Main

The emplacement of large igneous provinces (LIPs), the largest known volcanic events in Earth’s history, is commonly associated with severe environmental, climatic and ecosystem disturbance, including mass extinctions and oceanic anoxic events1. LIP-associated carbon emissions (primarily magmatic and thermogenic CO2) are hypothesized to have driven climate change, ocean acidification and anoxia–euxinia1. Although the conceptual framework for the Earth system effects of LIP carbon emissions is well established (Fig. 1), understanding these processes quantitatively remains a major challenge. The quantity and timing of LIP-associated gas emissions are usually estimated from geochronology on LIP materials and geochemical indicators of volatile content in the lavas (and affected country rocks)2,3,4. Several factors limit this approach: (1) current maximum level of precision for radioisotopic dating techniques (for example, 10s–100s of kiloyears for Jurassic igneous material2,5,6); (2) difficulty translating radioisotopic dates into volumetric emplacement rates through time, given the spatiotemporal complexity of LIPs5,7; (3) the unreliability of geochemical indicators of degassing in these ancient rocks, especially for carbon8. Thus, direct sampling of LIP materials alone cannot currently provide continuous sub-10 kyr emission data as required to link LIP activity quantitatively and mechanistically to climatic and carbon-cycle perturbations in detail.

Fig. 1: Maps and illustration of LIP gas emissions processes and environmental effects.
figure 1

a, Palaeogeographic reconstruction23 showing the Jurassic location of the Mochras borehole and locations of CAMP and KFLIP based on modern outcrop extent (approximate locations shown via yellow and white boxes/shading, respectively; primary outcrop locations for the KFLIP include present-day Africa (Karoo): South Africa, Lesotho, Namibia, Botswana, Mozambique, Zimbabwe and Zambia; Antarctica: Dronning Maud Land (Karoo) and Dry Valleys (Ferrar), and Tasmania (Ferrar))5,6,43. b, Inset map showing the present-day location of the Mochras borehole52. c, Illustration of direct and indirect feedback effects of LIP-associated CO2 emissions1 and simplified LIP Hg emissions pathways. Typical Hg/CO2 ratios are shown for each volatile source (eruptive, intrusive magmatic and intrusive thermogenic), and an approximate net LIP-associated Hg/CO2 value (Methods, Supplementary Information 4 and Supplementary Table 1; thermogenic sources are based on theoretical calculations and using natural (in situ) coal fires as an analogue). Basemap in a adapted with permission from ref. 23, Annual Reviews.

Full size image

As an alternative, carbon-cycle models have been used to estimate the total carbon emissions required to reproduce palaeoenvironmental records (for example, δ13C, temperature, pH and \(p_{{\mathrm{CO}}_2}\))9,10. This approach generally assumes that LIP-associated carbon emissions were entirely responsible for the observed perturbations and that carbon-cycle models adequately capture the Earth system response during the geological past. In such studies, LIP emission histories and feedback systems are not independently constrained, challenging our ability to test and validate our understanding of Earth system responses to large volcanic carbon emissions using palaeoclimatic reconstructions alone. For example, these Earth system responses include a number of both positive and negative feedbacks, such as amplification by perturbing exogenic carbon reservoirs (clathrates, terrestrial carbon reservoirs)10,11,12 and dampening via increased silicate weathering and organic-carbon burial13,14, which are incorporated into carbon-cycle models using differing parameterizations and mathematical simplifications. Independent carbon emissions estimates are required to assess whether carbon-cycle models sufficiently capture the full spectrum of Earth system sensitivity. Sedimentary mercury (Hg) concentration has been used as a qualitative indicator of large-scale volcanic activity during the geological past15 as volcanism is a primary natural source of Hg to the environment16. In this work, we use a chronologically well-constrained and high-resolution record to develop a sedimentary Hg-based quantitative proxy for LIP-associated (magmatic and thermogenic) gas emissions that aims to address this critical challenge.

The Early Jurassic period (201–175 million years ago (Ma)) witnessed a complex series of changes in climate, ocean chemistry, sea level and ecological diversity and included the emplacement of two LIPs: the Karoo–Ferrar LIP (KFLIP) and late-stage (post-Triassic) activity of the Central Atlantic magmatic province (CAMP6,17). The Toarcian Oceanic Anoxic Event (T-OAE), coeval within dating uncertainty with the KFLIP main phase, was among the most severe Mesozoic palaeoenvironmental events2,18. The T-OAE was characterized by global climate warming, regional oceanic anoxia–euxinia, ocean acidification, mass extinction and a major carbon-cycle perturbation18. In addition to the T-OAE, the Early Jurassic also contains a comparatively long-term (~5 Myr) negative carbon-isotope excursion (n-CIE) centred around the Sinemurian–Pliensbachian (S–P) stage boundary, hypothesized as related to late-stage CAMP volcanism19, and complex changes in δ13C in the latest Pliensbachian20.

While the overall temporal relationship between the main phase of KFLIP magmatism and the T-OAE is reasonably well established, the quantity of coeval LIP-associated carbon emissions is poorly constrained9,10. In addition, whether LIP eruptions coincided with other Early Jurassic CIEs remains far less certain, and whether a more general mechanistic relationship exists between environmental events during the Early Jurassic and LIP volcanism remains to be resolved19,20,21.

In this work, we present a compilation of new and literature sedimentary Hg concentration data from the Llanbedr (Mochras Farm) borehole, Wales, UK, creating an unprecedentedly long and high-resolution record spanning the Late Sinemurian age (~196 Ma) through the Late Toarcian age (176 Ma). This allows us to explore whether large-scale volcanism temporally coincides with the Early Jurassic CIEs. Further, we combine the data with knowledge of global Hg cycling and Hg/CO2 relationships for present-day volcanism to estimate excess LIP-associated carbon (CO2) emissions spanning the entire 20 Myr record. Focusing on the T-OAE, we use these emissions estimates to test whether current model scenarios adequately represent the environmental response to LIP-associated (magmatic and thermogenic) carbon emissions and feedback processes.

Geological setting

The Llanbedr (Mochras Farm) borehole22 (hereafter termed Mochras) in the Cardigan Bay Basin, Wales, offers one of the most stratigraphically expanded records of the hemipelagic Lower Jurassic. The Cardigan Bay Basin was within the Laurasian seaway, which connected the Arctic Sea and Tethys Ocean23 (Fig. 1). The timing of key events within the Mochras record is constrained via ammonite and foraminiferal biostratigraphy, astrochronology and δ13C chemostratigraphy12,19,24,25,26. There are no major hiatuses within the Lower Jurassic section25, and carbon-, oxygen- and osmium-isotope trends at Mochras are similar to those from Tethyan localities, suggesting regular seawater exchange with the global ocean12,27,28.

Lower Jurassic (Sinemurian–Toarcian) sedimentary Hg record

Volcanic systems emit mercury gas during eruptive and non-eruptive degassing, including intrusive magmatic activity29. Indeed, large-scale and prolonged subaerial volcanism30,31 can be traced by sedimentary Hg concentrations in the geological record: increases in Hg have been found to occur coeval with, for example, the main-phase eruptive intervals of the KFLIP30,32,33,34 and CAMP17.

Our new Hg data span the last ~2 Myr of the Sinemurian and the entire Pliensbachian, substantially extending the existing Mochras Hg data for the uppermost Pliensbachian and Toarcian27,30,34. We investigate the changes in Hg deposition through time as a reflection of the variable availability of environmental Hg and, hence, notable changes in large-scale global volcanic activity (probably dominantly LIP related).

As host phase complexity31 can influence the interpretation of sedimentary Hg records, we quantitatively account for related processes before using the Mochras record to estimate LIP-associated emissions. Although a simple ratio (Hg/total organic carbon (TOC)) is typically used for this purpose, this does not always fully correct for host-phase biases, nor does it quantify the additional environmental Hg present over background levels during periods of sedimentary Hg enhancement31. Instead, we developed a quantitative framework based on an empirical assessment of co-variation between Hg concentration and that of host phases (for example, TOC and S) and dilution effects due to sedimentary CaCO3—each statistically significant in the Mochras dataset (Supplementary Information 1.2). The framework relies on the observation that Hg behaves consistently throughout our dataset. Residual Hg (HgR) is found here by first calculating Hg and TOC concentrations on a CaCO3-free basis for each sample and then adopting a best-fit curve between them for the whole dataset. HgR is the difference between each sample’s CaCO3-free Hg concentration and the obtained best-fit curve (Methods and Supplementary Information 1.2). We expect this adaptable framework to be widely applicable as it can accommodate a range of Hg-host phase relationships, including, but not limited to, a linear relationship.

HgR hence represents the excess Hg loading beyond what is expected given the geochemistry of each sample. We use scaling factors from Hg-cycle box-model results35,36 to determine the increased atmospheric Hg flux associated with elevated HgR (Fig. 2 and Supplementary Information 4). The Hg-cycle model explicitly considers the critical effects of biogeochemical recycling of Hg within and between terrestrial and marine reservoirs before deposition in sediments. This is important because terrestrial sediments and organic material are major Hg sources to modern and Jurassic near-shore marine environments35,37. We compare the Mochras core dataset with the coastal marine sediment box as this locality received substantial amounts of terrigenous material26.

Fig. 2: Mercury-derived datasets, ammonite biostratigraphy and organic-carbon isotopic composition for the Mochras core versus age.
figure 2

Hg datasets (Supplementary Information 7)12,24,46,47: left plot shows residual Hg (HgR, ng g–1, ppb) and Hg enrichment (unitless) (Supplementary Information 1 and 3) with the blue shaded box indicating background (95%) Hg variability; middle plot shows Hg/TOC (in ppb/wt%) for comparison; right plot shows Hg emissions (Mg kyr–1) calculated from residual Hg using scaling factors from a Hg-cycle box model (Supplementary Information 4)35,36. Hg enrichment is HgR divided by the corresponding value on the calcium carbonate-free Hg versus calcium carbonate-free TOC best-fit curve. The slope of this curve is very shallow for our data (Supplementary Information 1.3 and Supplementary Fig. 6) and hence is nearly equivalent to dividing by a constant, enabling a secondary axis for approximate enrichment factor on the residual Hg plot. The S–P and T-OAE carbon-isotope excursions are labelled and indicated via green and red shading, respectively. δ13Corg, organic-carbon isotopic composition; VPDB, Vienna PeeDee Belemnite.

Full size image

We expect the primary control on HgR to be subaerial volcanic-associated gas emissions as submarine Hg emissions have limited geographical reach38. Although processes other than volcanism (including wildfires, enhanced weathering and redox changes) may result in excess Hg deposition, they are unlikely to have substantively affected our dataset (Supplementary Information 2). We therefore consider our data a representative estimate of subaerial Hg emissions in this instance, probably dominated by the volcanic-associated flux (Fig. 2).

Relationship between LIP activity and carbon-cycle events

We assess the relationship between LIP volcanism and carbon-cycle changes by comparing the HgR record and carbon-isotope stratigraphy in the Mochras core (Fig. 2). We find that the Sinemurian–Pliensbachian (S–P) transition, the upper margaritatus Zone, the Pliensbachian/Toarcian boundary and the T-OAE n-CIE each coincide with increases in residual sedimentary Hg levels above background (95%) variability (Fig. 2), albeit with different magnitudes. Thus, an increase in global volcanic-related (LIP) Hg emissions appears to have temporally coincided with all four n-CIEs.

To test whether a volcanic CIE driver is plausible, we use HgR to estimate the quantity of coincident LIP-associated carbon emissions. Measurements at present-day volcanoes, experiments and estimates of thermogenic emissions suggest that Hg and carbon emissions are closely coupled under most extrusive and intrusive volcanic scenarios29. Thus, using modern-analogue Hg/CO2 gas ratios from volcanism and sediment combustion, we estimate the carbon (CO2) emissions that probably corresponded with the excess Hg fluxes (Figs. 1 and 3, Methods, Supplementary Information 4 and Supplementary Table 1). To estimate the consequent increase in exogenic carbon reservoirs, we use a 300 kyr rolling window sum to approximate C accumulation before removal via weathering/burial13 and to calculate the corresponding resultant value of atmospheric \(p_{{\mathrm{CO}}_2}\) (ref. 39). We compare our carbon outgassing estimates with emission scenarios used in carbon-cycle model studies of the T-OAE interval (Fig. 4).

Fig. 3: Karoo and Ferrar geochronology data, with organic-carbon-isotope record, mercury-derived carbon accumulation estimates and osmium-isotope data (all from the Mochras core).
figure 3

The largest estimated carbon accumulation coincides with the T-OAE n-CIE12,24 and an increase in weathering demonstrated by a shift towards radiogenic osmium-isotope values27. Compilations of Karoo and Ferrar radioisotopic dates6 of extrusive lavas, dykes and sills are plotted as coloured bars that are kernel density diagrams (stronger colouration represents higher probability density). The uncertainty on these dates was incorporated by bootstrap resampling each date 100 times and plotting the distribution of the total results. Estimated carbon accumulation (Pg) was calculated by summing carbon emissions over a 300 kyr moving window to account for carbon accumulation in the ocean–atmosphere system, the decay timescale of which is determined by negative feedbacks13. Carbon emissions are a scaling of Hg emissions (Fig. 2), with a grey region spanning Hg/CO2 ratios between 10−6 as a minimum and 10−7 as a most likely net composition of LIP-associated emissions (Fig. 1). To contextualize the carbon accumulation, an equivalent resulting \(p_{{\mathrm{CO}}_2}\) estimate, assuming an initial \(p_{{\mathrm{CO}}_2}\) of 500–1,000 ppm, is included on the axis label. This \(p_{{\mathrm{CO}}_2}\) increase is calculated using equilibrium ocean carbonate chemistry partitioning of CO2 between ocean and atmosphere as a function of \(p_{{\mathrm{CO}}_2}\) (three-box model)39 (Methods and Supplementary Information 4). \(p_{{\mathrm{CO}}_2}\) increase is used as a metric of carbon-cycle influence, rather than the amplitude of potential (n-)CIE, as these carbon emissions estimates are agnostic to the proportion from magmatic versus thermogenic sources, and hence the net δ13C value12,24 of the emissions is undetermined.

Full size image
Fig. 4: Mercury-based carbon emissions estimates across the T-OAE n-CIE compared with literature estimates for the Karoo–Ferrar LIP and other LIPs.
figure 4

Literature carbon-cycle model-based estimates of total KFLIP carbon emissions across the T-OAE n-CIE based on GEOCLIM9 (purple point) and COPSE10 (red point) model scenarios matching δ13C, \(p_{{\mathrm{CO}}_2}\) and temperature proxy data with carbon emissions estimates for several other LIPs (green bars on the right of the plot represent the total emissions quantity; dotted lines are average emission rates in units of Pg yr–1). KFLIP model scenarios incorporate carbon fluxes with δ13C values indicative of a combination of CO2 and CH4 (Supplementary Information 4.4). Total carbon emissions and durations are compared to best represent an integrated view of the T-OAE, less dependent on sampling resolution and age models. Hg-based carbon estimates: the blue bar represents the range of carbon emissions estimated for Hg/CO2 ratios between 10−6 and 10−7 during the T-OAE n-CIE (wherein CO2 emissions and n-CIE durations are equivalent). A Hg/CO2 ratio of ~10−7 is hypothesized to be the most likely long-term net value, consistent with a combination of eruptive and diffusive (magmatic and thermogenic) degassing (Fig. 3). A Hg/CO2 ratio of 5 × 10−8 is shown to illustrate the maximum possible CO2 emissions that may occur in specific diffusive degassing scenarios53, although it is unlikely to represent a long-term net composition. Other LIP carbon emissions from a literature compilation3,45,54: the green bars on the right of the plot represent LIP-proxy-based estimates of the quantity of total carbon emissions for several other LIPs (based on a combination of CO2 content estimates, from melt inclusions and geochemical ratios, and thermogenic emissions estimates, and total LIP volumes), and the green dotted lines represent the average rate of carbon emission for those LIPs (based on the total quantity and the estimated emplacement duration). The Siberian Traps and CAMP represent LIPs with substantial associated thermogenic carbon emissions, and the Deccan Traps and Kerguelen Plateau represent LIPs with minimal thermogenic carbon emissions.

Full size image

Volcanism at the Sinemurian–Pliensbachian transition?

The 5 Myr duration, 3–4‰ (ref. 12) S–P boundary n-CIE in Mochras δ13Corg has previously been attributed to volcanic-associated CO2 emissions. However, the evidence for substantial LIP volcanism during this interval is sparse6 (Supplementary Information 5), and there is no conclusive geochronological or volcanological evidence of a large, sustained episode of volcanic activity/degassing across the S–P boundary as required to explain the long-duration n-CIE40.

The upper Sinemurian through lower Pliensbachian (~1,300–1,000 m core depth, ~195–188 Ma) generally shows background residual Hg concentrations. We propose that the few HgR outliers close to the S–P boundary (Fig. 2) correspond to a modest increase in volcanic-associated CO2 emissions between 194 and 190 Ma, estimated as cumulative ~2,420 PgC emitted over this 4 Myr period (Hg/CO2 of 10−7). Due to the extended duration and coeval carbon removal via negative feedbacks (for example, silicate weathering), a modest accumulation of 625 PgC is estimated for the ocean–atmosphere carbon pool (Fig. 3), much less than required to drive a 3–4‰ n-CIE40.

As submarine LIP volcanism generally leaves only regional sedimentary Hg signals38, we cannot exclude that submarine volcanic activity and carbon emissions played a role while leaving only the few Hg peaks recorded in the Mochras core. Although volcanic forcing cannot be excluded, our new Hg data do not support subaerial volcanism as the primary driver, and the ultimate cause of the S–P n-CIE remains enigmatic.

Late Pliensbachian volcanism

In contrast to the clear n-CIEs associated with the T-OAE and S–P, carbon-isotope records for the late Pliensbachian show spatially heterogeneous trends, which have previously been attributed to erosive sea-level fall in the latest Pliensbachian20. The Mochras record, considered among the more complete successions for this interval, shows an abrupt negative CIE at the margaritatus/spinatum zonal boundary, superimposed onto a positive trend, and a subsequent smaller n-CIE at the Pliensbachian/Toarcian boundary12,20. Previous studies have variably linked the n-CIEs to a combination of KFLIP volcanism and orbitally forced climate change20,21,30.

The upper Pliensbachian and Toarcian (1,000–700 m core depth, ~188–180 Ma) interval has several HgR peaks (100–200 ppb) starting in the margaritatus Zone and persisting through the lower Toarcian. This is consistent with existing Hg records from Switzerland, Dorset (UK) and Chile, which have Hg and Hg/TOC peaks within the margaritatus Zone33,41. Overall, these results strongly suggest that enhanced (LIP) volcanism occurred around this time. Both n-CIEs are coeval with HgR increases; however, other HgR peaks are not associated with notable changes in δ13Corg. This complexity may be explained via variability in the δ13C of LIP-associated carbon emissions (for example, magmatic −5‰ versus thermogenic < −20‰) or variations in the global C cycle in response to volcanic-related (or other) forcings.

The HgR peaks in the margaritatus Zone and the Pliensbachian/Toarcian boundary each correspond to a predicted exogenic carbon increase of up to 3,000 Pg (Fig. 3). The calculated increase in \(p_{{\mathrm{CO}}_2}\) is ~300–375 ppm (initial \(p_{{\mathrm{CO}}_2}\) of 500–1,000 ppm), consistent with a boron-isotope-based \(p_{{\mathrm{CO}}_2}\) record from the uppermost margaritatus Zone42, and probably resulted in climate warming and environmental change. Although the margaritatus Zone and the Pliensbachian/Toarcian boundary precede the main phase of KFLIP, they are within the known temporal range of early KFLIP intrusive activity5 (and, for note, the first stage of Chon Aike silicic LIP-style volcanism in Antarctica from 189 to 183 Ma (ref. 43); there are, however, substantial unknowns regarding Hg and carbon emissions from silicic LIP-style episodes, and thus we do not consider this further here). The magnitude of the HgR peaks and corresponding carbon emissions suggest that volcanic activity may be implicated in the late Pliensbachian environmental change.

T-OAE and C-cycle implications

The largest peaks in HgR, and hence largest potential CO2 emissions, of the entire 20 Myr record occur within the interval of the T-OAE n-CIE (Figs. 2 and 3), consistent with previous studies that documented elevated Hg concentrations at the Pliensbachian/Toarcian boundary and across the T-OAE interval30,32,33,37. The relative magnitude of these peaks compared with the rest of the HgR record suggests that KFLIP-associated gas emissions were greatest during this part of the Early Jurassic30,32.

We estimate that the HgR peaks during the T-OAE n-CIE corresponded with considerable CO2 emissions (a total of ~17,000 PgC emitted during the main phase of KFLIP, 184–181 Ma) and a maximum total accumulation of ~6,200 PgC, doubling to tripling atmospheric \(p_{{\mathrm{CO}}_2}\)—consistent with stomatal and boron-isotope-based estimates42,44. The timing of maximum Hg-based carbon emissions is consistent with geochronology from the KFLIP (Fig. 3), which suggests that the majority of KFLIP volcanism, both intrusive and extrusive activity, coincided with the T-OAE at 183–182 Ma (refs. 2,6).

We compare estimated carbon emitted during the T-OAE n-CIE (KFLIP main phase) between ~183 and 182 Ma with independent total carbon and average emission rate estimates from LIP-volcanology-based proxies (for example, melt inclusions, igneous geochemistry and modelling of magmatic processes) for several other LIPs45 (Fig. 4). Although the time history of gas emissions found using these proxies has inherent limitations due to uncertainty in volcanic stratigraphy, geochronology and gas content, the total summed carbon emissions reflect what is expected on the basis of our knowledge of LIP magmatic processes. Our Hg-based carbon quantity and rate estimates are somewhat higher than the ranges obtained for the Deccan Traps/Kerguelen Plateau but are smaller than those associated with the Siberian Traps/CAMP (Fig. 4).

Notably, we find that existing modelled LIP emission scenarios based on matching carbon-cycle model output with δ13C, temperature and/or \(p_{{\mathrm{CO}}_2}\) proxy data require higher average carbon input rates and total amounts of carbon across the T-OAE than supported by our emissions estimates and existing estimates based on LIP-volcanology-based proxies (Fig. 4). We compare average emissions rates as they are less sensitive to sampling resolution and age model (Supplementary Information 4.4). We estimate that a total of ~12,000 PgC was released (Hg/CO2 of 10−7) during the entire T-OAE n-CIE interval (~1 Myr), compared with estimates of 20,500 Pg over 262 kyr and 81,000 Pg over 150 kyr using published GEOCLIM9 and COPSE10 model results, respectively. Our most likely scenario, with Hg/CO2 by mass of 10−7 (a combination of eruptive and intrusive degassing, both magmatic and thermogenic; Fig. 1 and Supplementary Information 4.2), implies average excess carbon input rates across the T-OAE of 0.012 Pg yr–1 (0.015 Pg yr–1 during the falling limb of the n-CIE; Supplementary Information 4.4), with an upper estimate of 0.023 Pg yr–1 (Hg/CO2 of 5 × 10−8), compared with 0.08 and 0.54 Pg yr–1 for the GEOCLIM and COPSE scenarios, respectively. In addition, the duration of the T-OAE n-CIE using our updated Mochras age model is longer than in the GEOCLIM and COPSE model scenarios (~1 Myr in our record versus ~ 500 kyr; Fig. 4 and Supplementary Information 3). This difference has been noted in other records due to the choice of astronomical cycles46, but recent U–Pb dates (which we incorporate) support the longer duration47 (Supplementary Information 3). If a longer-duration n-CIE were modelled, more carbon input would be required to match the observed CIE amplitude. Alternatively, a shorter-duration T-OAE would require smaller Hg emissions to produce the observed Hg enrichment (given less dilution of the Hg emissions over time; Supplementary Information 4.1), similarly exacerbating the mismatch in carbon emission scenarios.

The discrepancies between our Hg-based estimates and the two carbon-cycle model scenarios suggest that our understanding of the early Jurassic carbon cycle is incomplete. Recent studies show that slow, state-dependent, positive-feedback processes leading to high Earth system sensitivity (for example, clay mineral formation48, oxidative weathering11, ocean circulation49 and biological productivity50) may be important factors. Both the complex feedback response to environmental change and the Early Jurassic background state (for example, the size of the initial exogenic carbon reservoir and weathering/erosional fluxes49,51) may not be adequately included in model studies. In addition, this discrepancy raises the question of whether the role of other co-emitted gases (for example, SO2 and CH4), which are not included in our Hg-based estimates or most carbon-cycle/climate studies, should be specifically considered in the climate response to LIPs.

Our results demonstrate the utility of combining Hg records with Hg box models to provide quantitative LIP-associated Hg and carbon emissions estimates. This approach provides important constraints on LIP-associated gas emissions at a resolution relevant for palaeoclimate studies and thus can be used to diagnose Earth system processes insufficiently captured within models or the need to investigate alternative carbon sources. The framework we describe here can be applied to other Phanerozoic LIPs, providing that long-term, high-resolution and chronologically and geochemically well-constrained Hg records are available. We envisage that this methodology will improve our understanding of how LIP emplacement and other large carbon emission events affect the Earth system.

Methods

Mercury concentration

Mercury concentration was determined on powdered material using a Lumex RA-915M portable mercury analyser with Pyro 915+ attachment at the University of Oxford. The calibration curve was determined using aliquots of National Institute of Standards and Technology Standard Reference Material 2587 (Trace Elements in Soil Containing Lead from Paint, Hg = 290 ± 9 ng g–1) of varying masses. Standards reproduced their accepted value within 5% on average. Sample duplicates (n = 21) were also analysed, and values were reproduced with an average percentage deviation of 11% and an average standard deviation of 1 ppb. We combine our new Hg dataset55 with previously published Hg data from the Mochras core from refs. 27,30,32,34,38. The TOC and total inorganic carbon (TIC) data are reproduced here from refs. 12,24. TOC and TIC were determined using a Rock-Eval 6 (ref. 56). The CaCO3 concentrations were then calculated from TIC (TIC × 8.333), assuming TIC resides in CaCO3. Details of these analytical methods are available within the respective studies.

Data analysis

Our goal is to investigate the evolution of Hg deposition through time as a reflection of the changing availability of environmental mercury. Quantitative assessment of temporal variability in Hg loading of the environment is complicated by co-variation between Hg and other geochemical parameters such as TOC and sulfur (S) that can act as host phases for mercury in sediments34. Although this co-variation is typically addressed via linear normalization by TOC (Hg is commonly associated with organic matter in the water column and in sediments deposited throughout geological history31), this approach does not always fully correct for host-phase biases and does not quantify how much additional Hg is present in excess of background levels during periods of Hg enhancement31 (Supplementary Information 1). We thus used a conceptual understanding of Hg sedimentary geochemistry to create a quantitative framework to calculate the amount by which the concentration of Hg in each sample deviates from the expected amount given the other chemical attributes of the sample (termed as residual Hg, HgR). This framework is based on an empirical assessment of the co-variation between Hg and other geochemical parameters (for example, concentration of TOC and S), including dilution effects due to sedimentary CaCO3—all of which are statistically significant in the Mochras dataset. The framework relies on the assumption that Hg behaves with consistent patterns throughout the dataset (which we demonstrate is true for our record; Supplementary Information 1.2) and thus requires a substantial dataset with a large proportion of ‘background’ periods without Hg-cycle perturbations.

The HgR is expressed in units of mass ratio (ng/g, ppb by mass). To calculate HgR, we first calculate calcium carbonate-free Hg and TOC (Hgcf and TOCcf) because Hg is most strongly correlated (inversely) with CaCO3 in the Mochras dataset. Hgcf does not co-vary statistically significantly with Scf, but it does with TOCcf. Hence, we use robust nonlinear regression57 to find a best-fit curve between Hgcf and TOCcf and calculate the residual (the distance from) to this curve for each sample (Supplementary Information 1). Robust regression (we use R package robustbase; see example code in Supplementary Information 6) is an optimal tool for fitting this curve since it weighs outliers less in the regression—outliers in Hgcf versus TOCcf most likely represent the presence of an additional Hg source that is not correlated with TOC. Robust nonlinear regression requires an initial choice of function for which the coefficients are found during regression. Here we find a hyperbolic tangent function to be the most useful as, with different coefficients, it can fit the range of curves that we frequently see in Hg versus TOC data, including quasi-linear co-variation. However, this choice is predicated on examining the behaviour present in each dataset as other functions may be appropriate in some cases (Supplementary Information 1).

There is no strong relationship between HgR and S, suggesting that, in this case, S is not a control on Hg concentration independent of co-variation with TOC and hence does not need to be explicitly considered in the calculation of HgR. If a relationship between HgR (with respect to TOCcf) and Scf had been present, a second residual would have been calculated between these parameters, which would then accommodate variations in Hg driven by both TOC and S (as well as dilution by CaCO3). However, HgR does not statistically significantly co-vary with TOC, CaCO3 or S (or the equivalent CaCO3-free parameters), lending confidence that we have successfully removed the effects of dilution and host-phase abundance. We hypothesize that the residual Hg in this dataset represents an increase in the environmental abundance of Hg that is ultimately driven by activity associated with large-scale magmatism, considering the timescales involved (~103 yr per sample). Other non-magmatic perturbations to the local or global environmental Hg load, such as recent Hg pollution from various industrial sources (unlikely to be relevant in the Mochras core record), might also be detected via this approach.

C emissions estimates and change in atmospheric \({\boldsymbol{p}}_{{\mathbf{CO}}_{\mathbf{2}}}\)

We estimate the CO2 emissions (then converted to equivalent mass of carbon) that may have corresponded with the peaks in the sedimentary Hg record by first estimating Hg emissions using previously published Hg box-model results35,36 by scaling Hg enrichment to total Hg emission mass, assuming each point represents 1,000 years (consistent with a median sedimentation rate of ~3.7 cm kyr–1 from our age model, sample size of a few centimetres, and Hg distribution over a few centimetres of sediment at time of deposition due to fluid infiltration). Then we use Hg/CO2 ratios from modern analogues to LIP degassing processes to estimate the CO2 emissions that probably corresponded with the Hg emissions (Fig. 1). We present a range of Hg/CO2 ratios from 10−6 to 5 × 10−8 and hypothesize a middle-of-the-range ‘net’ Hg/CO2 ratio of 10−7 by mass (Figs. 1 and 3), spanning the majority of volcanic degassing measurements (eruptive and non-eruptive degassing) and estimates of thermogenic gases (Supplementary Information 4 and Supplementary Table 1)53,58,59,60. We convert CO2 emissions to equivalent C by mass fraction (C/CO2 = 12/44 by mass).

Mercury has a shorter lifetime in the atmosphere and surface reservoirs (<10 kyr) than CO2, which accumulates for 100s of kiloyears13,35. We sum C emissions over 300 kyr rolling windows to approximate the timescale over which CO2 accumulates until removed from the ocean–atmosphere system via silicate weathering (weathering releases cations, in particular Ca2+, into the ocean, leading to the precipitation of carbonate minerals)13 and/or burial as organic carbon14. For comparison, see C emissions estimates with no rolling window in Supplementary Information 4.

To aid in comparison with the \(p_{{\mathrm{CO}}_2}\) record and model results, we calculate the approximate final \(p_{{\mathrm{CO}}_2}\) value following these emissions, assuming a constant initial \(p_{{\mathrm{CO}}_2}\) of 500–1,000 ppm (consistent with the T-OAE \(p_{{\mathrm{CO}}_2}\) record42) using ocean/atmosphere equilibrium carbon-partitioning equations39 (Supplementary Information 4). We plot the results of these equations, CO2 release quantity versus \(p_{{\mathrm{CO}}_2}\) increase, in Supplementary Information 4.