Allen and Pavelsky (2018) used Landsat and river stage data to determine river widths at mean annual discharge and judged their estimates accurate for widths wider than 90 m. Since the HydroBASINS dataset used for the river network is derived from a single flow direction algorithm, it cannot represent braided channels, such as occur in some reaches of Amazonian rivers. Their river channel, surface area estimate within the Amazon basin is ∼58,000 km². Mouthbays of rivers, such as the Tapajós, Xingu, Tefé and Coari, were considered lakes, and the analysis did not include the whole delta. Combining their river channel area with estimates for the delta (9,500 km², Castello et al., 2013 plus L.L. Hess, personal communication) and mouthbays of the Tapajós and Xingu (3,800 km², Sawakuchi et al., 2014) results in a river channel area of 77,500 km². An analysis for drainage areas from 1 to 431,000 km² and channels >2 m in width used hydraulic geometry and the drainage network to estimate the Amazon basin (excluding the delta and mouthbays) to have a combined area of about 60,000 km² (Beighley and Gummadi 2011). Given that this procedure included smaller rivers than Allen and Pavelsky’s analysis and used independent data, the two estimates are quite similar. As another example, Rasera et al. (2008) developed empirical relationships between drainage area and channel widths combined with river lengths derived from a digital river network to determine the area of streams and rivers in the Ji-Paraná River basin. Assuming their relationship applied to the whole lowland Amazon basin, an area of 23,000 km² would result for rivers from third to sixth order, similar to that estimated by Beighley and Gummadi (2011) for rivers in that size range. These estimates do not include headwater streams (zero or first orders) that require high resolution topography quite difficult to obtain under forest canopies or labor-intensive surveying. Riparian corridors, often with saturated soils, periodically flooded (Chambers et al., 2004), are also not included.

Large Amazon rivers are typically considered “white waters” with near-neutral pH and relatively high suspended sediment and nutrient concentrations (e.g., Amazon, Madeira, Purus, Juruá and Japurá), “black waters” with low pH, nutrients and suspended sediments and high dissolved organic carbon (e.g., Negro, Uatumã), and “clear waters” with low to neutral pH and low nutrients, suspended sediments and dissolved organic carbon (e.g., Tapajós and Xingu) (Mayorga and Aufdenkampe 2002; Junk et al., 2011). These three types of river water were classified for 6th to 11th order rivers based on field observations and visual inspection of optical images by Venticinque et al. (2016). Smaller rivers and streams are likely to vary considerably in chemical composition, but regional characterization is lacking.

If calculated as the difference between SAR analysis of open water areas at ∼100 m resolution for a period with near-maximum inundation in the central lowland Amazon basin (Hess et al., 2015) minus the river channel areas for rivers wider than 90 m (Allen and Pavelsky 2018), an area of ∼20,000 km² results. Since the river channel areas were derived for the whole basin and the SAR analysis applies only to the basin <500 asl, the lake area is probably under estimated. The HydroLAKES database (Messager et al., 2016) subsetted for the Amazon basin has a lake area of ∼23,000 km². This data set for lakes >0.1 km² used several data sources, including the SRTM waterbody data (Slater et al., 2006), and underwent manual removal of river and wetland polygons and other corrections.

Using side-looking airborne radar imagery acquired mostly during periods of low to moderate inundation by Projecto RadamBrasil (e.g., Departmento Nacional da Produção Mineral), Sippel et al. (1992) measured lake areas on the mainstem floodplain from 51^o to 70^o W and along the lower 400 km of the Japurá, Purus, Negro and Madeira rivers totaling ∼11,800 km².

Radar mosaics and maps generated from these images, both at 1:250,000 scale, allowed areas ≥∼0.05 km² to be estimated. Hamilton et al. (1992) found that the areas of lakes on the Amazon floodplains appear to follow the Pareto distribution, with censorship and truncation on both ends, which indicates that the lakes are statistically self-similar and that descriptive statistics for the lakes will vary with the scale of observation.

Seasonal and interannual variations in inundation lead to a range of lake areas and associated depths that are not represented by the available regional estimates. Results for well-studied lakes provide an indication of variations: Calado, 2–8 km² (Lesack and Melack 1995), Janauacá, 23–390 km² (Pinel et al., 2015), Curuai, 850–2,250 km² (Rudorff et al., 2014a); these values include some flooded vegetation at high stages. Limnological and ecological conditions in Amazon floodplain lakes are reviewed in Melack and Forsberg (2001), Melack et al. (2009), Melack et al. (2021), and Junk (1997).

The SAR-based analysis used by Hess et al. (2015) is well validated for detection of inundated forests. By combining the proportion of woody vegetation (forest, woodland and shrubs) compared to total inundated areas at the time of the high (70%) and low (62%) water levels analyzed by Hess et al. with the maximum (631,000 km²) and minimum (53,000 km²) inundated areas from Parrens et al. (2019), maximum (442,000 km²) and minimum (∼33,000 km²) flooded forest areas can be estimated. Seasonally flooded forests vary considerably in species diversity and composition, biomass and productivity depending on fertility of the sediments, duration of inundation and ecological factors not fully understood (Junk et al., 2010). Most flooded forests are associated with white-water rivers, and called várzea forests. Those along black-water rivers are called igapó forests, cover up to ∼84,000 km², and those near clear-water rivers that include várzea and igapó forests, are estimated to cover up to ∼50,000 km², based on the maximum flooded forest area calculated above and proportional areas from Melack and Hess (2010).

Seasonally inundated savannas with a variety of herbaceous plants and palms and other trees occur in Roraima (Brazil) and Llanos de Moxos (Bolivia) (Melack and Hess 2010; Junk et al., 2011). Interannual maxima, minima and mean inundation over an 8-year period in these two regions are provided by Hamilton et al. (2002): Roraima (16,500, 250 and 3,000 km²) and Llanos de Moxos (83,000, 6,100 and 34,000 km²). The Rupununi savannas, near Roraima are similar, and may reach a maximum area of ∼15,000 km² (Junk et al., 2011).

Other interfluvial wetlands occur in several regions within the Amazon basin and are not well studied (Junk et al., 2011). In the middle Negro basin ∼30,000 km² of wetlands with sedges, other aquatic plants, patches of palm, mainly Mauritia flexuosa, and open water occur; Frappart et al. (2005) estimated a similar inundated area for this region. A time series of SAR data was used to determine inundation duration, extent and vegetation in the Cuini (minimum, 784 km²; maximum, 964 km²), and Itu (minimum, 550 km²; maximum, 762 km²) wetlands, both located near the middle Negro River (Melack et al., 2009; Belger et al., 2011). Fleischmann et al. (2020) compared several inundation datasets for the Negro interfluvial wetlands associated with campinas and campinaranas (Rossetti et al., 2017), with maximum inundation extent reaching up to 20,000 km². The muted variation in area of these wetlands is likely owed to the limited hydrological connection to the rivers. Between the Purus and Madeira rivers wetland patches a few hectares to ∼150 km², totaling approximately 5,000 km² occur (Junk et al., 2011).

Herbaceous plants are abundant throughout the white-water floodplains of the Amazon basin. These plants grow profusely on sediments during low water and as rooted emergent and as free-floating plants, as inundation and water depths increase seasonally (Junk and Piedade 1997). Common C4 grasses are Echinochloa polystachya, Paspalum repens and Paspalum fasciculatum; other plants include the C3 grass, Hymenachne amplexicaulis, rice (Oryza perennis), and the genera Eichhornia, Ludwigia, Neptunia, Salvinia, and Pistia (Engle et al., 2008; Silva et al., 2013). SAR and other remotely sensed imagery can detect these plants, though their seasonal succession, growth and decline require time-series data. Also, narrow bands of floating plants fringing lakes and rivers and intermingled with woodlands can limit their accurate measurement. For the lowland basin, the combination of the proportion of inundated area represented by herbaceous vegetation in white-water river catchments and areas inundated permit an estimated area during high water of approximately 25,000 km² (Melack and Hess 2010; Hess et al., 2015). As an example of areal variations, in floodplain lakes totaling ∼9,400 km² along the eastern Amazon River, herbaceous plant coverage ranged from 770 to 2,900 km², based an analysis of Radarsat and MODIS images over 2 years (Silva et al., 2013). Herbaceous plants are rarely abundant in black-water river systems, and tend to be only moderately common in clear-water rivers (Junk et al., 2011).

Insufficient information is available to determine basin-wide variations and differences.

Among the floodplain and other wetlands delineated in sections 3.3 – 3.5, some are likely to be peatlands, though the classification of organic-rich soils in the region as peats is not standardized, and sampling is insufficient to determine their extent. Gumbricht et al. (2017) defined peat as a soil having at least 50% organic matter in the upper 0.3 m, while others have used different criteria, for example to identify minerotrophic and ombrotrophic sites (Lähteenoja et al., 2009). Among the possible peatlands based on the analysis by Gumbricht et al. (2017), the Pacaya-Samiria wetlands in the Peruvian lowlands, composed of seasonally flooded forests, palm swamps and peatlands, have had their inundated area and peat fairly well characterized (Lähteenoja et al., 2012; Jensen et al., 2018). By combing multi-sensor remote sensing with forest censuses and cores of peat thickness Draper et al. (2014) estimated that peatlands cover 35,600 ± 2,130 km² in the Pastaza-Marañon foreland basin (located in the Pacaya-Samiria region). A portion of this area is included within wetland areas described earlier, though during the period of the high water represented in Hess et al. (2015) fluvial flooding had receded in the Peruvian lowlands. Evidence for peat accumulation is also provided by sampling in parts of the Negro and Madre de Dios river basins, and suggested by model results (summarized in Gumbricht et al., 2017). However, the hybrid expert system used by Gumbricht et al. to estimate the regional distribution of wetlands and peatlands is not consistent with other estimates of inundation (Fleischmann et al., 2021), perhaps because of overestimation of soil moisture by the topographic index used or large rainfall in 2011, the year used. Their designation of floodplains and other wetlands as peatlands requires considerably more evidence, though recent studies are detecting peats scattered through the basin (e.g., Winton et al., 2021)

Hydroelectric reservoirs currently in the Amazon basin cover approximately 4,575 km² (Almeida et al., 2019; Kemenes et al., 2007 for Balbina). In Bolivia (50 km²), Ecuador (35 km²) and Peru (103 km²) almost all are above 1,000 m asl while in Brazil all are <500 m asl. The area of reservoirs (∼2,600 km²) estimated by Messager et al. (2016) is too low. Plans for more hydroelectric systems, especially in the Andes, could considerably expand their area (Almeida et al., 2019) and have substantial ecological consequences (Forsberg et al., 2017).

The majority of the studies in Amazonian aquatic environments, albeit sampled infrequently and usually in open waters of lakes or rivers, represent considerable effort given the immense area and remoteness of the basin. Few studies covered a complete hydrological cycle, and rarely included diel variations or multiple years. Related limnological, hydrological and meteorological measurements were seldom done concurrently, limiting the ability to evaluate the role of ecological factors, thermal structure and other processes on CH₄ fluxes and concentrations. Regions or habitats with few or no data include the Llanos de Moxos, coastal freshwater wetlands, riparian zones along streams, small reservoirs associated with agriculture, cultivated rice in Roraima and elsewhere, and habitats above 500 m. The recent findings of significant fluxes from trees, especially when inundated, included measurements in several types of forests (Pangala et al., 2017) and seasonal variations (Gauci et al., 2021), and point to the need for considerably more data on fluxes from trees given the large diversity of trees. Among hydroelectric reservoirs, measurements are available for a few (summarized in Melack et al., 2004; Guerin et al., 2006; Barros et al., 2011; Kemenes et al., 2016), but only Balbina reservoir has data collected from multiple reservoir and downstream stations as well as measurements of fluxes associated with turbines over a year (Kemenes et al., 2007, 2011). Abril et al. (2005) provide comparable, long-term data for Petit Saut reservoir in French Guyana.

Methane emissions based on floating chambers deployed in the Amazon River and major tributaries during low and high water, in most cases, reported by Sawakuchi et al. (2014) ranged from ∼0.2 to 297 mg CH₄ m⁻² d⁻¹ with averages (as mg CH₄ m⁻² d⁻¹): Amazon (21.6), Solimões (5), Negro (8.6), Madeira (0.6), Tapajós (38), Xingu (96), Preto (1.4) and Para (5). Barbosa et al. (2016) measured diffusive CH₄ fluxes using floating chambers and estimated fluxes based on the concentration of the gas in the water and calculated gas exchange coefficients along a 700-km transect including four stations in the Negro River and 21 of its tributaries at low and high water plus six stations on a 1100-km transect of the Solimões-Amazon River and one on the Maderia River occupied four times. Fluxes ranged from ∼0.2 mg CH₄ m⁻² d⁻¹ in the Solimões (during early falling water) to ∼3,900 mg CH₄ m⁻² d⁻¹ in the Jutaí River during low falling water with averages (as mg CH₄ m⁻² d⁻¹): Amazon and Solimões (18), Negro River and tributaries (54), Madeira (6.4), Purus (6.4), Juruá (5), Japurá (9.6). While ebullition may occur in rivers, as suggested by Sawakuchi et al. (2014), the indirect method used may be compromised by the wide ranges of gas exchange velocities expected in turbulent rivers.

Measurements in tributaries were made near their confluence with the Negro, Amazon or Solimões, and no upper reaches of rivers were sampled.

Additional data for the Solimões and Amazon rivers include those by Richey et al. (1988), who combined measurements of dissolved CH₄ concentrations in the main stem with estimates of air–water gas exchange to estimate a diffusive evasion of 3.2 mg CH₄ m⁻² d⁻¹, which is similar to 2.7 mg CH₄ m⁻² d⁻¹determined by Bartlett et al. (1990), but lower than measurements by Sawakuchi et al. (2014) and Barbosa et al. (2016). In the Uatumã River, downstream of Balbina dam, emission was 2,200 mg CH₄ m⁻² d⁻¹ (Kemenes et al., 2007), a value far above other rivers, with the exception of the Jutaí River. Interactions between the fringing floodplains and river channels are not well understood, though a gradient of increasing methane toward the margins of the Solimões River suggests inputs from the floodplains (Richey et al., 1988; Bartlett et al., 1990; Devol et al., 1994).

In a perennial first-order stream in the upper Xingu catchment, average fluxes based on floating chambers deployed monthly for a year were 108 ± 25 mg CH₄ m⁻² h⁻¹ or 2,600 ± 600 mg CH₄ m⁻² d⁻¹ (Neu et al., 2011). These very high fluxes could reflect input of groundwater high in CH₄, and may have over-estimated emissions because stationary chambers in streams can increase turbulent exchange. No other fluxes from streams are available. For comparison, average fluxes of 18 mg CH₄ m⁻² d⁻¹ summarized in Stanley et al. (2016) for tropical and subtropical streams are much lower. Clearly, many more measurements are needed.

Several studies have focused on open waters of lakes, though few included a complete year, adequately measured diffusive and ebullitive fluxes and rarely sampled diel variations. Over 2 years, nearly monthly measurements in an embayment and open water area of Lake Janauacá, located in várzea of the central basin, CH₄ fluxes were made with floating chambers connected to an off-axis integrated cavity output spectrometer and inverted funnels to capture bubbles (Barbosa et al., 2020; Barbosa et al., 2021). Diffusive fluxes, with measurements over diel periods combined for both sites, averaged ∼27 mg CH₄ m⁻² d⁻¹, and when combined with ebullition averaged 85 mg CH₄ m⁻² d⁻¹; diel variations were observed often with higher values during the day.

In várzea Lake Camaleão on Marchantaria Island in the Solimões River, Wassmann et al. (1992) deployed floating chambers connected to an automated system assaying CH₄ concentrations during a range of water levels, each with at least 3 days of sequential measurements, allowing bubble detection, and reported an average flux of 29 mg CH₄ m⁻² d⁻¹. Diel variations were not observed. As part of regular sampling over 18 months in two lakes on the floodplain of the Negro River and six on the floodplain of the Solimões River (Forsberg et al., 2017), Devol et al. (1990) reported average fluxes of 44 mg CH₄ m⁻² d⁻¹. Average diffusive fluxes during high water in várzea Lake Calado were 8.3 mg CH₄ m⁻² d⁻¹ (Crill et al., 1988), during rising water averaged 6.6 mg CH₄ m⁻² d⁻¹, increased up to 220 mg CH₄ m⁻² d⁻¹ during passage of a rare cold front, while during falling water averaged 54 mg CH₄ m⁻² d⁻¹; total flux averaged 163 mg CH₄ m⁻² d⁻¹ during falling water (Engle and Melack 2000).

As part of a transect along the Solimões and Amazon rivers, Barbosa et al. (2016) reported an overall average of 59 mg CH₄ m⁻² d⁻¹, ranging below detection to 298 mg CH₄ m⁻² d⁻¹ for 10 lakes, including white waters and black waters, sampled during four hydrological phases. Sawakuchi et al.’s (2014) few measurements of total fluxes from the eastern várzea Lake Curuai averaged 18 mg CH₄ m⁻² d⁻¹.

Methane fluxes within seasonally flooded forests occur from water surfaces, from the trunks of trees and from exposed soils during low water periods. In the flooded forests fringing Lake Janauacá, Barbosa et al. (2020) recorded an average of 110 mg CH₄ m⁻² d⁻¹, based on floating chamber and bubble trap measurements. Diffusive CH₄ fluxes within flooded forest measured by Wassmann et al. (1992) ranged from 1 to 12 mg CH₄ m⁻² d⁻¹, and ebullition averaged 69 mg CH₄ m⁻² d⁻¹. Fluxes at four water levels with floating chambers by Gauci et al. (2021) ranged as follows (expressed as mg CH₄ m⁻² d⁻¹) for plots along the Solimões (43–55 except for a high water value of 450), Negro (12–19) and Tapajós (36–55) rivers. Based on measurements with floating chambers in inundated igapó forests along the Jaú River Rosenqvist et al. (2002) calculated a mean annual emission of methane 30 mg CH₄ m⁻² d⁻¹.

Fluxes through trees in seasonally flooded forests can be high and quite variable when expressed per unit area of emitting surface. Based on sampling transects in twelve 0.4 ha forested plots along the Negro, Amazon, Madeira and Tapajós rivers during a period of rising water, Pangala et al. (2017) reported fluxes for mature and young trees per unit area of stem surface from 0.33 to 337 mg CH₄ m⁻² h⁻¹ and 0.39–581 mg CH₄ m⁻² h⁻¹, respectively. Similar measurements at four water levels in plots along the Solimões, Negro and Tapajós rivers by Gauci et al. (2021) ranged as follows (expressed as mg CH₄ m⁻² h⁻¹ per unit area of stem surface): Solimões (0.013–78.9), Negro (0.005–50.5) and Tapajós (-0.004–69.7). Soil CH₄ fluxes when the water table was below the surface in these plots were low and often negative, ranging from uptake of ∼1 mg CH₄ m⁻² d⁻¹ to evasion of ∼1 mg CH₄ m⁻² d⁻¹ (Gauci et al., 2021).

The Pantanal wetland can serve as a surrogate for the savanna wetlands in the Llanos de Moxos (Bolivia), which lack measurements of methane fluxes. Regular vertical profiles of methane and other gases obtained by aircraft from March 2017 to September 2019 in the Pantanal were combined with a planetary boundary layer budgeting technique to calculate a regional CH₄ flux of 50 mg CH₄ m⁻² d⁻¹ averaged over 1 year, assuming a planetary boundary layer-free troposphere exchange time of 3 days (Gloor et al., 2021). This flux integrates emissions from lakes, rivers and wetlands plus enteric fermentation by cattle and fires. Floating chambers deployed at five sites over a year in a Pantanal floodplain near the Miranda River yielded an average flux of 142 mg CH₄ m⁻² d⁻¹ (Marani and Alvala 2007).

Mean emission of monthly measurements with floating chambers in numerous wetlands near Boa Vista (Roraima) of about 14 mg CH₄ m⁻² d⁻¹ seem rather low compared to other similar Amazonian habitats (Jati 2014), perhaps because ebullition was not captured. Emissions from cultivated rice in Roraima are not available. In the mid-Negro basin, Belger et al. (2011) measured methane uptake on unflooded lands, evasion from flooded areas as diffusive and ebullitive fluxes with chambers and funnels, and as transport through rooted plants. Emission from wetland areas averaged 77 mg CH₄ m⁻² d⁻¹.

As water levels rise on riverine floodplains, herbaceous plants form floating mats. Several studies have deployed floating chambers to measure methane fluxes from the water surface or, rarely, from plants on the surface. In mats of floating herbaceous plants, Barbosa et al. (2020) measured an average diffusive CH₄ flux of 53 mg CH₄ m⁻² d⁻¹ and estimated an average ebullition of 97 mg CH₄ m⁻² d⁻¹, totaling 150 mg CH₄ m⁻² d⁻¹. Diffusive fluxes within similar mats reported by Bartlett et al. (1988; 1990) were similar, averaging 42 and 44 mg CH₄ m⁻² d⁻¹, respectively. Wassmann et al. (1992) reported diffusive fluxes from ∼2 to 28 mg CH₄ m⁻² d⁻¹, average ebullition of 23 mg CH₄ m⁻² d⁻¹ and did not detect plant-mediated transport via floating mats of Paspalum repens.

Fluxes in palm-dominated peatlands and nearby habitats in the western Amazon basin are variable and can be high. Eddy covariance measurements in a natural palm (Mauritia flexuosa) peatland near Iquitos (Peru) over a 2-year period averaged 22 g C m⁻² y⁻¹ (= ∼80 mg CH₄ m⁻² d⁻¹) (Griffis et al., 2020). Teh et al. (2017) measured fluxes in the Pastaza–Marañón basin in forests, a Mauritia flexuosa-dominated palm swamp, and a mixed palm swamp during wet and dry seasons in four campaigns. Among all data, diffusive CH₄ emissions averaged ∼48 ± 4 mg CH₄ m⁻² d⁻¹; fluxes in the M. flexuosa palm swamp averaged ∼49 ± 5 mg CH₄ m⁻² d⁻¹ (assuming their notation, CH₄–C, means the mass of C in the CH₄). As noted by the authors, their estimates of ebullition from short deployments of static chambers are probably not representative. In the same region, del Aguila-Pasquel (2017) sampled 8 times over 8 months spanning wet and dry seasons in a palm swamp and reported a mean flux of 73 ± 5.4 mg CH₄ m⁻² d⁻¹ (n = 129). In peatlands in the Madre de Dios River basin (Peru) Winton et al. (2017) found that open canopy Cyperacea-dominated areas emitted 4.7 ± 0.9 mg CH₄ m⁻² h⁻¹ (=113 ± 22 mg CH₄ m⁻² d⁻¹), and Mauritia flexuosa palm-dominated areas emitted 14.0 ± 2.4 mg CH₄ m⁻² h⁻¹ (= 336 ± 58 mg CH₄ m⁻² d⁻¹) during a short period in 1 month. That CH₄ can be emitted through M. flexuosa trunks suggests that emissions from palm-dominated peatlands based on soil flux chambers underestimate fluxes (van Haren et al., 2021). CH₄ fluxes were significantly correlated to pneumatophore density in M. flexuosa stands (van Lent et al., 2019).

Several hydroelectric reservoirs have measurements available while most, usually smaller ones, do not. In Balbina reservoir, measurements of diffusive and ebullitive fluxes from multiple stations within the reservoir (average 63 mg CH₄ m⁻² d⁻¹), degassing at the turbines and downstream were made over a year, when combined produce an annual CH₄ emission of 97 Gg for the whole system, excluding methane oxidation in the river (Kemenes et al., 2007). Additional measurements at Samuel and Curua-Una reservoirs indicated the significance of degassing at the turbines and downstream (Kemenes et al., 2016). Based on measurements using drifting chambers during four periods in the first 2 years after filling of the Belo Monte hydroelectric system, Bertassoli et al. (2021) reported averages of 104 and 283 mg CH₄ m⁻² d⁻¹, and whole systems annual totals of 20 and 50 Gg CH₄. About half the flux was attributed to ebullition, and degassing in turbines was deemed minor.

Paranaíba et al. (2021) combined measurements of diffusive methane fluxes and calculations of gas exchange velocities with floating chambers at several sites and concentrations of methane sampled and analyzed on continuous transects through the Curuá-Una reservoir. During the rainy season with rising water levels, they reported average fluxes of 9.6 mg CH₄ m⁻² d⁻¹ (range from 1.4 to 112) and during the drier season with falling water levels average fluxes of 14.4 mg CH₄ m⁻² d⁻¹ (range from 1.3 to 69). As part of a study of organic carbon burial in Curuá-Una reservoir, Quandra et al. (2021) found methane dissolved in pore water to be above saturation in about one quarter of their measurements, indicating a potential for ebullition. Porewater concentrations were similar during periods with rising and falling water, varied spatially, but were not related to C:N ratios or organic carbon burial rates.

Emission from other Brazilian reservoirs based on overall average diffusive and ebullitive emissions from the surfaces of ten reservoirs within southern portions of the basin, as summarized in Deemer et al. (2016) is ∼80 mg CH₄ m⁻²d⁻¹; this value does not include degassing through turbines or below the dam. Estimating the emissions from the reservoirs in Bolivia, Ecuador and Peru is more difficult because no measurements exist and temperatures will be less at higher elevations and the watersheds differ from conditions in Brazil; a value of half that from Deemer et al. (2016) is suggested. The extent that the reservoir emissions represent net emissions, i.e., emissions additional to those associated with the undammed rivers, are uncertain, with estimates at Belo Monte (Bertassoli et al., 2021) and Balbina (Kemenes et al., 2011); upland forest soils, before being inundated, are likely to be sinks for methane.

The seasonally flooded ecotone, called the aquatic-terrestrial transition zone, is a varying mixture of bare soil and cover by herbaceous and woody plants exposed during periods of low water. Areas occupied by the aquatic-terrestrial transition zone are especially large in regions with shallow seasonal flooding, such as in savannas. Few measurements of methane flux or related processes are available for these periods in the Amazon (Pangala et al., 2017; Gauci et al., 2021); those of Smith et al. (2000) for the Orinoco floodplain are relevant. Microbial activity in response to desiccation (Conrad et al., 2014) and CH₄ oxidation in exposed sediments on Amazon floodplains (Koschorreck 2000) have also been examined.

Sediments exposed as reservoir levels decline are analogous to the aquatic-terrestrial transition zone. Experimental measurements of methane emissions from sediment cores from a tropical Brazilian reservoir exposed to drying and rewetting indicated enhanced emissions from sediments with overlying water removed and when rewetted compared to sediments that had overlying water (Kosten et al., 2018). A comparative study for several types of aquatic systems including ones in tropical climates by Paranaíba et al. (2022) found emissions from portions of inland waters exposed to the atmosphere as water levels decline were consistently higher than emissions in nearby uphill soils. Statistical analyses revealed that methane emissions were negatively related to organic matter content and positively related to moisture of the sediments.

Riparian zones along streams often have soils with high water content but without standing water and occur throughout the Amazon basin. Topographic features are illustrated in Nobre et al. (2011). These environments are likely sources of methane emission. For example, soils in upland forests can release CH₄, depending on soil moisture levels (Sihi et al., 2021).

The biogeochemical and microbial processes involved in the production and consumption of methane are known though uncertainties remain (e.g., Segers 1998; Borrel et al., 2011; Bridgham et al., 2013; Schlesinger and Bernhardt 2013). However, the quantitative importance of these processes varies among habitats and regions, and needs further study in aquatic environments within the Amazon basin. The availability to methanogens of the varied organic compounds present in Amazonian waters has not been characterized, though studies of the use of these substances for metabolism, in general, are available (e.g., Waichman 1996; Mayorga et al., 2005; Amaral et al., 2013; Ward et al., 2013; Vihermaa et al., 2014). Using samples from sediment cores obtained in an Amazon reservoir and two other tropical reservoirs in Brazil and incubated over about 2 years, Isidorova et al. (2019) found that rates of CH₄ production had a strong statistical relation to sediment total nitrogen content and age of the sediment. Given the variations of sediment characteristics in Amazon floodplains and reservoirs (Hedges et al., 1986; Martinelli et al., 2003; Smith et al., 2003; Guyot et al., 2007; Moreira-Turcq et al., 2013; Cardoso et al., 2014; Sobrinho et al., 2016), further studies will likely reveal different methane production rates associated with these variations.

CH₄ oxidation has been measured in a tropical reservoir (Guérin and Abril 2007) and inferred to occur in floodplain lakes (Crill et al., 1988; Engle and Melack 2000) or exposed sediments (Koschorreck 2000). Based on stable isotopic mass balances of CH₄, Sawakuchi et al. (2016) estimated substantial rates of CH₄ oxidation in large Amazonian rivers, and found that genetic markers for methane oxidizing bacteria were positively correlated with CH₄ oxidation. Using incubations and measurements of δ ¹³C-CH₄ in a várzea lake, Barbosa et al. (2018) found that a large fraction of dissolved CH₄ was oxidized with volumetric CH₄ oxidation rates ranging from ∼1 to 175 mg CH₄ m⁻³ d⁻¹. Heavier values of δ ¹³C-CH₄ in surface waters when compared to bottom waters and sediment bubbles corroborate these high rates. They also found that CH₄ oxidation had a positive relation with CH₄ concentration and the presence of dissolved oxygen.

The microbial assembles in floodplain lakes (e.g., Melo et al., 2019) including methanogens and methanotrophs have received limited investigation in the Amazon (e.g., Finn et al., 2020; Bento et al., 2021; Gontijo et al., 2021). Conrad et al. (2010, 2011) and Ji et al. (2016) examined microbial communities and measured rates of methanogenesis in sediments by incubating sediment slurries from different floodplain lakes; CH₄ production was found to be mainly hydrogenotrophic based on isotopic fractionation. However, the congruence of laboratory incubations of sediment slurries with CH₄ production in intact sediments is unclear. More study is needed relating microbial activity to environmental conditions, such as the examination of responses to desiccation by Conrad et al. (2014). Anaerobic oxidation of methane and microbial methane production in oxygenated water also need attention, as indicated by studies elsewhere (e.g., Caldwell et al., 2008; Grossart et al., 2011; Roland et al., 2016). For example, Gabriel et al. (2020) used slurries of flooded Amazon forest soils to demonstrate the potential for anaerobic methane oxidation by Fe(III) reduction.

In the warm waters of the Amazon basin, high latent heat fluxes lead to convective mixing while diurnal heating under strong insolation leads to periods of stable stratification (e.g., Augusto-Silva et al., 2019). Exchange of methane between surficial water and overlying atmosphere depends on the concentration gradient between air and water and on physical processes at the interface, usually parameterized as a gas transfer velocity (k). Gas transfer velocities are influenced by atmospheric stability and, in water, are altered by currents, wind and convection, as well as rain, temperature and organic surficial films. Melack (2016) summarized estimates of k available for rivers and lakes in the Amazon basin. Results reported by Ulseth et al. (2019) indicate quite large k values can occur in high-energy montane streams due to bubble entrainment. Recent studies by MacIntyre et al. (2019) and MacIntyre et al. (2021) have used near-surface measurements of dissipation rates of turbulent kinetic energy and hydrodynamic theory to calculate k under low wind conditions in open water and within flooded forests.

Though gas transfer velocity can be parameterized based on wind speed (Wanninkhof 1992; Cole and Caraco 1998), the dependence of k on the dissipation rate of turbulent kinetic energy (ϵ) has theoretical and empirical evidence. The surface renewal model, k = c ₁ (ϵν)^1/4) Sc⁻ⁿ has an explicit dependence on dissipation rates, where dissipation rates have units of m² s⁻³, ν is kinematic viscosity (m s⁻²), Sc is the Schmidt number, n is −1/2 or −2/3, and c ₁ is an empirical coefficient. Dissipation rates depend on the shear just below the air-water interface and are augmented if cooling or heating are appreciable relative to shear (MacIntyre et al., 2021). This model has been applied to Amazon flooded forests (MacIntyre et al., 2019) and reservoirs (MacIntyre et al., 2021) and confirmed with indirect estimates of k using results of experiments with floating chambers (Amaral et al., 2020).

Stratification within tropical lakes and reservoirs can be appreciable as a result of intense heating when winds are light (e.g., Augusto-Silva et al., 2019). In that case, near-surface values of dissipation rate can be higher than observed under cooling. With stratification retarding the downward mixing of heat, much of the turbulence produced by wind is dissipated. Values of k computed using the surface renewal model during heating averaged 10 cm h⁻¹ but reached 18 cm h⁻¹ for winds up to 4 m s⁻¹, were independent of wind speed, and increased with heating (MacIntyre et al., 2021). Hence, the fluxes estimated from wind-based models for many of the lakes in the Amazon basin are likely underestimated.

CH₄ ebullition, a major mechanism for evasion to the atmosphere, is regulated by the production and accumulation of CH₄ in sediments and processes leading to the release of bubbles from the sediments. Release of bubbles can be influenced by variations in hydrostatic pressure caused by a drop in atmospheric pressure or drop in water level. Other factors include currents, waves, shear-stress at the sediment-water interface and possibly sediment disturbances by benthic organisms (Barbosa et al., 2021).

Vertical and horizontal water movements connect littoral, pelagic and benthic regions of lakes and wetlands (MacIntyre and Melack 1995) and are likely to influence CH₄ concentrations and fluxes. One-dimensional (e.g., DYRESM, Yeates and Imberger 2003) and three-dimensional (e.g., AEM3D, Hodges and Dallimore 2019) hydrodynamic models include processes induced by surface heat fluxes, wind, inflows and outflows. One-dimensional models characterize a lake as a single water column in the vertical and are computationally efficient permitting long-term and regional applications. Three-dimensional models divide a lake into grids in three directions, resolve spatial variability of thermal structure, internal waves, horizontal exchanges and calculate dissipation rates of turbulent kinetic energy, a key term in the surface renewal model of gas exchange. An example of an application of a three-dimensional model to a floodplain lake illustrates spatial differences in diel cycles of stratification and mixing (Figures 4A–D), and circulations driven by wind-induced basin-scale internal waves when the near-surface water is stratified (Figures 4E,F). Three-dimensional hydrodynamic modeling has demonstrated that diel differences in water temperature between floating plant mats and open water as well as basin-scale motions can cause lateral exchanges linking vegetated habitats to open water (Amaral et al., 2021). Higher CH₄ concentrations in herbaceous plant mats than in open water suggest that vegetated habitats can be sources of CH₄ to other regions (Bartlett et al., 1988; Barbosa et al., 2020).

In a summary of results from the Wetland and Wetland CH₄ Inter-comparison of Models Project (WETCHIMP), Melton et al. (2013) stated that the models disagreed in their simulations of wetland areal extent and methane emissions, in both space and time, and had parameter and structural uncertainty, and noted that lakes and rivers were not included. Moreover, mechanistic models of methane production and evasion appropriate for tropical floodplains are not available (Riley et al., 2011), though relevant conceptual models have been proposed (Cao et al., 1996; Potter 1997; Potter et al., 2014). While several models have potentially useful components or formulations (e.g., Walter and Heimann 2000; Tang et al., 2010; Bloom et al., 2012; Wania et al., 2013; Ringeval et al., 2014; Tan et al., 2015; Lu et al., 2016), no spatially explicit model exists that incorporates the inundation dynamics, ecological characteristics and limnological conditions of tropical floodplains and wetlands. Of special importance is inclusion of plant functional groups common in these habitats combined with appropriate algorithms to estimate their productivity, as these plants supply most of the organic carbon subsequently released as methane (Melack and Engle 2009; Melack et al., 2009). Algorithms are required that simulate stratification and mixing of the water column with concomitant influence on the extent of anoxia and on air-water gas exchange.

Current regional biogeochemical models of methane emissions from wetlands calculate grid-averaged methane fluxes based on soil temperature and carbon availability or heterotrophic respiration. Examples of these models include the Joint United Kingdom Land Environment Simulator (JULES) (Clark et al., 2011; McNorton et al., 2016), the Lund-Potsdam-Jena model (LPJ-WSL; Sitch et al., 2003; Zhang et al., 2016), a CH₄ biogeochemistry model based on the Integrated Biosphere Simulator (TRIPLEX-GHG; Zhu et al., 2014), the LPX-Bern model (Ringeval et al., 2014), a revision of TEM-MDM (Liu et al., 2020) and WetCHARTs (Bloom et al., 2017), JPL-WHyMe (Wania et al., 2010), the Dynamic Land Ecosystem Model (DLEM, Zhang et al., 2017), and CLM4Me (Riley et al., 2011; Meng et al., 2012). A methane model for Amazon peatlands is under development (Yuan et al., 2020). One problem with these models is their use of molecular diffusion through soil layers that is not appropriate for the turbulence-based gas transfer needed for conditions with surface inundation. Inundation is usually simulated as overland hillslope flows and an approximated terrain model, and does not represent the large seasonal inundation variations and hydrologic fluxes present in the Amazon basin. Climatic inputs are typically obtained from reanalysis products with insufficient temporal resolutions to force diel processes. The plant functional groups included do not represent the aquatic plants common in the Amazon basin. Furthermore, these models do not explicitly include biogeochemical processes, such as methanogenesis and methane oxidation, or physical processes, such as mixing through the water, lateral exchanges, ebullition, or air-water exchange via turbulent mixing. Also, they do not simulate dissolved oxygen concentrations and thus the impact of dissolved oxygen on the methane oxidation or prescribe bulk dissolved oxygen concentrations to the grids (Wania et al., 2010; Riley et al., 2011).

Results from an atmospheric transport model and ATTO mixing ratios (Botia et al., 2020) were compared to six versions of the extended ensemble available in WetCHARTs v1.3.1 (Bloom et al., 2017) to account for variability in wetland extent and the temperature-dependent CH₄ respiration fraction (Q₁₀) (Figure 5). All WetCHARTs’ simulations used precipitation from ERA5 (C3S, 2017) to drive the temporal variability of methane emissions and the CARDAMOM model for heterotrophic respiration, but the wetland spatial extent was varied. The simulations ending in three used the Global Lakes and Water Database (Lehner and Döll 2004) and those ending in four used the sum of all wetland and freshwater land cover types in GLOBCOVER (Bontemps et al., 2011). Each pair of simulations had a different temperature dependence of the CH₄ respiration fraction (Q₁₀), according to the second digit of each simulation (Q₁₀ = 1,2,3). The variability in the WetCHARTs simulations indicates that when using a different wetland extent but the same temperature dependence of the CH₄ respiration fraction (Q₁₀), the effect on the integrated CH₄ signal can be about 10 ppb (i.e., lines 913 and 914). This effect seems to be larger at lower Q₁₀ values. When using a different Q₁₀ and the same wetland extent (e.g., line 914 with 924 or 934), the differences are larger when comparing the Q₁₀ = 1 with the Q₁₀ = 2 simulations than between the Q₁₀ = 2 and Q₁₀ = 3 simulations. Furthermore, these comparisons suggest that WetCHARTs fluxes for the ATTO footprint are too high, and that the seasonality of WetCHART-derived mixing ratios at ATTO are 1 month out of phase when compared to the ATTO observations. When accounting for transport errors (not shown here), the simulated wetland signal at ATTO is still higher than the observations (Botia et al., 2020).

Potter et al. (2014) built a model for Amazon floodplain lakes based on the supply of organic carbon, as a key factor determining methane production. The LAKE model (Stepanenko et al., 2016) includes most major processes operative in lakes, but the methane module has only been tested on a small boreal lake. Zimmermann et al. (2021) used measurements and a one-dimensional physical model of a small temperate lake to examine how seasonal or more frequent thermocline deepening influenced methane emission and consumption by methanotrophs. Though conceptually relevant, this model would require revision because diel cycles of stratification and mixing are common in shallow Amazon lakes.

The Arctic Lake Biogeochemical Model (ALBM) is a one-dimensional process-based, biogeochemical model that simulates the thermal and methane dynamics of lakes (Tan et al., 2015). The model has been applied to arctic and boreal lakes on seasonal time scales (Tan et al., 2015; Guo et al., 2020), but the thermal module and several other aspects need revisions for conditions in shallow, warm waters. In section 6.2 we discuss alternative models of the relevant physical processes. By combining sensitivity analyses and calibration and validation with appropriate data, the models have been shown to perform reasonably well for specific lakes. However, to apply these models on a regional scale is more difficult as success depends on the calibration and validation sites being representative of the region, and on the availability of data for lakes in the region. An application of ALBM to the Amazon basin produced reasonable results for open water of lakes that further modifications are likely to improve (Figure 6). The parameters used for the Amazon basin were calibrated with data from two well-studied floodplain lakes (Janauacá and Calado) with relevant references cited above.

Several biogeochemical and physical processes need to be incorporated or improved in mechanistic methane models for floodplain lakes. New formulations of gas transfer velocity and inclusion of lateral exchanges and diel stratification and mixing, as described in section 6, are recommended. Developing parameterizations for fluxes through trees and herbaceous plants, oxic methane production, anaerobic methane oxidation and supply of carbon by autotrophic growth and hydrological inputs are challenging but important. To improve modeling of ebullition, inclusion of changes in hydrostatic pressure and from disturbances of the sediments may help.

Statistical models offer another approach for estimating lake and reservoir emissions, albeit with limitations related to data availability

DelSontro et al. (2016) analyzed the relationship between emission rates and the predictor variables of lake size, chlorophyll a, total phosphorus and total nitrogen using simple and multiple linear regression for lake and reservoirs distributed around the world. Though tropical reservoirs were included, Amazon floodplain lakes are not represented in this analysis. Among the variables evaluated only chlorophyll a had a positive effect on total methane emission. For comparison, a statistical model applied to Lake Janauacá found variables related to CH₄ production (temperature, dissolved organic carbon) and consumption (dissolved nitrogen, oxygenated water column), as important to dissolved CH₄ concentrations (Barbosa et al., 2020).

Using eddy covariance data from wetland sites largely in the north temperate zone, Knox et al. (2021) applied several statistical techniques to examine the importance of a variety of physical and biological predictors of the timing and magnitude of CH₄ fluxes. While quite informative with regard to the wetlands included, the only site in tropical South America is in the Pantanal, a seasonal savanna wetland. Though the Pantanal shares similarities to the Moxos wetlands, the majority of floodplain and other wetlands in the Amazon basin are not represented in the analyses.

While it is well known that all models have biases from limitations and assumptions within the models and as a result of the data used in calibration and validation, given the logistic and scientific challenges with measurements in the vast and complex Amazon basin, models of all types can surely contribute to understanding and projections of methane emissions.

Estimates of methane fluxes at different spatial scales illustrate varied approaches and limitations. Early estimates of regional methane fluxes made for the mainstem Solimões-Amazon floodplain and the whole basin were limited by lack of data on areas of inundation and aquatic habitats and few measurements of fluxes (e.g., Bartlett et al., 1988; Devol et al., 1990). The first estimates that incorporated remote sensing of inundation and habitat areas were done by Melack et al. (2004) for the mainstem Solimões-Amazon floodplain and a 1.77 million km² area in the central basin (Hess et al., 2003). They combined habitat-specific methane fluxes with seasonal changes in the area of aquatic habitats, derived from active and passive microwave remote sensing, and applied Monte Carlo error propagation to establish uncertainties. By assuming similar fluxes by habitat and adding approximations for particular regions (e.g., Moxos), an extrapolation to whole lowland basin was made. For the central basin quadrant, annual emission was estimated as 9 ± 1.7 Tg CH₄, and if extrapolated to the basin below 500 m asl, resulted in approximately 29 Tg CH₄ y⁻¹. The largest omission from these estimates is evasion from seasonally flooded forests. The measurements of emissions for trees, extrapolated to the Amazon basin, by Pangala et al. (2017; 15.1 ± 1.8 to 21.2 ± 2.5 Tg CH₄ y⁻¹) and Gauci et al. (2021; 12.7 to 21.1 Tg CH₄ y⁻¹ for inundated trees plus 2.2 to 3.6 Tg CH₄ y⁻¹ for riparian zones with water tables below the surface) substantially increase regional fluxes. However, considerable uncertainty in these estimates stems from the lack of adequate data on the abundance and emitting surface areas of the trees and the high variability in fluxes per unit area of trees.

As part of the Science Panel for the Amazon (SPA) report, released at COP26 (Gatti L. et al., 2021; Mahli et al., 2021), J. Melack, L. Basso and S. Pangala provided bottom-up estimates of methane emission from aquatic environments in the combined hydrological Amazon and Tocantins basins, and for the Amazon forest biome. A total area of aquatic habitats for these two basins of 970,500 km² was used to generate their bottom-up estimate of approximately 51 Tg CH₄ y⁻¹. However, this estimate does not include seasonal and interannual variations or a rigorous uncertainty analysis. A second estimate of methane emissions was done for the Amazon biogeographic province that encompasses tropical forests in much of Amapá, French Guiana, Guyana, Suriname, and southern Venezuela and eastern Colombia that drains into the Orinoco River (Albert et al., 2021). This estimate had added uncertainty because of the lack of data for most of the northeastern areas outside of the hydrological Amazon basin.

Improvements needed in bottom-up estimates are inclusion of seasonal variations in flooding and aquatic habitats, as noted in sections 2,3, and more measurements of all fluxes, summarized in section 4. An example of doing so is provided by Barbosa et al. (2021) for a floodplain lake with fringing inundated forests and herbaceous plants. They combined seasonal variations in ebullitive, diffusive and tree-mediated fluxes with variations in water depths and areal coverage of aquatic habitats, i.e., weighting fluxes by seasonal variations in habitat areas and habitat-specific fluxes. This approach was especially important for ebullition for which the fraction of total emission from water surfaces varied from 1% (rising and high water) to 89% (low water) in open water of the lake, and from 73% (rising water) to 93% (falling water) in flooded forests.

An example of the challenges and compromises in regionalization is offered for a reach of the mainstem Solimões-Amazon rivers and associated floodplains (Table 1). The values in Table 1 provide a sense of the relative importance of different habitats as sources of methane to the atmosphere based on areal fluxes and habitat areas for low and high water conditions. However, the fluxes should be viewed with caution because of the uncertainties associated with the data and calculations. As summarized in section 5, several studies provide measurements of methane fluxes from water surfaces in river channels and floodplain habitats. Those selected here were judged representative and offering the most complete data. To fully use data from all studies is not possible because actual datasets are not available in most publications. The fluxes from trees are especially difficult to extrapolate because of the large variance in per tree fluxes and the very limited measurements of the surface areas of trees. Fluxes from floodplain soils during low water periods are not included.

In an analysis for the reach of the mainstem Solimões-Amazon floodplain from 54^o to 70^o W, annual fluxes of 1.7 ± 0.4 Tg CH₄ were calculated with 67% of the uncertainty attributed to habitat-specific fluxes and 33% owed to uncertainties in areal estimates of inundation and vegetative cover (Melack et al., 2004). To do so, the methane emission rate per unit area for the mainstem Solimões-Amazon floodplain, representing a mixture of habitats, and using a mean annual flooded area of 42,700 km² (determined as monthly means from Sippel et al. (1998) summed over 12 months and divided by 12) was estimated as ∼40 Mg CH₄ km⁻² y⁻¹.

Several aircraft campaigns provide methane flux estimates for different regions and periods. Beck et al. (2012) estimated fluxes for the lowland Amazon during November as 36 ± 12 mg CH₄ m⁻² d⁻¹ and during May as 43 ± 19 mg CH₄ m⁻² d⁻¹. Miller et al. (2007) collected vertical profiles of methane over 4 years at sites near Santarém and Manaus and calculated average emissions of ∼27 mg CH₄ m⁻² d⁻¹. Wilson et al. (2016) applied the TOMCAT model using aircraft vertical profiles and estimated methane emissions of 36.5 to 41.1 Tg CH₄ y⁻¹in 2010 and 31.6 to 38.8 Tg CH₄ y⁻¹ in 2011 for an area of 5.8 × 10⁶ km⁻²; non-combustion emissions represented 92–98% of total emissions. Pangala et al. (2017) estimated emissions of 42.7 ± 5.6 Tg CH₄ y⁻¹ for an area of 6.77 × 10⁶ km² based on vertical profiles from 2010 to 2013; 10% of emission was attributed to biomass burning.

The most comprehensive record is provided by Basso et al. (2021), who used lower-troposphere vertical profiles of methane concentration from four sites in the Brazilian Amazon basin between 2010 and 2018 and a column budget technique to calculate emissions of 46.2 ± 10.3 Tg CH₄ y⁻¹ with no temporal trend. This finding of no temporal trend during a period with exceptionally high and low river stages and varying inundation areas suggests further examination of how methane fluxes could have varied through these years.

Basso et al. (2021) used the methodology of Gatti L. V. et al. (2021) to estimate fluxes for three subregions and combined these as a weighted flux scaled to an area of 7.25 million km², a region that includes parts of Venezuela, and all of Suriname, French Guyana and Guiana plus coastal watersheds in Brazil and upper Tocantins basin. The large region to the west of their sampling locations near Tefé and Tabatinga was not sampled and has some environments not represented by the regions to the east, such as the Peruvian peatlands and Andean highlands. Based on concurrent measurements of carbon monoxide, 17% of the sources were attributed biomass burning and 83% mainly to wetlands. Causes for seasonality in wetland fluxes and for the larger signals in the northeast are not fully understood, indicating that more studies are needed.

Based on SCIAMACHY data, Bergamaschi et al. (2009) calculated total Amazon emissions of 47.5 to 53.0 Tg CH₄ y⁻¹ in 2004 for an area of 8.6 × 10⁶ km². Based on an inversion model, Fraser et al. (2014) estimated an emission of 59.0 ± 3.1 Tg CH₄ y⁻¹ from tropical South America (approximately ∼9.7 × 10⁶ km²) in 2010. Tunnicliffe et al. (2020) using inverse modelling estimates and GOSAT measurements reported mean emissions for the Brazilian Amazon wetlands (and not clearly defined) lower than other estimates (9.2 ± 1.8 Tg CH₄ y⁻¹). Wilson et al. (2021) using GOSAT data and an inverse model estimated total emissions from natural, agricultural and biomass burning sources of 38.2 ± 5.3 (between 2010 and 2013) and 45.6 ± 5.2 Tg CH₄ y⁻¹ (between 2014 and 2017) for the Amazon basin (approximately ∼6.0 × 10⁶ km²). Given the different regions represented and the mixture of sources, it is difficult to compare these estimates with each other or with the aircraft or bottom-up values.

Methane emissions for the Amazon basin, extracted from the WetCharts ensemble of wetland models (Bloom et al. (2017) at monthly resolution for 2009 and 2010 by Basso et al. (2021), are 39.4 ± 10.3 Tg CH₄ y⁻¹, and results are similar to emissions derived from aircraft samples except for the northeastern portion of the basin. Gedney et al. (2019) provided estimates of methane flux for the Amazon basin using the JULES model that varied from ∼24 to ∼58 Tg CH₄ y⁻¹, a rather wide range that stems from problems with inundation estimates and parameterization of methane fluxes. As part of an application of ALBM to global lakes (Guo et al., 2020), an estimate for only lakes in the Amazon and Tocantins basins, based a static lake area from Messager et al. (2016) and calibration with data from lakes Janauacá and Calado, yielded estimates of ∼0.7 Tg CH₄ y⁻¹ and ∼0.1 Tg CH₄ y⁻¹, respectively. This estimate is similar to the estimate of ∼0.7 Tg CH₄ y⁻¹ in Mahli et al. (2021) for the same regions based on extrapolation of bottom-up measurements of fluxes and similar lakes areas.