Observed and Modeled Black Carbon Deposition and Sources in the Western Russian Arctic 1800–2014

Black carbon (BC) particles contribute to climate warming by heating the atmosphere and reducing the albedo of snow/ice surfaces. The available Arctic BC deposition records are restricted to the Atlantic and North American sectors, for which previous studies suggest considerable spatial differences in trends. Here, we present first long-term BC deposition and radiocarbon-based source apportionment data from Russia using four lake sediment records from western Arctic Russia, a region influenced by BC emissions from oil and gas production. The records consistently indicate increasing BC fluxes between 1800 and 2014. The radiocarbon analyses suggest mainly (∼70%) biomass sources for BC with fossil fuel contributions peaking around 1960–1990. Backward calculations with the atmospheric transport model FLEXPART show emission source areas and indicate that modeled BC deposition between 1900 and 1999 is largely driven by emission trends. Comparison of observed and modeled data suggests the need to update anthropogenic BC emission inventories for Russia, as these seem to underestimate Russian BC emissions and since 1980s potentially inaccurately portray their trend. Additionally, the observations may indicate underestimation of wildfire emissions in inventories. Reliable information on BC deposition trends and sources is essential for design of efficient and effective policies to limit climate warming.


S. 1. Study area and lake sediment coring
The Timan-Pechora basin is dominated by lowland tundra with continuous permafrost while the West Siberian Basin constitutes a large wetland area reaching from Arctic tundra in the north to the southern edge of the taiga forest zone in the south. Table 1 presents general information on the study lakes. The Vork 9 and Kharby sediments were collected in 2014 and the Podvaty sediment core in 1998 from the Timan-Pechora Basin in the vicinity of the three major industrial centers Usinsk, Inta and Vorkuta (Fig.  1). The cores were collected with a Glew gravity corer. 1 The Kharby sediment was sampled at 0.5 cm resolution for the top 10 cm, and 1 cm resolution below 10 cm, while the Vork 9 sediment was cut at 0.5 cm resolution for the top 20 cm, and 1 cm resolution below. The Podvaty sediment was extruded at 0.5 cm intervals for the first 5 cm and then at 1cm intervals below. The lakes are located in a flat lowland tundra environment with continuous permafrost, and have no major river in-or outfluxes.
The PONE sediment was collected in 2006 from the Putorana Plateau, a more pristine environment than the Timan-Pechora Basin, although at ca. 200 km proximity to one of the most polluted mining cities of Russia, Norilsk, with a HON-Kajak corer, 2 and subsampled at 0.25 cm intervals. PONE is an unnamed lake that was given the code name PONE in a previous study 3 . Due to the lack of sediment material, PONE samples were combined so that BC quantifications could be made at 0.5 cm resolution. The lake is situated in mountainous open woodland in a hanging valley between two mountain ridges and has a single outflow stream. 3 All sediment cores were collected at the deepest point of the lakes which was determined by transect water depth measurements with a hand-held echo sounder. At the deepest part of the lakes the potential sediment resuspension is expected to be lowest. None of the sediments were laminated or varved. However, the dating results of the study sediments showed no indication of bioturbation, hiatuses or vertical mixing of the sediments. Generally, bottomdwelling species that would affect the dating of the sediments are quite rare in Arctic lake sediments.

S.3. Soot Black Carbon (SBC) analysis with the chemothermal oxidation at 375 °C method (CTO-375)
The term black carbon comprises a myriad of carbonaceous particles formed under variable incomplete combustion conditions and from different fuel materials. These particles ranging from charred biomass to highly condensed and refractory soot owe different physical properties such as size, composition and structure. Thus, BC is not a precisely defined term. Currently, no analytical method is capable to quantify all BC particles simultaneously, and no standard method exists for BC quantification. Consequently, BC is an operational term and its precise definition depends on the method used for its quantification. 7,8

S7
Here, BC was quantified with the chemo-thermal oxidation method at 375 °C (CTO-375). The method effectively quantifies the most condensed high-refractory fraction of BC, the so called soot-BC (SBC). 9 SBC is formed by gas condensation in high-temperature flames of both biomass and fossil fuel burning when combustion temperatures are sufficient. Additionally, SBC may include some char-type BC that is a high-temperature combustion residue containing some morphological features of the burned material, such as Spheroidal Carbonaceous Particles (SCPs). 8 Generally, however, char-BC particles are less condensed forms of BC formed at lower combustion temperatures. As char-BC has lower thermal-oxidative stability it is mostly not quantified with the CTO-method. 9 Thus, SBC measurements represent a subset of total BC. Importantly, natural gas flaring (as well as for instance vehicle engines) produces BC only as a gas condensate in flames, and thus the CTO-375 method is well-applicable for our purposes to study the historical trend of flaring-derived BC, as well as all most refractory and smaller sized far travelling BC particles.
For the CTO-375 analyses and SBC quantification as in Refs 10-12 the dried sediments were homogenized to less than 100 µm particle size with a stainless steel ball grinder (Retsch, Mixer Mill 4000). To oxidize organic material, approximately 10 mg of sediment was precisely weighed into precombusted (12 h at 450 °C) silver capsules (5 × 8 mm), and combusted in a custom-made tube furnace at 375 °C for 18 hours under active airflow (200-300 mL min-1) at Stockholm University. Subsequently, carbonates were removed by in-situ microscale acidification with 1M HCl. Finally, the residual carbon, i.e. SBC, and nitrogen (N) concentrations were quantified with a Flash 2000 Organic Elemental Analyser (Thermo Fisher) at 950 °C at the University of Helsinki, Laboratory of Chronology. The measurements were calibrated, and the reliability and repeatability of the analysis determined, by including 5-6 standardized soil reference standard samples of known carbon and nitrogen content supplied by the device manufacturer in each analytical run of 32 samples.
To evaluate the potential risk of charring to the samples, the non-soot organic carbon (NS-OC) content of the samples was determined. Approximately 10 mg of each sediment sample was weighed into silver capsules and acidified with 1 M HCl, as above in the SBC analyses. The total organic carbon (TOC) content was quantified with the elemental analyzer and the NS-OC content of the samples determined by subtracting the amount of SBC from the TOC values. A clear positive correlation between NS-OC and SBC values could potentially indicate charring in the samples, as non-BC condensed organic matter could be charred and measured as SBC during the analysis. However, the CTO-375 method has been demonstrated to have a lower potential to give false positives compared to other BC quantification methods. 11,13 Also here, the SBC concentrations show no correlation with respective NS-OC concentrations, which strongly indicates that the SBC quantification was not affected by charring (Fig.  S1).

S.4. Evaluation of CTO-375 method accuracy and precision
The accuracy of the CTO-375 method can be evaluated by comparing SBC results of selected standard reference material reported in previous studies using the same or similar methods. Wood char, melanoidin, Baltimore Harbor sediment (NIST SRM-1941b), urban dust (NIST SRM-1649b), and diesel particulate matter (NIST SRM-2975) samples were prepared along with the lake sediment samples. These reference materials were chosen as they include 1) BC-containing environmental matrices (Baltimore Harbor sediment, urban dust aerosol and diesel particulate matter), 2) laboratoryproduced BC-rich materials (wood charcoal), and 3) potentially interfering materials originally devoid of BC but potentially creating BC during analysis (melanoidin). It is generally advised in the geochemical community to assess the BC analysis method accuracy and to specify the type of BC quantified by the analysis of these three reference material types in each study. This facilitates comparative analyses of BC in diverse environmental matrices and by different analytical methodologies. The wood char and melanoidin reference material were provided by Prof. Michael Schmidt, Univ. Zürich (https://www.geo.uzh.ch/en/units/2b/Services/BC-material.html) and the rest by the U.S. National Institute of Standards and Technology (NIST, Gaithersburg, MD, USA).
Our measurements and previously reported results of SBC and TOC concentrations of the five standard reference material samples are presented in Table S6. The SBC results of all standard materials are within the standard deviations of published mean data, and mostly have smaller relative standard deviations than previous studies. No SBC was detected in wood char that contains high amounts of organic carbon but no SBC, indicating that oxygen supply during the oxidation phase of the method was sufficient to avoid charring. Samples of the sediment reference material (SRM-1941b) were included in each lake sediment CTO-375 and elemental analyzer run to check for possible errors during the chemothermal oxidation or in the elemental analysis. The method precision was ascertained by analyzing randomly selected sediment samples in duplicates and triplicates. The average relative standard deviation was 3.7 % (range 0.06-9.5 %) for 15 different duplicate and triplicate SBC samples prepared from the lake sediments. Additionally, the relative standard deviation of SBC in the Baltimore Harbor sediment (SRM-1941b) was 3.9 % in 22 samples. Thus, the method precision can be estimated to be around 3.8 %. The method precision here is thereby better than previously reported values of up to 10 % 14 and 23 % 15 for SBC samples, but slightly lower than previously using the same instrumentation (1.7 %) 12 . Where duplicate to triplicate results are available for the SBC results, the mean concentrations are reported and error bars shown in Fig. S1. Several blank samples were prepared to check for contamination or crosscontamination of the samples during the sample preparation and elemental analysis. All had carbon contents below the detection limit (0.01 %, given by the device manufacturer). Figure S1 presents the SBC and NS-OC concentrations (g gDw -1 ) of the lake sediment samples. SBC represents a minor component of sediment and its concentrations may be strongly diluted or concentrated by the varying influx of other sediment components such as organic and mineral material. Consequently, SBC concentrations may vary substantially with time and from one lake to another, and their significance as such is minor for this study. However, the charring potential of the samples was assessed by testing the correlation between the SBC and NSOC contents of the samples. Based on the correlations none of the sediment cores suggest potential charring. S10 Figure S1. Non-soot organic carbon (NS-OC) and soot black carbon (SBC) concentrations (mg gDw -1 ) and their correlation. Note that the axes are not the same for all lakes. Samples of duplicate and triplicate analyses are shown as red asterisks and their average as an open red circle. S11

S.6. Detailed description on the gas-ion-source instrumentation in the radiocarbon isotope analysis at the Accelerator Laboratory of the University of Helsinki
The gas injection system of the Helsinki Accelerator Mass Spectrometry hybrid ion source was built to handle gas samples from 12 storage containers. It uses a pneumatically controlled syringe to inject the CO 2 , and helium as carrier gas to carry the CO 2 , through the capillary pipes. The CO 2 samples are moved from the sample intake, and between storage containers and the syringe, by cryogenic traps with liquid nitrogen. The efficiency of sample transportation done with this freezing -warming process is almost 100 %. Gas pre-pressures for He and CO 2 were set to 3 bar. With these pressures the carbon current from the ion source is 12 µA. Titanium cathodes were used and they were pre-sputtered for five minutes to remove carbon contaminants from the cathode surface. A similar gas-ion-source system for radiocarbon measurements has been presented in previous work. 16 Subsequently, the 16 sediment samples selected from the Vork 9 core were measured for their radiocarbon content with Accelerator Mass Spectrometry (AMS).

S.7. 14 C analyses with traditional AMS methodology
In addition to the 16 Vork 9 samples analyzed with the gas-ion-source, 6 parallel samples from the same sediment core were prepared by traditional graphitization for the 14 C analyses. For this, nine sediment subsamples of ca. 11 mg size were prepared with the CTO-375 method and pooled together into one CO 2 trap, and subsequently graphitized by the Helsinki Adaptive Sample preparation line, 17 and the radiocarbon content determined with AMS.
This was done to compare the results of the newly deployed hybrid ion source instrumentation with the more traditional 14 C analyses including the graphitization step (for results, see S.9.). The gas-ion-source step in the AMS analyses has several advantages compared to the traditional measurements using graphitization as a sample preparation step, as a) the required sample size for the gas-ion-source measurements is roughly one tenth of the traditional method and thereby extremely valuable for e.g. Arctic samples devoid of high BC concentrations, and b) because one sample preparation step potentially causing artificial isotopic fractionation (graphitization) is eliminated with the gas-ion-source in comparison to the old methodology. Samples are also easier and faster to prepare and the AMS measurements are substantially faster with the gas-ion-source compared to the traditional methodology. S12

S.8. Calculation of the bio-based carbon percentage of the extracted SBC, i.e. radiocarbon source apportionment
The radiocarbon source apportionment of the 16 selected Vork 9 sediment core samples was based on first determining the radiocarbon content (given as pMC, i.e. percentage modern carbon 18 ) of the samples by AMS and then calculating the percentage of the bio-based carbon (bio % SBC) ( Table S7).
The bio-based carbon percentage calculations of the extracted SBC were performed as described in the standard procedure ASTM D6866-20 19 by dividing the quantified pMC by an atmospheric reference pMC value and multiplying the result by 100.

% = 100 ×
The atmospheric radiocarbon concentration of the year 1950 is the reference year, to which all radiocarbon measurements are compared. This content is defined as 100 pMC (percent modern carbon, or 100 % of the radiocarbon content compared to 1950's level). Fossil material on the other hand has a value of 0 pMC, as it is completely devoid of radiocarbon. However, nuclear testing starting in the 1950s increased globally the naturally occurring atmospheric radiocarbon content which peaked in the 1960s and has subsequently decreased to 1950s values again. All biomass growing during the nuclear bomb testing period incorporated increased radiocarbon levels, depending on their growing period.
Here, the naturally varying atmospheric level of radiocarbon was compensated by applying an atmospheric correction factor. As the biomass component of the samples is expected to mainly consist of wood, and the turnover time of similar northern environment biomass has been estimated to be approximately 20 years in previous studies, 20,21 a model for average biomass radiocarbon content was created. For the model, the running averages of known atmospheric radiocarbon level of the previous 20 years were calculated based on the data of IntCal13 22 and Ref 23 , and the radiocarbon level corresponding to the year which each specific sediment sample approximately represented, was applied as an atmospheric correction factor. Thus, the bio % SBC results were calculated as the biomassderived carbon mass percentages of the total carbon of the SBC samples. Potential error sources affecting the calculations are those inherent to the pMC measurements, the assumption of the turnover time of biomass (ca. 20 years), and sediment dating uncertainties.
A major advantage of the radiocarbon-based source attribution is the analysis on the deposited BC particle itself, while previous source attribution has used, for instance, chemical speciation of associated major ions and metal particles in snow 23-26 that may behave differently than BC during longrange transportation to the Arctic. 27 S13  SBC source apportionment results obtained with these methods generally agree well, and the higher resolution gas-ion-source samples clearly support and expand on the pMC trend also observed in the fewer graphitized samples (Fig. 4 and Table S7). S14 Figure S2. The biomass-derived SBC content of the Vork 9 sediment samples prepared for the AMS analysis by the gas-ion source step (blue), and the traditional procedure including graphitization (red).

S.10. Evaluation of FLEXPART model uncertainties
The Lagrangian particle dispersion model FLEXPART is widely used around the world, 28 and has been shown to capture well both BC transport to the Arctic, 29-32 as well as Arctic atmospheric BC concentrations and their seasonality. [33][34][35] When comparing BC observations with FLEXPART model outputs it seems that discrepancies can be reduced significantly by changing the emission input data. 32,34 In particular, in northern Asia, uncertainties in the available emission inventories are large, and the available bottom-up emission inventories differ substantially both in magnitude and spatial disaggregation. 32 Thus, a substantial part of the uncertainties of the FLEXPART model outputs relate to the used emissions and their uncertainties.
The model uncertainties, other than the emission inventory uncertainties, consist mainly of two uncertainties: 1) uncertainties related to the meteorological input data (e.g., errors in winds, precipitation, etc.) and 2) errors in transport and removal within FLEXPART.
Errors of the meteorological input data could be assessed, at least to some extent, by using different meteorological input data sets. This could be done, for instance, by running an ensemble of model simulations based on different meteorological input data. However, to our knowledge no ensemble of meteorological analyses is available for the long time period (1900-1999) covered by our study.
Meteorological ensembles are normally only available for operational analyses (the most recent but shorter ERA5 reanalysis also comes with a mini-ensemble), and even then it is not clear whether these ensembles can quantitatively describe the uncertainty in the transport calculations. It is beyond the S15 scope of the current study to perform ensemble model simulations to test the influence of the meteorological input data. However, it is likely that these uncertainties have a larger influence on the BC deposition variability on synoptic time scales than on deposition rates averaged over longer time periods since systematic biases in the meteorological model (e.g., of precipitation rates) are not contained in the ensemble uncertainty.
Errors in transport and removal within FLEXPART can be assessed by varying critical parameters in FLEXPART, especially related to dry and wet deposition. For instance, a relative model-observation mismatch of 32 %-43% for a three year study period at high latitudes, created by perturbation of scavenging coefficients for BC in the simulated concentrations, has been shown. 32 As far as we know this is the only study that has evaluated uncertainties of the model itself, in particular scavenging efficiencies. Such studies are valuable to assess uncertainties of FLEXPART. However, these uncertainty estimates were made only for atmospheric BC concentrations 32 , and may not be directly applicable to BC deposition.
Model validation exercises require comparison with observational data. Unfortunately, BC deposition observations are currently scarce in the Arctic. To our knowledge, only one study 36 has quantified BC deposition from falling snow in the Arctic (Svalbard) for one winter season. For instance, ice core records or the current SBC flux data are not optimal for exact model validation as ice core and sediment BC deposition fluxes are affected by post-depositional processes and are not representative of only atmospherically deposited BC (although ice cores much better than sediments) and may have, for instance, dating problems. Thus, also the evaluation of uncertainties inherent for the FLEXPART model is beyond the scope of this study.
Importantly, the above listed uncertainties of the FLEXPART model and the input data mostly affect the exact amount of simulated BC deposition. Thus, as mentioned in the main text, we do not compare the exact values of modelled BC deposition and observed SBC fluxes. The listed uncertainties most likely do not affect the modelled BC deposition trend, and especially not in the yearly to decadal timescale which is relevant for our study. Thus, the potential model uncertainties should not affect the outcomes or discussion of the paper.   39 BC emissions (kt yr -1 ) from 1850 to 2015 by defined areas used in this study. Emission data from CMIP6. Note, that since 1997 the wildfire emissions are derived from satellite data instead of fire proxies (such as charcoal records) and thus the yearly variations are higher than for 1850-1996 (for more details see the original study 39 ). "Other" means the rest of the world.

S.13. Comparison of observed and modelled sources for the SBC and BC deposited at the study sites
The emissions used for the FLEXPART modelling included wildfire and anthropogenic BC emissions as shown above in Fig. S5. Optimally, these two emission groups of the emission inventory would be compared to the observed SBC radiocarbon-based biomass and fossil fuel-derived source classes. Unfortunately, such comparison is not feasible as they do not represent same emissions. For instance, S18 the radiocarbon-based biomass sources contain both emissions from wildfires and fuel woods, while fuel woods are classified as anthropogenic emissions in the emission inventory. Furthermore, peat combustion produces, depending on the age of the peat, a varying mixture of biomass and fossil fuelderived emissions that are separated by the radiocarbon measurements, but are grouped as wildfire emissions in total in the used CMIP6 emission inventory. The anthropogenic emission sources are not further divided into sub-categories (such as residential, industry, transportation etc.) for 1900 to 1999 in the CMIP6 emissions. Thus, it is impossible to directly compare the source attributions made by the model and produced by the observations. Figure S6. Observed SBC flux radiocarbon sources compared to modelled BC deposition sources to Vork 9. A) The temporal trend of SBC fluxes derived from biomass and fossil fuel sources. B) The contribution (%) of biomass vs. fossil fuel-derived sources to total SBC based on radiocarbon measurements of SBC. C) The temporal trend of modelled wildfire vs. anthropogenic sources to the BC deposited to Vork 9. D) The contribution (%) of wildfire vs. anthropogenic sources to total BC deposited at Vork 9. S19 Figure S6 shows that according to the radiocarbon measurements ca. 70 % of the SBC fluxes in the Vork 9 sediment originated from biomass combustion between ca. 1850 and 2014, while the modeling results suggest that between 1900 and 1999 only ca. 10 % of BC deposited at the site originated from wildfires. As mentioned, these source categories contain partly different sources.
Comparison of the observed and modelled results may suggest that the biomass-derived SBC contains to a majority SBC produced by combustion of fuel wood and other modern biomass. Indeed, BC emission from fuel wood combustion contributed around 61 % of the total residential BC emissions. 37 Thus, between 1850 to 2000 fuel wood combustion may constitute the bulk of biomass-derived SBC in our sediment record, as the domestic (i.e. residential) sector has caused by far the highest anthropogenic BC emissions from Russia, 38 particularly between 1850 to 1970 (ca. 80 %, Fig. S4b).
Consequently, it seems that wildfires may have had only a minor contribution to SBC fluxes and their temporal trend in the Vork 9 sediment record. However, we do not have any information on the temporal development of fuel wood use in Russia from 1850 to 2000 and thereby we cannot verify this hypothesis. As the currently best available wildfire database 39 does not suggest an increase in Russian wildfire BC emissions during the last few decades, we assume that the observed increase of biomassderived SBC fluxes in the sediment record between ca. 1980 and 2014 (Fig. S6) is mainly caused by a sharp drop in coal production and consumption and a potential switch to fuel wood in residential use in the study area in this time period. However, we also cannot preclude a possible increase in wildfire emissions in recent decades missed by emission inventories. Thus, our radiocarbon source apportionment data may suggest that BC emissions from wildfires are potentially underestimated in emission inventories, particularly since 1990. After 1990 our radiocarbon data show increased contribution of biomass-derived BC to SBC, but at the same time residential combustion has lost its dominance as the anthropogenic BC emitter (Fig. S4b) and thus the observed biomass-derived SBC is not likely only of residential origin. S20 Figure S7. Standardized SBC fluxes in four Northern Finland lake sediments compared to elemental carbon (EC) deposition in a Svalbard ice core. 12