Recent Warming Fuels Increased Organic Carbon Export From Arctic Permafrost

Climate‐driven thawing of Arctic permafrost renders its vast carbon reserves susceptible to microbial degradation, serving as a potentially potent positive feedback hidden within the climate system. While seemingly intuitive, the relationship between thermally driven permafrost losses and organic carbon (OC) export remains largely unexplored in natural settings. Filling this knowledge gap, we present down‐core bulk and compound‐specific radiocarbon records of permafrost change from a sediment core taken within the Alaskan Colville River delta spanning the last c. 2,700 years. Fingerprinted by significantly older radiocarbon ages of bulk OC and long‐chain fatty acids, these data expose a thermally driven increase in permafrost OC export and/or deepening of mobilizable permafrost layers over the last c. 160 years after the Little Ice Age. Comparison of OC content and radiocarbon data between recent and Roman warming episodes likely implies that the rate of warming, alongside the prevailing boundary conditions, may dictate the ultimate fate of the Arctic's permafrost inventory. Our findings highlight the importance of leveraging geological records as archives of Arctic permafrost mobilization dynamics with temperature change.

Currently, ecosystem and Earth system models incorporating permafrost carbon dynamics are principally informed by a combination of historical temperature records, in vitro and laboratory soil respiration experiments and observations of geomorphological change (Koven et al., 2011;Schädel et al., 2016;Smith et al., 2005). While valid, these observational approaches are restricted to the last few decades, which, in turn, precludes the rigorous evaluation of long-term simulations and, indeed, the identification of potential non-linearity within the climate system. To overcome this shortcoming, and to further our understanding of thermally modulated long-term permafrost carbon dynamics, high-resolution sub-centennial records remain paramount. While, in principle, deltaic sedimentary secessions offer integrated centennial to millennial-scale records of watershed-housed carbon cycling (Bianchi & Allison, 2009), within the Arctic, strong coastal currents and intense ice processing (e.g., ice gauging and strudel scour) often compromise their integrity and, where available (Tesi et al., 2016;Winterfeld et al., 2018), the sedimentary records often lack the resolution necessary to reveal high-amplitude dynamics.
To this end, the Colville River Delta, located on the northern coast of Alaska, provides a somewhat unique sedimentary archive; shielded from destructive cryospheric processes (e.g., ice gauging and strudel scour) by barrier islands, the delta features a well-laminated sedimentary record dominated by riverine OC inputs (Hanna et al., 2014;Schreiner et al., 2013;Zhang et al., 2015). Accordingly, to reconstruct northern Alaskan permafrost erosion over the last c. 2,700 years, we report radiocarbon ages of bulk OC from a sediment core (L1) collected from the Colville River Delta (Figure 1). Augmenting this record, targeting selected samples, we also conducted complementary radiocarbon analysis of long-chain fatty acids (LCFAs), which as major components in vascular plant leaf wax lipids (Bianchi & Canuel, 2011), serve as an age-resolved biomarker capable of tracing terrestrially derived OC through the Colville watershed. Radiocarbon ages of bulk OC from selected surface soil and upper permafrost samples from the Colville River watershed were ZHANG ET AL.  (Brown et al., 1997). Here, the Colville River watershed is outlined in orange in panel (a) and blue in panel (b). Filled black circles depict sampling locations within the (b) Colville River watershed and (c) delta. Maps were created using ArcGIS 10.6 with world ocean basemap. also measured ( Figure 1). While clearly not exhaustive in their geographic reach, these soil analyses provide a preliminary context for the likely range of permafrost ages that are present in the Colville River watershed, a region from which very few 14 C measurements exist.

Study Area and Sampling Strategy
The Colville River originates on the northern slopes of Brooks Range, flowing north through the Arctic Foothills and Coastal Plain, terminating via a delta on the inner shelf of the Beaufort Sea ( Figure 1). The watershed is underlain by continuous permafrost. Dwarf shrub coverage typifies the vegetation in the river's upper course, transitioning to greater moss cover toward its terminus (Walker et al., 2005). Like most Arctic catchments, ice break-up and snowmelt discharge occur in late May to early June, promoting an intense flood pulse with riverine discharge reaching up to 6,000 m 3 s −1 (Walker & Hudson, 2003). Although oil exploration activities have expanded in the watershed since 2001, particularly at and around the delta, the construction of pipeline, gravel pads, and roads that can potentially trigger thermokarst and thermal erosion are confined in very small zones near those infrastructures (Raynolds et al., 2014). The short-term oil development recorded in our core (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010) is unlikely to have significantly influenced sediment dynamics, especially when contextualized in terms of the seasonal bank erosion associated with the spring freshet (Walker & Hudson, 2003). In addition, minimal oil contamination has been found in nearshore sediments in this region (Venkatesan et al., 2013). More generally, human activity within the watershed is nominal representing a rare and relatively pristine region from which to reconstruct permafrost dynamics.
Simpson Lagoon, adjacent to the Colville River Delta, represents an attractive shallow water (<2.5 m) depocenter archiving Colville River sediments (Dunton et al., 2006;Hanna et al., 2014). Within the lagoon, sediment core L1 was collected in August 2010, using a Rossfelder P-3 submersible vibra corer abroad the R/V Annika Marie. After recovery, the core was split lengthwise, sectioned at 2 cm intervals and freeze-dried before analysis. Complementing core L1, selected soil samples were also analyzed in an attempt to, at least partially, represent the age profile for surface and upper permafrost in the wider Colville River watershed. Specifically, soils from Ivotuk and DUM were collected from the upper watershed in 1999 (Michaelson & Ping, 2003) and 2009, which were augmented with additional samples from Itkillik in 2011 (Kanevskiy et al., 2016). Samples from the first two sampling campaigns were air-dried at room temperature, while those from Itkillik were oven-dried at 90°C. While, hypothetically, these treatments could differentially lose volatile organic constituents, researches have estimated the loss to be less than 1% (Schumacher, 2002) and ramped-pyrolysis experiments verified that little CO 2 is lost at temperature below 90°C (Rosenheim & Galy, 2012;Vetter et al., 2017;Zhang et al., 2017), effectively minimizing preparative artifacts and bolstering the bulk OC comparison here.

Age Model
The age-depth model for core L1 is derived from 137 Cs sediment ages and radiocarbon dating of benthic foraminifera ( Figure S1). The age model and its uncertainty were determined using the R package BACON (Blaauw & Christen, 2011), which is a Bayesian technique for calibration and age modeling that relies on the Marine 13 radiocarbon calibration curve.
After wet sieving, 137 Cs dating was performed on the sub-32 μm sediment fraction. Here, after homogenization, the samples were packed and sealed in 39 mm vials for analysis. Samples were counted for c. 48 h on a Canberra low-energy germanium gamma spectrometer after at least 21 days of equilibration. Activities of 137 Cs were determined from the 661.6 keV photopeak. The 137 Cs activity profile of L1 is demonstrated in Figure 2 from Hanna et al. (2014; named differently as core SL2) and Figure S2, showing a steady-state accumulation of sediments. The reliability of sedimentation rates calculated by 137 Cs analysis was verified by 239,240 Pu analysis of other sediment cores in Simpson Lagoon (Hanna et al., 2014).
Benthic foraminifera (Elphidium sp. and Buccella frigida) were picked using a binocular microscope. The isolated foraminifera were rinsed and sonicated with DI water and methanol and sent to National Ocean Sciences Accelerator Mass Spectrometry (NOSAMS) facility at the Woods Hole Oceanographic Institution for radiocarbon dating. Reservoir ages were calculated via subtraction of the extrapolated 137 Cs age from radiocarbon age directly beneath the first appearance of 137 Cs (Hanna et al., 2014). Figure S1, the employed BACON approach incorporated both 137 Cs and 14 C age constraints, culminating an integrated age model. As is common when comparing sedimentation rates over different timescales (Baskaran et al., 2017;Davies et al., 2018), the upper 50 cm of core accumulated faster than its remainder: a feature driven by time-averaging of stratigraphic hiatuses combined with the evolution of fluvial load and oceanographic conditions (Miller & Kuehl, 2010). Nevertheless, as the depositional ages obtained from the two dating methods for upper samples are much younger than the actual 14 C ages of bulk OC and LCFAs (Tables S1, S2), the impact on the pre-depositional ages is limited.

Bulk TOC and δ 13 C Analysis
Aliquots (c. 5-40 mg) of homogenized sample powders were weighed into silver capsules and decarbonated via HCl fumigation (Harris et al., 2001). After drying at 60°C for 24 h, the capsules were encapsulated in tin and their total organic carbon (TOC) and stable carbon isotope (δ 13 C) values were determined via flash combustion using a Carlo Erba (NA1500 CNS) elemental analyzer interfaced with a Thermo Delta V Advantage isotope ratio mass spectrometer. Because of the wide range of soil and sediment TOC content, coefficient of variation was used to demonstrate the quality control on TOC measurement, which is within 3.0% for replicate measurements of selected samples. The standard deviation of δ 13 C data are within 0.1‰ for replicate measurements of selected samples.

Isolation of Long-Chain Fatty Acids
Lipids were extracted in dichloromethane:methanol (9:1, v/v) via accelerated solvent extraction and saponified with 15 mL 0.5M KOH in pure methanol and ∼150 μL Milli-Q water for 2 h (70°C). After hexane extraction of the neutral fraction, the remaining aqueous solution was acidified and extracted with hexane:dichloromethane (4:1, v/v). The acidic fraction was evaporated to incipient dryness and derivatized with methanoic BF 3 to form fatty acid methyl esters (FAMEs). FAMEs were purified initially via elution with dichloromethane from a 2 cm silica gel column. Following a second silica gel column-pass, the purified FAMEs were collected in hexane:dicholoromethane (1:1, v/v) subsequent to matrix elution in hexane. The hexane:dicholoromethane fraction was used for collecting individual long-chain FAME on a preparative capillary gas chromatograph (PC-GC) (Eglinton et al., 1996). Given their comparable 14 C ages (Feng et al., 2015), C 24 , C 26 , C 28 , C 30 , and C 32 fatty acids were ultimately combined for radiocarbon analysis. The ZHANG ET AL.  purity of the combined FAME mixtures was verified by elution from a silica gel column and subsequent GC-FID analysis.

Radiocarbon Analysis of Bulk OC and LCFAs
Aliquots of carbonate-free sediments and purified FAME fractions were graphitized at the University of Florida and measured at NOSAMS facility, Woods Hole Oceanographic Institution. Here, upon sealing under vacuum in quartz tubes, carbon was oxidized to CO 2 in the presence of CuO at 900°C for 3 h. The resultant gas was then cryogenically purified before isolation in Pyrex® reactor tubes and was converted to graphite via zinc reduction (Xu et al., 2007) with sequential combustion at 500°C for 3 h and 550°C for 4 h. Radiocarbon ages of bulk OC and LCFAs were corrected for preparative and instrumentally introduced biases using primary standard (oxalic acid 4990C), secondary standard oxalic acid C7 and coal blanks, assuming constant contamination introduced during GC collection and on vacuum line (Haghipour et al., 2018).
The remaining bulk sediments and FAME fractions were converted to CO 2 via combustion using an elemental analyzer (EA) interfaced with a Mini carbon dating system (MICADAS) at the accelerator mass spectrometry (AMS) facility at ETH, Zurich. In detail, bulk sediments were first decarbonated via a 72 h fumigation with 37% HCl (TM grade) at 70°C. Carbonate-free residues were neutralized with NaOH, dried and combusted in tin capsule where the product CO 2 was measured by EA-AMS. Naturally, the decarbonation process was not necessary for the purified FAME fractions, which were immediately transferred to liquid-specific capsules using dichloromethane:hexane (1:1, v:v) and analyzed by EA-AMS. Samples were normalized using oxalic acid II (NISTSRM4990C) and corrected for constant contamination introduced during fumigation and combustion (Haghipour et al., 2018).
For the LCFAs, the blank contribution associated with PC-GC FAME collection was obtained by identical treatment of a 14 C-dead C 28 alkane (O-504, Sigma-Aldrich) and a 14 C-modern C 32 alkane (D223107, Sigma-Aldrich) standard. These standards were collected via PC-GC preparation after 40 and 80 injections, demonstrating the procedurally induced addition of 1.9 ± 0.38 μg of C with a fraction modern (Fm) of 0.68 ± 0.21. The 14 C ages of FAMEs were then corrected for the addition of a radiocarbon-dead methyl group during derivatization to obtain mean 14 C age of LCFAs where Fm LCFA is the final fraction modern of fatty acids, Fm FAME is the fraction modern for FAMEs after correction for instrumental contamination and n c is the weighted average carbon number in the FAME mixture: where f c25 , f c27 , f c29 , f c31 , and f c33 are the relative proportions of FAMEs in a given mixture, corresponding to C 24 , C 26 , C 28 , C 30 , and C 32 fatty acids.

Calculation of Pre-Depositional Age and Its Significance
The calibrated sample age (X) describes the 14 C age of a given sample after correction for preparative and instrumentally induced biases, while the calibrated depositional age (Y) is the age obtained from age model. The difference between these ages yields the pre-depositional age (i.e., Z = X − Y). The uncertainty (σ Z ) associated with the pre-depositional age (Z), therefore, propagates from those associated with X and Y, summing in quadrature to give: ). For LCFAs, pre-depositional mean age was used as the 14 C composition actually reflects the weighted mean of the mixed fatty acids inventory of a given sample (Eglinton et al., 2021).

Statistical Analysis
We compared the pre-depositional mean 14 C ages of LCFAs of sediment samples using a t-test, a Wilcoxon rank sum test, and a Monte Carlo balanced bootstrap approach in R program. For Monte Carlo balanced bootstrap approach, the pre-depositional mean LCFAs ages and their associated uncertainties (σ Z ) were used to generate a normal distribution of 10 6 for each sample using rnorm function. These distributions were then sampled without replacement and subsampled into 10 6 bootstrap replicates with the intention of matching the original sample structure of the LCFAs data.

Results
The down-core stable carbon isotope (δ 13 C) record of bulk OC is relatively invariant over the past c. 2,700 years, with an average δ 13 C value of −26.5 ± 0.2‰ (1 S.D, Table S1). Contrasting with 13 C-enriched marine inputs, this average δ 13 C value is nearly identical to that of the fluvial deposits near the river mouth (−26.6‰) which most likely reflect sediments delivered during the spring freshet (Feng et al., 2015), implying the dominance of riverine inputs at this site. This agrees with the terrestrial biomarker results from the upper 50 cm of L1 (Zhang et al., 2015) and modeled dominance of OC sourced from the Colville River in L1 surface sediments (83%; Schreiner et al., 2013). The pre-depositional 14 C ages of bulk OC range from 977 to 14,248 years, either are younger than, or approximate, the bulk 14 C ages of the soil samples analyzed from the Colville River watershed (Table S2). Considering the dilution effect of OC from fresh surface soils and minor marine inputs on the bulk 14 C ages, these pre-depositional 14 C ages of bulk OC imply continuous permafrost OC inputs to the delta.
Over the past c. 2,700 years, the pre-depositional 14 C ages of bulk OC show moderate variability, with a pronounced outlier deposited at c. 900 cal. y BP (Figure 2; Table S2). Given that the OC at this depth is much older than the enveloping sediment and, indeed, the record as a whole, we suggest that this sample records an interesting but apparently atypical event that resulted in the significant addition of pre-aged carbon to the study site. Excluding this outlier, the pre-depositional 14 C age of bulk OC defines a gradual decreasing trend that parallels the long-term cooling trend seen between c. 1,200 cal. y BP and c. 100 cal. y BP (Figure 2). Based on the temperature reconstructions in central Brooks Range and the Arctic (Figure 2), c. 100 cal. y BP marks the termination of the Little Ice Age (LIA), beyond which the pre-depositional 14 C ages of bulk OC show an apparent increase that parallels the abrupt temperature rise seen over the past c. 160 years. The youngest pre-depositional 14 C age of bulk OC (977 ± 142 years; Figure 2) was observed near the end of LIA. Subsequent warming coincides with highly variable but generally older pre-depositional 14 C ages of bulk OC (4,311 ± 2,022 years, n = 7). No significant relationship was observed between TOC (Table S1) and the pre-depositional 14 C ages of bulk OC over the past c. 2,700 years. Since bulk sedimentary OC at this location mainly reflects a mixture of terrestrially sourced pools (e.g., surface and deep soils), we turn to the pre-depositional mean 14 C age of LCFAs to understand changes in pre-aged terrestrial matter delivery over the last c. 2,700 years (Table S3).
Based on the noted coupling between pre-depositional 14 C ages of bulk OC and temperature (Figure 2), we subdivided the pre-depositional mean LCFAs 14 C data into pre (n = 12) and post (n = 7) the end of LIA groups, with a discriminatory threshold at c. 100 cal. y BP. The post-LIA samples possessed significantly older LCFAs 14 C ages (t-test and Wilcoxon test, p ≤ 0.02) than their older counterparts. A similar conclusion was also reached via application of a balanced Monte Carlo-type bootstrap analysis. Here, LCFAs housed within post-LIA sediments were found to possess significantly older 14 C age (929 years) compared with their older counterparts (Figure 4).

Discussion
Radiocarbon ages of particulate permafrost OC provide perhaps the best means of tracing changes in permafrost OC release (Wild et al., 2019). For instance, the Kolyma River, within the continuous permafrost zone of eastern Siberia, features particulate OC (POC) with an extremely high pre-aged/modern ratio, signaling the dominance of permafrost OC export throughout its watershed (Wild et al., 2019). For Colville River, the radiocarbon ages of bulk OC from suspended particles collected at the river mouth (6,773 years; Eglinton et al., 2021) and that from river bed sediments collected in the main channel (10,000 y BP; Zhang et al., 2017) imply mobilization of permafrost via the fluvial system. As stated before, this site is dominated by riverine inputs based on δ 13 C and biomarker results compared to other sites in the Colville delta and Simpson Lagoon region (Schreiner et al., 2013;Zhang et al., 2015). The barrier islands shield L1 from intense ice process and make it a depocenter with highest sedimentation rate compared to other sites in the region (Hanna et al., 2014). Deposited at this delta, lateral transport-associated 14 C aging of terrestrial OC should be very limited (Bröder et al., 2018). The downcore profile of 137 Cs results also imply this site to be a stable deposition environment (Hanna et al., 2014;Lima et al., 2005). The accumulation rate of sediments at this site is relatively stable over the past ∼2,700 years according to the age model, except for the uppermost sediments deposited in recent warming period. Thus, the undisturbed sediment record provides an ideal archive from which to decipher changes of permafrost OC exported from the watershed.
Although evidence of permafrost export from Arctic river systems and subsequent coastal deposition in surface sediments has been reported (Feng et al., 2013;Schreiner et al., 2014), our results demonstrate that riverine mobilization of permafrost OC is a long-lived phenomenon that has been occurring at least to some degree for thousands of years. This is consistent with the mobilization of permafrost OC from soil to river and finally to the delta, as indicated by evidence from 14 C ramped pyrolysis and lignin biomarkers (Zhang et al., 2017). The close associationof the 14 C age of bulk OC with temperature reconstructions from Brooks Range and the Arctic over the past c. 1,300 years ( Figure 2) likely reveals the modulating role of climate in arctic permafrost dynamics. Dominated by riverine OC inputs, the observed post-LIA increase in pre-depositional 14 C age of bulk OC can, in principle, be attributed to either a decreased proportion of young OC or heightened percentage of aged OC. Considering contemporary vegetation expansion (Tape et al., 2006) and relatively higher TOC content in the post-LIA-aged sediments (Figure 3a), diminished fluxes of young and fresh surficial OC inputs are considered unlikely. Therefore, the observed increase in pre-depositional 14 C age of bulk OC is more readily attributable to a possible increase of permafrost-derived OC inputs and/or a progressive mobilization of an older, and presumably deeper, OC reservoir. Moreover, since sedimentary LCFAs have been shown to originate mostly from deep permafrost (Feng et al., 2013), the greater pre-depositional mean 14 C age of LCFAs in the aftermath of the LIA possibly lends further support to deeper permafrost-derived OC delivery over the last 160 years (Figure 4). These changes are not likely caused by diagenetic process which would be expected to preferentially remove more labile (fresher and younger) OC with increasing diagenesis/depth (Soetaert et al., 1996;Wakeham et al., 1997), contrary to our results showing older 14 C ages toward the top of the core. In contrast, post-depositional diagenesis may have dampened the change in the pre-depositional mean ages.
The increase of permafrost-derived OC inputs and/or mobilization of an older and deeper permafrost OC reservoir may stem from several processes, broadly reflecting thermally driven catchment-housed permafrost destabilization. Arctic river systems are highly seasonal, featuring disproportionately high discharge and sediment loads coincident with the spring thaw (Bring et al., 2016;Gordeev, 2006). During this period, ∼62% of Colville River's annual sediment load is estimated to transit its catchment, derived, principally, from bank erosion (Walker & Hudson, 2003). The rates of thermal erosion of riverbanks in the Colville main channel and its main tributary (Itkillik River) were averaged to be 0.9 m y −1 (Stephani et al., 2019) and 11.4 m y −1 (Kanevskiy et al., 2016), respectively. Here, in the absence of insulating vegetation and surface soils , permafrost exposed within the eroding bluffs is directly exposed to water and warming air ( Figure S3). Furthermore, within the eroding riverbanks, top soils are extremely thin compared to the underlying permafrost ( Figure S3), thereby bank erosion should enhance the preferential liberation of aged OC, potentially explaining the greater pre-depositional 14 C age of bulk OC and LCFAs over the last c. 160 years. Apart from bank erosion, widespread degradation of ice wedges, which form thermokarst ponds, has occurred across the Arctic (Jorgenson et al., 2015;Liljedahl et al., 2016). The increase in ice-wedge thermokarst in Alaska is likely related to the exceptionally warm and wet summers observed in 1989 and 1998 (Jorgenson et al., 2015). The intense destabilization of permafrost through thermokarst erosion may also facilitate bank erosion to some extent, providing another potential source of the pre-aged carbon found in post-LIA sediments.
Underlain by continuous permafrost, the lateral transport of OC in the watershed may not show apparent changes. Despite increased temperature in this region, thaw depth to the active layer has apparently remained stable (Hollister et al., 2006;Shiklomanov et al., 2012). Precipitation in northern Alaska has more or less remained constant over the same period (Bieniek et al., 2014;McAfee et al., 2013). In addition, the low permeability of the frozen soils will have largely curtailed water flow and storage (Bring et al., 2016), indicating limited changes on the export of OC by lateral transport. Even if there was some extent of increase of OC from the upper soils, most likely in the summer, the extremely low water discharge (Holmes et al., 2008) may not be able to deliver most of the upper soil OC to the delta.
Brooks Range and Greenland temperature reconstructions (Boldt, 2015;Cappelen & Vinther, 2014;McKay & Kaufman, 2014) reveal prior intervals of warming, yet these warming intervals show different characteristics when compared to the post-LIA warming episode (Figure 3). For example, during the Roman warming interval (c. 2,700-2,100 cal. y BP), TOC (n = 22) was significantly lower than that seen during the post-LIA (n = 18) warming episode (t-test, p < 0.01), yet the pre-depositional 14 C ages of bulk OC are not significantly different from the post-LIA (Figure 3). While only two pre-depositional mean 14 C ages of LCFAs are available from Roman warming period, they are younger (4,880 ± 681 y BP, n = 2) than those from the post-LIA warming era (6,138 ± 810 y, n = 7), potentially signaling a less pronounced release of pre-aged carbon. We speculate that the generally cooler prevailing temperatures at the beginning of the post-LIA warming ZHANG ET AL. 10.1029/2021AV000396 8 of 12 Figure 3. A geochemical and geochronological record of late Holocene permafrost thawing within the Colville River watershed. (a) Total organic carbon (TOC); (b) pre-depositional 14 C age of bulk organic carbon; (c) pre-depositional mean 14 C age of long-chain fatty acids (LCFAs); (d) summer temperature anomaly for Kurupa Lake in Brooks Range, Alaska (e) GISP2 ice-core-derived (black; Alley, 2004) and LOESS smoothed Greenland temperature record (gray; Cappelen & Vinther, 2014). Pre-depositional 14 C ages are plotted using the age depth model illustrated in Figure S1. Orange columns depict respective warming episodes discussed within the text and blue column highlights the Little Ice Age. Black horizontal lines in (a) and (c) demonstrate the mean value of parameters in those two warming episodes. episode compared with those for the Roman warming trend (6°C and 1°C cooler in Brooks Range and Greenland, respectively; Figures 3d and 3e), coupled with a more rapid temperature increase (551 and 12 times faster in Brooks Rang and Greenland, respectively; Figures 3d and 3e), may have played an important role in modulating permafrost dynamics and/or the balance between vegetation-induced carbon burial and thermally driven permafrost losses. Furthermore, mobilization of permafrost OC may have also been greater during the post-LIA warming because of preceding glacier expansion, and the corresponding growth of fresh-water stocks during the LIA (Solomina et al., 2016;Wanner et al., 2008). Accelerated warming may have triggered extensive glacial melting (Chen et al., 2006;Vargo et al., 2020), which would enhance infiltration and bank erosion, inducing more OC mobilization throughout the watershed. Rising Arctic temperatures should also increase plant carbon uptake (Elmendorf et al., 2012;Pearson et al., 2013), offering a negative feedback to offset the carbon losses from permafrost thawing. This effect, however, may be more important during a more modest warming regime, where the ecosystem would have more time to adapt to the climate perturbation (Gałka et al., 2018). Clearly the involvement and significance of these competing mechanisms requires thorough exploration; however, for now, it is clear that permafrost destruction cannot be linearly extrapolated from temperature reconstructions.
Mechanistic uncertainties notwithstanding, the observed increase in the 14 C ages of bulk OC and LCFAs archived in post-LIA sediments deposited in the Colville River delta indicates an increased thawing and mobilization of permafrost OC over the last c.160 years. This increase in 14 C age occurs in concert with a positive inflection in temperature reconstructions from Simpson Lagoon (Hanna et al., 2018), lakes within the Brooks Range (Bird et al., 2009;Boldt et al., 2015), and the wider Arctic region (Alley, 2004;Cappelen & Vinther, 2014;McKay & Kaufman, 2014). Permafrost thawing has recently been linked to CO 2 -driven temperature fluctuations operating over glacial-interglacial transitions (Crichton et al., 2016). Our results, therefore, lend support to the notion that such fluctuations can influence permafrost dynamics at sub-centennial timescales. Moreover, our down-core study provides relatively reliable evidence linking permafrost thawing and the export of pre-aged terrestrial OC via Arctic river systems, serving as a baseline for future pan-Arctic comparisons. The contrast between Roman and post-LIA warming trends imply that the rate of warming and the prevailing boundary conditions are important modulators of permafrost thawing; ZHANG ET AL.  . Bootstrap-generated 14 C age probability density distributions derived from pre-depositional mean long-chain fatty acids (LCFAs) ages in periods pre (>100 cal. y BP) and post (<100 cal. y BP) (a) the end of Little Ice Age and (b) their difference after subtraction. Mean ages and parenthesized 95% confidence intervals accompany each distribution. however, more work is clearly needed to refine these suppositions. Accordingly, the application of 14 C analysis of bulk OC and source-diagnostic compounds to long-term sediment records from other Arctic coastal regions remains a priority. In particular, the geological record offers unparalleled insights into how past changes in the magnitude and pace of warming and associated changes in boundary conditions (e.g., prevailing temperature, and glacial structure/dynamics) may influence the fate of the Arctic's permafrost reserves. Understanding of permafrost-driven climate feedbacks remains a fundamental prerequisite for accurate climate simulations.