Enrichment in 13 C of atmospheric CH 4 during the Younger Dryas termination

Introduction Conclusions References


Introduction
Ice core records from both Antarctica and Greenland demonstrate numerous abrupt climate changes throughout the Quaternary (Brook et al., 2000).One such event, the Younger Dryas-Preboreal transition (YD-PB) [∼ 1.6 kiloyears before 1950 AD; ka.BP], saw a rapid climate change leading a large increase in the atmospheric methane mixing ratio ([CH 4 ]) (Severinghaus et al., 1998).This increase could have been driven by changes in either the methane sources or sinks.However, modelling studies investigating interglacial-glacial changes in the predominant CH 4 tropospheric sink, the OH radical, indicate only a small variability in sink strength (Thompson et al., 1993;Martinerie et al., 1995).Interpretation of the YD-PB CH 4 record has thus focused on methane's natural sources with the debate centering on the roles of tropical wetlands (Chappellaz et al., 1990;Brook et al., 2000), marine gas hydrates (Kennett et al., 2003), and thermokarst lakes (Walter et al., 2007).
Wetlands are the most important natural methane source, accounting for approximately 75 % of global emissions in the Preindustrial Holocene (PIH) (Chappellaz et al., 1993;Wuebbles and Hayhoe, 2002); thus, the wetland hypothesis purports that an invigorated hydrologic cycle with higher precipitation allows expansion of wetland area, and increased wetland methane emissions, particularly in the tropics (Brook et al., 1996).
The clathrate (methane hydrate) release hypothesis comprises several scenarios.One contends that marine clathrates situated on the continental margins are capable of episodic destabilization events triggered by the warming of the upper thermocline waters.This scenario assumes that the majority of released methane passes through the water column, without oxidation, to the atmosphere (Kennett et al., 2003).A recent update on this hypothesis, based upon preserved tar records in sediments from the Santa Barbara basin, proposes that methane hydrates may act as a "climate sensitive valve system for thermogenic hydrocarbons", i.e. clathrate release allows for the increased release of methane from hydrocarbon seeps (Hill et al., 2006).
Thermokarst lakes have been shown to release very large amounts of methane from point-source locations with hotspots that produce up to 50 g CH 4 m 2 yr −1 (Walter et al., 2006).The large amount of available carbon (Zimov et al., 2006) and the high methane emission potential make thermokarst lakes a potential player in past abrupt climate changes (Walter et al., 2007).
The stable and radiogenic isotope ratios ( 13 C/ 12 C, D/H, 14 C/ 12 C) of atmospheric methane trapped in glacial ice can be used to probe the dynamics of methane's sources and sinks (Schaefer et al., 2006;Sowers, 2006;Fischer et al., 2008;Petrenko et al., 2009) and distinguish between these competing hypotheses.Primary methane sources have characteristic isotope signatures, although in some cases with similar and/or broadly ranging values for a single isotope system (Whiticar, 1999).However, the combination of all three methane isotopes overcomes many of their individual limitations, and is used here to investigate the dynamics behind the CH 4 flux increase during the YD-PB.

Sample collection and analysis
Ice samples were collected using un-lubricated chainsaws from the western margin of the Greenland ice sheet at Påkitsoq (69 • 825.830N, 50 • 815.200W) during the 2003-2005 sampling seasons.The start and end points of the Younger Dryas-Preboreal transition (YD-PB) in a sampling profile were determined in the field using gas chromatography with a flame ionization detector (GC-FID) analysis of [CH 4 ].This was later confirmed on parallel samples (collected along strike of the layers) by GC-FID [CH 4 ] analysis at Oregon State University (OSU), USA and δ 15 N and δ 18 O atm analysis at Scripps Institution of Oceanography, USA (Petrenko et al., 2006).The strike and dip of the ice was determined by visual cues as well as laboratory analysis (Petrenko et al., 2006).Ice from the actual YD-PB transition period is confined to 50 cm width and is directly accessible at the surface for precision sampling.Sample width was between 3 and 5 cm perpendicular to the strike of the profile.The YD-PB transition was resolved in great detail with ca.6 to 8 age horizons, and 1 to 5 replicates sampled at each age horizon per field campaign.Replicate samples were taken along the strike of the layers.Samples were shipped frozen to the University of Victoria (UVic), Canada, or OSU.
In the laboratory, we measured the methane stable carbon isotope ratio (δ 13 CH 4 ; referenced to Vienna Pee Dee Belemnite -VPDB) from occluded gas in 61 individual shaved ice samples (ca.100-200 g).Measurements were made via an improved wet-extraction, isotope ratio mass spectrometer (IRMS) method with analytical precision < 0.3 ‰ and no introduced error or offset due to the wet extraction procedure (Melton et al., 2011).The main improvements to the procedure of Schaefer et al. (2006) and Schaefer and Whiticar (2007) include (i) a He carrier gas pre-scrubbing trap, (ii) introduction of two carbon monoxide (CO) traps, and (iii) post-combustion trapping of the methane to produce a high amplitude sample peak, increasing the signalto-noise ratio of the sample.The blank contribution was, at most, ∼ 5 % of the sample signal, while more commonly < 3 %.Blanks were measured every two to four samples and consistently yielded a contaminant value with atmospheric δ 13 CH 4 .Following Schaefer and Whiticar (2007), sample δ 13 CH 4 was corrected in an isotope mass balance approach using the CH 4 concentration based upon m/z 44 peak height.
Atmospheric air tests were routinely measured at the start and end of an ice sample measurement day, as well as regularly between ice samples across a period of three years (2006, 2007, and 2009).These atmospheric air tests showed a mean δ 13 CH 4 value of −47.51 ± 0.29 ‰ (1σ ; n = 48; methane content varying between ∼ 160-830 pmol).The smallest atmospheric air samples measured (∼ 160 pmol) did not show significantly different values and no loss in precision compared to the total dataset (−47.50 ± 0.18 ‰; 1σ ; n = 5) (Melton et al., 2011).The Påkitsoq sample with the lowest CH 4 content measured contained ∼ 240 pmol CH 4 , which is well within the range measured by the atmospheric air samples.Additionally, samples of artificial reference ice were measured on most sample measurement days with a mean δ 13 CH 4 value of −47.51 ± 0.29 ‰ (1σ ; n = 32; methane content varying between ∼ 415-1080 pmol).These artificial reference ice measurements (Fig. 1) and the atmospheric air samples (Melton et al., 2011) show good linearity over the range in CH 4 content found in Påkitsoq samples, as well as over the range in sample mass for reference ice samples.The precision of our method is a significant improvement from Schaefer and Whiticar (2007) as these same tests yielded reported standard deviations of ± 0.47 ‰ and ± 0.52 ‰ for atmospheric air and artificial reference ice, respectively.As well, the calculated standard deviation across our 3-yr atmospheric air and artificial ice sample dataset is slightly greater than the true analytical standard deviation because of seasonal changes of the local atmospheric δ 13 CH 4 used as the reference air.The seasonal change in δ 13 CH 4 is ± 0.11 ‰ on the nearby Olympic Peninsula during the period from 1988 to 1995 (Quay et al., 1999).However, due to unknown offsets in δ 13 CH 4 between Victoria and Olympic Peninsula, we do not correct for inter or intraannual variability.
Tests were also performed on the remaining pieces of a shallow section of the Greenland Ice Sheet Project II (GISP2) Core #139 used in Schaefer and Whiticar (2007) and discussed in Melton et al. (2011).The quality of these samples is poor, with much of the ice highly fractured.The mean δ 13 CH 4 of the GISP2 ice is −49.37 ± 0.58 ‰ (1σ ; n = 5, corrected for gravitational fractionation) for a gas age of ∼ 225 ± 20 yr BP.Recently, Sowers (2010) published a δ 13 CH 4 record using GISP2 ice, reporting a δ 13 CH 4 of −49.24 ± 0.17 ‰ (1σ ; n = 3) for ice with a gas age of ∼ 200 yr BP.The good agreement of mean values is an indicator of our accuracy, although our standard deviation is much higher than Sowers (2010).However, as fractured ice has been reported by multiple investigators to lead to spurious values (Craig et al., 1988;Schaefer and Whiticar, 2007;Behrens et al., 2008), we view our artificial ice results as more indicative of our analytical method precision than those of the GISP2 ice.Additionally, there is no indication that the artificial ice samples would somehow lead to better precision than natural ice samples given that Schaefer and Whiticar (2007) reported poorer standard deviation for their artificial ice samples (± 0.52 ‰) than for their GISP2 samples (± 0.32 ‰).We note that Schaefer and Whiticar (2007) used higher quality parts of the GISP2 core section, so that the difference in precision to our study does not necessarily reflect on the quality of the new technique.
As Påkitsoq ice can contain anomalous [CH 4 ] (Schaefer et al., 2009), samples were filtered for anomalous methane concentrations.The assessment uses two criteria for exclusion: (i) samples that showed visible post-depositional features in the ice, such as fractures; layers or occlusions of dust; or samples with bubble-free bands in the ice (Petrenko et al., 2006), and (ii) samples that had IRMS-measured [CH 4 ] values more than 100 ppb different from the contemporaneous GISP2 [CH 4 ] (Brook et al., 2000).Sample [CH 4 ] for this data filter is derived by calculation against calibration curves from reference gases (serial dilutions of CH 4 gas in N 2 ) and air standards (1033 ppb methane (balance ultra zero air) purchased from Airgas Incorporated and calibrated at Washington State University against standard tanks previously calibrated at NOAA CMDL) (Petrenko et al., 2006) using the IRMS m/z 44 ion peak height (derived from the ice sample 12 CH 4 ) and air content of the ice.The air content of the δ 13 CH 4 ice samples is determined via parallel samples that were measured with GC at OSU as described in Brook et al. (2000Brook et al. ( , 2005) ) from pressure in the sample loop and ice sample volume.The standard deviation for [CH 4 ] measured with the IRMS is taken as the value determined using GISP2 section #139 in our lab.Due to the poor quality of the latter, this is a conservative estimate.The 100 ppb limit for inclusion of samples was selected by summing the magnitude of analytical standard deviation and uncertainty in the reference GISP2 [CH 4 ] (∼ ± 10 ppb) (Brook et al., 2000).This led to the removal of 21 unsuitable samples of the 61 total.For the actual ∼ 150 yr YD-PB transition, only three anomalous samples were removed of the total 46 samples that fall within the gas age limits for inclusion in the weighted linear regression (see Fig. 2).The three suspicious samples were not considered in our subsequent source analysis (Sect.3.3).However, the removal of those samples has no significant effect on the δ 13 CH 4 trend across the YD-PB (Fig. 3).
For replicate measurements (multiple ice samples of equal age), median values with standard deviations are reported.For single measurements, analytical precision as derived from the artificial reference ice measurements is shown (Melton et al., 2011).All raw δ 13 CH 4 values are presented in the Supplement in Table S1.

Corrections to δ 13 CH 4 values
The YD-PB transition δ 13 CH 4 values have been corrected for gravitational, thermal and diffusion fractionation (described below).The δD-CH 4 values of Sowers (2006) in Fig. 2 have been treated similarly.In addition, an isotopic disequilibrium correction is applied to the source reconstructions in order to account for the transient dilution effect on the source signal in the atmospheric reservoir as identified by Tans (1997) (Fig. 4).(Fischer et al., 2008).A weighted loess spline fit to the Greenland data (Påkitsoq and GISP2, shaded region is a 1σ standard error envelope, span 0.75) shows the general trends although it dampens the larger changes in the isotope record.
in the Supplement in Table S2.The European Project for Ice Coring in Antarctica (EPICA) ice core δ 13 CH 4 data from Dronning Maud Land (EDML) shown in Fig. 2 include their own gravitational fractionation correction (Fischer et al., 2008).As these values do not fall directly within the YD-PB, they do not require further correction (see further discussion in Sect.3.2).

Gravitational and thermal fractionation
δ 15 N of N 2 as measured at Scripps Institute of Oceanography (Petrenko et al., 2006) on parallel samples records approximately the same gravitational and thermal fractionation as CH 4 isotopes in the firn column prior to bubble closeoff.For the thermal fractionation component of this correction, we neglect different thermal diffusivities, T , of the isotopes of N 2 and CH 4 in air (Grachev and Severinghaus, 2003).The latter has not been determined experimentally for CH 4 , but T of 13 CH 4 in 12 CH 4 (Stevens and de Vries, 1968) suggests that the difference to N 2 isotopes is on the order of ∼ 12 % (Schaefer, 2005).The thermal component of the δ 15 N anomaly over the YD-PB termination is only ∼ 0.15 ‰ (Severinghaus et al., 1998).Therefore, the error from different T is likely around 0.02 ‰ with an uncertainty of a fraction of that value.

Diffusion fractionation
δ 13 CH 4 and δD-CH 4 values that have a gas age within the YD-PB transition, i.e. during periods of rapid and large atmospheric [CH 4 ] change, require correction for fractionation during firn diffusion processes.Under changing atmospheric concentrations, the atmospheric signal of the heavier methane stable isotopologue will diffuse to the firn closeoff zone more slowly than the light methane stable isotopologue (Trudinger et al., 1997).We quantified correction factors with a firn diffusion model (Schaefer, 2005) based upon other models (Herron and Langway, 1980;Schwander et al., 1993Schwander et al., , 1997) that calculate the effective diffusion coefficients for each isotopologue from physical properties of the firn on dependence of local temperature and accumulation rate.The applied correction factors are substantial (up to 0.7 ‰).However, sensitivity tests for environmental conditions (temperature and accumulation rate) and model input parameters (Schaefer, 2005) show that our observed trend in δ 13 CH 4 values is not an artefact of the diffusion model.The sensitivity tests quantify the absolute uncertainty as less than 0.2 ‰ (Schaefer, 2005).

Isotopic disequilibrium
A rapid change in atmospheric methane concentration and/or stable isotope values causes a temporary imbalance between stable isotope values of the atmosphere and that of the aggregated sources as predicted at steady-state after sink fractionation (Tans, 1997;Lassey et al., 2000).Accounting for isotopic disequilibrium is important if measured atmospheric stable isotope values are used to interpret methane sources during a period of rapid changes.We therefore calculated appropriate corrections using a 2-box atmospheric model (Lassey et al., 2000) that includes source, sink and interhemispheric transport terms.Isotopologues are treated as independent tracers, and the atmospheric isotope ratio is calculated for each time step (1 yr).The model simulates the 150-yr-long transition from a YD to a Preboreal source and sink budget taken from isotope-enabled 4-box atmospheric methane model simulations (Melton, 2010).Discrete correction factors are calculated for each data point by matching the corresponding [CH 4 ] values in the modelled and observed [CH 4 ] increase.The correction is small with maximum values of 0.09 ‰ for δ 13 CH 4 and 2.67 ‰ for δD-CH 4 (Table S2 in the Supplement).

Systematic offset between datasets
The values of Schaefer et al. (2006) have been adjusted for this study to account for a systematic offset from our new measurements.Measurements of outside air at UVic by the method of Schaefer et al. (2006) as described in Schaefer and Whiticar (2007) (−47.33 ± 0.47 ‰) and our method (−47.51 ± 0.21 ‰) (Melton et al., 2011) show a 0.18 ‰ offset.Schaefer and Whiticar (2007) also reported a 0.18 ‰ offset from a high-precision dataset of clean air measured on the nearby Olympic Peninsula (Quay et al., 1999), while our method showed no offset.Despite the fact that all three datasets were measured at different times and the seasonal δ 13 CH 4 cycle introduces uncertainty, we take this as an indication that the values of Schaefer et al. (2006) have to be adjusted to remove the offset through a simple linear addition.One anomalous point from the Schaefer et al. (2006) dataset has been excluded from the linear regression applied to the δ 13 CH 4 values during the YD-PB (Figs. 3 and 4).This is because a δ 13 CH 4 shift of that magnitude and speed is not possible due to firn diffusion processes.

Påkitsoq age scale
Field measurements of [CH 4 ] (Brook et al., 2000) were used to correlate the basic stratigraphy to that of GISP2 for ice sampling (Petrenko et al., 2006).To establish the age of the air bubbles within the Påkitsoq ice, we match geochemical records measured in Påkitsoq ice to those of well-dated ice cores from locations that are comparable in geography, as well as temperature and accumulation rate, to the snow deposition zone for Påkitsoq ice.Four reference records are used to determine the age scale of the Påkitsoq ice: (i) δ 15 N of atmospheric N 2 (Severinghaus et al., 1998), (ii) δ 18 O ice from the ice matrix (Grootes and Stuiver, 1997) et al., 2000), all from GISP2, Greenland, and (iv) δ 18 O atm of atmospheric O 2 from Siple Dome, Antarctica (Severinghaus et al., 2009).The δ 18 O atm value is globally well-mixed, and has been measured with high temporal resolution and analytical precision (Severinghaus et al., 2009).This dataset (from Siple Dome, Antarctica) is chosen over the current GISP2 dataset (Bender et al., 1999), due to its greater precision and inclusion of a gas-loss correction.
δ 15 N is the most consistent parameter between GISP2 and Påkitsoq, implying similar temperature and accumulation rates between the sites.This close correlation is found across all years and sampling locations (Melton, 2010).Thus, this parameter is relied upon extensively for age assignments.Due to the much improved δ 18 O atm record, this parameter is used more extensively in this work than previously (Petrenko et al., 2006;Schaefer et al., 2006).The Siple Dome record (Severinghaus et al., 2009) has a very high precision (pooled standard deviation after gas-loss correction of ± 0.012 ‰); however, the Påkitsoq measurement precision is not as good (± 0.028 ‰) (Petrenko et al., 2006).Therefore, the δ 18 O atm values are used as a secondary constraint on age tie-points set by the δ 15 N data.The δ 15 N data have peaks with similar δ 15 N values on both sides of the peak, and thus two possible ages.The δ 18 O atm values are then used to provide a secondary constraint of the age.The inflection points in the [CH 4 ] record provide excellent tie-points for the onset and end of climatic transitions.For example, they indicate the start (together with a peak in δ 15 N) and the end of the YD-PB.We use Påkitsoq [CH 4 ] data with high analytical precision measured by GC-FID at OSU (Petrenko et al., 2006) for the correlations.The δ 18 O ice parameter is used when the other parameters do not exhibit sufficiently unique features for an age determination.To account for the ice thinning and folding, the age scale is variable along the sampling profile.Fourteen age tie-points were used to create a continuous age scale (Table S3 in the Supplement) with linear interpolation between tie-points (Melton, 2010) for the 2001 sampling season.This 2001 sampling season age scale was then adapted for changes in the ice due to surface melt each sampling season on the basis of shifts in the horizontal positions of the geochemical markers referenced to the permanent markers left in the ice.Age uncertainty close to, and within, the YD-PB is estimated to be better than 1 % absolute uncertainty (Petrenko et al., 2006).
The absolute gas age scale (GISP2 depth to calendar age) used is that of Schaefer et al. (2006).This age scale differs from the common GISP2 gas age scale of Brook et al. (2000) by fixing the date of the YD termination to 11 570 ± 0 yr BP on the basis of tree ring records (Friedrich et al., 1999).All literature datasets presented in this study are converted to this age scale.The EDML dataset was converted to the Påkitsoq timescale by performing a linear regression on both of the GISP2 and EDML [CH 4 ] records during only the abrupt [CH 4 ] increase, and adjusting the EDML gas ages to allow the lines to overlap.This fitting and placement on the Påkitsoq timescale required an addition of 134 yr (no compression or expansion of the timescale was performed) to the GICC05 age of EDML points.
As both sample thickness and the gas age distribution due to diffusion and bubble enclosure contribute to an age range per sample, our per sample estimated range in age is ∼ 25-35 yr.

Results
The δ 13 CH 4 data are plotted in Fig. 2 on a common age scale with previous Påkitsoq data (Schaefer et al., 2006).The Schaefer et al. (2006) Påkitsoq dataset includes ice with an uneven distribution of gas age from each of the 2001-2003 sampling seasons (particularly at the start of the YD-PB) (Schaefer et al., 2006).In contrast, the combination of our dataset and the Schaefer et al. (2006) dataset now provides replicate samples over multiple field campaigns (2001)(2002)(2003)(2004)(2005) for each part of the record to minimize sampling biases and analytical uncertainty.The combined record contains 43 individual samples with gas ages within the ∼ 150 yr YD-PB.
Comparing the combined Påkitsoq δ 13 CH 4 dataset to the published records from the EDML core (Fischer et al., 2008) and GISP2 (Sowers, 2010) shows good agreement in the Preboreal period with all records showing a general pattern towards more 13 C-depleted values as the PB progresses (Fig. 2).Looking at the YD period, there is a greater divergence between the Påkitsoq and EDML records (Fischer et al., 2008) with the Påkitsoq values generally more 13 Cdepleted (and possibly with higher scatter than in the PB period).This could indicate that the inter-hemispheric δ 13 CH 4 gradient was greater in the YD than in the PB period.However, any interpretations of the magnitude of offset between datasets should be tempered by uncertainties due to (i) higher uncertainties in the age-scales outside the rapid transitions, and (ii) the inter-laboratory offset between the measurement labs.The inter-laboratory offset could be influenced by several factors including blank corrections, standard gases, and instrumental drift corrections.However, an inter-laboratory offset would result in a constant offset that influences the magnitude of the offset between datasets, but not the general pattern of each dataset.As a result, we will limit our discussion to the general pattern displayed by each dataset and not attempt to interpret the magnitude or changes in offset between datasets.

Observed Trend in δ 13 CH 4 values across the YD-PB
Throughout the late YD and early PB periods, while [CH 4 ] was stable, there is a trend of 13 CH 4 depletion (Fig. 2) that has been observed previously (Fischer et al., 2008).In contrast, the fast YD-PB [CH 4 ] rise coincides with strong enrichment in 13 CH 4 that reverses the long-term 13 CH 4 evolution.Outside of the transition, there appears to be relatively large movements in the δ 13 CH 4 values without a corresponding large change in atmospheric [CH 4 ].This phenomenon has been observed in other studies (Bock et al., 2010;Möller et al., 2012).Changes in δ 13 CH 4 values without significant changes in atmospheric [CH 4 ] could be caused by changes in the ratios of the different sources and sinks or even changes in the characteristic isotope values of the sources and sinks themselves due to changing environmental conditions or some unknown processes.While these changes are of interest, for this paper, we will focus our interpretation on the termination of the YD-PB as this period has a large increase in methane concentration coincident with dramatic changes in climate.We use the δ 13 CH 4 increase to delineate the driving processes of [CH 4 ] rise at the YD-PB.We focus on the methane budgets at the start and end of the [CH 4 ] rise to (i) minimize uncertainties due to fractionation that gases and their isotopologues undergo as they move through the unconsolidated snow (firn) prior to bubble close-off, and (ii) avoid transient signals, such as diffusional smoothing (see Sect. 2.2.2.).The change in δ 13 CH 4 across the YD-PB ( δ 13 CH 4 ) shows a mean 13 C-enrichment of 1.5 ‰ for the combined dataset as determined by a weighted linear regression with a 99 % confidence interval (CI) range of 1.0 to 1.9 ‰ (Fig. 4).This is a conservative estimate of δ 13 CH 4 because gas diffusion in the firn dampens the excursions that form the beginning and end of the reversal in the ice record and therefore smoothes the signal of atmospheric changes.The measured Påkitsoq δ 13 CH 4 over the ∼ 150 yr transition is similar in magnitude to the ∼ 2 ‰ 13 C-enrichment from 1850 to 2000 AD (Sowers, 2010), despite the absence of anthropogenic sources.

Robustness of the findings
Påkitsoq ice is known to sporadically produce anomalously high results for [CH 4 ] as compared to contemporary GISP2 ice (Petrenko et al., 2006;Schaefer et al., 2009).It is therefore necessary to test whether the finding of the 13 Cenrichment trend is robust.It can be shown that, although excess [CH 4 ] may lead to lower precision and introduce uncertainty, it does not systematically affect δ 13 CH 4 in the Påkitsoq ice (Schaefer et al., 2006).For the presented data, the maximal values of [CH 4 ] excess reach up to 480 ppb with two extreme values > 1000 ppb (defined as the difference between Påkitsoq IRMS-derived [CH 4 ] and GISP2 (Brook et al., 2000)).This is clearly higher than the conservative estimate of our [CH 4 ] precision (± 43 ppb) and excess values span three orders of magnitude, yet excess [CH 4 ] and δ 13 CH 4 are not correlated with r 2 = 0.092 (Fig. 5).Even if there was a systematic effect of excess [CH 4 ] on δ 13 CH 4 , the former would have to affect certain parts of the record more than others in order to create an artifactual trend.This is not evident with r 2 values for age versus [CH 4 ] excess over the ca.150 yr transition period of 0.02 and 0.01, with and without two major outliers ([CH 4 ] excess > 1000 ppb), respectively (Fig. 5).We cannot rule out that the lower-thanstandard precision of our method for [CH 4 ] masks a minor trend in [CH 4 ] excess.However, due to the lack of correlation between δ 13 CH 4 and the large range of detected [CH 4 ] excess, we consider it unlikely that such a masked trend in [CH 4 ] excess would cause the observed 13 C-enrichment trend.We also note that in Påkitsoq ice [CH 4 ] excess is more prevalent in ice from warm periods than in cold ones (Schaefer et al., 2009).As in situ produced CH 4 is expected to be 13 C-depleted due to the metabolic fractionation during methanogenesis (Whiticar, 1999), we would anticipate lower δ 13 CH 4 in the contamination prone PB ice compared to YD ice.The observed trend in our record is of opposite sign, so it is unlikely to be a natural artefact from in situ CH 4 production in the ice.We further note that the magnitude of the enrichment (1.5 ‰) clearly exceeds analytical precision (< ± 0.3 ‰) (Melton et al., 2011) and pooled standard deviation (± 0.56 ‰) of our new dataset (including points removed by the data filter).The 99 % CI of the trend, which is weighted by data uncertainty, shows that the trend is statistically significant with a minimum enrichment of 1.0 ‰.The reversal to higher δ 13 CH 4 is also not dependent on the applied data filter as it is evident in both the filtered and unfiltered datasets with almost identical magnitudes (Fig. 3).Additionally, the sum of all applied fractionation corrections (Sect.2.2.) takes the form of a flat-topped peak that is distinct from the near linear δ 13 CH 4 trend across the YD-PB (Fig. 6), i.e. uncertainties in the applied corrections cannot account for the trend.We conclude that the enrichment is a statistically significant and robust feature of our dataset.Two previously published δ 13 CH 4 studies covering the same time period have not reported an enrichment trend during the YD-PB.A closer look at the published data, however, reveals evidence for the enrichment in those records.2007). 3Values from Lassey et al. (2007). 4Proportional sink strengths are adjusted to accommodate the modern estimated sink strength of Allan et al. (2007) mean, low and high estimates.(Ferretti et al., 2005).These values integrate over long time spans and therefore present a strongly muted version of the atmospheric signal, but qualitative 13 C-enrichment is evident (Schaefer et al., 2006, Fig. 1), although it may not be statistically significant.The δ 13 CH 4 record from EDML, Antarctica, shows a near continuous 13 C-depletion trend from the start of the YD that persists far into the PB (Fischer et al., 2008).However, the record does not resolve the YD-PB in detail and presents only values before and after (but not during) the rapid [CH 4 ] increase (see further discussion below).The two bracketing values show the only reversal of the longterm depletion trend with an enrichment of 0.5 ‰ (Fig. 2).This value is significant with respect to the stated data uncertainty (± 0.15 ‰).
One of the EDML δ 13 CH 4 data points has a mean gas age (on the Påkitsoq timescale) of little more than a decade prior to the initiation of the rapid CH 4 transition.While we feel that our age determination is based upon the best approach available (see Sect. 2.2.5), given the uncertainty in gas ages, it is possible that the EDML sample falls within the abrupt CH 4 transition.If so, that data point would require correction for diffusion fractionation, giving a possible correction of between 0 and almost 1 ‰ (Buizert, 2012).Where exactly in that range the appropriate correction lies depends on how well the exact start of the abrupt [CH 4 ] rise can be determined.The Greenland records (GISP2 and Påkitsoq) have an advantage with detailed [CH 4 ] and δ 15 N records (Severinghaus et al. 1998), while the EDML [CH 4 ] record is relatively sparse (EPICA Community Members, 2006) and lacking a good gas phase indicator of the initiation of the warming event.Regardless, Buizert (2012) applied diffusion correction to the data point in question, which creates a plateau in the EDML δ 13 CH 4 record during the YD-PB, rather than a 13 C-enrichment.However, the next younger EDML datum is clearly younger than our maximum δ 13 CH 4 suggesting that the enrichment trend is simply not resolved in EDML.Yet, even a plateau would still mark a disruption of the long-term 13 C-depletion trend that initiates at the start of the deglaciation (Fischer et al., 2008), indicating different source/isotope dynamics during the [CH 4 ] rise than during the YD.
We conclude that a trend to higher δ 13 CH 4 during the [CH 4 ] increase of the YD-PB is a statistically significant feature measured using three different extraction and analytical techniques.The feature is also possibly evident in ice from a different location and measured with an independent set-up.Thus, while our record has higher uncertainty given the necessity of applying a contaminated sample filter, it appears that the observed 13 C-enrichment trend is robust enough to allow further interpretation.

Triple isotope mass balance
From the weighted linear regression, the atmospheric δ 13 CH 4 values at the start and end of the YD-PB are −46.9‰ and −45.4 ‰, respectively.The δ 13 CH 4 value of the total methane source can be derived from the atmospheric δ 13 CH 4 by applying a weighted aggregate isotopic fractionation of total sink processes (ε = −6.72 ‰), and correction for isotopic disequilibrium (Tans, 1997;Lassey et al., 2000)  the fractional proportion of each sink (including the highly fractionating atomic chlorine in the marine boundary layer (MBL) sink) and its ratio of the rate coefficients, α, for each isotope as ε = 10 3 (α − 1) in units of per mille.Using the increase in atmospheric [CH 4 ] across the YD-PB (ca.250 ppb; Brook et al., 2000), an isotope mass balance yields the stable carbon isotope ratio of the additional emissions (δ 13 CH 4 ↑ T ) to be −49.2± 1.3 ‰.
To further narrow the possible source(s) of the increased methane flux during the YD-PB transition, we can use the other isotopes of methane (δD-CH 4 and 14 CH 4 ).
The GISP2 stable hydrogen isotope (δD-CH 4 ) record for the YD-PB (Sowers, 2006) shows no significant trend within the transition with a near constant approximate value of −90 ‰, although there are relatively few measurements and thus higher uncertainty.The mean δD-CH 4 value of the additional emissions (δD-CH 4 ↑ T ) is −314 ± 8 ‰, after correction for sink processes (ε = −223.9‰) and isotopic disequilibrium (Table 1 and Sect.2.2.3).No primary CH 4 source has this combined δ 13 CH 4 and δD-CH 4 signature (Fig. 7).Therefore, it is likely that the Q T during the YD-PB cannot be attributed to variations of a single source and the observed Q T likely results from increased flux of two (or more) sources.
The YD-PB 14 CH 4 record provides a third constraint (Petrenko et al., 2009), although this record has high uncertainty due to a correction for in-situ 14 C production and because the necessarily large samples contain gas of ages beyond the actual transition (Fig. 2).The 14 CH 4 record shows a fossil methane contribution in the later stages of the YD-PB ( 14 CH 4 ↑ T = −138 ‰; range of 1.5 to −276 ‰, all values are converted from original age-corrected 14 C notation; Petrenko et al., 2009).The values from Petrenko et al. (2009) are from their two-end member scenario assuming a YD GEM contribution of 50 Tg CH 4 per year, which is close to other estimates of YD GEM emissions (Whiticar and Schaefer, 2007;Melton, 2010).Additionally, the values used in the triple mass balance have already accounted for the estimated shift in 14 CH 4 values due to declining atmospheric 14 CO 2 following Petrenko et al. (2009).
Using the isotope records of the YD-PB, three separate mass balances can be formulated for the sources of Q T : where Q n and δ( 13 C, D, 14 C)↑ n are the fractional mass flux change across the YD-PB and the field measurement-based characteristic isotope values (δ 13 CH 4 , δD-CH 4 , 14 CH 4 ) of the n-th source term, respectively (Table 2).Q C is the calculated total fractional mass flux (set as 1).δ ( 13 C, D, 14 C)↑ C are the calculated cumulative isotope values of the mixed sources, which can be compared to those derived from the ice records (δ ( 13 C, D, 14 C)↑ T ).As we have three mass balances  1 and 2).Unfilled circles are anthropogenic sources.
with three unknowns, the mass balances can be solved simultaneously for Q 1 , Q 2 , and Q 3 (see Supplement S1).
The results must account for the analytical uncertainties of the underlying isotope data.Therefore, error ranges have been conservatively selected for each of δ 13 CH 4 ↑ T , δD-CH 4 ↑ T , and 14 CH 4 ↑ T .The δ 13 CH 4 ↑ T range (± 1.3 ‰) was determined from the standard error-weighted linear regression model's 99 % confidence interval maximum range (Fig. 4).For δD-CH 4 , as there is little or no trend across the YD-PB, the δD-CH 4 ↑ T range (± 8 ‰) is taken as double the experimental external precision (± 4 ‰) (Sowers, 2006) to accommodate any possible errors due to under-sampling.The available 14 CH 4 record agrees well for the two YD measurements, while the two PB values differ substantially (Petrenko et al., 2009).The 14 CH 4 ↑ T range is calculated from the mean YD 14 CH 4 value and each of the PB 14 CH 4 values as 1.5 to −276 ‰.Given the 14 CH 4 record's high uncertainty and low temporal resolution, we investigate the influence of this record on our results in Sect.3.5.2.As no primary CH 4 source has the observed δ 13 CH 4 ↑ T (−49.2 ± 1.3 ‰) and δD-CH 4 ↑ T (−314 ± 8 ‰) signature, matching calculated δ 13 CH 4 ↑ C values to the ice records requires a mix of 13 C-enriched and 13 C-depleted sources.Only biomass burning and geologic emissions of methane (GEM) have δ 13 CH 4 that can serve as the 13 C-enriched source.Six primary sources can supply sufficiently 13 Cdepleted methane: aerobic plant methane (APM), tropical and boreal wetlands, thermokarst lakes, ruminants, and termites.To satisfy the requirement for a fossil methane contribution, as shown by the 14 CH 4 record, three sources emit 1 NBS OAS: oxalic acid standard (Donahue et al., 1990) 2 Modern values are adapted for the PB and YD as in Schaefer and Whiticar (2008) and Whiticar and Schaefer (2007).If the initial modern value chosen differs from those in Schaefer and Whiticar (2008), the absolute change between time periods is applied to the isotope value chosen.Source isotope values that are not expected to be sensitive to environmental and climatic changes are kept at their modern isotope values.Equivalent changes due to the palaeoenvironment have not been quantified for δD-CH 4 .Thus, modern values are used throughout (with the exception of the tropical wetlands sensitivity test outlined in Sect.3.5.1). 14CH 4 values are assumed to not systematically change across the time periods.
3 Table 1 of Etiope et al. (2008) lists the range of values for the different geologic sources. 4δ 13 CH 4 values from Lassey et al. (2007). 5δ 13 CH 4 and δD-CH 4 values from Fischer et al. (2008). 6To test the suggestion of MacDonald et al. (2006) that boreal wetlands during the deglaciation would be likely characterized by minerotrophic fens (e.g.−60 ‰ but ranging −50 to −73 ‰; (Bellisario et al., 1999) rather than the present Sphagnum bog complexes.a Whiticar and Schaefer (2007) b Based upon Nakagawa et al. (2002a,b) and Waldron et al. (1999) c Walter et al. (2008)  d Value is an average based on Keppler et al. (2006) and Vigano et al. (2009) for detached plant material.Whole plant values are not used due to the non-biotic nature of the CH 4 production (Nisbet et al., 2009).A weighted average between C 3 and C 4 plants was based upon 25 % global grassland coverage, 57 % of which are C 4 plants (Collatz et al., 1998).e The measurements of Vigano et al. (2009) are the only reported δD-CH 4 values.Isotope value is a weighted average of C 3 and C 4 plants in same proportion as noted above.f Lassey et al. (2007) g Walter et al. (2008).Estimate has high uncertainty as it is based on data from only two thermokarst lakes.Thermokarst lakes have been shown to emit methane with 14 CH 4 values ranging from modern to −993‰.Thus, we follow Petrenko et al. (2009) in assuming the 14 C in thermokarst lakes is a mixing of two-end member organic substrate materials: one modern and the other 14 C-free.We assume the fraction of these two end-members is the same during the YD and PB as present and use the same flux-weighted average (Walter et al., 2008; Petrenko et al., 2009).h We are not aware of any measured APM 14 CH 4 values.Therefore, we assign APM the 14 C value of ruminant CH 4 .i Quay et al. (1999).We assume that natural biomass burning has the same 14 CH 4 values as anthropogenic biomass burning measured 14 CH 4 values.j Winckler et al. (2002)  k Quay et al. (1999).We assume mined natural gas has the same 14 CH 4 value as GEM.l Wahlen et al. (1989b) Table 3. Triple isotope mass balance model results for YD-PB source scenarios.Fractional source contributions are calculated simultaneously via the triple isotope mass balance and compared to the mean ice record δ 13 CH 4 ↑ T , δD-CH 4 ↑ T and 14 CH 4 ↑ T values for scenario acceptance.All valid scenarios (bold font) pass the acceptance criteria outlined in the text.Solutions with negative emissions are physically meaningless and are listed in Table S4  fossil or 14 C-depleted methane: GEM, thermokarst lakes, and marine gas hydrates.There are then 29 possible threesource scenarios that include a 13 C-enriched, a 13 C-depleted and a fossil component, where the contribution from each source can be quantified from the triple mass balance (Table S4 in the Supplement).We impose two criteria on the mass balance calculation's source combinations to identify valid scenarios.First, the calculated cumulative isotope values, δ( 13 C,D, 14 C)↑ C , must reproduce the ice isotope records, i.e. δ( 13 C, D, 14 C)↑ C must equal δ ( 13 C, D, 14 C)↑ T .Second, the fractional contribution from each source, Q 1 , Q 2 , and Q 3 , must sum to 1.0 ± 0.1 (the closer to one, the more probable).The value of this second constraint is chosen arbitrarily.Obviously several simplifications and assumptions are intrinsic to the triple mass balance approach including constant and representative source isotope values through the transition, constant sink isotopic fractionation (the sink strength can vary; it just changes the magnitude of the CH 4 increase), and the assumption that the CH 4 increase can be represented adequately by the contribution of three or less sources.Representative source isotope values could also have changed from modern values and are investigated in Sect.3.4.
All valid scenarios derived from the triple isotope mass balances listed in Table 3 have in common that the increased CH 4 flux across the YD-PB, Q T , is predominantly sourced (> 50 %) from biomass burning with a strong contribution (∼ 43 %) from thermokarst lakes.The remaining sources (global wetlands, APM, termites, and ruminants) appear equally likely to have contributed a minor amount (> 10 %).A previous ice core δ 13 CH 4 study concluded that the biomass burning source flux remained stable between the YD and PB periods (Fischer et al., 2008).Their conclusion is based on the comparison of average YD and PB values using the long-term decreasing δ 13 CH 4 trend.In contrast, the pronounced rise in pyrogenic CH 4 reported here specifically applies to the short transition period while [CH 4 ] and δ 13 CH 4 change.The arbitrarily chosen selection criteria for valid scenarios ( Q 1 , Q 2 , and Q 3 must sum to 1.0 ± 0.1, or ± 10 %) appear to be reasonably chosen as the criteria would have to raised to 15 % to allow another possible scenario (one that includes thermokarst lakes, biomass burning, and marine gas hydrates).To bring in another source scenario that would challenge the dominant sources from the other valid scenarios would necessitate raising the inclusion limit to 22 % (for a scenario including biomass burning, GEM, and termites).By contrast, of the 29 possible source combinations, the five that pass the acceptance criteria all have a summed fractional contribution that falls within 4 to 7 % of unity.
Accounting for the uncertainties in the various isotope data from the ice records by investigating the minimum and maximum values of each isotope while holding the other two at their mean values (Table S5 in the Supplement) produces ranges for the various sources as ∼ 42-66 % for biomass burning, ∼ 27-59 % for thermokarst lakes and ∼ 0-29 % for the complementing third source.Given these probability ranges, the finding that Q T is composed mainly from biomass burning and thermokarst lakes with some contribution from a third source (or several minor ones) seems robust.However, additional uncertainty is introduced by incomplete understanding of present and past methane isotope budgets, which are investigated in the following section's sensitivity tests.
From Table S5 in the Supplement, three scenarios deserve further comment.If the YD-PB methane isotope values are best represented by the minimum δD-CH 4 ↑ T value (−322 ‰) or the maximum δ 13 CH 4 ↑ T value (−47.9 ‰), then the triple mass balance calculates a two-source mix of biomass burning and thermokarst lakes.While these source scenarios exceed the Q acceptance criteria (by 1 to 2 %), they fit the atmospheric isotope constraints better than any other source combinations for the minimum δD-CH 4 ↑ T or maximum δ 13 CH 4 ↑ T values (including all other source combinations listed in Table S6 in the Supplement).Also the minimum 14 CH 4 ↑ T value (−276 ‰) produces no valid scenarios.This implies that a fossil contribution over the YD-PB transition of this magnitude is unlikely.The source fractional contributions thus do not include this scenario in the calculation of their ranges.

Mass balance sensitivity tests
The source contributions quoted in Sect.3.3 are based on a standard scenario that uses modern isotope values adapted for the Younger Dryas conditions (Schaefer and Whiticar, 2008) (Tables 1 and 2).We varied these parameters to investigate the sensitivity of the triple mass balance results for (i) alternative source isotope values, proportional sink strengths, and sink fractionation factors; (ii) the impact of changing environmental and climatic conditions upon source isotope values, including changes in the distributions of, and methane produced by, C 3 and C 4 vegetation (Schaefer and Whiticar, 2008); and (iii) different strengths of the MBL atomic chlorine sink.All sensitivity tests using the mean ice core based δ 13 CH 4 ↑ T , δD-CH 4 ↑ T , and 14 CH 4 ↑ T values are described in detail below.The sensitivity tests produce different sets of source scenarios that are subjected to the same evaluation criteria as the standard scenario.Valid scenarios are shown in Table S6 in the Supplement.

Fen-dominated wetlands
At the YD-PB, boreal peatlands were in their early development stages, which are likely best characterized as fendominated wetlands (MacDonald et al., 2006), rather than ombrotrophic Sphagnum-dominated bogs that are extensive in the boreal region today with a δ 13 CH 4 value of about −65 ‰, as used in the standard scenario.Fen-dominated wetlands are arguably better represented by a more 13 Cenriched δ 13 CH 4 value of −60 ‰ (MacDonald et al., 2006) (Table 2).We investigated the triple mass balance sensitivity to this selection of the δ 13 CH 4 value and found the changes to be minor.None of the scenarios changed their status as valid or invalid (therefore not included in Table S6 in the Supplement).The maximum estimated fractional contribution from boreal wetlands increases from 0.07 to 0.10.

Changing environmental and climatic conditions
In our standard scenario, source isotope values are changed from present-day literature values following Schaefer and Whiticar (2008) and Whiticar and Schaefer (2007) to reflect different environmental and climatic conditions during the YD period.For δ 13 CH 4 , we account for changes between modern and Preboreal/YD environmental and climatic conditions including (i) changes in atmospheric δ 13 CO 2 ; (ii) vegetation patterns, including C 3 /C 4 plant distribution patterns, which influence the δ 13 C of the organic CH 4 precursor material; (iii) the impact of temperature change on CH 4 production itself, particularly in wetlands; and (iv) partial oxidation before emission to the atmosphere.To test the triple mass balance model sensitivity to these changes, we investigated two scenarios using δ 13 CH 4 values characteristic of (i) modern conditions and (ii) the Preboreal period (Table 2).All other model parameters are unchanged.The results from these two tests are shown as "Modern Scenario" and "Preboreal Scenario" in Table S6 in the Supplement with no significant changes to the results of the standard scenario.

Atomic chlorine in the marine boundary layer sink
A recent modelling study suggests that δ 13 CH 4 -enrichment on the order of 0.3 ‰ can result from glacial to interglacial changes in the sink configuration, primarily due to changes in the highly fractionating MBL atomic chlorine sink (Levine et al., 2011).As our observed trend far exceeds the modelled value, this supports the view that that the processes are primarily source driven, although a contribution from sink changes cannot be ruled out.The impact of the inferred atomic chlorine sink in the MBL (Allan et al., 2001) is thus investigated.The strength of the MBL sink is based upon the estimated modern strengths from Allan et al. (2007).The standard scenario assumes a MBL sink at the estimated minimum modern strength (Allan et al., 2007) (Table 1).Sensitivity tests explore the influence of assuming (i) no MBL sink, (ii) the mean, and (iii) the maximum MBL sink strength from Allan et al. (2007) (Scenarios "No MBL Sink", "Mean MBL Sink", "Max MBL Sink" in Table S6 in the Supplement).
All other model parameters remain the same as the standard scenario, except the fractional contributions from the other sinks are adjusted to accommodate the changes in the MBL sink.The ranges used in the sensitivity tests are −5.39 to −9.02 ‰ for total ε C and −218.3 to −233.6 ‰ for total ε D .
The results of these changes to the MBL sink are small (Table S6 in the Supplement).The most significant difference is that a scenario with biomass burning (contribution of 66 %), thermokarst lakes (40 %) and marine gas hydrates (4 %) is valid for a zero MBL sink; however, this scenario is on the limit of the acceptance criteria.We note that these tests, as well as all mass balance scenarios, assume that ε-values do not change through the transition.

Alternative δ 13 CH 4 and δD-CH 4 characteristic budgets
Alternative characteristic primary isotope value estimates for methane source types are used in the literature.Here we test model sensitivity of alternative source budgets compared to our chosen primary δ 13 CH 4 and δD-CH 4 values (Tables 1  and 2).The first alternative budget tested is that of Fischer et al. (2008).This budget does not include termites, GEM, thermokarst lakes or an MBL sink.Additionally, their APM characteristic isotope values are not supported by measurements (Keppler et al., 2006;Vigano et al., 2009).We thus do not investigate scenarios involving these sources/sink with the exception of thermokarst lakes where our standard value is used.
The second alternative budget is that of Lassey et al. (2007), which does not include source characteristic values for APM, marine gas hydrates, or thermokarst lakes.Additionally, they do not distinguish δ 13 CH 4 values between tropical and boreal wetlands (Lassey et al., 2007).We do not investigate scenarios involving APM or marine gas hydrates with the Lassey et al. (2007) budget.
While both Lassey et al. (2007) and Fischer et al. (2008) address different time scales than ours and thus have no direct bearing on our conclusions, they demonstrate the sensitivity of the triple mass balance results to different representative source isotope values.The mass balance results of the two alternative budgets by Fischer et al. (2008) and Lassey et al. (2007) (likewise named in Table S6 in the Supplement) do not produce additional valid scenarios compared to the standard scenario.This further supports the conclusion that the additional YD-PB CH 4 emissions were predominantly caused by biomass burning and thermokarst lakes with only small contributions from other sources.
From the sensitivity tests, the mass balances are most sensitive to changes in the primary source isotope values (Table S6 in the Supplement).However, these probable alternative isotope values result in valid scenarios with the same dominant sources as the standard scenario.While these sensitivity tests explore a relatively wide range in source characteristic isotope values (and even more so in Sect.3.5.1.),it could be possible for the characteristic isotope values to change through the course of the transition in response to environmental conditions, or some unknown processes, to values outside of the range of our tests.Given that the processes by which this could occur remain unknown, we are not able to accurately test for this possibility, but we do wish to acknowledge it and the associated uncertainty it brings.

Dominant source contributions
The fractional source flux changes from the triple mass balance calculations can be converted to absolute fluxes (in units of Tg CH 4 yr −1 ) by assuming a total flux increase Q T of 64 Tg CH 4 yr −1 (Petrenko et al., 2009).Due to the uncertainties in the ice core data and the model assumptions, the derived absolute fluxes for individual sources should be regarded as a best estimate only.However, the exercise may be informative to assess if individual source changes are probable in light of total source strengths and their estimated variability.
We calculate Q for biomass burning to be 27-42 Tg yr −1 .This indicates an increase of up to ∼ 100 % by the end of the YD-PB, compared to estimates of the YD biomass burning CH 4 flux (Fischer et al., 2008;Melton, 2010).For comparison, global methane budget modelling results covering the last two decades show modern pyrogenic emissions reaching almost 30 Tg yr −1 interannually (Bousquet et al., 2006;Drevet, 2008).While these studies necessarily include anthropogenic influences, they still illustrate the recognized sensitivity of biomass burning to climate and vegetation (Westerling et al., 2006;Bowman et al., 2009) even in the absence of an event the scale of the YD-PB.Changes in fire regimes can be related to fuel accumulation and vegetation distribution, as well as ignition sources, e.g.lightning (Westerling, 2009).Evidence of enhanced burning close to the YD-PB is observed in global charcoal datasets (Power et al., 2008;Daniau et al., 2010) and a North American charcoal and pollen dataset (Marlon et al., 2009).
It is also worthwhile to investigate if CH 4 isotopes can provide quantitative estimates of changes in global biomass burning.Carbon monoxide is a trace gas component that can be used to confirm estimates of biomass burning.CO is largely produced through biomass burning and oxidation of non-methane hydrocarbons, and the relative proportions of these sources can be distinguished using stable isotopes.Recent results from an Antarctic ice core record CO mixing ratio and isotopic ratios record spanning the last 650 yr (Wang et al., 2010) support the biomass burning estimates from two Antarctic ice core [CH 4 ] and δ 13 CH 4 records (Ferretti et al., 2005;Mischler et al., 2009) and are generally consistent with the tropical charcoal index (Daniau et al., 2010).The Wang et al. (2010) study thus supports the ability of δ 13 CH 4 to estimate changes in global biomass burning fluxes.
We estimate thermokarst lake Q to be 17-38 Tg yr −1 , i.e. slightly higher than derived from thermokarst lake basal initiation dates (∼ 10-16 Tg yr −1 ) (Walter et al., 2007).As well, a recent reanalysis of the thermokarst lakes basal initiation dataset using a different methodology finds that the main thermokarst lakes expansion occurred after the YD-PB (Reyes and Cooke, 2011), contrary to our results.We note that both of these studies using thermokarst lake basal 14 C initiation dates are somewhat uncertain as the database contains only 66 dates.As well, the initiation dates do not necessarily capture lateral expansion caused by climate changes (similar to peatlands; Korhola et al., 2010), which may at least partially account for the discrepancy between our estimates and those based upon thermokarst basal initiation dates.Thermokarst lake emissions into the Northern Hemisphere are consistent with the increase in inter-hemispheric [CH 4 ] gradient between the averages of the YD and PB periods (Brook et al., 2000;Dällenbach et al., 2000), although the latter may not be indicative of the short-term transition dynamics.In conclusion, there is no clear support for our estimated magnitude of increased thermokarst lake CH 4 production during the YD-PB from field evidence indicating a possible overestimate.
Reyes and Cooke (2011) also reanalyzed two large basal peatland 14 C datasets: one for Alaska (Jones and Yu, 2010) and the other spanning all circumpolar regions (MacDonald et al., 2006).They conclude that northern peatlands could not have been the primary drivers of the rapid increases in [CH 4 ] observed during the deglaciation.Their analysis is consistent with our triple mass balance results that show a minimal contribution from boreal peatlands, even if we assume the newly formed peatlands to be more productive minerotrophic fens rather than the present-day ombrotrophic Sphagnum-dominated bogs (MacDonald et al., 2006) (Table 2, Sect.3.4.1).Both their and our studies do not support an earlier suggestion (Jones and Yu, 2010) that Alaskan peatlands, less encumbered by ice sheet dynamics and thermal inertia than other circumpolar peatlands, could have contributed strongly to the CH 4 rise.
Scenarios with substantial contributions (> 10 %) from either GEM (including thermogenic clathrates) or biogenic marine clathrates overestimate Q T by at least 22 % and 45 %, respectively (Table 3) and do not satisfy the isotope constraints.Thus, our findings are not consistent with any significant clathrate release over hydrocarbon seeps (GEM) (Hill et al., 2006) during the YD-PB.

Tropical wetlands
Tropical wetlands have been discussed as the predominant driver of the atmospheric methane concentration increase over the YD-PB (Brook et al., 2000).These arguments are based on the size of this source at present day (Denman et al., 2007) and inferred changes to the low-latitude hydrologic cycle (Peterson et al., 2000).However, from the triple mass balance results (Tables 3 and Table S4 in the Supplement), tropical wetlands do not appear likely to have greatly increased their CH 4 flux over the YD-PB, while we note that stable isotope values of wetland sources are possibly sensitive to changes in the ratio of net to gross CH 4 production (Sowers, 2006) and source water δD-H 2 O values (Bock et al., 2010).
The hydrogen isotope ratio of wetland methane is ultimately derived from the hydrogen isotope ratio of meteoric water and hence precipitation (Whiticar, 1993) (recirculation notwithstanding).The hydrogen isotope ratio of precipitation is not constant through time.It is influenced by several factors, including the isotopic composition of the ocean, which, in turn, is largely determined by the size of terrestrial ice sheets.For example, the δD-H 2 O values of the ocean during the Last Glacial Maximum (∼ 21 kyr BP, LGM) are reported to be 6.5 to 9 ‰ more 2 H-enriched than at present (Schrag et al., 2002).In addition, the modulating Rayleigh distillation effect will be strengthened by a stronger pole-toequator temperature gradient in cold climates.Modelling of the LGM suggests that δD-H 2 O values were 15 to 30 ‰ more 2 H-depleted in northern latitudes, while values in low latitudes may have been 0 to 8 ‰ more 2 H-enriched than at present (Jouzel et al., 2000).The rapid changes in the hydrologic cycle proxies are suggested to reflect movement of the Intertropical Convergence Zone (ITCZ).Close to the ITCZ, the changes are unlikely to have a major effect on the δD-H 2 O as the water is rapidly recycled; however, the changes should lead to a latitudinal relocation of tropical wetlands.This effect has been assumed to be relatively small (Bock et al., 2010).For the carbon system, the δ 13 CH 4 changes due to environmental and climate changes of the YD have been estimated to be 1.7 ‰ (Schaefer and Whiticar, 2008).No similar estimate exists for the hydrogen system.
To test the mass balance results for tropical wetlands, the δD-CH 4 and δ 13 CH 4 characteristic values for tropical wetlands were changed to extreme values to test the maximum contribution from tropical wetlands.The tropical wetland δD-CH 4 value was changed to range between −300 ‰ and −330 ‰ (from a standard value of −315 ‰).Shifts of this magnitude are larger than present evidence indicates (Jouzel et al., 2000;Bock et al., 2010) and result in a maximum estimate of tropical wetland CH 4 flux increase.The tropical wetland δ 13 CH 4 value is changed by +8.9 ‰ from the modern value, or +7.2 ‰ from the estimated YD value, to −50 ‰ (see Table 1 for standard values) in the direction of the mean δ 13 CH 4 ↑ T value.To achieve a δ 13 CH 4 value this 13 C-enriched would require a disproportionately large contribution from C 4 organic matter to tropical wetland emissions as compared to estimated C 4 /C 3 distributions (Collatz et al., 1998;Schaefer and Whiticar, 2008).However, even with these extreme characteristic values, the fractional contribution only increases from 0.08 (standard scenario with mean δ 13 CH 4 ↑ T , δD-CH 4 ↑ T , and 14 CH 4 ↑ T values) to 0.22.As well, no additional scenarios pass the acceptance criteria besides those including biomass burning and thermokarst lakes as the main sources, as is the case for the standard tropical wetlands isotope values.Therefore, within the tested uncertainty ranges, our isotope mass balance results are not consistent with the hypothesis that tropical wetlands were the major source of increased methane flux during the YD-PB transition.

Test of the influence of the 14 CH 4 dataset
Our triple mass balance uses the 14 CH 4 dataset of Petrenko et al. (2009) as it is the only 14 CH 4 dataset presently available.This dataset is not ideal given the necessarily low temporal resolution of the samples and high uncertainty due to corrections for cosmogenic 14 C production.To test the sensitivity of the results based on the triple mass balance to the  Results are presented as percent mismatch between the fractional source contributions calculated independently by the δ 13 CH 4 and the δD-CH 4 mass balances (methodology is further described in Sect.3.5.2.).The characteristic source and sink isotope values and source fractional strengths are for the standard scenario as in the triple mass balance (see Tables 1 and 2).The mean percent mismatch is calculated using mean δ 13 CH 4 ↑ T (−49.2 ‰) and δD-CH 4 ↑ T (−314 ‰) values.Also presented are the minimums and maximums in percent mismatches between the mass balances for the range of δ 13 CH 4 ↑ T (1.3 ‰) and δD-CH 4 ↑ T (8 ‰) values (see Sect. 3.3).From this uncertainty range, the upper limit in percent mismatch values is consistently given by the constraint of a δD-CH 4 ↑ T value of −322 ‰ and δ 13 CH 4 ↑ T value of −47.9 ‰.The lower limit is consistently given by the constraints of a δD-CH 4 ↑ T value of −306 ‰ and aδ 13 CH 4 ↑ T value of −50.4 ‰.All percent mismatch values less than 10 % (similar to the criteria for an acceptable scenario in the triple mass balance; see Sec. 3.3) are considered valid scenarios and are highlighted in bold font.use of the 14 CH 4 dataset, we have performed a mass balance analysis using only the δ 13 CH 4 and δD-CH 4 datasets.This approach, of course, allows only a two-source mixture, which should be reasonable given that the triple mass balance consistently showed a predominantly two-source mix for the most likely scenarios (see Table 3).For this test, each mass balance is formulated individually (Eqs.4 and 5) and solved independently (in contrast to the triple mass balance which solves the three simultaneously): The sources included in the mass balances can be split into two groups.The first group ({δ 13 C, δD} 1 • Q 1 ) includes the sources that are more 13 C-enriched than the mean source δ 13 CH 4 ↑ T value (biomass burning and GEM), and the second group ({δ 13 C, δD} 2 • Q 2 ) includes sources more 13 Cdepleted (biogenic marine gas hydrates, tropical and boreal wetlands, thermokarst lakes, termites and APM) (see Fig. 7).
The mass balances can then be calculated for each combination of a source from the 13 C-enriched group and a source from the 13 C-depleted group.The δ 13 C C and δD C values are set to the δ 13 CH 4 ↑ T and δD-CH 4 ↑ T values, respectively.Q 1 and Q 2 are the fractional contribution of each source, and Q c is set to unity.
Since these mass balances are for the same system, if they are accurately representing the actual source combination that forms the δ 13 CH 4 ↑ T and δD-CH 4 ↑ T values, the fractional contributions of the sources ( Q n ) for each mass balance, δ 13 C and δD, should be equal, i.e.Q 1 from the δ 13 C mass balance should equal Q 1 from the δD mass balance.The larger the mismatch between the two mass balances' calculated source fractional contributions, the less likely those two sources accurately describe the source combination that formed the δ 13 CH 4 ↑ T and δD-CH 4 ↑ T values.The results of all possible source combinations for the standard scenario are presented in Table 4.We use the criteria of ± 10 % of mismatch between the two mass balance source contributions to denote valid scenarios (similar to 1 ± 0.1 used in the triple mass balance).With these criteria, no twosource mixes using GEM as the 13 C-enriched source are valid, with the closest being a 28 % mismatch between the fractional source combinations calculated in the δ 13 C and δD mass balances.For mass balances with biomass burning as the 13 C-enriched source, the valid scenarios suggest that the 13 C-depleted component could be any of APM, boreal wetlands, thermokarst lakes, or termites.These valid scenarios predict a biomass burning fractional contribution between 0.28 and 0.48.These values are less than those calculated by the triple mass balance; however, this test can only investigate two-component mixes, which will influence the fractional contribution values calculated.Exploring the uncertainty range around the δ 13 CH 4 ↑ T and δD-CH 4 ↑ T values shows no valid scenarios with GEM as the 13 C-enriched source, and no valid scenarios with either biogenic marine gas hydrates or tropical wetlands as the 13 C-depleted source.When the δ 13 CH 4 ↑ T and δD-CH 4 ↑ T values are assumed to be −50.4‰ and −306 ‰, the combination of ruminants and biomass burning becomes a valid scenario.Of the source combinations that are valid scenarios for the mean δ 13 CH 4 ↑ T and δD-CH 4 ↑ T values, none of them remain valid scenarios throughout the entire range of δ 13 CH 4 ↑ T and δD-CH 4 ↑ T values tested (see Table 4).
These mass balance tests support the conclusion, from the triple mass balance, of a strong role for biomass burning at the YD-PB transition.The triple mass balance result of a strong role for thermokarst lakes remains plausible; however, these tests demonstrate that the unique role for thermokarst lakes is dependent upon the validity of the 14 CH 4 dataset.If the use of, or data within, this dataset is not valid, then we are not able to distinguish the possible contribution of thermokarst lakes from that of APM, boreal wetlands or termites.This also adds uncertainty to the triple mass balance result of no strong role for boreal wetlands.Tropical wetlands and biogenic marine gas hydrates remain unlikely to have contributed greatly to the methane rise at the end of the YD.

Conclusions
We present a new δ 13 CH 4 dataset measured from Påkitsoq ice covering the YD-PB transition.A 13 C-enrichment trend across the transition is evident and well exceeds measurement uncertainty.The trend is also evident in previously published Påkitsoq measurements (Schaefer et al., 2006) using different experimental setups (Ferretti et al., 2005;Schaefer et al., 2006).While not unambiguous due to dating uncertainty, it appears that a similar disruption of the long-term deglaciation 13 C-depletion trend occurred in the presently available Antarctica δ 13 CH 4 record (Fischer et al., 2008).
A triple mass balance has been used to determine the most plausible source combination for the rapid [CH 4 ] increase during the YD-PB transition.The mass balance results show a strong role for biomass burning with an additional significant contribution from thermokarst lakes.The triple mass balance approach uses the best available δ 13 CH 4 , δD-CH 4 , and 14 CH 4 datasets, of which the 14 CH 4 dataset (Petrenko et al. 2009) has high uncertainty and much lower temporal resolution.To determine the influence of our use of the 14 CH 4 dataset, a separate mass balance approach using only δ 13 CH 4 and δD-CH 4 was performed.This test supports a strong role for biomass burning, but could not distinguish the corresponding 13 C-depleted source between aerobic plant methane, boreal wetlands, thermokarst lakes, and termites.These results then do not contradict the findings of the triple mass balance but do add extra uncertainty around the role of thermokarst lakes in the methane concentration rise.If the 14 CH 4 dataset can be used, our conclusion of the dominance of biomass burning, and perhaps thermokarst lakes, seems qualitatively robust in spite of analytical uncertainties of the ice isotope records they are based on, as well as assumptions on past methane budgets as confirmed by sensitivity tests.
In contrast, the quantitative interpretation of the fluxes is subject to the uncertainties of analyses and assumptions.Our analysis agrees with an increase in thermokarst lake methane emissions as previously suggested (Walter et al., 2007;Petrenko et al., 2009) as both mass balance tests find a possible role for thermokarst lakes, but neither test finds a dominant role for tropical wetlands or biogenic marine gas hydrates.The suggestion of a strong role for biomass burning in the rapid methane mixing ratio increase during the YD-PB transition is novel.If our results can be confirmed, they would indicate that, as climate warms into the future as projected, special attention should be paid to the dynamics of biomass burning and thermokarst lakes.Both of these methane sources appear capable of responding to a changing climate with substantial increases in methane flux producing a positive feedback on the climate system.Supplementary material related to this article is available online at: http://www.clim-past.net/8/1177/2012/cp-8-1177-2012-supplement.pdf.

Fig. 5 .
Fig. 5. Påkitsoq excess CH 4 plotted against gas age and δ 13 CH 4 values.Excess CH 4 is defined as the difference between Påkitsoq IRMS-derived [CH 4 ] and GISP2 (Brook et al., 2000).Linear trend lines and r 2 values are shown considering all points (solid black) or all points excluding the two major outliers (> 1000 ppb CH 4 excess)(dashed blue).Top panel: δ 13 CH 4 value as a function of methane excess.Bottom panel: methane excess as a function of gas age.The vertical dashed lines denote the start and end of the YD-PB transition as shown in Figs. 1 and 2.

Fig. 6 .
Fig. 6.The gravitational, thermal, diffusion, and isotopic equilibrium isotope corrections for the Påkitsoq δ 13 CH 4 record.The upper panel is the Påkitsoq δ 13 CH 4 record from this study -the Schaefer et al. (2006) values were treated identically but are not plotted here for clarity.The linear regression model (as in Fig.2) using Påkitsoq data points from the combined dataset (this study andSchaefer et  al., 2006)  is shown.The bottom panel shows the isotope corrections as applied to each data point.
The mean δ 13 CH 4 values of the Schaefer et al. (2006, Fig. 1) dataset show a continuous development to more 13 Cenriched values throughout the YD-PB, where the δ 13 CH 4 values at the end of the [CH 4 ] rise exceed those at the start by a margin bigger than the combined uncertainties.The authors did not state a trend, mostly because the start of the transition was represented by samples from only one field season, which does not cover the preceding YD.This introduced the possibility of a sampling bias.The combined Påkitsoq dataset now shows that the feature is reproducible in sample sets from various campaigns.Schaefer et al. (2006) also presented high precision δ 13 CH 4 measurements on large volume samples performed on a different experimental setup

Fig. 7 .
Fig. 7.Combination of characteristic δ 13 CH 4 and δD-CH 4 signatures for the major primary sources of methane to the troposphere, together with the calculated mean source isotope values of the additional emissions (δ 13 CH 4 ↑ T , δD-CH 4 ↑ T ) [black diamond] as derived from the ice core signal [white diamond] (Tables1 and 2).Unfilled circles are anthropogenic sources.
Artificial reference ice samples δ 13 CH 4 value vs. sample size across three years.The data demonstrate good linearity across the range of artificial ice sample sizes.The dataset mean δ 13 CH 4 value is −47.51 ± 0.29 ‰ (1σ ; n = 32; methane content varying between ∼ 415-1080 pmol).

Table 1 .
Carbon and hydrogen isotope fractionation factors and proportional strength values for the CH 4 sink terms used in the mass balance calculations.The standard scenario with a weak MBL is denoted in bold font.Alternative values tested are shown in parentheses.

Table 2 .
Characteristic isotope values for primary sources used in the triple mass balance calculations.Isotope values used in the standard scenario are denoted in bold font.Isotope δ 13 CH 4 values that are assumed not to change with time are denoted by "−".δD-CH 4 and 14 CH 4 values are unchanged from modern isotope values.Alternative source values for the sensitivity tests are shown in parentheses.
in the Supplement.

Table 4 .
δ 13 CH 4 and δD-CH 4 mass balance results for scenarios with biomass burning and geologic emissions of methane as the 13 Cenriched contributing source.