Black Creek watershed (BCW) is situated on the Yukon Coastal Plain in the western Canadian Arctic and drains into Ptarmigan Bay, which is a semi-open lagoon sheltered from the open Beaufort Sea (Fig. 1). Black Creek is a small coastal stream draining a polygonal tundra landscape underlain by continuous permafrost. The contributing watershed area is approximately 4 km2, estimated using ArcticDEM (digital elevation model), a publicly available 10 m resolution digital elevation model (Porter et al., 2018; accessed on 28 May 2020), from which we obtained a watershed delineation using GRASS GIS (Geographic Resources Analysis Support System graphical information system). The Yukon Coastal Plain stretches ∼ 300 km from the Mackenzie delta in the east to the Alaskan border in the west. The Quaternary surficial geology is mainly characterized by lacustrine, morainal, fluvial and colluvial deposits (Rampton, 1982). IWP tundra, moraine hills, wetlands, beaded streams and thermokarst lakes are the predominant landscape types. The land cover can be classified as low-shrub tundra, subzone E (Walker et al., 2018) with the occurrence of Betula nana, Salix polaris, mosses and lichens in HCPs, while graminoids dominate in LCP terrain. The mean summer temperature (June, July and August 1991–2020) is 7.7 ∘C (± 4.6 ∘C) based on available data for three nearby stations (Herschel Island – Qikiqtaruk, Komakuk Beach and Shingle Point). The mean annual temperatures at Shingle Point and Komakuk Beach are -9.9 and -11 ∘C, respectively, and precipitation means are 254 and 161 mm, respectively (Environment Canada, https://climate.weather.gc.ca/climate_normals, last access: 26 August 2021).

The region of interest is underlain by continuous permafrost and active-layer depths averaging around 30–40 cm in IWP terrain on nearby Herschel Island (Siewert et al., 2021). Ground ice volumetric content in the Yukon Coastal Plain averages around 46 % but reaches as high as 74 % in some areas (Couture et al., 2018; Couture and Pollard, 2017). The warm season in the western Canadian Arctic, during which the stream network is active, lasts approximately 4 months (Dunton et al., 2006). On average the sea ice break up in the southern Beaufort Sea region starts around mid-May, and freeze-up starts in early October with prolonged open-water periods around the Mackenzie delta area. Both winter and summer sea ice extent and concentration rapidly declined in recent decades (Galley et al., 2016). The lengthening of the sea-ice-free seasons leads to increased storm frequency and intensity. In combination with higher surface temperatures in the region (Screen et al., 2012), these environmental changes are expected to have a drastic impact on biogeochemical cycling and hydrological processes in the western Canadian Arctic (Parmentier et al., 2017).

During the sampling period (8–19 August 2018), we collected on-site weather data at a 5 min interval. Air temperature was measured at 1.5 m above the ground (BTF11/002 TSic 506; ± 0.1 ∘C accuracy). Precipitation was measured (Young Model 52203; 0.1 mm per tip; accuracy ± 2 %) at 0.5 ma.g. away from any objects causing potential wind shadow. Wind speed was measured using Thies Clima (4.3519.00.173; ± 0.5 ms-1 accuracy). Available weather data from outside the sampling period were downloaded from the Canadian government's website on the environment and natural resources (https://climate.weather.gc.ca, last access: 16 November 2021) for the three nearby stations mentioned in Sect. 2.1 (station IDs 2100636, 2100682 and 2100950, respectively).

Soil and water samples were collected between 9–19 August 2018. Soil samples were taken at 46 sites within the main polygon types in the watershed (HCP, LCP and flat polygon), which were classified based on field observations. Both the active layer (AL) and upper permafrost (PF) were sampled. Active-layer samples of a known field volume were collected from the main soil horizon types (O, A, B and Bf/Cf) and classified according to Schoeneberger et al. (2012). Samples with visible gley or cryoturbation were marked additionally. Permafrost samples were collected at 10 cm intervals below the permafrost table up to a depth of 100 cm from the surface (subject to practicality) using either a steel pipe and sledgehammer, SIPRE corer, or Hilti hammer drill. All soil samples (AL and PF; n=153) were stored frozen at -18 ∘C in ziplock bags until further processing in the lab, where porewater extraction took place. Stream water samples were taken every 6 h at the catchment outlet using an ISCO 3700 automatic water sampler (Teledyne). The sampler was placed ∼ 50 m upstream (± 1 elevation meter higher than the actual outlet), where water was shooting/free flowing for a significant part of the section downstream of the sampler. This location was chosen to ensure that no interference from the neighboring lagoon would occur under normal conditions. In addition, manual samples were taken along the main channel and at three tributary streams flowing into the main stem using pre-rinsed 500 mL Nalgene bottles, which were flushed with stream water three times prior to sampling. All water samples were filtered through pre-combusted and pre-weighed glass fiber filters (GF/F, Whatman, 47 mm diameter, 0.7 µm nominal pore size). Subsamples for DOC and δ13C DOC (dissolved organic carbon) analysis were acidified to pH < 2.0 using 36 % HCl (Suprapur) and stored in a cool (4 ∘C) and dark location. Subsamples for chromophoric and fluorescent dissolved organic matter (DOM; CDOM and fDOM) were stored frozen and in a dark location at -18 ∘C. All filters with suspended material designated for total suspended solids (TSSs), POC (particulate organic carbon) and δ13C POC analysis were stored frozen and in a dark location. Basic hydrochemical readings were taken at the stream outlet with an AP-5000 multiparameter probe (Aquaread Ltd.), which was deployed at the catchment outlet from 8 until 19 August. Measurements included relative water level (h, m), water temperature (T, ∘C), acidity (pH), electrical conductivity (EC, µSm-1), turbidity [NTU] (nephelometric turbidity unit), dissolved oxygen (DO, percentage of saturation), redox potential [mV] and CDOM abundance [µgL-1]. The CDOM sensor was calibrated using a quinine sulfate equivalent solution (CDOM-CAL-600, Aquaread Ltd.), yet units are given in micrograms per liter as provided by the instrument. An empirical stage discharge equation was derived using flow measurements at different stages and fitting a quadratic function within the measured range to estimate discharge (Q, Ls-1) as a function of the pressure head on the sensor (h, m) of the stream (Eq. 1). 1Q=-21927h2+7271.5h-524.89

The measurement was based on the creek's flowing cross-sectional area together with flow velocity (measured at two-thirds of the water depth and 25 cm increments using a M1 mini current meter and Z6 counting device; SEBA Hydrometrie GmbH & Co. KG) at the outflow at varying water levels.

Frozen soil samples were wet-weighed and slowly thawed at 8 ∘C. Porewater was then extracted from active-layer and permafrost samples (n=142) using Rhizon samplers (mean pore size of 0.6 µm, Rhizosphere, Wageningen, the Netherlands) under cold and dark conditions in a cooler room (4 ∘C). Subsamples for CDOM and fDOM analyses were taken from the extracted porewater and transferred into 15 mL falcon tubes and stored frozen and in a dark location until analysis. Subsamples for DOC and δ13C DOC were acidified (pH < 2) and stored cool (at 4 ∘C) and in a dark location in 40 mL pre-combusted glass vials until analysis.

All DOC samples have been filtered through glass fiber filters with 0.7 µm nominal pore size. The DOC concentration and δ13C DOC were measured with an Aurora 1030 DOC analyzer (OI Analytical, USA) connected via a Conflow V to an isotope ratio mass spectrometer (IRMS, Delta V, Thermo Scientific, Germany) at the University of Hamburg (Germany) (stream water incubation samples), the Alfred Wegener Institute Helmholtz Centre for Polar Research (Bremerhaven, Germany) (bulk porewater samples) following Hölemann et al. (2021) and at North Carolina State University (Raleigh, North Carolina, USA) (stream water bulk samples and porewater incubation samples). DOC concentrations were used to quantify the total amount of OC in a dissolved state within the watershed systems, and δ13C DOC was used to derive the origin and relative degradation state of OM. DOC concentrations from porewater were used to calculate yields in milligrams of DOC per gram of soil dry weight and milligrams of DOC per gram of soil organic carbon (SOC) using Eqs. (2) and (3). 2Ysoil=[DOC]×mwmsρw,3YC=[DOC]×mwmCρw, where mw is the mass in grams of water present in the sample, measured as the difference in the bulk wet soil sample weight minus the dry weight and assuming the density of water at 4 ∘C as ρw=1×103 gL-1; ms is the soil bulk dry weight in grams; mC is the total mass of SOC in grams; and [DOC] is the DOC concentration in milligrams per liter. SOC was measured as the fraction of the dry bulk weight lost on combustion at 550 ∘C multiplied by a conservative factor of 0.45 to convert from bulk soil organic matter (SOM) to SOC, in accordance with Jensen et al. (2018).

Carbonates were removed from freeze-dried filters using acid treatment. For this, filters were subsampled (16 filter punches of a 4 mm cross-section) into silver capsules, moisturized with 25 µL of distilled water, acidified with 25 µL of 1 M HCl and left for 30 min at room temperature. Then, 50 µL of HCl was added, and samples were dried in an oven for 3 h at 60 ∘C. After that, silver capsules were folded and analyzed for percentage of OC, percentage of N, and δ13C (‰ VPDB; Vienna Pee Dee Belemnite) at the Vrije Universiteit Stable Isotope Laboratory (Amsterdam, the Netherlands).

Deuterium and oxygen isotopes (δD and δ18O) were measured on water subsamples with a continuous-flow Delta Plus X IRMS coupled to a FlashSmart Elemental Analyzer and high-temperature conversion elemental analyzer (TC/EA) at Vrije Universiteit Amsterdam and are given as the per mil difference from Vienna Standard Mean Ocean Water (VSMOW). Deuterium excess was used to allocate the precipitation source (e.g., Fritz et al., 2016) in the same region (d excess = δD-8δ18O; Fritz et al., 2016; Dansgaard, 1964). Data were compared with water isotopic values from two local meteoric water lines (LWMLs) in Inuvik (200 km southeast) (δD = 7.3 ⋅ δ18O - 3.5) (Fritz et al., 2016) and Utqiaġvik (formerly Barrow), Alaska (600 km northwest) (δD = 7.5 ⋅ δ18O - 1.1) (Throckmorton et al., 2016).

The chromophoric fraction of DOM was used to characterize DOM and identify the source. The CDOM and fDOM (fluorescent DOM fraction) were used as indicators of DOM quality such as degradation status, molecular size and DOM source (Stedmon and Nelson, 2015). We used a range of absorbance and fluorescence indices for the characterization of DOM, as summarized in Table 1. Fluorescence data were processed using the drEEM toolbox (decomposition routines for Excitation Emission Matrices; Murphy et al., 2013) in MATLAB (R2017b). CDOM and fDOM were measured on a HORIBA Aqualog fluorescence spectrophotometer at the Technical University of Denmark (DTU, Copenhagen).

In this study we chose four indicators to the estimate degradation state and to infer lability:

Slope ratio (Sr) was calculated as the ratio between the slope of the absorbance between 275–290 and 350–400 nm and the absorbance ratio (a254:a365). These have been recognized as indicators of DOM molecular weight (MW) (Helms et al., 2008). The assumption is that lower-MW organic molecules will generally be more bioavailable.

Additionally, we performed incubations with stream and porewater samples to estimate the degradation potential of DOC by comparing DOC concentrations and δ13C DOC values before and after incubations.

Along with isotope tracers such as δ13C DOC, Δ14C DOC, δD and δ18O (Vonk et al., 2012; Grotheer et al., 2020), absorbance and fluorescence properties have been successfully used for source apportionment approaches and to characterize DOM and trace sources (Lee et al., 2020). We used endmember mixing model analysis (EMMA) to estimate the contribution of three potential sources (permafrost, active layer and in-stream primary production) to the Black Creek stream using δ13C DOC, Sr and a254/a365 as tracers. These parameters are considered semi-conservative and most distinctive in separating endmembers. We used a Bayesian mass-balance source apportionment model with Metropolis–Hastings Markov chain Monte Carlo sampling following and adapting code found in the supplementary information of Bosch et al. (2015) in MATLAB R2017b. To compute source contributions, we ran the model with a time series of measured δ13C DOC, Sr and a254/a365 at the catchment outlet over time. Although sample size is limited due to practical limitations, we chose to use δ13C DOC of porewater instead of δ13C SOC of soils to avoid fractionation effects from soil-to-water leaching (e.g., Kaiser et al., 2001; Boström et al., 2007) impacting the source apportionment. For the active layer, we used a δ13C DOC value of -26.4 ± 1.07 ‰ based on the porewater δ13C DOC measurements in the catchment (n=3). Active-layer values for Sr and a254/a365 were also based on porewater samples from the catchment and were 0.71 ± 0.08 (n=45) and 4.55 ± 0.8 (n=45), respectively. For permafrost we used a δ13C DOC value of -24.15 ± 1.03 ‰ (n=6) and Sr and a254/a365 values of 0.85 ± 0.08 (n=44) and 5.81 ± 1.2 (n=45), respectively, based on δ13C DOC and Sr and a254/a365 of permafrost porewater. Finally, the primary-production source was set to -28.48 ± 1.0 ‰ (n=9) based on δ13C DOC values of the tributaries where we observed primary production (algal mats). The standard deviation of the analyzed endmember samples was 0.237 ‰. However, we acknowledge that the tributary water δ13C DOC signal probably consists of a mixture of terrestrial and primary-production source leachates. Moreover, it is likely that fractionation takes place during leaching from OM sources, including primary-production sources. Hence, we expect a pure primary-production signal would be more depleted than what is found in the tributary water DOC, and we have therefore increased the standard deviation to ± 1 ‰, similar to the other δ13C DOC sources, to account for uncertainty. Our decision-making process herein is consistent with what others have found as representable δ13C primary-production endmembers (e.g., Winterfeld et al., 2015). The Sr and a254/a365 of the primary-production endmember were set at 0.78 ± 0.020 (n=9) and 5.40 ± 0.23 (n=9), respectively. To estimate mixing with terrestrial DOC, we also ran the simulation with δ13C POC (32.68 ± 2.00 ‰, n=9) instead of δ13C DOC by means of sensitivity analysis. We acknowledge that using the δ13C POC to trace DOM is not incorporating potential fractionation effects through leaching and that the most realistic primary-production δ13C value probably lies between the δ13C DOC and δ13C POC values used here.

We performed a sensitivity analysis by increasing and decreasing (± 5 %) permafrost endmember tracer means and standard deviations separately and comparing the effect (percentage change) on relative contribution of each source in the mixing model. The sensitivity analysis focused on permafrost endmember values, since we are primarily interested in the relative contribution of thawing permafrost. Additionally, it is assumed that changing other endmember tracer means and standard deviations by the same order of magnitude will result in same changes in order of magnitude of relative source contribution and testing, with only permafrost being representative of all endmembers.

All statistical analyses were performed in the Python 3 programming environment (Van Rossum and Drake, 2009) using the pandas (McKinney, 2010), SciPy (Virtanen et al., 2020) and statsmodels (Seabold and Perktold, 2010) packages. Significance statistics were calculated with a two-sided T test. Linear regression results were obtained using linear least-squares regression for two sets of measurements as described in the SciPy manual.

Weather conditions during the field campaign were variable with air temperatures ranging between -0.8 and 12.7 ∘C (mean = 4.2 ± 2.6 ∘C) (Fig. S1 in the Supplement). The predominant wind direction was northwest with mean wind speeds of 4.78 ± 2.92 ms-1 and gusts up to 15.1 ms-1 (7 Bft; Beaufort scale). Precipitation was generally low with notable rainfall recorded on 13 August (1.3 mm) and between 16 and 19 August (7.7 mm cumulative). Total precipitation during the monitoring period was 9.8 mm (Table S1 in the Supplement). Mean electrical conductivity (EC) at the catchment outlet was 954 µScm-1; pH ranged between 6.5 and 7.8 (mean = 6.9 ± 0.19); and water temperature ranged between 2.9 and 12.3 ∘C (mean = 6.6 ± 1.7 ∘C). During a storm event on 16 and 17 August, water levels at the outlet monitoring station peaked together with EC (19134 µScm-1). Apart from the elevated EC values during the storm event, the EC remained fairly constant during the monitoring period (100 ± 14.0 µScm-1). Discharge in the creek's main channel showed a decreasing trend (min = 9, max = 140, mean = 43 ± 27 Ls-1) during the measurement period. Discharge during the storm event of 16 and 17 August was disregarded due to uncertainty regarding tidal influence, which is indicated by the elevated EC values. Water isotope values within the main channel ranged between -130.3 ‰ and -127.1 ‰ (δD) and -16.35 ‰ and -16.25 ‰ (δ18O) with mean values of -128.0 ± 1.2 ‰ and 16.32 ± 0.03 ‰, respectively. In tributary streams stable water isotope values covered a wider range, between -132.4 ‰ and -94.38 ‰ (δD) and -17.11 ‰ and -11.94 ‰ (δ18O) with mean values of -122.9 ± 7.5 ‰ and 15.71 ± 0.98 ‰, respectively. Stable water isotope signals were grouped by their main source (i.e., permafrost and active-layer porewater, tributaries and the main channel) (Table 2). The correlation of δD and δ18O for the samples compared to the local meteoric water line (LMWL) in Inuvik shows that permafrost samples group in roughly three lines that are distinguished by their d excess (Fig. S2). The majority of the permafrost samples plot further from the LMWL at Inuvik than the modern samples (streams and active layer), which in turn also deviate from the Inuvik LMWL.

Incubation experiments show a high variability in DOC loss as well as δ13C shifts for the active layer and permafrost (Fig. 5). The degradability of DOC strongly differed between samples. The mean loss of DOC [%] from incubations with water from the three different tributary locations (A, B and C) was 4.30 ± 5.3 % after 7 d of incubation. During this period, mean isotopic values of δ13C DOC did not change significantly (p > 0.05) (-28.48 ± 0.2 ‰ at T = 0 d and -28.52 ± 0.3 ‰ at T = 7 d). However, looking at individual samples, δ13C DOC enrichment was observed in several cases (Table S3 and Fig. S5). Locations A and B showed comparable patterns with depletion of δ13C in the first 7 d and repletion between days 7 and 14; location C showed almost a linear trend toward a less depleted δ13C DOC signal over time. The incubation of porewater showed that DOC losses are in the same range as the tributaries and that there are no significant differences between HCP permafrost (5.0 ± 6 % DOC loss, n=3, T = 7 d) and LCP permafrost (7.0 ± 1 % DOC loss, n=2, T = 7 d) (p > 0.05). In contrast to that, DOC losses in the active layer (Oi horizon) were significantly higher (s ± 16 %, n=3, T = 7 d). The δ13C DOC signals are significantly different between the active layer, LCP permafrost and HCP permafrost at the beginning of the incubation (T = 0) and, on average, become slightly more enriched over time. Yet, there is no significant change in δ13C DOC between T = 0 and T = 21 (Table 4, Fig. S6).

Values of SUVA254 at the catchment outlet varied between ∼ 1.71 and ∼ 3.88 Lmg-1Cm-1 and averaged around ∼ 3.07 ± 0.5 Lmg-1Cm-1 (n=25). With the exclusion of two storm event extremes (SUVA254 = 11.02; 11.72 Lmg-1Cm-1), mean SUVA254 values were not significantly (p > 0.05) different from mean SUVA254 values in tributaries (∼ 3.86 ± 1.8 Lmg-1Cm-1). Permafrost porewater showed significantly lower SUVA254 (∼ 1.00 ± 0.6 Lmg-1Cm-1) (p < 0.05) than active-layer porewater (∼ 2.13 ± 0.9 Lmg-1Cm-1). The highest mean active-layer porewater SUVA254 is found in HCP polygon types (∼ 2.4 ± 0.9 Lmg-1Cm-1) followed by flat (∼ 2.14 ± 0.9 Lmg-1Cm-1) and LCP polygon types (∼ 1.55 ± 1.1 Lmg-1Cm-1). For permafrost the order was different, with the HCP polygon type showing the lowest values for SUVA254 (∼ 0.91 ± 0.7 Lmg-1Cm-1) followed by the LCP polygon type (∼ 1.06 ± 0.6 Lmg-1Cm-1) and the flat polygon type the showing highest SUVA254 (∼ 1.23 ± 0.6 Lmg-1Cm-1) in permafrost. Slope ratio (Sr), which negatively correlates with the MW of DOM, was negatively correlated with SUVA254 in our porewater samples (i.e., lower-MW molecules were less aromatic). Contrastingly Sr shows a weak positive correlation with SUVA254 (i.e., lower-MW molecules were more aromatic) in stream water samples. For the fluorescence index and biological index (FI and BIX), which are used to indicate terrestrial sources of DOM (i.e., soil organic matter and plant litter; low FI and BIX), and relatively fresh, more microbially derived DOM (i.e., leachates and products of algae and bacteria; high FI and BIX) (Fouché et al., 2017) the same holds: in porewater samples the aromaticity was highest in samples with a terrestrial and less fresh DOM signal, while in streams the aromaticity was highest in samples with a relatively fresh DOM signal.

Overall, our results show a predominantly terrestrial signature in streams which is slightly but significantly (p < 0.05) higher in tributaries (FI: ∼ 1.47, BIX: ∼ 0.50, Sr: ∼ 0.80) compared to the outlet (FI: ∼ 1.49, BIX: ∼ 0.55, Sr: ∼ 0.85). This signal is comparable to that of porewater in the active layer (FI: ∼ 1.46, BIX: ∼ 0.48, Sr: ∼ 0.81). In contrast, the permafrost porewater averages around a more fresh, microbial and lower-MW signature (FI: ∼ 1.65, BIX: ∼ 0.62, Sr: ∼ 0.93). Permafrost and the active layer are significantly different with respect to their SUVA254, FI, BIX and Sr values (p < 0.05). However, when looking more closely at the distribution of their values within the soil profile, we observe linear trends with depth respective to the permafrost table rather than clustering linked explicitly to each thermal layer (Fig. S7). With respect to polygon types, results show that HCP active-layer spectral indices are significantly different (p < 0.05) from the other polygon classes except for HIX and that differences between active layer and permafrost are most pronounced in HCPs (Table S6, Fig. 6). The humification index (HIX) peaks in the HCP active layer (∼ 0.86), while the lowest HIX value was found in HCP permafrost (∼ 0.75) (Table 5). A more detailed peak in HIX around the permafrost table suggests the prevalence of more degraded SOM (Fig. S7). With a deeper permafrost sampling depth, a decreasing HIX value is observed. Although distinctly different, values of HIX are significantly different neither between the active layer and permafrost nor between landscape classes. Values of HIX were highest in streams (∼ 0.95) and the active layer (∼ 0.85).

With our initial source apportionment setup, we modeled source contribution with mean isotopic (Fig. S8) and spectral-index values at the outlet (δ13C DOC: -25.40 ‰, Sr: 0.86, a254/a365: 5.55) for three endmembers (permafrost OM, active-layer OM and primary-production OM). Model results given these inputs indicate a relative contribution of these sources of ∼ 48 ± 20 %, ∼ 30 ± 21 % and ∼ 22 ± 15 %, respectively (Table 6, Figs. S9 and S10). We used the maximum (-26.08 ‰) and minimum (-24.73 ‰) δ13C DOC values and corresponding Sr and a254/a365 at the outlet to assess variation in source contribution over time. Correspondingly we find that permafrost, active-layer and primary-production OM contributions vary with between 31 %–67 %, 20 %–38 % and 14 %–31 %, respectively. When calculating the source apportionment considering HCP permafrost and LCP permafrost as different DOM sources (HCP permafrost δ13C DOC: -23.68 ± 1.2 ‰, Sr: 0.94 ± 0.1, a254/a365: 6.08 ± 1.1; LCP permafrost δ13C DOC: -25.05 ± 0.5 ‰, Sr: 0.92 ± 0.01, a254/a365: 5.67 ± 1.4), the mean relative contribution of permafrost sources increases from 48 % to 58 % at the catchment outlet. When using δ13C POC of the primary-production sources instead of δ13C DOC, this leads to a decrease in primary-production contribution from 21 % to 11 %, mostly resulting in an increase in permafrost contribution from 48 % to 55 % and an increase in active-layer contribution from 31 % to 34 % (Tables 6 and 7). A summary of the time series and computed source contributions using δ13C DOC of primary production can be found in Table S9.

From sensitivity analysis (i.e., changing input parameters with fixed relative amounts) we observed that modeled source contributions respond strongest to shifts in δ13C DOC. When decreasing the permafrost mean δ13C DOC from -24.15 ‰ to -25.36 ‰, this resulted in a shift from 48.7 % to 51.6 % of relative contribution, while active-layer contribution changes only 0.8 % from 30.8 % to 31.6 %, and primary production decreases from 20.6 % to 16.9 %. Inversely, when using a higher permafrost mean δ13C DOC (-24.15 ‰ to -22.94 ‰), this results in a decrease in its contribution from 49 % to 40 %, while primary production increased from 21 % to 27 % (Tables S7 and S8), and the active layer increased from 31 % to 33 %. Changing standard deviations of permafrost endmember values by ± 5 % leads to changes in contribution ranging from -3 % for permafrost to + 5 % for primary production.

Our data show that there are significant differences in terms of OM pools between the two thermal layers (i.e., active layer and permafrost) and between the main landscape features that define the terrain (i.e., HCP and LCP). The main differences between HCPs and LCPs are the microtopography and hydrological pathways, which may influence OC characteristics. The wetter soils in LCPs have higher thermal conductance; hence summer active-layer depths in the center of the polygon often reach deeper than in HCPs (Liljedahl et al., 2016; Walvoord and Kurylyk, 2016; Wales et al., 2020). LCPs' elevated ice wedge rims generally promote waterlogged conditions in the polygon center, while in HCPs, the degraded ice wedges form connected drainage channels result in well-drained polygon centers (Liljedahl et al., 2012).

Differences in drainage patterns are reflected in our observed DOC concentrations and yields. Mean SOC contents are higher in the active layer (LCP: 26 ± 13.2 %, HCP: 23 ± 12.8 %) compared to permafrost (LCP: 17 ± 10.7 %, HCP: 18 ± 6.2 %). This contrasts with the DOC concentrations, which are higher in permafrost (HCP: 171.20 ± 87.8 mgL-1, LCP: 150.62 ± 63.1 mgL-1) compared to active layer (HCP: 95.20 ± 53.7 mgL-1, LCP: 92.98 ± 53.2 mgL-1). The high mean DOC concentrations in permafrost porewater (162.44 ± 82.0 mgL-1) compared to those of the active layer (97.14 ± 52.54 mgL-1) indicate that the permafrost DOC pool is still freeze-locked (i.e., immobile), whereas the active layer has been flushed more regularly. Differences in DOC concentration between the active layer and permafrost were highest in the HCP and lower in the LCP polygon type category. This could indicate that in the active layer of HCPs, DOM was already subject to more degradation and flushing with runoff. This is also suggested in a simulation study by Liljedahl et al. (2012) showing that in LCP terrain 46 % of the snow water equivalent was flushed as runoff, while this was 73 % in HCP terrain.

Overall our data show a trend of increasing DOC yield (mgDOCg-1SOC) with increasing depth in the soil. The depth below the permafrost table proved to be a good indicator of DOC concentration as well ([DOC]=2.028⋅Δzp+134.60, R2 = 0.71, p=8.6×10-37). The SOC content showed a negative correlation with depth (Fig. 7), dipping around the permafrost table. This “dipping” effect could be ascribed to the increasing likelihood of waterlogging conditions occurring near the permafrost table of HCP soils (Harden et al., 2012). Signs of stagnant water were observed in several profiles of gleyed soil (typical brown-orange to grey-blueish color patterning caused by waterlogged anoxic conditions). We also noted that in samples of gleyed soil, comprising mostly mineral soils (B horizons), there often was a low SOC content relative to the DOC concentration (i.e., a high DOC yield; in mgDOCg-1soilOC). This suggests that, due to their waterlogged conditions, either these soils are efficiently leached out and/or DOC is flushed in from overlying O horizons and accumulates. The δ13C signature of DOC shows a more degraded (i.e., more enriched) signal than the bulk δ13C SOC, indicating that the degraded fraction of SOM is preferably leached and/or sorption processes are affecting δ13C DOC signatures. Moreover, we find that values of permafrost δ13C DOC are more enriched than those of the active layer, indicating a more processed OM pool or a stronger DOC leaching effect on 13C in mineral permafrost horizons than organic surface horizons.

The OC export from our investigated watershed is dominated by DOC (meanDOC:meanPOC = 16.03mgL-1:0.41mgL-1) with limited variability in concentration except from two storm events that diluted the DOC load. This may be explained by the flat topography in this area which minimizes impacts of bank erosion and thermo-erosion that enhance sediment mobilization (Costard et al., 2003), along with long residence times of groundwater within the soil facilitating DOC leaching (Connolly et al., 2018). Similar dominance of DOC has been shown elsewhere in the pan-Arctic watershed, both in large and small rivers (e.g., Holmes et al., 2012; Fabre et al., 2019; Coch et al., 2020), but this dominance may be even more pronounced for low-relief tundra plains. Based on satellite imagery (WorldView-2, DigitalGlobe Inc., acquired on 18 July 2018) an estimated ∼ 80 % to ∼ 90 % of our watershed consists of HCP terrain. The δ13C DOC and δ13C POC signatures from the stream water at the outlet compared to what was found in porewaters suggest that DOC is predominantly derived from HCP terrestrial sources (Mann et al., 2015), whereas POC likely stems from primary production of phytoplankton growth within the stream network or ice wedge troughs (Tank et al., 2011; Winterfeld et al., 2015) or fragments of sedge biomass transported into the stream (Wooller et al., 2007). According to our source apportionment ∼ 81 %–∼ 90 % is of terrestrial origin with 48 ± 19 % of the DOM/DOC stemming from permafrost. The remainder (22 ± 15 %) is most likely aquatic DOM/DOC produced by primary production (Fig. 8). The tributaries δ13C signal showing signs of primary production indicates that these small ice wedge trough streams are not necessarily important for transporting terrestrial OM. Presumably terrestrial OM is rather transported towards the main channel via supra-permafrost base flow, instead of via IWP troughs. Alternatively, terrestrial OM present in the IWP troughs may quickly be decomposed and/or incorporated into primary production. Increased hydrologic connectivity of IWP troughs as well as increased connectivity via active-layer deepening in permafrost watersheds (Lafrenière and Lamoureux, 2019; Evans et al., 2020) may lead to higher input of terrestrial OM.

At the outlet we measured the variable OC concentration and geochemical signature over time, with three observable patterns: (i) diurnal variation in CDOM abundance and optical properties; (ii) short, storm-induced peaks in POC and dips in DOC; and (iii) seasonal decline in DOM export (Fig. 3a and b, right).

i.

We observed a diurnal pattern in CDOM concentrations (range: ∼ 20 %–25 %) at the stream outlet (Fig. 3a, right). This diurnal pattern can be explained by both temperature- and light-dependent variability in productivity as well as variations in ground ice melt contribution and evapotranspiration-induced flow effects, which are in turn temperature dependent (Spencer et al., 2008; Ruhala and Zarnetske, 2017). We observe that CDOM fluctuates synchronously with water temperature (i.e., peaks in water temperature correspond with peaks in CDOM). Similarly, peaks in Sr (∼ 0.9) and FI (∼ 1.5) correspond with lows in temperature, while for HIX this pattern is inverted (Fig. S9). This dynamic may be explained by the increasing importance of primary production with high temperatures and decreasing importance of deeper baseflow when temperature decreases (i.e., flow becomes shallower, and the signal becomes less deeply terrestrial). The highest values of HIX are found around the permafrost table (presumably where baseflow takes place) (Fig. S7), and a decrease in HIX at the outlet in sync with lowering temperatures supports the idea of freezing up from below.

Several studies have looked at seasonal variability in DOC and DOM composition and concluded that antecedent winter DOC is flushed out during freshet, resulting in a DOC peak with relatively low Sr and FI and high SUVA254, indicative of DOM coming from the organic surface layer. As the summer season progresses, a steady increase in Sr and FI and decrease in SUVA254 and DOC continuing up to the very end of the season were found uniformly in rivers across the Arctic, linked to the increasing contribution of deeper soil horizons via deepening of the active layer (Neff et al., 2006; Spencer et al., 2008, 2009a; Holmes et al., 2008, 2012). In this respect our study shows similar trends in DOC, SUVA254, Sr and FI at the outlet over the course of the monitoring period.

Our source apportionment using δ13C DOC, Sr and a254/a365 as tracers showed that a high proportion of DOC within the stream originates from permafrost/deep-active-layer DOC contributions (∼ 48 %), outnumbering the DOC influx from the active layer (∼ 31 %) and primary production within the stream (∼ 21 %). This is in stark contrast to larger (Siberian) Arctic rivers (Wild et al., 2019), where fluvial DOM fluxes stem predominantly from recent terrestrial primary-production sources. Due to the small catchment size the contribution of permafrost OC is likely more evident in small streams, and the difference between these and larger Arctic rivers may indicate that permafrost DOC is likely degraded before it reaches larger rivers. Moreover, small streams like the one in this study drain a degrading continuous permafrost landscape exclusively, whereas large Arctic rivers drain also non-permafrost or discontinuous permafrost terrain, disproportionally contributing to the riverine OM fluxes (e.g., Frey and McClelland, 2009). The spatial and temporal extent of terrestrial permafrost inputs into stream networks may likely expand upon increasing severity of meteorological extremes. For instance, Schwab et al. (2020) found aged DOC downstream in the Mackenzie River main stem, following a warm summer and the second warmest winter on record.

This study shows that permafrost-DOM-related processes are most visible close to the terrestrial–aquatic interface, i.e., in the headwaters of Arctic rivers. With our results we also show that variability herein is highly seasonal and weather driven, as our measurements at the outlet show diurnal, storm event and seasonal-trend patterns within the duration of our relatively short (10 d) field campaign. This emphasizes the need for high-resolution long-term measurements (e.g., such as those ongoing at Cape Bounty Arctic Watershed Observatory (CBAWO); Lamoureux and Lafrenière, 2018) in order to fully understand the mechanisms at work in the (Arctic) permafrost watershed OM dynamics. Important is the notion that cascading effects and food web interactions resulting from permafrost release into Arctic headwaters may be difficult to detect but may have large impacts on sensitive Arctic ecosystems (e.g., Vonk et al., 2015a).

Small IWP tundra watersheds such as Black Creek, presented here, are abundant in and representative of the lowland regions of the coastal Arctic continuous permafrost zone. Due to their abundance and proximity to the Arctic Ocean, tundra streams have the potential to export large quantities of terrestrial permafrost organic matter into coastal waters. Our samples and measurements at the outlet give insight into the average discharge and OC flux of a typical tundra stream during summer. We acknowledge that uncertainty exists in these data and that extrapolation based on a single watershed will greatly increase these uncertainties. Still, we deem our study area representative of the majority of small catchments along the Canadian Yukon Coastal Plain (roughly from the Alaskan border in the west to Shingle Point in the east). As a first attempt toward upscaling from small stream data, we used mean discharge (67 ± 51.1 Ls-1), mean DOC and POC concentration (16.89 ± 0.7 and 0.36 ± 0.1 mgL-1), and a catchment area of ∼ 4 km2 to make an area-based estimated baseline OC flux for the region. This yields a DOC flux of ∼ 0.03 ± 0.021 gDOCm-2d-1 and POC flux of ∼ 0.0005 ± 0.00042 gPOCm-2d-1. In these flux estimates we excluded the variability caused by storm days. In contrast to DOC, POC fluxes are likely to be highly variable, since they are impacted by summer storm activity which can vary between 2 and 21 stormsyr-1 in the southern Beaufort Sea region (Hudak and Young, 2002). Arctic-type storms are most prevalent in July and August and average 8 stormsperseason. Assuming a duration of 1–2 dperstorm, this would lead to an additional export of ∼ 0.010–0.020 gPOCm-2 annually. The period through which lateral transport of OM occurs is dependent on the thaw season length and the rate at which the seasonal active layer deepens. To estimate seasonal fluxes, we use an average thaw season duration of 87.7 consecutive frost-free days calculated over the period 1950–2013 from “The Climate Atlas of Canada” (version 2, 10 July 2019, https://climateatlas.ca, last access: 18 August 2021). Using this average, we calculate that Black Creek watershed (∼ 4 km2) exports an average of 8.58 ± 6.6 tDOCyr-1 (or 2.14 ± 1.6 tDOCkm-2yr-1) and 0.24 ± 0.2 t POC yr-1 (or 0.061 ± 0.04 tPOCkm-2yr-1). A similar study by Coch et al. (2018) showed an average 17 d flux of 82.7 ± 30.7 kgDOCkm-2 and 3802.5 kgPOCkm-2. Making accurate extrapolations based on a few data points is debatable, yet by studying watersheds that are representative of IWP tundra we attempt to gain a first insight into the magnitude lateral OC flux components from this particular system on a larger scale. Our data compared with the nearby study site presented by Coch et al. (2018) show that OC fluxes from small streams are variable. With longer, warmer seasons and higher storm frequencies which are predicted for the Arctic (Day and Hodges, 2018), mobilization and export of POC and DOC towards the Arctic Ocean may substantially increase.

This study focuses on data retrieved during the latest stage of the thawing season, when active-layer depths are at their seasonal maximum. Our results show that OC quality and quantity varies between different soil horizons and landform (i.e., polygon type). Evolution of landforms via degradation of IWPs from LCPs to HCPs is likely to result in increased drainage and drying of the landscape and increased net runoff as a consequence. Our results show that the balance between lateral and vertical flux is therefore likely to shift toward lateral flux as mobilizable OM will be flushed out of the system more effectively. The shift towards drainage and export rather than within-ecosystem processing may have strong effects on the Arctic lowland tundra biodiversity and food web interactions, since these are to a large extent based on wetland ecosystems (e.g., Vonk et al., 2015a; Liljedahl et al., 2016).

In parallel to annual active-layer depth deepening, older and potentially more labile OM pools will become available for degradation. In this study we find indications that the most labile fractions are utilized within the soil, and we expect that even in a more well-drained HCP system, residence times within the soil will be long enough to allow for the utilization of the majority of labile DOM. The results show that concentrations of DOC in permafrost are much higher than in the active layer. This together with the notion that labile DOM would be converted quickly leads to the expectation that under current climate trends, small Arctic catchments affected by permafrost degradation may export higher loads of recalcitrant DOM. This in turn could impact aquatic food webs. Due to its strong coloration permafrost and deep-active-layer-derived CDOM could significantly impact light-dependent processes in the aquatic network. Moreover, chemical composition of permafrost and deep-active-layer DOM (e.g., high values of ammonium were found) may impact aquatic biogeochemistry. More research is needed to elucidate these ecosystem interactions.

To better understand and implement lateral permafrost OM dynamics in climate models, more quantitative and qualitative data on the distribution and behavior of small, pan-Arctic permafrost catchments are needed. Moreover, there is a need for more detailed mapping of these watersheds (e.g., high-resolution watershed delineation and assessment of watershed characteristics). Both the former and latter could be achieved by more longer-term (weeks or ideally months or years) monitoring on a larger spatial scale, e.g., by installing sensors and conducting repetitive field research in designated representative areas as well as by aggregating databases with field and remote sensing data (e.g., mapping of landscape-scale changes such as IWP degradation that help predict shifts in soil–stream dynamics and remote-sensing-derived SOC stock maps which could help predict soil DOC stocks and serve as a starting point to predict headwater stream DOC on a larger scale). Such an aggregated database would be valuable input for spatial modeling of lateral carbon fluxes. We also acknowledge that our and future studies would benefit from more extensive sampling to determine unambivalent endmember tracer values, especially for primary-production sources. Hence, we encourage future research efforts aiming to perform source apportionment to extensively test and select suitable endmembers and tracer values. Lastly, by focusing on optical properties of DOM it is relatively easy and cost-effective to trace changes in watershed biogeochemistry, as optical measurement techniques are relatively uncomplicated and readily available. Usage of these techniques together with the standardization of protocols and methods are therefore recommended in order to achieve a more harmonized approach toward understanding lateral permafrost OM dynamics.