Role of regional wetland emissions in atmospheric methane variability

Atmospheric methane (CH4) accounts for ~20% of the total direct anthropogenic radiative forcing by long‐lived greenhouse gases. Surface observations show a pause (1999–2006) followed by a resumption in CH4 growth, which remain largely unexplained. Using a land surface model, we estimate wetland CH4 emissions from 1993 to 2014 and study the regional contributions to changes in atmospheric CH4. Atmospheric model simulations using these emissions, together with other sources, compare well with surface and satellite CH4 data. Modeled global wetland emissions vary by ±3%/yr (σ = 4.8 Tg), mainly due to precipitation‐induced changes in wetland area, but the integrated effect makes only a small contribution to the pause in CH4 growth from 1999 to 2006. Increasing temperature, which increases wetland area, drives a long‐term trend in wetland CH4 emissions of +0.2%/yr (1999 to 2014). The increased growth post‐2006 was partly caused by increased wetland emissions (+3%), mainly from Tropical Asia, Southern Africa, and Australia.


Introduction
Two outstanding features of the recent global atmospheric CH 4 record are as follows: (i) prior to 2007 atmospheric CH 4 seemed to be approaching a stationary state with sources and sinks in balance and (ii) since 2007 growth has resumed [Rigby et al., 2008]. The reasons for these two features are not fully understood, but several indicators suggest that changes in wetlands contributed to both. First, wetland emissions are the largest single source of CH 4 to the atmosphere (~175 Tg/yr, 30% of total) and are known to be sensitive to climate variation. Thus, interannual variations will have an effect at the global scale [Bousquet et al., 2006;Ciais et al., 2014] and may produce a positive climate feedback in the future [Gedney et al., 2004;Melton et al., 2013]. Second, the isotopic signature of atmospheric CH 4 has become more depleted in 13 CH 4 since 2007, which may be explained by a shift toward a larger fraction of biogenic sources, i.e., wetlands [Nisbet et al., 2014], or agricultural sources [Schaefer et al., 2016]. However, both of these studies focused on the global isotopic signature of CH 4 and so were unable to spatially resolve the regions of main CH 4 emission changes.
Previous investigations into wetland CH 4 emissions using land surface models (LSMs) have shown large differences between estimates, both in magnitude and the spatial and temporal distribution [Wania et al., 2010;Riley et al., 2011;Kirschke et al., 2013;Melton et al., 2013]. Most of these studies were not evaluated against observations of atmospheric CH 4 concentrations, in particular recent column observations [Wunch et al., 2011;Parker et al., 2015]. Studies using atmospheric models forced by a combination of modeled wetland emissions and nonwetland source estimates were not fully consistent with both in situ and space-based CH 4 measurements [e.g., Patra et al., 2011;Fraser et al., 2013;Hayman et al., 2014]. Meng et al. [2015] showed good agreement between surface measurements and model simulations when using a wetland CH 4 model, but they did not include emissions after 2005. This was, in part, due to a reliance on the Global Inundation Extent from Multi-Satellites product for estimated wetland area extent and time variation, which only covers the period from 1993 to 2007 [Prigent et al., 2012]. It is therefore preferable to have an emission model that is able to predict wetland extent accurately and independently of observational data. model include improved representation of topography, which provides a more realistic spatial extent and temporal evolution of flooded areas [Marthews et al., 2015], and inclusion of dynamic soil carbon pools [Clark et al., 2011]. We first compare our modeled wetland CH 4 emissions with a top-down estimate available for 2003-2011 [Bloom et al., 2012]. We then combine the derived wetland CH 4 emissions with nonwetland emission estimates for use in an atmospheric chemical transport model (CTM), TOMCAT, and compare the simulated concentrations to long-term ground-based observations. We have taken advantage, for the first time, of the~5 year atmospheric column CH 4 measurements from the Greenhouse gases Observing SATellite (GOSAT) remote sensing mission [Parker et al., 2011[Parker et al., , 2015 to further evaluate the model. We demonstrate good agreement with both satellite-derived and surface CH 4 data, which allows us to investigate the role of wetland emissions in atmospheric CH 4 variability.

Wetland Flux Models
Wetlands are areas where the soil is saturated, either permanently or seasonally [Melton et al., 2013]. Within anaerobic wetland regions methanogenesis occurs [e.g., Christensen et al., 2003]. This CH 4 production process is primarily dependent on available substrate and soil temperature. Additional processes, which influence emissions, have been included in other LSMs [e.g., Wania et al., 2010;Riley et al., 2011]. We evaluated the importance of these within JULES by comparing simulations against surface flux observations and found that their inclusion currently provides either limited or no improvement on model performance (see supporting information) [Gauci et al., 2004;Turnock et al., 2015]. These comparisons highlight uncertainties in wetland emissions at a site-specific scale (R = 0.32). Therefore, due to process uncertainties, our analysis considers only temperature, wetland area, and organic substrate as controls of wetland CH 4 emissions.
We use JULES v3.4.1 [Clark et al., 2011] to derive spatially and temporally resolved global wetland CH 4 emissions from 1993 to 2014 (hereafter labelled JU). The model was forced by 3-hourly WATCH-forcingdata-ERA-interim (WFDEI) at a horizontal resolution of 0.5°× 0.5° [Weedon et al., 2014]. The rainfall component of WFDEI was bias corrected using Global Precipitation Climatology Centre (GPCC) and Climatic Research Unit measurements. Wetland fraction derivations were based on an improved topographic index within the JULES-TOPography-based hydrological MODEL, which produces significant improvements compared with previous versions [Marthews et al., 2015]. The topographic index and water table depth, which is controlled by precipitation, evaporation, and runoff, are the key controls of wetland area. For the wetland CH 4 flux estimation, equation (1) in Gedney et al. [2004] was updated to use the four carbon pools generated as part of both the RothC model [Jenkinson et al., 1990;Coleman and Jenkinson, 1999] and the Top-down Representation of Interactive Foliage and Flora Including Dynamics [Cox, 2001] model within JULES. The updated flux model is as follows: The total temperature independent methanogenesis (f sub , in kg m À2 s À1 ) is based on the four carbon pools of decomposable plant material (C dpm , kg m À2 ), resistant plant material (C rpm ), biomass (C bio ), and long-lived humus (C hum ), which are multiplied by their respective methanogenesis rates (R dpm/rpm/bio/hum , s À1 ). These rates used to calculate a CH 4 flux (F CH4 , kg m À2 s À1 ) are assumed to be the same as the respiration rates used in Clark et al. [2011] (see supporting information). The soil temperature (T soil , K) is averaged over the top 10 cm, and the reference temperature (T 0 ) is 273.15 K. At temperatures below 273.15 K the ground is considered frozen with zero surface flux. An effective Q 10 value, based on the reaction rate change for a 10 K change in temperature, was initially set at 3, based on McNorton et al. [2016a], and varies with temperature between 2.5 and 3 as given by equation (2)  First, we use top-down CH 4 emission data derived using the method of Bloom et al. [2012] (hereafter labelled BL) to evaluate JU. BL is based on column CH 4 (XCH 4 ) retrievals from the Scanning Imaging Absorption Spectrometer for Atmospheric Chartography, gravity anomaly-derived equivalent water height Geophysical Research Letters 10.1002/2016GL070649 from the Gravity Recovery and Climate Experiment, proxy soil temperature from the National Centers for Environmental Prediction/National Center for Atmospheric Research skin temperature analyses, and an assumed surface carbon input. This data set provides interannually varying emissions between 2003 and 2011; a climatology is used for 1993-2002 and 2012-2014. BL emissions, combined with nonwetland emissions, have previously been shown to compare well with observed surface concentrations when used as a boundary condition in an atmospheric CTM [Bloom et al., 2012]. We therefore use them as a benchmark for JU for the period 2003-2011. JU is suitable for the study of long-term trends because it is not limited by availability of satellite data. As with JU, we scaled BL to provide mean global emissions of 175 Tg/yr between 2000 and 2009.

Atmospheric CH 4 Model
We used the TOMCAT global atmospheric 3-D off-line CTM [Chipperfield, 2006] to predict atmospheric CH 4 concentrations given surface emissions. We performed three simulations, two of which used interannually varying wetlands (TOMCAT-JU and TOMCAT-BL) and the third used a climatology of JULES emissions between 1993 and 2014 (TOMCAT-CL). Simulations were forced by the 6-hourly European Centre for Medium-Range Weather Forecasts ERA-Interim reanalyses [Dee et al., 2011] for the period 1993 to 2014 with a horizontal resolution of 2.8°× 2.8°and 60 levels from the surface to~60 km. The model was spun-up from 1977 to 1992 using interannually varying emissions when available; these include JULES wetland fluxes then the mean global CH 4 was rescaled to match observed concentrations in 1993.
Anthropogenic emissions were taken from the EDGAR v4.2 FT 2010 database, which gradually increase between 1993 and 2010, with 2011-2014 emissions set to repeat 2010 [Olivier et al., 2012]. Previous studies have suggested a range of overestimates in the growth rate in EDGAR anthropogenic emissions Monteil et al., 2011;Patra et al., 2016]. The magnitude of the overestimated growth is highly uncertain; as an estimate, we reduced the growth rate between 2000 and 2010 by a factor of 2. Biomass burning emissions were taken from the GFED v4 database [van der Werf et al., 2010[van der Werf et al., ] between 1997[van der Werf et al., and 2014[van der Werf et al., , with 1993[van der Werf et al., -1996 emissions based on a climatology. Annually repeating hydrate, mud volcano, and termite and ocean [Matthews and Fung, 1987] emissions were taken from the TransCom CH 4 study [Patra et al., 2011]. JULES only simulates natural wetlands and does not account for agricultural practices (e.g., fertilization) altering the biogeochemical processes. To account for rice paddies in areas of natural wetlands, a mask (based on the MICRA2000 rice fraction map) [Portmann et al., 2010] was placed over the JU and BL wetlands and annual repeating emissions [Yan et al., 2009] were used instead. Any climate-driven influences on rice emissions are not accounted for. All nonwetland emissions are scaled to 2000-2009 values provided by Ciais et al. [2014] (see supporting information). Patra et al. [2011], where loss occurs through chemical reactions with OH, Cl, O( 1 D), and a soil sink. We use the same model OH fields as McNorton et al. [2016b]. These are a development of fields used by Patra et al. [2011], which were derived from a combination of semiempirically calculated tropospheric distributions and 2-D model stratospheric values [Spivakovsky et al., 2000;Huijnen et al., 2010]. We use CH 3 CCl 3 anomalies to vary the model OH field interannually between 1997 and 2007 [Montzka et al., 2011;McNorton et al., 2016b]. For the other years the original Patra et al. [2011] field is used. The Cl, O( 1 D), and soil sink fields do not vary interannually.

Satellite CH 4 Data
We use the University of Leicester Proxy dry air column-averaged mole fractions of CH 4 (XCH 4 ) v6.0 data [Parker et al., 2011[Parker et al., , 2015 derived from observations made by the TANSO-FTS instrument on board the GOSAT satellite [Kuze et al., 2009]. Retrievals of XCH 4 are sensitive to the surface and lower troposphere, hence providing information relating to emissions of CH 4 . GOSAT has provided global XCH 4 data since 2009 with a single-sounding precision between 0.4 and 0.8%, making it an ideal tool for the assessment of modeled atmospheric CH 4 .
To compare TOMCAT to GOSAT, we take the nearest model profile, both spatially and temporally, to the GOSAT retrieval. We also apply the sounding-specific GOSAT averaging kernel to the model profile to account for the differing sensitivities throughout the atmosphere, allowing the most robust comparison between measurement and model. Both GOSAT and model output are then binned into TransCom regions and temporally into monthly periods between 2009 and 2014.

Ground-Based CH 4 Observations
We use atmospheric CH 4 measurements from the Total Carbon Column Observing Network (TCCON) (see supporting information) [Wunch et al., 2011] and the National Oceanic and Atmospheric Administration/Earth System Research Laboratory (NOAA/ESRL) sampling sites [Dlugokencky et al., 2015]. These allow us to assess the model's ability to reproduce the observed long-term trends in CH 4 growth. To derive global surface CH 4 concentrations, we have used observations from 19 NOAA sites [see McNorton et al., 2016b, Table 1] interpolated across 180 latitude bins, which were then weighted by surface area. The process was repeated for model output sampled at the site locations. Figure 1 compares the JU and BL emissions globally and for defined TransCom regions [see Gurney et al., 2002]. Annual emissions from JU compare well with BL in most regions although JU emissions tend to be noticeably higher in tropical regions and lower in boreal regions, caused in part by JU (Q 10 (T 0 ) = 3) having a higher temperature sensitivity than BL (Q 10 (T 0 ) = 1.65). Methanogenesis temperature sensitivity varies depending on conditions; as a result there is uncertainty when assuming a global value [Segers, 1998].  (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011). Nearly all of the growth trends in JU originate from Europe (0.07 Tg/yr), Tropical Asia (0.13 Tg/yr), Southern Africa (0.05 Tg/yr), and Australia (0.10 Tg/yr), whereas trends in BL are largest from Eurasian Boreal (0.11 Tg/yr), Northern Africa (0.10 Tg/yr), and Southern Africa (0.13 Tg/yr). The spatial difference in trends occurs not only as a result of model differences but also as a result of the time periods covered by each model. When emissions are divided into the time periods before, during, and after the observed CH 4 pause, there appears to be no noticeable decrease in JU emissions during the pause (Figure 1b). However, modeled emissions are typically slightly above average early in the pause period (1999)(2000)(2001)179.9 Tg/yr) and smaller in the later pause period (2002)(2003)(2004)(2005)(2006)171.5 Tg/yr), indicating that changes in wetland emissions were not responsible for the initial stall in CH 4 growth but may have contributed later on.  Figure 2b shows positive long-term trends in JU wetland CH 4 emissions over Southeast Asia, northern Australia, Zambia, southern Democratic Republic of Congo, and the seasonally flooded savannas in tropical South America. Distinct regions with negative trends are less apparent, with the exception of the Iberá wetlands (Paraná river, NE, Argentina).

Results and Discussion
Variability in wetland fraction, soil temperature, and carbon substrate is the cause of trends in our modeled JU wetland CH 4 emissions. We investigated anomalies in these parameters both zonally (Figures 3a, 3c, and 3d) and globally (Figure 3e). Considerable wetland area anomalies occur at both boreal and tropical latitude bands, with an anomalously low tropical wetland area during both the 1997/1998 El Niño event and the later period of the pause in growth (2002)(2003)(2004)(2005)(2006) and anomalously high post-2006. Figures 3b and 3f show interannual variability in wetland area that occurs as a result of changes in the reanalysis rainfall over wetland regions [Harris et al., 2014;Weedon et al., 2014].
Soil temperature anomalies in wetland regions predicted by JULES are heavily weighted toward the tropics (Figure 3c), whereas carbon substrate flux anomalies are weighted toward the northern boreal regions (Figure 3d). Our results show positive Northern Hemisphere tropical wetland temperature anomalies during the 1997/1998 El Niño event and in 2010. Negative tropical wetland temperature anomalies are found early in the simulation from 1993 to 2001 (excluding 1997/1998).
To further investigate these trends, we calculated grid cell coefficient anomalies, which are based on the flux as a function of each variable, to satisfy the following rewrite of equation (2): ) weighted by the JULES climatological wetland area, taken from WFDEI using GPCC and CRU corrected rainfall [Harris et al., 2014;Weedon et al., 2014]. (c) Similar to Figure 3a but for upper layer soil temperature (°C) weighted by wetland area, from JULES. (d) Similar to Figure 3a but for the temperature-independent substrate carbon (gCm À2 s -1 ) weighted by wetland area from JULES. (e) Global monthly wetland fraction (f w , blue), temperature (f tem , red), and substrate flux (f sub , green) coefficient anomalies (%) weighted by wetland area between 1993 and 2014 and smoothed over a 12 month period. Note that the wetland area weighting varies with time; therefore, some temperature and substrate flux anomalies occur as a result of the variability in wetland area. (f) Global monthly rainfall anomaly (%) over the JULES climatological wetland area, taken from WFDEI using GPCC and CRU-corrected rainfall [Harris et al., 2014;Weedon et al., 2014], smoothed over a 12 month period.

10.1002/2016GL070649
The temperature and substrate coefficients were then weighted by grid cell wetland fraction and grid cell area to provide a global coefficient, from which monthly anomalies were calculated.
Globally, there is a statistically significant (95% level) positive trend in the wetland area CH 4 flux coefficient (f wet ) of 0.16%/yr over the 1993-2014 period (Figure 3e). There is a negative trend in rainfall over the model climatological wetland area, although this is not statistically significant (p value = 0.09) (Figure 3f). The trend in temperature flux coefficient (f tem ) anomaly is positive (0.13%/yr) and statistically significant. This trend causes high-latitude thawing, which controls the long-term trend in wetland area. The carbon substrate flux coefficient (f sub ) anomaly trend is slightly negative (À0.02%/yr) but is not statistically significant. Both the modeled CH 4 emission trend  and the interannual flux variability are dominated by changes in wetland area and temperature with small contributions from substrate, which is consistent with previous findings [e.g., Ringeval et al., 2010]. The reduction in our modeled wetland emissions from 2002 to 2006 is the result of a decrease in wetland area (f wet : À2.0%), while the increase since 2007 is mainly the result of increases in wetland area (f wet : +1.6%) and also temperature (f tem : +0.7%). Through comparison of simulated and observed atmospheric SF 6 concentrations, Wilson et al. [2014] showed that interhemispheric transport in TOMCAT is somewhat too slow [see also Patra et al., 2011]. This causes TOMCAT to slightly overestimate (underestimate) Northern (Southern) Hemisphere CH 4 concentrations. The correlation, mean bias, and root-mean-square error (RMSE) values taken from comparisons show that TOMCAT-JU produces better agreement with GOSAT (R = 0.88, mean bias = 11.1 ppb, and RMSE = 7.7 ppb), compared to TOMCAT-BL (R = 0.84, mean bias = 13.4 ppb, and RMSE = 8.9 ppb) and TOMCAT-CL (R = 0.87, mean bias = 11.2, and RMSE = 7.7 ppb) ( Figure 4b). Assuming that the model nonwetland CH Comparisons with observations indicate that TOMCAT simulations provide a reasonable representation of atmospheric CH 4 , although large uncertainties remain in atmospheric loss and nonwetland emission estimates. This suggests that JU emissions provide a reasonable representation of the spatial and interannual variability and are therefore suitable in the detection of wetland CH 4 trends. TOMCAT-JU (7.54 ppb/yr) captures most of the observed renewed growth post-2006 (8.12 ppb/yr), whereas TOMCAT-CL only captures some of the growth (6.48 ppb/yr), suggesting that increased wetland emissions have contributed an additional growth of approximately 1 ppb/yr (~13%) since 2007. The remaining growth is likely the result of nonwetland emissions and/or changes in atmospheric loss. The global CH 4 differences between TOMCAT_JU and TOMCAT_CL in Figure 4d show that the CH 4 concentration is 4 ppb higher by the end of 2014 when interannual variability in emissions is considered.

Summary
Recent trends in global wetland CH 4 emissions from 1993 to 2014 were investigated using a substantially improved version of the JULES LSM. The modeled emissions, used within the TOMCAT CTM, gave a good representation (R = 0.88) of the spatial and temporal variability in atmospheric CH 4 when compared with satellite and surface data. However, it should be noted that there remain significant uncertainties in modeling the processes in wetland CH 4 emissions. The wetland emission model predicts a statistically significant positive trend (0.2%/yr) in emissions. This is mainly driven by a long-term positive temperature trend, which increases the methanogenesis rate and, by thawing frozen high-latitude regions, the wetland area. Interannual variability (±3%/yr, σ = 4.8 Tg) superimposed on the long-term trend is dominated by changes in precipitation over wetland regions. However, we find no evidence for a long-term trend in precipitation over wetland regions.
We find that the slowdown in global CH 4 growth rate from 1999 to 2006 was not caused by decreased wetland emissions. Furthermore, our model suggests that increased wetland emissions of 3% (relative to [1993][1994][1995][1996][1997][1998][1999][2000][2001][2002][2003][2004][2005][2006] contributed to the renewed global CH 4 growth since 2007. The trends in atmospheric CH 4 growth are not exclusively a result of changes in wetland emissions. Over the period 1993-2014 anthropogenic emissions show large increases (~0.7%/yr) and the biomass burning emissions vary on an interannual basis, most noticeably during El Niño events. However, our study shows that even when accounting for those changes, the atmospheric loss and wetland emissions are keys to the interannual variability of CH 4 growth.