Production of carbon dioxide (CO2) in soils can lead to supersaturation of dissolved free CO2 (pCO2) in groundwater, which later evades to the atmosphere as groundwater enters streams and rivers. This process could be a significant pathway for return of terrestrially fixed C to the atmosphere. We measured pCO2 monthly over two years at multiple stations along three streams from their headwaters in remnant mature forests through multiple land covers in Pará, Brazil. The pCO2 averaged 19,000 μatm in headwaters and decreased to about 4,500 μatm downstream. Similar values were measured in headwaters of two small pristine mature forest catchments. Two approaches were used to estimate groundwater pCO2 evasion: assuming that headwater pCO2 measurements reflected incoming groundwater pCO2 or that all entering stream water was in equilibrium with previously measured deep soil CO2. With these assumptions, losses from the terrestrial environment through aquatic evasion of pCO2 would be 0.02–0.15 Mg C ha−1 of land area yr−1, which is about 2–3 orders of magnitude lower than annual estimates of soil respiration and net primary productivity. However, downstream pCO2 values that appear to be in quasi-steady state indicate contributions from other C sources, such as aquatic primary production, soil erosion, dissolved organic matter, or litter inputs from streamside vegetation. Hence, lateral pCO2 loss from groundwater to streams is minor for most of the terrestrial ecosystems of this region, although C loss to streams could be significant for net terrestrial budgets in riparian ecosystems or areas experiencing erosion.
 Mature forest ecosystems of the Amazon Basin are thought to be long-term sinks of atmospheric C, in the range of 0.5–0.8 Pg C yr−1 [Phillips et al., 2008]. However, Richey et al.  estimated that the efflux of CO2 from rivers and streams to the atmosphere in a 1.8 million km2 quadrant of the central Amazon Basin is about 0.2 Pg C yr−1 and that a scaled-up estimate for the entire basin would be on the order of 0.5 Pg C yr−1. These authors postulated that these fluxes were supported largely by terrestrial carbon entering streams and rivers, either as groundwater CO2 or as terrestrial organic matter that was subsequently mineralized by the aquatic ecosystem. They further suggested that these could be significant pathways for return of terrestrially fixed C to the atmosphere in the Amazon Basin, and one that is not measured or considered in most terrestrial C balance studies [Richey et al., 2002]. Carbon budgets for terrestrial ecosystems could be incomplete and net C sequestration could be overestimated if hydrologic export of evaded CO2, dissolved organic and inorganic carbon (DOC and DIC) and particulate organic carbon are not considered [Billett et al., 2004; Cole et al., 2007; Tranvik et al., 2009]. Here we examine the potential magnitude of pCO2 moving through groundwater from terrestrial Amazonian ecosystems into several small streams relative to other terrestrial C fluxes in the eastern Amazon region.
 Production of carbon dioxide (CO2) in soils can lead to supersaturation of dissolved CO2 (pCO2) in groundwater seeps [Johnson et al., 2006]. Concentrations of CO2 within deep soil air profiles in Amazonian forests and pastures often exceed 50,000 ppmv [Davidson and Trumbore, 1995; Davidson et al., 2004; Johnson et al., 2008], indicating that supersaturated groundwater may be common. As groundwater enters streams and rivers, much of the pCO2 rapidly evades to the atmosphere, thus transferring C that was fixed in terrestrial ecosystems to the atmosphere via an aquatic pathway.
 We measured pCO2 along three streams from their headwaters in mature evergreen forest remnants, through landscapes of mixed land use, including pastures, secondary forests, and agriculture, in the municipality of Paragominas, Pará, Brazil. Our objective was to measure pCO2 in headwaters and downstream stations in these small streams and to estimate the upper limit of plausible evasion of CO2 that could have been derived from the surrounding terrestrial ecosystems of these small catchments. We also present pCO2 measurements from two small watersheds within an intact mature forest in the neighboring municipality of Capitão Poço.
2.1. Research Site
 The region around Paragominas supported high-statured evergreen forest before it became a regional center for logging and ranching following the construction of the Belém-Brasilia highway [Nepstad et al., 1991]. More recently, intensive agricultural production for upland soybean, rice, and corn has also increased. These trends are similar to those for much of the Amazon region, particularly along the eastern and southern arc of deforestation [Fearnside, 2001].
 As a result of this extensive land-use change, it is impossible to find large areas of pristine forest that encompass an entire watershed basin of a perennial stream in the Paragominas area. The forested headwater areas of the three streams selected for this research (Table 1) were established as references relative to downstream samples in a larger study of soil water quality (see Figueiredo et al.  for delineation of watershed boundaries superimposed on Landsat images). Even these headwater areas, however, have been selectively logged, and have suffered some entry of pastoral and agricultural activities. As will be discussed later, the data demonstrate that one of these three headwater streams behaves much differently than the others, despite no obvious apparent difference during our site selection. More complete descriptions of land cover within these watersheds are given in a companion paper by Figueiredo et al. .
Table 1. Stream and Catchment Characteristics and Estimates of Possible CO2 Evasion Based on Two Methods of Estimating Groundwater pCO2 and Subsequent Evasion, as Described in the Discussion Section
Area Upstream of Gauging Station (ha)
Mean Discharge (m3 s−1)
Headwater pCO2 (μmole CO2 L−1)
Groundwater CO2 Evasion Based on Headwater (Mg C yr−1)
Groundwater CO2 Evasion Per Hectare Land Based on Headwater Concentrations (Mg C ha−1 yr−1)
Maximum Groundwater CO2 Evasion Per Hectare Land Based on Soil CO2 (Mg C ha−1 yr−1)
 In order to include streams that flow through fully forested catchments, two streams in small catchments, each about 20 ha, were sampled in the Capitão Poço municipality, which is about 100 km from Paragominas. These streams, including their headwaters, are located entirely within a 3,700 ha segment of unharvested mature forest within a single property. The watershed areas for these forested streams are much smaller than in the Paragominas study (Table 1), but are characterized by similar geologic and pedogenic history.
 According to the Geological Survey of Brazil (www.cprm.gov.br), both Paragominas and Capitão Poço watersheds overlay the Ipixuna and Barreiras formations, where the predominant clay mineral is kaolinite. Paragominas soils are classified as deeply weathered Oxisols (Haplustox, in the U.S. soil taxonomy, or Yellow Latosol in the Brazilian classification) in upper landscape positions, while in lower landscape positions some soils are classified as a clay-rich (40–60%) Plintosolos Haplicos (Plinthustults in the U.S. soil taxonomy), developed from both clay-rich colluvium from upslope and from the sandier Barreiras Formation [Markewitz et al., 2004; Moraes et al., 2006]. Soil studies in Capitão Poço municipality [Embrapa-SNLCS, 1990] also identified similar soils in this region (i.e., Yellow Latosols and Plinthosols), and clay contents are similar to Paragominas soils [Davidson et al., 2007]. Annual average rainfall for this region is 1800mm, most of which falls within a 6-month rainy season [Jipp et al., 1998]. Deep roots maintain an evergreen forest canopy, despite the long dry season [Nepstad et al., 1994].
2.2. Stream Water Sampling and Analyses
 Sampling along each stream was largely driven by issues of access. Land in this region is mostly privately held, and roads are typically unimproved dirt roads on private ranches. As such, sample points were identified at five locations on Igarape-54 (IG54), seven locations on Igarape-7 (IG7), and three locations on Igarape-Pajeú (IGP) where we were able to obtain permission for access (see Figueiredo et al.  for exact locations of sampling stations). For this reason, it was impossible to use consistent distances between sampling stations. For the Paragominas streams, we adopt a convention of calling the headwater sampling station within each watershed S1, and successive downstream sampling stations S2, S3, etc (e.g., IGP-S1 is the headwater sampling station in Igarape-Pajeú and IG54-S5 is the fifth downstream sampling station along Igarape-54). For each of the two Capitão Poço pristine watersheds (CP1 and CP2), we were able to find small groundwater seeps for the first sampling stations (S1), which were located uphill of the organized headwater streams. The organized headwater streams were used as the second sampling stations (S2). In all three Paragominas streams, sampling for pCO2 was initiated in November 2003 and continued on a nearly monthly basis until August 2005. In the Capitão Poço streams sampling was begun in February 2004 and continued on a nearly monthly basis until June 2005.
 Several measurements of stream water quality, including turbidity, pH, temperature, conductivity, alkalinity, dissolved oxygen and anion and cation concentrations are described in a companion manuscript by Figueiredo et al. . Here we take advantage of these data on stream temperature, pH and alkalinity in order to convert pCO2 values measured under laboratory conditions to pCO2 values under in situ stream conditions (described below). Temperature and pH were measured in situ using a field multimeter (WTW, model Multiline P3 pH/Oxygen, Gold River, CA). Stream water grab samples were collected in previously acid washed 250-ml polypropylene bottles. Bottles were filled to capacity to minimize headspace and were placed in cold storage (∼4°C) within a few hours of collection. Samples were returned to a field laboratory in Paragominas, where they were again analyzed for pH and for alkalinity by endpoint titrations with 1 mM HCl to pH 4.5 [Clesceri et al., 1998]. If possible, samples were analyzed for alkalinity on the same day of collection, but many were retained cold for up to 4–5 days until analysis.
 For pCO2 measurements, stream samples were collected in 60 mL syringes from 20 cm below the water surface, talking care to avoid air bubbles in the syringe before closing the syringe valve and transporting the samples to the field laboratory in Paragominas. The syringe samples were allowed to equilibrate to room temperature, and then 30 mL of water were injected into 60 mL serum bottles that had previously been flushed with CO2-free air. An extra needle in the serum stopper allowed excess air to vent while injecting the water sample. The half-filled serum bottles were shaken and allowed to sit at room temperature for one hour so that the headspace gas would come into equilibrium with the gases in solution. A 3 mL sample of the headspace gas was then injected through a 1 mL sample loop, and the gas in the sample loop was then injected into a stream of CO2-free air leading to a LiCor infrared gas analyzer [Davidson and Trumbore, 1995]. The analyzer integrated the area under the absorption peak as the sample passed through the analyzer. The area was converted to CO2 concentration using a standard curve generated from injecting standards of known concentrations (2,000 and 80,000 μatm CO2 in N2) through the sample loop in the same manner as the unknown samples.
 The concentrations of CO2 in the liquid phase in the serum bottles after equilibration with the headspace was calculated as follows:
where T is the equilibration temperature (K) in the lab. The dissolved free CO2 present in the original syringe sample was then calculated based on the sum of CO2 in air and water phases in the serum bottle, divided by the volume of the water sample injected into the bottle. This approach does not consider the redistribution of other carbonate species. Specifically, as dissolved free CO2 concentrations decrease during degassing, bicarbonate is converted to additional dissolved free CO2. However, this conversion is self limiting because of increasing pH with CO2 degassing. At the low pH values observed in the study streams (typically <5), most dissolved inorganic-C occurs as CO2 rather than bicarbonate and carbonate, so the bias in our estimations is minor. The in situ stream pCO2 was then calculated by reapplying Henry's Law (equation (1)) but using in situ stream temperature.
2.3. Stream Discharge
 Stream discharge was estimated monthly on a subset of points in the three Paragominas streams from July 2004 to June 2005 by measuring cross-sectional area and flow with a current meter (General Oceanic, model 2030W, Miami, FL). This measurement was done at the most downstream sampling stations in each stream for which a good cross section could be obtained (IG54-S5, IG7-S6 and IGP-S2) and followed the methods of Rantz . Due to the shallow depth of the small stream channel at the Capitäo Poço sites, it was not possible to measure stream discharge.
 For two of the Paragominas streams, IG7 and IGP, the pCO2 ranged from about 11,000 to 25,000 μatm at the headwater, and declined to around 4,000 to 5,000 μatm at downstream positions (Figures 1a and 1b). This decline is likely due to rapid evasion in shallow streams as supersaturated stream water is exposed to the atmosphere, combined with decreasing downstream proportions of newly emergent groundwater to total streamflow. The headwater pCO2 concentrations at the CP streams in undisturbed mature forest catchments were within a similar range. There was some tendency for the seeps (CP1-S1 and CP2-S1) to have higher pCO2 than the headwater streams (CP1-S2 and CP2-S2), but not consistently so (Figure 1d). In contrast, the IG54 stream in the Paragominas region gave unusual results at the headwater sampling station (IG54-S1) and the following two stations (IG54-S2 and IG54-S3; Figure 1c). While the average pH was about 4.5 in all other streams, the headwaters of IG54 had a mean pH of 6.2 and, also unlike all other streams, pCO2 concentrations were very low (averaging 1100 and 1600 μatm at IG54-S1 and IG54-S2). Only the downstream sampling stations, IG54-S4 and IG54-S5 had pCO2 and pH values in the range of the other streams. This result suggests that the headwater of IG54 either was not fed by groundwater, or that the groundwater was unlike any other sampled in the other Paragominas and Capitão Poço streams. Figueiredo et al.  also report unusual stream solute concentrations measured at IG54-S1, and they also were forced mostly to abandon attempts to use it as a reference point for downstream samples. This stream's headwater is located near a major highway and an agricultural field where there had been significant burning. The stream then passes through an urban area. We do not know the source of this variation and conclude that the headwater of this stream is uncharacteristic of most streams of the region that drain forested areas. Enough CO2-saturated groundwater apparently entered this stream by the time it reached our fourth and fifth sampling station (IG54-S4 and IG54-S5) for the pCO2 to be elevated there. The fifth sampling station is within a large cattle ranch, with a remnant forest in upslope positions [Davidson et al., 2000].
 Although the pCO2 varied temporally in the present study, no clear seasonal pattern is apparent. The annual mean of pCO2 for each of the three Paragominas streams is 20,000, 19,000, and 11,000 μatm, respectively, for IG7-S1 and IGP-S1 headwater stations and station IG54-S4 (which we substitute here for the IG54-S1 headwater station as a minimum indicator of headwater pCO2 of this watershed, since the headwater station had unexplained undersaturation of CO2).
 The pCO2 values reported for the streams in this study are in the range of other reported values for Amazonian streams and rivers, including, 52,000 μatm in a groundwater seep in a small forest catchment of Mato Grosso [Johnson et al., 2008], 10,000 to 31,000 μatm near headwaters of agricultural catchments near Igarapé-Açu, Pará [da Rosa, 2007], 5,000 μatm in the Acre River [Sousa, 2007], and 2,500–13,000 μatm in the main stem and major tributaries of the Amazon River [Richey et al., 2002]. Although the number of studies is limited, there appears to be a trend of higher pCO2 values reported for seeps and headwaters than for large rivers, indicating expected within-stream outgassing.
 Deep soil CO2 concentrations were previously measured in the IG54 watershed, increasing from 10,000 to 30,000 ppmv at 1 m depth to 60,000 to 70,000 ppmv at 8 m depth [Davidson and Trumbore, 1995]. These values are comparable to deep soil CO2 concentrations ranging from 30,000 to 90,000 ppmv measured elsewhere in the Amazon Basin [Davidson et al., 2004; Johnson et al., 2008]. Groundwaters supporting base flow in these watersheds should be in near equilibrium with these deep soil atmosphere CO2 concentrations [Johnson et al., 2008].
 One approach to estimating outgassing of pCO2 is to assume that all of the groundwater entering the stream throughout the watershed had the same pCO2 as was measured in the headwaters, and that all of the pCO2 in groundwater is eventually evaded to the atmosphere, either in the stream or river or in the estuary once the freshwater reaches the ocean. Gas evasion studies for small streams ranging in depth from 10 to 50 cm indicate gas exchange velocities (k600, normalized to the Schmidt number of 600 for CO2 in freshwater at 20°C) in the range of 10–25 cm h−1 [Bott et al., 2006; Melching and Flores, 1999; Raymond and Cole, 2001] and turnover times for dissolved CO2 of 0.4 to 5 h, with likely turnover distances of 150 to 3000 m. Hence, most groundwater pCO2 is likely evaded within a few kilometers of its entry into the stream, but the location of evasion need not be known for this analysis. Multiplying the mean headwater pCO2 concentration by the measured mean annual discharge rate (based on monthly measurements), and dividing this flux by the area of the watershed upstream of the discharge measurement point, the amount of terrestrial C lost via groundwater CO2 to stream evasion of pCO2 would be on the order of 0.02–0.04 Mg C ha−1 of land area yr−1 (Table 1).
 Another approach to estimating an upper limit of outgassing of pCO2 is to assume that all of the groundwater entering the stream throughout the watershed had a pCO2 equivalent to deep soil CO2, and that all of the pCO2 is evaded to the atmosphere [Johnson et al., 2008]. This approach assumes that our headwater measurements of pCO2 underestimate groundwater inputs, because of outgassing that may have occurred upstream of our headwater sampling locations. This is unlikely in the case at Capitão Poço, where we sampled directly in groundwater seeps, but is likely true for the Paragominas watersheds, where limited access on private property required sampling that was sometimes as much as 0.1 to 1.0 km from headwater seeps (which varied by season). Multiplying the reported average deep forest soil CO2 concentration (70,000 ppmv) [Davidson and Trumbore, 1995] by the measured mean annual discharge rate, and dividing this flux by the area of the watershed upstream of the discharge measurement point, the maximum amount of terrestrial C lost via groundwater CO2 to streams would be on the order of 0.08–0.15 Mg C ha−1 of land area yr−1 (Table 1). This is likely an overestimate, because it does not account for lower concentrations of pCO2 in groundwater from degraded lands [Davidson and Trumbore, 1995]. Also, quickflow into streams following precipitation events initially has lower pCO2, due to its lower residence time in the soil and lower soil CO2 concentrations near the soil surface, followed by high pCO2 water [Johnson et al., 2006, 2007], but we do not have temporal resolution to consider these dynamics or their associated errors.
 Depending on which of the two approaches described above is used, the flux of C via stream outgassing of groundwater pCO2 is about 2–3 orders of magnitude lower than annual estimates of soil CO2 efflux from pastures and forests of this region, which are in the range of 10–20 Mg C ha−1 yr−1 [Davidson et al., 2000] and estimates of aboveground net primary productivity of 4–5 Mg C ha−1 yr−1 [Nepstad et al., 1994]. Hence, while the groundwater is generally supersaturated in pCO2, this is a minor pathway of C loss from the terrestrial ecosystem in these small Eastern Amazon watersheds. Although root and microbial respiration in the soil generate impressively high concentrations of soil CO2 and groundwater pCO2, the mass of these gases transported in groundwater in this region is small relative to rates of photosynthesis by terrestrial plants and efflux of CO2 as soil respiration. In contrast, Johnson et al.  estimated groundwater-derived CO2 evasion of 0.4 Mg C ha−1 yr−1 in undisturbed small forested catchments near Juruena, Mato Grosso, Brazil, which could be in the range of 2–10% of soil respiration and NPP and could be more important relative to net ecosystem exchange of C. These higher losses of groundwater-derived CO2 estimated by Johnson et al.  compared to the present study could be due to either higher rates of runoff and/or the presence of productive, undisturbed, mature forests throughout the small Juruena catchments.
 Reported rates of CO2 evasion per unit area of water surface in the Ji-Paraná River network of Rondônia, Brazil, ranged from 0.7 to 12.7 μmoles CO2 m−2 s−1 [Rasera et al., 2008], which is similar to reported soil surface efflux rates in Amazonia [Davidson et al., 2000, 2004]. However, because stream surface usually covers on the order of only 0.5–2% of the area of the watershed, when the stream evasion of CO2 is normalized to the area of land that the stream drains and that may be the source of the stream pCO2, it becomes a minor amount of C per unit land area compared to the stocks and fluxes of C in those terrestrial ecosystems. Outgassing of CO2 in white water streams has been reported to be 2–40 times larger than hydrologic export of DOC [Johnson et al., 2006; Rasera et al., 2008], but this result does not contradict the conclusions reached here. Rather, both DIC and DOC export from these upland Amazonian terrestrial ecosystems, although potentially important for in-stream C processes, are small relative to other terrestrial C stocks and flows [Markewitz et al., 2004]. In other soils and ecosystems of other regions, aquatic export, processing, and burial can be a larger percentage of terrestrially fixed C than observed in the present study [Billett et al., 2004; Cole et al., 2007; Johnson et al., 2008, Tranvik et al., 2009].
 The persistence of 4000–5000 μatm pCO2 values at most downstream sites of the present study (Figure 1) suggests a downstream CO2 source that is in quasi-equilibrium with stream outgassing rates. To properly estimate and integrate total stream CO2 outgassing fluxes would require data on the gas exchange velocities (k600) for these streams at each reach and on the total stream surface area, neither of which we have. However, using literature values in the typical range of 10–25 cm h−1 for k600 [Bott et al., 2006; Melching and Flores, 1999; Raymond and Cole, 2001], and using the downstream average pCO2 for IG7 of 4200 μatm, the average respiration-driven outgassing from the stream is likely in the range of 3 to 9 μmoles CO2 m−2 of water surface s−1, which is in the range reported for other Amazonian rivers and streams [Rasera et al., 2008]. Assuming that 0.5% to 1.5% of the watershed area is covered by streams and reservoirs, total CO2 outgassing from streams could be equivalent to 0.06–0.5 Mg C ha−1 of land area yr−1. The lower end of this range includes the estimates based on groundwater inputs of pCO2 only (Table 1), but the upper-end of this range is several times larger than any of the estimates of groundwater inputs of pCO2. Not all of this additional C would be derived from land, however, as algae and macrophytes in the numerous reservoirs respire CO2 from plant parts below the water surface, and their growth creates organic matter that eventually decomposes, contributing an unknown fraction of the stream respiration. The few carbon budgets that have been derived for Amazonian floodplain ecosystems indicate that net primary productivity within the floodplain forest and by aquatic macrophytes is sufficient to account for the majority of within-stream respiration [Melack and Engle, 2009; Melack et al., 2009]. Other allocthonous C sources include DOC in surface water and groundwater, soil carbon eroded from stream banks or from degraded pastures and croplands, and litter inputs from stream-side vegetation [Hamilton et al., 1995, Mayorga et al., 2005]. Hence, the C budgets of watersheds drained by DOC-rich blackwater rivers, of upland areas experiencing significant soil erosion, and of riparian forests may need to account for export of C via an aquatic pathway.
 If all Amazonian land (7 million km2) loses C via hydrologic export of CO2 in groundwater at similar areal rates as reported in the present study, the basin-wide flux would be on the order of 0.01 to 0.10 Pg C yr−1. Scaling up with our second approach, if all of the water discharged by the Amazon River (long-term average of 209,000 m3 s−1; estimated for the mouth of the Amazon [Molinier et al., 1996]) originated from groundwater that was supersaturated at about 70,000 μatm, the flux would be about 0.18 Pg C yr−1. However, it is unrealistic to assume that all water entering the Amazon River is supersaturated groundwater. Higher published estimates of total CO2 evasion from the Amazon River system indicate a potentially important role for autotrophic inputs and for other allocthonous sources of carbon, including streamside vegetation, soil erosion, and DOC export from organic soils.
 We thank Patrício de S. Silva and Ewerton Cunha for assistance in the field. This research was supported by grants NCC5-686 and NNG06GE88A of NASA's Terrestrial Ecology Program as part of the Large-scale Biosphere-Atmosphere (LBA) project.