Report

A 21st-century shift from fossil-fuel to biogenic methane emissions indicated by 13CH4

See allHide authors and affiliations

Science  01 Apr 2016:
Vol. 352, Issue 6281, pp. 80-84
DOI: 10.1126/science.aad2705

Getting a rise out of agriculture

Methane, a powerful and important greenhouse gas, has been accumulating nearly uninterruptedly in the atmosphere for the past 200 years, with the exception of a mysterious plateau between 1999 and 2006. Schaefer et al. measured methane's carbon isotopic composition in samples collected over the past 35 years in order to constrain the cause of the pause. Lower thermogenic emissions or variations in the hydroxyldriven methane sink caused the plateau. Thermogenic emissions didn't resume to cause the subsequent rise. Instead, the ongoing rise is most likely due to biogenic sources, most notably agriculture.

Science, this issue p. 80

Abstract

Between 1999 and 2006, a plateau interrupted the otherwise continuous increase of atmospheric methane concentration [CH4] since preindustrial times. Causes could be sink variability or a temporary reduction in industrial or climate-sensitive sources. We reconstructed the global history of [CH4] and its stable carbon isotopes from ice cores, archived air, and a global network of monitoring stations. A box-model analysis suggests that diminishing thermogenic emissions, probably from the fossil-fuel industry, and/or variations in the hydroxyl CH4 sink caused the [CH4] plateau. Thermogenic emissions did not resume to cause the renewed [CH4] rise after 2006, which contradicts emission inventories. Post-2006 source increases are predominantly biogenic, outside the Arctic, and arguably more consistent with agriculture than wetlands. If so, mitigating CH4 emissions must be balanced with the need for food production.

Anthropogenic CH4 emissions have almost tripled [CH4] since preindustrial times (13). This contributes strongly to anthropogenic climate change through radiative forcing and impacts on atmospheric chemistry, particularly hydroxyl consumption, tropospheric ozone generation, and water vapor formation in the stratosphere (4). In a positive feedback to climate change, natural sources such as CH4 hydrates, tundra, and permafrost may increase (5). We must therefore understand how the CH4 budget responds to human activities and environmental change. The onset and end of the 1999–2006 [CH4] plateau (Fig. 1) (3, 6, 7) have been studied with inverse models (top-down) (814), as well as process modeling (6, 8, 1520) and emission estimates (bottom-up) (2123). These approaches are either not emission-specific or uncertain in scaling and process representation (8). In contrast, the 13C/12C ratio in atmospheric CH413C(Atm); expressed in δ notation relative to the Vienna Pee Dee Belemnite standard] is controlled by the relative contributions from source types with distinctive isotope signatures δ13C(So) [biogenic ~–60 per mil (‰), such as wetlands, agriculture, and waste; thermogenic ~–37‰, such as fossil-fuels; pyrogenic ~–22‰, such as biomass burning] (3, 24). Large and overlapping ranges for δ13C(So) in field studies of the main source types and even individual sources (such as wetlands) (24) average out at the global scale so that δ13C(So) is suitable to characterize emissions. Sink processes with characteristic isotopic fractionation ε (25) [for example, hydroxyl (OH) ε = –3.9‰; chlorine in the marine boundary layer (Cl-MBL) ε = –60‰; stratospheric loss ε = –3‰; or oxidation by soils ε = –20‰] (table S1) (26, 27) also influence δ13C(Atm). Therefore, δ13C(Atm) variations indicate changes in CH4 budgets, in which pertinent sources are industrial (thermogenic); agricultural, such as ruminants and rice cultivation (biogenic); and climate-dependent, such as biomass burning (pyrogenic) and natural wetlands, including freshwater and permafrost (biogenic). Other sources lack magnitude [termites, wild animals, ocean, and hydrates (8)] or known processes (geologic sources) to force abrupt and sustained changes (supplementary materials). Changes in the dominating OH sink may affect [CH4] and δ13C(Atm) trends, whereas substantial changes in other sinks are unlikely or uncertain (supplementary materials).

Fig. 1 Global trends in [CH4] and δ13C(Atm).

(A) Spliced records of globally averaged annual values for [CH4] from a historic spline (HS) (light blue) (1) and the NOAA-ESRL global monitoring network (dark blue) (3). The uncertainty range is indicated by the thickness of the connecting line. (B) Spliced records of globally averaged annual values for δ13C(Atm) from a HS (yellow) (2) and atmospheric time series from contributing Global Atmosphere Watch (GAW) stations measured in our three laboratories (green). Gray shading shows the 1σ confidence interval (CI). Details on the splicing and uncertainty estimates are provided in (25).

We reconstructed [CH4] and δ13C(Atm) time series by splicing measurements from ice cores, firn air, archived air (1, 2), and global networks (Fig. 1, fig. S1, and tables S2 and S3) (3) (25). 13C enrichment followed by stable δ13C(Atm) parallels [CH4] trends until the end of the 1999–2006 plateau. Afterward, [CH4] increases, whereas δ13C(Atm) becomes more 13C-depleted. This suggests that the increasing emissions before and after the plateau differ in δ13C(So).

We used a one-box model (25, 27) to quantify changes in the CH4 budget. An inversion run derives the history of global emission strength and isotopic source signature [δ13C(So)] from the [CH4] and δ13C(Atm) reconstructions and specified sink parameters (tables S1 and S3). In forward mode, this “base source” as input reproduces measured [CH4] and δ13C(Atm) until the start of an event (plateau or renewed increase). Afterward, the source is held constant, providing a “Stabilization Run” (Fig. 2A). A superimposed “perturbation source” then tests the effect of strengthening or weakening emissions with a prescribed perturbation δ13C(So) on δ13C(Atm). Alternatively, sink variability can be implemented for equivalent tests. The modeling design is detailed in section 1.3 of (25).

Fig. 2 Box-model results for the onset of the 1999–2006 plateau.

(A to C) Observed yearly averages for [CH4] and δ13C(Atm) [black dots; gray shading for 1σ CI of δ13C(Atm)]. SR92 (red dashed lines) indicates trends in [CH4] and δ13C(Atm) for emissions held constant at average 1991–1992 (553 Tg/year) and 1982–1992 (–53.35‰) levels, respectively. (A) Subtracting a source perturbation (yearly varying, average –9.5 Tg/year; PRs92/1 and PRs92/2) reconciles modeled [CH4] (blue line) with observations. (B) PRs92/1: δ13C(Atm) trends if δ13C(So) values between –25‰ and –60‰ (solid lines) are assigned to a set of perturbation runs starting in 1993. Thick line indicates best-fit scenario [δ13C(So) = –40‰] for 1999–2006 observations. (C) PRs92/2: as above, but assigning –53‰ to all perturbation runs for 1993–1995. Best-fit results from perturbation δ13C(So) = –35‰. Total δ13C(So) values for all runs are shown in fig. S14 and table S4.

Stabilization Run 92 (SR92) tests whether emissions simply stabilized to cause the [CH4] plateau (assuming constant sinks) (28). The base source is run from 1700 to 1992, during which time emission rates show steady trends (fig. S2); afterward, emissions are held constant at 1991–1992 rates and average 1982–1992 δ13C(So). These choices remove disruptions by the Mount Pinatubo eruption (supplementary materials). Model-data mismatches after the plateau onset (Fig. 2) suggest a changing source mix and emission reductions. The latter occur abruptly after 1992, for an average 7.2 to 11.2 Tg loss in annual global emissions over 1993–2006 relative to 1991–1992 (fig. S3). In “Perturbation Runs” (PRs92/1), this emission loss is superimposed on the SR92 source as a negative perturbation that decreases [CH4] from the SR92 values to observations (Fig. 2A). By assigning different perturbation δ13C(So) values in PRs92/1 to match observed δ13C(Atm), we fingerprinted the emissions that are no longer contributing to the total source. A perturbation δ13C(So) of ~–40‰ fits the plateau values within uncertainties, although without truly leveling out (Fig. 2B). The match improves for perturbation δ13C(So) = –53‰ for 1993–1995 and –35‰ from 1996 (PRs92/2) (Fig. 2C). Alternatively, Stabilization Run SR92/OH with OH variability, as reconstructed from methyl-chloroform (29), and constant 1992–2006 emissions, approximate measured [CH4] and δ13C(Atm) trends, so that additional source perturbations (PRs92/OH) are small and have little impact (Fig. 3). Combined OH variability and emission reductions fit observations better, but the relative weight of the two processes remains unknown, and perturbation δ13C(So) of –35 to –40‰ stays within uncertainties of the emissions-only scenario (figs. S4 to S7).

Fig. 3 OH-variability scenario for the onset of the 1999–2006 plateau.

Symbols are as in Fig. 2. (A and B) SR92/OH (red dashed lines) includes OH variability (29) for 1994–2007, in addition to constant emissions after 1992 at average 1991–1992 and 1982–1992 levels for emission strength and δ13C(So), respectively. (A) In SR92/OH, sink variability (29) produces [CH4] close to observations, leaving little room for source perturbations in PRs92/OH (average –2.6 Tg/year; blue line). This results from an OH-induced trend to lower atmospheric CH4 residence time τ for 1993–1999, whereas the longer-term average (1993–2007) of τ is almost identical to the value used for the runs in Fig. 2. (B) PRs92/OH: best fit is for perturbation δ13C(So) = –30‰, but δ13C(Atm) trends for values between –25‰ and –50‰ all fit observations throughout most of the 1993–2006 perturbation period because OH variability and the associated changes in ɛ and τ alone can account for observed [CH4] and δ13C(Atm) trends. Total δ13C(So) values for all runs are shown in fig. S14 and listed in table S4.

δ13C(So) ~ –35‰ is characteristic for thermogenic CH4 (3, 24, 26), which is mainly emitted from the production of oil, natural gas, and coal (21, 22). Simultaneous biogenic and pyrogenic reductions could produce the same signal as thermogenic reductions. This seems unlikely because climatic events such as El Niño–Southern Oscillation phases force opposite emission changes in wetlands (15)—the major biogenic source—and the total of natural and anthropogenic biomass burning (16). The perturbation δ13C(So) 1σ uncertainty (–28 to –42‰) (supplementary materials) allows for a small probability that the perturbation is pyrogenic. However, the required 20% pyrogenic drop is inconsistent with reconstructions (8, 17), and the preplateau reduction did not occur in the tropics (28), where most pyrogenic emissions originate (8). Sink changes may have contributed to the [CH4] plateau, but only in concert with stagnating total emissions.

To study the renewed [CH4] rise, another Stabilization Run SR06 runs the base source until 2006 and then holds it constant at 1999–2006 averages for emission strength and δ13C(So). This simulates a continuation of the [CH4] and δ13C(Atm) plateaus (Fig. 4). Superimposed Perturbation Runs PRs06 with additional emissions averaging +19.7 Tg/year (fig. S3) [which is consistent with (13, 30)] reproduce the [CH4] rise and need δ13C(So) ~ –59‰ (–56 to –61‰, 1σ) (supplementary materials) to match the post-2006 δ13C(Atm) decline (Fig. 4B). Alternatively, Stabilization Run SR06/OH includes available OH reconstructions (1994–2007; constant OH assumed afterward). Associated Perturbation Runs PRs06/OH prescribe Perturbation A (2007–2011) and Perturbation B (2011–2014) to capture the marked break in slope of 2011. Best fits for Perturbation A ~ –75‰ and Perturbation B ~ –60‰ account for a 13C-rich anomaly in 2008 and match the differing δ13C(Atm) slopes (Fig. 4C). This scenario suffers from uncertainties regarding the transition from reconstructed to constant OH in 2007 (alternative OH trends are examined in the supplementary materials). Also, matching the 2008 δ13C(Atm) anomaly could skew the trend for subsequent years. The resulting bias may be seen in perturbation δ13C(So) ~ –75‰ for 2007–2011. Such extremely 13C-depleted values are only found in some boreal biogenic sources, which are unlikely to dominate the global signal. Therefore, the more conservative result is perturbation δ13C(So) ~ –59‰ integrated over 2007–2014 (Fig. 4B), but during 2006–2011, δ13C(So) was potentially more 13C-depleted, as shown in Fig. 4C. OH variability may contribute to the post-2006 event, but perturbation δ13C(So) remains within uncertainties of the above estimates (figs. S8 to S12).

Fig. 4 Box-model results for the [CH4] increase after 2006.

Symbols are as in Fig. 2. (A to C) Two Stabilization Runs (red dashed lines) SR06 [(A) and (B)] and SR06/OH [(A) and (C)] produce the same trend in [CH4], but different δ13C(Atm) for emissions held constant at average 1999–2006 levels from 2007 on. (A) PRs06 and PRs06/OH (yearly varying with average 19.7 Tg/year) reconcile modeled [CH4] (blue line) with observations. (B) δ13C(Atm) trends for PRs06 with perturbation δ13C(So) values between –70‰ and –45‰. The thick line indicates best fit scenario for δ13C(So) = –59‰. (C) PRs06/OH for a combined scenario that includes (i) OH reconstructions (29) when available (1994–2007, no OH variability afterward); (ii) Perturbation A for 2007–2011 overlain on the OH variability (blue and orange lines), best fit for perturbation δ13C(So) = –75‰ (magenta line); and (iii) Perturbation B for 2011–2014 (green lines, using best-fit Perturbation A until 2011), best fit for perturbation δ13C(So) ~ –60‰ (bright green line). Total δ13C(So) values for all runs are shown in fig. S15 and table S4.

δ13C(So) ~ –59‰ is characteristic for biogenic sources (3, 24, 26). Thermogenic or pyrogenic emissions would require compensating changes in other sources or sinks. An atmospheric general circulation model (AGCM)–based chemistry-transport model (ACTM) study recently found post-2006 biogenic increases relative to pyrogenic ones (12). That study prescribed thermogenic emissions and therefore did not test all sources independently and whether thermogenic emissions are contributing to the [CH4] growth. We calculate the possible contribution of thermogenic CH4 in a combined increase with biogenic emissions as 0.9 ± 4.8 Tg/year (1σ) from 19.7 Tg/year total (supplementary materials). Larger thermogenic contributions require pyrogenic reductions. Process-based models (18) find average pyrogenic reductions by –1.5 Tg between 1999–2006 and 2007–2014. This accommodates <5.5 Tg/year additional thermogenic emissions. However, emission inventories suggest variable (21) or increasing pyrogenic emissions (22), negating thermogenic increases. Biogenic increases and pyrogenic decreases together facilitate thermogenic contributions to the total increase of 0 to 33%. A more 13C-depleted perturbation δ13C(So), as suggested in Fig. 4C, would lower this estimate further. In all scenarios of simultaneous pyrogenic, thermogenic, and biogenic changes, increasing biogenic emissions are causing most or all of the post-2006 [CH4] growth. This finding remains robust for potential sink changes (supplementary materials).

The global CH4 budget is underconstrained by [CH4] and δ13C(Atm). We have tested concurring and compensating changes in the pertinent and better-known [CH4] sources and sinks. We cannot rule out that other combinations, or less understood processes (such as stratospheric-tropospheric exchange), explain or contribute to δ13C(Atm) trends. Three-dimensional inversions of regional variability in our data may provide further insights. Nevertheless, our findings allow for a likely reconstruction of recent CH4 budget changes.

The δ13C(Atm) history suggests increasing 13C-rich anthropogenic emissions since the industrial revolution (2). We show here that in the 1990s and early 2000s, 13C-rich emissions likely stagnated or decreased. This signal is muted after the Mount Pinatubo eruption decreased OH and wetland emissions (10), as seen in model-data discrepancies (Figs. 2B and 3B) and possibly higher perturbation δ13C(So) (Fig. 2C) for 1993–1995. After 1995, the most parsimonious explanation for the observed emissions decrease is thermogenic reductions. Previous δ13C(Atm) studies provided contradicting results for thermogenic emissions (9, 11, 31); only (11) found reductions for ~1988–2002. Our result is consistent with combined bottom-up and top-down reconstructions finding “decreasing-to-stable” fossil-fuel sources during plateau onset (8). Also, ethane levels indicate declining annual thermogenic CH4 emissions between 1984 and 2010 by >10 to 21 Tg (32), which is consistent with our 7.2 to 11.2 Tg/year average decrease. One possible cause is reduced fossil-fuel CH4 emissions by 12 to 20 Tg/year through a collapse in production after the Soviet Union breakup in 1991 (28). Our results therefore support previous evidence for thermogenic reductions. Alternatively, OH variability with stagnant emissions provides an equally plausible explanation.

After 2006, the activation of biogenic emissions caused the renewed [CH4] rise. The exact nature of this source is less clear. Possibly, emissions from waste treatment contributed, although their δ13C(So) ~ –55‰ is somewhat too 13C-enriched, and inventories indicate no step change around 2007 (21, 22). Arctic warming could have enhanced emissions from wetlands, thawing permafrost, and CH4 hydrates (5). This was detected for 2007 but not afterward (14). The onset of the post-2006 trend in δ13C(Atm) in the 60° to 90°N latitude band seems to lag other latitudes (fig. S13), and post-2006 emissions rose mostly in the tropics, as shown in satellite CH4 measurements (13). This footprint and the perturbation δ13C(So) fit tropical wetlands and agricultural emissions. Natural wetlands, the single largest CH4 source, have been implicated in the post-2006 [CH4] growth (8, 14) under enduring La Niña conditions (33). The associated combination of higher wetland (15) and lower pyrogenic emissions (16) could explain strongly 13C-depleted perturbations δ13C(So) for 2007–2011 (Fig. 4C). However, tropical wetland emissions are higher in the southern hemisphere (19), whereas remote sensing shows that [CH4] increased mainly in the northern tropics and subtropics (13). Also, tropical wetlands are relatively 13C-enriched (–52 to –60‰) and match our post-2006 perturbation not as well as rice cultivation (–59 to –65‰) and C3-fed ruminants (–60 to –74‰) (3, 24, 26). This isotopic evidence against tropical wetlands is not strong, given the ranges of reported δ13C(So) for various sources. However, sustained source 13C-depletion over 7 years with the potential for strong 13C-depletion until 2011 is harder to reconcile with tropical wetlands as compared with other biogenic emissions,such as agricultural ones. Inventories report increased annual agricultural emissions over the 2000–2006 average of ~12 Tg by 2011, dominated by ruminants (21, 23). This can largely account for the post-2006 [CH4] growth, estimated at 15 to 22 Tg/year (30). Also, India and China’s dominance in livestock emissions (23) and Southeast Asian rice cultivation are consistent with the location of the source increase (13). Although we cannot identify the specific biogenic source driving the [CH4] increase with certainty, it is compatible with agricultural emissions. If so, feedbacks between climate change and natural CH4 emissions (5) are not yet evident.

The finding of a predominantly biogenic post-2006 increase is robust. Further, it seems likely that fossil-fuel emissions stagnated or diminished in the 1990s. They are a minor contributor to the renewed [CH4] rise. This contradicts emission inventories reporting increases of all source types between 2005 and 2010 with a major (~60%) thermogenic contribution (21, 22). The predicted δ13C(So) ~ –48‰ (or more 13C-enriched) produces a slight δ13C(Atm) increase that cannot be reconciled with the measured marked decline (Fig. 4B). The finding is unexpected, given the recent boom in unconventional gas production and reported resurgence in coal mining and the Asian economy (21, 22). Our isotope-based analysis suggests that the [CH4] plateau marks not a temporary suppression of a particular source but a reconfiguration of the CH4 budget. Either food production or climate-sensitive natural emissions are the most probable causes of the current [CH4] increase. These scenarios may require different mitigation measures in the future.

Supplementary Materials

www.sciencemag.org/content/352/6281/80/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S21

Tables S1 to S5

References (3447)

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: We thank S. Montzka for supplying OH-reconstruction data. M. Harvey and K. Steinkamp provided helpful discussions. Feedback from two anonymous reviewers improved the manuscript. Antarctica NZ supports air sampling at Arrival Heights, Antarctica (ARH). Raw data for individual stations measured by INSTAAR and NIWA are available from the World Data Centre for Greenhouse Gases http://ds.data.jma.go.jp/gmd/wdcgg/introduction.html; data from INSTAAR are also available from ftp://aftp.cmdl.noaa.gov/data/trace_gases. Data from Heidelberg University are available from http://www.iup.uni-heidelberg.de/institut/forschung/groups/kk/en/Data_html. Data measured by University of Washington and University of California, Irvine, were taken from http://cdiac.ornl.gov/ndps/quay.html and http://cdiac.ornl.gov/epubs/db/db1022/db1022.html, respectively. Yearly δ13C values for individual stations and global annual averages are presented in tables S3 and S4. This research was supported by the Marsden Fund Council from New Zealand Government funding, administered by the Royal Society of New Zealand. Further support came from NIWA under Climate and Atmosphere Research Programme CAAC1504 (2014/15 SCI). The authors declare no competing interests. H.S. designed the data analysis; G.W.B., T.M.B., R.J.M., J.B.M., D.C.L., B.H.V., C.V., and S.E.M. performed δ13C measurements; C.V., J.B.M., I.L., D.C.L., and J.W.C.W. designed sampling and analytical programmes and performed data quality control; E.J.D. provided [CH4] data; K.R.L. and H.S. designed the box model; S.E.M.F. and H.S. performed uncertainty analyses; and all authors contributed to the interpretation and the writing of the manuscript.
View Abstract

Navigate This Article