Geographical and environmental drivers of regional differences in the lake pCO2 versus DOC relationship across northern landscapes

Authors

  • Jean-François Lapierre,

    Corresponding author
    1. Groupe de Recherche Interuniversitaire en Limnologie et en Environnement Aquatique (GRIL), Département des Sciences Biologiques, Université du Québec à Montréal, Montréal, Québec, Canada
    Search for more papers by this author
  • Paul A. del Giorgio

    1. Groupe de Recherche Interuniversitaire en Limnologie et en Environnement Aquatique (GRIL), Département des Sciences Biologiques, Université du Québec à Montréal, Montréal, Québec, Canada
    Search for more papers by this author

Corresponding author: J.-F. Lapierre, Groupe de Recherche Interuniversitaire en Limnologie et en Environnement Aquatique (GRIL), Département des Sciences Biologiques, Université du Québec à Montréal, C. P. 8888, succursale Centre-Ville, Montréal, QC, H3C 3P8, Canada. (jfrancoislapierre@gmail.com)

Abstract

[1] Several recent studies have identified dissolved organic carbon (DOC) as playing a key role in determining surface water partial pressure of CO2 (pCO2) in northern lakes and, in particular, in shaping the commonly observed patterns of CO2 supersaturation. The nature of this role is unclear, however, and appears to vary regionally, as evidenced by the contrasting strength and shape of the diverse pCO2 versus DOC (pCO2-DOC) relationships. Here we combine original data on lakepCO2 from six boreal and temperate regions of Québec (Canada) with 13 studies from northern temperate and boreal aquatic landscapes published in the past 15 years to explore the factors that explain the differences in regional pCO2 baselines (pCO2 at low DOC) and in the slopes of the pCO2-DOC relationships. Mean elevation was the best predictor of the regionalpCO2 baselines, suggesting that lake position in the landscape determines the contribution of terrestrially derived CO2 to lake pCO2. In contrast, the slope of the pCO2-DOC relationships was strongly negatively correlated to the mean regional TP:DOC ratio. The relationship between DOC and TP varied at the cross-regional scale, and there was a large increase in the TP:DOC ratio at TP > 20μg L−1, resulting in negative slopes of the pCO2-DOC relationships for regions situated in that part of the TP gradient. These results highlight the interplay that exists between geographical gradients, large-scale biogeochemical patterns in regional lake trophic status, and the associated metabolic balance in determiningpCO2 dynamics in northern lakes.

1. Introduction

[2] Lakes act as funnels for materials derived from soils, including organic and inorganic nutrients and carbon exported from their terrestrial catchments [Cole et al., 2007; Prairie, 2008]. This supplementary material is responsible for a large fraction of the flow of carbon through aquatic biogeochemical cycles and impacts aquatic ecosystems in several ways. In particular, organic carbon exported from the terrestrial landscape contributes to the widespread supersaturation of CO2 in surface waters in lakes in northern landscapes, and to its subsequent evasion to the atmosphere [Jonsson et al., 2007; Kling et al., 1991], which is now recognized as a globally significant component of the biosperical CO2 budget [Battin et al., 2009; Tranvik et al., 2009].

[3] The drivers of this outgassing, however, are not well understood, which in turn limits our understanding of the role of lakes in temperate and boreal carbon budgets. There is a large degree of variability at different spatial scales across lakes, both in the partial pressure of CO2 (pCO2), and in the contribution of the different processes that supply pCO2. For example, it has been widely reported that biological mineralization of allochthonous organic carbon is a major source of excess CO2 in lakes [del Giorgio and Peters, 1994; Duarte and Prairie, 2005; Sobek et al., 2003], but other studies have shown that the apparent net heterotrophy is insufficient to sustain the observed CO2 supersaturation, and have pointed to inputs of CO2 from the catchment as playing a key role [Dubois et al., 2009; Finlay et al., 2009; Stets et al., 2009]. Likewise, photo-oxidation of DOC was found to account for essentially all of the DOC loss to CO2 in boreal Canadian lakes [Molot and Dillon, 1997], whereas respiration was shown to account for most of the mineralization in Swedish lakes with a comparable DOC range [Granéli et al., 1996; Jonsson et al., 2001].

[4] These apparently conflicting observations suggest that different processes are simultaneously active in all lakes and that their relative contribution to surface water pCO2 varies greatly, both temporally and along lake gradients, leading to a large heterogeneity both within [Åberg et al., 2004; Kelly et al., 2001] and across [Sobek et al., 2005] lakes. Several studies have nevertheless identified broad patterns in lake surface water pCO2 along, for example, gradients of lake size [Alin and Johnson, 2007; Marchand et al., 2009; Urabe et al., 2011], nutrients [Huttunen et al., 2003; Rantakari and Kortelainen, 2008], and DOC concentration [Bergström et al., 2004; Hope et al., 1996; Kritzberg et al., 2005; Larsen et al., 2011].

[5] These patterns with DOC concentration have generally been interpreted as evidence that C export from the terrestrial landscape fuels CO2 supersaturation in lakes, wherein DOC would either act as the direct substrate for CO2 production, or as a proxy for the loading of inorganic C [Humborg et al., 2010]. In this regard, the relationship between pCO2 and DOC appears to be a key property that may account for the regional differences in the regulation of lake pCO2. For example, the pCO2 at very low DOC concentrations may reflect, for any given region, the level of CO2 that is present at low rates of in situ DOC consumption and degradation, and thus may be an indication of landscape scale export of CO2 to lakes. Likewise, a strong positive relationship between DOC and pCO2would suggest a dominant role of in situ oxidation of terrestrial organic carbon (biological or photo-chemical) in generating excess CO2 in lakes, or a significant contribution of associated drivers of DOC with a similar impact on pCO2. Conversely, a weak, or a negative relationship would indicate either a strong contribution from other drivers that are not correlated to DOC, or from drivers that are correlated with DOC but have a contrasting impact on pCO2, respectively.

[6] In this context, it is interesting to note that whereas there have been a number of pCO2-DOC relationships reported for a variety of northern regions, there is a remarkable range in the strength and, more importantly, in the shape of these relationships [Roehm et al., 2009]. Several factors could explain these regional discrepancies. For example, regional patterns in DOC composition and susceptibility to biological and photochemical degradation could lead to contrasting pCO2 levels per unit of ambient DOC, and potentially to a different slope and intercept of the regional pCO2-DOC relationships. Likewise, regional patterns in landscape morphology and lake position in the landscape may dictate the amount and relative proportions of DOC and CO2 exports from land to lakes. Further, regional patterns in lake trophic status and the ensuing patterns in lake metabolism may modulate the impact of increasing DOC concentration, and presumably processing, on ambient pCO2 dynamics. Those different scenarios imply a very different pCO2 for any given amount of DOC present in lakes, and assessing such scenarios requires exploring how the pCO2-DOC relationship behaves at scales that are broader than those reported in individual field studies. This approach, however, also limits the number of variables available for an inter-regional comparison. Here we combine our own empirical results of a large-scale study across temperate and boreal lakes in Québec, with an extensive literature analysis of published data onpCO2for a wide range of northern temperate and boreal lakes across the world in order to explore the latter two scenarios that we introduced. We show that integrative geographical and environmental variables, such as elevation and the ratio of TP to DOC, predict well the inter-regional patterns in thepCO2-DOC relationships across northern landscapes.

2. Methods

2.1. Study Area

[7] The experimental portion of this study was conducted over 3 years in six boreal regions from Québec, Canada. We sampled the Eastmain region (52°N, 75°W) in 2009, the Abitibi (48.5°N, 79°W), James Bay (51°N, 79°W) and Laurentians (46°N, 74°W) regions in 2010, and the Chibougamau (49.5°N, 74°W) and Chicoutimi (48°N, 71°W) regions in 2011. These regions are part of the temperate and boreal biomes of Québec and cover a large span of characteristics in terms of climate, land cover and geomorphology. Sites within a region were typically found within a 50 km radius within which lakes and catchment properties were rather homogeneous, whereas different regions were several hundreds of kilometers afar and were represented by contrasting climate, vegetation cover and catchment morphometry. Briefly, from South to North, mean annual temperature ranges from 4.9°C in the Laurentians to −2.0°C in the Eastmain region. Mean annual precipitation ranges from 1050 mm y−1 in the Laurentians to 695 mm y−1 in the Eastmain regions, and between 900 and 1000 mm y−1for the remainder of the regions. Dominant vegetation cover shifts from deciduous forest in the Laurentians to mixed deciduous and conifer forest in the Abitibi and Chicoutimi regions; the James Bay, Chibougamau and Eastmain regions are characterized by spruce-moss forest. Relief is generally flat and lakes are typically shallow and turbid in the Abitibi and James Bay regions, which are situated in the Abitibi clay belt; elevation and landscape steepness were the highest in Laurentians and Chicoutimi regions. High densities of beavers (Castor canadensis) typically alter the landscape hydrology, especially in the Abitibi region, due to the construction of dams of various sizes that flood rivers shores and increase water residence time.

2.2. Sampling and Analyses

[8] A total of 198 different lakes were sampled from June to August over the three sampling years by boat or hydroplane depending on accessibility. Most lakes were sampled once, but some were sampled twice over a two-month interval; averages have been used in the latter case. Water was pumped at a depth of 0.5 m from the deepest measured point of the lake for in situ and lab measurements. In situpCO2was measured using a PP-Systems EGM-4 infrared gas analyzer (IRGA) receiving air equilibrated with the water by pumping lake water through a Liqui-Cel MiniModule contactor membrane. Total phosphorus (TP) was analyzed spectrophotometrically after persulfate digestion. Total nitrogen was analyzed as nitrates following alkaline persulfate digestion and measured on an Alpkem FlowSolution IV autoanalyzer. Chlorophylla (Chl a) was determined spectrophotometrically following filtration on Whatman (GF/F) filters and hot ethanol (90%) extraction. Filters were sonicated prior to extraction. Lake water filtered through 0.45 μm porosity PES cartridges (Sarstedt) was analyzed for DOC concentration on an OI 1010 TOC analyzer following sodium persulfate digestion. The surface and catchment areas, as well as elevation of the lakes we sampled were determined using the ArcView 3.2 and ArcGIS v.9 software applied on the DEM derived from (1:50000) maps.

2.3. Literature Analysis

[9] In addition to articles from our existing database, we searched “Web of Science” for articles containing “DOC,” “CO2” or “pCO2” or “carbon dioxide,” and “lake*” as topical keywords, and extended our search for the past 25 years. We selected studies that had focused on temperate and boreal regions that provided DOC and pCO2 data on enough number of lakes to allow building a significant pCO2-DOC relationship. In cases where published data were available for regions that we sampled ourselves, we gave priority to our own data. Although there have been substantial numbers of papers that have reported lake CO2 in the past decades, many fewer have reported both DOC and CO2 across a range of lakes. We ended up collecting published data on lake pCO2, DOC and ancillary lake and watershed properties taken directly from tables and figures from 13 papers published from 1996 to 2011. Five of the 13 studies did not simultaneously present DOC and pCO2 data but were still used to explore the link between pCO2and lake and watershed morphometry. The data cover lakes from boreal and northern temperate regions in Canada, the United States, Sweden and Finland. Because the studies presented in this paper have typically conducted sampling in a specific region of the temperate or boreal landscape, we use the term “regional” to refer to study-specific patterns or averages hereon, even though the spatial scale of the different regions differs across studies:Tables 1 and 2 describe the environmental and geographical gradients for the different regions comprised in this literature analysis.

Table 1. Mean and Range of the Major Variables for the Lakes Sampled in This Study and for the Published Data Used in This Analysis
VariableRange (min; max)
This studyPublished
Lake area (km2)0.01; 23350.0007; 378 400
Elevation (m)121; 85681; 1140
Maximum depth (m)0.50; 44.50.7; 680
DOC (mgL−1)0.76; 22.30.7; 21
TP (μgL−1)2.88; 153.20.25; 201
pCO2 (μatm)102; 2391103; 3629
 Average; Median
This studyPublished
Lake area (km2)54.9; 1.13128; 0.88
Elevation (m)368; 299404; 370
Maximum depth (m)8.5; 7.128.6; 8.4
DOC (mgL−1)8.4; 7.55.6; 5.2
TP (μgL−1)14.5; 8.813.0; 8.3
pCO2 (μatm)687; 606723; 590
Table 2. Parameters Obtained From the pCO2-DOC Relationships for Original and Published Data Among Northern Landscapesa
ReferenceRegionr2pCO2 BaselineSlopeGeographical Range
  • a

    The slope was obtained from the log (pCO2) versus log (DOC) relationships. The pCO2 baseline (μatm) corresponds to the estimated pCO2 (μatm) at DOC = 1.9 mgL−1 for each region, obtained from the corresponding regional equations. No pCO2 baseline was calculated for nonsignificant relationships; * = not significant. Parameters have also been estimated for the overall relationship when all data, published and from this study, are analyzed together (All data).

This studyAbitibi, Can.0.00*n.a.n.a.48.0°–49.0°N, 78.2°–79.6°W
 Chibougamau, Can.0.525530.2748.9°–50.9°N, 72.7°–74.7°W
 Chicoutimi, Can.0.216330.5247.7°–48.8°N, 70.5°–72.2°W
 Eastmain, Can.0.393500.6451.9°–52.4°N, 75.3°–76.5°W
 James Bay, Can.0.172250.5549.0°–50.6°N, 78.0°–79.6°W
 Laurentians, Can.0.02*n.a.n.a.45.8°–46.1°N, 73.8°–74.2°W
Bergström et al. [2004]Northern Swe.0.564740.3863°–67°N, 13°–19°W
del Giorgio et al. [1999]Estrie, eastern Can.0.452991.1045.0°–45.9°N, 71.0°–72.7°W
Finlay et al. [2009]Prairies, central Can.0.821410−1.0550.0°–51.3°N, 101.3°–107.1°W
Hope et al. [1996]Wisconsin, northern US0.181790.9245.5°–46.6°N, 89.2°–89.6°W
Jonsson et al. [2003]Abisko, northern Sweden0.624550.5567.5°–68.5°N, 18.0°–22.0°E
Kelly et al. [2001]Northwestern Ontario, Can.0.581361.1848°–51°N, 88°–94°W
Kritzberg et al. [2005]Wisconsin, northern US0.535910.7946.1°–46.3°N, 89.3°–89.6°W
Sobek et al. [2003]Sweden0.512101.1557°–64°N
 All data0.232820.8645°–69°N, 107°W–22°E

[10] Different techniques have been used across studies to determine lake pCO2, including calculation from DIC, alkalinity, temperature and pH [Rantakari and Kortelainen, 2005], direct measurement by gas chromatograph [Algesten et al., 2004] and infrared gas analyzer [Åberg et al., 2004] of CO2 equilibrated between water and air by headspace equilibration [Cole et al., 1994] or equilibrating devices such as contactor membrane [Roehm et al., 2009]. These methods have very different levels of resolution, but previous studies have shown that both the magntitude and the patterns are highly comparable across studies using different measurement techniques [del Giorgio et al., 1999; Prairie et al., 2002]. Although there is clearly additional noise introduced by comparing the pCO2values obtained from different techniques, we contend that the large gradients covered in this paper greatly limit the impact of these cross-method differences, and that the resulting noise does not introduce a systematic bias to the large-scale patterns presented here. We have thus used thepCO2 data from the different studies as reported without any further correction or manipulation to account for potential methodological issues.

[11] The sampling timing and frequency also varied among studies, but in all cases there was sampling at least during mid-summer. For studies with a single sampling date, we used the published data directly. Some studies reported several measurements throughout the ice-free period, and when averages for this period were presented in the study, we used those means. When data were available for different sampling dates but no mean was reported, we calculated averages for the ice-free season, more specifically, from May to October, such that there was no systematic seasonal effect among the studies that could bias the relationships presented here.

2.4. Data Manipulation and Statistical Analyses

[12] In this paper we compare how the pCO2-DOC relationships vary across regions and explore factors that may explain these differences. One major problem in this analysis is that the key environmental and geographic variables considered here, in addition topCO2 and DOC, were not systematically included or reported in the various studies. As a result, the use of multivariate analyses would necessarily lead to either the removal of a large number of regions (which had incomplete variables sets) or the use of a very restricted set of variables that were reported for the largest number of regions, which would clearly undermine the main objective and the reach of this study.

[13] To circumvent this problem, we first fit orthogonal (variance allocated to both explanatory and response variables) [Legendre and Legendre, 1998] linear regressions on log-transformed data from our own and published studies (log(pCO2) = a log(DOC) + b), using JMP software (SAS institute, NC, USA), and generated study-specific, regionalpCO2-DOC models. We then used the resulting regression equations to estimate a regionalpCO2 baseline at very low DOC (pCO2 at the 5th percentile of DOC concentrations for all available data; 1.9 mg L−1). No baseline was estimated in the absence of a significant regression. We also used the slopes of the relationships as indicators of the effect of DOC on pCO2 at the regional scale, more particularly, of how much pCO2 increases for a given increase in DOC in each study. These new response variables (pCO2 baseline and slope) were then analyzed individually against regional means of commonly measured geographical/morphological (latitude, elevation, mean annual temperature, lake size, mean depth), and lake environmental (Chl a, TP, TN, pH) predictors. Piecewise regressions [Toms and Lesperance, 2003] were performed to identify break points in the orthogonal relationships of DOC and pCO2 against TP, using the “Segmented” package [Muggeo, 2008] on R software [R Development Core Team, 2010].

[14] Variables were log-transformed to meet normality assumptions of linear regression. In one of the reported study, however, the non-transformedpCO2-DOC relationship was linear and would not require transformation. In this case, the coefficient of correlation differed by only 0.01 between transformed versus non-transformed data, and not transforming data would greatly limit the comparability of the resulting parameters with that of other regions. Therefore, we only present here the results obtained from the fits to log-transformed data.

3. Results

3.1. General Lake Properties

[15] Table 1 presents the ranges and averages of lake properties, both for our study lakes and for the lakes for which we gathered published data. There was approximately an order of magnitude in range for most variables, both for our own and published data, showing the wide gradients in environmental and geographical conditions of the systems included in our analysis. Range, average and median TP and pCO2 were comparable between Québec lakes and published estimates from other northern landscapes, but DOC tended to be higher and elevation to be lower in Québec lakes. Average lake area and maximum depth were skewed toward higher values due to a few extreme measurements, but medians suggest that the distribution of lake morphometric characteristics were also comparable between Québec and other northern temperate and boreal lakes (Table 1).

3.2. Regional pCO2-DOC Relationships

[16] Out of the six regional pCO2-DOC relationships that we obtained from our own field study, four were significant and positive while the remaining two (Abitibi and Laurentians regions) were not significant (Figure 1, Table 2). There were strong regional differences in the strength and in the shape (slope and pCO2 baseline) of these relationships, relative to each other and to those reported for other northern regions (Table 2). In particular, the pCO2 baseline for Chicoutimi and Chibougamau regions of Québec were among the highest when all studies were considered together, and the slope of the pCO2-DOC relationship for the Chibougamau region was the lowest positive relationship (Table 2).

Figure 1.

Relationships of partial pressure of CO2 (pCO2) versus dissolved organic carbon (DOC) for lakes sampled in temperate and boreal Québec, Canada. Regression lines are presented for significant relationships. The parameters are included in Table 2.

[17] Our literature analysis yielded a total of 8 additional regional studies that had reported concomitant surface water pCO2 and DOC data. Out of the 8 regional studies, seven yielded significant positive relationships and one yielded a significant negative relationship (Table 2, Figure 2). The average r2 for the 14 individual relationships (8 published + our 6) presented in Figure 2 (r2= 0.46) was twice that obtained by fitting a log-log curve to all the points together (r2 = 0.23, Table 2, “All data”), suggesting that the predictive power of DOC is much stronger at smaller, regional scales, and drastically declines as the scale is broadened. Taken together, our field data plus the published data show that there is over an order of magnitude range in pCO2 at any given concentration of DOC, but that there are regional trends in this scatter, reflected in widely varying slopes and pCO2 baselines at low DOC between regional models (Figure 2, Table 2).

Figure 2.

Relationships of pCO2 versus DOC for our own and published data among the boreal landscape. Error bars represent standard deviation for temporally replicated measurements, when available. The regression parameters are presented in Table 2.

3.3. Patterns in Regional pCO2 Baseline

[18] The regional pCO2 baselines were not correlated to water chemistry or other geographical variables but were strongly positively correlated to the average lake elevation in each study region (Figure 3). Lake elevation is presumably a proxy of other catchment and lake morphometric variables with a more direct impact on lake pCO2, but unlike elevation, such variables were not commonly measured in studies reporting concomitant pCO2 and DOC values. The relationship of lake elevation with catchment and lake morphometric variables, available for individual lakes in some regions, provides a more causative assessment of the potential role of lake elevation on pCO2 dynamics. Individual lake elevation was negatively correlated with lake catchment area (Figure 4a), which was, in turn, negatively correlated with the individual lakes pCO2 per unit DOC (Figure 4b). Lakes situated in higher catchments also tended to be smaller (Figure 5a), and smaller lakes tended to be more supersaturated in CO2 (Figure 5b). There was a large degree of variability in pCO2 for a given lake size, but small lakes tended to be systematically supersaturated, whereas pCO2converges to equilibrium and even under-saturation with increasing lake area (Figure 5b).

Figure 3.

Relationship between mean regional altitude and pCO2 baseline, which represents the estimated pCO2 at DOC = 1.9 mg L−1, calculated from the corresponding fit (parameters presented in Table 2). Lake altitude was not available for the study of Hope et al. [1996]; individual lakes altitude were not available for the study of Sobek et al. [2003], but we used pro-rated regional averages presented in the paper.

Figure 4.

Relationship between (a) lake elevation and catchment area and (b) lake catchment area and pCO2 per unit DOC. “Published” data originate from del Giorgio et al. [1999], Hope et al. [1996], Jonsson et al. [2003], and Kelly et al. [2001]. Equation for Figure 4a: log(catchment area) = −7.85log(elevation) + 20.7. Equation for Figure 4b: log(pCO2/DOC) = −0.24log(catchment area) + 2.02. Equations, p-values, and r2 for all data analyzed together.

Figure 5.

Lake area as a function of (a) elevation and (b) the relationship between pCO2 and lake area. “Published” data originates from del Giorgio et al. [1999], Hope et al. [1996], Jonsson et al. [2003], Kelly et al. [2001], Kritzberg et al. [2005], and Finlay et al. [2009]. Equation for Figure 5a: log(lake area) = −6.89log(elevation) + 17.56. Equation for Figure 5b: log(pCO2) = −0.18log(lake area) + 2.74. Equations, p-values, and r2 for all data analyzed together.

3.4. Patterns in the Regional pCO2-DOC Slopes

[19] The slope of the pCO2-DOC relationships, on the other hand, was not correlated to the regional averages of most geographical or morphometric variables, weakly correlated to the average regional concentrations of TP (r2 = 0.34, p = 0.04), but rather was strongly and negatively correlated to the average regional TP:DOC ratio (Figure 6a). This relationship suggests that a given increase in DOC is associated with a lower increase in pCO2 at higher TP:DOC ratios, coherent with the trend of the decreasing pCO2 per unit DOC as TP increases in individual lakes (Figure 6b), and with the observation that past a certain threshold in the TP:DOC ratio, increases in DOC do not imply higher pCO2.

Figure 6.

(a) The slopes of the regional pCO2-DOC relationships as a function of average regional TP:DOC ratio, and (b) thepCO2 per unit DOC as a function of TP for individual lakes. Equation for Figure 6a: y = −1.02 log(average regional TP:DOC) + 1.24. Equation for Figure 6b: y = −0.92 log(TP) + 2.93. Equations, p-values, and r2 for all data analyzed together.

[20] The relationships of DOC and pCO2 with TP (Figure 7) are consistent with this pattern. At TP < 20.2 μg L−1, DOC increases faster than TP (log slope = 1.22, Figure 7a), and in this same range, pCO2increases as does TP (log-slope = 0.83,Figure 7b). At TP = 20.2 μg L−1 (95% confidence interval = 13.8 μg L−1 − 29.0 μg L−1, based on the piecewise regression), there is a significant shift in the TP-DOC relationship (Figure 7a), wherein both variables become more loosely coupled and DOC increases at a slower rate than TP (log slope = 0.77), leading to higher TP:DOC values as TP increases in this part of the gradient. The relationship between pCO2 and TP also drastically shifts at a comparable TP value (14.8 μg L−1; 95% confidence interval = 13.1 μg L−1 – 16.8 μg L−1), switching from positive to negative. Therefore, at low TP:DOC ratios, there is a positive relationship between pCO2 and both DOC (Figure 7) and TP (Figure 7b), but the slopes become shallower, and even negative, as TP increases disproportionately relative to DOC (Figure 7a).

Figure 7.

Piecewise regression of (a) DOC and (b) pCO2 against TP. Figure 7a: TP < 20.2 (black symbols): log(DOC) = 1.22 log(TP) – 0.38, r2 = 0.34, p < 0.001; TP > 20.2 (white symbols): log(DOC) = 0.77 log(TP) – 0.15, r2 = 0.11, p = 0.03. Figure 7b: TP < 14.8 (black symbols): log(pCO2) = 0.83 log(TP) + 2.15, r2 = 0.24, p < 0.001; TP > 14.8 (white symbols): log(pCO2) = −0.99 log(TP) + 4.30, r2 = 0.17, p < 0.001. “Published” data for Figure 7a originate from Bergström et al. [2004], del Giorgio et al. [1999], Jonsson et al. [2003], Kritzberg et al. [2005], and Finlay et al. [2009]. “Published” data for Figure 7b originate from Bergström et al. [2004], del Giorgio et al. [1999], Hope et al. [1996], Jonsson et al. [2003], Kelly et al. [2001], Kritzberg et al. [2005], and Finlay et al. [2009]. All data have been analyzed together in the regression. Dash-dot line represents the break point, and the dotted bars represent the confidence interval (95%).

4. Discussion

[21] The dynamics of CO2 in lakes have received increasing attention in the past decade, and many different factors have been linked to lake pCO2 and CO2 fluxes. One of the few recurrent patterns across landscapes that have been identified is the relationship between surface water pCO2 and DOC concentration [Roehm et al., 2009]. Inspection of Table 2 quickly reveals that while these relationships exist for many regions, there is a large degree of variability in their strength and shape, and thus the interpretation of the underlying basis of this relationship is far from straightforward. We contend that the variability in the pCO2-DOC relationship reflects fundamental environmental gradients that can only be detected on a comparative basis over very large geographical scales, and there is now a sufficient mass and diversity of available data to allow this type of comparison. In this regard, our own results combined with the existing published data, when placed together in a broader context of northern landscapes, demonstrate that the different regional relationships in fact fit a cross-regional pattern that appears to be driven by a combination of geographic, morphometric and biogeochemical properties.

[22] Combining our own and published data allowed us to explore two basic scenarios that could account for the contrasting parameters of the regional pCO2-DOC relationships: regional differences in the amount and relative proportions of DOC and CO2 exports from land to lakes, and regional patterns in lake trophic status that may mediate the above patterns in C and nutrient loadings. In order to explore these scenarios, we used the slopes and calculated regional pCO2 baselines from the regional pCO2-DOC relationships as response variables and found that they were indeed well explained by proxies of trophic status and terrestrial influence, i.e., the TP:DOC ratio and lake elevation, respectively. The strength of the inter-regional relationships presented here (Figures 3 and 6), despite all the unaccounted variability induced, for example, by differing DOC composition, land use, lake mixing regime, climate, inter-seasonal and inter-annual patterns across the different studies, as well as by the inclusion ofpCO2 values obtained from different measurement techniques, suggest that these basic relationships effectively capture some of the key mechanisms that regulate lake pCO2 in northern landscapes.

4.1. The pCO2 Regional Baseline and Elevation

[23] Recent studies have provided evidence for a strong hydrologic control of lake pCO2 that is decoupled from net ecosystem production [Dubois et al., 2009; Stets et al., 2009], partly due to imports of terrestrially produced CO2 [Humborg et al., 2010; Jones et al., 2001; Walvoord and Striegl, 2007]. If this DIC subsidy were dependent on region-specific drivers, it would induce cross-regional variation in the concentration of CO2 for a given amount of DOC, regardless of the slope of the relationship between both variables. This pattern was indeed observed in the relationships we presented here (Figure 2), and we further showed that the regional pCO2 baseline was positively correlated to mean regional elevation (Figure 3).

[24] This result may seem counter-intuitive, as terrestrial net primary production, and the associated absolute exports of DOC and CO2 from land to water, tend to decrease with altitude [Jansson et al., 2008]. At higher elevation, however, aquatic ecosystems also tend to be more closely coupled to their surrounding landscape and thus more responsive to catchment processes. Riverine networks, especially small, headwater streams, carry large quantities of CO2 [Teodoru et al., 2009; Wallin et al., 2010], much of which originates from terrestrial respiration [Johnson et al., 2008]. Landscape position and topography are further related to key aspects of the delivery of water and C to lakes, including water transit time [McGuire et al., 2005], catchment and lake area (Figures 4 and 5) and position in the catchment [Tetzlaff et al., 2011], groundwater flow [Jones and Mulholland, 1998], and type of soils and land cover [Tetzlaff et al., 2009]. These elements favor higher hydrologic responsiveness in higher elevation catchments [Tetzlaff et al., 2011], resulting in shorter residence time and dominance of terrestrial over aquatic processes in determining the amount of organic and inorganic carbon circulating in surface waters. In this context, we interpret elevation as an integrative variable reflecting a large number of factors that couple lakes to the surrounding landscape. Higher proportions of terrestrially derived CO2 are thus expected to circulate in higher altitude watersheds with smaller catchments, independent of aquatic DOC content (Figure 4), whereas in lower elevation lakes with larger catchments in situ lake processes play a larger role in regulating surface water CO2 levels.

[25] Our results further indicate that elevation is not only linked to the watershed properties, but also correlates to the regional lake size distribution. On average, lakes in higher elevation landscapes tend to be smaller, and this pattern further has the potential to influence both DOC and pCO2 (Figure 5). Similar to higher altitude watersheds, smaller lakes tend to be more responsive to the delivery of organic and inorganic C from the catchment, due to higher drainage ratios [Hope et al., 1996], lower residence time [Kelly et al., 2001] and a higher relative contact surface with the basin. All these features result in higher terrestrial C import per unit lake volume or area and thus a higher potential for terrestrially derived C to influence both the DOC and the CO2 dynamics in these lakes. Conversely, larger lakes with longer residence times are expected to have smaller ranges of variation due to their capacity to buffer shifts in hydrology and C loadings through longer renewal time [Roehm et al., 2009]. Larger lakes are also exposed to stronger winds and have higher rates of degassing for a given wind speed (D. Vachon, personal communication), which would tend to dampen pCO2 levels.

[26] The idea of a direct terrestrial origin of lake pCO2 is not, however, in contradiction with the widespread notion that net heterotrophy supports excess CO2 in boreal lakes [del Giorgio et al., 1999; Duarte and Prairie, 2005; Sobek et al., 2003]. Together, these various studies rather show that there are several pathways supporting lake pCO2, and suggest that the balance between the contribution of these pathways differs not only among lakes, but also more broadly across regions. In this regard, it appears that high regional pCO2 baselines are related to a higher degree of coupling between lake and catchment processes, and this coupling is reflected here in the relationships of catchment and lake area with lake surface water pCO2.

4.2. Slope and TP:DOC Ratio

[27] The main reason DOC is an overall good predictor of pCO2 at the regional scale is because it acts both directly as a substrate for the production of CO2 within lakes, and as a proxy for variables that also influence pCO2 [Larsen et al., 2011]. Our results suggest that this is also the case at a cross-regional scale, over the temperate and boreal landscapes, but in a different manner. The “substrate” role of DOC would lead to the positive relationship that is generally found, while the “proxy” role would modulate the regional differences in this relationship. The drastic decrease in the predictive power of DOC at the cross-regional scale (Table 2) supports this view, and highlights the emerging role of associated drivers of pCO2 as the spatial scale broadens, especially if the relationship between DOC and these drivers differs across regions. For example, in the case of negative pCO2-DOC relationships, it is clear that more DOC cannot result in lesspCO2 per se, and that this pattern is driven by other variables and related processes.

[28] In this context, Figure 7 illustrates the importance of the relationship between DOC and TP in shaping pCO2dynamics. Because DOC and TP tend to co-vary, they also both tend have positive individual relationships withpCO2 (left portion of Figure 7). However, in the higher end of the TP gradient, where TP increases faster than DOC and the TP:DOC ratio is highest (past TP = 20 μg L−1 Figure 7a), the relationship between pCO2 and TP becomes negative (Figure 7b), and the increase in pCO2 resulting from any given DOC increase becomes gradually lower, and eventually negative at high TP:DOC (Figure 6a). This suggests that TP and DOC are strong, opposite drivers of pCO2, and that the manner in which they co-vary has a major impact on lake CO2 dynamics, both at the intra (Figure 6b) and an inter-regional (Figure 6a) scales.

[29] Our results suggest that the TP:DOC ratio may act as a proxy for the balance between primary production and respiration, and thus support the widespread idea that lake ecosystem metabolism is an important driver of lake pCO2 [Duarte and Prairie, 2005]. Our results further suggest that there are regional patterns in lake net metabolic balance that could account for the differences observed in pCO2 dynamics across northern landscapes, more specifically, in modulating the pCO2-DOC relationships. The relatively steep slope of the DOC-TP relationship below 20 ug L−1 suggests that terrestrial exports of organic matter constitute the major P source in more pristine lakes situated in less perturbed regions, while the weaker slope above 20 ug L−1may reflect the increasing contribution of inorganic P due to anthropogenic and other sources. In lakes with a higher average TP:DOC ratio, DOC-driven increases in respiration and photo-production of CO2would thus be compensated by P-driven increases in primary production, and the role of DOC as a substrate for mineralization would be secondary to its role as a proxy of nutrient inputs from the catchment, leading to the negative slopes of thepCO2-DOC relationships in the most productive regions. In addition, as TP increases and lakes become more productive, an increasing fraction of the ambient DOC available to respiration would be of autochthonous origin and would not contribute to further CO2super-saturation. Those combined effects would tend to maintain lower surface waterpCO2 for a given concentration of DOC.

[30] The intraregional patterns observed within the overall pCO2-TP relationship support this argument. On an individual basis, the studies that have measured TP concentrations beyond the thresholds presented inFigure 7 have been conducted in four regions: Abitibi and James Bay (this study), the Eastern Townships region [del Giorgio et al., 1999], and the Canadian Prairies [Finlay et al., 2009]. Each area is characterized by either anthropogenic perturbations or natural features in their landscapes that favor increased P but not necessarily DOC loadings, such as agriculture [Abell et al., 2011; Meriläinen et al., 2000], urban development [Schindler, 2006], forestry [Pinel-Alloul et al., 2002], or beaver dams [Roy et al., 2009]. As opposed to the significant positive relationship found in the lower part of the TP gradient (Figure 7b), the relationship between pCO2 and TP was not significant for the Abitibi and James Bay lakes, and was strong and negative for the other two regions mentioned (p < 0.001). These observations further support the conclusion that the role of DOC as a predictor of pCO2 is modulated by regional trends in lake metabolic balance, which are in turn driven by regional patterns in the proportion of the terrestrial loadings of TP and DOC.

4.3. Conclusions and Implications

[31] Our analysis shows that what appear to be scattered and sometimes contradictory results in fact yield a coherent portrait of the regulation of lake pCO2 in northern landscapes when placed together along common environmental and geographical gradients. The differences in the pCO2-DOC relationships appear to be driven by the same key factors which are integrated by the mean regional TP:DOC ratio and elevation. We found that proxies of lake metabolic balance and terrestrial CO2 exports account for a large portion of these patterns in the lake pCO2-DOC relationships, but additional underlying, correlated factors not reported here, such as regional patterns in DOC composition due to contrasting terrestrial influence and trophic status, also likely play a role and may further account for significant portion of the remaining, unexplained variability. Nonetheless, our results provide a synthetic view of the role of terrestrial exports of organic and inorganic nutrients and carbon in regulating northern lakes CO2 dynamics, and illustrate the importance of taking background regional geographic and environmental patterns into account at the time of evaluating how future changes in environmental and climatic conditions may affect CO2 dynamics in inland waters in northern landscapes.

Acknowledgments

[32] We thank two anonymous reviewers for constructive comments and P. Massicotte for help with analyses. J.-F.L. was supported by a doctoral scholarship from the National Science and Engineering Research Council of Canada (NSERC). This project is part of the program of the Industrial Research Chair in Carbon Biogeochemistry in Boreal Aquatic Systems (CarBBAS), and was co-funded by the National Science and Engineering Research Council of Canada (NSERC) and Hydro-Québec.