Emissions from Fossil Fuel Combustion and Cement Manufacture : 1751-2017

Abstract. Global and national scale inventories of carbon dioxide (CO2) emissions are important tools as countries grapple with the need to reduce emissions to minimize the magnitude of changes in the global climate system. The longest time series dataset on global and national CO2 emissions, with consistency over all countries and all years since 1751, has long been the dataset generated by the Carbon Dioxide Information and Analysis Center (CDIAC), formerly housed at Oak Ridge National Laboratory. The CDIAC dataset estimates emissions from fossil-fuel combustion and cement manufacture, by fuel type, using the United Nations energy statistics and global cement production data from the United States Geological Survey. Recently, the maintenance of the CDIAC dataset has been transferred to Appalachian State University, and the dataset is now identified as CDIAC-FF. This paper describes the annual update of the time series of emissions with estimates through 2017; there is typically a 2 to 3 year time lag in the processing of the two primary datasets used for the estimation of CO2 emissions. We provide details on two changes to the approach to calculating CO2 emissions that have been implemented in the transition from CDIAC to CDAIC-FF: refinement in the treatment of changes in stocks at the global level, and changes in the procedure to calculate CO2 emissions from cement manufacture. We compare CDIAC-FF's estimates of CO2 emissions with other global and national datasets, and illustrate the trends in emissions (1990–2015) using a decomposition analysis of the Kaya Identity. The decompositions for the top 10 emitting countries show that, although similarities exist, countries have unique factors driving their patterns of emissions, suggesting the need for diverse strategies to mitigate carbon emissions to meditate anthropogenic climate change. The data for this particular version of CDIAC-FF is available at https://doi.org/10.5281/zenodo.4281271 (Gilfillan et al. 2020).


Abstract. Global and national scale inventories of carbon dioxide (CO2) emissions are important tools as countries grapple with the need to reduce emissions to minimize the magnitude of changes in the global climate system. The longest time series dataset on global and national CO2 emissions, with consistency over all countries and all years since 1751, has long been the dataset 15 generated by the Carbon Dioxide Information and Analysis Center (CDIAC), formerly housed at Oak Ridge National Laboratory. The CDIAC dataset estimates emissions from fossil-fuel combustion and cement manufacture, by fuel type, using the United Nations energy statistics and global cement production data from the United States Geological Survey. Recently, the maintenance of the CDIAC dataset has been transferred to Appalachian State University, and the 20 dataset is now identified as CDIAC-FF. This paper describes the annual update of the time series of emissions with estimates through 2017; there is typically a 2 to 3 year time lag in the processing of the two primary datasets used for the estimation of CO2 emissions. We provide details on two changes to the approach to calculating CO2 emissions that have been implemented in the transition from CDIAC to CDAIC-FF: refinement in the treatment of changes in stocks at 25 the global level, and changes in the procedure to calculate CO2 emissions from cement manufacture. We compare CDIAC-FF's estimates of CO2 emissions with other global and national datasets, and illustrate the trends in emissions  using a decomposition analysis of the Kaya Identity. The decompositions for the top 10 emitting countries show that, although similarities exist, countries have unique factors driving their patterns of emissions, 30 1 Introduction 40 Monitoring emissions of carbon dioxide (CO2) to the atmosphere from fossil fuel combustion and other industrial processes is necessary due to the role of CO2 emissions in driving anthropogenic climate change, and because of the importance and prospects for reducing emissions. Emissions of CO2 impact climate systems, ecosystems, and human systems. Fossil fuel CO2 (FFCO2) emissions inventories are important tools as nations, corporations, and 45 individuals grapple with deciding appropriate reduction targets, and as verification that these reductions are occurring. The global carbon cycle is directly influenced by FFCO2 emissions, and periodic updates through emissions inventories provide information concerning the magnitude and extent of these impacts (Friedlingstein et al., 2019). Information from FFCO2 emission inventories reveals whether emissions are increasing or decreasing, which parties are 50 driving these trends, and what fuel types and economic factors are contributing to emissions. Current FFCO2 inventories are compiled using data from the production, consumption, and trade of fossil fuels. Data concerning production and consumption are assembled by multiple national and international agencies: the United Nations (UN), the International Energy Agency (IEA), the United States Energy Information Administration (EIA), and BP company being prominent 55 (Andres et al., 2012;Hutchins, Colby, Marland, & Marland, 2017). Depending on the emissions inventory focus, this fossil fuel data can be used to estimate CO2 emissions by fuel type (solids, liquids, and gases) and/or for economic sectors (energy, transportation, manufacture, etc.). Some inventories may also include emissions from additional industrial processes that emit CO2, such as cement manufacture, or emissions from the flaring of natural gas. 60 Emissions of CO2 from fossil fuel consumption are seldom measured directly, except in recent years at some power plants and other very large point sources, (e.g. (United States Environmental Protection Agency, 2018). FFCO2 emissions are generally estimated from the amount of carbon-based fuels that are consumed. Cement manufacture is often included in CO2 inventories because it is the largest industrial process leading to CO2 emissions that does not 65 involve combustion or the oxidation of non-fuel hydrocarbon products (Gibbs et al., 2000). Cement manufacture emits CO2 into the atmosphere through the process of converting calcium carbonate to lime, an essential ingredient of cement. The FFCO2 emissions from fossil fuels used to support cement manufacture are already included in CO2 emissions inventories (Andres et al., 2012;Andrew, 2019;Le Quéré et al., 2018). Although other industrial processes discharge CO2 70 into the atmosphere, e.g. iron and steel production, they are often not currently included in emissions inventories because of incomplete data and the recognition that their quantities are generally less than the uncertainty associated with FFCO2 emissions (Andres et al., 2012). Natural gas flaring occurs as a byproduct of petroleum and natural gas extraction and processing, such as in oil fields that are not well connected to natural gas markets, and the related CO2 75 emissions are often included in global and national inventories.
Although the ultimate goal of inventories is record keeping of FFCO2 emissions, the foci, boundary conditions, assumptions, and initial data sources make each of the currently existing inventories unique. Inventories can also differ on how to deal with fuel used in international trade (bunker fuels), which industrial processes are included, and sometimes even which 80 countries are included. However, consistency within a dataset is important, and changes to any of these aspects with time or place needs to be noted. It is also important to realize that while each https://doi.org/10.5194/essd-2020-337 of the current inventories presents estimates of emissions of CO2 for global, regional, and/or national totals, the independent verification of emissions is not presently possible. Estimates are based on survey data, derived average values, and large quantities of compiled data. Space-based 85 monitoring may eventually provide independent, third-party verification.
The longest, most consistent time series dataset on CO2 emissions has long been the time series of global and national emissions generated by the Carbon Dioxide Information and Analysis Center (CDIAC) at Oak Ridge National Laboratory (ORNL) (Andres et al., 2012;Marland & Rotty, 1984). The CDIAC emissions dataset extends from the beginning of the industrial era 90 (1751) to essentially the present, and estimates emissions from fossil-fuel combustion and cement manufacture for all countries (Andres et al., 2012;Friedlingstein et al., 2019;Le Quéré et al., 2018). The CDIAC annual inventories began in 1984 when global interest in CO2 emissions was limited to the scientific community, although focused estimates of global emissions had been produced earlier (Keeling, 1973). The CDIAC emissions estimates are based largely on 95 energy statistics from the UN Statistics Division (United Nations, 2019). The time requirement for the international data collection and processing are such that the UN releases this annual database on a two to three year time lag, which is subsequently reflected in the timeline of the CDIAC FFCO2 emission estimates.
The CDIAC FFCO2 inventory has a cosmopolitan user base; it is currently integral in the Global 100 Carbon Project's annual carbon budget (Canadell et al., 2007;Friedlingstein et al., 2019;Le Quéré et al., 2018), has provided data for the Intergovernmental Panel on Climate Change (IPCC) periodic reports, informs deliberations within the UN, and is utilized by the public and the media as a comprehensive resource for trends in CO2 emissions. However, the United States Department of Energy (USDOE) ceased support for this service at ORNL in 2017. The last 105 release supported by the USDOE included emissions estimates for the year 2014 (Boden, Marland, & Andres, 2017).The CDIAC CO2 emissions time series has been restored in 2019 with independent support from Appalachian State University. The most recent update (through 2017) is the focus of this paper. The historic emissions data from CDIAC at ORNL are stored at the USDOE's Environmental Systems Science Data Infrastructure for a Virtual Ecosystem (ESS-110 DIVE) data repository at the Lawrence Berkeley National Laboratory. CDIAC at ORNL supported a plethora of additional carbon related research, but this revival is aimed solely at the important dataset of CO2 emissions, so the Appalachian State University initiative is identified hereafter as CDIAC-FF.
Decomposition analysis is an important tool that can be used to characterize temporal drivers of 115 CO2 emissions, addressing issues such as why certain developed countries are declining in emissions (Le Quéré et al., 2019), assessing the socioeconomic aspects of emissions (Pui and Othman, 2019), or identifying drivers of emissions in specific countries using a variety of decomposition techniques (Brizga et al., 2014;O'Mahony, 2013). The most commonly used approach for this kind of analysis with regard to FFCO2 has involved the Kaya Identity, which 120 relates FFCO2 to four primary factors: population, per capita gross domestic product (GDP) (wealth), energy used per unit of GDP (energy intensity of the economy), and CO2 emitted per unit of energy used (carbon intensity of the energy system) (Kaya, 1989). The IPCC has used the Kaya identity to support analysis of emissions scenarios (Pachauri et al., 2014), although much of their focus on reducing emissions has been on the two elements of energy consumption and 125 carbon intensity. While the Kaya Identity has its limitations, it has regularly been employed due https://doi.org/10.5194/essd-2020-337 to the availability of quality data and its clear messages and general simplicity (O'Mahony, 2013;Pui and Othman, 2019).
In this paper we first review the methodology to produce the CDIAC-FF emissions estimates (section 2.2) and identify changes that have been implemented in the transition from ORNL to 130 Appalachian State University (Boden et al., 2017;Marland & Rotty, 1984). Two significant changes are noted: the method of including data on stock changes for calculating global totals of CO2 emissions (section 2.2.1) and the approach for calculating CO2 emissions from the production of cement (section 2.2.4). We also discuss trends in the 2017 time series of CO2 (section 3.1) and compare our estimates to other available global inventories (section 3.2). 135 Further, we decompose the Kaya Identity for the top 10 emitting countries to illustrate the drivers of emissions trends from 1990 to 2015 (the end date dictated by the availability of necessary supporting data) and the challenge that different countries face in making significant reductions in emissions (section 3.3).

Other global data sets of CO2 emissions from fossil fuel combustion
There are currently available four other prominent, annual, global FFCO2 emissions inventories that are "primary" emissions databases. This means that, like CDIAC-FF, the estimates are derived directly from energy data sources. There are also secondary inventories that synthesize their estimates from multiple primary sources (Andrew, 2020). These primary datasets are 145 available from the IEA, EIA, Emissions Database for Global Atmospheric Research (EDGAR), and the BP Statistical Review of World Energy. Andres et al. (2012) provide a brief discussion of their general characteristics and recently Andrew (2020) has provided a more detailed analysis of the similarities and differences of each of these primary and secondary datasets.
The IEA estimates emissions for both a reference approach (based on fuel type) and a sectoral 150 approach using their own energy questionnaire for member countries, data sharing with the UN for most other countries, national statistical publications, the best estimates from IEA staff experts, and follows the IPCC guidelines for emissions inventories (Andres et al., 2012;IPCC, 2006;IEA, 2019). The IEA data are for CO2 emissions from the energy sector and do not include emissions from fossil fuel products that are used for non-energy applications such as lubricants 155 and solvents, do not include emissions from gas flaring or cement manufacture, but do include emissions from bunker fuels in their estimates of global total emissions. Recently the IEA has published estimates of 2019 global emissions within 2 months of the year's end, based on partialyear data plus some national and market data releases (IEA, 2020).
The EIA collects their own energy statistics from annual, national-level reports from countries; 160 and uses an approach similar to the approach of CDIAC-FF (Andres et al., 2012). They use internally generated data on the carbon content of fuels and estimates of the fraction-oxidized coefficients in their calculations (Andres et al., 2012;EIA, 2019). EIA inventories do include bunker fuels in national totals, along with emissions from gas flaring and adjustment for non-fuel uses, but do not include cement manufacture. 165 EDGAR is produced as a joint effort of the Joint Research Centre of the European Commission and the PBL Netherlands Environmental Assessment Agency. EDGAR uses the energy balance statistics of IEA in a sectoral approach using the IPCC guidelines for emissions estimates, and represents the emissions from bunker fuels, gas flaring, cement manufacture, and non-fuel uses https://doi.org/10.5194/essd-2020-337 using tier I IPCC methods (Andres et al., 2012;Crippa et al., 2019;IPCC, 2006). Note that all of 170 the studies that estimate emissions from cement production rely on cement data from the United States Geological Survey (van Oss, 2019).
The BP Statistical Review of World Energy is the most current FFCO2 inventory, with estimates of emissions reported up to the most recent complete calendar year (BP, 2020). Their estimates for the two most recent years are often used by other inventories to extrapolate emissions values 175 for the two most recent calendar years (Myhre et al., 2009). This allows the Global Carbon Project, EDGAR, and other FFCO2 spatially-explicit inventories to report more-current estimates of global FFCO2 for researchers and the public (Crippa et al., 2019;Friedlingstein et al., 2019;Oda & Maksyutov, 2011;Oda, Maksyutov, & Andres, 2018). The BP dataset uses IPCC emissions factors but only considers fuels for combustion, with no distinction for bunker fuels 180 and no other industrial processes (BP, 2020).

Global fossil fuel CO2 emissions
CDIAC-FF uses the UN energy statistics, collected in an annual questionnaire to all countries, to estimate CO2 emissions (UN, 2019). The information contained in the UN dataset includes 185 production, imports, exports, and changes of stock for all fuels used for energy and non-energy uses. The UN also includes data on fuels that are used in international commerce, known as bunker fuels, and for fuels not categorized as fossil fuels, e.g. wood and other biofuels. Biofuels are not included in estimating CO2 emissions from fossil fuel combustion. The UN period of record dates from 1950 to essentially the present, with a two to three-year time lag between the 190 initiation of collection and final publication of each year's data. This is a dynamic dataset in which changes, additions, and deletions occur with each annual update of the energy statistics, based on reporting from each individual country. CDIAC-FF is a reference approach to CO2 emissions, meaning that we are focused on emissions from different types of fuel rather than from different economic sectors. We estimate emissions for three fuel types (solids, liquids, 195 gases) as well as for gas that is flared and for cement manufacture. CO2 estimates based on fuel type facilitate tracking mass flows among parties and makes possible ancillary estimates such as flows for C isotopes (Andres et al., 2000) Some key differences exist between the approach for estimating the global total of fossil fuel emissions and for estimating national totals. Fuel production data have traditionally been used by 200 CDIAC for global totals, whereas consumption data have been the standard for estimating national totals. The reason for this is the reduced uncertainty in production data at the global level; fewer data points are needed to calculate production totals rather than consumption totals. Calculations for CO2 emissions are conceptually simple and are the product of three terms: the amount of fuel i produced (Pi), the carbon content of the fuel (Ci), and the fraction of the fuel that 205 is oxidized each year (FOi) (Eq. 1). Units for Pi and values used for FOi and the Ci for each fuel type are summarized in Table 1.
A consequence of using fuel production data to estimate global total CO2 emissions is that all non-energy uses of fossil fuels are included in the global totals, as are bunker fuels. At the 210 national level, however, we deal with issues of trade, the portion of fuels used outside of national borders, and fuels that are not oxidized. National totals need to estimate the amount of fuel products that go into long-term products and specifically exclude fuels used in international commerce. A correction factor (part of FOi in Eq. 1) is included in the global total calculation to account for the effective fraction of fuel production that is not oxidized in the year of production 215 because of sequestration in long-lived, non-fuel products, i.e. we estimate that, on a global average, 6.7% of the carbon in liquid fuels produced in a given year is sequestered in long-lived products (Marland & Rotty, 1984). This implies that the balance between the production of long-lived products in any year and the oxidation of long-lived products produced in earlier years is such that the total amount of fuels sequestered in long-lived products increases by 6.7% of 220 annual production. In the 2016 update to the time series we implemented a change in our computation for the estimation of the global total of FFCO2 emissions. All CDIAC data sets prior to the CDIAC-FF data set for 2016 have used only production data, with a global-average value for FOi, for the estimation of global total emissions for solids, liquids, and gases, as well as for emissions from gas flaring. However, the 2016 UN energy statistics revealed a substantial drawdown of fuel 235 stocks already produced and on hand, especially for the solid fuels, and this inspired a refinement of the CDIAC-FF calculation. Historically, reporting of changes in stocks to the UN Statistics Division has been such that the data could be used for some countries but were incomplete for use on total global stocks. The assumption, in essence, was that at the global level there was no net change in stocks each year. 240 The reporting of stock change transactions in the primary UN energy data has been increasing with time and is now judged complete enough to use in the global FFCO2 emissions estimateswhile maintaining consistency with historic estimates. The data show two years in which the abundance of reported data on stock change transactions increased notably in richness -1970 and 1992 (Fig. 1a). By 1992 the data on stock changes approaches the completeness seen in 245 recent year accounts -and this also is the point at which the dissolution of the Soviet Union had occurred, the unification of Germany was complete, and the array of countries in the dataset was stabilizing. Thus, inclusion of stock changes is now part of the estimation of global CO2 emissions going back to 1992. Figure

National fossil fuel CO2 emissions
Fuel consumption data are more informative than fuel production data for scales smaller than 275 global totals because local specificity is needed to properly allocate emissions. At the national level fuel consumption (Eq. 2) is estimated using apparent consumption (ACi) and is substituted for Pi in Eq. 1. Apparent consumption is defined as: Where Pi represents production for a given fuel type i, Ii represents imports, Ei represents 280 exports, Bi represents bunker fuel loadings, NEi represents non-energy uses that are unoxidized (assumed to be zero for solids and gases), and SCi represents stock changes. CO2 emissions from bunker fuels are thus included in estimates of global total emissions but not included in the country totals except to designate the country where fuel loading took place. Emissions of CO2 will occur along international shipping lanes, not in the country where fuel loading took place. 285 Non-energy (non-fuel) uses involve fuel commodities that are used for applications that are not directly consumed for energy uses; examples would be petroleum liquids used to make plastics, lubricants, and asphalt or fertilizer production using natural gas. When the sum of emissions from all country totals does not equal the global total, there are three primary reasons; emissions from bunker fuels are included in the global, but not in national, totals; emissions from fuels 290 produced for non-energy uses are estimated in the global total, but at the national level nonenergy uses are explicitly subtracted out for liquids before estimation of CO2; and the sum of imports for all countries does not equal the sum of exports globally because of statistical errors and incomplete reporting.

Global and national emissions from cement manufacture
The manufacture of cement involves calcining carbonate rock, e.g. limestone, to produce CaOrich clinker, a primary ingredient in cement production. The production of clinker through calcination is one of the largest non-fossil fuel combustion sources of CO2 emissions. The clinker is then fine ground with gypsum and sometimes other additives to produce finished 305 cement. Calculations based on cement production were, and still are, facilitated by a global database of cement production by country maintained initially by the U.S. Bureau of Mines and subsequently by the USGS (van Oss, 2019).
The biggest change in CDIAC-FF is in the estimates of CO2 emissions from cement manufacture. The CDIAC emission factor for CO2 from cement manufacture has remained constant and time 310 invariant since 1987, with the assumption that all hydraulic cements had a high proportion of clinker (90-95%). Since that time, however, the quantity of additives in blended cements has increased broadly, that is the fraction of clinker in finished cements has decreased as additives such as coal fly ash and blast furnace slag have increased (Ke et al., 2013;Kim and Worrell, 2002). This made it clear that the original CDIAC methodology was overestimating CO2 from 315 cement manufacture, especially from China, which now produces over half of the world's cement (van Oss, 2019), and required a revaluation of the assumptions for our calculation. Since the clinker content of cement has been declining since before 1990, and varies with time and place, it follows that the best practice for calculating CO2 emissions from cement manufacture should be based on the amount of clinker in finished cements (IPCC, 2006). The 320 availability of good data on clinker production or the clinker content of cements really begins in 1990, so we have updated CO2 emissions estimates back to 1990 for the recent edition of the CDIAC-FF time series of emissions. To provide estimates of CO2 emissions from cement production that are transparent and consistent over time and space we rely, when possible, on clinker-production data that are publicly available and likely to be updated regularly (Case 1). 325 Where data on clinker production are not available we rely on data for cement production and best estimates of the clinker to cement ratio (Case 2). Emissions of CO2 from cement production, , are calculated as follows: Where 2 is the molecular weight ratio of CO2 to CaO, is the ratio of CaO in clinker (64.6%), is the clinker ratio, is the mass of clinker produced, and is the mass of the cement produced. Since the advent of widespread national reporting of greenhouse gas emissions to the United Nations Framework Convention on Climate Change (UNFCCC) 335 many countries have been reporting values for clinker production in their National Inventory Reports. Time series of clinker production back to 1990 are now available for 31 countries in these National Inventory Reports, and we use this clinker production data to calculate emissions in case 1. We also adopt the IPCC (2006) addition of 2% for cement kiln dust that is not captured in the cement product to generate a final emission factor ( 2 * ) of 0.52 kg CO2 per kg 340 clinker (0.142 kg C per kg clinker).
While cement manufacture is the third largest source of anthropogenic CO2 emissions (after fossil fuel use and land-use change) the availability of the data required for estimating emissions needs improvement (Andrew, 2019 Chinese cement production since 1990 (Cai et al., 2016;Gao et al., 2017;Ke et al., 2012Ke et al., , 2013Kim and Worrell, 2002;Liu et al., 2015;Shen et al., 2015;Wei and Cen, 2019). The IPCC 2006 inventory guidelines do not endorse the process of calculating CO2 emissions directly from cement production data, but the dearth of international data on clinker production and trade dictates that using a to estimate clinker production from cement data is often the best 360 choice commonly available.

Decomposition of recent CO2 emissions trends
The Kaya Identity, first described by Professor Yoichi Kaya (Kaya, 1989), is a way for us to evaluate factors that drive past and future trends in emissions. The Kaya Identity states that CO2 emissions (C) can be expressed as the product of four terms: where P is population, GDP is gross domestic product, and E is primary energy consumption. Data are available from the World Bank on each of these variables(World Bank, 2019). The four factors provide simple representations of population ( ), wealth ( ), the structure and efficiency of the economy ( ), and the carbon intensity of the energy system ( ). We i.e. ΔCx is the change in CO2 emissions over the interval t1 (reference year) to t2 which is attributable to Kaya factor x (Ang, 2005). We decomposed CO2 emissions attributable to each of 380 the factors annually from 1990 to 2015; data were not available to 2017 for each of the World Bank datasets.

Recent trends in global and national emissions
The global total for CO2 emissions from fossil fuel combustion and cement manufacture in 2017

400
The top 10 emitting countries now collectively emit approximately 65% of the world's total emissions. The top 10 emitters represent countries from North America, Europe, and Asia. These 10 countries' emissions and 2016-2017 growth rates as well as population changes and per capita emissions are summarized in Table 2. China has been the global leader in emissions since 2005 with emissions that have grown by 301% since 1990. The total Chinese CO2 emissions declined 405 from 2014 -2016, but saw a 1.7% increase in total CO2 emissions in 2017. Because of the implications of being such a large emitter of CO2, accurate accounting is important for Chinese emissions; however, there is uncertainty associated with Chinese data due in part to uncertainty in coal quality and to the improving quality of data on cement (Han et al., 2020).
The country with the largest reported growth in emissions from 2016 to 2017 in the top 10 410 emitters was Iran, increasing by 21 MtC. This is reportedly driven by a 74 % increase in emissions from the flaring of natural gas (8.9 MtC), followed by a 12.1% increase in emissions from liquid fuel combustion (6.6 MtC) and a 4.9 % increase in the emission from natural gas combustion (  As noted above, there are currently five primary sources for global estimates of CO2 emissions: CDIAC-FF, IEA, EIA, EDGAR, and BP. These emissions inventories have been prepared by different parties with different objectives, different emphases, different boundary conditions, and different results. Some, for example, include emissions from cement manufacture while some do not; but we compare the gross reported total of CO2 emissions as included in the respective 435 reports. Comparisons are not simple but we summarize briefly the alternate data sources and the differences that they convey (section 2.1). Figure 2 compares the final estimates of global total emissions for four years (1990,2000,2016,2017), and a sampling of data for six diverse countries that includes the three largest emitting countries.  Although systematic comparison of the alternate datasets has been undertaken (Andrew, 2020;Ciais et al., 2010;Hutchins et al., 2017;Macknick, 2009;Marland et al., 2007;Marland, Brenkert, & Olivier, 1999), the boundary conditions and assumptions used in the calculations make this comparison difficult. Andres et al. 2012 attempted to put them on common ground, and found that the global CO2 emissions agreed to within 3% of the mean (Andres et al., 2012), 455 and this estimate is similar to more recent comparative analyses (Andrew, 2020). Our goal here is to demonstrate a general accord that includes the reinvigorated CDIAC-FF. Absolute percent differences range from .27% to 20.6% depending on the country, and are less than 10% for the global totals for all four years (Figure 3). At the country level, all of the higher estimates of CO2 emissions (>10%), compared to CDIAC-FF, come from the EDGAR and EIA 460 datasets, while the lower estimates of CO2 (<-10%) come from the IEA, EIA, and the BP datasets. The larger underestimates are generally from the countries of Ecuador, Morocco, and India, while the larger overestimates, compared to CDIAC-FF, consist of China and France. We suggest that the differences are not indicative of accuracy but rather an indication of the different system boundaries and a measure of the uncertainty.

Decomposition of recent trends in CO2 emissions
To gain insight into what is driving changes in CO2 emissions at the country level, decomposition analysis was performed on the top 10 emitting countries for the period 1990-490 2015, or 1992-2015 for Russia and 1991-2015 for Germany. The results are presented as percentage contributions of the four Kaya-based factors (population, wealth, energy intensity of the economy, and carbon intensity of the energy system), to CO2 emissions changes based on the reference year estimates (Fig. 4). For sake of discussion, we will describe positive changes attributable to a specific Kaya factors as drivers of CO2 emissions, while negative change will be 495 described as offsets of CO2 emissions.  With the exception of the impacts of the dissolution of the Soviet Union on Russia, increasing wealth (per capita GDP) is a driving force on increasing emissions in each of the top 10 emitting countries. This is especially evident in China, where increasing wealth has contributed to a 561% increase in CO2 emissions from 1990 -2015. China's growth in wealth is partially offset by 510 decreases in energy intensity (250 % decrease in 2015, relative to 1990). Other countries that see this pattern of increasing wealth substantially driving emissions are India (312% increase 1990 -2015) and South Korea (243% increase 1990Korea (243% increase -2015. These are emergent, developing economies representing some of the fastest growing economies in the world since 1990. The dominant offsetting factors for these countries are decreasing energy intensity for India (116% 515 decrease) and decreasing carbon intensity for South Korea (106% decrease).
Saudi Arabia and Iran, the top 10 emitting countries from the Middle East, exhibit unique characteristics of the Kaya factors in which energy intensity is a driving force in increasing emissions in addition to population growth and increasing wealth. In Iran, 116% of the growth in emissions from 1990 to 2015 can be attributed to increasing wealth, 79% from increasing energy 520 intensity, and 61 % from population growth. These are modestly offset by decreases in carbon intensity of the energy system (50% decrease). Saudi Arabia is the only nation in the top 10 emitting countries in which population growth is the dominant driving force (132% increase, relative to 1990 values); decreasing carbon intensity of the energy system only provides modest offsets (33% decrease) to increasing CO2 emissions. 525 The remaining top 10 emitters (United States, Russia, Japan, Germany, and Canada) are all Annex-I countries with obligations to regularly report emissions to the UNFCCC; this potentially explains the minimal relative growth in CO2 emissions (<50% of 1990 emissions). The countries are characterized by increasing wealth having the largest magnitude influence on CO2 emissions, but this is offset by decreases in carbon intensity followed by decreases in energy intensity. 530 Population growth only contributes minimally to the trends in emissions in each of these countries, and in some cases (Russia) decreasing population is a small offsetting factor for CO2 emissions.

Data Availability
The exact version of the CDIAC-FF time series of CO2 emissions from fossil fuel 535 combustion and cement manufacture that is described in this publication is located here: https://doi.org/10.5281/zenodo.4281271 (Gilfillan et al., 2020) The historic record of CDIAC products from ORNL are archived here: https://data.essdive.lbl.gov/view/doi:10.3334/CDIAC/00001_V2017. Future and previous updates from CDIAC-FF produced at Appalachian State University will be included at https://data.ess-540 dive.lbl.gov/view/doi:10.15485/1712447. The most recent inventory year will also be located within the Appalachian Energy Center's website (https://energy.appstate.edu/research/workareas/cdiac-appstate). This includes .csv files for global and national totals as well as a ranking of each country with regard to total emissions and per-capita emissions for a given year.

545
FFCO2 emissions inventories are integral tools to evaluate sources of CO2 emissions, document trends concerning fuel and/or sectoral-based values, and verify that intended reductions are indeed occurring. While each of five available global emissions inventories is unique in approach, focus, boundary conditions, time interval covered, and application; the small differences in overall emissions estimates demonstrate the accuracy and integrity of the different 550 products and statistical approaches. Differences do not reflect the degree of accuracy since independent verification is not currently available at the global and national scales, especially for CO2 emissions for which there are both natural and anthropogenic sources. CDIAC-FF provides a long-term time series of FFCO2 emissions that is both comprehensive and consistent over time and countries. In continuing the CDIAC-FF data set at Appalachian State we provide long-term 555 continuity while continuing to provide updates and refinements as knowledge and available data permit. Improving availability of data on stock changes of global fuels and production of clinker have permitted improved estimates in the 2017 CDIAC-FF dataset.
In addition to evaluating changes in FFCO2 emissions over time, we consider what is driving recent changes for the top ten emitting countries. To evaluate the possibilities for limiting energy intensity while emissions growth in Saudi Arabia is being driven by population growth. The Kaya decomposition approach employed is simple but provides a framework for more extended analysis of the factors driving changes in emissions. While much of the previous analysis on a Kaya framework has focused on energy and carbon intensity, there is a need to 570 characterize the more difficult aspects of carbon mitigation: growth in population and wealth.
The future and equitable confrontation of climate change mitigation will rely on appropriate accounting of CO2 emissions across countries and across time. The top ten emitting countries each have a unique combination of drivers of changing emissions and the need for diverse strategies to mitigate carbon emissions. National and global inventories will provide evidence 575 whether planned emissions reductions are taking place.