Carbon Dioxide in Boreal Surface Waters: A Comparison of Lakes and Streams

Abstract

The quantity of carbon dioxide (CO₂) emissions from inland waters into the atmosphere varies, depending on spatial and temporal variations in the partial pressure of CO₂ (pCO₂) in waters. Using 22,664 water samples from 851 boreal lakes and 64 boreal streams, taken from different water depths and during different months we found large spatial and temporal variations in pCO₂, ranging from below atmospheric equilibrium to values greater than 20,000 μatm with a median value of 1048 μatm for lakes (n = 11,538 samples) and 1176 μatm for streams (n = 11,126). During the spring water mixing period in April/May, distributions of pCO₂ were not significantly different between stream and lake ecosystems (P > 0.05), suggesting that pCO₂ in spring is determined by processes that are common to lakes and streams. During other seasons of the year, however, pCO₂ differed significantly between lake and stream ecosystems (P < 0.0001). The variable that best explained the differences in seasonal pCO₂ variations between lakes and streams was the temperature difference between bottom and surface waters. Even small temperature differences resulted in a decline of pCO₂ in lake surface waters. Minimum pCO₂ values in lake surface waters were reached in July. Towards autumn pCO₂ strongly increased again in lake surface waters reaching values close to the ones found in stream surface waters. Although pCO₂ strongly increased in the upper water column towards autumn, pCO₂ in lake bottom waters still exceeded the pCO₂ in surface waters of lakes and streams. We conclude that throughout the year CO₂ is concentrated in bottom waters of boreal lakes, although these lakes are typically shallow with short water retention times. Highly varying amounts of this CO₂ reaches surface waters and evades to the atmosphere. Our findings have important implications for up-scaling CO₂ fluxes from single lake and stream measurements to regional and global annual fluxes.

Introduction

Despite the small fraction of the surface of the Earth occupied by streams, rivers, ponds, lakes, and reservoirs a variety of studies show that inland waters play an important role in the global carbon cycle (Richey and others 2002; Cole and others 2007; Battin and others 2009; Tranvik and others 2009; Kosten and others 2010; Aufdenkampe and others 2011). These studies clearly demonstrate that inland waters are highly active sites for transport, transformation, and storage of considerable amounts of carbon received from the terrestrial environment. For example, Tranvik and others (2009) have shown that the estimated annual loss of 2 Gt is similar to the extent of annual total global net ecosystem production. Such global estimates include, however, large uncertainties.

One key uncertainty in the role of inland waters for global carbon budgets concerns seasonal and daily variations of carbon dioxide (CO₂) fluxes from inland waters. So far, global estimates are based upon daytime CO₂ emissions from inland waters with a clear bias towards summer values in the Northern Hemisphere north of 40°N (Cole and others 2007; Tranvik and others 2009). Daytime summer CO₂ values might overestimate CO₂ emissions from heterotrophic systems, typical for the large boreal region (Cole and others 1994), because photo- and microbial carbon transformations, known to drive CO₂ emissions in heterotrophic systems (Kaiser and Sulzberger 2004; McCallister and others 2005) have been shown to be enhanced by increased sunlight and increased water temperatures (Vähätalo and others 2003; Bergström and others 2010; Gudasz and others 2010), typically occurring at daytime during the summer.

Thus, the potential for CO₂ emissions from inland waters, here expressed as CO₂ partial pressure (pCO₂), might increase towards summer and decline again thereafter provided that increasing efficiency in photosynthesis does not counteract the increasing efficiency in photo- and microbial transformation in heterotrophic systems towards summer. Such increases in CO₂ emissions towards summer have, for example, been observed in a lake in Northern Sweden (Jonsson and others 2007b).

CO₂ emissions from inland waters might, however, also decrease towards summer along with decreasing hydrological inputs of CO₂ and dissolved organic carbon (DOC) from terrestrial ecosystems that have been found to be important drivers for pCO₂ variation in surface waters (Striegl and Michmerhuizen 1998; Sobek and others 2003; Humborg and others 2009; Stets and others 2009; Teodoru and others 2009). The influence of hydrological inputs of CO₂ and DOC on pCO₂ in lakes is supported by a variety of studies showing that precipitation is one of the best predictors for CO₂ concentrations both in boreal and tropical lakes (Rantakari and Kortelainen 2005; Marotta and others 2010).

Decreasing pCO₂ towards summer has also been observed by, for example, Kelly and others (2001), Kortelainen and others (2006) and Atilla and others (2011). In addition to hydrological controls, Kelly and others (2001) attributed summer pCO₂ declines to the influence of thermal stratification which reduces the ratio of epilimnetic area to the epilimnetic volume (Ae/Ve). Considering that epilimnetic sediments are an important site of degradation of organic carbon to CO₂, they suggested that a reduction in Ae/Ve resulted in a dilution of CO₂ in the epilimnion and thus in a pCO₂ decline towards summer. A stratification effect on pCO₂ has also been observed by Åberg and others (2010). They found that as soon as the epilimnion deepened, pCO₂ in the surface water of a relatively small (3.8 km²) and deep (mean depth: 5 m, maximum depth 17 m) lake increased as a consequence of liberation of hypolimnion stored CO₂. Thus, according to previous studies, pCO₂ can either increase or decrease during summer.

We hypothesized that pCO₂ in small and shallow boreal lakes with short water retention times is controlled by processes that are similar to the ones occurring in streams with minimum pCO₂ values in lakes and streams during summer when hydrological inputs of CO₂ are low. To test this hypothesis, we used more than 850 boreal lake samples and more than 60 boreal stream samples from different water depths and different months as well as water discharge data. Data on rivers were not used for this study because most Swedish and Finnish rivers are either regulated or pass large agricultural regions, and are thus more heavily affected by human activity than streams and lakes.

Methods

Data Material

The results of this study are based on four databases. The first database comprised inventory data from 756 small (median size of the 756 lakes: 0.2 km²; 90 percentile: 2.3 km²) and shallow (median depth of the 756 lakes: 4 m; 90 percentile: 6 m) boreal lakes (catchment covered by agriculture <5% and by forest and lake ≥80%) distributed over Sweden (Figure 1). The lakes were sampled at the water surface (0.5 m) above the deepest part of the lake during early autumn when the water column was mixed and water temperatures were around 4°C. Each of the 756 lakes was sampled in 1995, 2000, and 2005. Although the year 2000 was exceptionally wet, year-to-year variations remained smaller than spatial variations. For the evaluation of spatial variations we used median values of 1995, 2000 and 2005 for each lake which gave us representative lake-specific data for the autumn period. The sampling and analyzing procedure was performed by the certified water analyses laboratory at the Swedish University of Agricultural Sciences. Variables considered in this study were surface (0.5 m) water temperature (WT), pH, alkalinity (Alk), conductivity (Cond), calcium (Ca), magnesium (Mg), sodium (Na), potassium (K), chloride (Cl), sulfate (SO₄), nitrate-nitrogen (NO₃-N), total nitrogen (TN), total phosphorus (TP), absorbance at 420 nm of 0.45 μm filtered water in a 5-cm cuvette (AbsF₄₂₀), total organic carbon (TOC), and reactive silica (Si). The ratio AbsF₄₂₀/TOC was used as a proxy for the quality of TOC. TOC in Swedish boreal lakes usually contains 97% ± 5% DOC (von Wachenfeldt and Tranvik (2008)), thus TOC in this study can be seen as equivalent to DOC. All analyses were done according to standard limnological methods. The data are freely available and can be downloaded at http://www.slu.se/vatten-miljo. In addition to lake water measurements, we used GIS-derived data on lake morphometry and catchment characteristics, that is, mean lake depth (D _m), lake surface area (A), size of catchment area of the lake (ADA), elevation of lake (Alt), catchment-specific runoff (average 1961-1990; R), catchment-specific air temperature (average 1961-1990; AirT) and percentage of forest (% forest) and lake surface cover (% water) in the catchment area. The catchment-specific runoff was further used to calculate an average water residence time (WRT, in years) for lakes according to:

$$ {\text{WRT}} = \frac{V}{{{\text{Area}} \cdot R}}, $$

(1)

where V is the lake volume, based on measured mean depth and lake surface area data (m³), Area is the size of the lake catchment area excluding the lake area (m²), and R is the modelled surface water runoff in the lake catchment, provided by the Swedish Meteorological and Hydrological Institute (SMHI, www.smhi.se) and available in GIS (in mm year⁻¹).

Figure 1

Map of Sweden and Finland, showing the locations of study lakes and streams. The biggest symbols represent the 14 lakes (large triangles) and 14 streams (large crosses) for which seasonal variations have been evaluated, the small triangles and crosses represent the 107 lakes and 64 streams, respectively, for which temporal data had been available and the smallest dots show all remaining lakes (>750) used in this study

The second database comprised 107 boreal lakes and 64 boreal streams distributed over Sweden. The lakes were sampled between 4 and 107 times during 1983 and 2010 and the streams between 4 and 512 times during 1966 and 2010. Both lakes and streams were sampled at a water depth of 0.5 m. For all sampling occasions, data on WT, Alk, pH, NO₃-N, TN, TP, and TOC were available. All analyses were done according to standard limnological methods by the certified water analyses laboratory at the Swedish University of Agricultural Sciences. The data are freely available and can be downloaded at http://www.slu.se/vatten-miljo.

The third database was a sub-database from the second database and used for the evaluation of seasonal pCO₂ variations. Fourteen boreal lakes and 14 boreal streams had complete monthly data on WT, pH, Alk, Cond, Ca, Mg, Na, K, Cl, SO_4, NO₃-N, TN, TP, AbsF₄₂₀, TOC, and Si available from May to October during 1995–2010. The data were from surface waters (0.5 m). In the Swedish lakes we also had data on chlorophyll a (Chla) concentrations in surface waters and on bottom water WT. In addition, we had for a variety of sampling occasions lake bottom water oxygen (O₂) available as well as pH, Alk, Cond, Ca, Mg, Na, K, Cl, SO_4, NO₃-N, TN, TP, AbsF₄₂₀, TOC, and Si which we used to evaluate vertical differences in the variables over different seasons. From surface and bottom water WT, we determined WT differences as a measure of thermal stratification. We further assumed that lake waters are mixed when temperature differences between bottom and surface water are less than 2°C and larger than −2°C.

The lakes and streams were about evenly distributed over Sweden (Figure 1) and represented small, shallow, and oligotrophic waters (Table 1). The majority of the lakes had an average water retention time (equation 1) of less than 1 year, a mean lake water depth less than 5 m and a lake surface area less than 0.4 km². All streams had lakes upstream, and in most streams more than 60% of the water at the location of the measuring site had flowed through lakes. Thus, stream water characteristics reflect processes occurring in the terrestrial environment as well as in upstream lakes.

Table 1 Physical and Water Chemical Characteristics of 14 Boreal Lakes (row 1–14) and 14 Streams (row 15–29)

The fourth database comprised data from 99 Finnish lake sites, all of them located above the deepest parts of Finland’s largest lakes, and sampled at different depths 1 to 34 times during the years 1998 and 1999. We used this database to evaluate vertical differences in pCO₂ in large lakes over different seasons. The lakes are mainly situated in central and eastern Finland and were usually sampled three times a year, at the end of the winter stratification just before spring circulation, at the end of the summer stratification just before autumn circulation, and during autumn circulation. The water chemistry was analyzed from unfiltered samples in the accredited laboratories of the Regional Environment Centers in Finland. For this study we used data on WT, pH and dissolved inorganic carbon (DIC, measured as total inorganic carbon) that were available from different depths. For further information on methods and water quality and catchment characteristics of the lakes, see Rantakari and Kortelainen (2005).

Finally, we had modeled monthly mean water discharge values available from 176 streams and rivers distributed all over Sweden from 1995 to 2008. The water discharge values are based on a large variety of actual measurements of water discharge, air temperature and precipitation. The values are available from the SMHI, http://www.smhi.se.

Estimates of pCO₂

Estimations of pCO₂ differed between the Swedish and the Finnish waters because measurements of DIC were available for the Finnish lakes whereas they needed to be calculated for the Swedish waters with available Alk, pH and WT data. For the DIC calculation in the Swedish waters, we first adjusted equilibrium constants (K ₁ and K ₂) for in situ WT according to Stumm and Morgan (1996):

$$ \log_{10} K_{1} = - 356.3094 - 0.06091964{\text{WT}} + \frac{21834.37}{\text{WT}} + 126.8339\log_{10} {\text{WT}} - \frac{1684915}{{{\text{WT}}^{2} }} $$

(2)

and

$$ \log_{10} K_{1} = - 107.8871 - 0.03252849WT + \frac{5151.79}{\text{WT}} + 38.92561\log_{10} {\text{WT}} - \frac{563713.9}{{{\text{WT}}^{2} }}, $$

(3)

where WT is the water temperature (in K).

As a next step, we calculated the concentrations of hydrogen [H⁺] and hydroxide [OH⁻] ions with measured pH values:

$$ \left[ {H^{ + } } \right] = 10^{{ - {\text{pH}}}} $$

(4)

$$ \left[ {{\text{OH}}^{ - } } \right] = 10^{{ - (14 - {\text{pH}})}} $$

(5)

With available [H⁺], K ₁, and K ₂, we calculated the ionization fractions (α₁ and α₂) reported by Stumm and Morgan (1996):

$$ \alpha_{1} = \left( {\frac{{\left[ {H^{ + } } \right]}}{{K_{1} }} + 1 + \frac{{K_{2} }}{{\left[ {H^{ + } } \right]}}} \right)^{ - 1} $$

(6)

$$ \alpha_{2} = \left( {\frac{{\left[ {H^{ + } } \right]^{2} }}{{K_{1} \cdot K_{2} }} + \frac{{\left[ {H^{ + } } \right]}}{{K_{2} }} + 1} \right)^{ - 1} $$

(7)

Finally, we calculated DIC (in μm):

$$ {\text{DIC}} = \frac{{{\text{Alk}} - \left[ {\text{OH}} \right]^{ - } + \left[ {\text{H}} \right]^{ + } }}{{\alpha_{1} + 2\alpha_{2} }} \cdot 1000, $$

(8)

where Alk is the alkalinity (in mEq l⁻¹).

The following steps were performed both for the Finnish and the Swedish data material. From DIC, we calculated the amount of carbon dioxide (CO₂; in μM):

$$ {\text{CO}}_{2} = \frac{\text{DIC}}{{\frac{{K_{1} }}{{\left[ {H^{ + } } \right]}} + \frac{{K_{1} \cdot K_{2} }}{{\left[ {H^{ + } } \right]^{2} }} + 1}} $$

(9)

Finally, we determined pCO₂ (in μatm):

$$ p{\text{CO}}_{2} = \frac{{{\text{CO}}_{2} }}{{K_{\text{H}} \cdot P \cdot 0.987}} $$

(10)

where K _H is the WT-adjusted Henry’s constant according to Stumm and Morgan (1996):

$$ \log_{10} K_{\text{H}} = 108.3865 + 0.01985076{\text{WT}} - \frac{6919.53}{\text{WT}} - 40.45154\log_{10} {\text{WT}} - \frac{669365}{{{\text{WT}}^{2} }} $$

(11)

and P is the air pressure (bar), adjusted for altitude (Alt; in meter above sea level)

$$ P = (1013 - 0.1Alt) \cdot 0.001 $$

(12)

A variety of the Swedish samples had negative Alk values causing negative pCO₂. These negative values were removed, reducing the database of 756 boreal lakes to a database of 709 lakes. No data had to be removed for the continuous time series of 14 boreal lakes and 14 boreal streams.

Statistical Methods

All statistical tests and calculations were carried out in JMP, version 9.0. Due to the non-normal distribution of many of our variables, tested by a Shapiro–Wilk test for normality, we restricted our statistical analyses to those that are robust against non-normal distributions, that is, non-parametric tests. To find important drivers for pCO₂ on a spatial and temporal scale, we used partial least squares regressions (PLS). PLS analyses were chosen because of the method’s insensitivity to X variable’s interdependency and the insensitivity to deviations from normality (Wold and others 2001). PLS is commonly used to find fundamental relations between two matrices (X and Y), where the variance in X is taken to explain the variance in Y. In PLS, X-variables are ranked according to their relevance in explaining Y, commonly expressed as VIP values (Wold and others 2001). The higher the VIP values, the higher is the contribution of an X variable to the model performance. VIP values exceeding 1 are considered as important X variables. In our PLS analyses, pH always was the most important X variable explaining pCO₂ variations both on a spatial and a temporal scale, but because pH, Alk, and WT have been used to calculate pCO₂, we removed these three variables from all PLS analyses to avoid autocorrelation.

Results

General pCO₂ Variations in Lakes and Streams and Co-varying Variables

Taking all available pCO₂ values from lakes (n = 11,538) and streams (n = 11,126), we found significantly higher pCO₂ values in streams than in lakes (Wilcoxon test: P < 0.0001). pCO₂ in lakes ranged from below atmospheric equilibrium (less than 10% of the samples) to a maximum of 20,868 μatm with a median value of 1048 μatm. In streams, pCO₂ had a higher median value with 1176 μatm and ranged from below atmospheric equilibrium (less than 10% of the samples) to a maximum of 26,891 μatm. Using PLS, we found that pCO₂ variations during autumn water column mixing between the 709 boreal lakes, for which we had the most complete lake and catchment database available (17 water chemical and 9 catchment variables), were best explained by TOC concentrations (positive relation, VIP value = 1.60). Additional highly important X variables (VIP values > 1.3) were AbsF₄₂₀ (positive relation), AirT (positive relation), and TN (positive relation). The variables Si, K, % forest, Ca, NO₃-N, D _m, % water, Mg, WRT, A, SO₄, and ADA were not important for pCO₂ variations in our lakes during autumn (VIP values < 1). With the PLS-model, 36% of pCO₂ spatial variations could be explained (excluding the variables pH, Alk, and WT).

We reached better model performance (up to 80% explanation of pCO₂ variations) when we modelled temporal pCO₂ variations in each of the 14 lakes for which we had monthly data available from 1995 to 2010. For the 14 lake-specific PLS models, we had in addition to the 17 water chemical variables that we used for the large-scale PLS model even Chla and WT differences between surface and bottom waters as X variables available. We found that in all lakes, AbsF₄₂₀ (positive relation), AbsF₄₂₀/TOC (positive relation), or WT differences (negative relation) were most important for pCO₂ temporal variations (excluding the variables pH, Alk, and WT). Chla was important for pCO₂ variations in surface waters of only 5 out of 14 lakes (VIP > 1). In four of the five lakes Chla was negatively related to pCO₂, but in the lake where Chla reached the highest VIP value, the relationship between Chla and pCO₂ was positive (R ² = 0.14, P < 0.001; Figure 2). In this lake (Figure 2), as in most other lakes, pCO₂ temporal variations showed a significant linear negative relationship to WT differences (R ² up to 0.34, P < 0.001).

Figure 2

Monthly data on water temperature differences (WT difference) between surface and bottom waters (a), carbon dioxide supersaturation (pCO₂; b), and chlorophyll a concentrations (c) in Stensjön from 1995 to 2010. d, e Relations between pCO₂ and WT difference and chlorophyll a, respectively, in Stensjön. Stensjön was the lake where chlorophyll a concentrations reached the highest VIP value in a PLS model that has been used to predict pCO₂ temporal variations.

Modelling pCO₂ temporal variations in each of the 14 streams with monthly data from 1995 to 2010 and using 17 water chemical variables as X variables showed different patterns from the lake PLS models. Unlike the lake models where either AbsF₄₂₀, AbsF₄₂₀/TOC, or WT difference came out as the most important variable for pCO₂ variations, the stream models indicated highly deviating most important variables, that is, AbsF₄₂₀, TOC, Cl, Si, Na, Cond, Ca, Mg, or NO₃-N. In addition, X variables that were important for pCO₂ variations in some streams where unimportant in others. How much of the pCO₂ variations in the different streams could be explained was, like for lakes, highly varying, ranging from less than 22% to more than 86% (excluding the variables pH, Alk, and WT).

Seasonal pCO₂ Variations in Lakes and Streams

Dividing the data material of 22,664 samples into four seasons, that is, a winter season where some samples have been taken below an ice cover (from January to March), a spring season where spring floods occur and the waters usually are highly turbulent (April, May), a summer season (from June to September), and an autumn season where waters are turbulent again (from October to December), we found the highest pCO₂ values during the winter season, both in lakes and in streams. During winter time, pCO₂ values were significantly higher in lakes than in streams (non-parametric Wilcoxon test: P < 0.0001). During spring time, pCO₂ showed lower values and there was no longer a significant difference in pCO₂ between lakes and streams (non-parametric Wilcoxon test: P > 0.05). Lowest pCO₂ values were reached during summer in both lakes and streams but lakes had significantly lower pCO₂ values than streams (non-parametric Wilcoxon test: P < 0.0001). Towards autumn, pCO₂ increased again in both lakes and streams with significantly higher pCO₂ in streams than in lakes (non-parametric Wilcoxon test: P < 0.0001).

We obtained the same results when we used pCO₂ values from surface waters of our 14 lakes and 14 streams from which we had monthly values from May to October during 1995–2010: high pCO₂ values and no significant difference in pCO₂ between lakes and streams in May (non-parametric Wilcoxon test: P > 0.05), high pCO₂ values in both lakes and streams in September and October, and low pCO₂ values during June to August in both lakes and streams but with significantly higher values in streams than in lakes, in particular during July (non-parametric Wilcoxon-test: P < 0.0001). Thus, pCO₂ in surface waters of lakes followed a clear sine function with high pCO₂ in spring and autumn and low pCO₂ during summer (Figure 3a). Clear seasonal variations, here and previously defined as variations following a sine function (Weyhenmeyer 2009), were in lake surface waters also observed for WT differences (Figure 3c), WT, pH, AbsF₄₂₀, AbsF₄₂₀/TOC, NO₃-N, TN, and Chla. In streams, we also found seasonal variations for the same variables with the exception of AbsF₄₂₀ and AbsF₄₂₀/TOC that did not show clear seasonal patterns in streams. Analyzing water discharge data that have been modeled for 176 streams and rivers distributed over Sweden showed maximum values during spring and minimum values during August (Figure 3d). August was also the month when Chla concentrations in lakes were highest and NO₃-N concentrations lowest. In contrast, pCO₂ started to increase again in August (Figure 3a).

Figure 3

Quantile plots of monthly values on carbon dioxide supersaturation (pCO₂) in surface waters of 14 Swedish boreal lakes (a) and 14 Swedish boreal streams (b), on water temperature differences between bottom and surface waters (WT difference) in the same 14 Swedish boreal lakes from a (c) and on monthly mean water discharges from 176 stream/river sites (d). All data are based on complete time series from 1995 to 2010. pCO₂ and water difference in the lakes show similar clear seasonal variations.

Although seasonal variations were observed in both lakes and streams, seasonality with clear minimum or maximum values during summer was generally more pronounced in lakes than in streams. As a consequence, we observed highly significant (non-parametric Wilcoxon test: P < 0.0001) differences between lake surface waters and streams in July for pH (higher in streams), Alk (higher in streams), WT (lower in streams), NO₃-N (higher in streams), AbsF₄₂₀ (higher in streams), AbsF₄₂₀/TOC (higher in streams), and pCO₂ (higher in streams). TP and TOC did not show significant differences between lake surface waters and streams in July (non-parametric Wilcoxon test: P > 0.05). Higher pH and lower WT in streams during summer are expected to result in lower pCO₂ in streams than in lakes, according to the equations used for pCO₂ calculations (equations 2–11). Variables that might explain our observed significantly lower surface water pCO₂ values in lakes than in streams during summer were Chla, AbsF₄₂₀, AbsF₄₂₀/TOC, Alk, and the WT difference between bottom and surface waters. According to our PLS models, we regard WT difference, AbsF₄₂₀, and AbsF₄₂₀/TOC as the most important variables for pCO₂ temporal variations in lake surface waters. Taking the monthly median pCO₂ of Swedish lake surface waters and relating them to the monthly median of the WT differences in the Swedish lakes, we observed a 40 μatm decrease in pCO₂ in surface waters per 1°C increase in the WT difference between surface and bottom waters (Figure 4a). The same pattern was observed in the Finnish lakes, but in these lakes, the intercept, which reflects lake water mixing conditions, was 489 μatm lower than in the Swedish lake surface waters (Figure 4a). The difference between pCO₂ in the surface waters of the large Finnish lakes and the small, shallow Swedish lakes corresponded well with pH differences between the two lake types (Figure 4b).

Figure 4

Monthly median values of available water temperature differences (WT difference) between bottom and surface waters (a) and monthly median values of pH (b) in relation to available monthly median values of carbon dioxide supersaturation (pCO₂) in surface waters of Swedish (14 lakes, data from 1995 to 2010; open circles) and Finnish (99 lakes, data from 1998 and 1999; black squares). For the Swedish lakes, the regression equation of panel a runs: y = −40x + 1258 (R ^s = 0.90, P < 0.01, n = 6), and for the Finnish lakes: y = −34x + 769 (R ² = 0.64, P < 0.05, n = 8).

Vertical pCO₂ Variations

In the 14 Swedish lakes, pCO₂ was significantly higher in bottom waters than in surface waters for all months (non-parametric Wilcoxon test: P < 0.001 for May, June, July, August, September, and P < 0.05 for October). Also in the Finnish lakes, pCO₂ in bottom waters significantly exceeded pCO₂ in surface waters (non-parametric Wilcoxon test: P < 0.001 for May, June, July, and August) except during the months of September and October when it was similar in bottom and surface waters (non-parametric Wilcoxon test: P > 0.05). The occasions where pCO₂ in surface waters exceeded that in bottom waters were rare and negligible in size, and they occurred during the mixing period (Figure 5). During all months, including the months when the water column was mixed, pCO₂ in bottom waters of the 14 Swedish lakes was significantly higher than pCO₂ in streams (non-parametric Wilcoxon test: P < 0.001). pCO₂ in bottom waters of lakes was significantly negatively related to bottom water oxygen concentrations (R ² = 0.37, P < 0.0001) but not related at all to bottom water temperatures (R ² = 0.00, P > 0.05).

Figure 5

Relationships between surface water carbon dioxide supersaturation (pCO₂) and bottom water pCO₂ in Swedish (a, b) and Finnish (c, d) lakes during the mixing and stratification period. The Swedish dataset comprises 14 boreal lakes with monthly data from May to October during 1995 to 2010 (due to missing bottom water data, the data series are not complete), and the Finnish datasets includes available data from 99 lakes samples during March and October in 1998 and 1999. The dashed lines represent the 1:1 relationship.

Discussion

Our results show clear differences and similarities in pCO₂ variations between boreal lakes and streams. In spring, prior to thermal stratification in lakes, pCO₂ measurements were similar in lakes and streams. In addition, overall seasonal patterns characterized by the highest pCO₂ in spring and autumn and lowest pCO₂ during summer following water discharge were comparable between boreal lakes and streams. The seasonal pCO₂ variations were, however, much more pronounced in lakes than in streams.

As outlined in the introduction, seasonal pCO₂ variations in surface waters, particularly in lakes, are not completely understood yet and have revealed contrasting patterns with either maximum (for example, Jonsson and others 2007a) or minimum pCO₂ values during summer (for example, Kelly and others 2001; Kortelainen and others 2006; Atilla and others 2011). Seasonal variations in pCO₂ are a result of CO₂ and bicarbonate inputs from the terrestrial environment, photosynthesis, photo- and microbial mineralization and water column mixing. In streams we expect pCO₂ to be primarily driven by hydrological conditions. Accordingly, we found the highest water discharges and highest pCO₂ values during spring and the lowest during summer. Hydrological conditions and catchment processes probably counteracted the influence of photo- and microbial transformations in streams which we expected to result in increased pCO₂ towards summer. An antagonistic effect of water discharge and photo- and microbial transformation on pCO₂ during summer might give an explanation why water discharge decreased faster towards summer than pCO₂ in streams (Figure 3).

Hydrological conditions and catchment processes are likely to have an important effect not only on pCO₂ in streams but also on pCO₂ in small and shallow boreal lake waters with short water retention times, as suggested by Humborg and others (2009). Accordingly, we found high pCO₂ values in lake waters at high water discharges in spring and low pCO₂ values at low water discharges in summer (Figure 3). The strong hydrological influence was probably why we found comparable pCO₂ variations in lakes and streams during spring. We suggest that also autumn pCO₂ variations in lakes are mainly influenced by water discharge patterns, because we found TOC, which in Swedish waters is highly influenced by water discharge (Erlandsson and others 2008), to be the most important variable for pCO₂ variations in lakes in autumn.

During summer, however, pCO₂ in streams and lakes seems to be affected by different processes, indicated by significantly lower pCO₂ values in surface waters of lakes than in streams. Most obvious differences between lakes and streams during summer are the effects of thermal stratification and photosynthesis. Photosynthesis is expected to result in decreasing pCO₂ in summer during the daytime. We do not have data on photosynthesis in our lakes but based on measurements on boreal lakes in general (for example, Jonsson and others 2001; Einola and others 2011) and the fact that Chla concentration that can be regarded as a measure of photosynthesis was only important for pCO₂ variations in 5 out of 14 lakes in our PLS models, we assume that photosynthesis is only a minor process influencing pCO₂ in our heterotrophic boreal lakes. This statement is further strengthened by the fact that the lake where Chla received the highest VIP value in the PLS model for pCO₂ variations showed a positive and not a negative relationship between Chla and pCO₂. We attribute the positive relationship to deviating seasonality patterns of Chla and pCO₂ in this particular lake. In Stensjön pCO₂ and Chla are out of phase with the highest Chla concentrations in August and lowest pCO₂ values in July (Figure 2). Such deviating seasonality patterns reflect that photosynthesis is not the dominant process determining pCO₂ variations in our boreal waters.

The influence of thermal stratification on pCO₂ in lake surface waters was, however, clearly detectable, despite our lakes being rather small and shallow. The influence of thermal stratification on biogeochemical cycling in lakes is well known (Keller 2007). Thermal stratification can result in nutrient depletion in the epilimnion, and oxygen depletion in the hypolimnion as element exchanges between epi- and hypolimnion are hindered by thermal stratification. Assuming that a large fraction of CO₂ is produced in the sediments or enters the hypolimnion via groundwater, a strong thermocline will result in CO₂ accumulation in the hypolimnion. The CO₂ accumulation in the hypolimnion occurs at the same time as O₂ is consumed, indicated by our observed negative relationship between pCO₂ and O₂. Consequently, O₂ concentrations have earlier been shown to be well related to CO₂ concentrations. In 177 randomly selected Finnish lakes, for example, as much as 79% of the variation in CO₂ could be explained by O₂ concentration only (Kortelainen and others 2006). Because we also found a clear negative relationship between O₂ concentrations and bottom water pCO₂ we suggest that thermal stratification plays a critical role for the distribution of pCO₂ in the water column of lakes.

A strong negative relationship between intensity of thermal stratification and pCO₂ in the epilimnion was not only found in Swedish lakes but also in Finnish lakes. Finnish lakes had, however, constantly lower pCO₂. The difference in pCO₂ between Swedish and Finnish lakes was reflected in pH differences (Figure 4b). We are, however, not able to differentiate between pH being a cause or a response variable. It is possible that pH in the large Finnish lakes is higher due to a larger lake volume giving the lakes a better buffering capacity and due to a higher relative importance of agriculture in the catchment area (Rantakari and Kortelainen 2005). In that case, high pH would result in low pCO₂. It is, however, also possible that primary production is higher in the large lakes, resulting in a higher CO₂ consumption with consequent low pCO₂. In that case, low pCO₂ would result in high pH. Independently of what the driver and the response is we found consistently higher pCO₂ values in the small Swedish lakes compared to the large Finnish lakes along the WT gradient (Figure 4a). Further studies on the relative importance of processes determining pCO₂ in small and in large lakes are needed to fully understand pCO₂ variations over large scales.

Once WT differences between surface and bottom waters become negligible towards autumn and the water column starts to mix, pCO₂ in surface waters of lakes reaches values that are close to the ones found in streams (Figure 3a, b). These results correspond to the findings of Bellido and others (2009), Huotari and others (2009), and Laurion and others (2010) who all found maximum gas losses during water mixing periods. Highest pCO₂ values, however, are observed below ice cover, as seen in this study and earlier reported by Kortelainen and others (2006). How far the accumulated CO₂ below an ice cover evades into the atmosphere at ice-off remains still unclear. We have indications from our results that CO₂ is accumulated in bottom waters throughout the year despite spring and autumn circulation. Such an accumulation of pCO₂ in bottom waters of lakes that exceeds pCO₂ in streams is likely a result of many processes: incomplete water column mixing, CO₂ input from groundwater into deeper parts of a lake and/or higher CO₂ production in the lower water column by remineralization of organic matter that has settled out from the epilimnion.

Although CO₂ in our small and shallow lakes might primarily be produced in the catchment or in the sediments, we have indications for epilimnetic CO₂ production because we found decreasing AbsF₄₂₀/TOC ratios in surface water of lakes towards summer. Decreasing AbsF₄₂₀/TOC ratios correspond to a preferential degradation of colored organic matter which has been observed for waters that have been exposed to solar radiation, suggesting a strong influence of photomineralization (Moran and others 2000; Vähätalo and others 2000. Seasonal variations in AbsF₄₂₀/TOC were not detectable for streams, probably because of other processes overriding the effect of photomineralization in typically shaded boreal streams.

From our results, we conclude that seasonal pCO₂ variation in boreal lakes and streams follows water discharge patterns, but that lakes during summer are additionally affected by WT differences between surface and bottom waters, causing pronounced differences in pCO₂ variations between lakes and streams and also between lake surface and bottom waters (Figure 6). According to our results, pCO₂ reaches minimum values in surface waters during summer. If such minimum pCO₂ values are the basis for annual flux estimates on a global scale, CO₂ fluxes from inland waters to the atmosphere will be underestimated. More research on seasonal and also on daily pCO₂ variation is needed to reduce uncertainties in global estimates of CO₂ released from surface waters to the atmosphere.

Figure 6

Probability density functions of carbon dioxide supersaturation (pCO₂) in lake surface waters (dark grey), lake bottom waters (light grey) and stream surface waters (black) during the water column mixing period in October (a) and during summer stratification in July (b). The figure is based upon monthly data from 14 boreal lakes and 14 boreal streams during 1995–2010.