Using satellite data to identify the methane emission controls of South Sudan’s wetlands

. The TROPOspheric Monitoring Instrument (TROPOMI) provides observations of atmospheric methane (CH 4 ) at 15 an unprecedented combination of high spatial resolution and daily global coverage. Hu et al. (2018) reported unexpectedly large methane enhancements over South Sudan in these observations. Here we assess methane emissions from the wetlands of South Sudan using two years (December 2017–November 2019) of TROPOMI total column methane observations. We estimate annual wetland emissions of 7.4 ± 3.2 Tg yr -1 , which agrees with the multiyear GOSAT inversions of Lunt et al. (2019) but is an order of magnitude larger than estimates from wetland process models. This disagreement may be explained 20 by the up to 4 times underestimation of inundation extent by the hydrological schemes used in those models. We investigate the seasonal cycle of the emissions and find the lowest emissions during the June-August season when the process models show the largest emissions. Using satellite altimetry-based river water height measurements, we infer that this seasonal mismatch is likely due to a seasonal mismatch in inundation extent. In models, inundation extent is controlled by regional precipitation, scaled to static wetland extent maps, whereas the actual inundation extent is driven by water inflow from rivers 25 like the White Nile and the Sobat. We find the lowest emission in the highest perception and lowest temperature season JJA when models estimate large emissions. In general, our emission estimates show better agreement, in terms of both seasonal cycle and annual mean, with model estimates that use a stronger temperature dependence. This suggests that temperature might be a stronger control for the South Sudan wetlands emissions than currently assumed by models. Our findings demonstrate the use of satellite instruments for quantifying emissions from inaccessible and uncertain tropical wetlands, providing clues for 30 improvement of process models, and thereby improving our understanding of the currently uncertain contribution of wetlands to the global methane budget.

3 by large emissions from wetlands in the region. Recently, Lunt et al. (2019) used methane observations from the Japanese Greenhouse gases Observing Satellite (GOSAT) in inverse modelling to infer emissions from tropical Africa during 2010-2016. They found that emissions from South Sudan were more than 3 times larger than the ensemble mean estimates from the Wetcharts process model (Bloom et al., 2017). They also found that emissions from the Sudd wetlands in the region increased rapidly from 2.4-4.2 Tg yr -1 in 2010-2011 to 5.2-6.9 Tg yr -1 in 2016, likely, because of an inundation extent expansion due 70 to an increase in water inflow from the White Nile river.
This study aims to infer the scale of the wetland methane emissions from South Sudan from TROPOMI observations using a simplified emission quantification method and investigate its relationship with the results of wetland process models and the seasonally varying climatological conditions. This study is structured as follows. Section 2 describes the method and data used including the TROPOMI data, wetland models and inundation extent data, and the emission quantification method. Section 3 75 presents our results and discussion including emission estimates from TROPOMI and their comparison with the process models, and an analysis of the differences between models and TROPOMI emission estimates using inundation extent and meteorological data. Our conclusions are given in Section 4.

TROPOMI methane data 80
TROPOMI is the single instrument onboard the Copernicus Sentinel-5 Precursor (S-5P) satellite, launched on 13 October 2017 in a sun-synchronous orbit at 824 km altitude (Veefkind et al., 2012). It is a push-broom imaging spectrometer, recording spectra along a 2600 km swath while orbiting the Earth every 100 min, resulting in daily global coverage. Total column methane (XCH4) is retrieved with near-uniform sensitivity in the troposphere from its absorption band around 2.3 m using earthshine radiance measurements from the Short Wave Infrared (SWIR) channel of TROPOMI . 85 TROPOMI XCH4 has a ground pixel size of 7 × 7 km 2 (7 × 5.5 km 2 since August 2019) at nadir with larger ground pixels towards the edges of its swath.
In this study, we use the operational two-band retrieval product of TROPOMI (Hasekamp et al., 2019). It uses 0.76 m O2A and 2.3 m CH4 bands in the Near Infrared (NIR) and SWIR spectra. XCH4 is retrieved using the full-physics RemoTeC 90 algorithm, which accounts for light path perturbations due to scattering by aerosol and cirrus cloud particles in the atmosphere (Butz et al., 2012;Hu et al., 2016). We only use high-quality XCH4 measurements retrieved under favourable cloud-free conditions. Also, XCH4 is filtered ("qa"=1) for solar zenith angle (< 70°), viewing zenith angle (< 60°), smooth topography (1-standard deviation surface elevation variability < 80 m within a 5 km radius) and low aerosol load (aerosol optical thickness < 0.3 in NIR band). Note that Hu et al. (2018) used two months of XCH4 data from the "scientific" retrieval product of SRON 95 Netherlands Institute for Space Research. Those measurements had a relatively sparse temporal coverage over South Sudan because they were performed during the commissioning phase of TROPOMI when algorithm tests and calibrations were 4 ongoing. The operational product used here provides a more temporally homogenous coverage and a surface albedo-dependent bias correction (Hasekamp et al., 2019).

Process model data 100
We compare TROPOMI emission estimates with two wetlands process models: Wetcharts (Bloom et al., 2017) and LPJ-wsl . These models calculate monthly methane emissions on a global grid of 0.5° × 0.5° resolution by simulating the microbial production and oxidation processes in the soil using temperature, inundation extent and heterotrophic respiration data. Wetcharts calculates wetland emissions using in total four inundation extent parameterizations, nine terrestrial biosphere models of heterotrophic respiration and three CH4:C temperature parameterizations (q10). Wetcharts version 1.0 105 provides two ensembles: (1) an ensemble with 324 emission estimates for 2009-2010, called the "Full Ensemble" and (2) an 18-member extended-in-time ensemble for 2001-2015, called the Wetcharts "Extended Ensemble". In the Wetcharts Full Ensemble, the set of four inundation extent estimates are calculated based on two maximum wetland area estimates, multiplied with two monthly varying scaling factors. The wetlands area estimates are taken from (1) the Global Lakes and Wetlands Database (GLWD; Lehner and Döll, 2004), and (2) the sum of all GLOBCOVER wetland and freshwater land types (Bontemps 110 et al., 2011). The scaling factors are calculated from (1) precipitation data from ERA-Interim meteorological data and (2) inundation extent data from the Surface WAter Microwave Product Series (SWAMPS) multi-satellite surface water product (Schroeder et al., 2015). The 18-member Wetcharts Extended Ensemble provides emission estimates for only the two inundation extent estimates that are based on ERA-Interim and only one terrestrial biosphere model CARDAMOM (Bloom et al., 2016). 115 LPJ-wsl methane model is based on the process-based dynamic global vegetation model Lund Postdam Jena (LPJ). It uses soil temperature, soil moisture-dependent fraction of heterotrophic respiration (Rh), and inundation extent to calculate wetlands methane emissions. The inundation extent of LPJ-wsl is calculated by the TOPography-based hydrological model (TOPMODEL) driven by meteorology from ERA-Interim. TOPMODEL simulates hydrologic fluxes of water, including 120 lateral transport, such as infiltration-excess overland flow, infiltration, exfiltration, subsurface flow, evapotranspiration, and channel routing through a watershed.

Inundation extent data
Earlier studies have indicated that the water availability is particularly important in the tropics (temperature is less limiting here in contrast to high latitudes), and hence, inundation extent is one of the main sources of uncertainty for tropical wetlands 125 Ringeval et al., 2010). We analyze the inundation extent data used in process models: TOPMODEL (used in LPJ-wsl); GLWD and GLOBCOVER with ERA-Interim (used in the Wetcharts Extended Ensemble). We compare these inundation extent estimates against the remote sensing-based high-resolution inundation extent data from Gumbricht et al. (2017), which maps wetlands and peatlands at 231 meters spatial resolution by combining three biophysical indices related to waterlogged soils, and (3) geomorphological position where water is supplied and retained. They use 2011 MODIS data to map the duration of wet and inundated soil conditions and Shuttle Radar Topography Mission (SRTM) for topography. In addition, we use satellite altimetry-based water height measurements from the Hydroweb database (Crétaux et al., 2011;Da Silva et al., 2010). The water height anomalies of the White Nile and Sobat rivers are used as a proxy for inundation extent variations in the Sudd and Macher wetlands, respectively. Fig. 1b shows the location of the river height measurement sites. 135 We also analyze temperature and precipitation data from the European Centre for Medium-Range Weather Forecasts' ERA5 meteorological reanalysis (Hersbach and Dee, 2016).

Emission Quantification method
The wetland distribution from Gumbricht et al. (2017) is shown in Fig. 1 for the region in South Sudan where a large TROPOMI XCH4 enhancement can be observed. This region, which includes Sudd, Machar and other smaller wetlands, is hereafter 140 referred to as the South Sudan wetland region (SSWR). To calculate emissions, we first prepare seasonally averaged TROPOMI XCH4 maps on a grid of at 0.1° × 0.1° resolution. Only grid cells with at least 5 high-quality TROPOMI measurements are used in the season average map. We apply the mass balance method of Buchwitz et al. (2017) to calculate emissions from December 2017 to November 2019. The emission Q (Tg yr -1 ) from the SSWR box in Fig. 1a for a given period is calculated using the following equation: 145 Where, ∆ ! is the "source XCH4 enhancement", i.e., the mean XCH4 difference between the source and the surrounding background. C is a dimensionless factor of 2.0 derived by Buchwitz et al. (2017) based on the concentration difference of air 150 parcels before and after entering a source area. M (5.345 Tg CH4 km -2 ppb -1 ) is the atmospheric total column mixing ratio-tomass conversion factor for a surface pressure of 1013.25 hPa, which is the standard atmospheric pressure. #$% is a dimensionless factor used to correct for the changes in column air mass with surface elevation, calculated as the ratio of surface pressure in the source and standard atmospheric pressure (1013.25 hPa). L is the "effective size" of the source region (632 km), calculated as the square root of its area (4.0 × 10 5 km 2 ). V (km yr -1 ) is the ventilation wind speed derived from the ERA5 155 meteorological reanalysis vertical wind speed profile. Surface elevation variations change the contribution of tropospheric to the total atmospheric column, which influences XCH4. TROPOMI XCH4 maps are corrected for this effect by adding the correction factor 7 ppb km -1 from Buchwitz et al. (2017), using GMTED2010 elevation data shown in Fig. A1 (Danielson et al., 2010). We remove the large scale latitudinal XCH4 gradient from the seasonal average TROPOMI XCH4 maps by subtracting a 3 rd order polynomial fit from the background region, excluding the source region (see Figure 2a). Figure 3 shows the monthly average ERA5 wind speed at 10:00 UTC (TROPOMI overpass time) in the SSWR as a function of pressure within the local boundary layer during select months. To calculate V, the pressure-weighted average of these 165 boundary layer wind speed is calculated over SSWR using monthly average ERA5 boundary layer height data. We use average boundary layer winds instead of 10-meter winds because it was found to better represent the ventilation wind speed in the source region (see Varon et al., 2018). For SSWR, the 10-meter wind speed is on average 35 % lower than the boundary layer wind speed, consistent with the diminishing influence of the surface friction with height.

170
The uncertainty of Q is calculated as sum-in-quadrature of uncertainties associated with ∆ ! and V. The ∆ ! uncertainty is estimated as sum-in-quadrature of (1) 1-standard deviation of ∆ ! estimates calculated by sequentially increasing the size of the background box from 1° to 10° longitude and latitude in 1° interval, and (2) the XCH4 uncertainty of a single 0.1° × 0.1° grid cell in the average map (= 22 ppb), taken as 1-standard deviation XCH4 of all the grid cells in Fig. 1a. Note that this approach overestimates the XCH4 uncertainty of the grid cells as XCH4 variations within the grid are also caused by 175 emissions and surface elevation variations in addition to measurement errors. The uncertainty of is estimated from the variation in wind speed during 4 consecutive hours (09:00 UTC, 10:00 UTC, 11:00 UTC, 12:00 UTC) centered around the TROPOMI overpass time. Note that we use the mass balance method equation from Buchwitz et al. (2017), but not their empirical equation to estimate the uncertainty of Q. They derive that equation using a fixed wind speed of 1.1 m/s globally, which would give a larger uncertainty in comparison to our approach of using location and time-specific wind information: 180 ERA5 average V is 2.5 ± 0.42 m s -1 in SSWR during 2018-2019.

XCH4 enhancements
We first assess the XCH4 enhancements over South Sudan in the two-year average map of TROPOMI XCH4 shown in Fig. 1a in relation to the SSWR wetland distribution in Fig. 1b. Similar to previous remote sensing studies (Frankenberg et al., 2011;185 Lunt et al., 2019;Hu et al., 2018), we observe a large XCH4 enhancement over the Sudd wetlands. In addition, the TROPOMI data also resolve another distinct enhancement over the Machar and Lotilla wetlands in eastern South Sudan, indicating large emissions from these wetlands too. The second enhancement was also observed by Hu et al. (2018) using two months of TROPOMI XCH4.

190
The Sudd wetlands are flooded by the main White Nile tributary originating from Lake Victoria, whereas the wetlands in Southeast Sudan are along smaller rivers like the Kangen and Sobat, originating from the Ethiopian mountains. Lunt et al.
(2019) attributed their GOSAT inversion emission estimates only to Sudd and evaluated the emissions using auxiliary data for Sudd. However, as the wetlands in the east are flooded by a different set of rivers and have a substantial contribution to the overall XCH4 enhancement, they also need to be considered when studying the mechanisms driving the large emissions in this 195 region.
The XCH4 enhancement for SSWR in the two-year average is 18.8 ± 2.8 ppb, which is more than 3 times the enhancement over the Permian basin in the USA as reported by Zhang et al. (2020). It is very unlikely that the SSWR enhancement is an artefact of the known aerosol or surface albedo biases in the TROPOMI XCH4 data. We elaborate further on this in Appendix 200 Sect. A1. Figure 2 shows seasonally average XCH4 maps over SSWR, and Table 1 quantifies the seasonal XCH4 enhancement and areal coverage of the TROPOMI data. TROPOMI has good coverage in SSWR, ranging from 40 % in JJA to > 90 % DJF.
It is higher than 70 % in all seasons except JJA, likely due to persistent cloud cover during the wet season. The lowest enhancements are observed in JJA in both 2018 (10.5 ± 4.1 ppb) and 2019 (2.6 ± 3.7 ppb). It is unlikely that these low enhancements are artefacts of the low coverage as there is still sufficient TROPOMI data (> 40 %) and measurement are not 205 systematically missing over the large emissions areas of SSWR, the Sudd and Machar wetlands. SON-2019 has the largest enhancement of 29.3 ± 4.0 ppb, likely due to low wind speeds.

Emissions quantification
We use the mass balance method of Buchwitz et al. (2017) to estimate emissions from SSWR for each season (see Table 1).
Emissions during most seasons are close to 10 Tg yr -1 , except for the low emissions in JJA. We find very low emission in JJA- Tg yr −1 ). Direct application of the mass balance method on the two-year average XCH4 map shown in Fig. 1a yields an annual emission of 10 ± 1.7 Tg yr -1 . However, this is likely an overestimate as the two-year average temporally under samples the low emissions of JJA seasons due to low coverage during these seasons. Therefore, to ensure uniform temporal sampling of 215 all seasons, we calculate annual SSWR emissions by averaging the seasonal emission estimates, resulting in 8.2 ± 3.2 Tg yr -1 .
Moreover, this approach is likely less sensitive to error due to mean-of-products vs product-of-means effect. A caveat of the mass balance method is that it ignores two factors: (1) the influence of emissions in the background region, and (2) the contribution of emissions in the source region to the background average XCH4. Both factors increase the background XCH4 and ignoring them results in an underestimation of the emission estimate. However, this underestimation is large when the 220 ratio of the area of the background region and the source region is small, and as we apply the method using a large background, we do not expect a significant impact on our emission estimates. Lunt et al. (2019) report methane emissions for all sources (including wetlands, biomass burning, anthropogenic, wild animals) from the Sudd wetlands using multiyear GOSAT inversions. Their emission estimate of 5.2-6.9 Tg yr -1 for 2016 is within the 225 uncertainty bounds of our SSWR total emission estimate of 8.0 ± 3.2 Tg yr -1 for 2018-2019. Note that some difference in the emission estimates can be explained by the difference in the definition of the source region between the two studies as their region extends more north and less east than ours. TROPOMI shows a large XCH4 enhancement over Lottila and Machar wetlands in the east SSWR, indicating large emissions from these wetlands. As the source region in Lunt et al. (2019) only partially covers these wetlands, our emission estimates are expected to be higher. 230 To calculate wetlands emissions from SSWR, we account for other methane emissions in the region using bottom-up data.
The region has small emissions from wastewater management (0.03 Tg yr -1 ), energy for buildings (0.01 Tg yr -1 ) and manure 235 management (0.01 Tg yr -1 ). EDGAR does not report any significant emissions from fossil fuel exploitation sector in the region.
Recently, Scarpelli et al (2020)  source is the emission from termites (0.16 Tg yr -1 , Sanderson, 1996). We subtract the total of these non-wetlands emissions to calculate wetland emissions of 7.4 ± 3.2 Tg yr -1 from SSWR in 2018-2019. This estimate is an order of magnitude larger than the 0.5 Tg yr -1 wetlands emissions from the prominent Pantanal wetlands of South America in 2010-2018 which are estimated using GOSAT inversions by Tunnicliffe et al. (2020).

245
Our SSWR wetlands emission estimate of 7.4 ± 3.2 Tg yr -1 can be an overestimate if the emissions from the abovementioned non-wetlands sectors are underestimated in the inventories. However, this is unlikely as it would require a very large underestimation in the inventories for the two years studied here. For example, for the oil and gas sector, the annual emissions (0.05 Tg yr -1 ) will need to be underestimated by two orders of magnitude to have a significant error impact on the wetland emission estimates. Moreover, the strong seasonality shown by the TROPOMI emission estimates is not expected in 250 oil and gas emissions. The SSWR biomass burning emissions are higher in comparison to the other sectors, but a large  (FAOSTATS., 2020). This amount is twice of what we use to calculate the wetlands emissions for SSWR. In the extreme case that all these additional emissions are located in SSWR, it would slightly reduce our wetland emission estimate, however, well within its uncertainty margin. Manure management emissions in SSWR (0.01 Tg/yr) are small even though there is a large cattle population in South Sudan due to lack of effective management practices. This is reflected in the small emission factors used for the country by EDGAR for dairy cattle: 1 kg 260 CH4 head -1 for South Sudan vs 48 kg CH4 head -1 for USA.

Annual means
SSWR integrated mean methane emission estimates from the process models are nearly an order of magnitude lower than those from TROPOMI (Table 2) Table 2 also presents the maximum inundation extent (i.e., sum of seasonal and permanent wetland areas) used by the process models. It range from 25,000 to 69,000 km 2 across the models. These inundation extent are up to 4 times lower than the 275 observation-based maximum inundation extent estimates of 99,000 km 2 by Gumbricht et al. (2017). Huges & Huges (1992) give the permanent wetland area of the different wetlands in SSWR (Table A1). The sum of these areas is 36,000 km 2 , significantly larger than the permanent inundation extent (i.e., minimum inundation extent ) used in the models (Wetcharts Extended Ensemble: 1,000 km 2 ; LPJ-wsl: 14,000 km 2 ; SWAMPS: 16,000 km 2 ). Rebelo et al. (2011) used remote sensing data to characterize inundation extent of the Sudd wetlands over a 12 months period, yielding a total wetlands area of 50,000 km 2 280 (41,000 km 2 of seasonally inundated and 9,000 km 2 of permanent inundated). According to Huges & Huges (1992), other wetlands in the SSWR have a total permanent wetlands area of >20,500 km 2 , meaning that Sudd accounts for only about a third of the SSWR's total wetland area. As other wetlands in SSWR are also along rivers like Sobat, their inundation extent likely has a large seasonality, and assuming that the relative seasonal amplitude of inundation extent of these other wetlands is similar to that of Sudd would give a total (seasonal + permanent) flooded area of 134,000 km 2 . Adding the Sudd inundation 285 extent yields a total SSWR inundation extent of 164,000 km 2 , which is larger than the total estimate of 99,000 km 2 from Gumbricht et al. (2017). Overall, we find substantial evidence of underestimations of SSWR inundation extent in the process models, which may explain their emission underestimations as they assume that inundation extent is a strong control of the emissions.

290
We now look at variations in annual mean inundation extent to find a possible cause of high emissions in 2018-2019. Lunt et al. (2019) attribute the emission increase in South Sudan between 2010 and 2016 to an inundation extent increase in the Sudd owing to an increased water inflow from the White Nile river found in satellite altimetry-based river water height measurements. To investigate this for the period 2018-2019, we look at trends in water height (see Fig. 4) of Lake Victoria, and White Nile and Sobat rivers. Similar to Lunt et al. (2019), we observe a rapid water height increase during 2011-2014. 295 After this period, water levels stabilize and slightly decrease but remain significantly higher than in 2009-2010. 2019 shows the highest water level for the Sobat river due to a renewed positive trend from 2017 onward. This suggests that the total SSWR inundation extent was significantly higher in 2018-2019 than the pre-2011 levels. In contrast, the inundation extent data used in the process models, shown in Fig. 4a, have negative trends, which means that the process models do not account for the emission increase during 2010-2016 due to increasing inundation extent, as suggested by Lunt et al. (2019). 300 Inundation extent estimates from the remote sensing-based SWAMPS also do not show the increase and underestimate annual means. Schroeder et al. (2015) have recommended not to use SWAMPS absolute inundation extent as the microwave sensors used in SWAMPS have limited capability to detect water underneath the soil surface or beneath closed forest canopies. This effect can impact also the temporal changes, in addition to the absolute inundation extent, as such flooding beneath the forest 305 canopies would also not be observed. It is unclear why TOPMODEL, which accounts for lateral water transport processes, does not capture the trend in river outflow. These are interesting topics for follow-on investigations.

Seasonal cycle
Next, we assess the seasonal cycle of the TROPOMI-derived emission estimates. Figure  where R is correlation coefficient), indicating that the seasonality of emissions is driven by inundation extent. In fact, the differences in inundation extent seasonality between LPJ-wsl and Wetcharts are consistent with the emissions differences; for 315 example, both inundation extent and emissions in LPJ-wsl are lower than in Wetcharts during MAM.
The seasonality of the altimetry-based river water height measurements, shown in Fig. 5d, is highest in SON and is very different from Wetcharts inundation extent (highest in JJA). This can partially explain the difference in the seasonal cycles of Wetcharts and TROPOMI emissions. The seasonal cycle of Wetcharts inundation extent is strongly correlated with local 320 precipitation (Fig. 5b), as the intra-annual inundation extent variation is calculated using precipitation. However, this method would not accurately account for inundation extent variation due to lateral water fluxes and evapotranspiration. Surface runoff is especially important for river-fed wetlands like Sudd, whose inundation extent is controlled by water inflow from the White Nile because the evapotranspiration rate exceeds rainfall in the region (Lunt et al., 2019;Sutcliffe and Brown, 2018). LPJ-wsl inundation extent seasonality shows better agreement with the river height data as it is calculated using TOPMODEL, which accounts for the lateral fluxes and evapotranspiration. However, LPJ-wsl emissions still show large differences with the seasonal cycle of TROPOMI emissions. Previous remote sensing studies for the Sudd wetlands have found the largest inundation extent during September-January in 2007-2008(Robelo et al., 2012 and during December-January in 1991-1992 (Travaglia et al., 1995), in better agreement with river height measurements than the process models. Overall, inundation extent seasonality of models appears to be significantly off, which can explain part of the mismatch between TROPOMI and model 330 emissions. In both 2018 and 2019, TROPOMI emission estimates are the lowest during JJA, while river height measurements are the lowest in MAM. A similar seasonal cycle mismatch in the GOSAT emission estimates and inundation extent, derived using MODIS Land Surface Temperature (LST) as a proxy, is shown in Lunt et al (2019). Furthermore, they find the highest emissions trend during SON, which had the smallest trend in inundation, but no trend in emissions during MAM, which has the highest inundation extent trend (i.e. strongest negative LST trend). 335 An explanation for the difference in seasonal phasing can be a higher temperature dependence of emissions than suggested by the models as temperatures are lowest during JJA. We evaluate this hypothesis using Wetcharts Full Ensemble, which provides a total of 324 emission estimates for three temperature dependences q10 (=1, 2, 3; see Bloom et al., 2017). Figure 6 compares the average seasonal cycle of TROPOMI emissions with Wetcharts emissions using different q10's (see also Table 3). 340 Wetcharts emissions with q10 = 1 have the poorest agreement with the seasonal cycle of TROPOMI (R = -0.62). Interestingly, these emissions also have the lowest annual means (= 0.5 Tg yr -1 ). Conversely, Wetcharts emissions with q10 = 3 have the best correlation with TROPOMI (R = -0.28) and have the largest annual mean (=1.0 Tg yr -1 ). In fact, the member estimate-out of the 324-member Full Ensemble --with the largest annual emissions of 3.7 Tg yr -1 has the best correlation with TROPOMI (R = 0.00). As expected, this member uses q10 = 3. The agreement of TROPOMI with the larger q10 model 345 estimates, in terms of both annual mean and seasonal cycle, suggests that wetland emissions from SSWR have a large temperature dependence. In their study of wetlands in the Amazon Basin, Tunnicliffe et al. (2020) pointed to temperature as a more important control on methane emissions than inundation extent. They find a simultaneous, spatially correlated emission and temperature increases in the west Brazilian Amazon during the El-Nino of 2015, with unchanged inundation extent.
Moreover, Wilson et al. (2016) found a negligible impact on wetlands emissions in the Amazon basin despite the large 350 difference in precipitation between 2010 and 2011, which impacted inundation extent significantly. Note that it is also possible that the higher q10's we find for SSWR emissions are simply compensating for errors due to a misrepresentation of inundation extent or other factors covarying with temperature. Figure 7 shows the emission anomalies time series from TROPOMI along with temperature and inundation extent, which we 355 assume to be proportional to river height. A small lag between the river height and inundation extent is expected, but we expect it to be negligible in comparison to a full season. We observe that the emissions show a strong correlation with temperature (R= 0.49), but a poorer correlation with inundation extent (R= 0.24). The emissions peak a full season later than inundation extent, and accounting for this seasonal lag improves the correlation significantly (R= 0.80). An explanation for this can be the higher temperature dependence of emissions discussed earlier. Another explanation could be the "activation" time of 360 methanogenesis, as after flooding it takes time for anoxic conditions to develop and alternative electron acceptors to be depleted. Jerman et al. (2009) documented that methane emissions from water-saturated soil slurries remained very low for a long time: methane production started after a lag of 84 days at 15° C and a minimum of 7 days at 37° C, the optimum temperature for methanogenesis. They found that the lag was inversely related to iron reduction, which is expected as iron reduction outcompeted methanogenesis. Similarly, Itoh et al. (2011) investigated methane emissions from rice paddy fields 365 and found a time lag of a few weeks between the onset of inundation and peak emissions.
Process models assume that wetland emissions are instantaneously regulated by inundation extent, and they do not account for the time lag as information on the availability of alternate electron accepters is generally not available. This results in an incorrect temporal allocation of the wetland emissions. Furthermore, some models assume inundation extent is instantaneously 370 regulated by precipitation. In river floodplains like Sudd, inundation extent is mostly controlled by river inflow, and not the local precipitation, as the evapotranspiration rates exceed the rainfall in the region. Therefore, scaling with precipitation would even worse emission estimates. Overall, a combination of temperature and inundation extent dependences that are used in the models can explain their seasonal cycle mismatch with TROPOMI emissions.

Conclusions
XCH4 enhancements over South Sudan have been observed in remote sensing studies suggesting large emissions from the Sudd wetlands as the cause (Lunt et al., 2019, Hu et al., 2018, Frankenberg et al., 2011. We observe two large enhancements in the region in a 2-year average map of TROPOMI XCH4--over Sudd, and Machar and Lotilla wetlands. Sudd Wetlands are 380 flooded by the White Nile river originating from Lake Victoria, while the wetlands in the east are around smaller rivers like the Sobat originating in the Ethiopian mountains. In this study, we examine these wetlands, and their river systems, together to understand the controls of the emissions causing the large XCH4 enhancements. We estimate methane emissions of 7.4 ± 3.2 Tg yr -1 from wetlands in South Sudan during 2018-2019 using a mass balance 385 approach applied to TROPOMI data. We find large differences between the emission estimates from TROPOMI and wetland process models LPJ-wsl and Wetcharts. The annual mean estimates from TROPOMI are an order of magnitude larger than mean estimates of the models, which may be explained by the up to 4 times underestimated inundation extent in the models. We find differences in interannual variability and average seasonal cycles of TROPOMI and models, which can be again partially explained by the strong dependence of model emissions on poor inundation extent estimates. We find the lowest 390 emission in the highest perception and lowest temperature season JJA, when models estimate large emissions as they incorrectly assume an instantaneous influence of the precipitation-derived inundation extent. We find that the Wetcharts emission estimates that use a stronger temperature dependence (q10 = 3) show a better agreement with TROPOMI concerning both seasonality and annual emissions. This indicates that the models may also underestimate the temperature sensitivity of the methane emissions. 395 The inundation extent of SSWR is analyzed using satellite altimetry-based river height measurements of White Nile and Sobat rivers at locations within the Sudd and Macher wetlands. The inundation extent estimates used in models are based on the local precipitation, whereas, the actual inundation extent of SSWR is driven by water inflow from the rivers as evapotranspiration exceeds the precipitation in the region. As a result, both the seasonal cycle and trend of model inundation extent disagree with 400 river height data. The seasonal cycle of inundation extent from river height data shows better agreement with the TROPOMI emissions when a full season-long lag between the two is assumed. This time lag can be explained by the time needed for methanogenesis to develop in the seasonally flooded areas of the wetlands. A more precise estimate of the lag is not possible due to the coarse temporal resolution of our TROPOMI emissions estimates.

405
The lack of information on the correct relationship of wetland emissions with inundation extent and temperature results in large model uncertainties. Such large gaps in our understanding of the processes driving wetland emissions call for further investigation. As shown here for the wetlands of South Sudan, TROPOMI provides valuable observations over remote and inaccessible wetland regions of the world, which future wetland studies can take advantage of.

Section A1. Systematic Measurement Uncertainties
Surface albedo and aerosols can alter the optical light path, introducing biases in XCH4 (Butz et al., 2011). Therefore, the XCH4 enhancement over South Sudan can be affected by the differences between the source and background region values of these parameters. The average retrieved aerosol optical thickness (AOT) and surface albedo in the SWIR band of TROPOMI 415 are shown in Fig. A1. For SSWR and its background, the AOT and albedo differences in two-year average data are 0.001 and -0.10, respectively. The average differences for seasonal average maps are -0.01 ± 0.01, -0.15 ± 0.02 and 16.3 ± 8.4 ppb for AOT, albedo and XCH4 respectively. The negative albedo difference for SSWR occurs due to the high albedo the Sahara in the background. This small albedo difference is unlikely to influence the SSWR XCH4 enhancement significantly, especially, as an albedo-based bias correction is applied to the XCH4 in operational TROPOMI product (Hasekamp et al., 2019). We also 420 examined the possibility that the XCH4 enhancement over South Sudan is an artefact of sun glint geometry of TROPOMI observations due to refection on standing water of the Lakes and inundated areas in the region. This can happen when the observation geometry over a water body surface is at the specular reflection angle (i.e., the viewing zenith angle matches the solar zenith angle) causing a spike in the level 1 radiance measurements. However, this was found not to occur over the 29.3 ± 4.0 11.2 ± 1.9