Glacial/interglacial wetland, biomass burning, and geologic methane emissions constrained by dual stable isotopic CH₄ ice core records

Significance

Polar ice is a unique archive of past atmosphere. Here, we present methane stable isotope records (used as source fingerprint) for the current and two past interglacials and their preceding glacial maxima. Our data are used to constrain global emissions of methane. Tropical wetlands and floodplains seem to be the dominant sources of atmospheric methane changes, steered by past variations in sea level, monsoon intensity, temperature, and the water table. In contrast, geologic emissions of methane are stable over a wide range of climatic conditions. The long-term shift seen in both isotopes for the last 25,000 y compared with older intervals is likely connected to changes in the terrestrial biosphere and fire regimes as a consequence of megafauna extinction.

Abstract

Atmospheric methane (CH₄) records reconstructed from polar ice cores represent an integrated view on processes predominantly taking place in the terrestrial biogeosphere. Here, we present dual stable isotopic methane records [δ¹³CH₄ and δD(CH₄)] from four Antarctic ice cores, which provide improved constraints on past changes in natural methane sources. Our isotope data show that tropical wetlands and seasonally inundated floodplains are most likely the controlling sources of atmospheric methane variations for the current and two older interglacials and their preceding glacial maxima. The changes in these sources are steered by variations in temperature, precipitation, and the water table as modulated by insolation, (local) sea level, and monsoon intensity. Based on our δD(CH₄) constraint, it seems that geologic emissions of methane may play a steady but only minor role in atmospheric CH₄ changes and that the glacial budget is not dominated by these sources. Superimposed on the glacial/interglacial variations is a marked difference in both isotope records, with systematically higher values during the last 25,000 y compared with older time periods. This shift cannot be explained by climatic changes. Rather, our isotopic methane budget points to a marked increase in fire activity, possibly caused by biome changes and accumulation of fuel related to the late Pleistocene megafauna extinction, which took place in the course of the last glacial.

Past variations of the atmospheric methane mixing ratio ([CH₄]) have been documented using ice cores on various timescales (1–5). Compared with preindustrial levels of 715 parts per billion (ppb) (6, 7), anthropogenic emissions have led to a 2.5-fold increase since about the year 1800 CE as documented by the National Oceanic and Atmospheric Administration Earth System Research Laboratory. The lowest values of around 360 ppb are measured in ice cores during glacial maxima, whereas for the last eight interglacials, [CH₄] typically ranged between 600 and 700 ppb (4). Fast and significant stadial/interstadial [CH₄] increases occurred within a few decades during glacials in parallel to Dansgaard–Oeschger events (8). Furthermore, [CH₄] levels in the Northern Hemisphere are mostly higher than those in the Southern Hemisphere (9). These data taken together with our understanding of the present day methane budget (10–15) show that wetlands and primarily those located in the tropics dominate natural CH₄ emissions. On long timescales, wetland extent (16) and CH₄ fluxes are also strongly influenced by the position of the Intertropical Convergence Zone (ITCZ) (17) and the global monsoon system (18, 19), which in turn, is steered by changes in solar insolation. There is an ongoing debate on the relative contributions from tropical vs. boreal wetlands (ref. 9 and references therein). The latter is possibly overestimated, because other high-latitude sources, like thawing permafrost, thermokarst lakes (20), or (cryospherically capped) geologic seeps (21), also contribute to this source from a hemispheric (ice core) point of view.

Additional constraints on the contribution of the various methane sources can be derived from the carbon and hydrogen stable isotopic signature of CH₄ in ice cores. Microbially produced CH₄ is depleted in both stable isotopologues (¹³C and ²H or D; “light” isotopic sources) compared with the source mix. However, there are two important natural sources relatively enriched in ¹³C and D (“heavy” sources): biomass burning (BB), an important and climatically variable process (22–26), and “geologic” emissions of old methane (GEMs; called GEM by, for example, refs. 27–29).

Accordingly, the isotopic fingerprint of methane has been successfully used to shed light on relative source mix changes (30–33). However, biogenic sources (such as wetlands and BB) may change their isotopic signatures in parallel to changing climate and environmental conditions (25, 34). For the last glacial, Möller et al. (25) showed that the ¹³C content of CH₄ (δ¹³CH₄) is higher at low CO₂ levels and concluded that biome changes play a major role in this isotopic change. The deuterium content of CH₄ [δD(CH₄)] is also a function of the isotopic signature of the water used for methanogenesis (31, 35–38). However, the net impact of global meteoric water isotope changes caused by the transfer of water from the ocean to ice sheets is only of secondary importance (refs. 31 and 36 and references therein). Both δ¹³CH₄ and δD(CH₄) are influenced by at least one other factor, namely the ratio of net to gross CH₄ production in wetlands (i.e., the fraction of CH₄ consumed by methanotrophs before emission to the atmosphere). This process varies as a function of the water table and site conditions, leading to changes in strength and isotopic composition of CH₄ emissions (31, 39).

In this contribution, we greatly extend the existing carbon and hydrogen isotopic information of CH₄ from ice cores in terms of temporal coverage, temporal resolution, accuracy, and precision. Our data cover three interglacials and their preceding glacial terminations and glacial maxima. The interpretation of these data centers around the discussion of the source side of the methane cycle. This assumption is justified in the light of recent work on sinks of methane (40–43), with net variations that are estimated to be relatively small (SI Text). Hence, dual stable isotope records of methane provide important insights into the suite of terrestrial and marine processes emitting methane and their changes in the past.

Ice Core Measurements

We measured records of methane stable carbon (δ¹³CH₄) and hydrogen [δD(CH₄)] isotopes from four Antarctic ice cores (Fig. 1 and Figs. S1 and S2–S5): i.e., the two European Project for Ice Coring in Antarctica (EPICA) ice cores from Dronning Maud Land (EDML) and from Dome Concordia (EDC), the Talos Dome [Talos Dome Ice Core Project (TALDICE)], and the Vostok (core 5G; Vostok Station) ice cores. Note that these values are representative of the tropospheric isotope signature of methane in high southern latitudes and that an interpolar difference (IPD) in [CH₄] and its isotopic signatures exists.

Fig. 1.

Paleoclimatic records for three interglacials and preceding glacial maxima (note the breaks in the x axis). From top to bottom, the panels show (A) solar insolation in June at 30° N (133) and atmospheric δ¹⁸O from Vostok (purple) (134), EDC (light pink) (51, 135–138), and Siple Dome (red) (84); (B) [CH₄] (ref. 4 and data from this study); (C) δD(CH₄) from EDML and EDC (this study; error bars are 1-sigma SDs of reference air measurements); (D) δ¹³CH₄ from Talos Dome, EDC, and Vostok (5G; this study; the error [based on 1-sigma SDs of replicate ice core measurements (47)] is approximately the size of the symbols) and data from EDML and Vostok (25, 32); (E) [CO₂] (110); and (F) relative sea level as reconstructed from Red Sea sediment cores (108). Time intervals indicative of MIS (45) are given next to the sea-level curve. Ice core records are given on the Antarctic ice core chronology (AICC2012) gas age scale (137, 139), and insolation and sea level on their individual age scales. Note the inverse direction of all isotope axes.

Fig. S1.

Ice core records of δ¹⁸O_atm (51, 135–138), [CH₄] (4), its stable isotopes [this study and data measured in refs. 25 and 32; note that the dataset by Fischer et al. (32) was corrected for a Kr effect as presented in ref. 25], and [CO₂] (110). The red and blue shading indicates interglacial and glacial time slices, respectively, used to calculate numbers for Table S1. This plot also shows additional Bern δ¹³CH₄ data between 25 and 80 ka BP, which are not presented in Fig. 1.

Fig. S2.

Highlighted data for the investigated interglacial periods (note the breaks in the x axis). (A) Solar insolation in June at 30° N (133) and atmospheric δ¹⁸O from EDC (51, 135–138); (B) speleothem δ¹⁸O data from Sanbao cave (70); (C) [CH₄] (4) (green line) and data from this study (open diamonds are from TALDICE, open circles are from EDC, and open triangles are from Vostok samples); (D) δ¹³CH₄ from TALDICE, EDC, and Vostok [5G; this study; symbols are chosen as for [CH₄]; the error bars represent the 1-sigma SD of ice core replicates (47): 0.15‰]; and (E) relative sea level as reconstructed from marine sediment records from the Red Sea (108). Ice core records are given on the Antarctic ice core chronology (AICC2012) gas age scale (137, 139), and insolation, speleothem data, and sea level are given on their individual age scales. Note the inverse direction of all isotope axes.

Fig. S5.

Paleoclimatic records of Fig. 1 zoomed for MIS 12 and MIS 11.3. From top to bottom, the panels show (A) solar insolation in June at 30° N (133) and atmospheric δ¹⁸O from Vostok (purple) (134), and EDC (light pink) (51, 135); (B) [CH₄] (ref. 4 and data from this study); (C) δD(CH₄) from EDML and EDC (this study; error bars are 1-sigma SDs of reference air measurements); (D) δ¹³CH₄ from Talos Dome, EDC, and Vostok (5G; this study; the error bars represent the 1-sigma SD of ice core replicates (47): 0.15‰) and data from EDML and Vostok (25, 32); (E) [CO₂] (110); and (F) relative sea level as reconstructed from Red Sea sediment cores (108).

The investigated time periods are from 25 to 0.5 kilo years (ka) before present [(BP), where present refers to 1950], from 160 to 80 ka BP, and from 440 to 370 ka BP (Fig. 1 and Fig. S1). With reference to marine sediment records, these time periods are approximately equivalent to Marine Isotope Stage (MIS) 2&1, MIS 6&5.5, and MIS 12&11.3, respectively (44, 45).

The δ¹³CH₄ data were measured on the TALDICE (170 samples) and EDC (90 samples) ice cores and using seven samples from Vostok for MIS 5.5. For δD(CH₄), we present data from EDML (54 samples) for MIS 2&1 and MIS 5.5 and from EDC (47 samples) for MIS 6&5.5 and MIS 12&11.3. Altogether, this dataset presents dual stable isotope records from three glacial maxima, the following terminations and interglacials, and two glacial inceptions.

Information on measurement procedures, accuracy, and precision can be found in Materials and Methods and our technical papers (46–48). δ¹³CH₄ and δD(CH₄) data are reported using the δ notation on the international Vienna Pee Dee Belemnite (VPDB) and Vienna Standard Mean Ocean Water (VSMOW) scales, respectively. Information on data handling with respect to firn diffusion effects is described in SI Text.

Results

The main results of this study, presented in Fig. 1 and Fig. S1, confirm previous results but largely extend the time coverage and are based on data with improved precision and accuracy (46–48). Our results are consistent with δ¹³CH₄ data from Greenland Ice Sheet Project 2 (GISP2) (49) and EDML (32) for the Holocene (MIS 1), the last termination, and the last glacial maximum (LGM) (Fig. 1 and Figs. S3 and S6), as well as throughout the last glacial period (25) (Fig. S1). For the last 25 ka, we present a greatly improved view of the evolution of δD(CH₄) compared with the pioneering work by Sowers (31, 49), which was characterized by higher sample to sample variability (Fig. S6).

Fig. S3.

Paleoclimatic records of Fig. 1 zoomed for the LGM and the Holocene. From top to bottom, the panels show (A) solar insolation in June at 30° N (133) and atmospheric δ¹⁸O from Vostok (purple) (134), EDC (light pink) (51, 135–138), and Siple Dome (red) (84); (B) [CH4] (ref. 4 and data from this study); (C) δD(CH₄) from EDML and EDC (this study; error bars are 1-sigma SDs of reference air measurements); (D) δ¹³CH₄ from Talos Dome, EDC, and Vostok (5G; this study; the error bars represent the 1-sigma SD of ice core replicates (47): 0.15‰) and data from EDML and Vostok (25, 32); (E) [CO₂] (110); and (F) relative sea level as reconstructed from Red Sea sediment cores (108).

Fig. S6.

Comparison of previously published datasets with this study for the last 25,000 y. (A) δD(CH₄) by Sowers (31, 49) from the Greenland core Greenland Ice Sheet Project (GISP2) and data from this study from EDML (Antarctica). Note that the offset of the two datasets is caused by a not well-quantified IPD in δD(CH₄) (approximately −16‰) plus an interlaboratory-scale offset (46). Additional slight differences might occur, because the datasets by Sowers (31, 49) are not free from a Kr effect, whereas δD(CH₄) data from this study are measured without any Kr interference (46). (B) δ¹³CH₄ from the Greenland core GISP2 (49), data from EDML (32), and data from this study from TALDICE and EDC (all three cores from Antarctica). Also for δ¹³CH₄, the IPD is not well-known (approximately −0.5‰). We stress that only the δ¹³CH₄ data of this study are measured without any Kr interference (47, 48). The data by Sowers et al. (49) and Fischer et al. (32) have been corrected for a Kr effect by ref. 25 using two different approaches: the correction (Δδ¹³C_Kr) of Vostok and GISP2 data measured at the Pennsylvania State University were inferred indirectly from CH₄ mixing ratios (referred to as type 1), whereas for the EDML record, the correction was based on the Kr-induced anomaly derived from the ion current ratios (type 2; section 1.3 of ref. 25 has a detailed description of both approaches).

In our data, we document a rather smooth deglacial δD(CH₄) decrease of 18‰ and only small long-term variations over the Holocene, with a mean value of −71‰ for the Southern Hemisphere. Note that there is a north–south IPD for δD(CH₄) of roughly −16‰ for the Holocene (46). Interestingly, glacial/interglacial amplitudes in δD(CH₄) can be either of similar amplitude (LGM–Holocene: ∼14‰) as swings from stadial to interstadial conditions (e.g., during the glacial inception around 390 ka BP in Fig. 1) or considerably larger (MIS 6–MIS 5: ∼25‰). The decrease of δD(CH₄) into MIS 5.5 is also faster compared with the much smoother MIS 2/1 and MIS 12/11.3 transitions. On the contrary, all of the δ¹³CH₄ transitions investigated are gradual (Figs. S3–S5).

Within full glacial periods and interglacials, δD(CH₄) shows rather small variations [standard deviation (SD) is typically around 5‰ (Table S1)] compared with the large leverage of sources and sinks, indicating little changes in the source composition. On the contrary, δ¹³CH₄ shows pronounced trends that differ for all of the interglacials investigated. Most importantly, interglacial δ¹³CH₄ is not correlated to [CH₄]. Next, while for δD(CH₄) and δ¹³CH₄, the interglacial MIS 5.5 and MIS 11.3 mean levels are comparable, a clear shift of the mean values for the Holocene is evident (Fig. 1 and Fig. S2). A shift of similar size is also found for the mean level during the LGM compared with MIS 6 and MIS 12 (Fig. 1 and Table S1).

Table S1.

Quantitative estimates of the mean methane stable isotopic signature (this study) and mixing ratios (4) characterizing the time slices used in the box model

Note the large SDs, especially for δ¹³CH₄ and [CH₄] during interglacials, indicative of the large signal ranges observed within these time periods (compare Fig. 1 and Figs. S2–S5). Red and blue colors represent interglacial and glacial time slices, respectively, in line with colored bars in Fig. S1. Columns from left to right give the minimal and maximal gas ages of the analyzed time periods on the Antarctic ice core chronology (AICC 2012) age scale; the names of the ice cores used; the number of samples (N), average (mean), SD, median, minimum value, and maximum value for [CH₄], δ¹³CH₄, δD(CH₄); and the MIS roughly corresponding to the ice core time slices.

Our data confirm earlier findings (25, 49) that δ¹³CH₄ is evolving independently from [CH₄] for large parts of the ice core record. This feature is substantiated by new δ¹³CH₄ data from EDC over the MIS 12/11.3 transition and the penultimate glacial termination as well as for variations during MIS 11.3. In all of the terminations investigated, δ¹³CH₄ drops strongly when [CH₄] increases only slowly, well before the major rapid [CH₄] rise (Fig. 1 and Figs. S3–S5). Only about one-half of the amplitude of the δ¹³CH₄ change is covariant with the rapid methane rises into the interglacial periods. The decoupling between δ¹³CH₄ and [CH₄] is even more evident during rapid CH₄ rises connected to Dansgaard–Oeschger (DO) events, which have no counterpart in δ¹³CH₄. This observation is indicative of small source mix changes (25). Moreover, there is no abrupt δ¹³CH₄ shift connected to the [CH₄] peak during the early MIS 5.5 (128.5 ka BP) (Fig. 1). On the contrary, δ¹³CH₄ continues its downward trend during MIS 5.5, when [CH₄] decreases toward lower glacial values, whereas [CO₂] and δ¹⁵N₂ in ice cores (a proxy for firn temperature) indicate interglacial conditions (50, 51). δ¹³CH₄ only reverses its trend around 115 ka BP when [CO₂] and δ¹⁵N₂ start to drop (Fig. S4). At that point in time, [CH₄] is already below 500 ppb, a level typical of stadial intervals.

Fig. S4.

Paleoclimatic records of Fig. 1 zoomed for MIS 6 and MIS 5.5. From top to bottom, the panels show (A) solar insolation in June at 30° N (133) and atmospheric δ¹⁸O from Vostok (purple) (134), EDC (light pink) (51, 135–138), and Siple Dome (red) (84); (B) [CH₄] (ref. 4 and data from this study); (C) δD(CH₄) from EDML and EDC (this study; error bars are 1-sigma SDs of reference air measurements); (D) δ¹³CH₄ from Talos Dome, EDC, and Vostok (5G; this study; the error bars represent the 1-sigma SD of ice core replicates (47): 0.15‰) and data from EDML and Vostok (25, 32). The graph is extended by E showing δ¹⁵N of air (51). Panel (F) shows [CO₂] (110); and (G) relative sea level as reconstructed from Red Sea sediment cores (108).

To confirm the very low δ¹³CH₄ values of our EDC and TALDICE records at the end of MIS 5.5 (which were not seen in the very few Vostok samples previously measured at Pennsylvania State University and presented in ref. 25), we analyzed an additional seven samples from the Vostok core with our improved method. These samples correspond well to our data from other ice cores (Fig. 1 and Fig. S4), thus indicating that the lowest values for δ¹³CH₄ of −52‰ must have been missed in the older Vostok record (25), most likely because of methodological problems. Other small differences in our record compared with the older Vostok time series are presented in figure 9 of ref. 47, but the overall conclusions of ref. 25 are confirmed in the light of the improved dataset presented here (Fig. 1 and Fig. S1).

Discussion

The Potential of Marine Clathrates and Other Geologic CH₄ Emissions.

Decreasing δD(CH₄) over all three glacial terminations (Fig. 1) supports the conclusion in refs. 31 and 36 that marine clathrates (gas hydrates) do not significantly contribute to the altered atmospheric CH₄ budget during transitions. However, it is not only catastrophic emissions caused by destabilization events of marine clathrates that have been proposed to explain past [CH₄] variations (52, 53) but also, more steady emissions through, for example, natural marine hydrocarbon seeps that may have been exposed during times of sea-level low stands. Together with seeps and mud volcanoes, clathrate releases constitute the so-called GEM (27, 28, 54). Concerning the modern Arctic, there is also an ongoing debate on the origin and importance of CH₄ releases from the East Siberian Shelf (55–59), which are thought to stem from both organic carbon in thawing subsea permafrost and geologic reservoirs.

In the following section, we show that our isotope data (Fig. 1) are incompatible with the strong role of GEM proposed by, for example, refs. 29, 60, and 61. Luyendyk et al. (61) argue that lowered methane concentrations along with high δ¹³CH₄ values during the glacial could be a result of increased GEM, whereas wetland emissions were reduced to a minimum. However, our δD(CH₄) constraint indicates that this hypothesis is incorrect for two reasons. First, the differences in δD(CH₄) between glacials and interglacials are too small to be compatible with a dominant GEM source, and second, a covariation (e.g., within MIS 5.5) of δ¹³CH₄ and δD(CH₄), expected under the dominant GEM hypothesis, is not seen in our data (Fig. 1 and Figs. S2–S5). An unambiguous assessment of the geologic emissions based on ¹⁴CH₄ measurements performed on Antarctic ice over the last termination is pending, but results presented by Petrenko et al. (62) indicate only a 10% contribution from GEM to the atmospheric methane budget during the Younger Dryas period and no strengthening of this source for the [CH₄] rise into the Preboreal.

To quantify the maximum contribution of GEM based on our data, we used our previously described box model (32, 36). However, to be consistent with work on the recent atmospheric methane budget (63), we differentiated only three source categories (microbial, GEM, and BB). Emissions and isotopic signatures of these three sources were varied in our model within predefined limits in a Monte Carlo approach. Moreover, we also included a Cl sink for CH₄ in the marine boundary layer. Equilibrium results of each model run were compared with the ice core constraints (Table S1), and 10,000 valid runs were recorded for each time slice (details are in Materials and Methods, SI Text, and Fig. S7). Hence, all box model runs accepted in this study are consistent with the presented ice core constraints within the error limits of the data.

Fig. S7.

Box model results fulfilling the ice core constraints. Each line encloses 10,000 valid realizations of the Monte Carlo model covering the parameter spaces of six time periods (Table S1). The Matlab function convhull() was used to determine the envelope around the solutions for each time slice. Legends are valid for all subpanels. (A and B) Shown are emission strengths of the microbial model source in relation to (A) BB and (B) GEM for interglacial and glacial times. (C and D) Shown are emission strengths of the microbial model source in relation to its isotopic signature: (C) δ¹³CH₄ and (D) δD(CH₄) for interglacial and glacial times. δ¹³CH₄ of the microbial source is chosen to be heavier by 2.6‰ for glacials (in the text and Table S2). (E and F) Box model results of the standard setup (lines) and the sensitivity runs with zero IPD of [CH₄] (dashed lines). Shown are emission strengths of BB in relation to GEM. E shows results for interglacials, and F shows results for glacials.

Here, we focus on the model results for GEM and BB presented in Fig. 2, showing a clear and expected anticorrelation of the emission strengths of the two sources (because both are enriched in ¹³C and D relative to the microbial source). Moreover, Fig. 2 shows that, for any given GEM value, BB emissions are higher in the Holocene and the LGM compared with previous interglacials and glacials, respectively (Biome and Fire Regime Changes Caused by Megafauna Extinction). We can use Fig. 2 to constrain possible GEM, where we can safely assume that GEMs are the same during the Holocene compared with previous interglacials and the same for the LGM compared with previous glacial maxima. Our model results allow individual scenarios with higher GEMs for glacials compared with interglacials (Fig. 2). However, because BB emissions must always be larger than zero, an absolute upper limit of interglacial (glacial) GEM is 90 (70) Tg CH₄ a⁻¹ (Terragram methane per year) according to our data constraints. Using independent estimates of Holocene and LGM BB emissions (10, 32, 36, 63–66) of 25 and 15 Tg CH₄ a⁻¹, respectively, GEMs are in fact smaller than 47 (Holocene) and 41 (LGM) Tg CH₄ a⁻¹. It is important to stress that our mean Holocene estimate is based on krypton-free δ¹³CH₄ data, resulting in lower δ¹³CH₄ values and thus, slightly lower GEMs compared with previous assessments (63, 64) of the Late Holocene (discussion is in SI Text).

Fig. 2.

Box model results fulfilling the ice core constraint. Shown are emission strengths of BB and GEM. Each envelope encloses all 10,000 valid realizations of the Monte Carlo processes searching the parameter spaces of the six time periods (SI Text and Table S1). The Matlab function convhull() was used to determine the envelope around the solutions for each time slice. Upper shows results for interglacials, and Lower shows results for glacials. Model results for the Holocene and the LGM are considerably shifted toward higher GEM and/or BB because of the ¹³C- and D-enriched isotope targets of these younger time slices (Fig. 1 and Table S1). Arrows indicate possible readouts to assess maximal GEM for all investigated time periods (solid lines) and increased BB for the Holocene and the LGM compared with previous time periods (dashed lines).

Overall, we conclude that GEMs (seeps and marine clathrates) are at no point the dominant contributor to the global methane budget, and they are not strongly variable players that could explain the observed glacial/interglacial [CH₄] variations over the last few hundreds of thousands of years (Figs. 1 and 2 and SI Text) (25, 31–33, 36, 62). Thus, we infer that microbial sources must represent the dominant control for natural atmospheric CH₄ changes.

The Role of High Northern Latitude Microbial Emissions.

Recently, Köhler et al. (67) calculated that—dependent on assumptions on the gas age distribution of the bubbles in the ice—up to 14 Tg CH₄ a⁻¹ could have been released into the atmosphere from permafrost thawing (a source relatively depleted in ¹³C and D) over the Oldest Dryas–Bölling/Alleröd (OD-BA) transition. Unfortunately, the resolution of our δD(CH₄) data for this event is insufficient to give a direct and qualified answer. For the Younger Dryas–Preboreal (YD-PB) transition, Melton et al. (68) argued that it is possible to close the isotopic budget by a parallel increase in ¹³C- and D-enriched emissions from BB and depleted emissions from thermokarst lakes. However, given our information on a gradual and modest decrease of δD(CH₄) over this transition decoupled from the strong [CH₄] increase (Fig. S3), this argumentation is only one valid scenario. We argue that a dominant contribution of D-depleted high-latitude emissions (in which we include thawing permafrost/thermokarst lakes/boreal wetland emissions) to the rapid CH₄ increases is unlikely, because the gradual decrease in δD(CH₄) during the Preboreal starts only after [CH₄] is already high (Fig. S3). For the Younger Dryas, δD(CH₄) stays at −62‰ and does not change over the YD-PB transition. Also, there is no imprint of rapidly increasing ¹³C-depleted emissions from northern high-latitude wetlands (as proposed in ref. 69) in the rather smooth evolution of δ¹³CH₄. For both rapid [CH₄] rises (OD-BA and YD-PB), explanations are preferred that invoke increasing emissions of sources with small isotopic leverage on δD(CH₄) and δ¹³CH₄ as the main drivers. This kind of emission change is fulfilled by strengthening low-latitude microbial sources, such as predominately C4-fed tropical wetlands, which incorporate increasingly depleted water (as seen in δ¹⁸O records) during methanogenesis at that time (12, 35, 70).

For the Holocene, several authors have suggested higher CH₄ emissions from high-latitude ecosystems relative to the LGM (71–73). Based on pollen analyses, Yu et al. (73) proposed a protopeatland phase as the precursor for the succession from wetlands to fens and later bogs. Because peatlands evolve from fens to bogs, this succession is accompanied by decreasing CH₄ emissions and a shift to lower δ¹³CH₄ signatures. The latter may be linked to trophic status, degree of methanotrophy, plant types, or type and quality of organic substrates (49, 74–76). However, because there is a strong reduction of methane emissions during this succession, the source signature effect does not leave a sizable imprint in the CH₄ isotopic signature of the atmosphere. Accordingly, the observed leveling out of δ¹³CH₄ changes during the Holocene is in line with decreasing emissions from northern peatlands (49). To close the budget, the increase in [CH₄] over the latest 4.5 ka calls for an additional source. Several theories have been put forward (42, 72, 77, 78), on which we comment below.

Control of (Sub-)Tropical Wetland and Floodplain Emissions on Atmospheric CH₄.

Multiple lines of evidence suggest that (sub-)tropical (sporadically/seasonally inundated) floodplains, wetlands, and peatlands dominate global natural methane emissions (2, 4, 9–12, 14, 17, 25, 79–87). Our data provide additional support for this hypothesis. To underscore the tropical (Ax) or wet and warm temperate (Cax) climatic boundary conditions (where x denotes any of the second characters of the classical Köppen–Geiger climate classification in table 1 of ref. 88, necessary for such wetlands), we refer to this group of methane-emitting systems as AxCax wetlands throughout. The AxCax wetlands are largely located in regions influenced by seasonal swings of the ITCZ (17, 89–95). Because the sizes of AxCax regions are not evenly distributed in both hemispheres, a good portion of their emissions (located, for example, in Southeast Asia) also contributes to the IPD in methane mixing ratio over a wide range of climate states (9).

Temperature, precipitation, the water table, and net primary production (NPP) are regarded to be the main factors controlling CH₄ fluxes in AxCax wetlands (80, 92, 96, 97). Low [CO₂] during glacials reduces NPP (98), and we expect decreased fluxes from AxCax wetlands as a direct consequence. However, lower glacial sea level led to newly exposed AxCax landmasses (such as the Sunda Shelf), where wetlands could develop (16). On top of an overall reduction in AxCax CH₄ fluxes during glacial times, substantial ecosystem shifts leading to a larger C4/C3 plant ratio may explain parts of the glacial δ¹³CH₄ evolution (ref. 25 and references therein).

Specific evidence for the key role of AxCax wetland CH₄ emissions comes from the rather small amplitudes in the δD(CH₄) response for stadial/interstadial (36) and glacial/interglacial changes (Fig. 1 and Table S1) (31). Because extratropical methane sources have a stronger leverage on the integrated hydrogen isotopic source signature, they are expected to experience larger glacial/interglacial δD(CH₄) changes than their AxCax counterparts. Records from speleothems (85, 99) and plant waxes (100) located in AxCax climates suggest amplitudes for (meteoric) waters used for methanogenesis that are in line with our atmospheric δD(CH₄) data (Fig. 1). High-latitude changes in δD of precipitation are much stronger (35, 101) and would lead to δD(CH₄) changes too large compared with our data if this source were to control the observed CH₄ variations.

The earlier proposal by Ridgwell et al. (102) that flooding of the continental shelves is a main contributor to initial steep methane rises is in line with our dual isotope records.

Apart from the overall glacial/interglacial shifts, the variations in δ¹³CH₄ are largely decoupled from the changes in δD(CH₄). To understand the observed variations in δ¹³CH₄ (Fig. 1), we discuss in the following changes on the Indonesian archipelago, a region for which wetland history since the LGM has been studied in great detail and that can, therefore, serve as a blueprint for our process understanding. Recently, Dommain et al. (103) presented local sea level as the key player controlling Sundaland’s wetland extent since the LGM. Rising sea level during the deglaciation and Early Holocene lowers the regional hydraulic gradient, leading to higher water tables for peatlands in this region. Falling local sea levels after 5 ka BP lead to an expansion of peatlands located in the coastal lowlands (103). Furthermore, sea-level changes in the Sunda Shelf region may also control moisture supply in the Indo-Pacific Warm Pool and intensity of monsoonal rainfall (85, 99, 104, 105). Taken together, these findings suggest that AxCax wetland CH₄ emissions from the Indonesian archipelago may vary over precessional timescales because of sea level and precipitation changes. We note that Sundaland represents only a fraction of tropical wetland area, and we suggest that other large-scale tropical methane-emitting systems (like the Amazon and the Congo basins) responded similarly to (local) sea level and hence, hydrological gradient changes. Taken together, wetland methane emissions of South America, Africa, and Sundaland can explain the observed ice core signals.

Dommain et al. (103) also suggest that the exposure of the Sunda Shelf led to drier conditions after MIS 3, causing degraded inland wetlands during the LGM. We propose that the baseline level of atmospheric methane (2, 106) is in fact determined by AxCax wetlands (located in Sundaland and other AxCax regions) and that its decline to the lowest levels observed in ice cores after MIS 3 is caused by the drying of tropical wetland systems. We observe that the CH₄ response reported for Greenland interstadials (DO 2/3, 18/19/20, and 22/23) during periods of falling sea level is generally small (8, 107, 108). In other words, [CH₄] only shows large stadial/interstadial increases during periods of rising (local) sea level when insolation and increased monsoon precipitation could, in principle, boost wetland CH₄ emissions (2, 4, 19, 25, 70, 106, 109). Hence, only under the prerequisite of a low(ering) hydraulic gradient in AxCax wetland regions can any forcing (temperature or precipitation) lead to strong methane production increases during DO events and glacial/interglacial terminations.

One open question remaining is why MIS 5.5 and MIS 11.3 [CH₄] and δ¹³CH₄ histories differ drastically in their temporal evolution, whereas δD(CH₄) is rather constant: near −89‰ for both interglacials (Fig. 1). Most importantly, a pronounced minimum in δ¹³CH₄ is found at the end of MIS 5.5 (116 ka BP) during the time of minimal northern insolation. At the same time, [CH₄] decreases continuously toward glacial levels, with a steeper decrease from the minimum δ¹³CH₄ values onward. The following scenarios cannot be used to explain the δ¹³CH₄ minimum during MIS 5.5. (i) A proportional reduction of all sources. Because this scenario would lead to no signal in δ¹³CH₄, it can be ruled out. Thus, a change in the source mix or a shift in the isotopic signature of the (dominant) source(s) is required. (ii) Reduced microbial emissions while keeping geologic and BB emissions constant. Because this combination would produce higher δ¹³CH₄ when total CH₄ decreases, it also cannot explain our observations. (iii) Stable or even increasing emissions from microbial, isotopically light sources [such as ¹³C- and D-depleted (boreal) wetlands and thermokarst, permafrost]. This setting would call for overcompensation by decreasing ¹³C-enriched emissions to meet falling [CH₄]. Accordingly, decreasing emissions from BB and/or GEM could result in the observed low δ¹³CH₄ values. The last two scenarios can be ruled out on the basis of our δD(CH₄) constraint, because we would expect coevolving trends for both isotopes. In contrast, our δD(CH₄) data show no clear trend during this time period.

The δ¹³CH₄ minimum occurs at the very end of MIS 5.5 at a time when δ¹⁵N₂ and [CO₂] indicate the end of the warm period (Fig. S3) (50, 51, 110). At that time, [CH₄] is already close to glacial levels of below 500 ppb (4). We propose that the lowest interglacial CH₄ levels (coincident with northern insolation minima at the start and end of MIS 11.3, the end of MIS 5.5, and the mid Holocene) are also mainly because of decreased AxCax wetland emissions. At the same time, these periods are connected to δ¹³CH₄ minima. To close the isotope budget, a simultaneous reduction in a relatively ¹³C-enriched source is required. Hence, additionally, declining BB emissions are proposed to meet all constraints. Such a scenario is in line with lower BB emissions under cooling climate conditions (24) and recently published speleothem data (70) showing reduced Asian monsoon strength corresponding to low [CH₄] and low δ¹³CH₄ during MIS 11.3 and for the late MIS 5.5 δ¹³CH₄ minimum.

Interestingly, CH₄ and δ¹³CH₄ decrease during the first one-half of the Holocene but reverse their trend during the second one-half, when northern insolation is still declining. To explain this feature, Ruddiman (77) proposed an early human influence. An alternative scenario meeting all of the constraints presented is that of stronger southern insolation, leading to increasing CH₄ emissions of AxCax wetlands in the tropical Southern Hemisphere (e.g., South America). This scenario is consistent with proxy data (111, 112) and a CH₄/climate model study (78). Moreover, our CH₄ and δ¹³CH₄ data during MIS 11.3 show the same behavior as in the Holocene. We can conclude that the Holocene trends operate similarly to MIS 11.3, with strong southern insolation causing increased southern tropical AxCax wetland emissions during the last 5 ka. For MIS 5.5, this insolation boost from the south would have come too late and already during falling sea level, causing [CH₄] to drop continuously.

In summary, we conclude that tropical methane-emitting systems are the key players among all natural methane emitters, reflecting changes in (local) sea level, monsoon strength, and temperature induced by orbital changes.

Biome and Fire Regime Changes Caused by Megafauna Extinction.

Arguably, the most surprising feature of our records (Fig. 1) is the pronounced difference in absolute levels for both δ¹³CH₄ and δD(CH₄) for the Holocene compared with MIS 5.5 and MIS 11.3 and for the LGM compared with MIS 6 and MIS 12. Shifts of ∼2–3‰ for δ¹³CH₄ and 10–18‰ for δD(CH₄) toward higher numbers are found, with no obvious difference in CH₄ mixing ratio between these time slices (Table S1) (42). Straightforward explanations for similar [CH₄] accompanied by shifted isotope records require changes in the source signatures or changes in emission strength of a source with strong leverage. To our knowledge, no general isotope shifts of that size have been described in precursor materials for methanogenesis before MIS 2. It is also unlikely that the source strength or signature of GEM or biogenic emissions changed markedly compared with previous glacial/interglacial cycles. In fact, GEM is expected to change in response to sea level or ice sheet extent, but the two parameters remain within a similar range for all glacials and all interstadials considered in this study. One possibility to reconcile the observations is CH₄ emission changes related to changes in biomes and fire regimes, because BB is a CH₄ source strongly enriched in ¹³C and D (13, 22, 65). BB is an ancient and persistent feature throughout the geologic record (113), and there is evidence of net changes in fire regimes as a consequence of the megafauna extinction that was presumably caused by rapid climate changes in combination with human interference in the course of the last glacial (refs. 114–119 and references therein). The review by Johnson (120) on the timing of the arrival of humans on different continents and the ecological consequences of megafauna extinction supports the idea that increased fire frequency was caused by increased vegetation density and the accumulation of plant material not consumed by herbivores. For example, records from Australia of charcoal, different plant pollen types, and spores of the fungus Sporormiella are used by Rule et al. (121) to indicate large herbivore activity and conclude that megafauna extinction caused increased fire activity after 41 ka BP. Furthermore, these Australian records show that fires were common during the Holocene but much less frequent in the previous interglacial. We note that responses might be different in other parts of the globe (122–124) and that, today, Australia accounts for only roughly 6% of global fire carbon emissions (125). However, other authors reported similar observations of fire activity changes on other continents (126, 127), but a global synthesis is not available yet.

Assuming similar GEM for all of the time periods investigated, we can derive the net change in BB emissions for different time periods from Fig. 2, where we compare model interglacial (glacial) runs with the same interglacial (glacial) model parameters (all identical except for a small shift from the microbial source to BB). For example, an assumption of 30 Tg CH₄ a⁻¹ of GEM results in a shift in BB emissions for the Holocene by 15 Tg CH₄ a⁻¹ (or +83%) compared with MIS 5.5 and MIS 11.3. For the LGM, the model results show an increase in BB emissions by about 7 Tg CH₄ a⁻¹ (or +54%) compared with MIS 6 and 3 Tg CH₄ a⁻¹ (or +18%) compared with MIS 12. Hence, our dual stable isotope records (Fig. 1) directly support the hypothesis (120, 121) of higher fire activity during the Holocene and the LGM compared with previous interglacials and glacials, respectively. At the same time, the largely unchanged CH₄ levels suggest that direct CH₄ emissions from large animals are confined to the lower end of values found in the literature (128).

Conclusions

Stable isotopic methane records from polar ice cores offer insights into past methane emission inventories and significantly improve our quantitative understanding of past atmospheric methane changes. With our dual isotope data, we can rule out a dominant role for GEM in the glacial methane budget and especially, past emission changes. In fact, methane emissions from tropical wetland and seasonally inundated floodplain systems seem to be the strongest source for not only interglacials but also, glacials. These emissions play the major role in the waxing and waning of atmospheric methane mixing ratios recorded in polar ice. Key parameters to control CH₄ emissions in these wetland systems (AxCax wetlands) are temperature and the water table as steered by a combination of solar insolation, (local) sea level, and monsoon strength. Between 80 and 25 ka BP, the CH₄ emissions experienced a shift in both stable isotopes, leading to higher (heavier) values for the younger part. This observation is hard to explain by climate-driven wetland changes. We propose that this shift is caused by biome changes that foster BB emissions in the course of the late Quaternary megafauna extinction.

Materials and Methods

δ¹³CH₄ Analysis of Ice Core Samples.

All presented δ¹³CH₄ data were measured at the University of Bern using the system described in detail in ref. 47. All data are free of a krypton interference (48). Blank ice measurements indicate no artifacts associated with ice processing (47). The external precision (1-sigma SD) is estimated based on the long-term repeatability of replicate ice core analyses to be better than 0.15‰ (47). Our reference is a bottle of Pacific air calibrated at the Stable Isotope Laboratory of the Institute of Arctic and Alpine Research (University of Colorado) (47). All δ¹³CH₄ data are given using the δ notation on the VPDB scale. δ¹³CH₄ data have been corrected for gravitational settling in the firn using published δ¹⁵N₂ records (SI Text). In polar firn, diffusive isotopic fractionation takes place when large and rapid changes of the CH₄ mixing ratio (129, 130) occur. We calculated the expected influence of this effect according to the procedure presented in ref. 129 and apply this correction to the data plotted in our figures. Note that these effects are rather short-lived and very small for our selected samples and do not affect any of our conclusions. [CH₄] data shown in our figures were obtained using the isotope analysis and have an estimated 1-sigma error of 5 ppb based on TALDICE replicate measurements (47).

δD(CH₄) Analysis of Ice Core Samples.

All presented δD(CH₄) data have been measured at the University of Bern using the system described in detail in ref. 46. All data are free of a krypton interference. Blank ice measurements indicate no artifacts associated with ice processing (131). For δD(CH₄), the external precision (used as error bars in the figures) is estimated based on the 1-sigma SD of our daily standard gas measurements used to calibrate the sample. This uncertainty ranged between 0.9‰ and 3.8‰, with a median of 2.1‰. Replicate analyses of ice core samples indicate a reproducibility of better than 2.3‰ (pooled 1-sigma SD for measurements performed during a time period of 3.5 y) (46). All data are given in the commonly used δ notation on the VSMOW scale. Note that no international reference standard for δD(CH₄) in air/ice core samples exists so far. Our data are tied to the scale of the Institute for Environmental Physics in Heidelberg, Germany using a cross-calibrated bottle of recent air (46). A laboratory offset to measurements from the United States is evident (46). δD(CH₄) data have been corrected for gravitational settling in the firn using published δ¹⁵N₂ records (SI Text). No corrections concerning diffusive isotopic fractionation (129, 130) have been performed for δD(CH₄) data (SI Text), because the size of this correction is negligible compared with the δD(CH₄) variations.

Box Model.

We used the box model presented in refs. 32 and 36 to assess maximal GEM and increased emissions by BB for the last 25 ka compared with previous time periods. The atmosphere of the model consists of four boxes (northern and southern troposphere and stratosphere) with prescribed air mass exchange. CH₄ emissions into the tropospheric boxes as well as their source signatures and sink fractions are varied in a Monte Carlo approach. The model is run into steady state, and the equilibrium value of the southern troposphere box is compared with the data constraints (Table S1). If the modeled [CH₄], δ¹³CH₄, and δD(CH₄) values are compatible [within the uncertainty of the measured data (Table S1)], the input values are recorded as a possible CH₄ budget solution. For each time slice, 10,000 valid runs are collected. To be consistent with work on the recent methane budget (63), we distinguish only three source categories here: GEM, BB, and a microbial source including natural emissions by wetlands, ruminants, and termites. Isotopic signatures of the sources are based partly on our previous work (refs. 32 and 36 and references therein) and mainly on the collection by Schwietzke et al. (63) with some modifications (details are in SI Text and Table S2). In particular, we adjusted the geographical distribution of the sources to be consistent with IPD estimates for [CH₄] (9, 79, 89, 132) and lowered the isotopic signature of the Northern Hemisphere microbial source. To account for biome shifts during the glacial, we shift the microbial and BB source signatures by +2.6‰ in line with the interpretation given in ref. 25. The model uses four sinks (tropospheric OH, stratospheric loss, soils, and tropospheric Cl), with fractions that have been varied according to Table S3. Overall, our atmospheric isotope constraints are free of a methodologically caused krypton artifact perturbing previous studies, leading to generally lower δ¹³CH₄ constraints and hence, lower GEM/BB estimates in our study.

Table S2.

Ranges of isotopic source signatures and source strengths used as model input

Source category	δ13C-CH4 minimum (‰ wrt VPDB)	δ13C-CH4 maximum (‰ wrt VPDB)	δD-CH4 minimum (‰ wrt VSMOW)	δD-CH4 maximum (‰ wrt VSMOW)	Global source strength minimum (Tg CH4 a−1)	Global source strength maximum (Tg CH4 a−1)
Standard model setup
Interglacials
Microbial source (south)	−62.4	−60.4	−332.8	−292.8	50	250
Microbial source (north)	−63.6	−61.6	−343.1	−303.1
BB	−24.7	−20.0	−255.0	−195.0	0	60
GEM	−44.9	−43.2	−205.0	−165.0	0	120
Glacials
Microbial source (south)	−59.7	−57.7	−332.8	−292.8	20	150
Microbial source (north)	−60.9	−58.9	−343.1	−303.1
BB	−22.0	−17.3	−255.0	−195.0	0	60
GEM	−44.9	−43.2	−205.0	−165.0	0	120
Adjusted microbial source for sensitivity runs (assuming no IPD of [CH4])
Interglacials
Microbial source (south)	−62.3	−60.3	−332.8	−292.8	50	250
Microbial source (north)	−64.0	−62.0	−347.3	−307.3
Glacials
Microbial source (south)	−59.6	−57.6	−332.8	−292.8	20	150
Microbial source (north)	−61.3	−59.3	−347.3	−307.3

For every Monte Carlo model run, values for δ¹³CH₄, δD(CH₄), and the source strengths of each of three model source categories have been randomly picked from the given intervals. wrt, with respect to. The Matlab function unifrnd() has been used to generate continuous uniform random numbers. References and adjustments for the given isotopic signatures are described in detail in SI Text (refs. 13, 22, 32, 36, and 63 and references therein). Note that, concerning the microbial source, different isotopic signatures for the Southern Hemisphere and Northern Hemisphere are used. All source strengths are given as global values and distributed between the hemispheres according to information given in the text. Note the changed isotopic signatures for the microbial source for the sensitivity runs that assume no IPD of CH₄.

Table S3.

Model sink fractions, isotopic fractionation factors, and hemispheric distribution

Sink	Sink fraction minimum (%)	Sink fraction maximum (%)	Fractionation factor ε for δ13CH4 (‰)	Fractionation factor ε for δD(CH4) (‰)	Northern Hemisphere fraction of sink (%)
Tropospheric OH	78.8	84.8	−5.4	−231	50
Stratospheric loss	7.6	11.0	−22	−80	50
Soils	3.9	5.6	−12	−160	73.8
Tropospheric Cl	3.4	5.0	−60	−470	43.1

For every Monte Carlo model run, sink fractions of the individual sink processes have been randomly picked from the given intervals to account for uncertainties in the sink apportionment. Fractionation factors and hemispheric distribution have been kept constant for all model runs.

SI Text

Corrections of Isotope Data to Address Fractionation in the Firn Caused by Diffusion.

Air in firn is subject to diffusion (140). Accordingly, we applied corrections to our measured ice core isotope data as explained in the following two sections.

Correction of δ¹³CH₄ and δD(CH₄) data because of gravitational settling in the firn.

Gravitational settling creates a gradient with heavier isotopologues accumulating at the bottom of the diffusive zone. For the heavier isotopes, an enrichment in the range of 0.2–0.5‰ per mass difference is observed in air bubbles of an ice core and can be corrected for using δ¹⁵N₂ measurements (141, 142). Because only the mass difference is decisive, this approach holds for both δ¹³CH₄ and δD(CH₄). For δ¹³CH₄, we used interpolated δ¹⁵N₂ records from EDC (51, 136, 143), TALDICE (144), and Vostok (145). Concerning δD(CH₄), we used the same procedure and datasets for EDC (51, 136, 143). However, for EDML, no complete dataset covering all our samples is available, and we used the mean value (0.44‰) of all EDML δ¹⁵N₂ values given in ref. 146. Note that the glacial/interglacial difference in the last mentioned dataset is only about 0.05‰, and hence, the error introduced because of this simplified approach is much smaller than other measurement uncertainties for δD(CH₄). All δ¹³CH₄ and δD(CH₄) data presented in the figures of this contribution have been corrected for gravitational settling in the firn.

Correction of δ¹³CH₄ data because of diffusive isotopic fractionation in the firn.

Isotopologues of a trace gas species (e.g., δ¹³CH₄) have a different diffusion constant; hence, if a concentration gradient between the free atmosphere and the bottom of the firn is present, the isotopic signature of the original atmospheric signal changes while it is carried down to the lock-in depth. The phenomenon called diffusive fractionation is described in detail in ref. 129. The authors also provide the mathematical tools to quantify the effect. Using this approach allowed us to calculate the diffusive fractionation correction for our δ¹³CH₄ data.

The diffusive column height for each data point was calculated using interpolated δ¹⁵N₂ records from EDC (51, 136, 143), TALDICE (144), and Vostok (145). The CH₄ mixing ratio and its annual changing rate have been determined using a spline approximation [1,000-y cutoff frequency (147)] of the EDC CH₄ data (4). Other than the values given in ref. 129, we further used estimates for the mean annual site temperatures [EDC: −54 °C (129), TALDICE: −41 °C (148), and Vostok: −55 °C (134)] and the mean annual surface air pressure [EDC: 694 mbar (129), TALDICE: 721 mbar (149), and Vostok: 624 mbar (150)] to calculate the diffusion fractionation correction. All new δ¹³CH₄ data presented in the figures of this contribution have been corrected for diffusive isotopic fractionation in the firn. The effect for all of our samples is small, because we did not measure δ¹³CH₄ samples from time periods with rapidly changing CH₄ mixing ratio. The changes of the measured values are between −0.18‰ and +0.08‰, hence not relevant for any of our conclusions.

The effect is of the same size for δD(CH₄). Because the uncertainties of this parameter are larger, we choose not to correct our δD(CH₄) data for diffusive isotopic fractionation in the firn.

Definition of Time Slices for “Typical” Glacial and Interglacial Levels.

We defined time slices that are intended to represent “typical levels” within our isotope records. These time slices are used to quantitatively describe glacial/interglacial amplitudes of [CH₄], δ¹³CH₄, and δD(CH₄) and assess the shift of both isotopes for the Holocene and the LGM compared with earlier interglacials and glacials, respectively. The time interval definition in Fig. S1 uses data for glacial maxima characterized by CO₂ concentrations below 210 ppm and CH₄ concentrations below 420 ppb (4, 110). Similarly, data used for interglacial periods have CO₂ concentrations above 260 ppm and CH₄ concentrations above 500 ppb (4, 110). The time periods used are illustrated in Fig. S1, where red and blue shading highlights data for interglacial and glacial time slices, respectively. In Table S1, we summarize [CH₄], isotope mean and median values of the used time slices.

Note that the intention of this definition is only a simplified first-order view of the presented records, whereas our data show variations within the chosen time slices rather than stable levels. Note that the time interval with the most ¹³C-depleted values between 115 and 120 ka BP is not included in the MIS 5.5 typical level, because [CH₄] is already below typical interglacial levels. This δ¹³CH₄ minimum is discussed as a special feature in the text.

Box Model to Constrain the CH₄ Budget.

Natural methane sources can be differentiated according to their isotopic signatures. For instance, BB and GEM emit relatively isotopically heavy methane (i.e., enriched in ¹³C and deuterium compared with the averaged source mix). However, the largest natural source type—microbial emissions largely from wetlands—emits methane with an isotopic fingerprint slightly lower in δ¹³CH₄ and δD(CH₄) than the source mix. In addition, isotopic fractionation by the sinks leads to heavier methane in the atmosphere compared with the emissions.

Box model: Setup.

To constrain the (isotopic) CH₄ budget, we used the box model presented in refs. 32 and 36, which consists of four boxes (northern and southern troposphere and stratosphere) with prescribed air mass exchange. This model allows one to assess maximal GEM and increased emissions by BB for the Holocene and the LGM compared with previous interglacials and glacials, respectively. To this end, CH₄ sources (both emission strengths and isotopic source signatures) (Table S2) and relative sink contribution (Table S3) are varied within prescribed ranges in the model (refs. 13, 22, 32, 36, and 63 and references therein). The model is run into steady state, and the equilibrium value of the southern tropospheric box is compared with our data constraints (Table S1) for each of the time slices. If the modeled [CH₄], δ¹³CH₄, and δD(CH₄) values are compatible within the uncertainty (Table S1) with the data constraint, the emission values are recorded as a possible CH₄ budget solution. For each time slice, 10,000 valid runs have been collected.

Box model: Sources.

To be consistent with recent work on the current methane budget (63), we distinguish only three source categories here: GEM, BB, and a microbial source including natural sources, such as wetlands, termites, and naturally occurring ruminants. Isotopic signatures of the sources are based on previous work (refs. 13, 32, and 36 and references therein) and the collection in ref. 63 with some modifications. Table S2 shows the δ¹³CH₄ values used for the microbial source, BB, and GEM according to ref. 63 with some modifications (see below) and δD(CH₄) values (refs. 13 and 22 and references therein).

For the interglacial δ¹³CH₄ source signatures of the microbial source, BB emission, and GEM, we used the global value given in ref. 63 with two modifications for the microbial source. (i) The modern microbial source used in ref. 63 includes anthropogenic emissions from ruminants and landfills, which bias the natural isotopic source signature. As we investigated the natural budget in this contribution, we removed the waste CH₄ source (which leads to a heavier source mix signature) and decreased emissions from ruminants, which at present largely reflect livestock in ref. 63, by 80 ± 20% (which leads to a lighter source mix signature). In the end, these adjustments essentially canceled each other and led to a global mean microbial source signature that is the same as the recent value given in ref. 63 within the error limits. Accordingly, we used a global mean microbial source signature of −62.2 ± 1.0‰ in our model. (ii) The value given in ref. 63 represents a global average; however, previous work (refs. 32, 35–37, 75, 151, and 152 and references therein) has shown that high-latitude wetland sources are depleted in ¹³C and deuterium. Because the high-latitude wetland sources are mainly located in the Northern Hemisphere, a difference in the isotopic source signatures of the two hemispheres is observed. To account for this difference, we used the geographically resolved wetland emission estimate in ref. 10 and the published data on the IPD of the methane mixing ratio (9, 79, 89, 132) to assess how much of the total wetland emissions are located in the Northern Hemisphere. Aggregating the distribution in ref. 10 into three categories (tropical south, tropical north, and boreal north) leads to a northern fraction of wetland source emissions of 0.6. However, our model runs are consistent with the IPD constraint (9, 79, 89, 132) only when using a northern fraction of 0.7 for the microbial source. This higher ratio is achieved by shifting 17 Tg CH₄ a⁻¹ from the southern tropical to the northern tropical region compared with ref. 10. We stress that this small difference is not in conflict with the estimate in ref. 10, which assessed the present day emissions dominated by the anthropogenic sources. In a next step, we assigned δ¹³CH₄ and δD(CH₄) source signatures for the boreal (high-latitude) source, which are 6 and 50‰ lower, respectively, than the low-latitude sources (refs. 32, 35–37, 75, 151, and 152 and references therein). Using the global source signature value in ref. 63 (see above), this adjustment led to a Northern Hemisphere microbial source signature of −62.6‰ and a Southern Hemisphere microbial source signature of −61.4‰ in δ¹³CH₄. For δD(CH₄), we use −323‰ for the Northern Hemisphere and −313‰ for the Southern Hemisphere for the microbial source signature in our model (Table S2). These numbers lead to global source signatures of our microbial source of −62.2 and −320‰ for δ¹³CH₄ and δD(CH₄) as proposed in refs. 63 and 13, respectively.

As shown in refs. 25 and 34, isotopic signatures of biogenic sources (microbial and BB) may have changed on glacial/interglacial timescales. According to ref. 25, we estimate that about one-half of the glacial/interglacial amplitude of δ¹³CH₄ is caused by environmental changes leading to source signature changes, whereas the other one-half is accounted for by source mix changes. Hence, we attributed one-half of the averaged δ¹³CH₄ differences of MIS 12 minus MIS 11 and MIS 6 minus MIS 5 (Table S1) to a δ¹³CH₄ shift of our microbial and BB source signatures and used these shifted source signatures in our model for glacial time slices. Accordingly, the microbial and the BB δ¹³C source signature ranges for glacial time slices are heavier (higher δ¹³CH₄) in our model by 2.6‰ compared with interglacial runs (Table S2).

The geographic distribution of the emissions (i.e., the fraction that is emitted into the northern model troposphere compared with the Southern Hemisphere) has been adjusted, such that the modeled IPD of [CH₄] is in line with data constraints for the Holocene and the LGM (9, 79, 89, 132). Our best estimates for both interglacial and glacial model setups are northern emissions fractions of 0.7 for the microbial source, 0.7 for BB, and 0.6 for GEM. Note that no IPD information is available for the oldest four time slices used in this study, because no or no reliable atmospheric CH₄ values are available for the Northern Hemisphere from Greenland ice cores. Hence, we used the same source distribution for older time slices (identical input parameters for the older interglacials, MIS 5.5 and MIS 11.3, as for the Holocene and identical input parameters for the older glacials, MIS 6 and MIS 12, as for the LGM).

To exclude an overestimation of the emissions into the Northern Hemisphere, we additionally performed sensitivity runs where we assumed no latitudinal difference in the CH₄ emissions (i.e., northern emission fractions of 0.5 for all three source categories). Because an IPD is evident at present and persistent for the last 25,000 y (9, 79, 89, 132), this exercise is considered a minimum conservative endmember for the true hemispheric source distribution. Accordingly, we expect true BB/GEM to be between the results of our best guess model runs and these sensitivity runs. For the sensitivity runs, identical emission values (Table S2) and targets (Table S1) have been used with one exception. The calculation of isotopic source signatures according to the procedure described above yielded slightly higher numbers for the microbial source for both δ¹³CH₄ and δD(CH₄), because 15 Tg CH₄ a⁻¹ had to be shifted from the northern tropical to the southern tropical region compared with the distribution given in ref. 10. The changed isotopic signatures of the microbial source for the sensitivity model runs are given in Table S2.

Krypton measurement artifacts and consequences for the CH₄ budget.

We stress that our new ice core δ¹³CH₄ and δD(CH₄) measurements presented in this study are free of the krypton (Kr) measurement artifact described in refs. 46–48, which arises if Kr interferences are not excluded during the mass spectrometric (MS) analyses. Previous studies relied on measurements including a Kr effect, which leads to biased δ¹³CH₄ values that are higher. Specifically, we note that the assessment in ref. 63 is based on Greenland ice core data for the Late Holocene in ref. 64, which were not corrected for a Kr interference during the measurement. Using the differences of Sapart’s (64) δ¹³CH₄ values with and without Kr effect for standard air bottles with different [CH₄]/Kr ratios given in table 2 of ref. 48, we can roughly assess the bias of the Late Holocene δ¹³CH₄ dataset. For this exercise, we assume a mixing ratio around 700 ppb (64). Furthermore, assuming no additional amount dependence of δ¹³CH₄ in this dataset (which could amplify or dampen the observation), we calculate an offset of these data of approximately +1.15‰. If true, the Late Holocene atmospheric value was lower by this amount. As a consequence, lower δ¹³CH₄ constraints lead to lower BB and/or GEM estimates compared with the assessment in ref. 63. Using our observed differences of older time periods (Table S1) to scale emissions based on Fig. 2 leads to a reduction of GEM by roughly 6 Tg CH₄ a⁻¹ compared with the value given in ref. 63. Note that the total GEM estimate is also strongly dependent on the assumed BB emissions as discussed in the text.

Box model: Sinks.

The atmospheric lifetime of methane has been kept constant at 8 y for all performed runs. This approach is in line with several atmospheric chemistry modeling studies, which show little change in the overall lifetime of methane for different climate periods (40–42, 153). In our revised box model, four sinks are implemented (OH oxidation in the troposphere, methanotrophy in soils, a stratospheric sink, and a Cl sink in the marine boundary layer) with fixed fractionation factors (Table S3) (13, 154–156). Note that these sinks differ greatly in their fractionation factors. Accordingly, although the overall lifetime may have stayed constant over time, a change in the relative fraction of each sink to the total lifetime may have had an impact on the isotopic composition of the atmosphere. Moreover, the effect of such sink contribution effects would have a different impact on δ¹³CH₄ and δD(CH₄). Atmospheric methane removal through the four sink processes in our model is scaled according to ref. 10; however, in our Monte Carlo approach, we allow the relative contributions of each sink to vary within certain limits to account for uncertainties in our understanding of the sink attributions (10). Accordingly, the fractional sink of each sink process (in Tg CH₄ a⁻¹) was varied independently by ±15%. The sum of all individual sink contributions in each Monte Carlo run was then scaled to balance the total emission for the lifetime of 8 y required to obtain the targeted CH₄ mixing ratio. The resulting ranges of the relative contribution of each sink are given in Table S3.

Also, the sink contributions are different in both hemispheres. The hemispheric distribution of the model sinks is constant for all runs and summarized in Table S3. The tropospheric and stratospheric sinks are split half and half between the northern and southern boxes. The soil sink and the marine chlorine sink are distributed according to the difference in land and ocean coverage in both hemispheres. We choose a partitioning according to the Global Land Cover Facility with unevenly distributed northern fractions of 0.74 and 0.43 for the soil and marine chlorine sinks, respectively (Table S3) (157, 158).

Box model: Results.

Contrary to our previous work (32, 36), we do not present normalized probability density functions (nPDFs) for the box model results. This change is because of the fact that using nPDFs is misleading and suggests a likelihood for emission strengths of different sources, which in fact, is an artifact of the Monte Carlo process. Without additional knowledge, each accepted box model solution in line within the uncertainties with the (isotopic) data constraint is equally likely. For example, the higher numbers of accepted model solutions in ref. 32 for short lifetimes in the glacial do not imply that a shorter lifetime is more realistic. Instead, they reflect only that, in the box model approach, where all parameters are varied independently, it is much easier to achieve low glacial CH₄ concentrations by reducing only one parameter, the lifetime, instead of reducing the emissions of all source types at the same time by the correct amount, while not violating the (isotopic) CH₄ budget. The only valid information that should be drawn from the model results is the field of possible emission strengths for each of the given sources. Accordingly, we can exclude all scenarios that do not fulfill the ice core constraint. In this study, we avoid these pitfalls of nPDFs but go beyond previous Monte Carlo approaches by analyzing the accepted model runs to find functional relationships between the emissions of different source types (Fig. 2 and Fig. S7). For example, Fig. 2 shows the variation of possible GEMs under given BB emissions. Similarly, one can derive information for the microbial source for given numbers of BB and/or GEM as shown in Fig. S7 A and B. For instance, selecting 25 Tg CH₄ a⁻¹ for BB and 30 Tg CH₄ a⁻¹ for GEM during interglacials, the microbial source strength is roughly between 150 and 230 Tg CH₄ a⁻¹.

The uncertainty of the isotopic signature of the sources and the fractions of the sinks is intrinsically implemented in the model results, because we allowed the Monte Carlo process to pick values from broad ranges of isotopic source signatures (Fig. S7 C and D and Table S2) and the sink fractions (Table S3). Fig. S7 C and D shows that the model favors lower emissions from the microbial source for the Holocene and the LGM compared with older time slices, because the heavier isotope targets are more easily achieved by higher emissions of BB and/or GEM.

Results for the sensitivity analysis on the hemispheric distribution of sources are presented in Fig. S7 E and F. It is not straightforward to reliably quantify the IPD in the past, but it is clear that, for a minimum (zero) IPD in [CH₄], our BB and GEM maximum estimates are not significantly higher compared with the standard model runs.

Note that, for emissions from marine clathrates (and generally, submarine GEM), which are part of our GEM model source, our scenarios do not take into account isotopic fractionation during oxidation in the water column. This process leads to heavier (higher) numbers in both isotopes for methane reaching the air/sea interface (27, 29, 31, 36, 39, 54, 159). Assuming heavier fingerprints for clathrate emissions into the atmosphere reduces their share to the global budget. The same is true if a heavier carbon isotopic signature is assumed for some sources of GEM as listed in table 1 of ref. 29.

Supplementary Graphs.

An overview of all data presented in this study is shown in Fig. S1, which highlights the time intervals used to determine the targets for the box model runs. We underline that Fig. S1 presents additional Bern δ¹³CH₄ data between 25 and 80 ka BP, which are not presented in Fig. 1 and have been partly shown in our technical article on the measurement system (47). Furthermore, we present the data shown in Fig. 1 in enhanced and zoomed versions. First, the interglacial periods are highlighted (Fig. S2) and set into context with a recently published speleothem monsoon proxy record (70). Second, we zoom into three presented time intervals: LGM/Holocene, MIS 6/MIS 5.5, and MIS 12/MIS 11.3 (Figs. S3–S5). Finally, we present a figure comparing previously published data to our records (Fig. S6).