Here we examine the patterns in carbon dioxide partial pressure (pCO2) measured in a number of small boreal streams (<5 km in length) in the northwestern boreal region of Québec during the ice-free season and compare these to the patterns found in a major river (Eastmain River) and in a tributary in the same region. All systems were consistently supersaturated in CO2 (range 450 to 5000 μatm) streams having both higher (mean 1850 μatm) and more variable pCO2 than that of rivers (range 550 to 800 μatm). Stream pCO2 was positively related to DOC concentration and stream segment length, both suggesting a direct influence of the surrounding landscape. Calculated stream water-air CO2 fluxes ranged from 700 to over 3000 mg C m−2 d−1, up to 2 orders of magnitude higher than those measured in large rivers and lakes of the same region. Small streams, despite their extremely reduced areal coverage (1% of the aquatic surface), accounted for 25% of the total aquatic C emissions, and the resulting areal stream fluxes were comparable to those measured in different soils or wetlands in the region.
 The current trends in global climate change and their potential links to greenhouse gas dynamics (carbon dioxide and methane) have prompted increasing efforts to quantify and predict long-term trends in greenhouse gases emissions from both human activities and natural sources. In this context, it is of particular importance to understand natural C sources and sinks, and how these might shift under scenarios of climate and environmental change. It is now widely recognized that freshwater ecosystems play a biogeochemical and ecological role that is largely disproportional to their contribution to total surface [Cole et al., 2007]. Continental waters transport, process and store large amounts of terrestrially derived materials, and far from being passive recipients or inert conduits, they can profoundly affect regional C dynamics [Battin et al., 2008; Cole et al., 2007]. Among the key features of continental systems, it has now been well established that rivers and lakes tend to be supersaturated in CO2 and thus act as sources of CO2 to the atmosphere [Cole et al., 2007; Kling et al., 2000]. A handful of studies have shown that these aquatic C fluxes may be significant relative to other components of the regional C budget [Algesten et al., 2004; Jonsson et al., 2007; Kling et al., 2000], but the overall magnitude and importance of aquatic emissions remains to be determined for the majority of landscapes.
 As second largest terrestrial biome in the world [Sun et al., 2008], and representing up to 22% of the global forest area [Dunn et al., 2007] boreal ecosystems are key components of global C cycle. Generally considered to be net carbon sinks [Chapin et al., 2000; Dunn et al., 2007], boreal forests contain up to 13% of global C biomass but store in their peat-rich soils more than 40% of global soil carbon [Dunn et al., 2007]. Aquatic ecosystems are likely to play a particularly important role in the vast boreal ecosystems, which are characterized by a high density of lakes, rivers and wetlands. The boreal biome in northern Québec, in particular, has among the highest freshwater densities in the world, and is characterized by an extremely high number of lakes of all sizes [Downing et al., 2006], and a diversity of wetlands (bogs, mires and ferns), which collectively occupy nearly 20% of the total surface. In addition, this landscape, as other boreal regions [Agren et al., 2007; Temnerud et al., 2007], has a large number of rivers, which cover up to 1% of the total surface. Most of the river surface is concentrated in either a few very large rivers (with discharge > 200 m3 s−1, each draining tens to hundreds of thousands of km2), or in a vast network of small streams, with comparatively few intermediate rivers. In this regard, one of the most outstanding features of this boreal landscape is the dominance of small streams that connect the innumerable lakes and wetlands, and the relatively low number of true headwater streams. These connecting stream segments are generally short, typically ranging from hundreds to a few thousands of meters, and traditional stream classifications and models [Battin et al., 2008] are difficult to apply, as is the use of digital geographical tools because, for the most part, they appear as 1-D representations in the GIS maps.
 The CO2 dynamics of boreal lakes has been relatively well studied [Algesten et al., 2004; Dillon and Molot, 1997; Jones et al., 2003], and from these studies it is clear that boreal lake CO2 emissions are significant at the regional scale. In contrast, the patterns in surface water CO2 in these complex boreal stream networks have been seldom explored. Previous studies of CO2 dynamics in low-order rivers in other regions have shown that these are extremely dynamic in terms of DIC [Guasch et al., 1998; Waldron et al., 2007; Worrall et al., 2005], and generally highly supersaturated in CO2 [Finlay, 2003; Hope et al., 2001; Kling et al., 1991]. Resulting from inputs of DIC groundwater and from the mineralization of terrestrial organic carbon [Battin et al., 2008], supersaturation in CO2 has also been reported for large rivers in temperate and tropical areas [Cole and Caraco, 2001; Raymond and Cole, 2001; Richey et al., 2002]. Rapid flow and associated turbulent conditions further enhance the gas exchange with the atmosphere, and thus small streams tend to emit large amounts of CO2 per unit area [Genereux and Hemond, 1992]. Although it is expected that the boreal streams also follow this pattern, direct extrapolation of results from other regions is not possible, and there are presently no models for boreal stream pCO2 that can be used to assess the importance of these systems in regional C budgets.
 Here we present a 2-year study of CO2 dynamics in a range of streams in the northern boreal region of Québec (Eastmain River basin). In this paper we explore the magnitude and temporal variation in stream pCO2, and their links to morphometric and physicochemical attributes of the streams, and we present empirical tools that allow the estimation of stream pCO2 and associated CO2 fluxes at the regional scale. For comparative purposes, we have further surveyed pCO2 in one of the major rivers of the region, the Eastmain River, as well as in another river of intermediate length and discharge (La Rivière à l'eau Claire). We use these data to assess the relative importance of small streams to the total C emissions from the combined river surfaces in this boreal landscape. We conclude that the contribution of streams to the regional aquatic C emissions is far from negligible and must therefore be incorporated into future assessments of the role of aquatic systems in regional carbon processes.
2. Material and Methods
2.1. Study and Site Description
 The study area is located in the Eastmain River region (51–52°N, 75–76°W), of northern Québec, Canada (Figure 1). This boreal part of the Canadian Shield is relatively homogeneous in geology, topography and climate. The vegetation is dominated by black spruce (Picea mariana), and to a lesser extent, white spruce (Picea glauca), American larch (Larix laricina), Jack pine (Pinus banksiana) and the balsam fir (Abies balsamea). Deciduous trees, which dominate early forest successional stages, include the quaking aspen (Populus tremuloides), the black cottonwood (Populus trichocarpa) and the paper birch (Betula papyrifera). The region has an average altitude of 250 m and is characterized by an average temperature varying between 0 and −2.5°C, with 600 to 1000 mm of annual precipitations. Freshwaters cover over 20% of the territory, with hundreds of thousands of lakes, extensive bogs and peatlands in addition to complex networks of rivers and streams. The Eastmain River (ER hereafter) is the main river in the region, and with an average discharge of 635 m3 s−1, is one of half a dozen major boreal rivers in northern Québec. It originates in north central Québec region, flows 800 km west, draining a total area of about 46,400 km2, and discharges into the James Bay. The river was dammed in 2006, to create the Eastmain-1 reservoir, which is 35 km in length, a surface area of 603 km2 and a total volume of 6.94 km3, with an annual output of 2.7 TWh. The Rivière à l'Eau Claire (REC hereafter), is an intermediate river which is approximately 28 km in length, draining an area of about 1200 km2, and with an average discharge of 15 m3 s−1 (Table 1). This river was entirely flooded as the result of the construction of the Eastmain-1 reservoir in 2006.
Table 1. General Description of Stream and Riversa
Aquatic System Type
Q (m3 s−1)
DOC (mg L−1)
DIC (mg L−1)
TP (μg L−1)
TN (mg L−1)
Physical characteristics (length, width, depth, watershed, and flow rate (Q)), and chemical characteristics (dissolved organic carbon (DOC), dissolved inorganic carbon (DIC), pH, total phosphorus (TP), total nitrogen (TN), and partial pressure carbon dioxide (pCO2)). N represents the number of samples for each variable.
Large River (Eastmain)
 The study that we present here was carried out as part of a larger project whose aim was to assess the net impact of the creation of the Eastmain-1 reservoir on the regional C budget. For this purpose, rivers, lakes and wetlands that existed within the reservoir basin were sampled prior to flooding, with the aim of developing a reference database on C dynamics in boreal aquatic ecosystems of this region (preimpoundment stage), that can be then compared to the postflood C emissions. Here we present only our results from flowing waters (streams and rivers), the lake C dynamics are presented in two companion papers [Marchand et al., 2009; Roehm et al., 2009].
 Sampling of streams was carried out between July and September, and a total of seventy streams of different sizes within the Eastmain basin (Figure 1) were visited for this study: 50 during 2005 and an additional 20 in 2006. Of the streams sampled in 2005, 20 were visited only once, the rest were sampled 2 or 3 times over the period from July to September, so as to assess seasonal variability. In 2006, 15 streams were visited only once and 5 were sampled twice. Because of the remoteness of the region, the sampling areas were reached either by helicopter or by hydroplane and the actual sampling site then accessed by foot. Once on site, stream depth and width were measured or estimated (depending on the size of the stream) and current velocity was measured with a Flow Tracker ADV (Acoustic Doppler Velocimeter); these three variables were then used to calculate water discharge. The REC river was also sampled three times during 2005 at a 3 sites, whereas ER was sampled at 7 different points along a 110-km stretch, three times during 2005 (Figure 1).
 Temperature, conductivity, dissolved oxygen and pH were measured in situ using a YSI combination probe. Water samples were collected at approximately 5 cm depth in a 2-L Nalgene bottle, and used for both CO2 measurements in situ, and for additional chemical analyses in the laboratory. For the determination of dissolved CO2 (pCO2), a 100-mL syringe, fitted with a two-way valve, was completely filled with water from the sampling bottle, half the volume was then discarded so as to create a head space and the sampling valve closed. The syringe was then gently shaken so as to equilibrate the gases in the water and air. The headspace was then sampled using a 10-mL syringe connected to the second valve, and the gases directly injected into a PPSystems (Model EGM-4) infrared gas monitor. Sample pCO2 was calculated to take into account the CO2 movements between the water and gas phase during equilibration.
2.3. CO2 Fluxes
 CO2 fluxes (mg C m−2 d−1) to the atmosphere were calculated on the basis of the interfacial mass transfer mechanism from water to air expressed mathematically as first-order equation dependent on the difference between the concentrations in the water and the air:
is the CO2 exchange velocity coefficient, in m d−1.
is Henry's constant corrected for temperature, kh = 10−(1.11+0.016*T−0.00007*T^2), with temperature (T) in °C.
is partial pressure CO2 measure in the water, in μatm.
is partial pressure CO2 measured in the air, in μatm.
 Fluxes from lakes, rivers and streams were all calculated from equation (1) but the gas transfer velocity coefficient (k) was chosen separately for each aquatic component according to their specific characteristics. Because the water velocity and specific flow regime, k for the rivers was assumed to be 2 m d−1. This value is higher than the proposed k of between 0.7 to 0.8 m d−1 of Cole and Caraco  for large rivers but consistent with a larger range of between 0.7 and 3 m d−1 reported by various authors for large rivers and estuaries, as reviewed by Raymond and Cole . This assumed k value of 2 m d−1 is conservative for these large riverine systems, which have high turbulent energy and water velocities ranging between 8 and 12 m s−1 in the Eastmain River, and over 1 m s−1 for the Rivière à l'Eau Claire. For streams we have assumed a k of 4 m s−1, in line with the value of 3.8 m s−1 used by Jonsson et al.  for boreal streams in Sweden. Finally, we assumed an exchange coefficient for lakes of 0.8 m d−1, derived from the gas transfer coefficient model [Cole and Caraco, 1998] for an average wind speed of 3 m s−1 as recorded during our sampling. Ambient air pCO2air concentration was measured at each sampling location, and ranged between 326 and 426 μatm. To simplify the calculation, we used a mean value of 380 μatm, slightly higher than the value of 360 μatm used by Jones et al.  for a pCO2 survey in the United States. However, a decrease in atmospheric pCO2 from 380 to 360 μatm would lower the calculated CO2 fluxes with less than 3%. Total emissions to the atmosphere were calculated for the ice-free period only (from May to October), and clearly underestimate the annual emissions because they do not include winter CO2 dynamics and under-ice CO2 accumulation, which have been shown to be significant in boreal landscapes [Jonsson et al., 2007].
2.4. Laboratory Analyses
 Total phosphorus was determined spectrophotometrically following potassium persulphate digestion. Total nitrogen was determined as nitrates following alkaline persulphate digestion and measured on an Alpkem Flow solution IV autoanalyzer. Dissolved inorganic and organic carbon (DIC and DOC) concentration were measured in 0.2 μm-filtered water samples in an OI-1010 Total Carbon Analyzer using wet persulphate oxidation.
2.5. Stream and Watershed Morphology
 The length of each stream segment, the width of the larger rivers, and the watershed area above the sampling point were obtained from digitized maps (scale 1:50,000) using the hydrological extensions in ArcMap GIS 9.2.
 The fluvial network in this area is composed mostly of two types of streams: (1) streams connecting lakes and wetlands and (2) outflows of lakes and wetlands that merge into other streams. Together, these different categories of stream form complex branching patterns, with very few true headwater streams. For the purpose of this study, we have classified the streams in terms of total length in the case of connecting streams, or the length of the smallest individual stream segment that can be identified using GIS, in the case of branching streams. The streams sampled ranged from less than 20 m to a maximum of approximately 5 km in length, which corresponds approximately to the upper limit of the length of individual stream segments in this region (Table 1). Stream width ranged from 0.5 to 20 m, and there was a significant positive relationship between segment length and width:
with both stream width and length in meters (m).
 This relationship allows the estimation of total stream surface from GIS-derived data of stream length, as we show in sections below. There was a 4-order of magnitude range in water discharge, from 0.0003 to 1.25 m s−1 (Table 1). The sampled streams had a wide range in DOC, (4 to 25 mg L−1), pH (4 to 6), and nutrient concentrations (TP from 4 to 29 μg L−1) (Table 1). The two larger rivers that we sampled for comparative purposes were much less variable in terms of water chemistry, with mean pH of and DOC concentrations of 6.0 and 6.1, and 8.0 and 8.1 mg L−1, for ER and REC, respectively (Table 1).
3.2. Magnitude and Variability of Stream and River pCO2
 All systems sampled were consistently supersaturated in CO2, but streams had both the largest range and also the highest values, ranging from 500 to over 5400 μatm (Table 1). Some of this range resulted from temporal variability within a given site. The average coefficient of variation in surface pCO2 for the 17 streams for which we had 3 samples (July, August, and September) was 29%. Some of this temporal variability within streams appeared related to changes in discharge, since the highest CO2 concentrations tended to occur at times of low discharge. In contrast, both ER and REC were much less variable in terms of CO2 than the streams (Table 1). Figure 2 shows the surface water pCO2 along the (Figure 2a) 110-km stretch of the Eastmain River in three different sampling dates during 2005 (Figure 2a) and (Figure 2b) 10-km stretch of the Rivière à l'Eau Claire. There was a slight decreasing trend in surface pCO2 downstream Eastmain River, but except for one site in August, all the samples were within 17% of the grand mean of 610 μatm. A sharper decreasing trend was measured downstream REC and likewise, the pCO2 for the two sampling dates were within less than 17% of the grand mean of 845 μatm. Overall, there was a trend of declining pCO2 from streams (overall average 1858 μatm) to the intermediate (average 845 μatm) and largest (610 μatm) rivers (Figure 3a).
3.3. Patterns in Stream pCO2
 Surface water CO2 was related to both landscape and stream water chemical variables (Figure 4). Among the physical variables, there was a negative correlation between stream pCO2 and both the area of the proximate watershed (Figure 4a) and stream discharge (Figure 4b):
 There was a significant positive relationship between stream length and pCO2 (Figure 4c):
 These relationships indicate that pCO2 tends to increase with stream segment length, suggesting a cumulative effect of the surrounding basin, but that this influence is modulated by the amount of water discharged, and by the relative size of the basin directly influencing the loading to the stream.
 Stream pCO2 was also related to several chemical properties of the water, in particular a positive relationship with both DOC and total N concentrations (Figures 4d and 4e):
In addition, there was a strong negative relationship with pH (Figure 4f):
Additionally, we developed two multiple regression models which increased the coefficient of correlation (r2) up to 0.63:
 Surface water pCO2 was relatively constant in the Eastmain River, both spatially and temporally, at least along the 110-km stretch that we sampled. These large rivers integrate numerous watershed processes, and the extremely large discharge probably swamps any features that could affect CO2 locally, thus maintaining relatively constant surface CO2 concentration. The pCO2 in the intermediate river that we studied was somewhat more spatial variable than that in the Eastmain River (700 to 1170 μatm), but nevertheless all samples were within 25% of the grand mean. This suggests that the CO2 emissions from these large and intermediate boreal rivers can be scaled up regionally quiet precisely by combining mean pCO2 values taken from a few representative sites and GIS-based estimates of total river surface. In contrast, CO2 concentrations in small streams were more variable, temporally within a given site and especially across a range of stream lengths and discharge. Others have reported large variations in stream pCO2 at very different temporal scales, from diurnal [Dawson et al., 2001; Guasch et al., 1998], to seasonal [Dawson et al., 2002; Waldron et al., 2007] and interannual [Dawson et al., 2002]. Larger variability of pCO2 values in stream compared to intermediate and large rivers may reflect higher dynamics in term of discharge, disturbance or connection with the terrestrial ecosystems of the small streams, whereas the rivers, carrying the combined signatures of upstream ecosystems, are well buffered and therefore, have lower pCO2 range. This implies that streams CO2 concentrations and fluxes are generally more sensitive to hydrological changes (i.e., discharge) and material loading compared to rivers.
 In this regard, we have shown that stream pCO2 relates to water chemistry, most notably to DOC, and to certain stream morphometric variables. We found a strong negative relationship with pH, which is to be expected in these poorly buffered systems where high concentrations of CO2 can strongly influence water pH [Neal et al., 1998; Waldron et al., 2007]. Stream pCO2 was also strongly positively related to both DOC and TN. The latter is not surprising since in these waters TN is largely in the form of dissolved organic N, and is therefore expected to follow a similar empirical pattern as DOC. Neal et al.  also reported a strong positive correlation between DOC (and with several forms of N) and pCO2 in lowland streams in northern United Kingdom, and this type of relationship has been previously reported for lakes, both in this same region [Roehm et al., 2009], and elsewhere [Jones et al., 2003; Prairie et al., 2002]. In the case of lakes, this relationship has been interpreted as indicating both in-system decomposition of DOC loaded from the watershed, as well as reflecting overall C loading into the system, including groundwater injection of soil CO2, and the same probably applies for streams as well [Battin et al., 2008]. Interestingly, our results show that the pCO2 - DOC relationships reposted here for streams differs markedly from that reported for the lakes in the region by [Roehm et al., 2009]. At low DOC concentrations, boreal lakes and streams tend to have very similar average pCO2, whereas with increasing DOC, pCO2 in streams tends to increase much faster than in lakes, so that at a DOC concentration of 10 mg L−1, for example, the average pCO2 is approximately 80% higher in streams than in lakes (1200 versus 740 μatm, respectively). The difference in the pCO2 versus DOC relationships in streams and lakes would suggest that groundwater DIC inputs dominate over DOC mineralization in streams, whereas mineralization plays a larger role in lake CO2 dynamics.
 As we pointed out earlier, the overwhelming majority of streams originate as outflows of lakes (and sometimes of wetlands), so that the chemistry of the stream network is constantly being reset as the water flows through standing waters. Evidence of this is the fact that the intercept of the stream pCO2 versus length regression is 1130 μatm, which is remarkably similar to the overall average pCO2 that we measured for the lakes in the region (1260 μatm) [Roehm et al., 2009]. Thus, the shortest stream segments (generally < 200 m) tend to reflect mostly lake conditions, but as the length of the segments increases, so does the influence of the additional C loading from the surrounding watershed, which explains the positive relationship between pCO2 and segment length. A comparable pattern of increasing DIC downstream was reported by Kling et al.  for connecting streams in Alaska. This pattern does not apply to all landscapes, however. For example, Dawson et al.  found that stream pCO2 in a Scottish moorland catchment declined downstream from the source, suggesting that in this landscape degassing of CO2 that is loaded in the headwaters predominates over the emission of CO2 loaded during transit. Temnerud et al.  reported a similarly complex pattern in distribution in TOC in a boreal stream network in northern Sweden. There was also weak but significant negative relationship between stream discharge and stream pCO2, which suggests that the influence of C loading from the watershed is mediated by stream physical processes. It is likely that this relationship reflects the connections that exist between water velocity, flow regime and the vertical mass transport. At any given external CO2 input, a stream characterized by high flow (Q) would tend to reduce its CO2 pool by increased emission associated with high vertical mass transport typical of turbulent flow regimes [Genereux and Hemond, 1992].
 Perhaps the most useful of these relationships in term of upscaling stream pCO2 and estimating overall stream CO2 fluxes at the landscape level is the regression between stream segment length and pCO2, because the former is a landscape feature that is relatively easily obtained from digitized maps, whereas stream chemistry or discharge cannot be easily estimated for the ensemble of streams. However, the relationships with DOC, TN and discharge that we present here may eventually be used to assess the direction of changes in boreal stream CO2 dynamics under scenarios of major shifts in hydrology and material loading that are predicted for northern latitude landscapes [Schindler, 1998], although these models still lack both sufficient temporal resolution and generality to be used at the scale of the entire Québec boreal landscape.
 These models that we present for small streams, particularly those based in stream length and discharge, clearly cannot be extrapolated much beyond the range of conditions from which they were developed. For example, we show here that pCO2 tends to increase with segment length, but clearly this relationship does not apply to the larger rivers, which we have shown, typically have much lower CO2 concentrations that most streams. Where, along the continuum of river sizes, do these relationships break down? We cannot answer this question because we have very few rivers longer than a few thousands of meters, but our sampling distribution actually reflects the distribution of river sizes in the landscape. Our GIS-based analysis of river distribution in these regions shows that between the small stream network, and the major rivers there are comparatively few rivers of intermediate size and discharge. For instance, a GIS-based survey of a large boreal area of 50,000 km2 in the Eastmain region revealed that intermediate rivers (between 10 and 90 km in total length) account for less than 3% of aquatic coverage, the remainder comprised by lakes (>90% of aquatic surface), a few large rivers (>6% of total aquatic), and the complex network of small streams (<0.1% of aquatic area) (C. Teodoru et al., Integrating stream, river and lake CO2 fluxes into a new emergent propriety of boreal landscape, manuscript in preparation, 2009).
4.2. Scaling up Stream and River CO2 Emissions at the Landscape Level
 The data and models presented above can be combined to estimate the total CO2 emissions from streams and rivers in this boreal region, so as to assess the importance of these systems relative to each other and to other components of the regional C budget. We selected a block of approximately 970 km2 (square in Figure 1) that corresponds to a GIS unit (033B04), and which covers most of the sampling areas. We determined the length of all the individual stream segments contained within this entire block of territory, from digitized maps using GIS. Figure 5 shows the frequency distribution of segment lengths in this unit of landscape, as well as the contribution of three stream length categories to total stream surface in this area. There is an overwhelming dominance of short streams in this region, but the total stream surface area is dominated by both intermediate and longer stream segments.
 For each stream segment we estimated the average pCO2 using the relationship between stream length and pCO2 presented above (equation (5)). For streams shorter than 200 m, pCO2 was fixed at 1130 μatm which represents the intercept value of the stream pCO2 versus length relationship. The areal CO2 flux (mg C m−2 d−1) for each stream segment was then estimated from pCO2 as described in the Methods section; the area of individual stream segments was obtained by estimating the width from the measured length using equation (2); total CO2 flux per stream was calculated from the average areal CO2 flux times the total segment area. In the case of rivers, the area could be determined directly using GIS. We used the average pCO2 measured in REC and ER (Table 1) to determine average areal fluxes for the two categories of rivers, and multiplied the resulting flux by the total area occupied by each river type to estimate total river flux. For comparative purposes, we further estimated the total number of lakes in the block of territory, and their respective surface areas. We used the relationship between lake surface area – lake pCO2 developed by Roehm et al.  for the lakes in this region, to determine the average pCO2 of each lake; CO2 fluxes were then determined from pCO2 for each lake using equation (1), as described in the Methods section, and then summed to derive total lake CO2 flux.
 The results of this upscaling exercise are presented in Table 2. Aquatic ecosystems (stream, rivers and lakes, wetlands were not considered) covered a total of 119 km2, or 12.3% of the total area of the block, comprised by 790 stream segments, 2 segments of the Eastmain River, one intermediate river, and 1029 lakes, representing 1.1%, 30.9%, 0.4% and 67.6% of the total aquatic surface, respectively (Table 2). The mean area-weighted pCO2 was highest in streams (2234 μatm), and lowest in ER and lakes (610 and 608 μatm, respectively). The resulting average areal CO2 fluxes differed by more than 2 orders of magnitude between systems, from over 3100 mg C m2 d−1 in streams to around 70 mg C m2 d−1 in lakes (Table 2). The total CO2 emissions for a 180-day (ice-free) period from the combined aquatic ecosystems amounted to approximately 3200 tons of C, of which 24% were emitted by the stream network, 40% by the Eastmain River alone, 35% by lakes, and around 1% by the intermediate REC river.
Table 2. Analysis of Total Aquatic Fluxes in a Block of Boreal Territorya
Number of Individual Components
Aquatic System Type
Aquatic Area (km2)
Total Block Area (%)
Aquatic System (%)
C Flux (mg C m2 d−1)
Total C (tons)
Total C (%)
BNDT 033B04, 967.7 km2, on the basis of a combination of GIS analyses and empirical models of stream and river (this study) and lake [Roehm et al., 2009] surface water pCO2. Described are the number of systems in each category, their total area, and their contribution to total block area (percent of Total Block Area) and to total aquatic system (percent Aquatic System); average area-weighted surface CO2 concentrations (pCO2), the estimated average area-weighted CO2 flux (C Flux), and the total CO2 emission (Total C) for each system type for a 180-day ice-free period.
 The main outcome of this exercise is the realization that whereas streams occupy a very small portion of the boreal landscape and contribute very little to the total aquatic surface, total stream CO2 fluxes appear to be of the same magnitude as the estimated total lake and river emissions in this landscape. There is no doubt that the distribution of both surface area and CO2 fluxes among aquatic components will vary across the boreal landscape, but the block of territory that we analyzed for this exercise is representative of a much larger boreal area. If anything, the relative contribution of streams to the total aquatic C emissions is likely to be larger in other areas, for example, that do not have large rivers flowing by, as the one we analyzed for this example. One of our main conclusions is that the boreal stream network is an important component of the regional C budget, and that these small streams play a role in the C budget that is largely disproportionate to their surface area. Of all the boreal aquatic components, streams are probably the one most closely coupled to the surrounding landscape, in terms of chemistry and material loading [Agren et al., 2007; Worrall et al., 2005] and therefore the one that is more sensitive to changes in terrestrial ecosystem processes. In particular, changes in soil C dynamics and hydrology are likely to profoundly affect stream CO2 fluxes, and conversely, the stream network acts as an integrator and reflection of the surrounding landscape [Jenerette and Lal, 2005]. We have also shown that the large boreal rivers are also key players in the C budget, at least locally, because although they are on average less saturated in CO2 than small streams, they cover a much larger total surface area and thus generate large C emissions. We suggest that the carbon budget of boreal regions cannot be effectively quantified or predicted unless the major aquatic ecosystems, including streams networks, are included, because these are sites of extremely high biogeochemical activity, and an integral component of the terrestrial C cycle. This observation may also be significant to terrestrial NEP measurements derived from eddy-covariance towers, as they are placed in upland areas with fewer stream segments. Streams have also been often ignored in regional C budgets, but our results show that these systems are hot spots of CO2 emission and that they play a role in boreal C dynamics that is much larger than what their small size would suggest.
 This research was supported financially and logistically by Hydro-Québec, through the Hydro-Québec/UQÀM Eastmain-1 Research project. We especially thank A. Tremblay (Hydro-Québec) for his continuing support to this project. We would also like to thank C. Roehm, A. St. Pierre, M. Camire, A. Blain, D. Marchand, M. Genest, and S. Barette for field assistance and C. Beauchemin for laboratory analysis.