Rivers across the Siberian Arctic unearth the patterns of carbon release from thawing permafrost

Significance High-latitude permafrost and peat deposits contain a large reservoir of dormant carbon that, upon warming, may partly degrade to CO2 and CH4 at site and may partly enter rivers. Given the scale and heterogeneity of the Siberian Arctic, continent-wide patterns of thaw and remobilization have been challenging to constrain. This study combines a decade-long observational record of 14C in organic carbon of four large Siberian rivers with an extensive 14C source fingerprint database into a statistical model to provide a quantitative partitioning of the fraction of fluvially mobilized organic carbon that specifically stems from permafrost and peat deposits, and separately for dissolved and particulate vectors, across the Siberian Arctic, revealing distinct spatial and seasonal system patterns in carbon remobilization.


Supplementary Information Text Sampling and analyses of DOC and POC
Samples for DOC and POC analyses were collected at Salekhard (Ob), Dudinka (Yenisey), Zhigansk (Lena) and Cherskiy (Kolyma) between July 2003 and November 2013 as part of the PARTNERS and ARCTIC-GRO programs (1). Water was collected with US Geological Survey D-96 depth-integrating water samplers from 2003 through 2011, and with Van Dorn bottles deployed at discrete water depths starting in 2012 during ice-free conditions. In both cases, multiple casts across the river channel were combined in a Teflon churn to create composite samples for analysis. During ice-covered periods, surface water samples were collected through holes cut into the ice. Water for 13 C and 14 C analyses of DOC was filtered through Whatman QM-A filters (2.2 µm pore size) and stored frozen in acid-leached polycarbonate bottles until analysis. DOC samples were UV oxidized and cryogenically purified before 13 C and 14 C analysis at the US-NSF National Ocean Sciences Accelerator Mass Spectrometry (NOSAMS) facility of the Woods Hole Oceanographic Institution (Woods Hole, Massachusetts, USA). For 13 C analyses of POC, water was filtered through pre-combusted Whatman GF/F filters (0.7 µm pore size). Filters were frozen and transported to the Marine Biological Laboratory in Woods Hole, where they were dried at 60°C, triple acidified with sulfurous acid (H2SO3) to remove inorganic carbon, redried, and packed into tin capsules. The δ 13 C values of POC were analyzed with Elemental Analysis -Isotope Ratio Mass Spectrometry at the Marine Biological Laboratory between 2003 and 2011 and at the University of Texas Marine Science Institute thereafter. For 14 C analysis of POC, water was filtered through pre-combusted Whatman QM-A quartz filters (2.2 µm pore size) that were then acidified with H2SO3 and analyzed for 14 C at the NOSAMS facility. Details on DOC and POC sampling and analysis methods can be found in previously published papers (2)(3)(4), as well as in the metadata provided with the publicly available datasets (www.arcticgreatrivers.org).
Main sources of organic carbon to Siberian rivers Overview. We distinguished four potential organic carbon sources for source apportionment, (i) recent terrestrial primary production, (ii) active layers (including non-permafrost surface soils), (iii) Holocene permafrost, peat and thermokarst deposits, and (iv) Pleistocene permafrost deposits such as Ice Complex Deposits. Note that aquatic primary production in high-latitude rivers represents mostly recycling of terrestrially-derived, mineralized carbon and is not considered an independent carbon source (see main text for discussion). The isotopic composition of potential carbon sources was constrained based on meta-analyses of previous publications, focusing on Siberia, but extending the spatial coverage where data from Siberia were scarce. Conventional 14 C ages (before 1950) reported in the literature were converted into Δ 14 C values. The Δ 14 C values of organic carbon sources were used for quantitative statistical source apportionment, and the δ 13 C values for qualitative comparison to assess the degree of processing of organic carbon in Siberian rivers. Individual carbon sources, and their Δ 14 C and δ 13 C values are briefly summarized in Table  S1, and in more detail below.
Terrestrial primary production. We estimated the Δ 14 C and δ 13 C values of carbon recently fixed by terrestrial plants from measurements of litter and surface organic layers in arctic, subarctic, and boreal systems. Since data from Siberia were scarce (n = 1 for Δ 14 C, n = 10 for δ 13 C; see Table S2), we included also observations from the European part of northern Russia, northern Scandinavia, northern Canada, and Alaska. We followed the description of the original authors for identifying litter and surface organic layers; data for mineral surface soils from Siberia are included in the active layer category. Data and references are presented in Table S2. Note that for most samples, only δ 13 C or Δ 14 C values are reported. The δ 13 C values of organic and litter layers were thus constrained as -27.7 ± 1.3‰ (n = 94) and the Δ 14 C values as 97.0 ± 124.8‰ (n = Pleistocene deposits. We constrained Δ 14 C and δ 13 C values of Pleistocene deposits using observations from Pleistocene Ice Complex Deposits (also known as "Yedoma") that are most vulnerable to degradation due to their high ice content. The δ 13 C values were estimated based on a previous review (19) as -26.3 ± 0.7‰ (n = 374), and Δ 14 C values by updating the database used by Vonk et al. (20) and Tesi et al. (21). This new database includes a set of more recent publications, but excludes samples where contamination by Holocene material was indicated (see also ref. (21)). As for Holocene deposits, we restricted our database to observations from river bank and coastal exposures in order to capture the Δ 14 C range of material realistically remobilized to aquatic systems. Only data from Siberia were considered. The final database is presented in Table S5, and Δ 14 C values were estimated as -954.8 ± 65.8‰ (n = 329).

Source apportionment
Fractions of organic carbon from recent primary production as opposed to permafrost and peat deposits (PP-C) in river samples were calculated using Equations (1) and (2), based on the Δ 14 C values of samples (Δ 14 Csample), recent primary production (Δ 14 Crecent) and PP-C (Δ 14 CPP). frecent and fPP are the fractions of recent carbon and PP-C, respectively.
The Δ 14 C values of the PP-C endmember were calculated for three scenarios, assuming different contributions of organic carbon from active layer, Holocene deposits and Pleistocene deposits.
The Best Estimate scenario represents, in our opinion, the most realistic estimate as it assumes that all PP-C compartments contribute to fluvial PP-C. A least biased approach was used where all fractional combinations of individual compartments are set to be equally likely. Formally, we thus integrated over all possible fractional combinations of the individual compartment distributions to compute the Best Estimate distribution. Compared to the assumption of equal contribution of all compartments, this approach results in a wider spread of the combined PP-C probability density function compared to the assumption of a fixed mixing ratio, and a larger uncertainty that considers not only the uncertainties of the Δ 14 C values of individual compartments but also the uncertainty of their relative proportions, and is therefore more conservative. To account for the uncertainties of the Δ 14 C endmember distribution, a Bayesian approach was implemented. A uniform ("objective") prior distribution was used to represent our initial knowledge of the fractional source contributions. The prior then was combined with the knowledge from the measurements assuming isotopic mass-balance ("the likelihood") to compute the posterior distribution of the fractional contributions, from which statistical parameters such as mean, median and standard deviations can be derived. Here, the method presented in Andersson et al. (22) was adjusted to accommodate flux weighting, with fluxes calculated from discharge and POC or DOC concentration at the respective time point. The weight from data point i (wi) was computed where J is the flux and N the number of observations. The posterior distribution of the fractional source contribution can then be expressed as: where p(f) is the prior and ∏ (∆ | ) • the likelihood. A Markov chain Monte Carlo (MCMC) algorithm (22) implemented in Matlab (ver. 2014b) was used to compute the posterior. To ensure good convergence and a low computation uncertainty, the calculations were run using 1 000 000 iterations, a burn-in of 10 000 and a data thinning of 10.
Response of fluvial Δ 14 C values to changes in PP-C release An additional simulation was performed to test the sensitivity of fluvial organic carbon Δ 14 C values to changes in PP-C release. To that end, the flux of PP-C in rivers was changed by a factor x ranging from 0.5 to 2.0 in 0.25 increments, while recent carbon flux was kept constant. The thus adjusted PP-C fraction of total fluvial carbon (fPP-change) and the resulting shift in Δ 14 C values of fluvial organic carbon were calculated using Equations (5) and (6).
The minimum resolvable change in PP-C flux was calculated for each river by changing measured Δ 14 C values in 1‰ increments and comparing the resulting dataset with baseline values using flux-weighted t-tests. The minimum change in Δ 14 C that resulted in a statistically significant difference (p < 0.05) was then inserted into Equations (5) and (6) to derive the corresponding factor x. Calculations were performed in R 3.5.1 (23) with the packages 'Hmisc' (24) and 'weights' (25).     Table S1. Overview of Δ 14 C and δ 13 C values of potential organic carbon sources.            Table S7. Statistical analysis of DOC and POC δ 13 C and Δ 14 C values. Table S7. Statistical analysis of DOC and POC δ 13 C and Δ 14 C values. Flux-weighted differences between rivers and seasons were analyzed with two-way ANOVA followed by Tukey's HSD as post hoc test for individual rivers or seasons in R 3.5.1 (23) with the packages 'Hmisc' (24) and 'HH' (82). Differences were considered significant at p < 0.05 (n.s., not significant), and significant differences between categories are indicated by different letters for the post hoc tests.