Stable atmospheric methane in the 2000s: key-role of emissions from natural wetlands

Two atmospheric inversions (one fine-resolved and one process-discriminating) and a process-based model for land surface exchanges are brought together to analyse the variations of methane emissions from 1990 to 2009. A focus is put on the role of natural wetlands and on the years 2000-2006, a period of stable atmospheric concentrations. From 1990 to 2000, the top-down and bottom-up visions agree on the time-phasing of global total and wetland emission anomalies. The process-discriminating inversion indicates that wetlands dominate the time-variability of methane emissions (90% of the total variability). The contribution of tropical wetlands to the anomalies is found to be large, especially during the post-Pinatubo years (global negative anomalies with minima between -41 and -19 Tg yr(-1) in 1992) and during the alternate 1997-1998 El-Nino/1998-1999 La-Nina (maximal anomalies in tropical regions between +16 and +22 Tg yr(-1) for the inversions and anomalies due to tropical wetlands between +12 and +17 Tg yr(-1) for the process-based model). Between 2000 and 2006, during the stagnation of methane concentrations in the atmosphere, the top-down and bottom-up approaches agree on the fact that South America is the main region contributing to anomalies in natural wetland emissions, but they disagree on the sign and magnitude of the flux trend in the Amazon basin. A negative trend (-3.9 +/- 1.3 Tg yr(-1)) is inferred by the process-discriminating inversion whereas a positive trend (+1.3 +/- 0.3 Tg yr(-1)) is found by the process model. Although processed-based models have their own caveats and may not take into account all processes, the positive trend found by the B-U approach is considered more likely because it is a robust feature of the process-based model, consistent with analysed precipitations and the satellite-derived extent of inundated areas. On the contrary, the surface-data based inversions lack constraints for South America. This result suggests the need for a re-interpretation of the large increase found in anthropogenic methane inventories after 2000.

From 1990 to 2000, the two inversions agree on the time-phasing of global emission anomalies. The process-discriminating inversion further indicates that wetlands dominate the time-variability of methane emissions with 90 % of the total variability. Top-down and bottom-up methods are qualitatively in good agreement regarding the global emission anomalies. The contribution of tropical wetlands on these anomalies is 10 found to be large, especially during the post-Pinatubo years (global negative anomalies with minima between −41 and −19 Tg y −1 in 1992) and during the alternate 1997-1998 el-Niño / 1998-1999 la-Niña (maximal anomalies in tropical regions between +16 and +22 Tg y −1 for the inversions and anomalies due to tropical wetlands between +12 and +17 Tg y −1 for the process-based model). 15 Between 2000 and 2006, during the stagnation of methane concentrations in the atmosphere, total methane emissions found by the two inversions on the one hand and wetland emissions found by the process-discriminating-inversion and the process model on the other hand are not fully consistent. A regional analysis shows that differences in the trend of tropical South American wetland emissions in the 20 Amazon region are mostly responsible for these discrepancies. A negative trend (−3.9 ± 1.3 Tg y −1 ) is inferred by the process-discriminating inversion whereas a positive trend (+1.3 ± 0.3 Tg y −1 ) is found by the process model. Since a positive trend is consistent with satellite-derived extent of inundated areas, this inconsistency points at the difficulty for atmospheric inversions using surface observations to properly con-

Introduction
The growth rate of atmospheric methane (CH 4 ) has experienced large variations since the early 1990s: after a decade of decrease, interrupted by two peaks in 5 and 2002(Dlugokencky et al., 1998Cunnold et al., 2002;Wang et al., 2004;Bousquet et al., 2006), the growth rate of atmospheric methane remained small from 1999 to 2006, only to increase again since 2007 (Rigby et al., 2008;Dlugokencky et al., 2009;Bousquet et al., 2011). As methane is emitted by a large variety of sources, the explanations for the observed atmospheric variations have generally implied changes in several source or sink types.
The sources involved in the variations of the 1990s are now well understood, although their relative magnitude may still be debated.
The 1991-1993 growth rate anomaly is linked to the Pinatubo eruption, which led to a decrease in methane loss, due to reduced tropospheric hydroxyl radical (OH) concentrations and stratospheric chemistry changes (Bândȃ et al., 2012), followed by a decrease in natural wetland emissions. The negative impact of this volcanic event on wetland CH 4 emissions is due to its effects on the climate, with both cooling (Dlugokencky et al., 1996) and precipitation anomalies in the Northern hemisphere (Walter et al., 2001b;Dlugokencky et al., 2011), likely slightly amplified by sulfur deposition 20 (Gauci et al., 2008). The collapse of the economy of the former USSR and Eastern Europe has also led to decreased anthropogenic emissions starting in 1991 and spanning most of the 1990s (Dlugokencky et al., 2003).
The 1997-1999 large growth rate anomaly is explained by a positive anomaly in emissions from biomass burning, both in the Tropics and at high latitudes (van der Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | by a positive anomaly in the same in 98-99, linked to wet la-Niña conditions (Bousquet et al., 2006;Chen and Prinn, 2006).
Overall, the role of wetlands in changes in atmospheric methane has often been identified as dominant (Chen and Prinn, 2006;Bousquet et al., 2006Bousquet et al., , 2011, linked to meteorological conditions of temperatures and precipitations (Dlugokencky et al., 5 2009;Bousquet et al., 2011). These conditions are thought to lead to changes in both the wetland extent (e.g. Ringeval et al., 2010) and the CH 4 flux per wetland area (e.g. Bloom et al., 2010). The impact of wetland emissions is often thought to be combined with changes in OH concentrations (Wang et al., 2004;Monteil et al., 2011), although it is now well accepted that inter-annual changes in OH concentrations are limited to 10 1-3 % and can therefore explain only a limited part of the observed changes in the methane growth rate (Montzka et al., 2011).
The analysis of the variations of atmospheric methane after 1999 is still largely debated.
Various scenarios have been suggested to explain the stabilization of methane con- 15 centrations between 2000 and 2006: (i) reduced global fossil-fuel-related emissions, estimated from AGAGE and NOAA (Chen and Prinn, 2006) or from ethane emissions used as a proxy to fossil-fuel-related CH 4 emissions (Aydin et al., 2011;Simpson et al., 2012); (ii) compensation between increasing anthropogenic emissions (as inferred by inventories such as EDGAR4 (EDGAR 4, 2009)) and decreasing wetland emissions 20 (Bousquet et al., 2006); (iii) a decrease in emissions due to rice-paddies, attributed to changes in agricultural practices (Kai et al., 2011); (iv) stable microbial and fossil-fuelrelated emissions in the early 2000s (Levin et al., 2012) and/or (v) significant (Rigby et al., 2008) to small (Montzka et al., 2011)  Methane emissions can be investigated using process-based models and inventories (bottom-up or B-U approach), and atmospheric inversions (top-down or T-D approach). Atmospheric inversions combine atmospheric observations of CH 4 , an atmospheric chemistry and transport model, and prior information about sources and sinks. The flux estimates that give the best fit to the observed atmospheric concentrations are 5 derived by optimization. However, atmospheric inversions provide a limited insight into the underlying biogeochemical processes controlling emissions, particularly over regions where several processes and sources overlap. B-U models computing wetland (Melton et al., 2012) or biomass-burning (van der Werf et al., 2010) methane emissions incorporate knowledge of small-scale processes but they need additional information 10 and constraints to project their local emission estimates to larger scales compatible with the atmospheric signals. B-U emission inventories (EDGAR 4, 2009;EPA, 2011) are based on country-scale energy use and agricultural statistics and usually provide yearly to decadal estimates of anthropogenic emissions at global and national scales.
In this study, we present an analysis of methane fluxes for the 1990-2009 period.

Process model for natural wetland emissions
The model of CH 4 emissions by natural wetlands used here (hereafter ORCHIDEE) is based on the ORCHIDEE global vegetation model (Krinner et al., 2005) which simulates land energy budget, hydrology, and carbon cycling. It has then been developed (Ringeval et al., 2010(Ringeval et al., , 2011 to simulate:

10
-the CH 4 -emitting wetland area dynamic, using some TOPMODEL concepts (Ringeval et al., 2012a). The wetland extents are here normalized to match Papa et al. (2010). A global, multi-year data set giving the monthly distribution of flooded areas at a ≈ 25 km resolution has been generated. It is based on multiple satellite observations which are optimized specifically for surface water detection (Papa 15 et al., 2010;Prigent et al., 2012). It is built using a combination of satellite data including passive microwave observations and a linear mixture model to account for vegetation. Three wetland classes, differing by the value of the water table depth, are simulated for each grid cell and at each time step: saturated wetland and wetland with a mean water table at 3 and 9 cm below the soil surface.

20
-the flux density, which is computed following Walter et al. (2001a) and Ringeval et al. (2010) and results from three processes: production, oxidation and transport (via diffusion, ebullition and through plants).
A summary of the ORCHIDEE methodology used to compute CH 4 emissions is given in Wania et al. (2012) The wetland emissions are driven by the CRU-NCEP climate forcing data set (Viovy and Ciais, 2009). In this study, four scenarios of wetland emissions are given based on: -accounting or not for the non-saturated wetlands (i.e. all classes of wetlands or saturated wetlands only)

5
-multiplying or not the simulated wetland area by peat-land map in boreal regions. This map is obtained by using soil organic carbon data from IGBP DIS at high resolution (5 ′ × 5 ′ ), by dividing each pixel of this database by 130 kg m −3 (which is the maximum soil carbon density of peat), and then by regridding the result at 1 • × 1 • resolution (Lawrence and Slater, 2007). The hypothesis underlying the 10 multiplication of the two products (map of peat-land cover and map of inundated areas) is that the inundated fraction is the same for an entire grid-cell as for a subgrid peat-land into this grid-cell (see Bousquet et al., 2011, for more details).
ORCHIDEE provides CH 4 emissions by natural wetlands as the product of a flux density by a wetland extent, at a 1 degree by 1 degree resolution for the time period

Inverse methods
The two inversion models used here are based on the Bayesian formalism: they assim-20 ilate surface observations of CH 4 and methyl-chloroform (CH 3 CCl 3 or MCF) concentrations (measurements from various surface monitoring networks: AGAGE (Prinn et al., 2000(Prinn et al., , accessed: 2012, CSIRO (Francey et al., 1999) (Hourdin et al., 2006) with prior information on the spatio-temporal distribution and uncertainties of CH 4 sources and sinks to estimate the magnitude and the uncertainties of optimized surface emissions. The two methods differ by the inversion setup and the resolution method: see Bousquet et al. (2005) and  , 1994, 2009)) with monthly uncertainties ranging from ±5 ppb to ±50 ppb, with a median of ±10 ppb. The transport model is LMDZt v3 offline nudged on analyzed horizontal winds (Uppala et al., 2005;Hourdin et al., 2006) with OH pre-optimized using methyl-chloroform concentrations (Bousquet et al. (2005) used MCF concentration measurements by AGAGE (Prinn et al., 2005) and NOAA/ESRL (Montzka et al., 2000(Montzka et al., , 2011). The prior emissions are elaborated from various inventories by Matthews and Fung (1987), Olivier 15 and Berdowski (2001) and van der Werf et al. (2006). Monthly uncertainties on fluxes are set at 150 % in each region; there are no error correlations but month-to-month changes are limited to ±250 % if the process follows a seasonal cycle or to ±50 % otherwise (Peylin et al., 2000(Peylin et al., , 2002. The main advantage of this inversion scheme is the low cost of computation, which 20 makes it possible to test various scenarios, varying OH concentrations, constraints, uncertainties. In this study, eleven scenarios are used, as described in Bousquet et al.  (Schmidt et al., 2005), NIWA (Lowe et al., 1991), NOAA/ESRL (Montzka et al., 2000(Montzka et al., , 2011Dlugokencky et al., 1994Dlugokencky et al., , 2009 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | spectively 4) scenarios; since INVVAR consists in only one case, no such range is given.

10
Variations in the latitudinal distribution of wetlands in ORCHIDEE are significant (±7 %) and due to the use or not of a soil carbon map overlaid to the wetland extent given by ORCHIDEE to reduce the simulated CH 4 emitting areas to peat-lands (see Sect. 2.1).
At the global scale, the inter-annual variations of the total net CH 4 fluxes of INVVAR and the various INVANAs are in good agreement, in the phasing through time and in the 15 magnitude with time correlations ranging from 66 % to 84 % over the 19 common years (Fig. 1, top). Since these correlations are for deseasonalized anomalies, which are a relatively small signal, they are considered as good. This result is not simply a check of a correct mass balance as the two inversions use different OH fields and different approaches to optimize the atmospheric sink of CH 4 (off-line in INVANA or simulta-20 neously in INVVAR). INVVAR retrieves a lower variability than INVANA with standard deviations over all monthly anomalies of 12 and 18 Tg y −1 respectively. The standard deviation for wetlands in INVANA is 16 Tg y −1 , which indicates that wetlands explain about 90 % of the variability of total methane emissions. In the following sections we therefore focus on changes in natural wetland emissions.   Fig. 1 top), in agreement with the literature. These anomalies are mainly due to the contribution of the tropical areas (minima for latitudes less than 30 • N of −21 Tg y −1 for INVVAR and −19 to −35 Tg y −1 for the 11 INVANA scenarios in 1992, Fig. 1). This is 5 consistent with the negative impact of this volcanic event on wetland CH 4 emissions already noticed in the literature (Hogan and Harriss, 1994). The one INVANA scenario producing smaller changes in global and tropical emissions is the one assuming constant OH concentrations with time; it is closer to INVVAR in the early 1990s. In IN-VANA, the optimized OH concentrations are sensitive to the errors on MCF emissions, 10 which are proportional to MCF fluxes. Since MCF emissions are large during the 1980-90s, their errors are also large and INVANA find large year-to-year OH changes (up to −15 % for the total column between 1996 and 1997 for one scenario). In INVVAR, since the optimization is simultaneous, OH concentrations are constrained by both MCF and CH 4 measurements. Therefore, OH variability is smaller than in most INVANA scenar- 15 ios, closer to the constant-OH one, and more consistent with Montzka et al. (2011).
A large positive anomaly of global total emissions is inferred by INVVAR andINVANA in 1997-1999 (maxima at +37 and +19 to +25 Tg y −1 respectively, Fig. 1 top), related to the el-Niño/la-Niña events. Most of the signal comes from tropical regions (maxima at +27 and +16 to +22 Tg y −1 in INVVAR and the INVANAs). That could be linked to 20 fires in 199720 fires in -199820 fires in (van der Werf et al., 2004 and exceptionally high CH 4 emissions from natural wetlands during la-Niña event in 1998 (Hodson et al., 2011). These high wetland emissions are actually consistent with 1998 natural wetland emissions as seen by ORCHIDEE, with anomalies of +11 to +17 Tg y −1 at the global scale ( Fig. 1) and +12 to +17 Tg y −1 in the tropical regions (Fig. 1 bottom). Note that for the four ORCHIDEE a negative trend over the same period (−4.2 ± 1.2 Tg y −1 ). To assess the cause of this discrepancy between the two approaches, we studied the regional distributions of methane emissions within the Tropics.

Latitudinal and regional break-down of emissions
We find that the disagreement between ORCHIDEE and INVANA at the global scale af-10 ter 2000 is mainly due to the anomalies south of 30 degrees North, an area where emissions are mainly due to tropical land regions (Fig. 1). Of the four ORCHIDEE scenarios described previously, only two are relevant in the tropics (the other two are obtained by varying boreal peat-lands) and are thus retained in the following: ORCHIDEE-all accounting for all classes of wetlands and ORCHIDEE-sat accounting only for satu-  (Fig. 1 bottom). Moreover, the trend over 2000-2006 in this band (Table 1) is −4.1±0.9 Tg month −1 for INVANA whereas the trend for the mean between ORCHIDEE-all (scenario accounting for all wetlands) and ORCHIDEE-sat (scenario with saturated wetlands only) is +1.4 ± 0.3 Tg month −1 (Table 1).

25
In order to investigate the regional contributions to the flux anomalies and trends of wetland emissions in the Tropics, we computed the yearly flux anomalies for OR-CHIDEE and the mean of the eleven scenarios of INVANA for eight key-regions South 9029 Introduction  (Fig. 2). Interestingly, ORCHIDEE and INVANA agree on the fact that variations in methane emissions from the Amazon basin (tropical South America in Fig. 2) drive the trend of tropical emissions, but with opposite signs. Indeed, the ratio between the slopes for tropical South America and the whole < 30 • S region (Table 1) indicates that tropical South America explains 93 % of the trend for ORCHIDEE and between 5 89 and 95 % in INVANA (depending on the scenarios). However, the inferred trends for the Amazon region are of opposite signs with −3.9 ± 1.3 Tg y −1 for INVANA and +1.3 ± 0.3 Tg y −1 for ORCHIDEE (Table 1). When an atmospheric inversion cannot provide insights about the underlying emission processes and the sensitivity of wetland emissions to the climate, the ORCHIDEE process-based model can. The monthly anomalies of CH 4 wetland emissions for the Amazon region for ORCHIDEE and INVANA, together with the precipitations and temperature changes are shown in Fig. 3. Not surprisingly, ORCHIDEE and INVANA phases of the inter-annual variations are not in agreement at this regional scale, although the magnitudes of the changes are close. As noted before, the trends between In tropical South America, the trend in ORCHIDEE wetland emissions is driven by precipitations (Fig. 3, middle panel), which show a positive trend of 0.05 mm d −1 between 2000 and 2006 over the Amazon. When the IAV of wetland extent is removed 5 (black curve in Fig. 3, bottom panel), the IAV of emissions over 1990-2009 decreases from 3.8 (ORCHIDEE-sat) to 2.1 Tg y −1 and appears to be driven by air temperature with a strong correlation of 0.86 over 1990-2009 (Fig. 3, bottom panel). Temperature could have an effect on both methanogenesis rate and substrate supply but large uncertainties remain on the contribution of each process (e.g. White et al., 2008, for northern peat-lands).
Overall, ORCHIDEE infers an increase of CH 4 emissions in the Amazon between 2000 and 2006, which is opposite to the decreasing emissions inferred by INVANA inversions. The difference of IAV between ORCHIDEE-simulated emissions when the time variability of wetland extent is either prescribed or computed (Fig. 3, upper panel) 15 underlines the difficulty to capture a good IAV of the processes involved in the extension/retraction of wetland areas. A way of improvement would be to implement the floodplains into the model (e.g. like Decharme et al., 2008Decharme et al., , 2012, the relevant processes leading to wetland formation in the Amazon basin (Hess et al., 2003). Uncertainties linked to precipitations (magnitude and spatial distribution) in the Amazon 20 basin and their huge effect on the floodplains extent simulated by land surface models (Guimberteau et al., 2012) is also problematic.
The lack of wetland Plant Functional Type into the ORCHIDEE model and the use of the mean grid-cell simulated labile carbon as methanogenesis substrate's proxy could also lead to an overestimated sensitivity of CH 4 emissions to precipitations (Ringeval 25 et al., 2012b). However, in our study, the sensitivity of CH 4 emissions to precipitations seems to mainly happen through the wetland extent (Fig. 3) and make the lack of floodplain representation the main caveat of the B-U approach used here. 13,2013  The IAV of the remote-sensed inundated extent in the Amazon basin over 2000-2006 and of the in-situ river discharge are in agreement and do not exhibit a clear negative trend (see Fig. 10a in Papa et al., 2010). This points out at a possible issue with INVANA global inversions. Indeed, the Amazon region is poorly constrained by the surface networks, which only provide routine observations over the neighboring 5 oceans. This lack of nearby observations means that large changes in emissions can be tolerated in the inversion when assimilating these observations. Also, using large regions may induce aggregation errors as explained in Kaminski et al. (2001). It should be noted that for this area INVVAR (which works at the grid cell's scale but only optimizes total net emissions) obtains emissions which are close to the prior; the only  [2007][2008]. This seems to confirm that the changes made by INVANA in the large region are due to constraints which are not directly relevant to the Amazon basin. 15 Overcoming the lack of observations requires to assimilate more regional data in the Amazon area (such as aircraft data (Miller et al., 2007), satellite data (for example SCIAMACHY (Frankenberg et al., 2008) Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | match the stable global concentration, infers a −4 ± 1 Tg y −1 decrease, mostly in South American wetland emissions. This result might be sensitive to the large scale transport between the Northern latitudes and the tropics as stated in Stephens et al. (2007) for carbon dioxide. If the increase in anthropogenic CH 4 emissions is real, a decreasing source or an increasing sink must be identified to match the stable concentrations. 5 There is no indication of a positive trend in OH concentrations after 2000 that could have induced an abnormally increasing methane loss in the troposphere (Rigby et al., 2008;Montzka et al., 2011). The increasing wetland emissions found by ORCHIDEE indicate that South America might only be an opportunistic candidate for reduced emissions and that another region and/or another process is indeed decreasing. The analyses of 13C in CH 4 recently brought elements to the debate. The possible decrease in microbial-related emissions in the Northern hemisphere since the 1990s inferred by Kai et al. (2011) is contested by the analysis of another more consistent 13C data set, which infers no trend in the inter-hemispheric difference of 13C (Levin et al., 2012). This last result questions the validity of the large increase in anthropogenic methane emis-