Impact of interannual and multidecadal trends on methane-climate feedbacks and sensitivity

Methane is the second most important anthropogenic greenhouse gas. After relative stability in the early 2000s, atmospheric methane resumed growth after 2007 and accelerated since 2014, with records in 2020–2021. Concurrent declines in δ13CH4 have been attributed to (i) surging biogenic emissions, (ii) rising fossil-fuel emissions with reduced biomass burning, and (iii) weakening methane sinks. Methane-climate feedbacks arise via sources (positive, e.g., wetland and permafrost emissions) and sinks (negative, chiefly via hydroxyl radical, OH). However, complexities include secondary positive feedbacks mediated by CO and BVOCs that reduce OH, precipitation effects on wetlands and fires, SST influences on OH and Cl, and possible oceanic methane production by cyanobacteria and phytoplankton. The methane-lifetime feedback contributes the largest uncertainty to feedback strength. The research question is to quantify the causal contributions of temperature and precipitation to observed changes in CH4 and δ13CH4, differentiate positive versus negative feedbacks across space and time, and assess methane-climate feedback sensitivity over interannual to multidecadal scales.

Prior work reviewed identifies multiple drivers of the post-2007 methane increase: biogenic sources (wetlands, agriculture), fossil fuels, and changing sinks (OH, soil). Studies highlight uncertainties in wetland and permafrost feedbacks, ENSO impacts on wetland emissions, wildfire contributions, limited role yet potential variability of chlorine sinks, and the importance of atmospheric chemistry (OH, NOx, H2O vapor) in determining methane lifetime. Recent findings suggest widespread aerobic methane production in oceans by cyanobacteria and phytoplankton, adding uncertainty to oceanic feedbacks. IPCC AR6 summarizes net methane feedbacks but with substantial uncertainty, especially regarding OH trends and wildfire-related processes. These motivate a causal, data-driven quantification to constrain past CH4 and δ13CH4 variability.

The study partitions climate-driven (c) and non-climate-driven (nc) contributions to changes in atmospheric CH4 and δ13CH4 using a material balance framework and a causal analysis based on normalized information flow (nIF). The methane budget is expressed with emissions Q(T, Pr, Qnc) and lifetime τ(T, Pr, CH4), differentiating c- and nc-contributions with fractions σ and 1−σ. Monthly causal contributions from temperature and precipitation to changes in CH4 (and δ13CH4) are estimated using: ∂CH4(T)/dt = α × ηIF_T × dT/dt and ∂CH4(Pr)/dt = α × ηIF_Pr × dPr/dt, where ηIF is a practical normalized causal sensitivity derived from information flow (IF) between cause and effect time series normalized by the overall uncertainty flow to the effect. The sign of contributions is set by the sign of the interannual covariance (correlation) to distinguish positive versus negative feedbacks. A calibration factor α (maximal instantaneous causal sensitivity) is chosen by equating the largest observed peak in dCH4/dt (and d(δ13CH4)/dt) to the estimated total climate contribution at zonal and global scales (peak around 1997–1998 for CH4 and around 2009 for δ13CH4). Two spatial approaches separate direct oceanic and terrestrial influences: (i) exclusive means using land-only (LSAT, Pr) or sea-only (SST) fields at each latitude, and (ii) area-weighted land–sea means proportional to net flux contributions. C-contributions are estimated over moving 49-month windows (center ±24 months), enabling interannual analysis while capturing lagged effects; CH4 data span July 1983–December 2020 (centered months July 1985–December 2018), δ13CH4 spans 2000–2015. All variables are transformed to anomalies (CH4 and climate relative to 1984–2020 mean; δ13CH4 relative to 1998–2017 mean), seasonality removed, and rates computed as the difference between forward 12-month and backward 12-month means (e.g., dT/dt in °C yr−1). Zonal means are derived from gridded 3D fields via exclusive or area-weighted averaging, and global means from area-weighted zonal averages. Data sources: NOAA GML marine boundary layer CH4, WDCGG δ13CH4 from 23 station datasets, NOAA GHCN CAMS V2 LSAT (0.5°), NOAA OI SST V2 (1°), PREC/L precipitation over land, and GMST computed as area-weighted LSAT+SST. IF and nIF are computed using maximum-likelihood estimators; an empirically modified normalizer better approximates practical causal sensitivity than the traditional Liang normalizer. The methodology also reconstructs nc-contributions by subtracting area-mean c-contributions from observations. Methane-climate feedback sensitivity is computed as the integrated global c-contribution per °C of GMST change (ppb °C−1) and as annual mean c-contribution per °C (ppb yr−1 °C−1). Projections of long-term c-contributions versus GMST are obtained by regressing ln(sensitivity in ppb yr−1 °C−1) against ln(AGMST), filtering negative/extreme values, and extrapolating to 0.01–5 °C GMST anomalies. Sensitivity is analyzed for the entire record and post-2012 period to represent climate-stabilizing versus accelerated-warming decades.

- Interannual alternation of feedbacks: The climate-driven contributions to dCH4/dt and d(δ13CH4)/dt oscillate between positive and negative correlations interannually, especially in the tropics, reflecting alternating warming–drying versus cooling–wetting periods. These alternations involve sequences where positive LSAT-driven emissions (wetlands, fires) are followed by negative SST anomalies that weaken OH sinks, yielding additional positive contributions to CH4. - Peaks reproduced: Two major observed dCH4/dt peaks (1997–1998 and 2013–2016) are well captured by estimated climate contributions; 2020 shows a strong peak likely dominated by climate given reduced fossil sources. - Spatial patterns: Strong c-contributions align with known wetland regions (North America, western Siberia/Russia) and tropical sources (Southeast Asia, Amazon floodplains, NW Latin America). Paddy rice regions in South Asia show strong negative δ13CH4 signals consistent with biogenic emissions. Oceanic SST-related contributions show alternating positive/negative patterns, with strong signals in the East Pacific around 30°N during 2013–2015. - Mechanistic sequence: Identified processes include (i) negative contributions in cool years via reduced microbial emissions and strengthened OH sink; (ii) during warming–drying, either negative contributions via reduced anaerobic emissions/increased sinks or positive contributions via enhanced wetland emissions and fires emitting CH4/CO/BVOCs; (iii) secondary positive feedback via CO/BVOCs consuming OH; (iv) subsequent positive contributions via negative feedback as SST cools, lowering OH/Cl sinks; (v) precipitation-driven rewetting increasing emissions. - Role of oceans: Evidence suggests direct oceanic methane emissions linked to biological production (cyanobacteria and phytoplankton) contribute to positive SST feedbacks, with optimal temperatures ~27–37°C; methanotrophic oxidation peaks at 25–35°C. Positive SST feedbacks are more likely above ~10°C. SST influences also propagate indirectly via LSAT, Pr, and terrestrial OH. - Decadal regime shifts: From late-1980s through 1990s, negative feedbacks strengthened (higher OH anomalies) reducing climate contributions; after ~2007, δCH4(T)/dt increased and δ(δ13CH4(T))/dt decreased, indicating weakening sinks and/or stronger biogenic sources, with sharp strengthening post-2012/2013. - Sensitivity estimates: Long-term methane-climate feedback sensitivity stabilizes around ~200 ppb °C−1, equivalent to ~0.08 W m−2 °C−1, exceeding the mean net IPCC AR6 estimate (−0.02 W m−2 °C−1 net; 0.05 positive − 0.03 negative). Decomposition implies roughly 0.05 + 0.03 W m−2 °C−1 when accounting for positive contributions from lagged negative feedbacks. Sensitivity in ppb yr−1 °C−1 decreases over the record (with a post-2012 uptick), consistent with increasing OH sink and decadal feedback switching. - Variability with warming: Projections indicate widening gaps between c-contribution trajectories inferred from full-period versus post-2012 sensitivities as GMST rises, implying amplified interannual–decadal variability of methane-climate feedbacks at higher temperatures. - Isotopic constraints: Climate contributions better capture negative trends in δ13CH4 than positive ones, supporting the role of climate-driven processes in lowering δ13CH4. Post-2007, observed and climate-driven δ13CH4 trends align, suggesting offsetting nc contributions from fossil and agricultural sources.

The causal analysis demonstrates that climate-driven methane feedbacks are not monotonic but oscillate on interannual and multidecadal timescales. Alternating positive and negative feedbacks, especially in the tropics influenced by ENSO-related SST and precipitation variability, jointly raise atmospheric CH4 via lagged and nonlinear processes. Positive terrestrial feedbacks (wetlands, fires) often initiate sequences that, through chemistry-climate couplings (CO/BVOCs reducing OH), set up conditions where subsequent cooling reduces OH sinks, adding further positive contributions. Multidecadal shifts show periods dominated by stronger OH sinks followed by phases with weakening sinks and enhanced biogenic emissions. Oceanic influences include both indirect effects (via LSAT, Pr, OH) and likely direct methane production from cyanobacterial and phytoplankton blooms at suitable SSTs, explaining strong SST-linked positive feedbacks outside warmest tropical waters. The derived methane-climate feedback sensitivity (~0.08 W m−2 °C−1; ~200 ppb °C−1) suggests IPCC AR6 may underestimate net methane feedbacks because it does not fully account for positive contributions arising from lagged negative-feedback phases. Sensitivity peaks in boreal and tropical latitudes reflect wetland feedback dominance. With warming, the amplitude of feedback oscillations likely increases due to higher H2O vapor and OH variability, implying potential amplification of climate variability. Wildfire feedbacks may be underestimated if only direct emissions are considered; secondary chemistry and lagged sink responses can substantially enhance CH4 increases following fire years.

This study provides a causal, observation-constrained quantification of how temperature and precipitation drive interannual to multidecadal methane feedbacks affecting both CH4 concentrations and δ13CH4. It reveals alternating positive–negative feedbacks that collectively increase atmospheric CH4 through nonlinear, lagged processes, highlights significant oceanic contributions (both indirect and likely direct biological emissions), and estimates a higher historical methane-climate feedback sensitivity (~0.08 W m−2 °C−1; ~200 ppb °C−1) than reported in IPCC AR6. The findings imply that methane-related climate variability may intensify with further warming and that wildfire-associated feedbacks are larger than currently represented when secondary chemical effects are included. Future work should: (i) better separate direct oceanic from indirect SST-driven terrestrial influences; (ii) improve constraints on OH variability and its drivers; (iii) extend δ13CH4 observations for stronger isotopic attribution; (iv) quantify the duration and magnitude of lagged responses; and (v) assess how mitigation (rapid anthropogenic CH4 cuts, reduced ocean eutrophication) could dampen feedback variability.

Key limitations include: (1) Potential underestimation of negative ∂CH4(T,Pr)/dt due to concurrent anthropogenic emissions affecting observed peaks; (2) Use of a single calibration factor α across different drivers (LSAT, SST, Pr) and units; (3) Overlap between LSAT and Pr effects despite being treated as mutually exclusive when summed; (4) Short 49-month windows limit uncertainty quantification of IF/nIF; (5) Assumed locality of zonal CH4 for attributing gridded T and Pr influences; (6) Limited δ13CH4 data coverage increases reconstruction uncertainty; (7) Inability to fully distinguish direct oceanic from indirect SST effects on terrestrial processes and OH; (8) Region-specific anomalies (e.g., Northern Australia, Angolan uplands) lack definitive physical explanations; (9) Methodological caveat that proportionality to T or Pr (instead of dT/dt and dPr/dt) performs poorly, emphasizing reliance on rate-of-change which assumes certain hysteresis behaviors; (10) No formal uncertainty bounds for IF estimates due to short time series.