonates, and then the solution was centrifuged at 3000 rpm for 10 min and the solid residue was washed with > 18 MΩ·cm⁻¹ deionized water. This procedure was repeated 7–8 times until the solution pH reached ~7. Finally, the residual solid was dried at 60 °C for sulfur isotope analysis. Sulfur isotopes were analyzed with a MAT253 Plus mass spectrometer and Flash EA elemental analyzer coupled with a Conflo IV multi-purpose interface instrument. The pre-treated samples were combusted at high temperatures in an O2-permeable environment, reduced and converted to SO2 in high-purity He carrier gas, and passed through a redox reaction tube to fully oxidize all the gases. Finally, the SO2 gas was separated using a chromatographic column and analyzed by mass spectrometry. The S isotope results are expressed in standard delta notation as per mil (‰) relative to the Vienna Canyon Diablo Troilite (V-CDT) standard, with an analytical accuracy of ±0.2‰.

$${{{\rm{\delta }}}}^{34}{{\rm{S}}}({\textperthousand })=(({}^{34}{{\rm{S}}}/{}^{32}{{\rm{S}}}){{\rm{sample}}}\,/\,({}^{34}{{\rm{S}}}/{}^{32}{{\rm{S}}}){{\rm{standard}}}-1)\times 1000$$
(1)

Chemical index of alteration and land surface temperatures

The chemical index of alteration (CIA) is widely used to assess deep-time changes in climate and chemical weathering, and is calculated as follows81:

$${{\rm{CIA}}}=[{{{\rm{Al}}}}_{2}{{{\rm{O}}}}_{3}/({{{\rm{Al}}}}_{2}{{{\rm{O}}}}_{3}+{{\rm{CaO}}}^{*} +{{{\rm{Na}}}}_{2}{{\rm{O}}}+{{{\rm{K}}}}_{2}{{\rm{O}}})]\times 100$$
(2)

where CaO* is the CaO bound in silicates, excluding the CaO in carbonates or apatite82, and CaOexcess = CaO – P2O5 × 10/3. If CaOexcess < Na2O, then CaO* = CaOexcess; otherwise, CaO* = Na2O83.

The chemical index of weathering (CIW) and modified CIA (CIX) can correct for the effects of K and Ca on the CIA parameters. The CIW is used to remove K2O from the CIA to avoid the effect of K metasomatism84, and the CIX is used to remove CaO* from CIA to avoid the effects of carbonates85. The formulae are as follows:

$${{\rm{CIW}}}=[{{{\rm{Al}}}}_{2}{{{\rm{O}}}}_{3}/({{{\rm{Al}}}}_{2}{{{\rm{O}}}}_{3}+{{\rm{CaO}}}^{*} +{{{\rm{Na}}}}_{2}{{\rm{O}}})]\times 100$$
(3)
$${{\rm{CIX}}}={{{\rm{Al}}}}_{2}{{{\rm{O}}}}_{3}/({{{\rm{Al}}}}_{2}{{{\rm{O}}}}_{3}+{{{\rm{Na}}}}_{2}{{\rm{O}}}+{{{\rm{K}}}}_{2}{{\rm{O}}})\times 100$$
(4)

Furthermore, the CIA values of modern suspended particulate matter in large global rivers are sensitive to land surface temperatures (LSTs), estuary latitude, and soil depth in the watershed86. As such, CIA values can be used to estimate LSTs using the following quantitative relationship33:

$${{\rm{LS}}}{{\rm{T}}}\left({}^\circ {{\rm{C}}}\right)=0.56\times {{\rm{CIA}}} - 25.7({{\rm{R}}}^{2}=0.50)$$
(5)

which is applicable at temperatures of 3–25 °C, corresponding to a CIA range of approximately 50–90. This LSTs estimation has been successfully applied to quantify paleotemperatures of Permian and Triassic terrestrial successions20,33,87.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.