Enhanced methane emissions from tropical wetlands during the 2011 La Niña

Year-to-year variations in the atmospheric methane (CH4) growth rate show significant correlation with climatic drivers. The second half of 2010 and the first half of 2011 experienced the strongest La Niña since the early 1980s, when global surface networks started monitoring atmospheric CH4 mole fractions. We use these surface measurements, retrievals of column-averaged CH4 mole fractions from GOSAT, new wetland inundation estimates, and atmospheric δ13C-CH4 measurements to estimate the impact of this strong La Niña on the global atmospheric CH4 budget. By performing atmospheric inversions, we find evidence of an increase in tropical CH4 emissions of ∼6–9 TgCH4 yr−1 during this event. Stable isotope data suggest that biogenic sources are the cause of this emission increase. We find a simultaneous expansion of wetland area, driven by the excess precipitation over the Tropical continents during the La Niña. Two process-based wetland models predict increases in wetland area consistent with observationally-constrained values, but substantially smaller per-area CH4 emissions, highlighting the need for improvements in such models. Overall, tropical wetland emissions during the strong La Niña were at least by 5% larger than the long-term mean.


Uncertainty estimation
In Figures 1 and 2 of the main text, the shaded regions represent ±1σ uncertainty of the respective time series. These uncertainties were calculated using the Monte-Carlo method with 100 simulations. In Figure 1, the retrievals errors, provided by RemoteC, were assigned as uncertainty of individual GOSAT XCH 4 measurements. For model output, model representation errors were used. In Figure 2, the emission uncertainties are the posterior uncertainties calculated by TM5-4DVAR [Basu et al., 2013]. δ 13 C-CH 4 measurements were assigned an uncertainty of 0.02 . Similar procedures were used in SM Figure 1  Surface measurements of CH 4 are shown in Figure 1. There are considerable differences between the zonal mean CH 4 variations derived from these data and the GOSAT FP retrievals (see Figure 1 of [Hunter , 2007] The differences in between the interannual variation in surface measurements and satellite retrievals likely reflect differences in spatial coverage of the two datasets. GOSAT FP has a more even spatial coverage than the surface networks (see Figure 2). A larger number of tropical surface measurements are taken in the northern hemisphere compared with southern hemisphere, which can bias the zonal average. Also, the signal from land takes quite a long transport path-including upward transport by convection, etc.-before reaching a marine site in the Tropics, increasing the chance of transport variations modifying the signal.
After subtracting the TM5-Meteorology, influences of sampling and transport are removed from the two measurement sets and their residual anomalies are in better agreement (surface ∆ CH 4 TRO = 2.34 ppb vs. GOSAT ∆ XCH 4 TRO = 2.04ppb). This is consistent with the posterior emissions from the inversion assimilating only surface measurements being similar to those obtained when assimilating GOSAT FP and surface measurements (see Figure 3).

Biomass burning emissions
We use the GFED4s inventory to account for CH 4 emission from biomass burning (BB).  In LMDz-PYVAR-SACS, OH fields were optimized using methyl chloroform (MCF) measurements. These results should be treated with caution as 1. The MCF-based OH optimization becomes increasingly uncertain with MCF levels dropping to only a few ppt in recent years.
2. It is difficult to determine the correct relative uncertainties of CH 4 and MCF, which introduces a temporally varying weight of the MCF measurements on the solution of the coupled inversion system.
3. We make a comparison between different inversion systems. Doing so complicates the comparison, especially for the absolute optimized emissions as different inversion systems may give a wide range of estimates depending on their setup and boundary conditions. However, inversion-optimized temporal emission variations are known to be less sensitive to differences in inversion setup than the mean state.
In addition, recent studies have pointed out that atmospheric OH is well buffered against changes in its driving parameters [Lelieveld et al., 2016]. The δ 13 C-CH 4 influence of a 3 TgCH 4 yr −1 enhanced sink will only be 0.005 , which is within the error margins of the δ 13 C-CH 4 anomalies. If the whole anomaly was caused by OH this would lead to an isotopic effect that was less than observed, suggesting the observed anomaly is driven by changes in the sources rather than the sinks. 7 Transport impact on δ 13 C-CH 4 Figure 2 in the main text shows the XCH 4 variability due to variability in atmospheric transport. Meteorological variability can influence δ 13 C-CH 4 due to the isotopic fractiona- tion of the OH sink, changes in the strength of inter-hemispheric exchange, etc. Figure 6 shows the simulated δ 13 C-CH 4 variability in response to variations in transport. Overall, they are an order of magnitude less than the variability in the δ 13 C-CH 4 measurements.

Process-based wetland models
Process-based wetland models estimate CH 4 emissions from natural wetlands using information about precipitation, temperature, biomass availability, etc. We analyzed the CH 4 emission output from two such models: LPJ-wsl [Hodson et al., 2011;Zhang et al., 2016] and CLM4.5 (referred as CLM from here on) [Riley et al., 2011;Xu et al., 2016]. These models show a weaker enhancement of CH 4 emissions during LN11 than the TM5-4DVAR inversion (See Figure 7). A poor correlation is seen between these emissions and precipitation anomalies. This happens despite general agreement between the inundated area calculated by the hydrological schemes of these bottom-up models and SWAMPS (Surface WAter Microwave Product Series). As shown in the main text (see Figure 3) the inundated area in SWAMPS correlates well with the inversion derived emission anomalies in TRO.
Two mechanisms, that are implemented in the process-based wetland models, might explain the disagreement between inundated area and modeled CH 4 emissions: 1. CH 4 emission is directly related to ecosystem respiration, which increases with increasing temperature. During LA11 the temperature anomaly in TRO was slightly negative (µ temperature TRO = −0.05 • C), and hence, it will decrease the strength of inundation-driven positive CH 4 emission anomaly.
2. The relation between extent of inundated area and CH 4 emission is complex. In general, wetland CH 4 emission increases with increase in inundated area. However, the reverse can also happen if the increase in precipitation causes a higher water table depth, which will increase the chances of CH 4 getting oxidized before reaching the atmosphere.

Other retrieval/inversion methods
An important source of systematic error in satellite retrievals is the scattering of light by aerosols and thin cirrus clouds along the measured light path. The full-physics (FP, Butz et al. [2010]) and the proxy [Frankenberg et al., 2005] retrieval methods have been developed in the past to account for such atmospheric scattering. Additionally, the so-called ratio method assimilates X ratio (XCH 4 :XCO 2 ) directly to optimize the surface fluxes of CH 4 and CO 2 [Fraser et al., 2014;Pandey et al., 2015Pandey et al., , 2016. Hence, it avoids the errors introduced in translating retrieved X ratio to XCH 4 using modeled XCO 2 (XCO model 2 ).
The proxy and ratio method generally yield twice as many valid CH 4 retrievals as FP, because the latter requires stricter cloud filtering criteria. In this study, we still use FP retrievals to avoid potential correlations between the inter-annual variations of CH 4 and CO 2 in response to ENSO. The proxy retrieval method might erroneously attribute an CO 2 anomaly, that is not captured in XCO model 2 , to XCH 4 . Figure 8 illustrates that this is indeed what happens. The ratio inversion method does not depend on the XCO model 2 , however, it can still wrongly assign a CO 2 anomaly to CH 4 emissions to fit the X ratio in the atmosphere  depending on the relative uncertainty assigned to the a priori CO 2 and CH 4 fluxes. Therefore, the ratio and proxy methods find larger anomalies than the FP inversion. The opposite is seen during EN10 and after LN12.  Figure 11: Pearson product-moment correlation (R) between monthly anomalies of TM5-4DVAR CH 4 emissions, derived by assimilating FP XCH 4 , and anomalies of temperature, precipitation and inundated area.