Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

The processing of carbon by aquatic ecosystems of inland waters is now recognised as a significant component of regional and global carbon dynamics. In particular, the high rates of sedimentation in lakes and reservoirs and considerable evasion of carbon dioxide and methane from many rivers, lakes, and wetlands lead to fluxes disproportionately large relative to the area of inland waters (Cole et al. 2007; Downing 2009; Battin et al. 2009; Aufdenkampe et al. 2011; Butman and Raymond 2011; Raymond et al. 2013; Stanley et al. 2015). Although the magnitude and variability of these fluxes remain uncertain, especially in tropical regions, recent studies are improving our understanding of carbon dynamics in the streams, rivers, lakes, reservoirs, and wetlands of the Amazon basin. The Large-scale Biosphere Atmosphere Experiment in Amazonia (LBA) and related activities have resulted in numerous relevant publications, many of which have been summarised in a recent monograph (Gash et al. 2009). In particular, Richey et al. (2009) described carbon processing from streams and rivers, and Melack et al. (2009) examined ecosystem processes in inundated areas. From these studies, it appears that evasion of carbon dioxide from Amazonian rivers, lakes, and temporally inundated aquatic habitats is of similar magnitude to net ecosystem exchanges in non-inundated upland forests (terra firme forests) derived from eddy covariance measurements. In addition, Amazonian aquatic ecosystems account for a significant proportion of the global methane flux from natural wetlands.

With a focus on exchanges of carbon dioxide and methane between inland waters and the atmosphere, the purpose of this chapter is to extend the regional extent of previous analyses to the whole lowland Amazon basin (regions <500 m above sea level; 5.05 million km2; Fig. 7.1; not including the Tocantins basin) by synthesis and critical evaluation of recent publications. Remote sensing analyses of inundation and wetland habitats, inundation modelling, measurements in rivers, reservoirs and other types of wetlands, and improved estimates of gas exchange coefficients for standing and flowing water are included. Regional estimates of gas fluxes are complemented with consideration of sources of the carbon being released, and remaining uncertainties and research needs are discussed. A discussion of climate trends and variability, exceptional events, and human impacts, as factors potentially altering carbon dynamics in the Amazon basin, is presented.

Fig. 7.1
figure 1

Lowland Amazon basin (area ≤500 m; grey) with floodable areas (white), as described in Melack and Hess (2010) and Hess et al. (2003). Small areas of wetlands on the northern and southern edges are not shown because remote sensing data used to develop the wetland distribution were not available there. Ba Balbina Reservoir; Ca Lake Calado; Cu Lake Curuai; Cn Lake Canaçari; Ja Jau River; JiP Ji-Paraná River; Ju Juruena watershed; Moxos Llanos de Moxos; Ro Roraima; UN upper Negro interfluvial wetlands

Full size image

Several current projects are in the process of publishing their results, and will continue to contribute to our understanding of carbon dynamics in the Amazon, though only a portion is available for inclusion here. These projects include HYBAM (http://www.ore-hybam.org/index.php/eng/documents), CARBAMA (http://carbama.epoc.u-bordeaux1.fr/PUBLICATIONS.html), the ‘Rede Beija Rio’ network (http://boto.ocean.washington.edu/story/Amazon), A Biogeoquímica do Carbono e Mercúrio na Bacia Amazônica (Barbosa 2013) (http://www.biologia.ufrj.br/limnologia/projeto-carbono.php), and GEOMA (http://www.geoma.lncc.br/).

2 Inundation and the Variable Extent of Aquatic Habitats: Remote Sensing and Modelling

Carbon dynamics and carbon dioxide and methane exchanges in the Amazon basin vary among the diverse aquatic habitats. Junk et al. (2011) used information on climate, hydrology, water and sediment chemistry, and botany to delineate 14 major types of naturally occurring wetlands in the lowland Amazon. The amplitude, duration, frequency, and predictability of inundation are key criteria in this classification. While the classification has broad utility, data on all the criteria, including variations in inundation, are lacking for many parts of the Amazon. Hence, assigning spatial detail and areal extent to the various wetland types remains a challenge.

With a focus on gas exchange at the air–water interface, it is essential to have estimates of the surface area of the aquatic habitats, a daunting requirement given the large size of the Amazon basin and the wide range in dimensions of the habitats from headwater streams (<1 m across) to floodplains fringing major rivers (tens of km wide). While remote sensing can provide excellent information, currently available data on a regional scale have a spatial resolution of about 100 m, and sensors that allow seasonal and inter-annual variations in inundation to be recorded have spatial resolution of tens of km. Therefore, additional approaches are needed, especially for small rivers and streams, including geomorphology and modelling.

2.1 Remote Sensing

A variety of remote sensing approaches using passive and active microwave, laser, visible, and near-infrared and gravity anomaly detection systems are available and have been applied to tropical aquatic environments (Melack 2004). Melack and Hess (2010) applied the methodology of Hess et al. (2003), based primarily on mosaics of synthetic aperture radar (SAR) data obtained during a period of low and a period of high river stage, to the lowland Amazon basin to determine floodable area, inundated area, and areal extent of major habitats permanently or periodically inundated. Total floodable area within the lowland basin was estimated as 800,000 km2. However, portions of southern Bolivia were not covered by available SAR data, and floodplains in the south-western Brazilian Amazon were not well delineated owing to the timing of data acquisitions. Open water, floating macrophytes or grasslands, and flooded or unflooded forests were distinguished, and the spatial resolution of the products was about 100 m. Areal extent of the aquatic habitats was reported for 31 river basins and five reaches of the Solimões-Amazonas River. While not so specific as the classification proposed by Junk et al. (2011), these habitats and divisions by river basin are relevant to the biogeochemical processes considered here. High-resolution, remotely sensed products are available for specific locations in the Amazon and provide information appropriate for validation of basin-scale products (e.g. Silva et al. 2010; Renó et al. 2011; Hawes et al. 2012; Arnesen et al. 2013).

Several remote sensing approaches have been used to estimate the areas of open water in the lowland Amazon basin. Open water area was determined by Melack and Hess (2010) as 64,800 km2 from high water data. The Shuttle Radar Topography Mission (SRTM) offers digital elevations with 90 m horizontal postings (Jarvis et al. 2008) and a water body product (http://dds.cr.usgs.gov/srtm/version2_1/SWBD/). Open water area derived from the SRTM data is 72,000 km2, with river channels wider than approximately 300 m covering 52,000 km2. Hanson et al. (2013) provide a global composite of Landsat data at 30 m resolution from which an open water of 92,000 km2 can be derived (Forsberg, personal communication). Global datasets of lakes, such as those in Lehner and Döll (2004), need refinement to be applicable to the dynamic and spatially complex conditions in the Amazon.

Remote sensing of seasonal variations in inundated area depends on systems with coarse spatial resolution, and time series products are available at a 25-km scale. Hamilton et al. (2002, 2004) used data from satellite-borne passive microwave sensors to determine monthly inundation on the mainstem Amazon River floodplain (Brazil), the Llanos de Moxos (Bolivia), the Bananal (Brazil), and Roraima savannas (Brazil and Guyana). Regressions between flooded area and stage heights in nearby rivers were used to extend the records of inundation for nearly a century for the Amazon mainstem and for several decades in the other floodplains. Prigent et al. (2007), Papa et al. (2008, 2010), and Prigent et al. (2012) provide inundation estimates at 0.25° resolution, derived from several satellite-borne sensors for the period from 1993 to 2007. Comparison of these products with SAR-based estimates of inundation areas in the Amazon basin indicated generally good correspondence for moderate to large inundated units. Aires et al. (2013) used the 25-km resolution products in conjunction with SAR-based images to develop a wetland dataset for the Amazon for a 15-year period with a resolution of about 500 m. Schröder and McDonald (pers. comm.) are producing a monthly inundation product for the period from 2002 to 2009 from AMSR-E (passive microwave) and QSCAT (radar scatterometry) data at 0.25° resolution.

Gravity anomalies detected by the GRACE satellites provide estimates of changes in water volumes partially associated with seasonal variations in inundation at a scale of 100,000 s of km2, and have been used basin-wide in the Amazon (Alsdorf et al. 2010; Xavier et al. 2010). Frappart et al. (2005) combined satellite-derived SAR and altimetry data with in situ gauges to calculate water storage and inundation for the Negro River basin

2.2 Geomorphological Approaches to River Areas

Beighley and Gummadi (2011) combined relationships developed from hydraulic geometry with a high-resolution drainage network to estimate cumulative channel lengths and surface areas. Their analysis was done for drainage areas from 1 to 431,000 km2. For channels >2 m in width, they estimated that the Amazon basin contains c. 4.4 million km of channels with a combined area of 59,700 km2. Channels over 150 m in width were estimated to represent 29,500 km2 of the combined area.

To determine the area of streams and rivers in the Ji-Paraná basin (Fig. 7.1), Rasera et al. (2008) developed empirical relationships between drainage area and channel width combined with river lengths derived from a digital river network (Mayorga et al. 2005a). Areas of rivers from third to sixth order covered on average 342 km2 within the 75,400 km2 Ji-Paraná basin, representing 0.45 % of the basin. Assuming that a similar relationship applied to the whole lowland Amazon basin, an area of 22,700 km2 would result for rivers from third to sixth order. This value is similar to that estimated by Beighley and Gummadi (2011) for rivers in that size range.

Richey et al. (2002) estimated areas of moderate to small rivers and streams as a geometric series relating stream length and width to stream order. They calculated a total channel area of 21,000 km2 at low water for a 1.77 × 106 km2 area in the central Amazon, representing a fractional area of 1.2 %. If this fractional proportion is extrapolated to the whole lowland basin, a total channel area of 60,600 km2 would result.

Downing et al. (2012) employed stream network theory combined with data on stream width to approximate the areal extent of streams and rivers, when within their channels, on continental scales, and estimated that rivers and streams were likely to cover 0.30–0.56 % of land surfaces. Stream and river areas in the conterminous United States represent 0.52 % of the land surface (Butman and Raymond 2011). If these percentages are applied to the lowland Amazon basin with an area of 5.05 × 106 km2, areas for all rivers and streams range from 15,150 to 28,300 km2.

These different areas for river and stream channels vary depending on approach and region considered, and given the basin-wide, explicit method used by Beighley and Gummadi (2011), their values are adopted here. Headwater streams are not represented by their values. For large rivers, the area derived from the SRTM data is also used. Further refinement of stream areas is required.

2.3 Modelling of Inundation

Several models of river discharge and associated inundation dynamics have recently been developed and applied to the Amazon basin (Coe et al. 2007; Beighley et al. 2009; Victoria 2010; Yamazaki et al. 2011; Getirana et al. 2012; Paiva et al. 2013). Miguez-Macho and Fan (2012) compiled a groundwater dataset for the Amazon and combined it with a hydrological model. While generally successful at calculating river discharges, the models’ ability to represent inundation in the full range of floodplain and wetlands is only moderate, largely because of the lack of sufficiently accurate and detailed digital elevation models (DEMs). For example, Paiva et al. (2013) compared their modelled inundation in 3-month intervals with 0.25° remotely sensed estimates (Papa et al. 2010) averaged for the period from 1999 to 2004. The match was good in the central basin, but modelled areas significantly underestimated flooded area in the Bolivian Amazon and lower mainstem in Brazil. In the Peruvian Amazon, where the SAR analyses of Melack and Hess (2010) indicated a large area of wetland, the model also did so, but the coarse remotely sensed data did not. To better evaluate modelled results, it would be beneficial to examine SAR-derived products for specific periods and locations. Hydraulic models of flooding applied on a mesoscale have done well in locations with high-resolution DEMs (Wilson et al. 2007; Rudorff et al. 2014a, 2014b).

3 Gas Transfer Velocity Between Water and Atmosphere

Exchange of carbon dioxide and methane between surficial water and overlying atmosphere depends on the concentration gradient between air and water and on physical processes at the interface, usually parameterised as a gas transfer velocity (k), also called a piston velocity or gas exchange coefficient. Gas transfer velocities are a function of turbulence, kinematic viscosity of the water, and the molecular diffusion coefficient of the gas; the Schmidt number is the ratio of the latter two terms and is gas specific (MacIntyre et al. 1995). Schmidt numbers used here are normalised to carbon dioxide in freshwater at 20 °C and referred to as k(600) (Engle and Melack 2000). Gas transfer velocities are influenced by atmospheric stability and, in water, are altered by currents, wind, and convection, as well as rain (Ho et al. 2007), temperature, organic surficial films, and changes in hydrostatic pressure. Methane can also exit via bubbles (ebullition; Crill et al. 1988) and pass through tissues of rooted aquatic plants, both herbaceous and woody (Brix et al. 1992; Rice et al. 2010).

In lakes, direct measurements of exchange can be made with floating chambers, as has been done in the Amazon since the 1980s (e.g. Crill et al. 1988; Guérin et al. 2007). Alternatively, measurements of gas concentrations can be combined with estimates of k to calculate diffusive fluxes. While collecting and assaying samples for carbon dioxide and methane are fairly straightforward, the selection of appropriate k values remains a challenge. In a study at Lake Calado (Fig. 7.1), Crill et al. (1988) used a surface renewal model, which has a sound theoretical basis (Banerjee and MacIntyre 2004), as the basis for determination of k. Empirical relations between wind speed and k have been applied as well (e.g. Engle and Melack 2000; Guérin et al. 2007). Rudorff et al. (2011) used three different models of k: a simple wind-based equation, a small eddy version of the surface renewal model, and a wind-based model that includes diel heating and cooling (as described by MacIntyre et al. 2010). In a short-term experiment in the open water of a floodplain lake in the central Amazon basin, Polsenaere et al. (2013) applied an eddy covariance technique to calculate fluxes and k values for CO2.

Results from Rudorff et al. (2011) and Polsenaere et al. (2013) indicate that k values for standing waters in the Amazon are underestimated if based on simple wind-based relations commonly used. Rudorff et al. (2011) reported gas transfer coefficients that take into account wind as well as heating and cooling were on the order of 10 cm h−1. Polsenaere et al. (2013) reported k values ranging from 1.3 to 31.6 cm h−1, averaging 12.2 ± 6.7 cm h−1. Under conditions with high sensible and latent heat fluxes, but low wind speeds (<2.7 m s−1), k values near or above 20 cm h−1 were recorded. Based on floating chambers deployed in Balbina Reservoir (Fig. 7.1), Kemenes et al. (2011) calculated k values from 1.1 to 24.7 cm h−1, with an average of about 12 cm h−1. In contrast, Guérin et al. (2007) reported k values of about 2.5 cm h−1, based on chambers and eddy covariance in Petit Saut Reservoir, but noted that rain led to k values for CO2 from 0.8 to 13.4 cm h−1 at rain rates of 0.6–25 mm h−1. These results indicate that k values in lakes are generally higher than those used in prior regional extrapolations; e.g. Richey et al. (2002) used k values of 2.7 ± 1 cm h−1 for floodplains and lakes.

In warm tropical waters, such as those in the Amazon basin, latent heat fluxes are especially important and lead to convective mixing and enhanced k values. Conversely, diurnal heating under strong insolation can cause stable stratification of the water column that may lead to low or high k values. Hence, given the pronounced diel cycle of heating and cooling often observed in shallow tropical lakes, it is important to measure stratification and mixing, and gas exchanges, over these diel cycles, although such measurements are very seldom done.

The lack of studies of k values and gas concentrations in vegetated habitats adds further uncertainty, especially because of the large areas of flooded forests and floating macrophytes throughout the Amazon basin (Melack and Hess 2010; Junk et al. 2011). Though winds and direct heating are lower in vegetated habitats than in open waters, convective mixing and horizontal exchanges driven by differential heating and cooling and associated eddies, as water moves through the vegetation (Ortiz et al. 2013), will likely increase k values. Release of hydrophobic organic molecules by aquatic plants may reduce gas exchange within flooded vegetation.

Spatial variations in CO2 and CH4 concentrations can be large, as reported for the Amazon basin (Rudorff et al. 2011; Polsenaere et al. 2013; Abril et al. 2014) and elsewhere (Roland et al. 2010; Hofmann 2013). The high variability in time and space of bubbling adds further variance to methane evasion rates. Though floating chambers can capture bubbles, to increase spatial and temporal coverage submerged funnels are usually used. Recent applications of hydroacoustic measurements have allowed significant improvements in estimation of ebullition (Del Sontro et al. 2011), though this method has yet to be applied in the Amazon basin.

In flowing waters, most calculations of gas exchange are based on k in combination with concentration measurements, though direct measurements with floating chambers have also been used. Values of k for Amazon waters have been derived from 222Rn mass balances and floating chambers (Devol et al. 1987; Alin et al. 2011; Kemenes et al. 2011; Rasera et al. 2013). Alin et al. (2011) reported that values of k were significantly higher in small rivers and streams (channels <100 m wide), where current velocities and depth were found to be important, than in large rivers (channels >100 m wide), where wind was important. The range of k(600) values reported by Alin et al. (2011) for large rivers in the Amazon (1.2–31.1 cm h−1) is quite similar to that reported by Rasera et al. (2013) from a multi-year study in six non-tidal rivers in the Amazon basin (1.3–31.6 cm h−1). For the Uatumã River below Balbina Reservoir, Kemenes et al. (2011) reported average k values of 10.5 cm h−1. As in the case for lakes, these results indicate that k values are generally higher in flowing waters of the Amazon basin than those used in prior regional extrapolations; e.g. Richey et al. (2002) used k values of 9.6 ± 3.8 cm h−1 (Amazon mainstem) and 5 ± 2 cm h−1 (major tributaries).

4 Carbon Dioxide and Methane Concentrations and Fluxes

Recent measurements of carbon dioxide and methane concentrations and fluxes have been made in streams, rivers, wetlands, and a few lakes and reservoirs. These results are summarised by region with the intent of providing the basis for improved basin-wide estimates. Hence, areal fluxes are calculated for each habitat as Mg C km−2 year−1, kg C km−2 day−1, or g C m−2 year−1, as averages and/or ranges, if appropriate or possible.

4.1 Streams and Rivers

Working in remote headwater streams in the southern Amazon basin, Johnson et al. (2006, 2007, 2008) determined that most of the CO2 in the streams had been terrestrially respired within soils and that almost all was evaded to the atmosphere within headwater reaches. Baseflow delivered groundwater highly supersaturated in CO2, while during storms surface run-off and direct precipitation were relatively low in CO2. Concentrations of CO2 near the source of these headwater streams were 10,000–50,000 μatm. In a year-long study of a perennial, first-order stream in southern Mato Grosso draining a forested catchment, Neu et al. (2011) recorded pCO2 concentrations from 6490 to 14,980 μatm and evasion rates from the stream surface of c. 6490 ± 680 g C m−2 year−1; methane concentrations in the stream ranged from about 290–440 μatm and evasion averaged 990 ± 220 g C m−2 year−1. Similarly, Davidson et al. (2010) measured high pCO2 levels (average value of 19,000 μatm) in headwaters in remnant forests in northeastern Pará. Vihermaa et al. (2014) sampled two small streams and the La Torre and Tambopata rivers in the Madre de Dios region of the western Amazon; they reported CO2 evasion rates from 1866 to 82,900 mg C m−2 day−1 for the Tambopata River and a perennial stream, respectively. Working on small to moderate-sized rivers in the Ji-Paraná basin (Rondônia), Rasera et al. (2008) measured CO2 evasion rates per unit of river area from 695 to 13,095 mg C m−2 day−1 for third- and fourth-order rivers and from 622 to 4686 mg C m−2 day−1 for fifth- and sixth-order rivers in the basin.

Richey et al. (2009) reported pCO2 concentrations ranging from 500 to 20,000 μatm and illustrated a positive correlation of pCO2 with discharge over 4 years for the Solimões (at Manacapuru), the Madeira (at Porto Velho), and Ji-Paraná rivers. Borges et al. (2015) sampled along the mainstem Solimões and Amazon rivers and at the mouths of major tributaries and reported a range in pCO2 concentrations from 70 to 16,880 ppm. Ellis et al. (2012) measured concentrations from 860 μatm in the Acre River to 12,900 μatm in a stream in campina vegetation. Based on a 5-year study with seasonal sampling of six non-tidal rivers (Negro, Solimões, Teles Pires, Cristalino, Araguaia, and Javaés) and one tidal river (Caxiuanã), Rasera et al. (2013) reported a range of pCO2 concentrations from 259 to 7808 μatm and demonstrated a strong correlation between pCO2 and discharge. They reported a range in CO2 flux from uptake of 830 mg C m−2 day−1 to evasion of 15,860 mg C m−2 day−1. Uptake occurred in clear water rivers at low water in conditions conducive to algal growth. Alin et al. (2011) reported a similar range of evasion rates (41–14,720 mg C m−2 day−1), as did Ellis et al. (2012) with a range from 830 to 13,170 mg C m−2 day−1. Abril et al. (2014) conducted eight 800-km cruises along the main channel of the Solimões-Amazon River and portions of its major tributaries in the central basin and measured pCO2 every minute while underway. Values of pCO2 were similar to previous studies and varied from approximately 1000–10,000 ppmv, except for those in the Tapajós River which were lower.

Raymond et al. (2013) included regional information in their global estimates of carbon dioxide emissions from streams and rivers. Carbon dioxide concentrations in streams and rivers of whole Amazon basin (including the Andes) averaged about 6890 μatm and their calculated efflux, using a k(600) of 29 cm h−1, averaged 19,100 mg C m−2 day−1, expressed in relation to surface area of rivers and streams, not in relation to land area, as given in Raymond et al. (2013). The k(600) value used is at the high end of those reported by others for the rivers of the lowland Amazon as it includes streams and rivers in highland portions of the basin. Hence, the areal efflux is probably too high as a basin-wide average.

Sawakuchi et al. (2014) reported measurements of methane concentrations and fluxes from the mainstem Solimões-Amazon River and five tributaries (Negro, Madeira, Tapajós, Xingu, and Para) based on floating chambers. Sixteen of the 34 sites were sampled during low and high water. Dissolved, near-surface methane concentrations ranged from 0.02 to 0.5 μM. Overall, average riverine flux was 16.8 kg C km−2 day−1.

4.2 Lakes

In Lago Grande de Curuai (Fig. 7.1), a floodplain composed of interconnected lakes with a flooded area ranging seasonally from 850 to 2274 km2, Rudorff et al. (2011) noted gradients in CO2 concentration with higher concentrations near littoral regions with floating macrophytes than farther off shore as well as seasonal differences in concentrations. Polsenaere et al. (2013) used an equilibrator connected to an infrared gas analyser to measure pCO2 on transects in Lake Canaçari (Fig. 7.1), 450 km2 in area, during a 4-day measurement period. Incorporation of extensive spatial sampling in these two studies permitted recognition of spatial patterns not possible from other studies having far fewer samples. Based on transects of pCO2 and eddy covariance-based k values, Polsenaere et al. (2013) calculated mean evasion of 612 kg C km−2 day−1 during a 4-day low water period. Rudorff et al. (2011) estimated mean fluxes of CO2 from open water in L. Curuai of 2930, 4180, 4450, and 4370 kg C km−2 day−1 during receding, low, rising, and high water levels, respectively. Abril et al. (2014) include five large lakes as part of their continuous transects of pCO2 in the central Amazon and reported variations from approximately 20–20,000 ppmv. They found that the carbon dioxide efflux increased as the percentage of floating, emergent aquatic vegetation in the lakes increased.

Raymond et al. (2013) calculated carbon dioxide concentrations in lakes of whole Amazon basin (including the Andes) as averaging about 1906 μatm and an efflux, using a k(600) of 5.8 cm h−1, of 1230 kg C km−2 day−1, expressed in relation to surface area of rivers and streams, not in relation to land area, as given in Raymond et al. (2013).

4.3 Wetlands

The Negro River basin includes extensive flooded forests (locally called igapó) and large areas of interfluvial wetlands (Hess et al. 2003). Based on multi-temporal synthetic aperture radar data and field measurements with floating chambers of methane flux in the Jau River basin (Fig. 7.1) made by Rosenqvist et al. (2002), mean annual emission of methane from igapó was 23 Mg C km−2 year−1. Upper Negro interfluvial wetlands are a mosaic of emergent grasses, sedges, shrubs, and palms with shallow permanent water or seasonal flooding. 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. Carbon dioxide fluxes were calculated from measurements of CO2 concentrations in air and water and a k value of 2.7 cm h−1. Based on annual emission from two interfluvial wetlands representative of the region (Fig. 7.1), Belger et al. (2011) estimated average areal emission from wetland areas as 770 Mg C km−2 year−1 for CO2 and 21 Mg C km−2 year−1 for CH4.

Large savanna floodplains occur in the Llanos de Moxos (Bolivia) and in Roraima (Brazil) (Fig. 7.1) (Hamilton et al. 2002; Ferreira et al. 2007). Based on measurements in similar systems elsewhere, Melack et al. (2004) approximated mean annual methane emission from these two areas as 70 Mg C km−2 year−1. Jati (2013) made monthly measurements of carbon dioxide and methane flux with floating chambers in 80 wetlands near Boa Vista (Roraima); mean values from his results were about 9670 kg C km−2 day−1 and 9.6 kg C km−2 day−1, for CO2 and CH4, respectively. These are quite high CO2 fluxes and rather low CH4 fluxes compared to other Amazonian habitats. Emissions from cultivated rice in Roraima are not available.

4.4 Reservoirs

Five hydroelectric reservoirs (Tucuruí, Balbina, Samuel, Curuá-Una, Serra da Mesa), covering about 6300 km2, currently exist in the lowland Amazon in Brazil. All were constructed decades ago and continue to release both carbon dioxide and methane from their surfaces and through their turbines and to enhance releases in downstream rivers. Only Balbina has data collected from multiple upstream and downstream stations over a full year as well as measurements of fluxes associated with turbines (Kemenes et al. 2007, 2011). Though not in the Amazon basin, multiyear studies at Petit Saut (French Guiana), located in tropical forest, include measurements in the reservoir and downstream (Abril et al. 2005). Data scattered through the years at other Amazonian reservoirs are also available (see citations in Melack et al. 2004; Guérin et al. 2006). Kemenes et al. (2016) report carbon dioxide and methane evasion via degassing through turbines and downstream for Tucuruí, Samuel, and Curuá-Una reservoirs. Barros et al. (2011) summarised much of the data from Amazonian reservoirs, though methane emission from Balbina is listed as 10 mg C m−2 day−1 rather than 47 mg C m−2 day−1, as reported in Kemenes et al. (2007). Moreover, degassing through turbines and downstream is not included for Balbina or other Amazonian reservoirs.

Carbon dioxide emissions from the surface of Balbina and Petit Saut reservoirs averaged 1296 Mg C km−2 year−1 and 473 Mg C km−2 year−1, respectively. In the case of CO2 total annual emission from the Balbina hydroelectric system, including the reservoir, turbine outflow, and river channel extending 30 km downstream, was 1340 Mg C km−2 year−1, when expressed relative to the average reservoir area. Average CH4 emissions from the reservoir surface were 18 Mg C km−2 year−1, and total emissions were 39 Mg C km−2 year−1, indicating the importance of degassing of methane through the turbines and downstream. Though Barros et al. (2011) present mean fluxes from Amazonian reservoirs for CO2 of 400 Mg C km−2 year−1 and for CH4 of 50 Mg C km−2 year−1, these values should be used with caution since the data from which they were obtained are based on only a few samples from most of the reservoirs without adequate seasonal sampling and without data on downstream fluxes.

4.5 Airborne Surveys

Airborne campaigns have provided integrated coverage of subregions of the Amazon basin and permitted calculation of carbon balance (Lloyd et al. 2007) and methane emission (Beck et al. 2012; Miller et al. 2007). Miller et al. (2007) collected vertical profiles of methane over 4 years at sites near Santarém and Manaus and calculated average emissions of 20 kg C km−2 day−1. Wetlands are likely the major source of methane, at least during seasons with extensive inundated areas. Other sources include fires, urban areas, termites, and, perhaps emissions associated with terra firme forests. Beck et al. (2012) described results of airborne campaigns in November and May, periods representing generally low and large inundation, during which continuous, in-flight measurements of CH4 and sampling for isotopic analyses were conducted. The flights extended over much of the lowland Amazon in Brazil and were concentrated in the central basin. Isotopic measurements indicated that biogenic methane predominates, and wetlands are likely the major source though near Manaus anthropogenic sources, such as waste decomposition, contribute. A signature of biomass burning was detected in samples collected during the dry season, but this source appeared to be minor for CH4. Beck et al. (2012) estimated a CH4 flux for the lowland Amazon during November as 27 ± 9 kg C km−2 day−1 and during May as 32 ± 14 kg C km−2 day−1.

5 Regionalisation of Fluxes

5.1 Prior Estimates

Though various estimates of regional carbon dioxide and methane fluxes have been made through the years, only recent estimates that used data available through the beginning of the twenty-first century are summarised here. Richey et al. (2002) and Melack et al. (2004) were the first to use regional analyses of microwave remote sensing data to establish inundated areas and habitats. Both applied Monte Carlo error propagation to establish uncertainties.

Richey et al. (2002) used carbon dioxide measurements primarily from the Solimões-Amazon River, its fringing floodplain and mouths of major tributaries, and conservatively low piston velocities to calculate an outgassing rate of 830 ± 240 Mg C km−2 year−1 for the annual mean flooded area. The areal flux was combined with remote sensing-derived inundated areas at high and low stages for rivers and floodplains over 100 m across (Hess et al. 2003) to determine total outgassing in a 1.77 million km2 quadrant in the central basin. The seasonal variation in inundated area was assumed to track river stage, as measured by the sparse network of gauges. A comparison of inundated area-derived passive microwave data versus stage for large floodplains supports this assumption (Hamilton et al. 2002). Areas of moderate to small rivers and streams in this quadrant were approximated as a geometric series relating stream length and width to stream order. To extrapolate to the whole Amazon basin (6.07 million km2), an areal flux of half that used for the central quadrant was applied to the 4.3 million km2 outside the central quadrant to yield a value of 470 Tg C year−1 and that applied to the area of the lowland basin yields an evasion of 390 Tg C year−1.

Rasera et al. (2013) extrapolated their results to a central Amazon quadrant (1.47 million km2). To do so, they combined (1) areas of streams (<100 m wide) and the areas of rivers and floodplains (>100 m wide) for high and low stage as reported in Richey et al. (2002); (2) k(600) values based on recent work for rivers and streams; and (3) pCO2 values from measurements and from a relationship between measured values and average soil cation exchange capacity. If the annual total evasion of CO2 calculated in this manner were increased in proportion to the slightly larger central basin area used by Richey et al. (2002), 432 ± 78 Tg C year−1 would result. This annual rate is about twice that reported by Richey et al. (2002) and reflects higher k(600) values and improved data for streams. It is important to recognise that the fluxes calculated for the mainstem Amazon and tributaries in Rasera et al. (2013) include floodplain areas, not just river channels; if extrapolated to the area of the lowland basin a flux of 1240 ± 206 Tg C year−1 results.

Regional extrapolation of fluxes of carbon dioxide from streams and small rivers is especially difficult because of the few measurements among the millions of kilometres of these systems and the large spatial and temporal variations observed. Johnson et al. (2008) approximated potential evasion of CO2 from headwater streams basin-wide (an area of 6.07 million km2, which included the Andean highlands) as 114 Tg C year−1 or 19.5 Mg C km−2 year−1, excluding inundated areas and accounting for human modified land uses, and where the areal rate is expressed per km2 of total land area, not stream areas. Variations in annual water balances for the period from 1976 to 1996 would introduce about a 10 % increase or decrease between wet or dry years. Their regionalisation approach is based on groundwater fluxes, determined as the difference between average annual precipitation and evapotranspiration, and estimates of soil pCO2 from carbonate equilibrium reactions at a spatial scale of 0.1°. Though an interesting approach, it requires validation based on actual measurements for a variety of headwaters, such as Andean, blackwater, or savanna, streams that are different from those examined by Johnson et al. (2006). Rasera et al. (2008) extrapolated from the Ji-Paraná River basin to the Amazon basin and arrived at a value substantially higher than can be calculated from the data presented. Based on the total CO2 evasion from the Ji-Paraná River basin divided by the area of the basin and multiplied by the area of third- to sixth-order rivers (derived from the fractional area of these rivers in the Ji-Paraná River basin; see above), the annual evasion is about 21 Tg C for these rivers in the lowland Amazon basin.

Melack et al. (2004) used available measurements of habitat-specific methane fluxes; all the data were from the central Amazon basin and based mainly on methane captured in floating chambers. By combining these values with remote sensing-based estimates of areal extent of aquatic habitats (open water in lakes, rivers, flooded forests, and aquatic macrophytes), they calculated regional estimates for the mainstem Solimões-Amazon in Brazil and for the same 1.77 million km2 quadrant used by Richey et al. (2002). Seasonal changes in the areal extent of the aquatic habitats were approximated by interpolating between remote sensing-derived areas obtained for low and high water periods. A time series of inundation extent along the mainstem Solimões-Amazon floodplain, based on passive microwave data, was used to calculate inter-annual variations in methane flux for 3 years. To extrapolate to the whole lowland basin (5.05 million km2), a single, habitat-averaged value of 30 Mg C km−2 year−1, calculated from the mean annual emission estimated for the mainstem Solimões-Amazon in Brazil and the mean annual flooded area of this reach, was used, resulting in a flux of c. 22 Tg C year−1. If expressed as the greenhouse gas warming potential equivalence of CO2, this mean flux amounts to about 0.2 Pg C year−1 (not 0.5 Pg C year−1, as noted in the original paper).

Sawakuchi et al. (2014) subtracted estimates of Landsat-based lake areas from the water body category in an AVHRR 1-km regional land cover product (Brown et al. 2003) to obtain a large river channel area of approximately 91,000 km2, a value considerably larger than others discussed earlier. When combined with their methane fluxes, they calculated an annual average flux of 0.37 Tg C as methane from the large rivers of the Amazon basin. If the areal estimate of Beighley and Gummadi (2011) is used, the flux is 0.12 Tg C year−1, and if the SRTM areal estimate is used, the flux is 0.21 Tg C year−1. These methane fluxes represent 0.5–1.6 % of the basin-wide emission calculated by Melack et al. (2004) and indicate a minor role for the large river channels. Methane emission from the mainstem Solimões-Amazon River in Brazil was estimated by Melack et al. (2004) to represent only 0.06 % of the total flux for that reach including the fringing floodplains.

Barros et al. (2011) estimated emission from extant tropical Amazonian reservoirs, based on an area of 20,000 km2, of 8 Tg C year−1 as CO2 and 1.0 Tg C year−1 as CH4. We question these values based on the area of Amazonian reservoirs being 6300 km2 and based on areal emission estimates available from Balbina and Petit Saut, as described above.

Global or continental scale calculations derived from satellite retrievals of atmospheric concentrations and inverse modelling provide coarse spatial resolution methane emission values for the Amazon basin (e.g. Frankenberg et al. 2008). Beck et al. (2012) examined the performance of CH4 inversion models, constrained by observations from surface stations and SCIAMACHY retrievals, in comparison to their airborne campaigns for the Amazon basin. In this comparison, the global models used the same transport model but different prior CH4 inputs from wetlands, none of which were well suited to the Amazon basin. One result derived from the transport modelling suggested that the Amazon basin is influenced by an atmospheric region larger than the basin. Furthermore, Beck et al. (2012) concluded that a reliable annual methane budget will require regional airborne campaigns over a full year.

5.2 New Estimates

In principle, total basin-wide fluxes of CO2 and CH4 (F, in units of Tg C year−1) could be calculated using a general expression similar to that in Melack et al. (2004):

$$ F\ {=}^6{\varSigma_{j=1}}^{12}{\varSigma}_{i=1}{t}_i{A}_{ij}{f}_{ij} $$
(7.1)

where F is the flux of CO2 or CH4 for each habitat (expressed as kg C km−2 day−1); j is each habitat: (1) headwaters, (2) streams and moderate-sized rivers, (3) large rivers, (4) floodplains, (5) wetlands, and (6) hydroelectric reservoirs; A is an estimate of average flooded area of each habitat per month; t is the number of days per month; and i is each month incremented from 1 to 12. Depending on available data, several habitat categories could be subdivided: e.g. large rivers (e.g. Solimões-Amazon and white water tributaries, Negro and black water tributaries, Tapajós, and other clear waters), floodplains (e.g. open water lakes, flooded forests, aquatic macrophytes), and wetlands (e.g. upper Negro interfluvial, Roraima savanna, Llanos de Moxos). If data from multiple years were available, such as the inundation time series of Papa et al. (2010) or Paiva et al. (2013), or calculated from an empirical or mechanistic model of carbon dioxide and methane dynamics, Eq. (7.1) could be evaluated repeatedly to determine inter-annual variability.

Several challenges make it difficult to apply Eq. (7.1). In particular, sufficient information about the spatial and temporal variations of inundated areas, habitat characteristics, and associated fluxes on a basin-wide scale for multiple years is lacking. A modelling system that combines climatic and hydrological processes with biogeochemical and ecological processes is required.

As an alternative, a combination of the published values, summarised in prior sections, for specific regions and calculations based on averaged measurements of carbon dioxide from (1) moderate-sized rivers, (2) large rivers, (3) floodplains and wetlands, and (4) hydroelectric reservoirs as noted in Table 7.1 are used to estimate carbon dioxide emissions. These fluxes include those measured with floating chambers and those calculated with new estimates of gas exchange velocities and in situ gas concentrations. Remote sensing-based estimates of inundated areas at low and high water levels and modelled variations in inundation are used to approximate annual values for floodplain and wetland environments. River channel areal estimates from Beighley and Gummadi (2011) and SRTM are used. The estimates are standardised to the lowland basin below 500 m as delineated from JERS-1 synthetic aperture radar data (5.05 million km2). Insufficient new measurements for methane are available to improve upon Melack et al. (2004).

Table 7.1 Spatial components used in calculation of lowland Amazon basin carbon dioxide evasion
Full size table

Carbon dioxide evasion from rivers and streams is estimated to be c. 200 Tg C year−1. Though fluxes per unit area for rivers and streams are based on updated k(600) and recent data from a range of stream and river sizes, a smaller surface area for these habitats than previous basin-wide estimates is being used. The annual flux from the category called floodplain and wetland habitats, which includes lakes, plus those from other wetlands and reservoirs is estimated to be approximately 1600 Tg C year−1. These fluxes incorporate updated k(600) values; however, the lack of information about k values in vegetated areas is a concern because Melack and Hess (2010) estimate that about 79 % of the lowland basin is characterised by woody vegetation with another 13 % predominately herbaceous vegetation. Since k values are likely to be less in these areas in comparison to those in lakes, the annual fluxes would be lower. Furthermore, basin-wide modelled inundation fractions and remote sensing-based areas do not represent well the seasonality in areas or fluxes. With these and other issues noted above acknowledged, total evasion of carbon dioxide for the lowland Amazon basin is estimated to be c. 1.8 Pg C year−1. Though a formal analysis of uncertainty cannot be done based on the heterogeneous information used for these estimates, based on spatial and temporal variability and uncertainty in measurements, an overall uncertainty of at least ±50 % is reasonable.

Raymond et al. (2013) calculated global annual evasion of carbon dioxide from streams and rivers ranging from 1.5 to 2.1 Pg C and evasion from lakes and reservoirs ranging from 0.06 to 0.84 Pg C. The ranges represent 5th and 95th confidence intervals derived from a Monte Carlo analysis. Wetlands were not included. That the value reported here for the lowland Amazon basin is similar to the sum of these global estimates indicates the importance of including tropical floodplains and other wetlands in calculations of carbon dioxide evasion from inland waters.

6 Sources and Decomposition of Organic Carbon

Floodplains and other wetlands are productive aquatic environments in which most of the production and evasion of CO2 and CH4 is likely derived from metabolic processing of the carbon fixed by aquatic plants. These environments also export considerable amounts of carbon to rivers and accumulate sediments. Estimates of carbon balances for floodplains at several spatial scales provide supporting evidence: Calado (Melack and Engle 2009), Curuai (Rudorff 2013), the mainstem Solimões-Amazon floodplain in Brazil (Melack and Forsberg 2001), and a 1.77 million-km2 quadrant in the central Amazon (Melack et al. 2009; Abril et al. 2014). In particular, Abril et al. (2014) suggest that Amazonian wetlands export about half of their primary productivity to neighbouring waters where it is metabolised and much is released to the atmosphere. Further evidence is provided by estimates of root respiration by herbaceous and woody aquatic plants (Hamilton et al. 1995; Worbes 1997), isotopic studies of microbial respiration (Waichman 1996), calculation of aquatic macrophyte growth and decay (Engle et al. 2008; Silva et al. 2009, 2013), rates of methane oxidation in exposed wetland sediments (Koschorreck 2000), and the enrichment of δ13C of CO2 in lowland rivers, expected if C4 grasses are significant sources of respired carbon (Mayorga et al. 2005b).

Understanding the relevance of evasion of carbon dioxide from aquatic environments to the carbon balance of terra firme forests requires (1) determination of the proportion of the carbon fixed within aquatic ecosystems versus that imported from uplands as inorganic or organic carbon and (2) measurements of aquatic respiration and of decomposition of these sources of organic carbon. Though further work is needed, recent results indicate that the carbon inputs to aquatic systems that are eventually emitted as CO2 or CH4 vary among habitats and are related to hydrological conditions and proximity of carbon sources.

Headwater streams in the Juruena watershed (Fig. 7.1), and presumably elsewhere, receive most of the CO2 degassed in their uppermost reaches directly from terrestrially derived respiration in soils with small contributions of organic C from riparian and upland litter that is gradually processed downstream (Johnson et al. 2006, 2007). In slightly larger streams and small rivers, relatively labile dissolved and particulate organic carbon and lateral inputs of dissolved CO2 support evasion (earlier work summarised in Richey et al. 2009; Davidson et al. 2010). There is regional heterogeneity in the carbon sources, with 13C-depleted CO2 in streams draining sandy soils in the Negro basin indicative of C3 plants, while the 13C-enriched CO2 in streams passing through pastures in Rondônia indicating C4 grasses. Based on direct assays of 14CO2 outgassed from small streams and rivers in the western Amazon, Vihermaa et al. (2014) report that a portion of the carbon dioxide is derived from sedimentary rock and carbonate weathering.

Carbon dioxide evasion from large rivers appears to be supported by a wide variety of carbon compounds from a combination of nearby and distant organic carbon sources. Especially relevant results are reported by Ellis et al. (2012), who determined the δ13C of the CO2 evaded from the Amazonian rivers and found that organic carbon from C3 and C4 plants and phytoplankton was evident and spatially and temporally variable. Another valuable result from Ward et al. (2013) demonstrated that the degradation of lignin and associated macromolecules in water from the lower Amazon River was sufficient to support considerable respiration and presumably CO2 evasion. Since these compounds have been thought to be refractory and are often of terrestrial origin, this finding supports the notion that metabolism in the large rivers is supported by diverse carbon sources. Fatty acid and stable isotope analyses by Mortillaro et al. (2011, 2012) as well as studies by Kim et al. (2012), using the branched and isoprenoid tetraether index, offer further evidence of multiple carbon inputs to the lower Amazon River. Oxidation of petrogenic organic carbon has also been documented (Bouchez et al. 2010). Results by Mayorga et al. (2005b) indicated that the main source of respired carbon was <5 years old and that the dissolved CO2 was isotopically different from organic carbon in the rivers sampled at the same time. These results imply that inputs of labile carbon, which is rapidly oxidised, support the generation of the high pCO2 values observed. However, Ellis et al. (2012) found that the respired carbon was isotopically similar to that in the water. Further, as noted by Rasera et al. (2013), the high concentrations of CO2 during high river stages may reflect export of labile organic carbon from fringing floodplains. In the Amazon, Madeira and Solimões rivers Ellis et al. (2012) found that riverine respiration could account for most or all of the CO2 evaded, while in the Negro River it could account for only 15–34 %. Photo-oxidation of organic carbon appears to make small contributions to CO2 in large rivers (Amaral et al. 2013; Remington et al. 2011).

In summary, the relative importance of carbon sources originating in terra firme forests versus aquatic habitats varies as a function of position in the continuum from headwaters to large rivers with fringing floodplains or associated wetlands. Hence, a general conclusion regarding the proportion of terra firme net productivity that is emitted to the atmosphere after lateral transport to aquatic environments is difficult to state. However, it is clear that CO2 evasion, in particular, is supported by carbon fixed by terra firme plants in many streams and that this carbon contributes to metabolic processes in the largest rivers. In contrast, floodplains and wetlands likely represent environments where the CO2 and CH4 emitted to the atmosphere are derived largely from carbon fixed by aquatic plants and, to a lesser extent, algae, and that a portion of organic carbon metabolised in rivers is supplied by their floodplains. Based on this logic, the values in Table 7.1 lead to the conclusion that the majority of the basin-wide evasion of carbon dioxide is derived from plants in aquatic environments.

7 Uncertainties and Research Needs

7.1 Field Measurements

The largest uncertainties stem from the sparseness of measurements in time and space. For example, based on a Monte Carlo error analysis, Melack et al. (2004) noted that the uncertainty in actual methane fluxes, largely because of the high spatial and temporal variability, compounded by the episodic nature of ebullition, was larger than the uncertainty associated with remote sensing-based habitat extent. This result applied to the best sampled floodplains of the central basin. In many wetlands of the basin, few or no data are available. Particularly large gaps with no data exist in the Llanos de Moxos, Bananal, peatlands in the western Amazon (Lahteenoja et al. 2011), Peruvian lowlands, and habitats above 500 m. The extensive network of streams and medium-sized rivers is significantly under-sampled, and Richey et al. (2009) made a plea for collecting many spatially distributed measurements given the large variability observed among the data gathered. While reservoirs are receiving increased attention, time series data and measurements above and below the dams are required to better guide the management of greenhouse gas evasion from hydroelectric projects.

Within lakes, reservoirs, and wetlands, diel and seasonal variations in vertical stratification and horizontal advection influence the concentration gradients of gases and the transfer velocities. With the increasing availability of in situ sensors and automatic recording systems, it is possible to incorporate temporal and spatial variations in physical and chemical conditions in calculations of gas exchanges. For example, deep convective mixing at night, common in tropical waters, is likely to increase both transfer velocities and concentrations of carbon dioxide and methane in surficial waters, thus increasing fluxes.

Other components of the carbon balance in aquatic habitats can be challenging to measure, and data are lacking for many areas or are without sufficient spatial and temporal coverage even at better studied sites. Key processes that need attention are primary productivity by algae and higher plants, sedimentation in floodplains and other wetlands, transport of dissolved and particulate inorganic and organic carbon from uplands to inland waters, and rates of respiration and decomposition.

7.2 Modelling

While models of river discharge and inundation are improving, the current basin-scale models do not properly flood some important and large habitats, such as the largely rain-fed interfluvial wetlands in the upper Negro basin, the Roraima wetlands, or the Llanos de Moxos. One key to improvement is higher spatial resolution DEMs for floodplains. Biogeochemical models of carbon dioxide and methane production and evasion appropriate for conditions in the Amazon are not available (Riley et al. 2011), though relevant models are under development (Potter et al. 2016). While several models have potentially useful components or formulations (Ringeval et al. 2010; Bloom et al. 2010; Cao et al. 1996; Potter 1997; Walter and Heimann 2000), no spatially explicit model exists that incorporates the inundation dynamics and biogeochemical processes of aquatic environments in the Amazon.

8 Climate Change, Exceptional Events, and Human Impacts

Climate warming, climate variability, including exceptional droughts and floods and severe wind storms, and fires are influencing the Amazon basin (Davidson et al. 2012). Additionally, human alterations include agricultural expansion with associated deforestation, construction of dams, and fossil fuel exploration (Melack 2005; Costa et al. 2009; Pires and Costa 2013; Renó et al. 2011; Castello et al. 2013; Finer and Orta-Martinez 2010). A review of current climatic conditions in the context of aquatic environments in the Amazon basin by Melack and Coe (2013) noted the paucity of meteorological records in floodplains and other wetlands. Basin-wide, long-term warming trends in air temperature are generally 0.1–0.3 °C per decade (1960–2009; Burrows et al. 2011). Rainfall has negative trends in the northern Amazon and positive trends in the southern Amazon (1949–1999; Marengo 2004). Severe droughts occurred in 2005 and 2010 and were associated with tropical Atlantic warming and ENSO events (Marengo et al. 2008, 2009, 2011; Zeng et al. 2008; Villar et al. 2011). In contrast, an exceptional flood occurred in 2009 (Chen et al. 2010). Furthermore, increased variability in climatic conditions with increased frequency and severity of droughts and storms is projected by global models (Malhi et al. 2008; Gloor et al. 2013; Huntingford et al. 2013; Lau et al. 2013; Liu et al. 2013; Reichstein et al. 2013).

Another consequence of severe weather can be substantial disturbance of forests as trees are blown down (Negrón-Juárez et al. 2010; Chambers et al. 2007). Droughts can lead to increased incidence of fires (Aragão et al. 2008; Longo et al. 2009), even in igapó forests (Flores et al. 2014).

Melack and Coe (2013) ran simulations of inundation under altered climate and land uses specifically for the Amazon basin using a basin-wide hydrological model forced with observed climate data from 1950 to 2000. Simulations of inundation with 10 and 25 % decreases in rainfall (without seasonal and spatial variations included) resulted in reductions in inundation similar to reductions in rainfall: −5 to −20 % (10 %) and −12 to −30 % (25 %). Based on 35 % deforestation coupled to a global climate model, rainfall decreased and evapotranspiration decreased more; hence average maximum flooded area along mainstem Amazon increased slightly compared to current land cover. Others have considered actual or potential climate change or land use impacts on hydrological conditions in the Amazon basin (Coe et al. 2009, 2011; Casimiro et al. 2011, 2012; Langerwisch et al. 2012; Cox et al. 2013). In addition to the existing hydroelectric reservoirs, others are under construction (Belo Monte on the Xingu River; Santo Antonio and Jirau, run-of-river dams on the Madeira River), and many more are planned throughout the Amazon basin including the Tapajós hydroelectric complex which could inundate about 2000 km2 (Finer and Jenkins 2012). To estimate the emissions of CO2 and CH4 from these new and planned reservoirs is difficult because of differences in construction and operation and because economic and environmental issues will likely play roles. Furthermore, road construction and agriculture create numerous small impoundments (Macedo et al. 2013), of unknown total area with unmeasured fluxes.

9 Conclusion

The updated value for annual carbon dioxide emission from aquatic habitats in the lowland Amazon basin of 1800 Tg C is larger than previous estimates. Almost 90 % of this flux is likely associated with lakes, floodplains, and other wetlands. Carbon fixed by terra firme plants contributes most of the carbon dioxide emitted from streams and adds organic carbon to rivers. Lakes, floodplains, and other wetlands represent environments where the CO2 emitted to the atmosphere is derived largely from carbon fixed by aquatic plants with lesser contributions from algae, and a portion of organic carbon metabolised in rivers is supplied by their floodplains.

To further improve the updated estimates of carbon dioxide and methane evasion from aquatic habitats in the Amazon basin requires several activities. The largest uncertainties stem from the sparseness of measurements in time and space. Deployment of eddy covariance systems, continuous measurements along transects, and regional airborne campaigns would help considerably. Inclusion of habitats not well characterised, such as flooded forests, savannas, intermittently flooded regions along streams, and depressions within terra firme forests is needed. Hydroacoustic techniques and sampling releases via turbines in hydroelectric dams will improve estimation of ebullition. Seasonal and inter-annual variability of inundated areas derived from remote sensing and modelling should be incorporated. The lack of high-resolution digital elevation data are a serious limitation throughout the basin. Given the pronounced diel cycle of heating and cooling often observed in shallow tropical waters, it is important to measure stratification and mixing, and gas concentrations and exchanges, over these diel cycles. Biogeochemical models of carbon dioxide and methane production and evasion appropriate for conditions in the Amazon require further development. Climate warming, climate variability, including exceptional droughts and floods and severe wind storms, fires, agricultural expansion, and construction of dams are all influencing the Amazon basin with consequences for the role of aquatic environments in the carbon cycle and release of carbon dioxide and methane.