Phenology is the dominant control of methane emissions in a tropical non-forested wetland

Tropical wetlands are a significant source of atmospheric methane (CH4), but their importance to the global CH4 budget is uncertain due to a paucity of direct observations. Net wetland emissions result from complex interactions and co-variation between microbial production and oxidation in the soil, and transport to the atmosphere. Here we show that phenology is the overarching control of net CH4 emissions to the atmosphere from a permanent, vegetated tropical swamp in the Okavango Delta, Botswana, and we find that vegetative processes modulate net CH4 emissions at sub-daily to inter-annual timescales. Without considering the role played by papyrus on regulating the efflux of CH4 to the atmosphere, the annual budget for the entire Okavango Delta, would be under- or over-estimated by a factor of two. Our measurements demonstrate the importance of including vegetative processes such as phenological cycles into wetlands emission budgets of CH4. Tropical wetlands are a significant but understudied source of methane. Here, methane emissions were measured over three years in a perennial tropical swamp in the Okavango Delta, Botswana, finding phenology was the overarching control of emissions.

A tmospheric concentrations of methane (CH 4 ), the second most important greenhouse gas after carbon dioxide (CO 2 ), have increased steadily since 2007 1-3 after nearly a decade of zero growth 4 but the causes for this renewed increase are not fully understood. Possible explanations include (1) an unidentified increase in CH 4 emissions from anthropogenic sources such as oil and natural gas 5 , (2) a reduction in CH 4 destruction due to changes in the oxidative capacity of the atmosphere 6,7 , (3) an increase in biogenic emissions inferred from a shift in the isotopic signatures of atmospheric CH 4 8 and (4) a decrease in biomass 9 burning reconciling the combined effects of (1) and (3). The isotopic shift supports the idea of an increase in biogenic emissions, particularly from the tropics 10 that contribute~65% of the global CH 4 budget. Estimates of global emissions have large uncertainties, with bottom-up (BU) budgets (inferred from process-based models and inventories) overestimating top-down (TD) budgets (estimated through inversion modelling constrained by atmospheric CH 4 concentration measurements) by ca. 30% 11 . Wetland budgets are particularly uncertain: for example, during the 2000-2009 period, global estimates were 9% larger for BU than TD budgets 12 , but this figure was recently revised and BU are currently 22% smaller than TD budgets 11 . Despite these uncertainties, the current estimate is that emissions from tropical wetlands have increased by~10% between the 2000s and 2017, and account for ca. 20% of the global CH 4 budget. These BU and TD estimates provide little information about the underlying processes so there is an urgent need to consider understudied emission mechanisms and processes that might help reconcile estimates from the two approaches.
This is particularly timely, given that revisions of conventional model parameters such as wetland extent [13][14][15] and temperaturesensitivity of methanogenesis 16 fail to reconcile BU and TD estimates or explain recent inter-annual variations in emissions. Net fluxes of CH 4 to the atmosphere result from complex and sometimes competing processes, which underpin production, oxidation and transport, and the magnitude and temporal dynamics of these terms are intimately linked to vegetative processes and growth cycles. For example, the availability of C for CH 4 production in soils, either from plant litter or photosynthates, can control CH 4 production and contribute to the modulation of daily to seasonal emissions in ecosystems spanning subarctic to subtropical latitudes 17,18 , and the level of methane oxidation in soils has been shown to be larger in vegetated soils due to plant-mediated oxygenation of the rhizosphere 19,20 . Plant-mediated transport of CH 4 can be the dominant transport mechanism 21 , but the efficiency of this pathway can be speciesdependent and variable [22][23][24] .
Recent studies in the tropics have shown that trees can be substantial sources or pathways of CH 4 to the atmosphere 25,26 , but less is known about the role of emergent macrophytes in nonforested tropical wetlands, which account for 20-37% of the global land surface of vegetated wetlands [27][28][29] . Whilst parameters such as soil C and inundation are commonly used in processbased models of CH 4 emissions, transport, and particularly plantmediated transport of CH 4 is a relatively poorly represented pathway. Regional and global estimates of wetland CH 4 emissions from models with and without explicit treatment of the transport pathway can vary by up to a factor two 30 , and uncertainties on emissions from the data-poor tropics are particularly large.
In this work, we report on three years of measurements of land-atmosphere exchange of CH 4 in permanent and seasonally flooded wetlands in the Okavango Delta, Botswana, and demonstrate that net CH 4 emissions are broadly controlled by seasonal changes in regional hydrology and, particularly, by the associated phenological cycle of the wetland vegetation. Furthermore, the marked diel cycles in methane emissions observed at the perennial swamp study site, challenge the common practice of upscaling daytime fluxes to higher temporal statistics. We also demonstrate that the seasonality in CH 4 fluxes measured on the ground is observed by satellite data and that we can reconcile annual BU and TD CH 4 budgets for the entire Okavango Delta. We established two eddy-covariance (EC) sites in the Okavango Delta in August 2017 to quantify CH 4 fluxes (F CH4 ) from a seasonal floodplain and from a permanent wetland. These EC systems were still active at the time of writing and our EC dataset is, to the best of our knowledge, the longest continuous record of wetland CH 4 fluxes in Africa.
Results and discussion CH 4 methane emissions at the perennial swamp. The largest CH 4 emissions (0.10-0.64 g m −2 d −1 ) in the Okavango Delta are from the perennial swamp areas, which are dominated by emergent Cyperus papyrus and Phragmites australis. Monthly median fluxes of CH 4 (F CH4 ) measured at the Guma Lagoon papyrus swamp (18°57′ 53.01″S; 22°22′16.20″E) are linearly and positively correlated (R 2 = 0.89, p-value from t distribution <0.05) with monthly maximum gross primary production (GPP MAX , "Methods") as shown in Fig. 1a. GPP MAX, which represents the maximum photosynthetic potential of the papyrus stand, is a measure of the vigour/health of the stand at a particular time and the strongest predictor for F CH4. F CH4 and GPP MAX tend to increase with air and water temperature but these relationships are not statistically significant. This is consistent with other studies, which documented significant correlations between CH 4 fluxes and GPP at monthly to seasonal timescales. At such timescales, the net CH 4 fluxes to the atmosphere are most likely modulated by the CH 4 production term, which is stimulated by carbon (C) deposition in the rhizosphere from photosynthates [31][32][33] .
We also found a positive correlation between F CH4 and monthly estimates of enhanced vegetation index (EVI; Fig. 1b, R 2 = 0.69, p-value < 0.05), and a comparable correlation (R 2 = 0.66, p-value < 0.05) between GPP MAX and EVI. EVI, a proxy for leaf phenology or biomass 34 was obtained from Sentinel-2 imagery for the portion of the papyrus stand within the flux footprint of the EC tower. The phenological cycle of papyrus is complex: instead of a seasonal full stand die-back, senescence and recruitment of new shoots co-occur throughout the year, but the ratio of mature to senescing plants is variable 35 . The phenological cycle of papyrus inferred from EVI is characterised by high EVI values (green vegetation/more aboveground biomass) during summer months (December-February) followed by a gradual decline (senescence) until winter (June-August). We propose that GPP MAX, a metric comparable to the photosynthetic capacity of Wu et al. 34 is an indicator of the maturity of the papyrus stand or of leaf ontogeny. GPP MAX might hence be more representative of, or biased towards, mature plants than EVI, because these are more productive/photosynthetically active than juvenile and senescent ones; as demonstrated by Wu et al. 34 , leaf ontogeny is a strong predictor for GPP, but EVI cannot resolve leaf age which results in a weaker correlation. CH 4 emission mechanisms and processes in the perennial swamp. The correlation between F CH4 and EVI is likely an artefact of the circular dependency of F CH4 on GPP MAX and GPP MAX on EVI, but based on the available data it is not possible to rule out an element of seasonality in the plant-mediated transport term of CH 4 . Indeed, a small study on papyrus reported negligible CH 4 fluxes through the culms of juvenile and senescing individuals. By extrapolation, this could mean that the largest CH 4 fluxes for an entire papyrus stand occur when the proportion of mature plants reaches a maximum, i.e. when GPP MAX , but not necessarily EVI, is at its maximum. This is consistent with the findings of Whiting and Chanton 18 , who explained the strong positive linear correlations between F CH4 and biomass in a variety of plants, as a combination of rates of the plant-mediated organic substrate supply and plant-mediated transport. Working on the assumption that CH 4 is mainly lost through non-stomatal pathways such as pores in plant stems, they suggested that more biomass could equate with larger conduits for CH 4 transport and hence greater fluxes. In conclusion, whilst variations in physiology or plant-mediated transport efficiency are plausible, the dominant control of the seasonal cycle in CH 4 emissions is probably C-deposition to the rhizosphere.
We observed marked diel cycles from October to March (Fig. 2), characterised by a decrease in F CH4 during the central daylight hours when vapour pressure deficit (VPD) was high and CO 2 uptake peaked. Overall, emissions of CH 4 were 52 ± 26% larger at night than during the day (median ± inter-quartile range of monthly night to day ratios), and this night-to-day ratio increased with increasing EVI (Extended Fig. 2). Both day and night-time fluxes were linearly correlated with EVI but the night-time slope was almost double the daytime one. Night-time fluxes also had a weak statistically significant correlation with mean air temperature, but that was not the case for daytime fluxes. The implication of these differences between night and day fluxes is that estimating daily or higher temporal budgets from daytime measurements would lead to significant underestimations. To the best of our knowledge, there is no published literature on the dominant plantmediated transport pathway for CH 4 in Cyperus papyrus, but a limited study documented constant CH 4 efflux from mature culms throughout daylight hours, and negligible emissions from juvenile and senescent plants 35 . Such constant emission patterns do not fit the marked diel trends observed for most of the year, but stomatal flux regulation is a possibility. Jones and Muthuri showed 36 that the stomatal conductance in papyrus canopies exhibits a sharp early morning rise followed by partial closure around midday as VPD increases. Partial stomatal closure reduces transpiration, even when root zone water availability is high, but has a lesser impact on photosynthesis in C4 plants 37 such as papyrus 38 . Consequently, partial stomatal closure could reduce CH 4 emissions as well as transpiration, while affecting CO 2 uptake, and by extension GPP max , to a smaller degree. CH 4 fluxes through the umbels of the plants, which are comprised of hyperstomatal bracteoles and rays, and their diel cycles are however unknown. Although plausible, stomatal control of CH 4 fluxes in papyrus at short timescales remains speculative.
Pressurisation of the internal lacunae found in many aquatic macrophytes in response to increasing VPD and, to a lesser degree, air temperature, can give rise to convective gas flow 39 , albeit with temporal cycles opposite to the ones we measured over papyrus 39,40 . Some authors have attempted to reconcile the concept of convective flow with observed trends of decreasing CH 4 emissions during daylight hours as venting of CH 4 accumulated overnight inside the plants until fluxes become limited by pressurised ventilation, or by root-shoot gas transport rather than stomatal conductance 23,31,41 . To the best of our knowledge, pressurisation has not been studied in Cyperus papyrus, but other members of the Cyperaceae family (e.g. C. involucratus and C. eragrostis) are known not to produce significant convective flow 39 .
Diel cycles of O 2 fluxes through plants, and concentrations within the rhizosphere, which can decrease by 30% at night in papyrus as a result of metabolic O 2 demand and the absence of photosynthetic production 42 , offer an additional mechanism for the measured trends. A concomitant reduction in CH 4 oxidation at night could shift the balance of the CH 4 production and consumption terms towards a net increase in CH 4 , giving rise to diel emission trends consistent with our measurements 43 . This could also explain the strong correlation between CH 4 fluxes and EVI, particularly at night (Supplementary Fig. 1), as more live biomass (high EVI) could equate with higher metabolic O 2 demand and higher net CH 4 fluxes. In all likelihood, the trends in net CH 4 emissions result from a superposition of several processes, but our dataset cannot resolve them.  The seasonal trend seen in EC data was also observed in satellite-derived fluxes, albeit with an apparent time lag of ca. one month between EC and satellite ( Fig. 3). This temporal difference is likely due to an asynchronicity between the environmental controls of the CH 4 fluxes observed locally (EC) and at the scale of the entire Delta (satellite-derived fluxes). Furthermore, interannual variations in surface properties may have impacted the seasonality of the CH 4 fluxes: for example, in 2019, the onset of greening and browning of the papyrus stand at Guma Lagoon (inferred from EVI, Supplementary Fig. 2) lagged other years by one to two months, which indicates a variability in the timing of vegetation development. This was not tested at the broader basin scale, but basin-wide means of both surface temperature and EVI account for 30% of the variance in estimated CH 4 emissions.

Controls of CH 4 emission dynamics at the seasonal floodplain.
At Nxaraga seasonal floodplain (19°32′53″S; 23°10′45″E, site description in Methods), which is dominated by grasses such as Panicum repens, Cynodon dactylon, and Sporobolus spicatus, monthly median CH 4 emissions ranged from 0.006 to 0.262 g m −2 d −1 and exhibited a complex dependence on both water level (WL) and air temperature (T air , Fig. 4). Figure 4 suggests that water level and air temperature are predictors of F CH4 , until water levels drop below a critical threshold, at which point the these variables uncouple from F CH4 . In 2017 and 2018, CH 4 emissions increased from winter to summer as T air increased (R 2 = 0.06, p-value = 0.4) and WL decreased (R 2 = 0.43, p-value = 0.05). Whilst the correlation between F CH4 and T air was not statistically significant (Fig. 4b, p-value = 0.4), both typically increased from ca. June until the end of the year. This was accompanied by a gradual decrease in WL. 2019, a year of severe drought, did however not fit this trend: both WL and CH 4 fluxes remained relatively low throughout the year, and the fluxes were decoupled from air temperature and WL. Microbial activity and CH 4 production scale with temperature 16 , whilst soil water content, or WL, regulates several processes contributing to the net CH 4 flux measured above the surface: (1) it creates the anoxic conditions necessary for CH 4 production, (2) it decreases CH 4 oxidation by reducing the amount of oxygen available to methanotrophic microbial communities, and (3) it impacts the efficiency of the diffusion of CH 4 out of the soil. The combined effect of these terms, which operate in parallel, determines the magnitude and sign of the net surface-atmosphere flux of CH 4 . The Okavango Delta receives pulsed seasonal flooding leading to alternating periods of soil wetting (April-August) and drying (September-March) in some parts the Delta; this hydrological regime impacts the balance of CH 4 production, oxidation and diffusion terms and thereby the magnitude of the net fluxes that alternate between periods of net emissions of CH 4 to the atmosphere, and periods of low emissions or even net oxidation, which we have observed at Nxaraga seasonal floodplain. Our findings are consistent with published experimental results and modelling approaches of wetland CH 4 Fig. 4 Relationship between monthly methane fluxes, water level and air temperature. Monthly median CH 4 fluxes (FCH 4 , g m −2 day −1 ) ± inter-quartile range (IQR) from August 2017 to August 2020 as a function of (a) monthly water level, and (b) mean air temperature at the seasonal floodplain measurement site (19°32′53″S; 23°10′45″E). The number of half-hourly flux data points (n) from which median and IQR were calculated changed from month to month because of the variability of points filtered out by the micrometeorological quality control filter (see Methods). Consequently, n ranged from 33 (August 2020) to 570 (September 2019). The solid lines represent linear regressions on a temporal subset of the data (austral winter to summer 2017 and 2018) excluding the 2019 drought period (equations and t-test two-sided p-value given in the panels): standard error of (a) ±0.10 and 0.09, and (b) ±0.01 and 0.16 for slope and intercept, respectively.
Methane flux upscaling from local to Delta scale. We upscaled the measured emission budgets to the entire Okavango Delta by defining three ecohydrological areas with distinct CH 4 flux characteristics (Methods): (a) the papyrus swamp at Guma Lagoon was used as proxy for all permanently flooded areas, (b) the seasonal floodplain at Nxaraga was the proxy for all seasonally flooded areas, and (c) we assumed that net oxidation fluxes, as measured using closed chambers over dry sandy soil at the seasonal floodplain (see Gondwe et al. 46 for methodology), prevailed at the occasionally flooded (defined as flooded at least once per decade) areas throughout the year. Based on these assumptions, we estimate using our EC data that the Okavango Delta was a net annual source of CH 4 to the atmosphere of 0.48 ± 0.09 and 0.24 ± 0.03 Tg year −1 in 2018 and 2019, respectively (Figs. 5, 6). The 2019 budget compares well with the independent emission estimate inferred from satellite observations (0.23 ± 0.05 Tg year −1 ). The 2018 EC budget was more than double the satellite estimate (0.18 ± 0.05 Tg year −1 ), which could be due to an overestimation of the contribution of the seasonal floodplain. The EC flux footprint entrains a highly heterogeneous landscape of seasonally flooded grasslands traversed by a major river, which means that the fluxes are not fully representative of the land area classified as seasonally flooded. Selecting only data points for which 90% of the flux originated from within 200 m from the EC tower to restrict the contributions from riparian and aquatic source areas, the local annual budget for Nxaraga is reduced by 34% in 2018 and 78% in 2019, but uncertainties are large (0.12 ± 0.05 Tg in 2018 and 0.003 ± 0.010 Tg in 2019). In addition, we used maximum wetland extent to upscale local fluxes to the regional scale, but this merely approximates the seasonal dynamics because peripheral areas of the alluvial fan experience shorter flooding durations. As a result of these approximations, the upscaled budgets for the seasonally flooded wetlands are likely overestimated. Wetland extent was the dominant control of interannual variability in the upscaled budget for perennial swamps.
Whilst spatial uncertainties could not be quantified, the extent of the perennially flooded areas varies little at the seasonal timescale, and we therefore expect more robust upscaled CH 4 budgets. However, it must be also be noted that the uncertainties arising from using papyrus as a proxy for other major macrophyte communities (e.g. phragmites) are unknown. Were et al. 47 , did not observe statistically significant differences in the magnitude and seasonality of soil CH 4 emissions between papyrus and phragmites plots in Uganda, but van den Berg et al. showed 48 that Phragmites australis possess strong diel cycles characterised by elevated emissions of CH 4 during daylight hours (a reverse emission cycle to that we observed in papyrus) during the growing season, and CH 4 venting through dead culms has been documented 49,50 . The percentage coverage of phragmites and papyrus in the Okavango Delta being unknown, the impact different emission mechanisms and patterns on upscaled fluxes cannot be estimated.
Using simplified assumptions, CH 4 emissions from the permanent, vegetated wetlands accounted for 63% and 97% of the overall CH 4 emission budget for the Okavango Delta during 2018 and 2019, respectively. These wetlands hence play a disproportionate role in the overall budget compared to their source area (24% of total extent).
We estimate that CH 4 emissions (4.8 Tg year −1 ) from African Cyperus papyrus alone (0.1% land cover) 51 account for 6% of the continent's total CH 4 emissions (85 Tg year −1 ) 12 . However, without considering seasonality in fluxes as observed at the perennial swamp, these estimates could range from 2% to 10% of the continent's budget (calculated using annual minimum and maximum emissions, respectively). This is a significant uncertainty range, equivalent to 6-30% of the inter-decadal global increase in CH 4 emissions from all sources (TD, 2010s compared to 2000s) 11 . This exemplifies the magnitude of the uncertainties on emissions from tropical vegetated wetland, and the urgency to better constrain them. This will require the development of a detailed knowledge of the level of emission modulation performed by wetland vegetation globally, and identify direct observables (e.g. GPP) or proxies thereof (e.g. EVI) to upscale locally derived budgets and refine process-based models. In particular, understanding the vegetative controls on CH 4 production and oxidation, the environmental controls of plant-mediated transport at species level, and how climate and management impact them, will be key to forecasting future emissions of CH 4 in tropical wetlands. Instrumentation. The eddy-covariance instrumentation consisted of a Campbell Scientific IRGASON and a LI-COR 7700 open-path methane (CH 4 ) analyser. The IRGASON consists of a 3D ultrasonic anemometer and open-path infrared gas analyser, providing co-located measurements of the wind vector and mass densities of carbon dioxide (CO 2 ) and water vapour. The IRGASON was oriented into the prevailing wind direction at each site, and the LI-COR 7700 CH 4 analyser was mounted onto a horizontal boom, 0.3 m from the anemometer in the crosswind plane. A Vaisala WXT520 weather station recorded air temperature, pressure, relative humidity, wind speed and direction. Total solar radiation and photosynthetically active radiation (PAR) were measured by a Skye Instruments pyranometer (model SKS1110) and quantum detector (model SKP215), respectively. A Campbell Scientific CR3000 datalogger logged all sensors (sampling rate of 10 Hz for the EC variables and 10-s interval for the meteorological parameters) using a custom data acquisition program written in CR Basic (version 3.7).

Methods
At Guma lagoon, the instrumentation was mounted onto a 3-m high tripod, which was itself located on a 3-m high platform (effective measurement height 5.5 m). The EC system was installed on land, ca. 30 m to the west of a predominantly floating papyrus (Cyperus papyrus) mat. The papyrus mat was partially grounded to the West, where it met the shore of Guma Lagoon and it extended ca. 300 m into the lagoon in an easterly direction. The canopy height was on average 2.5 meters above water level.
At Nxaraga, a 2.5-m high EC mast was erected on the SW edge of Chief's Island to sample greenhouse gas (GHG) fluxes from the seasonal floodplain, which extends several hundred meters to the W, S and SW. The portion of floodplain within the flux footprint of the EC mast was bounded by a permanent, meandering water channel fringed by reeds and grasses such as Phragmites spp and Miscanthus junceus. The vegetation of the floodplain, which is dominated by grasses (e.g., Panicum repens, Cynodon dactylon, Sporobolus spicatus), attracts many types of herbivores (e.g. impala, buffalo) and is grazed for most of the year.
Flux calculations and data quality control. The EC data were processed into halfhourly fluxes using EddyPro ® v.7.0.6. The eddy-covariance theory is well documented 57,58 and will not be discussed here. The core flux processing options ARTICLE were raw data detrending using block averaging, double rotation of the wind vector (i.e. aligning the u-component streamwise and nullifying the vertical and crosswind components), application of the Webb-Penman-Leuning correction 59 to correct for density fluctuations, time lag determination using the covariance maximisation with default approach, and fully analytical methods of spectral corrections for low-60 and high-pass 61 filtering effects.
Half-hourly flux data were rejected from further analysis if any of the following criteria was fulfilled: • Failure of the micrometeorological data quality controls based on the assessment of steady state conditions and integral turbulence characteristics (flag value of 2, following the 0-1-2 flagging system proposed by Foken et al. 62,63 ). Monthly and annual budgets. For each month of the study, half-hourly data were aggregated into hourly bins to construct 24-hour mean and median cycles of CO 2 and CH 4 fluxes. The uncertainty on each hourly mean and median data point was taken as the standard deviation and inter-quartile range, respectively. Aggregating data into hourly bins ensures that each of the 24 hourly points of the monthly diel cycle had the same weighting. This reduces the risk of biasing higher temporal flux statistics (e.g. daily, monthly or annual budgets) towards daytime values, as night time points are more likely to fail the data quality control criteria.
Daily budgets were calculated as the sum of hourly values and the associated uncertainties (σ day ) were obtained using standard error propagation rules (Eq. (1); σ i denotes the uncertainty on the flux value at hour i, with i ranging from 0 to 23).
Monthly budgets and uncertainties were calculated by multiplying the daily values by the number of days in a typical year (365) and dividing by 12; annual budgets were obtained by summing the monthly values. Following error propagation rules, monthly uncertainties were summed in quadrature as in Eq. (1) to obtain the total annual uncertainty.
Carbon dioxide flux partitioning. The fluxes of CO 2 were partitioned into ecosystem respiration (R eco ) and gross primary production (GPP) following the procedure of Saito et al. 64 . For R eco , a two-step procedure was applied 65 : • A non-linear function of temperature (Eq. (2)) was fitted to night time fluxes of CO 2 .
• Assuming that the temperature dependency also extends to daytime, R eco was calculated for all available half-hourly time points using the non-linear parameterisation (Eq. (2); with A and B being fitting coefficients) on temperature (T) obtained with night time data.
GPP was calculated as the difference between R eco and measured net CO 2 flux (F CO2 ; Eq. (3)): Maximum monthly gross primary production (GPP max ) was obtained by fitting the hyperbolic function of the photosynthetically active radiation (PAR) given in Eq. (4) 66 to the values of GPP calculated using Eq. (3).
Temporal data aggregation, CO 2 flux partitioning and plotting were done using R version 4.0.3.
Mapping of the ecohydrological zones of the Okavango Delta. Zones with different flood regimes for the Delta have previously been mapped based on a combination of interpretation of satellite imagery, analogue aerial photography, ground-truthing and rule-based modelling [67][68][69][70] . These zone maps are however dated and of relatively low spatial resolution. Given the dynamic nature of the system, for this study we produced a new distribution map of ecohydrological zones based on a recent time-series of higher resolution remote sensing (Fig. 5) and statistically determined plant communities 71 . The communities were aggregated into larger groups for the purposes of this study, in which we needed to distinguish between perennially flooded areas and floodplains that experience regular seasonal flooding.
Mapping of the annual flood frequency in the Delta was based on a maximum inundation extent dataset derived from Landsat imagery 72 spanning the period 1990 to 2019. In five years of this sequence, one or more of the six annual images needed to produce the mosaicked composite image for the year were not available (1993, 2000, 2009, 2010 and 2012) and these years were excluded from the dataset. Frequencies are thus expressed as a fraction of the 25 time-step sequence, with each time step representing 4% of the total record. Frequency thresholds for transitions from one floodplain vegetation community to the next were identified through a combination of cluster analysis, indicator species analysis and species distribution modelling 73 .
The areas of perennially, seasonally and occasionally inundated floodplain used for the upscaling are summarised in Table 1. The inflow in 2019 was the lowest on record (1934-present) and resulted in a contraction of the perennially and seasonally flooded area for that year, effectively increasing the area under occasional flooding. To capture this shift for upscaling, the annual perennially flooded area was subtracted from the total inundated area derived from high temporal resolution MODIS imagery 68,69 .
Upscaling of eddy-covariance fluxes. An annual CH 4 budget for the entire Okavango Delta was calculated from the temporally and spatially weighted monthly budgets of the two measurement sites in the permanent and seasonal swamps (Eq. (5)). In addition, an annual budget for CH 4 oxidation in the occasional wetland was estimated from static chamber measurements of CH 4 fluxes over dry, sandy soil at Nxaraga. The dry soil chamber measurements were taken monthly (with a few exceptions) from August 2017 until August 2020 ca. 10-20 m inland from the EC mast overlooking the seasonal floodplain.
In Eq. (5), A P , A S and A O stand for the surface area of the permanent, seasonal and occasional wetlands, respectively, summarised in Table 1. F P,i , F S,i and F O,i denote the mean methane flux during month number i in the permanent, seasonal and occasional wetland, respectively.
Determination of the enhanced vegetation index (EVI). We used SNAP (SNAP -ESA Sentinel Application Platform v8.0, http://step.esa.int) to estimate mean EVI values for the fraction of the papyrus stand within the flux footprint of the EC tower at Guma Lagoon (18°57′53.01″S; 22°22′16.20″E) from Sentinel-2 level 2A (bottom of atmosphere) reflectance imagery (tile number 34KFE). Where level 2A data was not available for download from the Copernicus Open Access Hub (http://scihub.copernicus.eu), we used the SNAP plugin Sen2Cor to process level 1C (top of atmosphere) data into level 2A. EVI was calculated for a set region of interest (Fig. 7) from Sentinel-2 spectral bands B2, B4 and B8 using Eq. (6). EVI ¼ 2:5*ðB8 À B4Þ=ð1 þ B8 þ 6*B4 À 7:5*B2Þ ð 6Þ Emission estimates inferred from satellite observations of CH 4 . We use CH 4 emission estimates from the Okavango Delta between December 2017 and March 2020 generated from satellite CH 4 column data from the TROPOspheric Monitoring Instrument (TROPOMI). The methods and estimates we use follow those outlined in a recent study 74 , which we briefly summarise here. The TROPOMI data were quality controlled for cloud coverage, surface albedo, aerosol optical depth and surface topography 73,74 . We use the GEOS-Chem chemistry transport model to relate prior surface emission estimates to atmospheric CH 4 concentrations and ultimately to TROPOMI CH 4 column data. The model was run in a nested configuration with a high-resolution simulation at 0.25°× 0.3125°in the domain 36°S to 20°N and 20°W to 55°E. The concentrations at the edge of this regional domain were informed by the corresponding global model simulation with a coarser horizontal resolution of 2°× 2.5°and empirically fitted to satellite data outside of the regional domain to ensure realistic model CH 4 concentrations at the lateral boundary conditions of the regional domain.
To infer the distribution and uncertainty of CH 4 emissions, we use an Ensemble Kalman Filter (EnKF) inverse method. A prior model simulation was driven each month by independent emission inventory estimates, including from wetlands, anthropogenic sources and biomass burning. To directly compare the GEOS-Chem model and TROPOMI data, we sample the high-resolution model CH 4 concentration distribution at the local time and location of individual TROPOMI scenes and convolve with scene-specific averaging kernels that describe the instrument vertical sensitivity to changes in CH 4 . For our EnKF calculations we used 140 ensemble members, an assimilation window of 15 days, and a lag period of 1 month. Monthly CH 4 emission estimates from the Okavango delta were calculated as the sum of emissions inside a domain of 18.25°S-19.75°S and 22°E-24°E, a region covering the delta. Prior emissions in this region averaged 0.1 Tg yr −1 , of which over 90% were from wetlands during austral summer. Therefore, the posterior total is likely to be largely representative of wetland CH 4 emissions, as opposed to other sources.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The eddy-covariance and meteorological data generated in this study for the Guma Lagoon perennial wetland have been deposited with the UK Environmental Information Data Centre 75 . The eddy-covariance and meteorological data generated in this study for the for Nxaraga seasonal floodplain have been deposited with the UK Environmental Information Data Centre 76 .

Code availability
The data acquisition (Campbell Scientific CR Basic) and data analysis (R) codes used in this study can be requested from the corresponding author. The approaches being standard, we did not deem the codes to merit deposition in online repositories.
Received: 27 April 2021; Accepted: 10 December 2021; The region of interest, from which pixels were sampled to calculate the enhanced vegetation index (EVI) of the floating papyrus stand, is shown as a blue polygon. The approximate location of the water level measurement sensor is also indicated.