Significant feedbacks of wetland methane release on climate change and the causes of their uncertainty

Emissions from wetlands are the single largest source of the atmospheric greenhouse gas (GHG) methane (CH4). This may increase in a warming climate, leading to a positive feedback on climate change. For the first time, we extend interactive wetland CH4 emissions schemes to include the recently quantified, significant process of CH4 transfer through tropical trees. We constrain the parameterisations using a multi-site flux study, and biogeochemical and inversion models. This provides an estimate and uncertainty range in contemporary, large-scale wetland emissions and their response to temperature. To assess the potential for future wetland CH4 emissions to feedback on climate, the schemes are forced with simulated climate change using a ‘pattern-scaling’ system, which links altered atmospheric radiative forcing to meteorology changes. We perform multiple simulations emulating 34 Earth System Models over different anthropogenic GHG emissions scenarios (RCPs). We provide a detailed assessment of the causes of uncertainty in predicting wetland CH4–climate feedback. Despite the constraints applied, uncertainty from wetland CH4 emission modelling is greater that from projected climate spread (under a given RCP). Limited knowledge of contemporary global wetland emissions restricts model calibration, producing the largest individual cause of wetland parameterisation uncertainty. Wetland feedback causes an additional temperature increase between 0.6% and 5.5% over the 21st century, with a feedback on climate ranging from 0.01 to 0.11 Wm−2 K−1. Wetland CH4 emissions amplify atmospheric CH4 increases by up to a further possible 25.4% in one simulation, and reduce remaining allowed anthropogenic emissions to maintain the RCP2.6 temperature threshold by 8.0% on average.


Introduction
CH 4 is the second largest contributor to the anthropogenic greenhouse gas (GHG) effect after carbon dioxide (CO 2 ), contributing about 21% of the increased radiative forcing since pre-industrial times (Myhre et al 2013). Over a 100 year time period CH 4 global warming potential is about 28 times that of CO 2 (Myhre et al 2013). Surface atmospheric CH 4 concentrations reached 1810 ppb in 2012 (Saunois et al 2016), which is about 2.5 times as large as pre-industrial concentrations. In the late 1990s and early 2000s there was a near decade of minimal growth in atmospheric CH 4 , which has recently been followed by a renewed and sustained period of increase (Nisbet et al 2014).
Natural wetland emissions currently contribute up to approximately 40% of the global CH 4 emissions total (Saunois et al 2016), and are thought to cause much of the yearly atmospheric CH 4 concentration variability Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
The Saunois et al (2016) review shows that present-day wetland CH 4 global emissions remain highly uncertain, though there is some consistency between estimates using top-down (atmospheric inversions which incorporate observations into dynamic and chemistry models) and bottom-up (process-based land models) approaches, with ranges of 127-202 TgCH 4 yr −1 and 153-227 TgCH 4 yr −1 , respectively. The range in bottom-up estimates of Saunois et al (2016) is smaller than reported in similar earlier studies (Kirschke et al 2013, Melton et al 2013, which is mainly due to the use of a common wetland area map across models. Amazon basin observations have recently found that tropical trees act as significant conduits for wetland CH 4 emissions (Pangala et al 2017), transferring about half of the total wetland flux here. These measurements have resolved the previously large discrepancy between CH 4 flux estimates from scaled-up flux measurements and top-down approaches over this large regional source (Pangala et al 2017). However, lack of detailed knowledge of present-day wetland extent remains a large contributor to current emissions uncertainty in process-based models, as it is often used for scaling up (Kirschke et al 2013, Saunois et al 2016. Uncertainties in atmospheric chemistry and transport, and non-wetland CH 4 budgets contribute significantly to inversion model estimate uncertainty (Saunois et al 2016).
From a limited number of studies over different climate change scenarios, global wetland CH 4 emissions are projected to increase by up to 78% (with simulated climate change between present day and doubling of atmospheric CO 2 ), and enhanced future feedback raises radiative forcing by up to 0.1 Wm −2 K −1 (Arneth et al 2010, Ciais et al 2013. To extend this we take a more rigorous approach in assessing CH 4 wetland feedback under climate change. We consider separate parameterisations of the two primary wetland CH 4 generation processes, include CH 4 transfer through trees and a new representation of tropical soils, and adopt better constrained parameters. We then apply this in a framework of multiple emissions scenarios and 34 different Earth System Models (ESM) climate change estimates.

JULES wetland model
CH 4 consumption (methanotrophy) in the nonsaturated zone above the water table is far more efficient than methanogenesis occurring within the saturated zone (Roulet et al 1992). Hence we assume that CH 4 produced in the soil saturated zone can only reach the atmosphere directly when the water table reaches the soil surface (incorporated via the grid-box wetland fraction; Gedney et al 2004). CH 4 in the soil saturated zone may still be transferred to the atmosphere via vegetation when the water table is lower however. With emissions being so dependent on water table depth, it is necessary to simulate hydrology and wetland extent adequately.
Large-scale surface models traditionally use soil hydraulic properties based on sand, silt and clay percentages (Best et al 2011). In order to improve the hydrological depiction in the tropics, which is a major contributor to the global wetland CH 4 budget, we extend our soil properties to include tropical soils (ferralsols, which are weathered soils with micro-aggregated particles), as these are poorly represented by standard particle distribution functions (Marthews et al 2014).
JULES (Clark et al 2011) simulates the wetland fraction fw by combining simulated grid-box mean water table depth and sub-grid topographic distribution to produce a sub-grid water To incorporate CH 4 transfer through vegetation, including the significant flux via tropical trees (Pangala et al 2017), we extend the standard modelled fw to include an effective wetland extent fwe. We assume this flux through vegetation can occur where roots exist below the water table. The flux originating at soil depth z is therefore proportional to the relative occurrence of water table depth at z within the grid box fw z , ¢( ) and the fractional number of roots below z, P z . ( ) Root distribution decays exponentially up to total soil depth z soil, according to root depth z r : Integrating P z fw z ¢ ( ) ( ) gives: by definition. Summing over vegetation fractions fv i , we obtain an effective vegetation wetland fraction fwe v which encompasses vegetation transfer: Here i represents each vegetation type. Efficiency factor A relates to how effectively CH 4 is removed from the soil via vegetation. Over vegetation when the water table is at or above the surface, we assume that any soil CH 4 not lost through vegetation is emitted directly into the atmosphere. Hence the effective wetland fraction for soil is: Due to lack of detailed, available measurements, such as how vegetation flux varies with water table depth, A can only be calculated by ensuring the total fraction of CH 4 emitted via tropical Amazon trees agrees with that observed (∼0.5; Pangala et al 2017). Since Pangala et al (2017) is the only comprehensive, large-scale study on this (Covey and Megonigal 2019) we assume that the tuned 'A' value is valid over all vegetation types. Applying this 'A' value to different plant functional types is non-ideal, however its impact on total flux over other vegetation types is small partially due to shallower rooting depths. ( Figure S5 shows that including vegetation transfer has little effect outside tropical South America). This finding is consistent with McNorton et al (2016), which shows that incorporating additional processes, such as CH 4 transport via short vegetation, does not improve overall CH 4 flux simulation.

JULES CH 4 parameterisation
CH 4 production in wetlands occurs via two main pathways: acetoclastic and hydrogenotrophic methanogenesis. Acetic acid from root exudates (related to net primary productivity NPP; Bridgham et al 2013) is the substrate for the former, whereas CO 2, from the soil carbon store is a source for the latter. There is considerable uncertainty about the relative importance of the two substrates in CH 4 generation at the large-scale however, causing speculation that this is a large contributory factor in future wetland CH 4 flux uncertainty (Bridgham et al 2013). To place bounds on this uncertainty, JULES is run with two CH 4 parameterisations, where the source for wetland CH 4 flux into the atmosphere (F CH4 ) is based on either soil carbon Cs (kgm −2 ) or NPP (kgm −2 s −1 ). Due to the inclusion of vegetation-mediated flux, total CH 4 fluxes become dependent on an effective wetland fraction, fwe (where fwe fwe fwe : where T (K ) is the mean top 1 m soil temperature and To a reference temperature (273.16 K). F CH4 (NPP) is restricted from going below zero, as would otherwise occur if NPP becomes negative. (Soil uptake is included within the atmospheric methane lifetime). The choice of two substrates is important, as they are likely to have different climate sensitivities: F CH4 (Cs) is proportional to a large pool, which we assume is time invariant; F CH4 (NPP) is sensitive to a change in flux, with initial soil CH 4 emissions occurring less than twelve hours after photosynthetic assimilation (Megonigal et al 1999). Both schemes are sensitive to climate through their Q10 temperature relationships and effective wetland fraction ( fwe), with F CH4 (NPP) having an additional substrate sensitivity to climate and atmospheric CO 2 . These two schemes are chosen to see if uncertainties in the temporal production of available substrate is important in the climate-CH 4 feedback response. K CS and K NPP are tuned to produce the appropriate global total wetland flux to cover the total range for all models in Saunois et al (2016), 127-227 TgCH 4 yr −1 . (K CS varies from 3.2-5.0×10 -12 s −1 and K NPP 4.1-8.0×10 −3 (unitless) from UPP to LOW). (2) and (3) describe the amounts by which reaction rates increase with a 10 K temperature increase. These factors are themselves a function of temperature, so as to directly follow the Arrhenius equation which describes the temperature dependence of a biological process. Thus using a temperature dependent Q10 allows us to fit the Arrhenius equation beyond a specific climatic region. Hence Q T Q 10 10 ,

Q10(T) factors in equations
where Q10 To is the Q10 temperature sensitivity at To. Q10 CS,To and Q10 NPP,To are specific to the methanogenesis pathways considered.
The above scheme does not include some more detailed processes, such as CH 4 suppression due to sulphate deposition, some transport mechanisms, and multiple CH 4 pools. McNorton et al (2016) demonstrate that their inclusion does not improve overall model performance, and there are insufficient observations to adequately constrain many such processes

JULES wetland and CH 4 calibration and validation
In order to calibrate the JULES wetland CH 4 relations (JULES-CS, JULES-NPP) off-line runs are carried out using observation-based meteorology (Weedon et al 2014) at 0.5°resolution for all land points.
The wetland fraction simulated by JULES compares well spatially (figure 1) and temporally (figure S2) with the SWAMPS-GLWD (Poulter et al 2017) observation-based product, and is within observational spread for global total wetland area (figure 1(e), Davidson et al 2018). A detailed assessment of simulated wetland uncertainty and its impact on the magnitude of wetland CH 4 feedback on climate change is provided in the supplementary material SI sections S1 and S7.
To calibrate the soil carbon-based emission relation JULES-CS, temperature sensitivity Q10 To,CS is determined by fitting against Q10's in the multiple wetland site data analysis (Turetsky et al 2014, Q10 Tur , section S2). To directly map Q10 To,CS onto Q10 Tur , JULES-CS must use fixed (observed) soil carbon (Nachtergaele and Batjes 2012), which is equivalent to assuming a highly recalcitrant and/or large soil carbon pool that changes negligibly over time. Using this fitting procedure we produce lower (LOW), mid (MID) and upper (UPP) estimates of Q10 CS,To of 3.0, 3.7 and 4.7 (equivalent to poor and rich fens, and bogs in Turetsky et al 2014), respectively (SI table S1).
The intermediate Q10 CS,To value of 3.7 obtained from the above bottom-up approach here is very close to that found in Gedney and Cox (2003) where the CH 4 parameterisation was calibrated against global inter-annual variations in atmospheric CH 4 (a top-down approach). Although Turetsky et al (2014) focuses on bogs and fens, Yvon-Durocher et al (2014) demonstrate that for multiple eco-systems spanning the globe, CH 4 emissions have a consistent mean temperature dependence (0.96 eV with a 95% confidence of 0.86-1.07 eV). This is numerically equivalent to Q10 To =4.5 and range of 3.8-5.3 (as 1 eV ∼1.6×10 −19 J), and consistent with our estimates of 3.0-4.7. The resulting regional distribution of JULES CH 4 fluxes compare well against other  table S2). (As we have tuned the tree-mediated fluxes to be 50% of the total, and our flux totals agree with observations, it follows that there is also good agreement between the tree-mediated fluxes: JULES 12.3-27.0, Pangala et al 2017 11.5-26.2 TgCH 4 yr −1 ). The aforementioned comparisons demonstrate consistency across very different calibration approaches and different spatial and temporal scales, indicating that Q10 CS,To is appropriately calibrated for a globalscale study.
We are able to use the regionally-calibrated Q10 CS,To globally because CH 4 fluxes are observed to have very similar temperature dependencies for a wide range of eco-systems and regions (Yvon-Durocher et al 2014). The above mapping approach cannot be directly applied to the NPP substrate-based CH 4 emissions relation (JULES-NPP) however, because NPP is itself dependent on soil moisture and temperature. Instead, we calibrate JULES-NPP by comparing against regional fluxes across the optimised JULES-CS (section S2). The resulting values of Q10 NPP,To for poor and rich fens, and bogs are 1.3, 1.6 and 2.3, respectively (SI figure S3(c), table S1).
From these comparisons with multiple measurements and models over different spatial and temporal scales, we demonstrate that both the JULES wetland extent and CH 4 emission parameterisations respond appropriately to temperature and precipitation and are therefore suitable for the analysis of large-scale CH 4 emissions in future climate scenarios.

Climate change simulations
To analyse the potential for wetland CH 4 emissions to feedback on climate change, multiple simulations are performed using JULES coupled to a climate change impacts model (IMOGEN, section S5

Analysing the main feedback drivers and sources of uncertainties
To rigorously assess the relative importance of the drivers of CH 4 emission changes, JULES-CS and JULES-NPP, equations (2) and (3) S8). Hence future global growth in wetland CH 4 flux is dominated by the tropics as this is the main source of present-day emissions.
To understand the causes of these flux changes and their uncertainties, we present the impact of temperature, precipitation and NPP (light and dark blue and yellow symbols, respectively; figure 2) on global CH 4 flux changes (equations (6) and (7)). Wetland extent responses are dominated by differences in precipitation projections (as demonstrated by comparing figures 2 and S7) which are themselves highly uncertain. Consequently the simulations do not predict a consistent expansion or contraction over most wetland regions (figures 1(g) and (h)). There is even a lack of consensus as to whether the global total wetland area is likely to increase or decrease in the future. (For example, for RCP8.5 simulations it varies from a 2% decrease to a 19% increase between 2000 and 2100). Hence projected changes in precipitation result in simulated CH 4 flux changes of uncertain sign. The resulting flux changes are of relatively small magnitude however. Projected temperature change is a more important driver than precipitation, and results in both a larger CH 4 flux response and a greater uncertainty (figure 2). For JULES-NPP, some of the temperature response is within the NPP term. As well as being dependent on temperature, NPP is also strongly driven by atmospheric CO 2 (through increased CO 2 fertilisation), which is in itself highly correlated with temperature.
In both substrate parameterisations the spread in projected wetland CH 4 flux by year 2100 is dominated by uncertainty in the wetland emissions parameterisation (figure 3). The uncertainty due to simulated climate change (within a specific RCP) plays a smaller, but still significant role (63%-65% versus 35%-37% over the three RCP scenarios-only RCP4.5 shown in figure 3(a)). Within the wetland scheme, uncertainty due to present-day global wetland flux estimate is larger than that from temperature sensitivity (Q10 To , figure 3(a)). Moreover the projected spread due to global present-day wetland emissions uncertainty alone is  (6) and (7)): precipitation (dark blue); temperature (light blue); NPP (yellow) and total (red). 5%-95% confidence intervals shown for: IMOGEN simulated modelled flux (vertical lines), and flux due to each/combined drivers (horizontal lines), with climate only (solid), and climate+JULES CH 4 scheme (dotted-dashed). One-to-one and zero x-axis are shown as black dotted and dashed, respectively. comparable to that from simulated climate change. Critically, in terms of CH 4 generation process uncertainty, the impact of substrate source used (i.e. Cs versus NPP), is relatively small (between 5% and 13% for the different scenarios). We also find that the wetland physics sensitivity experiments have little impact on the projected future wetland flux (section S7).
Enhanced wetland CH 4 emissions cause substantial 21st century increases in atmospheric CH 4 concentration of up to 25.4% above that with fixed wetland emissions (and 23.6% without including the CH 4 atmospheric lifetime sensitivity experiments) (figures 3(c) and 4). Depending on the scenario, the corresponding 21st century increases in total and CH 4 radiative forcings are further enhanced by between 0.7%-10.0% and 3.5%-40.0%, respectively. The percentage changes in CH 4 radiative forcing are higher for lower RCP scenarios (figures 3(c) and (d)) because CH 4 radiative forcing is nonlinearly dependent on concentration (Etminan et al 2016). This radiative forcing enhancement leads to an additional increase in global air temperature of up to 0.21 K at 2100 (and land air temperature of up to 0.27 K -not shown). The relative causes of uncertainty in the enhanced total radiative forcing and percentage temperature changes are similar to those in wetland emissions (figure 3). In addition, atmospheric CH 4 lifetime uncertainty contributes around 7% of the total spread in radiative forcing and temperature changes.
Due to wetland emissions feedback, significant reductions in allowed anthropogenic emissions are required to maintain projected temperature changes between 2020 and 2100 (appendix SB, table 1). For RCP2.6 these reductions are 21.4 GtC, with a likely value (greater than 68% probability) of between 12.4-30.4 GtC (4.6%-11.3%). This is consistent with the Comyn-Platt et al (2018) estimate of 19.6 GtC for the comparable 1.5°C stabilisation threshold. (This is despite Comyn-Platt et al 2018 using a different substrate generation scheme, thereby further demonstrating that the impact of substrate used is relatively small). Near-term implications are significant with the RCP2.6 mid 21st century global temperature peak reached nine years earlier (4-16 years) (not shown). CH 4 emissions from wetlands increase significantly from present day to 2100 in all simulated climate change patterns and RCP scenarios. In the higher sensitivity wetland emission relation, they may more than double between 2000 and 2100, with an associated change in atmospheric CH 4 concentration of up to 952 ppbv (figure 4). The subsequent feedback on radiative forcing ranges from 0.01 to 0.11 Wm −2 K −1 . Despite the improved physical representation in JULES, this has little impact on the feedback magnitude of the wetland emissions on climate at the global scale (section S7).

Discussion and conclusions
There is also considerable uncertainty in other inland water CH 4 source emissions (Saunois et al 2016) and these are also likely to respond to climate change (Ciais et al 2013). Present-day rice agriculture emissions are estimated to be around 36 TgCH 4 yr −1 which is between 16% and 30% of natural wetland emission estimates (Saunois et al 2016). Moreover, even without rice agriculture expanding, their associated CH 4 emissions are likely to increase in response to temperature (Khalil et al 1998), adding an additional radiative feedback.
We have included new physical processes: the CH 4 transfer through tropical trees and the inclusion of tropical-specific soils, to improve wetland representation  in this important region. This allows us to make more accurate assessments of the feedback magnitude between wetland CH 4 release and climate change. We have determined this feedback uncertainty by constraining wetland parameterisations, using data from multiple sources over local to regional scales. Despite this, there remains a sizeable feedback uncertainty, which is dominated by relatively poorly known wetland extent and CH 4 parameters, rather than that due to climate change. Our more rigorous and constrained approach helps explain the IPCC's assessment of 'low confidence' in the magnitude of the wetland CH 4 -climate feedback (Ciais et al 2013). Furthermore, we identify the main causes of the uncertainty range, which can be further constrained through future measurement campaigns focusing on detailed wetland processes, including those determining the vegetation transport of CH 4 (Covey and Megonigal 2019). Given that the remaining lack of knowledge in contemporary global wetland CH 4 flux is the single largest cause of wetland scheme uncertainty, improved estimates of present-day global wetland emissions remain a research priority. The CH 4 feedback on climate may be enhanced or reduced through interactions with other parts of the earth system through the intricate coupling of biogeochemical cycling and reactive atmospheric chemistry. For instance, we do not consider the interaction with the carbon cycle in which changing atmospheric CO 2 and climate may alter the size and distribution of vegetation and soil carbon, and in turn the available substrate for methanogenesis. Understanding these interactions, and their potential nonlinearities, will become possible as the processes analysed in this study are routinely incorporated in to the next generation of ESM, and we suggest this should be a priority.
Under a range of future socio-economic assumptions, from high GHG emission mitigation to 'business-as-usual' scenarios, we find that wetland CH 4 feedbacks in a warming climate will significantly augment atmospheric CH 4 concentrations by up to 25.4% over the 21st century. This raises CH 4 radiative forcing significantly beyond that caused by direct human activity, producing a positive feedback that further enhances climate change. Over the scenarios considered, the wetland feedback amplifies the 21st century CH 4 radiative forcing on average by 14.4%, with uncertainty estimates indicating values as high 40.0% for the RCP4.5 scenario. This generates an average additional warming of 2.4%. Under the RCP2.6 emissions scenario (which gives an approximately 1.5°C increase in temperature from pre-industrial to 2100), the 21st century temperature rise is amplified by up to 5.5%, and furthermore rises more quickly in the near term. This corresponds to a likely reduction in remaining allowed anthropogenic emissions of between 4.6%-11.3% in order to maintain the same temperature profiles. The more rapid warming and carbon budget reduction, imply that a combination of enhanced short-term emission cuts and longer-term increased use of negative emission technology, are required to be consistent with the Paris climate targets.