Identification of Missing Anthropogenic Emission Sources in Russia: Implication for Modeling Arctic Haze

Any comprehensive simulation of air pollution in the Arctic requires an accurate emission inventory. Using a community global emission inventory EDGAR v4.2 (Emissions Database for Global Atmospheric Research), GEOS-Chem modeling underestimated aerosol optical depth by 150–300% when compared to ground-based sites in Russia. Emissions from power plants, gas flaring, and mining were found significantly underestimated or even missing in EDGAR’s Russian emission inventory. Approximately 70% of Russian provinces had lower NOx and PM10 emission from power plants in EDGAR as compared to a Russian federal emission inventory. Emissions from gas flaring dominated in Russia’s main oil and gas producing regions. However, it is completely missing in EDGAR. In addition, EDGAR underestimated Russia’s mining emissions in most of its remote areas. Overall, we find EDGAR underestimated Russia’s emissions especially at high latitudes and this could overlook the impact of Russian emissions on the Arctic if EDGAR is used as input for models.


INTRODUCTION
The Arctic region is vulnerable to the transport and deposition of particulate matter (PM), such as black carbon, and sulfate. The Arctic Circle (north of 66°33′44′′N) includes parts of Alaska, Europe and vast regions of Canada and Russia. Anthropogenic emissions and biomass burning originating from these countries have been shown to be the main cause of Arctic haze (Law and Stohl, 2007). Most of the Arctic region countries have relatively reliable emission inventories except Russia due to the difficulties of quantifying the local emission factors and locating emission sources. Results from trajectory models (e.g., FLEXPART (Hirdman et al., 2010a, b), WRF (Harrigan et al., 2011), HYSPLIT (Huang et al., 2010), PSCF (Eleftheriadis et al., 2009), and Canadian Meteorological Centre model (Sharma et al., 2006)), concluded that northern and central Russia was the major source region contributing to Arctic haze. The former USSR contributed to the haze measured at a Canadian high † Oak Ridge Institute for Science and Education Fellow * Corresponding author.
Tel.: 1-865-974-2503; Fax: 1-865-974-2669 E-mail address: jsfu@utk.edu Arctic site with the dominant fraction of 67% during a 16year period, followed by the European Union (18%) and North America (15%) (Huang et al., 2010). Meanwhile, the contribution from Asia or Southeast Asia was negligible (Stohl, 2006;Hirdman et al., 2010b). Recent 3-D chemical transport modeling efforts, however, have shown contrasting and inconsistent results. For instance, Asian anthropogenic emissions were suggested to be the dominant source of Arctic CO pollution by using GEOS-Chem (Fisher et al., 2010). A GISS model (ModelE) study also suggested that south Asia (industrial and biofuel emissions) and biomass burning were the predominant sources of Arctic soot (Koch and Hansen, 2005). A multimodel (17 models) research effort determined that the European emissions dominated at the surface of Arctic but East Asian emissions were more dominant in the upper troposphere (Shindell et al., 2008). Models generally underpredicted black carbon concentrations in the Arctic (Koch and Hansen, 2005;Shindell et al., 2008), and the largest divergence in model results occurred in northern Eurasia and the remote Arctic from the AeroCom model inter-comparison (Koch et al., 2009). For the latter study, anthropogenic emissions from Russia and Asia had to be doubled to match with the observations (Wang et al., 2011). Recently, Stohl et al. (2013) used a Lagrangian particle dispersion model FLEXPART to greatly improve the simulated black carbon over the Arctic by using daily-varying residential combustion emissions and introducing a global gas flaring emission inventory. Lacking of certain emission sources and improper treatment of emission temporal profiles are pointed out against previous studies of ascribing unsatisfactory model performances to physical process problems in aerosol models.
In this study, we aim to identify several underestimated and missing emission sources in the Russian part of the global emission inventory EDGAR, with a specific focus on the energy, gas flaring and mining sectors. Model simulation using EDGAR was evaluated against observations in Russia. Large point sources (i.e., thermal power plants) in EDGAR were compared to a global power plant database CARMA and satellite detection of NO 2 columns. Gas flaring areas were retrieved from satellite imagery and the emissions from gas flaring were estimated. This study first demonstrated the differences of multiple emission sectors between EDGAR and a Russian federal emission inventory. It should be noted that this study doesn't aim to demonstrate how the identified gaps between various emission databases will translate into modeling results, but draw conclusions on the importance of improving the Russian emission inventory on modeling the origin and impact of the Arctic haze.

METHODS
EDGAR was first used to determine whether using this emission dataset as input to transport modeling would represent measured concentrations of pollutants. Then, other data sets, including a database of power plants, an emission inventory from the Russian federal government, and data on gas flaring were utilized to verify possible underestimated or missing emission sectors investigated in this study.

Emission Data EDGAR Global Emission Inventory
EDGAR is a global database for anthropogenic emissions of greenhouse gases and air pollutants with a spatial resolution as fine as 0.1° × 0.1°. Sectoral emissions are available, including energy production, transportation, industries, residential, and biomass burning. Biomass burning emissions are based on the Global Fire Emissions Database (GFED). It is generated on the monthly basis (van der Werf et al., 2010) with temporal scaling profiles of daily and 3-hourly . Conversions from carbon emissions to various species are based on emission factors from Andreae and Merlet (2001). The methodology for the EDGAR emission calculations is well established (EDGAR, 2013). It has been used as a default emission inventory for various models, e.g., GEOS-Chem, MOZART, the unified EMEP model, ECHAM5-HAMMOZ, GISS-PUCCINI, etc. We used the newest version of EDGAR, v4.2, and the most recent year available, which is 2008. However, due to the lack of up-to-date local activity data, emissions estimates in recent years are not necessarily reliable for Russia. For instance, compared to EDGAR v4.1, the corrections of power plant emissions were only made for sources in China (EDGAR, 2011). In this study, we make a distinction between the global EDGAR database, which we use as input for the global chemical transport model GEOS-CHEM and the Russian part of the EDGAR database, which we compare to various other databases (hereinafter called "RUS_EDGAR").

Russian Federal Emission Inventory
The Russian Federal State Statistics Service (FSSS, http://www.gks.ru) provides its national emissions inventory (hereinafter called "RUS_FSSS") of air pollutants for a limited range of emission sectors. RUS_FSSS inventory data is available for fossil-fuel fired power plants and mining, but not gas flaring. We compared 2008 data from RUS_FSSS to be consistent with RUS_EDGAR. The pollutants reported by FSSS, include solid particles, carbon monoxide, nitrogen oxides, and hydrocarbons. The methodologies for estimating the pollutant emissions were established by various Russian research institutes. A list of the approved emission calculation methodologies currently in use by the Russian Federation are documented by SRI-Atmosphere (SRI Atmosphere, 2012). The emission of a specific air pollutant into the atmosphere is estimated by using the following equation: where i represents a specific economic sector; M raw,i is the annual total raw emission of sector i prior to technology controls; η i is the removal efficiency of sector i; and M i is the annual total emission released into the atmosphere. To convert the RUS_FSSS total suspended particulate (TSP) emission data to the more commonly used metric of particulate matter with a diameter of 10 µm or less (PM 10 ), we multiplied the original RUS_FSSS TSP data by a scale factor of 0.675, which was calculated based on an average over multiple estimates of emission factors for PM 10 and TSP from coal combustion in power plants (EEA, 2013). Table 1 shows these calculated nationwide emissions of PM 10 , and reported NO x and CO, for the power and mining sectors in the Russian Federation, from FSSS for 2008. In order to allocate the nationwide emissions to the provincial level, we used the same relative provincial distribution of the Russian category entitled "social-economic indicators" (FSSS, 2011) to allocate national emissions by province. This is equivalent to: where k represents a specific Russian province; E i,k is the annual revenues rendered in sector i in province k (unit: million rubles); And M i,k is the annual emission in sector i in province k.

CARMA -A Global Power Plant Dataset
Carbon Monitoring for Action (CARMA) is a database under the operation of the Confronting Climate Change Initiative at the Center for Global Development (http://www. carma.org). It contains information about the energy production, intensity, carbon emission and locations of the power plants worldwide. CARMA compiles data from both national publicly disclosed databases and a commercial database of the world's power plants (Wheeler and Ummel,

Gas Flaring Dataset
National and global gas flaring volumes are estimated based on satellite sensor observations from the U.S. Air Force Defense Meteorological Satellite Program (DMSP) Operational Linescan System (OLS). NOAA NGDC (National Geophysical Data Center) serves as the long-term archive for DMSP (http://www.ngdc.noaa.gov/dmsp/interest/ gas_flares.html). Gas flaring activity is detected from the visible band signal at night and identified based on a parameter called "lights index". More detailed descriptions of the methodology can be found in (Elvidge et al., 2009).

GEOS-Chem Simulation and Model Evaluation
A 3-D global chemistry model GEOS-Chem (v8-02-03) was used to evaluate the reliability of RUS_EDGAR in this study. The model is driven by GEOS-5 (Goddard Earth Observing System) assimilated meteorological inputs from the NASA Global Modeling and Assimilation Office (GMAO) for 2008. The model's resolution is set at 2 by 2.5 degrees with 47 vertical layers. The global anthropogenic emission input to the model is from EDGAR, superseded by regional emission inventories, including the NASA INTEX-B inventory for South and East Asia, the EMEP inventory for Europe, the BRAVO inventory for Mexico, and the USEPA's NEI for the US. The emission inventory for the rest of the world is from EDGAR v4.2.
The model performances of GEOS-Chem have been intensively evaluated in the United States (Fu et al., 2011;Zhang et al., 2012) and East Asia (Lin, 2012), suggesting its applicability for simulating global air pollution transport and transformation given reliable emission inventory. Currently, limited model simulations have been conducted for Russia (Makarova et al., 2011). To evaluate the GEOS-Chem model performance in Russia, AERONET (Aerosol Robotic NETwork, (Holben et al., 1998)) observations were used as they are the only available publicly accessed aerometric network in Russia with assured quality. In 2008, data from five AERONET sites were available. These sites in Russia are shown in Fig. 1, including Moscow, Yekaterinburg, Tomsk, Irkutsk, and Yakutsk. Aerosol optical depth (AOD) at the visible wavelength of 500nm was chosen as the parameter for the model evaluation. The method of converting modeled aerosol chemical species to AOD is described in Huang et al., 2012).

OMI NO 2 Column Concentrations
In this study, the NO 2 columns retrieved from the Ozone Monitoring Instrument (OMI) is used to verify the locations of large point sources. The capability of detecting large point emission sources by using the OMI satellite retrieves have been well demonstrated (Zhang et al., 2009;Li et al., 2010). OMI is a nadir-viewing near-UV/Visible spectrometer aboard NASA's Earth Observing System's (EOS) Aura satellite. It can measure the sunlight at a spectral region of 264-504 nm with a spectral resolution between 0.42 nm and 0.63 nm and a nominal ground footprint of 13 × 24 km 2 at nadir. In this study, the NO 2 column concentrations are from the OMI Level 3 daily global products with a spatial resolution of 0.25° × 0.25°. Spring data are used as higher pollutant concentrations occur during the cold season compared to the warm season. In addition, the satellite had very limited coverage over the high latitudinal regions of Russia from late autumn to winter due to the low solar zenith angles.

Underestimated Russian Emission from Perspective of Modeling
The comparisons between GEOS-Chem modeled AOD by utilizing EDGAR and the observed AOD at the five Russian AERONET sites are shown in Fig. 2. The left panels show the temporal variations (shown in Julian days) and the right panels represent the scatter plots between observation and simulation. Significant discrepancies between the observed and modeled AOD are evident. As shown in the left panels of Fig. 2, the model predicted relatively flat temporal variations and missed almost all the peaks. Modeled and observed AOD differed by a factor of 5-10 during intensive pollution episodes. Observed AOD peaked at different times depending on the site locations, possibly due to the variations of local emission intensities and local meteorology. Biomass burning was investigated to be insignificant for the high AOD events by conducting sensitivity simulation with zeroing out emissions from biomass burning. Lowest biomass burning emissions in this study year (2008) during the 2000s should be responsible for this (Fig. S1). Overall, neither peak episodes or the spatial differences of AOD could be reproduced by the model at all. Fig. 2 (right panels) showed evidence that AOD at all sites in Russia were significantly underestimated by the model. AOD were biased low by about 2-3 folds for Moscow and Yekaterinburg, two of the biggest cities in Russia. In some smaller cities (e.g., Tomsk and Irkutsk) and remote areas (e.g. Yakutsk), AOD were underestimated by about 1.5-2 folds. As stated earlier, GEOS-Chem performed relatively well in various regions, where emissions are relatively reliable. The unsatisfactory model performance in Russia strongly suggested that Russia's emissions in EDGAR (RUS_EDGAR) needs substantial improvement.

Emissions from Fossil-Fuel Fired Power Plants Detection of Missing Power Plants in RUS_EDGAR from CARMA and OMI
In this section, we compared CARMA and RUS_EDGAR to locate possible regions of Russia where the two databases differ in their co-locations of fossil-fuel fired power plants. The locations of fossil-fuel fired power plants in RUS_EDGAR (black squares) and CARMA (pink dots) for 2007 are shown in Fig. 3(a). In most areas, pink dots were surrounded by the squares, indicating that a power plant was identified in both databases. However, we found that CARMA contained more sites than RUS_EDGAR in some regions, e.g., the two sub-regions highlighted in Figs. 3(b) and 3(c).
To further evaluate the reliability of these "additional" fossil-fuel fired power plants from CARMA, the spatial distribution of NO 2 column concentrations observed from OMI during the spring of 2007 is overlaid in Fig. 3. As shown in Fig. 3(a), the areas where there were co-located fossil-fuel fired power plants from both RUS_EDGAR and CARMA generally showed high NO 2 concentrations. Emissions from electricity and heat production contributed 56% to the total NO x emissions in Russia according to RUS_FSSS. Specifically, we found that areas where there were additional fossil-fuel fired power plants in CARMA compared to RUS_EDGAR also showed high NO 2 column concentrations, further suggesting that there were indeed fossil-fuel fired power plants missing in EDGAR.
Two sub-regions with apparent missing power plants were selected for discussion as marked by the red rectangles in Fig. 3(a). Sub-regions 1 (Fig. 3(b)) refers to the Urals Federal District (which contains the Khanty-Mansiysk and Yamalo-Nenets Autonomous Okrugs) and Sub-regions 2 (Fig. 3(c)) refers to Chukotka Autonomous Okrug. The high NO 2 columns are likely not related to the regional/longrange transport, as the adjacent areas around these two regions were accompanied with relatively low NO 2 concentrations, hence suggesting local emission sources. Also, it is unlikely that residential emissions contributed significantly to the high NO 2 concentrations there, due to the fairly low population densities (fewer than 3 persons/km 2 ) in these areas as shown in Fig. 1.  Fig. 3(b) shows that a total of 16 fossil-fuel fired power plants (some power plants were closely located and couldn't be differentiated clearly in the figure) were missing in EDGAR over Sub-region 1. High NO 2 concentrations were observed around the areas where power plants were located, especially around co-located power plants. For instance, relatively high NO 2 column concentration of about 5.0-5.5 × 10 15 molecules/cm 2 were observed around the P11-P15 power plants (Fig. S2 and Table S1). In the Khanty-Mansiysk Autonomous Okrug of the Urals Federal District (Fig. 3(b)), it was noted that one large capacity power plant (P13: SURGUT-2) was missing in RUS_EDGAR with its annual intensity of 1.56 × 10 7 MWh based on CARMA. It ranked as the fifth largest fossil-fuel fired power plant in Russia. Also, its fuel type was investigated to be fuel oil/diesel, which had high emission factors for pollutant gaseous (e.g., NO x and CO) and particulate matter. The total missing energy production in the Khanty-Mansiysk and Yamalo-Nenets Autonomous Okrugs reached 2.35 × 10 7 and 2.23 × 10 6 MWh based on CARMA, respectively, which could contribute significantly to the air pollutants emission. It is noted that Sub-region 1, specifically the Yamal-Nenets Autonomous Area, shows widely dispersed high NO 2 spots besides at locations near the missing fossil-fuel fired power plants. As discussed in Section 3.3, this is a region of significant gas flaring emissions. This explained the large scale high NO 2 zone in Sub-region 1.
Sub-region 2 is in the northeast part of the Chukotka Autonomous Okrug (Fig. 3(c)). Only one fossil-fuel fired power plant appeared in RUS_EDGAR, while there were five more power plants (P17-P21, Fig. S2) found in CARMA. The spatial distribution of NO 2 columns in Fig. 3(c) verified the existence of these plants. The total energy production of missing power plants in Sub-region 2 reached 5.19 × 10 5 MWh based on CARMA, which was about 400% higher than that in RUS_EDGAR. In addition to these two regions, there was also considerable absence of power plants distributed in other regions. Fig. S2 and Table S1 illustrate the locations and information of individual missing power plants. The total underestimated energy production in EDGAR as compared to CARMA reached 6.18 × 10 7 MWh (42% from coal-fired power plants), accounting for 9.6% of the total energy production in Russia. In other words, RUS_EDGAR underrepresented about 10% lower energy input, which could be significant when translating the country's energy inputs into emissions.

Comparison of RUS_EDGAR to RUS_FSSS
We further compared RUS_EDGAR and RUS_FSSS at the provincial level. Fig. 4 shows the scatter plot between the two emission inventories for NO x and PM 10 emissions from fossil-fuel fired power plants. Each scatter represents one Russian province. For NO x emissions ( Fig. 4(a)), approximate one third of provinces were in relatively good agreement between the two datasets as indicated by the dots within the 1:2 and 2:1 lines. However, approximate two-thirds of provinces fell above the 2:1 line, indicating significant underestimation of RUS_EDGAR compared to RUS_FSSS. Khanty-Mansiysk Autonomous Okrug was the province that we found to be the most underestimated in RUS_EDGAR. Its NO x emissions from fossil-fuel fired power plants in RUS_FSSS reached 140.3 Gg, which ranked the third highest of any province within the Russian Federation. However, this source area was not even registered in RUS_EDGAR. In addition, NO x emissions from power plants in the Yamalo-Nenets and Chukotka Autonomous Okrugs were also lower in RUS_EDGAR compared to RUS_FSSS by a factor of ~30. This finding corroborated the results from the OMI observation that hot spots of NO 2 columns occurred in areas where many fossil-fuel fired power plants were missing in RUS_EDGAR. Fig. 4(b) shows the comparison for PM 10 emission from fossil-fuel fired power plants between the two emission inventories. PM 10 emissions were even more underestimated than for NO x . The total national PM 10 emission from fossilfuel fired power plants was approximately 830 Gg in RUS_FSSS in 2008 (Table 1), about 2 times higher than that in RUS_EDGAR. The regional differences between the two databases in the distribution of PM 10 are similar to NO x . In Section 3.1, we show that GEOS-Chem simulations, using EDGAR, significantly underestimated AOD for the five AOD measurement sites in Russia. The missing PM 10 and NO x emissions identified here likely explain part of the observed AOD underestimation.

Emissions from Gas Flaring
Gas flaring is a widely used practice for the disposal of associated gas in oil production and processing facilities where there is insufficient infrastructure for utilization of the gas. Flaring causes hazards to human health and also contributes to global anthropogenic emissions (McEwen and Johnson, 2012). Russia possesses the largest gas flaring volume in the world. In 2008, the amount of gas flaring from Russia reached about 42 BCM (billion cubic meters) and contributed 29% of total gas flared worldwide (World Bank, 2012). Although included in the EDGAR inventory, gas flaring emissions were estimated to be almost zero for Russia.  Fig. 5) were the other four largest regions that contributed to Russia's national flaring emission. Relatively low APG (Associated Petroleum Gas) utilization rates were the major reason for the large gas flaring emissions in these regions. For instance, the oil and gas fields in the Urals and Western Siberia (e.g., Khanty-Mansiysk Autonomous Okrug, Yamalo-Nenets Autonomous Okrug and Tomsk Oblast) only had a moderate APG utilization rate of 55-78% (FNI, 2010), and the oil and gas fields in the Northwestern Federal District (e.g., Komi Republic, Nenets Autonomous Okrug) had a APG utilization rate of only slightly above 35% (Knizhnikov and Poussenkova, 2009). Table 3 shows the results of the 2008 emissions for PM 10 , NO x , and CO in the five major flaring regions and the whole Russian Federation based on the annual gas flaring volumes and emission factors. To estimate the emissions from flaring, we applied the mean emission factors (EFs) for PM 10 , NO x , and CO as listed in Table 2. The EFs in Table 2 have wide ranges and high uncertainties due to differences in fuel type, heating values, combustion efficiency, etc. (RTI International, 2011). We estimate the Russian national PM 10 , NO x , and CO emissions from gas flaring to be 132, 126, and 601 Gg, respectively. Compared to Russia's emissions from the energy industry at a national scale, the magnitudes of flaring emission were a factor of 8 and 26 lower for PM 10 and NO x , respectively, because flaring occurred in a relatively limited geographic region in Russia (Fig. 5). National CO emissions from flaring were close to that from power plants, mainly due to incomplete combustion.
For the regions where flaring occurred, its emission dominated over other sources. For example, gas flaring emissions of PM 10 , NO x , and CO from Khanty-Mansiysk were 63, 60, and 286 Gg in 2008 as compared to those of 34, 26, and 140 Gg from the energy industry, respectively. In Yamalo-Nenets, gas flaring emissions were even more dominant with emissions of 27, 25, and 121 Gg for PM 10 , NO x , and CO as compared to those of 8, 6, and 34 Gg from the energy industry, respectively.   Table 3. 2008 gas flaring emission for PM, NO x , and CO in five major flaring regions and the whole Russian Federation. Uncertainties of estimated emission (one standard deviation) are shown in the parentheses.

Russian Regions
Flaring Emission (Gg) Index (Fig. 5 The large NO x emission from flaring should also partly account for the observed NO 2 columns from satellite in the flaring source regions. As shown in Fig. 3(b), high NO 2 columns spread over the central and southern parts of Yamalo-Nenets where less or even no power plants located there. This could be explained by the intense flaring emissions there. Fig. 5 demonstrates that the flaring source region in Yamalo-Nenets corresponded relatively well to the high NO 2 columns there. The Nenets Autonomous Okrug is another example demonstrating the possible impact of flaring emission on the air pollutant levels. As shown in Fig. 3(a), there were two power plants in this region while neither of them located in the high NO 2 column zone. As we compare the flaring source area in Nenets to the spatial distribution of NO 2 , it showed relatively good consistency between them.

Emissions from Mining Activities
In addition to the underestimated emission from power plants and the missing emission from gas flaring as discussed in the previous two sections, emission produced during mining processes could be another potential source that is neglected or underestimated in RUS_EDGAR. Russia is one of the world's leading mineral producing countries. The mineral raw material sector in Russia, which included mineral extraction and processing, produced about 30% of the country's gross domestic product (GDP) (USGS, 2007). The national mining emission from RUS_FSSS reached 275, 151 and 2229 Gg for PM 10 , NO x , and CO, respectively (Table 1). However, in RUS_EDGAR, the mining emissions for the above three species were much lower to be 19, 47, and 379 Gg for PM 10 , NO x , and CO, respectively. Thus, RUS_EDGAR underestimated mining emissions by a factor of ~16, 2, and 4 for PM 10 , NO x , and CO, respectively, as compared to RUS_FSSS.
In order to find out which part of Russia had the biggest discrepancy for mining emissions between the two inventories, we distributed the total Russian mining emission to the provincial level by using the provincial economic revenues from mining and quarrying as a proxy (method described in Section 2.1.2). Fig. 6 shows the ratio of RUS_FSSS vs. RUS_EDGAR in the mining emission sector for PM 10 , NO x , and CO at the Russian provincial level, respectively. The ratios above and below 1.0 are indicated by different colors. It is found that the regions with ratios less than 1.0 (means RUS_FSSS is lower than RUS_EDGAR) mainly distributed in areas with higher human population density (Fig. 1), e.g., the European and southern parts of Russia. While most underestimations of mining emission occurred in the remote areas, e.g., the Urals, Siberia, the Far East, and parts of the Northwestern Federal District. Of which, Khanty-Mansiysk and Nenets were again among the most underestimated regions. A factor of over 1000 lower in RUS_EDGAR was indicated in the figure for the PM 10 and CO mining emission. In some other regions, e.g., Yamalo-Nenets, Sakha, Evenk, and Chukotka, where mining activities were active, their mining emissions were also underestimated significantly.

OUTLOOK
Based on the results presented above, the regions identified with most considerable emissions missing were all located at high latitudes, e.g., Nenets,and Chukotka (Figs. 3 and 5). Fig. S3 ( (Cheng, 2013), unpublished plot from the same project) plots the probabilities of emission source regions by using the Potential Source Contribution Function (PSCF) and measurement data of black carbon at Tiski Bay in 2010. Hot spots were evident in Urals, central part and eastern tip of Russia, corroborating well with our results above. Thus, if by using EDGAR as the model input, the impact from Russian continental anthropogenic emissions on the Arctic could be possibly underestimated. The addition of the significant Russian emissions identified in this study might shift the view of the source apportionment of the Arctic pollution and will be especially important for the impact assessment of long-range transported pollutants on air quality and climate change in the Arctic region. It is important to note that there may be other significantly under-reported emissions, such as industrial, residential and/or transportation emissions. An up-to-date and comprehensive Russian emission inventory requires close collaboration with local authorities. Detailed local information of fuel usage, emission factors, and efficiencies dependent on economic sectors and regions and are necessary. Currently, we are working on the speciation of total PM emissions with spatial/temporal allocation and will further implement the newly constructed Russian emissions into multiple 3-D models. It is expected that the improved Russian emissions will significantly advance the simulation of air pollutants and climatic impacts over the Arctic region. This study and our future works will also provide insights on how emission controls should be targeted to alleviate the climate change over the Arctic.

CONCLUSIONS
In this study, our aim was to elucidate the differences between the Russian part of the global emission inventory EDGAR and a Russian federal emission inventory. From the view of the GEOS-Chem modeling, AOD at multi-AERONET sites in Russia were underestimated by about 150%-300%, indicating a significant underestimation of Russian local emissions. Three emission sectors including fossil-fuel fired power plants, gas flaring and mining emissions were specifically investigated. Considerable fossilfuel fired power plants were found missing in RUS_EDGAR in comparison to CARMA and the spatial pattern of NO 2 columns observed from OMI. Most missing power plants were found in the Urals Federal District and the Chukotka Autonomous Okrug. The total underestimated energy production in EDGAR reached 6.18 × 10 7 MWh, accounting for about 9.6% of the total energy production in Russia. In comparison to a Russian federal emission inventory (RUS_ FSSS), around 70% of the Russian provinces showed lower NO x and PM 10 emission in EDGAR. The total NO x and PM 10 emission in RUS_FSSS reached 3419 and 801 Gg, compared to that of 2267 and 382 Gg in RUS_EDGAR, respectively. Emissions from the Khanty-Mansiysk, Yamalo-Nenets, and Chukotka Autonomous Okrugs had the largest discrepancies between RUS_EDGAR and RUS_FSSS.
Russia's gas flaring combustion was another emission source that was neglected in RUS_EDGAR although Russia is the world's largest gas flaring country. Its national PM, NO x , and CO emission from gas flaring reached 132, 126, and 601 Gg in 2008, respectively. The Urals Federal District and the Northwestern Federal District are Russia's main oil and gas producing bases and also the major gas flaring regions. Khanty-Mansiysk Autonomous Okrug, Yamalo-Nenets Autonomous Okrug, Komi Republic, Nenets Autonomous Okrug, and Tomsk Oblast were the five largest gas flaring areas in Russia. The gas flaring emissions of PM, NO x and CO in those regions were estimated to overwhelm the other emission sources. And it could partly explain the widespread high NO 2 columns detected by OMI where there were no or very few power plants.
Lastly, Russia's mining emissions in RUS_EDGAR were also found significantly underestimated with a factor of ~16, 2, and 4 lower for PM 10 , NO x , and CO, respectively, as compared to RUS_FSSS. The largest underestimated emissions occurred in remote areas, e.g., the Urals, Siberia, the Far East, and parts of the Northwestern Federal District.    Table S1.

Description of Potential source contribution function
The PSCF is a technique for source region identification that requires both ambient air chemistry data and backward air mass trajectory. PSCF analysis yields a two-dimensional map that shows a synthetic probability field describing the source strength of a geographical area (i.e., a grid cell), which is called as the "Potential Source Contribution". The total numbers (n i,j ) of trajectory endpoints (i.e. coordinates of the back trajectory for each hour before arriving at the receptor site) falling within grid cell [i,j] during the study period are counted. Also, the number of those in the same grid cell with pollutant level higher than a set threshold was calculated as m i,j . Then, the ratio between n i,j and m i,j is the PSCF value for this grid cell: PSCF i,j = m i,j /n i,j . To minimize the biased PSCF caused by the low n i,j values, PSCF ij was weighted with w ij by setting at 0.1 for n i,j < 9, and 1.0 for n i,j ≥ 10. Note that PSCF didn't incorporate any emission input and couldn't resolve detailed small-scale features while it was an indication of the likelihood that a given region contributed to the receptor site.  Table S1. Name, energy capacity, and locations of missing power plants as indicated in Figure S3.