Arctic Amplification of Global Warming Strengthened by Sunlight Oxidation of Permafrost Carbon to CO2

Once thawed, up to 15% of the ∼1,000 Pg of organic carbon (C) in arctic permafrost soils may be oxidized to carbon dioxide (CO2) by 2,100, amplifying climate change. However, predictions of this amplification strength ignore the oxidation of permafrost C to CO2 in surface waters (photomineralization). We characterized the wavelength dependence of permafrost dissolved organic carbon (DOC) photomineralization and demonstrate that iron catalyzes photomineralization of old DOC (4,000–6,300 a BP) derived from soil lignin and tannin. Rates of CO2 production from photomineralization of permafrost DOC are twofold higher than for modern DOC. Given that model predictions of future net loss of ecosystem C from thawing permafrost do not include the loss of CO2 to the atmosphere from DOC photomineralization, current predictions of an average of 208 Pg C loss by 2,299 may be too low by ~14%.


Introduction
Current estimates are that 5% to 15% of the ∼1,000Pg of the soil organic carbon (C) stored in surface permafrost soils could be emitted as greenhouse gases by 2,100 given the current trajectory of global warming (Plaza et al., 2019;Schuur et al., 2015), with additional C lost in lateral transfer from soils to surface waters (Plaza et al., 2019). Models assessing the sensitivity of the climate system to thawing permafrost soils estimate that decomposition of organic C in these soils could result in 0.3°C to 0.4°C additional global warming (i.e., arctic amplification) by 2,100 to 2,299, respectively (McGuire et al., 2018). However, none of these predictions include the oxidation of organic C upon export to sunlit surface waters.
Oxidation of dissolved organic carbon (DOC) to carbon dioxide (CO 2 ) by sunlight (photomineralization) currently accounts for up to 30% of the CO 2 emitted to the atmosphere from arctic surface waters (Cory et al., 2014). As permafrost DOC is exported to sunlit waters, its oxidation to CO 2 will depend on whether permafrost DOC is labile to photomineralization, which is currently debated (Selvam et al., 2017;Shirokova et al., 2019;Stubbins et al., 2016;Ward & Cory, 2016). The lability of terrestrially derived DOC to photomineralization is hypothesized to depend on iron and DOC chemical composition (Gao & Zepp, 1998;Gu et al., 2017;Miles & Brezonik, 1981;Ward & Cory, 2016;Xie et al., 2004). To test these hypothesized controls, we made the first direct measurements of the amount, source, and age of CO 2 produced from photomineralization of permafrost DOC collected on younger and older glacial surfaces, and from two common vegetation types in the Arctic (Table S1; Ping et al., 1998;Trusiak et al., 2019). Here we show that (i) the lability of permafrost DOC to photomineralization depends on sunlight wavelength, (ii) iron controls the lability of permafrost DOC to photomineralization, and (iii) old carboxylic acid C (4,000 to 6,300 a BP) derived from lignin and tannin is mineralized to CO 2 by sunlight. Collectively, our results support the inclusion of photomineralization in model predictions and experimental studies of arctic amplification of climate change.

Permafrost Soil Collection
Soils were collected from the frozen permafrost layer (>60 cm below the surface) at five sites underlying moist acidic tussock or wet sedge vegetation, and on three glacial surfaces on the North Slope of Alaska during summer 2018 (Table S1; Hobbie & Kling, 2014;Mull & Adams, 1989;Walker et al., 2005). See the Supporting Information for soil collection protocols, including precautions to minimize radiocarbon ( 14 C) contamination, and for a summary of the experimental design. DOC was leached from each permafrost soil (leachate) at the Woods Hole Oceanographic Institution (WHOI) as described in the Supporting Information.

Δ 14 C and δ 13 C Analyses of DOC
The 14 C and stable carbon ( 13 C) isotopic compositions of DOC were analyzed from each permafrost leachate at the National Ocean Sciences Accelerator Mass Spectrometry (NOSAMS) facility at WHOI (Table S2) following Beaupré et al. (2007). Each permafrost leachate was diluted with UVC-oxidized MilliQ water (Millipore Simplicity ultraviolet, UV, system; 1.5 hr; 1,200-W medium pressure mercury arc lamp) to achieve a total C mass between 800 and 2,000 μg. The diluted permafrost leachate was acidified with UVC-oxidized trace-metal grade phosphoric acid (85%) to pH < 2 in a precombusted quartz reactor (450°C; 4 hr), and the dissolved inorganic carbon (DIC) was purged with high-purity helium gas in the dark. The DOC was then oxidized with UVC light to DIC for 4 hr (1,200-W medium pressure mercury arc lamp), and the resultant CO 2 was extracted cryogenically. On average, 1,370 ± 240 μg of C were extracted from each permafrost leachate (±1 standard error, SE; n = 6; Table S2). A subsample of the CO 2 was analyzed for 13 C using a VG Prism-II or Optima stable isotope ratio mass spectrometer (instrumental precision of 0.1‰; Coplen et al., 2006), and the δ 13 C (‰) was calculated as follows: where 13 R is the isotope ratio of a sample or standard (VPDB), as defined by 13 R ¼ 13 C= 12 CÞ: À The remaining CO 2 was reduced to graphite with H 2 and an iron catalyst, and then analyzed for 14 C isotopic composition using an accelerator mass spectrometer at the NOSAMS facility (Longworth et al., 2015). The Δ 14 C (‰) and radiocarbon age of DOC were calculated from the fraction modern (McNichol et al., 2001;Stuiver & Polach, 1977) using the oxalic acid I standard . Δ 14 C analyses of DOC had an instrumental precision of 2-6‰ (Longworth et al., 2015;McNichol et al., 2001).
DOC leached from one permafrost soil (Toolik moist acidic tundra) was prepared and analyzed for 14 C and 13 C twice to quantify the standard error of duplicate analyses (Table S2). Δ 14 C and δ 13 C analyses of DOC had standard errors of 1‰ and 0.1‰, respectively (1 SE; n = 2; Table 1). A procedural blank was quantified manometrically by oxidizing MilliQ water with UVC light in a precombusted quartz reactor (450°C; 4 hr) for 1.5 hr, acidifying to pH < 2, and purging the DIC as described above. The procedural blank was 4 μg of C, which was <0.5% of the total C masses extracted from the permafrost leachates.

Apparent Quantum Yield Spectra
The CO 2 produced from photomineralization of permafrost DOC was measured as a function of sunlight wavelength. The lability of DOC to photomineralization is defined as the apparent quantum yield spectrum (CO 2 produced per mol photon absorbed by DOC; hereafter called the yield spectrum, φ PM,λ ). Yield spectra of permafrost DOC were directly measured for the first time with a custom-built high-powered (≥100 mW), narrow-banded (±10 nm) light-emitting diode (LED) system from soils collected in 2018 ( Figure S1). Each permafrost leachate was equilibrated to room temperature (~24 hr) and then placed in 20 gas-tight, flat-bottomed precombusted (450°C; 4 hr) 12-mL quartz vials with butyl rubber septa and GL-18 caps (light-exposed vials) and four gas-tight precombusted (450°C; 4 hr) 12-mL borosilicate exetainer vials (dark control vials; Labco, Inc.). Vials were placed in an inner aluminum housing (painted black matte to minimize light scattering), with the flat bottom facing upward toward the light source, and then exposed to ≥100-mW, narrow-banded (±10 nm) LEDs at 278, 309, 348, 369, and 406 nm alongside the dark controls for 12 or 30 hr (Table S3). The LEDs were tuned such that each permafrost leachate absorbed the same amount of light at each wavelength. After LED exposure, light-exposed and dark control waters were immediately analyzed for DIC (Apollo SciTech, Inc.) and for chromophoric dissolved organic matter (CDOM; Cory et al., 2014). The experiment above was then repeated for the analysis of photochemical oxygen (O 2 ) consumption to quantify the apparent quantum yield spectra of photo-oxidation (φ PO,λ ) from each permafrost DOC. Dissolved O 2 was measured in each light-exposed and dark control vial on a membrane inlet mass spectrometer (Bay Instruments; Kana et al., 1994).
At each LED wavelength, φ PM,λ and φ PO,λ were calculated as the concentration of DIC produced and O 2 consumed, respectively, divided by the light absorbed by CDOM. The amount of light absorbed by CDOM (mol photon m −2 nm −1 ) was quantified for each vial exposed to a LED using absorption coefficients of CDOM (a CDOM,λ ) and the photon flux spectrum (Cory et al., 2014). The photon flux spectrum was quantified from the solar irradiance spectrum from each LED source, which was measured by radiometry and chemical actinometry (see the Supporting Information). φ PM,λ and φ PO,λ are reported as the average ± 1 SE of experimental replicate vials (n = 4).

Δ 14 C and δ 13 C of CO 2 Produced From Light
The Δ 14 C and δ 13 C of DIC produced following exposure of DOC to UV and visible light were quantified from permafrost leachates prepared from each permafrost soil collected in 2018, except for Sagwon moist acidic tundra ( Figure S1). Each permafrost leachate was equilibrated to room temperature and then placed in up to four precombusted (450°C; 4 hr) 600-mL quartz flasks with ground glass stoppers and no headspace. The flasks were exposed to custom-built LED arrays consisting of ten ≥100 mW, narrow-banded (±10 nm) 309-or 406-nm chips alongside one or two foil-wrapped dark control flasks (Table S2). Exposure times ranged from 8 to 25 hr to achieve similar concentrations of DIC produced from each permafrost DOC sample and at each wavelength (Table S4).
After LED exposure, foil-wrapped light-exposed and dark control flasks were immediately transferred to foilwrapped, precombusted 500-mL borosilicate glass bottles (450°C; 4 hr) in a N 2 -filled glove bag, preserved with saturated mercuric chloride, and plugged with gas-tight ground glass stoppers (McNichol et al., 1994). Those bottles were stored in the dark at room temperature for ≤1 week until preparation for carbon isotope analyses at the NOSAMS facility. Bottles were kept foil-wrapped while each water sample was acidified with trace-metal grade phosphoric acid (85%) to pH < 2 and stripped of DIC with high-purity N 2 gas. The resultant CO 2 was trapped and purified cryogenically and its concentration was quantified manometrically. The 14 C and 13 C of the CO 2 were analyzed at the NOSAMS facility (Table S2) and converted to Δ 14 C and δ 13 C values as described above. Δ 14 C analyses of DIC had an instrumental precision of 1-2‰ (Longworth et al., 2015;McNichol et al., 2001). The reported precision of δ 13 C is 0.1‰ (Coplen et al., 2006).
The Δ 14 C and δ 13 C of CO 2 produced from the photomineralization of DOC were calculated as follows: The Δ 14 C and δ 13 C of the CO 2 produced from photomineralization of permafrost DOC are reported following exposure to LEDs at 309 (UV) or 406 nm (visible). When available, all values are reported as the average ± 1 SE of experimental replicates (n = 2). (3) The Δ 14 C and δ 13 C of CO 2 produced in each light-exposed flask were calculated relative to one or two dark controls (Tables S2 and S4) and are reported as the average ± 1 SE of replicate values for the experiments conducted alongside two dark controls (Table 1). The concentration, Δ 14 C, and δ 13 C of DIC in the dark controls are reported as the average ± 1 SE of replicate flasks (n = 2; Table S4). The 14 C age of CO 2 produced is the age of DIC in light-exposed flask minus the 14 C age of DIC in the dark control (Table 1). This approach to quantify the Δ 14 C and δ 13 C of CO 2 produced from photomineralization of organic C was previously described in detail for polystyrene (Ward et al., 2019). In this previous study, experimental reproducibility of Δ 14 C and δ 13 C of CO 2 produced from photomineralization was 5‰ and 0.1‰, respectively (±1 SE; n = 3).

Results and Discussion
All permafrost DOC was labile to photomineralization at all wavelengths measured (Figure 1a), and the yield spectrum always decreased exponentially with increasing wavelength from the UV to the visible (p < 0.05; Figure 1a). The magnitude of the photomineralization yield varied up to eightfold among permafrost DOC samples ( Figure 1a) and was significantly, positively correlated with the concentration of dissolved iron (p < 0.001 as shown at 309 nm in Figure 1b; see the Supporting Information). There were no significant correlations of the photomineralization yield with dissolved cations other than iron or with any measure of DOC concentration or composition (see the Supporting Information).
Our results are the first to demonstrate in natural samples that the lability of permafrost DOC to photomineralization is controlled by dissolved iron. Although photomineralization of terrestrially derived DOC has

Geophysical Research Letters
been shown to increase with addition or decrease with removal of iron in the laboratory (Gao & Zepp, 1998;Gu et al., 2017;Miles & Brezonik, 1981;Xie et al., 2004), evidence is lacking for a mechanism by which iron enhances photomineralization. Here we present new experimental evidence and a synthesis of literature results that collectively support a mechanism of iron-catalyzed photo-decarboxylation of lignin-and tannin-derived carboxylic acids within old permafrost DOC.
Iron is hypothesized to catalyze the photo-decarboxylation of organic acids by a ligand-metal-charge-transfer reaction (Mangiante et al., 2017) where Fe(III) is a cyclic catalyst that is photo-reduced to Fe(II) while the C in carboxylic acids (hereafter "carboxyl C") is oxidized to CO 2 (Gao & Zepp, 1998;Miles & Brezonik, 1981;Xie et al., 2004). Two lines of evidence from our study strongly support iron-catalyzed photo-decarboxylation of permafrost DOC to CO 2 . First, loss of carboxyl C (quantified by 13 C nuclear magnetic resonance) upon exposure of permafrost DOC to sunlight was significantly, positively correlated with the dissolved iron concentration (p < 0.05; Figure S2), as expected if photo-decarboxylation is the mechanism of CO 2 production. The only other study that quantified photochemical loss of carboxyl C from permafrost DOC concluded that it accounted for up to 90% of the CO 2 produced from photomineralization (Ward & Cory, 2016). However, this prior study used DOC from one site (with only one iron concentration) and thus was unable to link iron abundance to DOC photo-decarboxylation. Second, the ratio of photochemical CO 2 production per dissolved O 2 consumption by DOC was ≥1 for all permafrost DOC that also contained >1-μM total dissolved iron ( Figure S3). A ratio ≥1 for photochemical CO 2 produced per O 2 consumed is considered evidence for photo-decarboxylation because this reaction is expected to proceed with a stoichiometry of ≥2:1 mol CO 2 produced per mol O 2 consumed (Miles & Brezonik, 1981;Xie et al., 2004). While ratios ≥1 for photochemical CO 2 produced per O 2 consumed have previously been observed in high-iron waters (Cory et al., 2015;Miles & Brezonik, 1981;Xie et al., 2004), here we show ratios ≥1 for DOC from various permafrost soils concurrent with photochemical loss of carboxyl C.
Isotopic signatures of the CO 2 produced by sunlight indicate that iron is catalyzing the oxidation of carboxyl C attached to organic matter derived from lignin and tannin. Photochemical production of 13 C-depleted CO 2 (Table 1) increased significantly with the ratio of photochemical CO 2 produced per mol of O 2 consumed (p < 0.05; Figure S4). Therefore, as photo-decarboxylation accounts for more of the total CO 2 produced (as indicated by increasing CO 2 /O 2 ; Figure S4), the source of the CO 2 is increasingly 13 C-depleted carboxyl C, such as that derived from lignin and tannin (Ball & Aluwihare, 2014). Photochemical production of 13 Cdepleted CO 2 is interpreted as resulting from photomineralization of lignin-or tannin-derived DOC (Franke et al., 2012;Spencer et al., 2009) because lignin and tannin are relatively more depleted in 13 C compared to other fractions of DOC (Benner et al., 1987; see the Supporting Information). In addition, the 13 C enrichment of DOC remaining after photomineralization has been correlated with photochemical loss of lignin (Spencer et al., 2009), and high-resolution mass spectrometry revealed that lignin-and tannin-derived compounds within permafrost DOC are preferentially degraded by sunlight compared to other fractions of DOC (Ward et al., 2017;Ward & Cory, 2016). Thus, our results indicate that iron-catalyzed photodecarboxylation of lignin and tannin in permafrost DOC is producing CO 2 (Table 1, Figures S2 and S4).
The carboxyl C derived from lignin and tannin that was photomineralized to CO 2 was old, from 4,000 to 6,300 a BP (Table 1, Figure 2). The Δ 14 C composition of CO 2 produced from photomineralization of permafrost DOC (−546‰ to −397‰) was always ≤70‰ different from the initial, bulk Δ 14 C-DOC signature (−585‰ to −411‰; Table 1, Figure 2). The linear relationship between the initial, bulk permafrost Δ 14 C-DOC and the Δ 14 C-CO 2 produced by photomineralization (p < 0.05; Figure 2) indicates that the bulk age of permafrost DOC was a strong predictor of the age of DOC photomineralized to CO 2 . Collectively, our

10.1029/2020GL087085
results demonstrate that old carboxyl C (4,000 to 6,300 a BP) derived from lignin and tannin and associated with iron is photomineralized to CO 2 .
The presence of iron may explain contrasting literature results from high (Selvam et al., 2017;Shirokova et al., 2019;Ward & Cory, 2016) to little or no (Shirokova et al., 2019;Stubbins et al., 2016) lability of permafrost DOC to photomineralization. For example, undetectable photomineralization of permafrost DOC from Russian arctic thaw slumps (Stubbins et al., 2016) may have been due to the 100-fold dilution of the DOC with deionized water. Although dissolved iron was not reported (Stubbins et al., 2016), dilution likely also resulted in the precipitation of iron (oxy)hydroxides and thus lower dissolved iron concentrations (Miller et al., 2009). In a study of Russian arctic surface waters that likely contained permafrost DOC (Shirokova et al., 2019), up to 13% of the DOC pool was photomineralized to CO 2 , consistent with the presence of dissolved iron (3-7 μM). Provided that all permafrost DOC contains carboxyl C (Feng et al., 2017;Ward et al., 2017;Ward & Cory, 2016) and that permafrost soils generally contain high levels of leachable iron (Herndon et al., 2015;Heslop et al., 2019;Ping et al., 1998;Trusiak et al., 2019), arctic permafrost DOC is labile to photomineralization in proportion to the iron present. Given that the export of iron is currently strongly, positively correlated with DOC export from arctic soils to surface waters (Trusiak et al., 2019), we expect that iron and DOC export may continue to covary as permafrost soils thaw. Therefore, we predict that the yield spectrum of permafrost DOC will be within the range reported here in the future (Figure 1a).
The photomineralization yield spectra of permafrost DOC directly measured in this study have significantly shallower spectral slopes compared to those quantified indirectly for arctic surface water DOC (two-tailed, unpaired t test; p < 0.001; Figure S5). Thus, permafrost DOC has relatively lower lability to photomineralization at UV wavelengths and higher lability at visible wavelengths (different at 95% confidence interval; Figure S5). If permafrost DOC comprises 100% of the DOC in surface water, photomineralization rates will increase by twofold compared to current rates (Figure 3a) due to the higher lability of permafrost DOC at visible wavelengths multiplied by the approximately tenfold greater photon flux in the visible versus the UV light region (Table S5). It follows that photomineralization rates increase in proportion to the permafrost DOC exported to surface waters (Figure 3b).

Implications
The uncertainty in model predictions of future ecosystem C gain or loss crosses zero (McGuire et al., 2018). For example, under the RCP8.5 scenario, a net ecosystem loss of C of 208 ± 307 Pg C is predicted by 2,299 (average ± 1 SD). The large uncertainty in model predictions of permafrost C storage in this scenario includes a 20% probability that the net C storage is between +100 (gain) or −100 (loss) Pg C (see the Supporting Information). Photomineralization of DOC to CO 2 is always a loss to the atmosphere, and as the net C gain or loss for any particular year or over time nears zero, the relative importance of photomineralization increases. Given that photomineralization rates of permafrost DOC are nearly twofold higher than for modern DOC in arctic surface waters (Figure 3a), and assuming from 2,010 to 2,299 that 75% of DOC in surface waters was delivered from permafrost soils, then the photomineralization rates . Photomineralization rates were higher for permafrost DOC than for surface water DOC due to higher lability in the visible light region. (a) Wavelength-dependent water column rates of photomineralization for Imnavait moist acidic tundra permafrost DOC and Imnavait Creek DOC (turquoise vs. black line, respectively). Solid lines show the average photomineralization rate spectrum and the similar color shading shows the upper and lower 95% confidence intervals. Rates of photomineralization were calculated using the different wavelength-dependent apparent quantum yields for photomineralization (φ PM,λ ) for Imnavait moist acidic tundra permafrost DOC or Imnavait Creek DOC ( Figure S5; see the Supporting Information). The red dashed and solid lines mark the wavelength of peak photomineralization rates (330 nm vs. 402 nm) for Imnavait Creek and Imnavait moist acidic tundra permafrost DOC, respectively. (b) Calculated photomineralization rates increased with an increasing fraction of permafrost DOC in the surface water DOC pool. Percent increase in photomineralization rates as permafrost DOC comprises 0-100% of the DOC pool in Imnavait Creek, Kuparuk River, and Toolik Lake (compared to no permafrost DOC in the DOC pool). Only the φ PM,λ was varied in the water column rate calculations, using a "mixed" φ PM,λ calculated as a mixture of the φ PM,λ for permafrost DOC with the φ PM,λ for the surface water DOC (see the Supporting Information). All values are shown as the average ± 1 SE of calculated rates for surface water DOC mixed with each of the five permafrost DOCs in this study (n = 5).

Geophysical Research Letters
of 20 g C m −2 y −1 reported in Cory et al. (2014) would increase to 39 g C m −2 y −1 . Using a surface area of water in permafrost regions of 6% (Cooley et al., 2019),~9 Pg CO 2 could be produced from the photomineralization of permafrost DOC by 2,299 (see the Supporting Information). In addition, if potentially more than half of future terrestrial C losses are lateral in hydrologic flow (Plaza et al., 2019), a pathway missing from models used in McGuire et al. (2018), photomineralization of that C would occur upon exposure to sunlight. For example, taking the predicted net ecosystem loss of 208 Pg C by 2,299 under RCP8.5 (McGuire et al., 2018), potentially another~100 Pg C could be lost laterally and produce~21 Pg C as CO 2 from photomineralization in surface waters (Cory et al., 2014; see the Supporting Information). CO 2 from photodegradation of permafrost DOC is conservative because it does not account for (a) increased lability of permafrost DOC to microbial respiration following exposure to sunlight (Cory et al., 2013;Ward et al., 2017) and (b) increased annual sunlight exposure due to more ice-free days for surface waters in a warmer Arctic (Šmejkalová et al., 2016). Therefore, reducing the uncertainty on whether permafrost thaw will be a net sink or source of C to the atmosphere requires representing processes such as photochemistry in models of the future arctic C balance.

Data Availability Statement
All data are available in the manuscript or the Supporting Information, and data are available at the Arctic LTER (https://arc-lter.ecosystems.mbl.edu/arctic-coupled-biological-and-photochemical-degradationdissolved-organic-carbon/data-photo-coupled).