Spatiotemporal variations of pCO2 and δ13C-DIC in subarctic streams in northern Sweden

Authors


Abstract

[1] Current predictions of climate-related changes in high-latitude environments suggest major effects on the C export in streams and rivers. To what extent this will also affect the stream water CO2 concentrations is poorly understood. In this study we examined the spatiotemporal variation in partial pressure of CO2 (pCO2) and in stable isotopic composition of dissolved inorganic carbon (δ13C-DIC) in subarctic streams in northern Sweden. The selected watersheds are characterized by large variations in high-latitude boreal forest and tundra and differences in bedrock. We found that all streams generally were supersaturated in pCO2 with an average concentration of 850 µatm. The variability in pCO2 across streams was poorly related to vegetation cover, and carbonaceous bedrock influence was manifested in high DIC concentrations but not reflected in either stream pCO2 or δ13C-DIC. Stream water pCO2 values were highest during winter base flow when we also observed the lowest δ13C-DIC values, and this pattern is interpreted as a high contribution from CO2 from soil respiration. Summer base flow δ13C-DIC values probably are more affected by in situ stream processes such as aquatic production/respiration and degassing. A challenge for further studies will be to disentangle the origin of stream water CO2 and quantify their relative importance.

1 Introduction

[2] During the past decades, increased waterborne losses of dissolved carbon (C) from terrestrial sources have been manifested at northern latitudes [Frey and Smith, 2005; Dutta et al., 2006; Frey and McClelland, 2009] caused by the polar amplification of climate change and changes in precipitation patterns. Thawing permafrost [Osterkamp, 2007], changes in hydrology [Peterson et al., 2002; Déry and Wood, 2005; Lyon et al., 2010], and shifts in vegetation cover [Olefeldt and Roulet, 2012] can all potentially affect these waterborne C losses. Part of this stream water C occurs as CO2 that can be emitted to the atmosphere and contribute to the land-atmosphere C exchange [Kling et al., 1991; Jones and Mulholland, 1998; Hope et al., 2001; Teodoru et al., 2009; Koprivnjak et al., 2010]. To what extent the CO2 part of the stream C also is affected by the climate-related effects on C fluxes is currently not clear. A key issue in understanding landscape C budgets is thus to achieve a better understanding on how landscape features affect aquatic CO2 and how these influences may change in the future.

[3] The few studies carried out suggest that streams at high latitudes are generally supersaturated with CO2 [Kling et al., 1991] and CO2 can account for a substantial part of the dissolved inorganic C (DIC) in streams. The contribution of CO2 to DIC in about 12,300 mainly boreal streams in Sweden was estimated to be about 33% (percent of median concentrations) [Humborg et al., 2010]. In arctic and subarctic rivers, DIC is a major component of stream C and can account for as much as 70% of the dissolved C flux [Striegl et al., 2007]. In fact, most large Siberian and Canadian rivers are richer in DIC than total organic carbon (TOC). On average, Siberian rivers have some 7.2 mg L−1 C as TOC and about 9 mg L−1 C as DIC, and Canadian rivers as the Mackenzie have even higher DIC concentrations(>20 mg L−1 C) [Gordeev et al., 1996]. At a landscape level, degassing of streams and lakes can contribute to a significant part of the watershed CO2 loss [Kling et al., 1991; Cole et al., 2007; Humborg et al., 2010; Karlsson et al., 2010; Butman and Raymond, 2011], and future changes in DIC and its components may thus have large implication for CO2 exchange with the atmosphere. There are already strong indications of increased export of DIC from high-latitude regions, mainly due to an increase in the importance of subsurface flow pathways [Striegl et al., 2005; Walvoord and Striegl, 2007; Lyon et al., 2010]. To what extent this also will affect the pCO2 in streams is, however, less clear.

[4] The supersaturation of CO2 in streams can be attributed to both the aquatic respiration and photooxidation of terrestrial OC [Graneli et al., 1996; Karlsson et al., 2007] and to the inflow of groundwater that is supersaturated in CO2 [Worrall et al., 2005]. The latter is mediated via respiration from heterotrophic activity and root respiration elevating soil pCO2. Part of the soil CO2 that gets dissolved will be consumed via carbonic acid (H2CO3)-mediated weathering and converted to dissolved carbonate ions (HCO3, CO32−) which are all components of DIC. Thus, weathering can be regarded as a long-term sink for atmospheric carbon [Berner, 1992]. Studies from boreal and temperate regions show that groundwater dominated influxes of DIC to streams seem to dominate smaller streams [Finlay, 2003; Worrall et al., 2005] and have higher DIC and/or pCO2 values as compared to larger, higher-ordered rivers [Finlay, 2003; Teodoru et al., 2009; Humborg et al., 2010]. Kling et al. [1991] reported values from four rivers in northern Alaska with summer pCO2 values ranging from 369 to 845 µatm putting these in the same range as those reported from larger rivers in a boreal region [Teodoru et al., 2009].

[5] Although there is a lack of stream data on pCO2 from tundra-dominated environments, a likely expectation is that stream pCO2 values are lower in these environments as compared to those in streams in boreal regions. This is because the factors that are potentially affecting soil pCO2 are lower in tundra-dominated environments as compared to boreal forest ecosystems. Such factors include lower plant productivity, shorter growing season, and decreased soil heterotrophic activity; all of which are contributing to the soil pCO2. Consumption of soil CO2 may, however, also be less due to relatively lower weathering rates which are positively related to temperature [Lasaga, 1998]. This could thus counteract effects of lower soil CO2 production, because less C is consumed during weathering. In situ processes in streams may also affect stream pCO2 but can be assumed to be of less importance [Finlay, 2003]. Climate-related changes in both terrestrial and aquatic ecosystems may thus have a combined influence on stream water CO2 concentrations.

[6] The fringe between tundra ecosystems and ecosystems dominated by boreal forests are areas where changes in climate, and related ecosystem effects, have been observed in the last decades [Wilmking et al., 2004; Sturm et al., 2005; Tape et al., 2006; Callaghan et al., 2010], and it is likely that an environment where changes in stream pCO2 may be observed. The aim of this study was to quantify gaseous and dissolved DIC in subarctic streams and to characterize environmental variables controlling the potential sink/source patterns of atmospheric carbon in a transitional environment of tundra and boreal forest. We selected a set of 49 streams in a subarctic landscape in the vicinity of Abisko in northern Sweden encompassing watersheds with a varying degree of forest and tundra cover and differences in bedrock. Measured stream water pCO2 and water chemistry (a strong proxy for weathering influence) were used to test the influence of spatial variations in vegetation cover and bedrock properties. In addition, temporal variations in six streams were monitored in detail during a hydrological year to elucidate the influence of water flow paths and seasonality on pCO2. To better relate the DIC observed in the streams to terrestrial and other sources (biogenic, atmospheric, and geological), we also measured 13C-DIC in all streams as such isotope signature likely provides some insights on the origins of DIC [Hitchon and Krouse, 1972; Amiotte-Suchet et al., 1999; Finlay, 2003].

2 Material and Methods

2.1 Study Sites

[7] We selected 49 streams across a subarctic landscape gradient in northern Sweden (68°21′36″N, 18°46′48″E) located between the town of Kiruna (92 km east of Abisko) and the Norwegian border 30 km west of Abisko (Figure 1). The mean annual temperature in Abisko was about −1°C (1961–1990 [Åkerman and Johansson, 2008]), but the mean annual temperature during the recent decade has been above 0°C [Callaghan et al., 2010]. Precipitation is around 300 mm yr−1 in Abisko but increases eastward and is about 424 mm yr−1 at Bergfors located 40 km east of Abisko [Åkerman and Johansson, 2008]. The precipitation change westward is more pronounced and is about 1000 mm at the Norwegian border [Alexandersson et al., 1991]. More than 50% of the precipitation falls as snow [Kohler et al., 2006]. The watersheds and their streams are all draining into the upper reaches of the Torne river system. The streams encompass a range of different vegetation covers and stream lengths. A set of 15 streams were selected above the tree line with stream lengths ranging from 0.6 to 31 km (median 5.4 km). For the remainder of this study, we refer to these as the “tundra” streams. The remaining 34 streams ranged in stream length from 1 to 531 km (median 4.9 km) and had watersheds that differed in their forest cover composition. The forest cover ranged from 1 to 58% with the remaining nonforested vegetation being tundra. As such, these watersheds and their streams are referred as “mixed” streams for the remainder of this study to separate them from clearly nonforested tundra streams. Some additional background information is given in Tables 1 and 2 and Table A1 in the Supporting Information. The bedrock in the region is dominated by mica schist with segments of marble west and north of Abisko, whereas schist and quartzite are predominant eastward. The vegetation is dominated by deciduous forest at lower altitudes (Betula pubescens Ehrh. spp. czerepanovii) and dwarf shrub heath tundra at altitudes above approximately 550 m.

Figure 1.

Map showing streams and watersheds for six numbered streams that were used for the study on seasonal patterns. The larger lake is Torneträsk, and the star denotes Abisko.

Table 1. Background Data for the Streams and Watersheds in This Study
 StreamWatershedForestaDOC 
Length (km)Area (km2)(%)(µmol L−1)pH
  1. aOnly watersheds with forest are included.

Average22.321.2161207.51
Median5.45.0121077.4
Max565.3530.8583408.4
Min0.30.61227.0
Table 2. Common Names, Locations, and Watershed Characteristics for the Six More Intensely Monitored Streams
 Watershed Characteristics
Stream No.Common NameLat, LongAreaAverage ElevationStream LengthaSlopeForest Cover
   (km2)(MASL)(km)(deg)(%)
  1. aStream lengths are estimated as the length of all connected running waters upstream of each sampling point. MASL, meters above sea level.

1Homojokka68.00°,19.85°16575125.639
2 68.23°,19.61°565149.225
3Pessijokka68.30°,19.23°10096793103
4 68.32°,19.17°675659.518
5Miellajokka68.36°,18.79°529554714.511
6Abiskojokka68.34°,18.96°56595653113.19

2.2 Stream Water Sampling

[8] Grab samples for the study of spatial variation across the streams were taken in mid-September 2008, whereas six more intensely studied streams were sampled from mid-April 2008 to the end of April 2009. In these more intensely studied streams, samples were taken more intensely during the spring freshet (2 to 3 times per week) and thereafter weekly. From December to April, only monthly samples were taken. Samples for DOC analyses were filtered through a 0.45 µm Millex HA filter, Millipore, whereas a 0.22 µm Whatman Nuclepore filter was used for cation and anion samples. Samples for DOC analyses were thereafter acidified with 1.0 M hydrochloric acid and samples for cation analyses acidified with concentrated nitric acid. Samples for DOC, cation, and alkalinity analyses were stored in a cooler, whereas samples for anions and alkalinity were kept frozen until further analyses. Water for analysis of CO2 concentration was collected in 60 mL plastic syringes in the field. Three syringes with 30 mL of water and no air space were collected for each stream at each sampling occasion. The samples were analyzed within 4 h after sampling.

[9] Daily stream flow estimates were based on monitored pressure changes in the stream and air at each sampling point using Hobo water level data loggers and empirical rating curves for streams 1 to 5 (Figure 1). For these streams, daily stream flow was thus monitored from 1 June 2008 through 1 October 2008. The daily stream flows for the remainder of the year were scaled from Abiskojokka (stream 6) to provide a first-order approximation of actual flows. This allowed us to extend from the limited period of direct observations at streams 1 to 5 to year-round estimates. For Abiskojokken, daily stream flows for the entire year were measured and are available through the Swedish Meteorological and Hydrological Institute (Gauge ID 957).

2.3 δ13C-DIC Analyses and Calculations

[10] For determination of δ13C-DIC isotope compositions in stream water, filled 1-L samples were taken as grab samples and transported back to the laboratory. An aliquot of 4 mL stream water was taken from each 1-L sample with a syringe and was injected into a 12 mL septum-sealed glass vials (Labco Limited), which had been flushed with N2 gas for 3 min prior to injection. The vials were prepared with 100 µl of 99% H3PO4 to act as a preservative [Taipale and Sonninen, 2009] and to transform all the HCO3 and CO32− ions to CO2(g). Transfer by needle injection through the septa avoided both air entering the vial and CO2 leaving it. The samples were stored under cold (+4°C) and dark conditions until analysis within four months from sampling. The stable carbon isotopic compositions were determined using a Gasbench II extraction line coupled to a Finnigan MAT 252 mass spectrometer. Results are given as per mil deviations from the standard (PDB) and denoted δ13C, where R is the ratio of 13C/12C:

display math(1)

[11] From repeated measurements of standards, the reproducibility was calculated to be better than 0.1‰ for δ13C. We also tested if the sample handling using 1-L plastic bottles could cause any degassing and fractionation during transport and handling at the laboratory as compared to injecting stream water into the vials in the field. We used stream samples from three streams and three replicates and found no difference between the two sample handling procedures; the sample variation for the replicates was 0.1‰ for each stream.

2.4 Analyses

[12] DOC was measured on a TOC analyzer (Shimadzu TOC-VcPH total organic carbon analyzer). Alkalinity and pH were measured using a Mettler Toledo automated titration system using a Metrohm Aquatrode Plus (6.0257/000) pH electrode (Metrohm AG, Switzerland). The samples were titrated to pH 4.0 with 0.1 M HCl and then back-titrated to 5.6 using 0.1 M NaOH. Alkalinity was calculated from the difference in the amount of NaOH and HCl used and the sample volume. Analysis of CO2 concentration was done using a headspace equilibration technique [Åberg et al., 2007]. A 30 mL gas headspace was created, after which the syringes were shaken vigorously for 1 min and then left standing for 1 min for equilibration of the gas and water phases. The concentrations of CO2 in the headspace were analyzed using an infrared gas analyzer (EGM-4; PP-Systems Inc.). The pCO2 was calculated from the mean headspace pCO2 by using Henry's law according to Weiss [1974], and knowing the water temperature, the volume relationship between liquid and gas phases, the atmospheric pressure, and the amount of CO2 added with headspace, air DIC was calculated from alkalinity and pCO2 values using PHREEQCI [Parkhurst and Appelo, 1999]. Cations were analyzed on an ICP-OES (Varian Vista Ax.). Accuracy and precision was better than 4% based on certified standard measurements. Anions were analyzed on an ion chromatography system, Dionex DX-300, equipped with an AS14 column using electrical suppression (Dionex Corp., Sunnyvale, CA).

[13] We also tested the ability to use predictive models for stream water pCO2 based on boreal forest stream networks dominated by coniferous forests using stream water DOC concentrations as the independent variable [Teodoru et al., 2009]:

display math(2)

[14] All geographical estimations and analyses were made using ArcGIS 9.2 (ESRI). Stream lengths were estimated as the length of all connected running waters upstream of each sampling point. Geographical raw data were downloaded from the digital database at the Swedish Mapping, Cadastral and Land Registration Authority, Sweden.

2.5 Statistics

[15] Comparisons for differences between mixed and tundra streams were tested using Student's t test. Relationships between landscape variables, stream water chemistry, and pCO2, DIC, and δ13C-DIC were explored using linear regression. Relationships between stream water and watershed properties were also characterized by using principal component analysis (PCA) after standardizing to unit variance. Resulting factor scores of the first and second principal components (PC1 and PC2) were tested by one-way analyses of variance (ANOVA) followed by Tukey's multiple comparison to test for differences in seasonal patterns. Statistical analyses were performed using SPSS (SPSS 12.0.1 software, Chicago, IL, USA). Significant differences refer to the p ≤ 0.05 level unless otherwise stated.

3 Results

3.1 Spatial Variations in Stream Water pCO2 and DIC Across an Arctic Landscape

[16] The average value of stream water pCO2 in tundra and mixed watersheds was about 850 µatm, and we found no significant difference in pCO2 between both types of watersheds (Table 3). The stream water pCO2 values ranged between 383 and 3590 µatm across the landscape gradient and were mostly supersaturated with respect to atmospheric CO2 (380 µatm, Figure 2). The highest values were found in smaller streams with a stream length less than 5 km, whereas lower values (<1000 µatm) could be found across the whole range of stream lengths in the study, i.e., from 600 m to 530 km (Figure 2). There was a weak negative linear relationship between the ln-transformed pCO2 and stream length (lnpCO2 (µatm) = −0.18 × ln stream length (km) + 6.90; r2 = 0.19, p = 0.001). Variations in pCO2 were best explained by DOC (lnpCO2 (µatm) = 0.0035 × DOC (µmol L−1) + 6.09; r2 = 0.34, p < 0.001), the two highest values treated as outliers (Figure 2). Using the model by Teodoru et al. [2009], equation (2) consistently underestimated pCO2 in our streams.

Table 3. Comparison of Streams With Only Tundra Catchment Vegetation and Streams With a Partial Catchment Forest Cover (September Sampling)
Catchment TypeNo. of SamplespCO2DICDOCδ13C-DIC
(ppm)(µmol L−1)(µmol L−1)(‰)
Tundra15876301121−5.4
Mixed34845640124−6.5
t test p = 0.900p = 0.003p = 0.974p = 0.343
Figure 2.

The relationship between stream water pCO2 and a) stream length and b) dissolved organic carbon (DOC). The lower panels show the relationship between the molar Ca/Na ratios and c) dissolved inorganic carbon (DIC) and d) stream water pCO2.

[17] We found no significant relationship between pCO2 and any of the other stream variables measured, such as pH, DIC (not shown), or the Ca/Na ratio (Figure 2). DIC concentrations were positively related to the molar Ca/Na ratio (Figure 2); the higher Ca/Na values were always found in the areas west and northwest of Abisko where calcareous bedrock is found, which is easily weatherable compared to silica bedrocks dominating the remaining watersheds. DIC and alkalinity were strongly positively related to stream water concentrations of Ca2+, Mg2+, and SO42− (r2 = 0.99, respectively); DIC = 2.01 × Ca2+ + 1.57 × Mg2+ − 1.48 × SO42− (concentrations in µmol L−1). On average, stream water CO2 contributed to 17% of the DIC. The predictive model for stream pCO2 using DOC consistently underestimated pCO2 in our streams, and the predicted values were unrelated to measured values.

3.3 Temporal Variations in Stream Water pCO2

[18] The water flow in all streams indicated a typical hydrological year for high-latitude watersheds with three major hydrological periods, winter base flow, spring flood, and summer-autumn base flow as exemplified by flow measurements conducted in stream 5 (Figure 3). We also plotted the average temperature and DIC values from all six streams indicating a clear dilution effect of spring flood (Figure 3). Stream water DIC concentrations across the six streams showed the lowest values during the spring peak flow, a gradual buildup during summer and early autumn base flow conditions and the highest concentrations during winter base flow (Figure 3).

Figure 3.

Seasonal variation in (upper panel) dissolved inorganic carbon (DIC) and the stream water flux for stream 5. The lower panel shows seasonal variation in water temperature. The DIC and temperature values are average values for six streams, and the error bars denote 95% confidence interval.

[19] The seasonal variation in pCO2 across the six studied streams showed a somewhat different picture. The highest pCO2 values were always found during winter base flow conditions (Figure 4) for the intermediate (Figure 4c) and the smaller streams (Figures 4d–4f), whereas seasonal variations were less pronounced in the two larger streams (Figures 4a and 4b). Especially high values were observed before ice-break in several of the streams (Figure 4). The overall range in stream water pCO2 varied from 312 to 4078 µatm, the largest variation was found in stream 5 with a watershed area of 52 km2 (Figure 4c). The relative difference in stream pCO2 between the individual streams was similar, independent of season, between the six streams. Streams 6 and 3 had the overall lowest pCO2 values, whereas stream 5 generally exhibited high values throughout the season. Seasonal variations in pCO2 were in most cases unrelated or only weakly related to DOC and/or DIC with one exception (data not shown). Seasonal variations in pCO2 in stream 5 were strongly positively related to DIC and could explain 86% of the variation in pCO2; pCO2 (µatm) = 314 × e3.10 × DIC (µmol L−1), r2 = 0.82, p < 0.001. This was also the stream that had the largest variation in pCO2. The CO2 contribution to DIC was also the largest in this stream with a seasonal median of 24% of DIC as compared to values ranging from 8 to 20% for the other five streams.

Figure 4.

The seasonal variation in δ13C-DIC and pCO2 across six streams with varying watershed sizes: a) stream 6 (565 km2), b) stream 3 (100 km2), c) stream 5 (52 km2), d) stream 1 (16 km2), e) stream 4 (6 km2), and f) stream 2 (5 km2). Dashed line denotes the approximate atmospheric concentration of CO2.

3.3 Relationships Between pCO2, δ13C-DIC, Stream Chemistry, and Watershed Properties

[20] The spatial variation in δ13C-DIC across streams with different watershed properties (September sampling) ranged from −0.6 to −13.7‰ (Figure 5). There was no difference in average δ13C-DIC values between streams in tundra and mixed watersheds (Table 3), and the range in δ13C-DIC values was similar (Figure 5). An almost similar seasonal variation in δ13C-DIC values as compared to the spatial variation was found in four of the intensely monitored streams (Figure 4); the difference between maximum and minimum values in individual streams ranged between 6 and 12‰.The lowest δ13C-DIC values were found during winter base flow conditions and the highest values during summer and autumn base flow conditions (Figure 4). The values were remarkably stable during the summer/autumn period except for stream 5 (Figure 4).

Figure 5.

The left panel shows the relationship between 13C-DIC and the ln-transformed pCO2: filled squares and solid line represent 49 streams sampled in September 2008, and crosses and dashed line represent stream samples across six streams sampled over 1 year. The right panel shows the relationship between 13C-DIC and DIC for the 49 streams sampled in September 2008: filled triangles represent tundra watersheds, and unfilled squares represent mixed watersheds.

[21] A PCA of the seasonal data for all six streams (Figure 6) also shows a clear separation between three major hydrological periods: winter base flow, spring flood, and summer-autumn base flow (Figure 4). The loadings for principal component 1 (PC1) shows that lnpCO2 and δ13C-DIC were the most important factors separating the three hydrological periods across the PC1, whereas pH and DOC were the two most important factors for principal component 2 (PC2). The first two principal components could explain 75% of the variation with PC1 accounting for 39% of the variation. The average values for each period were significantly separated for both components (p < 0.001, one-way ANOVA followed by Tukey's multiple comparison). Seasonal variations in δ13C-DIC during the ice-free season for the individual streams were best explained by DOC and to a lesser extent by DIC (Table 4). Stream 5 was an exception, and DIC and lnpCO2 were better predictors of δ13C-DIC in this case. The δ13C-DIC values and DOC were negatively related, whereas DIC was positively related, the exception being streams 5 and 6 where δ13C-DIC and DIC were negatively related.

Figure 6.

Stream water and watershed properties across an arctic landscape gradient; the data were subjected to a principal component analyses (PCA). The upper panels show the a) scores for each sampling occasion for the six streams sampled for 1 year and b) the loadings for the individual stream water properties. The lower panels show the scores c) for the individual streams from the September 2008 sampling and d) the loadings for the different watershed and stream properties.

Table 4. The Linear Relationship (r2) Between δ13C-DIC and DOC, DIC, and lnpCO2a
StreamDOCDIClnpCO2
  • a

    Data from six streams during the ice-free season. The direction of the slope (positive or negative) is indicated by the symbol within the parentheses.

  • ***

    denotes significance at the p < 0.001 level. ns, not significant.

  • *

    denotes significance at the < 0.05 level.

10.71*** (−)0.63*** (+)0.65*** (−)
20.78*** (−)0.52*** (+)ns
30.82*** (−)0.32*** (+)ns
40.81*** (−)0.43*** (+)ns
5ns0.62*** (−)0.69***(−)
60.78*** (−)ns0.17* (−)

[22] The spatial variation in stream water δ13C-DIC was strongly positively related to pCO213C-DIC (‰) = −5.88 × lnpCO2 (µatm) + 32.9, r2 = 0.69, Figure 5). We found no relationship between DIC and δ13C-DIC (Figure 5); again, the higher DIC-values were indicative of easily weatherable calcareous bedrock (Figure 2). Other factors, such as stream water temperature and pH, were also unrelated to δ13C-DIC (data not shown). It should be noted that the spatial variation in stream water temperature and pH was rather low in the September sampling; the average stream water temperature was 4.3 ± 1.0 (average ± standard deviation) and pH was7.5 ± 0.4. The seasonal δ13C-DIC showed a similar general relationship using the whole data set for the six streams (n = 223), however, with a lower degree of explanation (δ13C-DIC (‰) = −5.83 × lnpCO2 (µatm) + 31.7, r2 = 0.53, p < 0.001). A PCA including stream water and watershed properties (Figure 6) showed a clear separation for PC1, separating streams with a strong influence of calcareous bedrock from those with less along PC1, the three most important factors being DIC, Ca/Na ratio, and pH. Factors such as lnpCO2, stream length, DOC, and δ13C-DIC were the most important for PC2, whereas landscape factors such as mixed and tundra vegetation did not separate the streams.

4 Discussion

4.1 Stream pCO2 Concentrations in Subarctic Environments

[23] Arctic and subarctic environments are currently undergoing rapid changes with consequences for waterborne C losses [Frey and McClelland, 2009]. Thawing permafrost, thermokarst erosion, and conversion of permafrost regions to fens may all change terrestrial losses of DOC and DIC [Frey and McClelland, 2009]. Counter to this, deepening of flow paths due to the permafrost thaw can produce a reverse effect whereby decreasing DOC and instead increasing DIC [Walvoord and Striegl, 2007]. To what extent these changes also affect the partial pressure of CO2 in arctic and subarctic stream networks is, however, still unclear.

[24] In this study we show that (i) streams and intermediate rivers in a subarctic landscape are supersaturated in CO2 for the most part of the year and serve as a net source of CO2 to the atmosphere, similar to previous findings from other high-latitude or boreal streams [Kling et al., 1991; Teodoru et al., 2009; Koprivnjak et al., 2010]; (ii) stream pCO2 in high-latitude boreal forests are comparable to those of tundra-dominated environments [Kling et al., 1991] and are thus lower than those of previously reported results from boreal forest stream networks [Teodoru et al., 2009]; and (iii) the highest stream pCO2 in small streams and intermediate rivers occur during winter when the relative deep groundwater contribution to stream water is at its maximum [Lyon et al., 2010]. This is also the time when we see the most depleted stream δ13C-DIC indicating respiration of organic matter.

[25] The average pCO2 values we report are just half of the average for lower-latitude boreal streams and rivers reported by Teodoru et al. [2009] but somewhat higher than that for larger Arctic rivers [Kling et al., 1991]: 1852 and 611 µatm as compared to about 850 µatm for our subarctic streams and rivers. The landscape variation is, however, considerable, and the range in pCO2 values we found where almost as high as in the lower-latitude boreal forest [Teodoru et al., 2009; Humborg et al., 2010; Wallin et al., 2010; Koprivnjak et al., 2010]. The observed difference between the mixed and tundra streams is likely persistent throughout the year as indicated by the similarity with respect to the relative difference between the more intensively monitored individual streams. The absence of difference in pCO2 between nonforested and mixed watersheds may at first seem surprising since we assumed that an increased forest cover would also increase stream water pCO2. Estimates of CO2 emissions from lakes in the same region, however, reveals that lakes located in catchments with coniferous boreal forests tend to have much higher CO2 emissions than those located in catchments with extensive deciduous mountain birch [Jansson et al., 2008] or tundra and suggests that the shift between tundra and high-latitude forest exerts a less strong influence than that of coniferous boreal forest. Taken together with the results of this current study, this indicates that vegetation-related landscape variations at higher latitudes (where the boreal forest is dominated by deciduous trees or possibly low-productive coniferous forest) may not exert a sufficiently strong influence to allow for clear separation of tundra from forest-dominated catchments. These findings are important since they indicate that current observed vegetation changes at the boreal forest-tundra ecotone [Wilmking et al., 2004; Sturm et al., 2005; Tape et al., 2006] only may have minor effects on stream pCO2.

[26] We also found that stream water pCO2 were only weakly related to DOC which most likely is related to the low mire coverage and subsequent low DOC concentrations. In low-latitude boreal forest landscapes, variations in stream water pCO2 have been explained by DOC concentrations [Teodoru et al., 2009] or peatland coverage [Wallin et al., 2010; Koprivnjak et al., 2010], the latter strongly related to stream DOC [Ågren et al., 2007]. These headwater streams are also more acidic due to the higher DOC concentrations than our relatively high pH streams (circumneutral and above) and will give a proportionally higher CO2 of the DIC [Wallin et al., 2010]. As such, extending results from boreal regions dominated by coniferous forests may not necessarily be applicable when moving into regions where dominant forest cover shifts from coniferous to high-latitude forests or tundra.

4.2 Seasonal Variations in pCO2 and Hydrological Flow Paths

[27] The high pCO2 values and low δ13C-DIC values that we found during winter base flow suggest that there is a buildup of CO2 with a stronger respiratory signal [Striegl et al., 2001]. It is not likely that these high winter pCO2 values are explained by in situ stream processes such as degradation of DOC because these low-flow winter conditions typically involve much smaller volumes of flowing water in the streams and generally low DOC concentrations. The elevated stream CO2 instead coincides with a deep groundwater contribution with a longer residence time [Lyon et al., 2010] and suggests that the buildup of CO2 could be related to soil processes. A shift from no-frozen to frozen conditions could, for instance, decrease soil gas diffusion and promote a buildup of groundwater CO2. Wintertime soil respiratory activities could potentially also contribute to elevated groundwater CO2, although the respiratory activity is much smaller than that during the summer [Grogan and Jonasson, 2006]. A recent study from the boreal forest also showed that a concomitant mineralization of organic matter and a progressive buildup of DIC seem to occur in deeper groundwater during its lateral transition across a hillslope [Klaminder et al., 2011]. The freezing of the streams would also decrease CO2 evasion, but it seems unlikely that a decreased degassing solely could explain an isotopic shift of up to 6‰ (stream 1, Figure 4d) since CO2 only is a minor part of DIC (about 10%) during this time of the year and only part of the CO2 will be degassed.

[28] There seems to be a general agreement that DIC concentrations and pCO2 in small streams are more influenced by the relative contribution of groundwater inputs [Finlay, 2003; Worrall and Lancaster, 2005; Öquist et al., 2009]. We can also show that this can be true for relatively large streams such as stream 5 in this study where temporal variations in pCO2 were strongly correlated to DIC. This likely indicates a potential role of seasonal changes in hydrological flow pathways on influencing pCO2 as previous work has demonstrated a clear connection between groundwater fluxes and subsurface flow pathway distributions and stream water DIC concentrations [Lyon et al., 2010].

4.3 Possible Explanations to Stream Water δ13C-DIC Variations

[29] The strongest single factor related to stream water δ13C-DIC values was stream pCO2. In addition, the relationship shown in Figure 5 is strikingly similar to the relationship Striegl et al. [2001] found for a large set of Finnish and North American lakes sampled before spring ice melt (Figure 7). Below we discuss how δ13C-DIC values become a function of both in situ stream/lake processes and soil factors and how this could explain the common relationship we found for stream and lake 13C-DIC values.

Figure 7.

A schematic view of factors that can influence the relationship between δ13C-DIC and pCO2. The filled squares and solid line represent stream data from this study (September sampling), and the crosses and dashed line represent lake data from the USA and Finland in Striegl et al. [2001, Figure 3A]. In-stream processes such as degassing and photosynthesis will increase δ13C-DIC and lower pCO2. Mineralization of terrestrial organic matter will lower δ13C-DIC and increase pCO2 and can occur in the aquatic system or in the soil. The proportion between silicate and carbonate weathering will determine the δ13C-DIC of the groundwater that the aquatic system receives and is influenced by the isotopic value of the soil atmosphere.

[30] The range for δ13C-DIC values found in streams is often due to a mix between weathering of silicate and carbonate minerals and the respiratory δ13C-CO2 value weathering those minerals. Superimposed on this signal is the influence of degassing of CO2 (which increases the δ13C-DIC value and decreases pCO2), photosynthesis (which increases the δ13C-DIC value and decreases pCO2), and respiration of terrestrial OM (which decreases the δ13C-DIC value and increases pCO2) [Waldron et al., 2007, and references therein]. In situ stream processes can probably explain δ13C-DIC values for the two larger streams (streams 3 and 6) where variations in δ13C-DIC were small and pCO2 values were low and where we also commonly found lakes within the stream network. The more lakes and open waters, the more CO2 will be degassed and more photosynthesis will take place, and both will shift δ13C-DIC values to higher values. We also see similar trends in the smaller streams considered in this study with higher δ13C-DIC values during summer/early autumn base flow conditions. These results are, however, contradictory to what may be expected with respect to seasonal variation in terrestrial respiration. During the summer period, soil CO2 concentrations can be expected to be high due to an increased heterotrophic activity and root respiration. As such, we would expect that stream δ13C-DIC values would be lower due to a stronger respiratory influence [Amiotte-Suchet et al., 1999]. This potentially stronger respiratory influence is not visible in the streams due to higher temperature, i.e., lower CO2 solubility, lower groundwater table, and lack of the ice cap preventing degassing of CO2.

[31] Assuming a simple mixing model and using a carbonate value of +2‰ from the area (E. Lundin, unpublished data, 2012) and that soil CO2 is the proton source (as carbonic acid) having a δ13C value close to the stream δ13C-DOC value of −28‰ (R. Giesler, unpublished data, 2012), we would expect δ13C-DIC values in the range of approximately −8 to −15‰ in our streams depending on the carbonate/silicate weathering distribution [see, for instance, Finlay, 2003 and Amiotte-Suchet et al., and references therein]. However, observations presented in this study exhibited δ13C-DIC values higher than −8‰ with no discernible spatial pattern in stream water δ13C-DIC values that could be related to carbonate bedrock across the watersheds considered. There are studies indicating that the δ13C values connected to soil respiration of CO2 in the boreal part of northern Sweden can range between −22 and −26‰ during the year [Ekblad and Högberg, 2001] and values as high as −10‰ have been reported from the soil atmosphere [Rightmire, 1978]. Adopting the upper limit of this range would shift the expected δ13C-DIC values in our catchments and could potentially also give higher values than expected based on a respiratory signal close to terrestrial organic matter. A challenge for coming studies to unravel the various processes affecting δ13C-DIC will require a concert of measurements covering, for example, respiratory δ13C-CO2 values in soils, δ13C-DIC values (14C values) along hydrologic flow pathways, and quantification of in-stream processes.

5 Concluding Remarks

[32] Overall, our data emphasize that streams in high-latitude environments are sources of CO2 to the atmosphere although potentially to a lesser extent than those draining lower-latitude boreal forest environments. We were not explicitly able to disentangle the sources of stream water CO2 in this study based on δ13C-DIC values alone. It is seems very likely, however, that there is a strong respiratory influence during winter base flow conditions, whereas summer base flow δ13C-DIC values probably are more affected by in situ stream processes. Current predictions of changes in arctic and subarctic environments suggest that thawing of permafrost will lead to deeper flow paths and a larger groundwater contribution to streams and rivers [Frey and McClelland, 2009, and references therein]. Our results indicate that this also could result in increased stream CO2 concentrations and potentially increased aquatic CO2 emissions. A key in this context is where the buildup of the supersaturation in CO2 occurs: does it take place mainly in soils and groundwater or does it occur mainly in the surface waters and lakes? Degradation of terrestrial DOC in aquatic ecosystems is a C source (albeit often unaccounted for) and can thus be seen as an extension of the terrestrial heterotrophic respiration. Soil respiration can, however, be both a large potential CO2 source and a large potential CO2 sink since part of the soil CO2 can be transformed by weathering and locked in the aqueous phase as HCO3 for geological timescales. The direction of future changes in high-latitude streams may thus have large implications for landscape C budgets, and understanding the processes behind inorganic C in this context will be a key issue.

Acknowledgments

[33] This study was also supported by the Swedish Research Council (VR; 2007–3841 and 2009–2999) and the Swedish Research Council for Environment, Agricultural Sciences, and Spatial Planning (FORMAS; 214-2008-202). We thank Thomas Westin and Albin Månsson for their help in the field and laboratory, Tyler Logan for the help with analyses, Heike Siegmund and the Stable Isotope Laboratory (SIL) at the Department of Geological Sciences, Stockholm University, for their help with isotope analysis, and the Abisko Scientific Research Station where most laboratory work was performed. We also thank Robert Striegl for sharing data in Figure 7 and two anonymous reviewers for their helpful comments.