Exploring differences in FFCO2 emissions in the United States: comparison of the Vulcan data product and the EPA national GHG inventory

Quantitative assessment of greenhouse gas emissions is an essential step to plan, track, and verify emission reductions. Multiple approaches have been taken to quantify U.S. CO2 emissions from fossil fuel combustion (FFCO2), the primary driver of global climate change. A 2020 study analyzing atmospheric 14CO2 observations (a key check on bottom-up estimates) and multiple inventories found significant differences in the U.S. total FFCO2 emissions. The specific reasons for the differences were left for future work. Here, we take up this task and explore the differences between two widely used U.S. FFCO2 inventories, the Vulcan FFCO2 emissions data product and the Environmental Protection Agency (EPA) GHG inventory, developed using mostly independent data sources. Where possible, we isolate definitions and data sources to quantify/understand discrepancies. We find that the initial 2011 emissions difference (104 MtC yr−1; RD = 10.7%) can be reduced by aligning the two estimates to account for differing definitions of emission categories or system boundaries. Out of the remaining 90.6 MtC yr−1 gap (RD = 6.2%), we find that differences can be largely explained by data completeness, emission factors, and fuel heating values. The remaining difference, 45.4 MtC yr−1 (3.2%), is difficult to isolate due to limited EPA documentation and disaggregation of emissions by sector/fuel categories. Furthermore, the final net difference obscures countervailing gross differences (∼40 MtC yr−1) within individual sectors. Nevertheless, this comparison suggests the potential for a national estimation approach that can simultaneously satisfy reporting at the national/global scale and the local scale, maintaining internal consistency throughout and offering detailed decision support to a much wider array of stakeholders.


Introduction
Anthropogenic emissions of carbon dioxide from the combustion of fossil fuels (FFCO 2 ), the primary driver of global climate change, accounted for 64% of global greenhouse gas (GHG) emissions in 2019 [1].Climate change impacts are already manifest as a 1.09 • C increase in global temperature during the 2011-2020 time period relative to the 1850-1900 time period [2], an accelerated sea level rise [3][4][5], more frequent and severe drought [6,7], and increased heat-related human mortality [8], among others.As of May 2022, atmospheric CO 2 concentration has shown a nearly 1.5 times increase over pre-industrial levels [9].
Countries, including the United States, adopted the Paris Accords on climate change which constituted agreement to limit global warming to no more than 2 • C compared to the pre-industrial era [10].In April, 2021, the U.S. agreed to cut GHG emissions 50%-52% relative to 2005 levels by the year 2030 and achieve net zero emissions no later than 2050 [11].One essential component of assessing the current GHG trends and GHG mitigation potential is an accurate accounting of FFCO 2 emissions from local to national to global scales [12].
The U.S. is a signatory to the United Nations Framework Convention of Climate Change (UNFCCC), and this framework requires a submission of an annual national GHG inventory [10].The Inventory of U.S. Greenhouse Gas Emissions and Sinks prepared by the U.S. Environmental Protection Agency (EPA) is a comprehensive 'bottom-up' assessment of GHG emissions in the U.S. since 1990 [13].The USEPA GHG inventory (EPA inventory here after) is the official emissions data submitted to the UNFCCC and reports annual total FFCO 2 emissions across economic sectors using national level data.
Independent of the UNFCCC process, the scientific community has also pursued estimation of FFCO 2 emissions, often to support investigation into aspects of the global carbon cycle where key questions remain regarding uptakes/emissions from forests and soils and their potential sensitivity to climate change (e.g.[14,15]).Principle among the bottom-up estimation efforts in the U.S. is the Vulcan Project, the first comprehensive bottom-up U.S. national space/timeresolved estimate of FFCO 2 emissions from different economic sectors [16,17].Begun in 2004, the Vulcan Project constructs emissions estimates from a large number of datasets including direct flux monitoring, vehicle activity, fuel statistics, and building attributes.The gridded Vulcan emissions is widely used in scientific and policy analysis as a key constraint to atmospheric CO 2 inversion studies [18][19][20][21].
A critical means to evaluate these bottom-up estimation approaches is via comparison to a topdown estimation approach which uses a combination of atmospheric transport models and atmospheric GHG concentration measurements.Furthermore, top-down approaches using atmospheric 14 CO 2 concentration measurements, a near-perfect tracer for FFCO 2 [22], eliminates the confounding impact of biological or oceanic carbon exchange [23,24].In a recent study, an estimate of US FFCO 2 emissions using 14 CO 2 monitoring, found a 1.4% difference to the Vulcan 2010 emissions estimate.The comparison to the EPA inventory showed a ∼3X larger difference (4.4%).This translates into a 6% difference between Vulcan and the EPA inventory (Vulcan > EPA) for the year 2010.
Understanding the difference between the two independent datasets, the Vulcan data product and the EPA inventory, offers potential insight into the FFCO 2 inventory construction and its underlying data and estimation algorithms.A 6% difference, for example, amounts to an absolute emissions value larger than the total annual FFCO 2 emissions of 177 countries.Here, we explore the Vulcan and EPA sector and fuel-specific estimation methods and, where possible, isolate key definitions, parameters, and data sources, to quantify and better understand the difference.This exploration will not only provide insights into the best practices in estimating emissions, but offer additional numerical support for uncertainty estimation on inventories in general.Isolating and ultimately reducing these differences also offer the potential for an estimation system that simultaneously meet UNFCCC obligations and the large number of sub-national decisionmakers in need of accurate GHG emissions estimation.

Vulcan and the EPA GHG Inventory
Vulcan version 3.0 data represents direct FFCO 2 emissions on a 1 km × 1 km grid in the 50 U.S. states and District of Columbia and includes information on emission sector, technology, and fuel for 2010-2015 period [17].The Vulcan data product and associated documentation is publicly available from the Oak Ridge National Laboratory Distributed Active Archive Center [25].For more details about the estimation procedure, see Gurney, et al [17].The Vulcan annual total FFCO 2 emissions for 2010-2015 were aggregated by sector and fuel, and the sectorspecific uncertainty estimates, which represent upper and lower confidence intervals at the 95% confidence levels (high and low), were also used to estimate a one-sigma uncertainty.
The EPA's annual national FFCO 2 emissions were obtained from the April 2020 version of the EPA GHG inventory, a reporting year consistent with the published year of Vulcan version 3.0 data [13].Though various vintages of the EPA inventory show revisions to emissions in previous years, the changes to the EPA estimated total emissions for the 2010-2015 time period, when examining the reports between 2017 and 2022, amount to less than 0.4% of the total emissions in any given year.The Energy chapter includes emissions for IPCC's Common Reporting Framework (CRF) Source Category 1A & 1B, which are the direct emissions from fossil fuel combustion from stationary and mobile sources (1A) and fugitive emissions (1B).Cement production is reported separately under IPPU chapter (CRF 2A1).All CO 2 emissions were converted to carbon equivalent units (MtC yr −1 ) for comparisons purposes.
We include ethanol emissions as a component of gasoline (gasohol) consumed for onroad transportation and some for the commercial and industrial activities.All other biogenic CO 2 exchange categories are excluded.

Source category crosswalk
The emission source categories (or sectors) used in the Vulcan data product and the EPA inventory were paired according to their documented definitions ('sector-alignment' , figure 1).Definitional differences were accounted for by reallocating specific emission categories in a consistent compatible fashion (see text S1 and S2, and table S1 for the detailed descriptions).
There are nine different sectors used here comprised of seven traditional sectors (electric power, onroad, airport, Commercial Marine Vessel: CMV, railroad, residential, cement production), pleasure craft, and one aggregate sector (commercial, industrial, and nonroad: CIN).For the airport and CMV sectors, we consider emissions only from aviation taxiing and landing-taking-off (LTO) cycles and only at the U.S. seaports (see equation in text S3 for the detailed crosswalk and estimation methods).Both sectors include international flight and shipping (bunker) emissions.The CIN emissions were aggregated because the EPA inventory treats nonroad emissions (off-road petroleum consumption) as a part of commercial or industrial emissions and not reported separately as is done in Vulcan.The emissions from flaring and non-energy use of lubricants are also added to the CIN and onroad (lubricants only) categories, respectively.See text S1, S2, and S3 for a more detailed description of the sector alignment method.
After the sector alignment is performed, we explored the FFCO 2 emissions calculation methodology (e.g.primary data source, fuel heat content (HC)) in an attempt to account for identifiable differences such as missing data in one or other dataset or differences in the choice of key parameters.

Statistical analysis
We define the fuel-specific, sector-specific, and total annual emissions differences (TDs) between Vulcan and EPA as follows: TD sector [ y] = where D i,j [y] is an annual emissions difference between the emissions from Vulcan (FFCO 2Vulcan,i,j [y]) and EPA (FFCO 2EPA,i,j [y]) for sector i (i = 1, 2… n sector ) and fuel j (j = 1, 2… n fuel ) calculated for the year y (y = 2010, 2011…2015).
The n sector and n fuel represent the number of sectors (n sector = 9) and the number of major fuel types (i.e.coal, NG, and petroleum; n fuel = 3).TD [y] is the overall annual total emissions differences and calculated by aggregating D i,j [y] across sectors and fuels.We also calculate relative differences (RDs) for each sector and total: where RD i [y] is a sector-specific, annual relative difference (in %) between the emissions from Vulcan (FFCO 2Vulcan,i [y]) and EPA (FFCO 2EPA,i [y]) calculated for sector i for the year y.
The one-sigma uncertainty was estimated in the Vulcan and EPA results.For Vulcan, the upper and lower uncertainty bounds were calculated by halving the reported confidence interval.For EPA, those values were obtained from the EPA GHG inventory Annex tables.

Total emission differences
The annual total FFCO 2 emissions (50 states and DC, excluding U.S territories) from the Vulcan data product and EPA inventory prior to any sector alignment show similar year-to-year changes throughout the 2010-2015 time period (figure 2(a)).Emissions declined about 50 MtC yr −1 in both estimates from 2010 to 2012, then turned upwards in 2014 and downward again in 2015.These interannual variations are identical to those from the Energy Information Administration's (EIA) annual energy consumption.Vulcan's annual total FFCO 2 emissions were consistently larger than those from the EPA inventory for all analysis years.The Vulcan uncertainty was about two times larger than the EPA uncertainty (table 1 for 2011, see tables S2 and S3 for all years), and the lower boundary of the Vulcan uncertainty encompassed the EPA emissions estimate for all years.
When examined by fuel type (table 1, figure 3), petroleum FFCO 2 emissions constituted the largest contribution to the 2011 total (2011 is the Vulcan 3.0 'base' year) in both Vulcan (44.9%) and the EPA inventory (40.5%), followed by coal (34.6% and 35.0%, respectively) and natural gas (NG, 20.5% and 24.5%, respectively).To be comparable with Vulcan, the small biofuel-based emissions in the EPA inventory (ethanol blended in gasoline reported separately from gasoline, 19.9 MtC yr −1 ) were added to the gasoline total.
Among the three major fossil fuels, the FFCO 2 emissions differences were largest and positive (Vulcan > EPA) for petroleum (+102.3MtC yr −1 ) and coal (+25.9MtC yr −1 ) but negative (−37.7 MtC yr −1 ) for NG.We further investigated the petroleum FFCO 2 emissions by individual fuel.All of the individual Vulcan petroleum fuel FFCO 2 emissions were larger than EPA, except for petroleum coke (−11.7 MtC yr −1 difference).

Sector-specific differences
The 2011 sector contribution to the total annual emissions in each data product (from this point forward all results reflect the sector alignment) was also examined (table 2).Electricity production emissions account for the largest contribution to total emissions in both Vulcan (40.4%) and EPA (41.7%), followed by the onroad (both 29%), and CIN sectors (both 21%).Hence, the combination of the electricity production, onroad, and CIN sectors account for >90% of the 2011 annual total emissions.The relative contributions of the remaining sectors (other mobile sources, residential, and cement) were <10% (<144 MtC yr −1 ) of the Vulcan and EPA 2011 emissions total.Similar results were seen in years other than 2011 (table S5).
The remaining results show timeseries comparison results for the electricity production, onroad, and CIN sectors only (>90% of total emissions).All other smaller sectors are presented in the supplementary information only (figure S1).

Exploring differences
The electricity production FFCO 2 emissions from the Vulcan data product and EPA inventory showed similar interannual variability (figure 4(a)).The emissions declined 64.8 MtC from 2010 to 2012, stayed at the same level until 2014, then dropped again in 2015.These temporal changes were mainly driven by the decline in coal-fired electricity production emissions in both Vulcan and EPA (>71% of the total, figure S2).By fuel type, the electricity production FFCO 2 emissions difference was greatest for petroleum, though in 2015, the difference in coal emissions slightly exceeded that from petroleum combustion (figure 4(b)).
Examination of the electricity production data sources common to the Vulcan data product and the EPA inventory indicates that most of the 2011 difference (10.4 MtC yr −1 out of the total difference of 18.2 MtC yr −1 ) was explained by individual power plants present in the Vulcan data product but not included in the EPA inventory (table 3).Table 1.Fuel-specific 2011 FFCO2 emissions (MtC yr −1 ) for the Vulcan data product and EPA inventory after sector alignment.The fuel share within each data product is shown in parenthesis (%).For petroleum fuels, the fuel share of individual refined fuels to the total petroleum emissions is shown (in italics).The signed difference (MtC yr −1 ) was calculated as Vulcan-EPA.

Fuel types
Vulcan (MtC yr a Vulcan ethanol emissions are included in the gasoline category (gasohol).when these identifiable differences are accounted for is effectively zero (table 3, figure 4(a)).See table S6 for the years other than 2011.The onroad FFCO 2 emissions in both data products followed similar variability over the 2010-2015 time period, reaching a minimum value in 2012 (figure 5(a)).The Vulcan onroad emissions remain larger than EPA across all years.Gasoline emissions contributed most to the total onroad differences (>60% of the absolute difference, figure 5(b)) followed by diesel onroad emissions.The lubricant combustion (i.e.engine oil, reported in the EPA inventory only) contributed little to the emissions differences.Note that neither Vulcan nor EPA diesel onroad emission include biodiesel and the compressed natural gas (CNG) emissions in Vulcan are included in the gasoline emissions sub-category due to the NEI2011v1 fuel categorization [27].
Greater than 90% of the onroad sector differences were explained by the use of different fuel HC values (30% of the 2011 total difference) and inclusion/exclusion of ethanol emissions in the onroad total (table 4).Fuel HC values are commonly needed when transforming emission factors from physical (e.g.emissions per gallon of gasoline) to thermodynamic units (e.g.emissions per joule of gasoline).Throughout Vulcan, other than the onroad sector, high heating values (HHV) are used which incorporate the heat of vaporization of water during the combustion process.The EPA inventory similarly uses HHV.However, because the MOVES2010b model, whose output constitutes the onroad Vulcan input, uses lower heating values (LHV) to estimate the onroad emissions [28], these were reflected in the final Vulcan results.This use of LHV in Vulcan explains 27 MtC yr −1 of the 2011 total onroad difference (table 4).The final 2011 difference after accounting for the HC difference is 10.1 MtC yr −1 (table 4).These differences were similarly present in years other than 2011 (see table S7).
After accounting for these often-countervailing differences, the remaining 10.1 MtC yr −1 difference could reflect uncertainty in the parameters used in  the MOVES model such as fuel efficiency, vehicle population and age, and fuel blending, which is difficult to access and beyond the scope of this comparison.
The FFCO 2 emissions from the CIN sector differed by an average of 18 MtC yr −1 over the 2010-2015 time period with both Vulcan and EPA data products increasing slightly starting in 2012 (figure 6(a)).While the Vulcan emissions were consistently larger than EPA, the uncertainty bounds overlapped.
The 2011 FFCO 2 emissions from petroleum consumption accounted for the largest share of total CIN emissions (45.6%) in the Vulcan data product, whereas emissions from NG accounted for the largest share in the EPA inventory (56.0%) (table 5).NG and petroleum fuels contributed near equally to the CIN sector difference.However, the differences were of the opposite signs, with Vulcan petroleum CIN emissions larger than the EPA estimate and Vulcan NG emissions smaller than the EPA estimate (figure 6(b), table 5).
We found that the largest emission difference in CIN was associated with diesel (32.2 MtC yr −1 , 46.2% of the total absolute petroleum difference, table 5).Upon closer examination, this difference was driven by diesel fuels consumed for off-road purposes, and was possibly due to the uncertainty in the combustion processes of agricultural, construction, and mining equipment (e.g.engine types, sizes).The EPA's fuel-specific adjustments applied to the commercial and industrial gasoline and diesel fuels consumption (to match those of the EIA national total fuel consumption) might partly explain this difference.A similar result was seen in years other than 2011 (table S8).

Discussions
A variety of agencies, NGOs and academic researchers are generating estimates of GHG emissions at various temporal and spatial scales.Because of the challenges associated with validating these estimates, comparisons where overlaps exist, are a critical step in better understanding the differing methods and attaining more accurate final emissions estimates.In the U.S., the EPA constructs a national inventory of GHG emissions, including FFCO 2 , in compliance with the provisions of international policy [13].The Vulcan Project, similarly generates estimates of FFCO 2 emissions, but was built for different purposes, namely incorporation into systems that use atmospheric monitoring combined with 'bottom-up' estimates of fluxes (e.g.[29,30]).An additional Vulcan objective is to provide emissions information at spatial and temporal scales useful for mitigation management by cities, states or province, and private sector users.In a recent study, it was noted that the nationalscale FFCO 2 emissions estimate from the Vulcan data product and the EPA inventory differed by 6% [18].Though a small relative difference, this amounts to an absolute emissions difference of roughly 100 MtC, an emissions value larger than the total annual FFCO 2 emissions of 177 countries (92% of the countries in the world) [14,31].Therefore, it is critical to understand the specific circumstances driving the differences in these two commonly-used FFCO 2 estimates.
Given the complex source composition of CO 2 emissions from fossil fuel combustion, a key first step in a comparison is ensuring that the categorical and system boundary definitions are consistent.In the case study described here, we reduced the difference between the Vulcan U.S. FFCO 2 emissions data product and the EPA U.S. FFCO 2 inventory considerably by aligning the emissions according to sector definitions and fuel categories.Using 2011 as an example year within the time series examined, we found that 13.5 MtC yr −1 of the annual total FFCO 2 difference (TD = 104.0MtC yr −1 ) could be explained by these alignment issues.When quantifying this using absolute (unsigned) difference, it was 66.3 MtC yr −1 of reduction from the initial difference (156.8MtC yr −1 ), and this was equivalent to 42.3% of the difference.This emphasizes the need to always consider alignment of the underlying definitional differences among GHG emissions estimates.
After the sector alignment, the 2011 difference between EPA and Vulcan is 90.6 MtC yr −1 (Vulcan > EPA) or a RD of 6.2%.This difference was accommodated by the uncertainty bounds of both data products (figure 2(a)).This was also true, not surprisingly, for the three largest emitting sectors (electricity production, onroad, and CIN; figures 4(a), 5(a), and 6(a), respectively).Nevertheless, the absolute differences, though smaller after alignments, are worth examination for what they might reveal in terms of methodological differences and additional insight into the data sources used.For example, the persistent offset between the two data products when the relative temporal changes are consistent is suggestive of a broad, time-independent methodological difference.
Nearly half of the 90.6 MtC yr −1 national FFCO 2 emissions difference was attributed to a few explicit data and estimation method elements best enumerated by examining each sector individually.The different characterization of powerplant facilities in the electricity production sector and the use of different HC in calculating emissions from gasoline and diesel fuels in the onroad sector explained 57.1% and 72.7% of the sector total differences.A central example of the impact of HC choice is the use of LHV versus HHV values.While the choice of whether to use a LHV or HHV is driven by particular interests and requirements of different inventory developers, it is important to use that choice consistently across all sectors in an emissions data product.A key recommendation for the Vulcan data product is to consistently use an HHV throughout the emissions estimation in future versions, something not done in the version 3.0 used in the comparison here.As Vulcan aims to accurately quantify the amount of CO 2 that enters the atmosphere from fuel combustion activities, HHV is the better choice.By accounting for those numerical differences, the final 2011 difference between Vulcan and EPA is 45.4 MtC yr −1 (RD = 3.2%, figure 7).Further explanation of this residual difference remains a task for future research.Further investigation at a finer space/time scales (e.g.state-and sub-annual levels) would help narrow down issues not visible at the national level and reduce unnecessary emission biases that are missed or not covered in this work.It is also worth noting that the difference in the combined CIN sector (TD = 16.9MtC yr −1 ) is the result of larger, countervailing differences between CIN NG and CIN petroleum emissions (nearly 40 MtC yr −1 , table 5).These differences, in particular, are worth exploring further with more detailed data and disaggregation by fuel/sector, which will help better understand the underlying causes.For example, the emissions from nonroad mobile activities are an independent emission category in the Vulcan estimate, while the EPA reports those as part of either commercial or industrial emissions (owing to the data source used).Some fuels, such as petroleum coke (table 5), show very large percentage differences.This may be due to the fact that these fuels are used both as feedstocks to material production and as fuels for energy.Data on these proportions are limited and uncertain.
Additional work is recommended as it remains unclear what is causing the remaining differences in results, especially at the sub-sector and fuel levels.Also, because we focused on the U.S. national domain, the key elements identified here may not be applicable nor fully explain the emissions differences that may arise when comparing inventories for other countries.
For the purposes of policymaking at national scales, the small relative difference between the Vulcan and EPA estimates is unlikely to have implications for emissions mitigation choices.This is particularly the case given that the two data products show very similar trends over time with persistent offsets, making the comparison of a target to a baseline level equally consistent.Unexplored, however, is whether or not that consistency at the national level remains at small spatial scales (i.e.urban) where there is very active policy engagement.While the Vulcan data product estimates emissions at these finer scales, the EPA inventory does not and hence, such a comparison is not currently possible.
The small relative difference (3.2%), however, masks what amounts to a large absolute difference (45.4 MtC yt −1 ), important when the emissions are placed in the context of consistency with atmospheric measurements [32].Consistency with the atmosphere is critical as atmospheric measurements, and flux estimates retrieved thereof, naturally integrate all sources and sinks of emissions and offer an independent estimate of emissions amounts.Thus, the absolute accuracy of emission data products such as Vulcan or the EPA inventory are critical when being considered as part of a larger integration of activity and atmospheric-based estimates [32].

Conclusions
Inventories of greenhouse gases are critical to understanding, tracking, and verifying GHG emissions and emission reductions.These are also a critical component of fully integrated emissions estimation approaches such as those that incorporate atmospheric concentration measurements [32].Hence, the accuracy and completeness of inventories must be explored and improved, where possible.Inventory inter-comparisons, where more than one exist, are essential tools to explore and assess quality and methodological choices.Here, we explore the difference between two well-known U.S. annual FFCO 2 emission inventories, the Vulcan v3 FFCO 2 emissions data product and the EPA GHG inventory.We find that nearly half of the initial emissions difference is due to the differing definitions of emission categories or system boundaries.We align the two estimates to account for these differences before exploring the remaining numerical and methodological differences at the sector and fuel scales.Out of the 90.6 MtC yr −1 difference (RD = 6.2%) in 2011, we find that additional half (49.8%) can be explained by differences in data completeness, emission factor choices and fuel heating values.The remaining 2011 difference, 45.4 MtC yr −1 (50.2%, RD = 3.2%), is difficult to isolate mechanistically due to limited documentation and disaggregation of emissions by sector/fuel categories.Furthermore, this final net difference obscures countervailing gross differences (∼40 MtC yr −1 ) in individual sectors that remain important to understand and worthy of further investigation.
This study does not purport to establish which of the two estimates are more 'correct'-indeed the uncertainties are large enough to encompass the final 3.2% difference between the two estimates.Rather, the results provide deeper insight into the methods and data sources used with the aim of converging on agreed approaches and data.Given the differing initial aims of the Vulcan data product (integration into atmospheric inverse system approaches) and the EPA inventory (compliance with international policy), this comparison also sets the stage for a single estimation approach that can satisfy both the reporting from the global scale to the local scale, maintaining internal consistency across scales.

Figure 1 .
Figure 1.The sector crosswalk process depicting the Vulcan data product sector classification (rectangles with rounded corners) and the EPA inventory (filled rectangles).NG: natural gas, CMV: commercial marine vessel, LTO: landing-taking-off.

Figure 2 .
Figure 2. Annual total FFCO2 emissions (MtC yr −1 ) for the Vulcan data product and EPA inventory, and emissions differences for 2010-2015; (a) FFCO2 emissions time series for Vulcan (orange) and EPA (green) after sector alignment.The shaded areas represent one-sigma error bounds; (b) annual total emission difference (TD, columns) and relative difference (RD, lines) before (gray) and after (black) the sector alignment.

Figure 3 .
Figure 3. Fuel-specific 2011 FFCO2 emissions (MtC yr −1 ) for the Vulcan data product and EPA inventory after sector alignment.(a) Fuel-specific FFCO2 emissions for coal, NG, petroleum, and biofuel; (b) emissions difference by fuel.For petroleum only, the fuel share of individual refined fuels to the total petroleum emissions are shown.

Figure 4 .
Figure 4. Vulcan and EPA annual electricity production FFCO2 emissions differences by fuel type (TD fuel ) for 2010-2015.(a) FFCO2 emissions time series for Vulcan (gray) and EPA (green).The shaded areas represent one-sigma error bounds; (b) Vulcan versus EPA emissions difference by fuel type (TD fuel ).

Figure 5 .
Figure 5. Vulcan and EPA annual onroad FFCO2 emissions and differences for 2010-2015: (a) FFCO2 emissions time series for Vulcan (gray) and EPA (green).The shaded areas represent one-sigma error bounds; (b) Vulcan versus EPA emissions difference by fuel type (TD fuel ).

Figure 6 .
Figure 6.Annual commercial, industrial, and nonroad (CIN) FFCO2 emissions and differences for 2010-2015 after.(a) Emissions time series for the Vulcan data product (orange) and EPA inventory (green) and the one-sigma error bounds (shaded area); (b) difference between Vulcan and EPA by fuel type (TD fuel ).

Figure 7 .
Figure 7. Breakdown of annual total emission difference (TD in MtC yr −1 , column) and relative difference (RD in %, line) for 2010-2015.The column colors represent identifiable differences (filled gray) and remaining differences (filled black).The line represents RDs after accounted for the identifiable differences.

Table 3 .
Factors contributing to the 2011 electricity production sector FFCO2 emission differences (MtC yr −1 ) between the Vulcan data product and EPA inventory.The difference was calculated as Vulcan-EPA inventory.Vulcan data sourceVulcan (MtC yr −1 ) EPA (MtC yr −1 ) Difference (MtC yr −1 )Vulcan includes emissions from small power plants (generation below 1 MWh yr −1 ) not reported to EIA-923 (the EPA inventory data source).Furthermore, instances occurred where power plants present in both Vulcan and EPA, were sourced to different input datasets (4.3 MtC yr −1 difference).In the Vulcan data product, dual reporting from the Clean Air Markets Division (CAMD) and EIA-923 datasets were handled by prioritizing the CAMD data over the EIA-923 (estimate from fuel consumption) [26].The EPA inventory, by contrast, uses the EIA-923 data exclusively.Other sources of difference include the use of different carbon emission factors (3.5 MtC yr −1 ).Thus, the final emission difference

Table 4 .
Fuel breakdown and factors contributing to the 2011 onroad FFCO2 emissions differences (MtC yr −1 ) between the Vulcan data product and EPA inventory.The difference was calculated as Vulcan-EPA.Vulcan (MtC yr −1 ) EPA (MtC yr −1 ) Difference (MtC yr −1 )a Vulcan ethanol emissions are included in the gasoline category (gasohol).b Includes small amount of emissions from Compressed Natural Gas (CNG) and Liquefied Petroleum Gas (LPG).

Table 5 .
The 2011 annual FFCO2 emissions difference (MtC yr −1 ) between the Vulcan data product and EPA inventory for the CIN sector with disaggregation of petroleum fuel into refined fuels.The difference was calculated as Vulcan-EPA.
a EPA ethanol emissions are included in gasoline.b Includes petroleum emissions from lubricants and incineration of waste.