Abundant pre-industrial carbon detected in Canadian Arctic headwaters: implications for the permafrost carbon feedback

It is estimated that Arctic permafrost soils contain between 1330 and 1580 Pg organic carbon (C) (Schuur et al 2015), equivalent to roughly half the total fossil fuel C emissions since the start of the Industrial Revolution (~AD1750, Boden et al 2010). Rising air temperatures in the Arctic are warming permafrost soils, which can lead to deepening of the seasonally thawed 'active' layer (AL), and exposure of previously frozen organic matter to in-situ mineralization and lateral transport (Schuur et al 2009, Vonk et al 2013a 2013b, Abbott et al 2016). The organic C stored in permafrost can be up to, or greater than, ~50 000 years old, having been fixed into plant matter during past snow-free seasons and accumulating in soils (Mann et al 2015).

Arctic headwater streams, ponds and lakes are closely connected to the landscape via hydrological flow paths, both along the surface and through the AL (Paytan et al 2015, Dean et al 2016), and consequently carry signals of change in permafrost conditions (Neff et al 2006, Raymond et al 2007). Old C (defined here as C assimilated into plant material from the atmosphere and sequestered in organic-rich soils prior to the 1950s) has proven notoriously difficult to detect in aquatic systems due to the overwhelming presence of 'modern' C (assimilated since 1950). Post-1950s C has a significantly enriched ¹⁴C signature as a result of atmospheric nuclear weapons testing (Dutta 2016). This ¹⁴C enrichment has severely hindered attempts to provide high resolution dating of C assimilated over the past ~200 years, especially when the material is derived from multiple years (Evans et al 2014). Only systems affected by significant permafrost thaw, usually in the form of rapid thaw-related erosion (thermokarst), have shown clear signals of old dissolved or particulate organic C (DOC and POC, respectively) being mobilized (Vonk et al 2015).

Recent studies highlight headwater systems as hotspots for preferential mineralization of old DOC during transport (Drake et al 2015, Spencer et al 2015), resulting in the release of greenhouse gases (GHGs) driving the 'permafrost carbon feedback' (PCF, Walter Anthony et al 2016). However, the release of GHGs derived from old permafrost C has only been inferred by aquatic ¹⁴C studies of DOC and POC, with the few studies that consider aquatic GHG ¹⁴C signatures themselves focusing only on ebullition fluxes, and none combining these measurements (Negandhi et al 2013, Bouchard et al 2015, Walter Anthony et al 2016).

To address this, we measured the ¹⁴C content of CO₂ alongside DOC in a range of inland headwaters (streams, ponds and lakes) in the western Canadian Arctic to determine the contribution of old permafrost C to these C fluxes. Additional sampling of ¹⁴CH₄ at two locations supplemented the assessment of C sources in GHG fluxes from headwaters draining permafrost soils. ¹⁴CO₂ and ¹⁴CH₄ samples were collected using newly developed techniques, and alongside DO¹⁴C, represent the first such combined aquatic ¹⁴C measurements in the Arctic. Previous work at the study site showed that the main pathway of water delivery to the streams and lakes is via the organic-rich soils, which can be up to 50 cm deep and contain maximum bulk organic C ages of 2200–6200 years B.P. (Quinton and Pomeroy 2006, Dean et al 2016). Here, we focus on the aquatic mobilization of C assimilated prior to the Industrial Revolution, as it is demonstrably older than the AD1950 'old C' cut-off.

2.1. Sample collection and analysis

Samples were collected from a series of headwater systems (six lake sites, four stream sites and four pond sites) from a ~14 km² area in the continuous permafrost zone of the Northwest Territories, Canada (68° 44^' 54.5^'' N, 133° 29^' 41.7^'' W; figure 1). The main study catchment within this area is Siksik Creek (0.94 km²), which forms part of the Trail Valley Creek catchment. The study site is an upland tundra, consisting of hummock and inter-hummock morphology, where the inter-hummocks are loosely filled by organic matter and act as the dominant hydrologic pathway in the system; the hummocks are primarily formed from mineral soils (Quinton and Pomeroy 2006, Dean et al 2016). The aquatic biogeochemistry of the site is dominated by the organic inter-hummocks (Quinton and Pomeroy 2006, Dean et al 2016).

The catchments are predominately covered in dwarf and low shrubs, sedges (Eriophorum and Carex spp.), bryophytes and lichens, with some tall shrub patches (Alnus viridis and Betula spp.) on the hillslopes. The riparian zones are vegetated with Betula glandulosa and Salix spp. The region belongs to the Circumpolar Arctic Vegetation Map (Walker et al 2005) bioclimate sub-zone E, and is dominated by low-shrub tundra (map unit S2) out-with riparian zones. While the study site is considered to be relatively pristine (Dean et al 2016), the annual average air temperature in Inuvik (c. 45 km south of the study site) has increased by 2.5 °C since 1970, and active layer depths at Illisarvik on Richards Island, c. 120 km north-northwest of the site, have increased by around 8 cm since 1983 (Burn and Kokelj 2009).

Aquatic sampling took place in the snow free season of 2014. Repeated sampling took place within Siksik creek catchment, with the main stream and two polygon sites sampled five times throughout the season (figure 1). A series of lake and polygon sites outside Siksik catchment were sampled once in August 2014 (figure 1). Samples for DOC, CO₂ and CH₄ concentrations and isotopic content were collected at each site (see supplementary methods; Dean et al 2016). 500 ml water samples were collected and filtered for DO¹⁴C analysis (n = 26). ¹⁴CO₂ (n = 36) and ¹⁴CH₄ (n = 2) samples were collected by equilibrating sample waters with a large headspace, following Garnett et al (2016a), (2016b)—see supplementary methods. Low CH₄ concentrations prevented ¹⁴CH₄ sample collection on all but two occasions (figure 2(c); Dean et al 2017).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

2.2. Age distribution analysis

Aquatic ¹⁴C samples represent an integrated signal from the upstream catchment area, and therefore a source-apportionment approach (such as an age distribution) should be used to describe the relative contributions of C from different soil layers and locations within the catchment (Sierra et al 2017). We therefore applied the approach described below to represent soil C export dynamics at the catchment scale.

At the end of each growing season, the organic matter from senescing vegetation contributes to the upper soil layers, although deep rooting of vegetation will contribute C to a range of depths, both through exudations and root turnover (meaning that age estimates derived with an exponential distribution are conservative, as dilution with recent C is deeper than assumed by this approach). This occurs each year, and results in the formation of organic soil horizons across the Arctic tundra. Each year's input of plant-derived organic matter loses C through the activities of detritivores and decomposers, as well as through leaching, exporting progressively less C per year over time as the residual organic matter tends to become more recalcitrant. Here, based upon previous work (Raymond et al 2007, Moore et al 2013, Evans et al 2014), we chose an exponential distribution to represent these soil dynamics (equation (1)), assuming that the largest C contributions come from the upper youngest soil layers, and that the contribution from each successive older soil layer decreases exponentially as they become thinner through time (Raymond et al 2007, Moore et al 2013, Evans et al 2014)—note that changes in lability of C at increasing depth can also be anticipated, but defining a simple relationship may be elusive, based on current understanding (Jansson and Taş 2014, Ward and Cory 2015) (calculations made in R, version 3.3.2; figures 2(a) and (b)).

Table 1. ¹⁴C values for each sampling point (percent modern C [pmC] and conventional ¹⁴C age in years B.P.), and associated numerical age distribution outputs ('solution'—years before sampling date [years BSD]); uncertainties in the pmC values reflect the uncertainty from the AMS instrument and sample preparation; uncertainties in the age distribution output reflect the propagation of this error into the age distribution analysis.

			14CO2				DO14C
					Age distribution (years BSD)				Age distribution (years BSD)
Site	Water type	Sampling date (2014)	14C (pmC ± 1σ)	Conv. Age (y B.P.)	Younger solution	Older solution	14C (pmC ± 1σ)	Conv. Age (y B.P.)	Younger solution	Older solution
S1	Stream	12/06	105.06 ± 0.51	modern	no sol.	120 ± 63	105.53 ± 0.49	modern	3 ± 58	113 ± 51
S3	Stream	12/06	108.52 ± 0.49	modern	7 ± 1	76 ± 5	106.45 ± 0.53	modern	4 ± 1	100 ± 7
S5	Stream	12/06	—	—	—	—	106.87 ± 0.50	modern	5 ± 1	94 ± 6
S7	Stream	12/06	108.97 ± 0.49	modern	7 ± 1	71 ± 5	109.64 ± 0.48	modern	8 ± 1	65 ± 4
L17	Polygon	12/06	106.72 ± 0.47	modern	5 ± 1	96 ± 6	104.45 ± 0.49	modern	no sol.	131 ± 9
L19	Polygon	12/06	106.66 ± 0.48	modern	5 ± 1	97 ± 6	106.16 ± 0.49	modern	4 ± 1	104 ± 7
S1	Stream	13/07	103.88 ± 0.48	modern	no sol.	142 ± 9	—	—	—	—
S3	Stream	13/07	104.57 ± 0.48	modern	no sol.	129 ± 9	—	—	—	—
S5	Stream	12/07	104.70 ± 0.48	modern	no sol.	126 ± 9	—	—	—	—
S7	Stream	12/07	105.45 ± 0.49	modern	3 ± 59	114 ± 51	—	—	—	—
L17	Polygon	12/07	104.40 ± 0.46	modern	no sol.	132 ± 9	—	—	—	—
L19	Polygon	12/07	104.16 ± 0.48	modern	no sol.	137 ± 8	—	—	—	—
S1	Stream	28/07	103.95 ± 0.46	modern	no sol.	140 ± 9	—	—	—	—
S3	Stream	28/07	103.84 ± 0.48	modern	no sol.	142 ± 9	—	—	—	—
S5	Stream	28/07	103.10 ± 0.45	modern	no sol.	158 ± 10	—	—	—	—
S7	Stream	28/07	104.56 ± 0.48	modern	no sol.	129 ± 9	—	—	—	—
L17	Polygon	25/07	104.90 ± 0.48	modern	no sol.	123 ± 64	—	—	—	—
L19	Polygon	26/07	104.02 ± 0.46	modern	no sol.	138 ± 9	—	—	—	—
S1	Stream	30/08	103.62 ± 0.48	modern	no sol.	147 ± 9	100.64 ± 0.88	modern	no sol.	219 ± 25
S3	Stream	30/08	104.20 ± 0.45	modern	no sol.	136 ± 8	100.66 ± 0.88	modern	no sol.	218 ± 25
S5	Stream	30/08	104.86 ± 0.49	modern	no sol.	123 ± 65	101.07 ± 0.88	modern	no sol.	210 ± 24
S7	Stream	30/08	104.89 ± 0.49	modern	no sol.	123 ± 64	101.57 ± 0.89	modern	no sol.	195 ± 23
L17	Polygon	29/08	102.17 ± 0.47	modern	no sol.	179 ± 12	97.37 ± 0.85	214 ± 70	no sol.	336 ± 37
L18	Polygon	29/08	103.01 ± 0.45	modern	no sol.	160 ± 10	100.73 ±0.88	modern	no sol.	216 ± 25
L19	Polygon	29/08	104.23 ± 0.48	modern	no sol.	135 ± 9	100.76 ± 0.88	modern	no sol.	219 ± 25
L1	Lake	01/09	104.43 ± 0.48	modern	no sol.	131 ± 9	101.59 ± 0.89	modern	no sol.	194 ± 23
L2	Lake	01/09	104.78 ± 0.49	modern	no sol.	125 ± 65	100.27 ± 0.87	modern	no sol.	230 ± 26
L3	Lake	01/09	100.94 ± 0.47	modern	no sol.	214 ± 12	101.05 ± 0.88	modern	no sol.	210 ± 24
L5	Polygon	01/09	105.44 ± 0.49	modern	3 ± 59	114 ± 51	99.72 ± 0.87	22 ± 70	no sol.	248 ± 28
L6	Lake	01/09	105.15 ± 0.49	modern	no sol.	119 ± 62	100.76 ± 0.88	modern	no sol.	219 ± 25
L7	Lake	02/09	104.09 ± 0.48	modern	no sol.	138 ± 9	100.94 ± 0.88	modern	no sol.	214 ± 24
L9	Lake	02/09	98.03 ± 0.45	160 ± 37	no sol.	311 ± 18	98.98 ± 0.86	82 ± 70	no sol.	277 ± 32
S1	Stream	08/09	—	—	—	—	102.16 ± 0.89	modern	no sol.	179 ± 23
S3	Stream	08/09	103.48 ± 0.45	modern	no sol.	149 ± 9	100.67 ± 0.88	modern	no sol.	218 ± 25
S5	Stream	08/09	104.25 ± 0.45	modern	no sol.	135 ± 9	102.11 ± 0.89	modern	no sol.	181 ± 23
S7	Stream	08/09	105.53 ± 0.46	modern	3 ± 58	113 ± 51	103.06 ± 0.90	modern	no sol.	158 ± 20
L17	Polygon	08/09	102.96 ± 0.48	modern	no sol.	161 ± 11	97.15 ± 0.85	233 ± 70	no sol.	345 ± 35
L19	Polygon	08/09	103.35 ± 0.48	modern	no sol.	152 ± 10	98.98 ± 0.86	83 ± 70	no sol.	277 ± 32
			14CH4
					Age distribution (years BSD)
Site	Water type	Sampling date (2014)	14C (pmC ± 1σ)	Conv. Age (y B.P.)	Young solution	Old solution
S3	Stream	13/07	100.78 ± 1.00	modern	no sol.	218 ± 57
L17	Polygon	25/07	72.23 ± 0.89	2613 ± 102	no sol.	2168 ± 199

modern = no conventional radiocarbon age possible when >100 pmC; no sol. = no younger solution available from the age distribution analysis.

A more sophisticated distribution is not necessarily better due to the complex and highly heterogeneous soil age profiles that exist within even small stream catchments, and therefore detailed soil profile data can be challenging to extrapolate at the catchment scale in a way that is relevant to hydrological export (Hrachowitz and Clark 2017). We assume that the youngest C contribution is from the 2013 soil layers, as the majority of soil organic matter accumulation occurs at the end of the growing season during plant senescence (Raymond et al 2007), but incorporating potential contributions from the sampling year (2014 in this study) into these distributions had a negligible impact on the derived median ages. We assigned each soil layer a ¹⁴C value in pmC based on atmospheric ¹⁴CO₂ values during past northern hemisphere summers (NH zone 1) (Hua et al 2013, Levin et al 2013, Reimer et al 2013), representing the ¹⁴C signature of the CO₂ fixed into vegetation during the growing season. Published records only go up to 2012; values for 2013–14 were generated assuming the same rate of depletion in atmospheric ¹⁴CO₂ as occurred between 2011 and 2012. This approach assumes unlimited organic matter depth in the soil profile; we did not assign a maximum basal age as in Moore et al (2013) and Evans et al (2014), but no sample's age distribution in this study incorporated any C older than the known maximum age at the site (2200–6200 years B.P), with the exception of the oldest CH₄ sample (table 1).

To determine the exponential distribution for the measured ¹⁴C content of each sample, we solved equation 1 for λ (the rate parameter):

$$\begin{equation} {\rm{pm}}{{\rm{C}}_{{\rm{stream}}}}\left( t \right) = \,\mathop \smallint \limits_0^{{\rm{inf}}} \lambda {e^{ - \lambda T}}\,{\rm{pm}}{{\rm{C}}_{{\rm{air}}}}\left( {t - T} \right)\,{2^{\frac{T}{{5730}}}}\,{\rm{d}}T \end{equation} \tag{ 1 }$$

where pmC_stream is the aquatic sample's measured ¹⁴C content in percent modern carbon (pmC), pmC_air is the atmospheric ¹⁴CO₂ content in pmC at time (t), and T is the age of the sample in 'years before sampling date' (years BSD; i.e. years before 2014 in this study), and should not be confused with conventional radiocarbon ages reported in years B.P. All aquatic ages in the manuscript are given in years BSD based on the distributions, with years B.P. values confined to the conventional ¹⁴C reporting in table 1 and the supplementary information available at stacks.iop.org/ERL/13/034024/mmedia. Note that λ can have two solutions in some circumstances (see section 3).

Each sample of DOC, CO₂ and CH₄ had a bulk ¹⁴C value in pmC which can be described by a single exponential distribution of soil ¹⁴C contributions (figure 2(b)). We report the median value of the distribution (years BSD, figure 2(b)), which has an intuitive interpretation: Half of the C is younger and half of the C is older than this median age. The mean of an exponential distribution gives an older age than the median by a factor of 1.44 (equation (2)).

$$\begin{equation} {\rm{Median}}\,\,{\rm{Age}} = \,{\lambda ^{ - 1}}\,{\rm{ln}}\left( 2 \right). \end{equation} \tag{ 2 }$$

We incorporate the analytical uncertainty of each individual sample's ¹⁴C value (table 1) into a median uncertainty range (figure 3, table 1). In some cases, both a young and an old distribution can describe a given sample's ¹⁴C content. In figure 3 and in the statistical analyses we use only the old distributions due to the prevalence of old C in the system throughout the study period (see section 3).

In June, following the spring snow melt, nine out of the 11 samples have ¹⁴C values (in pmC) that are only slightly above the atmospheric ¹⁴CO₂ signature at the time of sampling (103.12 pmC for this study). This means that the ¹⁴C values of these samples can represent C assimilated either side of the atmospheric ¹⁴C bomb peak (figure 2(b)). Therefore, two possible exponential distributions can describe their ¹⁴C value, giving both a younger (i.e. 3–8 years old) and an older (i.e. > 70 years old) solution. The aquatic C in the study system in June could therefore be very young, sourced primarily from net primary production in the last few years (Raymond et al 2007) or from C fixed predominantly just prior to the bomb-peak (Evans et al 2014). In some cases (DOC, n = 1; CO₂, n = 8), the analytical uncertainty propagated into the age distributions results in a high uncertainty as the distribution cannot distinguish between the old and young solutions (figure 3).

For the June samples, we cannot definitively conclude that one solution is correct over the other. Raymond et al (2007) concluded that the younger solution is most likely following the spring melt, as the AL is shallow, and snow melt will flush out a pulse of modern C from recently accumulated material and microbial biomass in the upper unfrozen soil layers. However, two samples at this time point (one DOC and one CO₂, but from different sites) cannot have a younger solution, indicating that there is significant C in these samples derived from pre-bomb peak material—this may be due to flushing of old C mobilized at the end of the previous season during the snow melt. After this time point, no younger distributions can describe the DO¹⁴C samples, and only three ¹⁴CO₂ samples can have younger distributions (table 1). Evans et al (2014) suggest that the younger solution in these scenarios is more likely given the undisturbed history of the site (Dean et al 2016). But based on our results showing that younger solutions are not possible for the substantial majority of samples after June, we assume the older solutions are also more likely for the early season samples given that this assumption yields a smooth increase in age over the season, as opposed to a jump in age between early and late season. A jump in age from predominantly young C being exported in June to older C later in the season is not reflected in the seasonal evolution of the DOC, CO₂ or CH₄ concentrations (supplementary figure S2), or the overall aquatic biogeochemistry (Dean et al 2016). Further, the positive relationship between δ¹³C and CO₂ age (R² = 0.40, P < 0.001; figure 2(c)) also indicates that some of the CO₂ in the water samples is produced at depth (this increase in δ¹³C with depth is due either to the mineralization of C deeper in the soil profile that generally has a higher δ¹³C value, or isotopic fractionation by methanogenesis in the deeper, less oxic soil layers; Clymo and Bryant 2008, Corbett et al 2013, Hicks Pries et al 2013), supporting the exclusion of the younger distributions as deeper peat layers will generally contain older C. However, even if the younger solutions are assumed to be correct in the June samples, this would only further emphasize the seasonal evolution of C age exported by the study waters seen in figure 3. Further, it should be noted that the old solutions still contain a substantial proportion of young C (figure 2(b)).

Thus, earlier in the season, young C may form a significant component of the aquatic DOC and CO₂ export, but the seasonal evolution of the DO¹⁴C and ¹⁴CO₂ in the system demonstrates that old C becomes more prevalent later in the season (figures 3 and 4). In the following analysis, we assume that the older solutions are true.

Zoom In Zoom Out Reset image size

The ¹⁴C age distributions indicate the prevalence of old C in most samples collected in this study (figure 2(c), table 1). DOC and CO₂ ages were youngest in June following the spring snow-melt, and were not significantly different (P = 0.27, figure 3). By August/September, on average, CO₂ ages had risen from 92–151 years BSD (59%–63% increase, confidence interval: 37%–94%, see supplementary methods), while DOC ages rose from 101–228 years BSD (120%–125% increase, confidence interval: 70%–185%), showing an increase over the season that is more pronounced in the DOC than CO₂ (figure 3).

However, the seasonal increase in C age was not matched by an equivalent trend in aquatic DOC, CO₂ or CH₄ concentrations at the sites sampled for ¹⁴C analyses in 2014 (supplementary figure S2), or across the entire study area in 2013–2014 (Dean et al 2016). The observed age increase represents proportionately higher contributions from older soil layers later in the season that would be undetectable without ¹⁴C.

The majority of C in both CO₂ and DOC in our samples (60%–85%) was fixed in the industrial era post ~AD1750 (figure 4). However, the modeled relative contribution of C fixed in the pre-industrial era increased from 15% to a maximum of 40% over the growing season, demonstrating the mobilization of old C in both DOC and CO₂ as the thaw front deepened. Seasonal increases in contributions of C from older soil layers to respired CO₂ have also been observed in the Alaskan tundra (Hicks Pries et al 2013), demonstrating parallels between terrestrial and aquatic CO₂ export in permafrost zones. This contrasts with recent work in an organic-rich permafrost-free boreal catchment in northern Sweden, where little seasonal variability was seen in the age of aquatic DOC, CO₂ and CH₄ export, showing that aquatic C export from non-permafrost catchments can differ significantly (Campeau et al 2017a).

There is a significant positive correlation between DOC and CO₂ age (P < 0.01; R² = 0.38), but DOC is generally older than CO₂, and there is a clear difference in δ¹³C between CO₂ and DOC samples (figure 2(c)). It is possible that CO₂ is younger than DOC due to the intrusion of atmospheric CO₂ into the study waters, and that ¹³C fractionation may occur during this process (Doctor et al 2008). However, aquatic CO₂ concentrations were consistently much higher (by a factor of ~11) than atmospheric values (Dean et al 2016), suggesting that CO₂ predominantly moves from the water to the atmosphere, rather than the reverse (and by convention, isotopic fractionation of ¹⁴C due to invasion is corrected for). The increasing trend between CO₂ age and δ¹³C (figure 2(c)) indicates a positive relationship between organic C age and its production at depth (see section 3, Clymo and Bryant 2008, Corbett et al 2013, Street et al 2016). Previous work at the study site indicated that there was limited input to the aquatic biogeochemistry of the site from weathering of the local bedrock and tills (Dean et al 2016), and there is no evidence of a carbonate contribution to the ¹⁴CO₂ samples collected in this study (supplementary figure S1), supporting the conclusion that this relationship between ¹⁴C and δ¹³C in the CO₂ samples is caused by the production of CO₂ at depth in the soil profile.

DOC likely represents C sourced from the entire thawed soil profile due to the vertical infiltration of water that then moves laterally through the soils until discharged into open water. CO₂, however, is predominantly produced by soil organic matter metabolism in the younger, unsaturated soil layers (Hicks Pries et al 2016, Street et al 2016), with some contribution of CO₂ generated in the older saturated soil layers; increased CO₂ generation from the older deeper layers results in higher δ¹³C values compared to δ¹³C-DOC (Clymo and Bryant 2008, Corbett et al 2013, Hicks Pries et al 2013). The indication of organic matter degradation at depth in this study is further evidence for the presence of old C in bulk DOC and CO₂ samples, with relative contributions of C from deeper soil layers increasing through the season as the AL deepens (figures 3 and 4).

Only two ¹⁴CH₄ samples could be collected during the sampling campaign due to highly heterogeneous and often low dissolved CH₄ concentrations in this upland tundra study system (supplementary figure S2; Dean et al 2016, Street et al 2016). A sample collected from a deep ice-wedge polygon (which exposed the oldest soil layers) was much older (2168 ± 199 years BSD) than a stream sample (218 ± 57 years BSD), but both contained a large pre-industrial C component (43%–92%, table 1). This indicates that old C may be exported from the system as CH₄, although both the age and flux may be highly variable.

The aquatic ¹⁴C ages presented here clearly demonstrate that, as the thaw front advances deeper into the soil profile during the thaw season, older C is mobilized into inland waters as both DOC and CO₂. We also demonstrate the potential for aquatic mobilization of old dissolved CH₄. The export of aged C as aquatic CO₂ is an important finding as direct release of CO₂ can form part of the PCF, whereas the potential for aged DOC to contribute to the PCF depends on its fate during freshwater transport and residence in the ocean (Vonk et al 2015).

This first look at dissolved CO₂ age dynamics in Arctic headwaters presented here shows that it generally resembles DOC export (see section 4). However, the disconnect between DOC and CO₂ age and δ¹³C suggests that little of the aquatic CO₂ was generated from DOC mineralization in the aquatic zone. The CO₂ is, therefore, most likely generated in the soil profile and transported laterally, in aqueous phase, into the headwater systems (Campeau et al 2017b). This contrasts with previous studies that demonstrated the rapid mineralization of old DOC (Vonk et al 2013a, Drake et al 2015, Mann et al 2015, Spencer et al 2015); however, these studies were from Arctic regions with extremely old ice-rich yedoma deposits containing C that can be ~50 000 years B.P. or older, whereas the oldest C observed in the present study area was 2200–6200 years B.P. Therefore, inland water ¹⁴C age distributions may be significantly different in yedoma regions.

POC, in comparison, is generally older than DOC in aquatic systems (Guo et al 2007, Marwick et al 2015), and can form a substantial component of the carbon exported by Arctic catchments affected by permafrost thaw, particularly where thermokarst activity is occurring (Vonk et al 2015). Thermokarst activity was not observed at our study site, so may not be a major component of C export in the study catchments. However, to fully account for the age of C exported from Arctic aquatic systems, it is important to consider PO¹⁴C as well as more extensive ¹⁴CH₄ measurements in future studies.

For pan-Arctic regions where modern aquatic ¹⁴C ages are common, the results presented here suggest that up to 30% of CO₂ and 40% of DOC aquatic fluxes (figure 4), equivalent to 16 (range: 6–25) and 10 (range: 5–14) Tg C yr⁻¹, respectively (McGuire et al 2009, Holmes et al 2012, Vonk et al 2015, Wik et al 2016), may be derived from organic matter formed in the pre-industrial era. These values can be considered conservative estimates, as the proportion of pre-industrial C observed in this study does not represent POC or CH₄ export, or areas that are affected by more severe permafrost thaw and thermokarst (Olefeldt et al 2016), where a larger proportion of old C would be expected in aquatic C fluxes due to the exposure of deep organic matter, especially in yedoma regions (Vonk et al 2015, Walter Anthony et al 2016).

To check that the aquatic C ages exported from the site, based on the age distributions, are within the right order of magnitude, we apply a basic estimate for the expected C export if the study site were in equilibrium (i.e. the C inputs to the soils as NPP equal soil outputs, and total soil C stocks remain at steady-state; see supplementary methods). Under these assumptions, we would expect to observe median aquatic C ages ranging from 21–117 years BSD, with ~21% of this C of pre-industrial origin. This indicates that the ages we see in the aquatic export from this site are within the expected range. In order to take this analysis a step further and to quantify whether the aquatic C export is contributing to the PCF (i.e. whether the observed ages are significantly older than the expected ages, and therefore inconsistent with a terrestrial C cycle at steady-state), we need a more detailed model of soil age distributions, and to be able to partition the ages of CO₂ and CH₄ lost vertically from the soil versus the lateral aquatic DOC and POC components. The analyses presented here serve as a first step in this direction, and future work should consider upscaling point-source soil data to the catchment scale to relate the measured vertical soil distributions directly to aquatic C export.

There is, therefore, no clear evidence that the aquatic systems in the immediate study area are currently contributing to the PCF. Given this 'baseline' state of the study catchment, the findings presented here suggest that pre-industrial C is already present in substantial amounts in the Arctic C cycle, and that old C export will likely be enhanced by permafrost thaw in response to climate warming.

This work was primarily supported by the UK Natural Environment Research Council (NERC), grant numbers NE/K000217/1, NE/K000225/1, NE/K000268/1 and NE/K000284/1. JFD acknowledges partial support by the program of the Netherlands Earth System Science Centre (NESSC), financially supported by the Ministry of Education, Culture and Science (OCW). We would like to thank the Aurora Research Institute in Inuvik, Prof Philip Marsh of Wilfrid Laurier University, Dr Oliver Sonnentag of Université de Montréal, and Dr Mark Cooper of the University of Exeter, for support during the field campaigns, Environment Canada for data access, Dr James Weedon of Vrije Universiteit Amsterdam for support with the statistical analyses, and the staff at the NERC Radiocarbon Facility (NRCF010001) and the SUERC AMS Laboratory for the ¹⁴C sample processing. The authors declare no competing financial interests. Requests for additional materials and data should be sent to J F Dean.