A global anthropogenic emission inventory of atmospheric pollutants from sector- and fuel-specific sources (1970-2017): An application of the Community Emissions Data System (CEDS)

. Global anthropogenic emission inventories remain vital pollution, as well as the resulting impacts on the environment, human health, and society. Rapid changes in today’s society require that these inventories provide contemporary estimates of multiple atmospheric pollutants with both source sector and fuel-type information to understand and effectively mitigate future impacts. To fill this need, we have updated the open-source Community Emissions Data System (CEDS) (Hoesly et al., 2019) to develop a new global emission inventory, CEDS GBD-MAPS . This inventory includes emissions of seven key atmospheric pollutants (NO x , CO, SO 2 , NH 3 , NMVOCs, BC, OC) over the 25 time period from 1970 – 2017 and reports annual country-total emissions as a function of 11 anthropogenic sectors (agriculture, energy generation, industrial processes, transportation (on-road and non-road), residential, commercial, and other sectors (RCO), waste, solvent use, and international-shipping) and four fuel categories (total coal, solid biofuel, and the sum of liquid fuels and natural gas combustion, plus remaining process-level emissions). The CEDS GBD-MAPS inventory additionally includes global gridded (0.5  0.5  ) emission fluxes with monthly time resolution for each compound, sector, and fuel-type to facilitate 30 their use in earth system models. CEDS GBD-MAPS utilizes updated activity data, updates to the core CEDS default scaling procedure, and modifications to the final procedures for emissions gridding and aggregation to retain sector and fuel-specific information. Relative to the previous CEDS data released for CMIP6 (Hoesly et al., 2018), these updates extend the emission estimates from 2014 to 2017 and improve the overall agreement between CEDS and two widely used global bottom-up emission inventories. The CEDS GBD-MAPS inventory provides the most contemporary global emission estimates to-date for 35 these key atmospheric pollutants and is the first to provide global estimates for these species as a function of multiple fuel-types across multiple source sectors. Dominant sources of global NO x and SO 2 emissions in 2017 include the combustion of oil, gas, and coal in the energy and industry sectors, as well as on-road transportation and international shipping for


Introduction
Human activities emit a complex mixture of chemical compounds into the atmosphere, impacting air quality, the environment, and population health. For instance, direct emissions of nitric oxide (NO) rapidly oxidize to form nitrogen dioxide (NO2) and can lead to net ozone (O3) production in the presence of sunlight and oxidized volatile organic compounds (VOCs) (e.g., Chameides, 1978;Crutzen, 1970). In addition, direct emissions of organic and black carbon-containing particles (OC, BC), as 55 well as secondary reactions involving gaseous sulfur dioxide (SO2), NO,ammonia (NH3), and VOCs can lead to atmospheric fine particulate matter less than 2.5 m in diameter (PM2.5) (e.g., Mozurkewich, 1993;Jimenez et al., 2009;Saxena and Seigneur, 1987;Brock et al., 2002). PM2.5 concentrations were estimated to account for nearly 3 million deaths worldwide in 2017 (Stanaway et al., 2018), while surface O3 concentrations were associated with nearly 500,000 deaths in 2017 (Stanaway et al., 2018) and significant global crop losses, valued at $11 billion (USD2000) in 2000(Avnery et al., 2011Ainsworth, 2017). 60 In addition, atmospheric O3 and aerosol both impact Earth's radiative budget (e.g., Bond et al., 2013;Haywood and Boucher, 2000;US EPA, 2018). Other pollutants, including carbon monoxide (CO), NO2, and SO2 are also directly hazardous to human health (US EPA, 2018), while NO2 and SO2 can additionally contribute to acid rain (Saxena and Seigneur, 1987;US EPA, 2018) and indirectly impact human health via their contributions to secondary PM2.5 formation. In addition, NH3 deposition and nitrification can also cause nutrient imbalances and eutrophication in terrestrial and marine ecosystems (e.g., Behera et al., 65 2013;Stevens et al., 2004). While these reactive gases and aerosol have both anthropogenic and natural sources, dominant global sources of NOx (= NO + NO2), SO2, CO, and VOCs include fuel transformation and use in the energy sector, industrial activities, and on-road and off-road transportation . Global NH3 emissions are predominantly from agricultural activities such as animal husbandry and fertilizer application (e.g., Behera et al., 2013) and OC and BC have large contributions from incomplete or uncontrolled combustion in residential and commercial settings (e.g., Bond et al., 2013). 70 Emissions of these compounds and the distribution of their chemical products vary spatially and temporally, with atmospheric lifetimes that allow for their transport across political boundaries, continuously driving changes in the composition of the global atmosphere.
Global emission inventories of these major atmospheric pollutants, with both sectoral, and fuel-type information are paramount for 1) understanding the range of emission impacts on the environment and human health and 2) for developing 75 effective strategies for pollution mitigation. For example, spatially gridded emission inventories are used as inputs in general circulation/climate (GCM) and chemical transport models (CTM), which are used to predict the evolution of atmospheric constituents over space and time. By perturbing emission sources or historical emission trends, such models can quantify the impact of emissions on the environment, economy, and human health (e.g., Mauzerall et al., 2005;Lelieveld et al., 2019;IPCC, 2013;Liang et al., 2018;Lacey and Henze, 2015), provide mitigation-relevant information for polluted regions (e.g., GBD 80 MAPS Working Group, 2016RAQC, 2019;Lacey et al., 2017), and anchor future projections (e.g., Shindell and Smith, 2019;Venkataraman et al., 2018;Gidden et al., 2019;Mickley et al., 2004). economic, health, and environmental impacts, and mitigate future burdens, computational models require emission inventories with regionally accurate estimates, global coverage, and the most up-to-date information possible. Though global bottom-up inventories can lag in time due to data collection and reporting requirements, the incorporation of smaller regional inventories provides the opportunity to improve the timeliness and regional accuracy of global estimates. 120 To further increase the policy-relevance of such data, it is also important that global emission inventories not only provide contemporary estimates, but report emissions as a function of detailed source sector and fuel type. For example, the recent air quality policies in China have included emission reductions targeting coal-fired power plants within the larger energy generation sector (e.g., Zheng et al., 2018). Decisions to implement such policies require accurate predictions of the air quality benefits, which in turn depend on simulations that use accurate estimates of contemporary sector-and fuel-specific emissions. 125 While the EDGAR, ECLIPSE, and CEDS inventories all provide varying degrees of sectoral information (Table 1), there are no global inventories to-date that provide public datasets of multiple atmospheric pollutants with both detailed source sector and fuel-type information. Crippa et al. (2019) do describe estimates of biofuel use from the residential sector in Europe using emissions from the EDGARv4.3.2 inventory (EC-JRC, 2018), but do not report global estimates or regional emissions from other fuel-types. Similarly, Hoesly et al. (2018) describe fuel-specific activity data and emission factors used to develop the 130 global CEDSv2016-07-26 inventory, but do not publicly report final global emissions as a function of fuel-type. In contrast, a limited number of regional inventories have provided both fuel-and sector-specific emissions. These inventories, for example, have been applied to earth system models to attribute the mortality associated with outdoor air pollution to dominant sources of ambient PM2.5 mass, such as residential biofuel combustion in India and coal combustion in China (GBD MAPS Working Group, 2018Group, , 2016. As countries undergo rapid changes that impact fluxes of their emitted pollutants, including population, 135 emission capture technologies, and the mix of fuels used, fuel and source-specific estimates are vital for capturing these contemporary changes and understanding the air quality impacts across multiple scales. As part of the Global Burden of Disease -Major Air Pollution Sources (GBD-MAPS) project, which aims to quantify the disease burden associated with dominant country-specific sources of ambient PM2.5 mass (https://sites.wustl.edu/acag/datasets/gbd-maps/), we have updated and utilized the CEDS open source emissions system to 140 produce a new global anthropogenic emission inventory (CEDSGBD-MAPS). CEDSGBD-MAPS includes country-level and global gridded (0.50.5) emissions of seven major atmospheric pollutants (NOx (as NO2), CO, NH3, SO2, NMVOCs, BC, OC) as a function of 11 detailed emission source sectors (agriculture, energy generation, industry, on-road transportation, non-road/offroad transportation, residential energy combustion, commercial combustion, other combustion, solvent use, waste, and international shipping) and four fuel groups (emissions from the combustion of total coal, solid biofuel, liquid fuels and natural 145 gas, plus all remaining process-level emissions) for the time period between 1970 -2017. Similar to the prior CEDS inventory released for CMIP6 , CEDSGBD-MAPS provides surface level emissions from all sectors, including fertilized soils, but does not include emissions from open burning. In the first two sections we provide an overview of the CEDSGBD-MAPS system and describe the updates that have allowed for the extension to the year 2017 and the added fuel-type information. These include updates to the underlying activity data and input emission inventories used for default estimates and scaling 150 procedures (including the use of two new inventories from Africa and India), the additional scaling of default BC and OC emissions, as well as the use of updated spatial gridding proxies, and adjustments to the final gridding and aggregation steps that retain detailed sub-sector and fuel-type information. The third section presents global CEDSGBD-MAPS emissions in 2017 and discusses historical trends as a function of compound, sector, fuel-type, and world region. The final section provides a comparison of the global CEDSGBD-MAPS emissions with other global inventories, as well as a discussion of the magnitude and 155 sources of uncertainty associated with the CEDSGBD-MAPS products.

Methods
The December 23, 2019 full release of the Community Emissions Data System ) provides the core system framework for the development of the contemporary CEDSGBD-MAPS inventory. The CEDSGBD-MAPS inventory is developed for the GBD-MAPS project and is not an updated release of the core CEDS emissions inventory. As detailed in Hoesly et al. 160 (2018), the original version of the CEDS system was used to produce the first CEDSv2016-07-26 inventory (hereafter called CEDSHoesly) (CEDS, 2017a, b), which provides global gridded (0.50.5) emissions of atmospheric reactive gases (NOx (as

Overview of CEDSGBD-MAPS System
The CEDS system has five key procedural steps, illustrated in Fig. 1. After the collection of input data in Step 0, Step 1 calculates default global emission estimates (Em) for each chemical compound using a bottom-up approach shown in Eq. 175 (1). In Eq. (1), emissions are calculated using relevant activity (A) and emission factor (EF) data for each country (c) and year (y), as a function of 52 detailed working sectors (s) (sub-sectors used for intermediate steps in the CEDS system) and nine working fuel-types (f) ( Table 2). CEDS conducts these calculations for two types of emission categories: 1) fuel combustion sources (e.g., electricity production, industrial machinery, on-road transportation, etc.) and 2) process sources (e.g., metal production, chemical industry, manure management, etc.). We note that the distinction between these source categories is 180 reflective of both sector definition and CEDS methodology, as described further in Sect. S2.1. This results in some working sectors that include emissions from combustion, such as waste incineration and fugitive petroleum and gas emissions, to be characterized in the CEDS system as process-level sources (further details in Sect. S2.1). In contrast to CEDS combustion source emissions, which are calculated in Eq. (1) as a function of 8 fuel types, emissions from CEDS process-level sources are combined into a single 'process' category, as described in Sect. 2.4.
For emissions from CEDS combustion sources, annual activity drivers in Eq. (1) primarily include country-, fuel-, and sector-specific energy consumption data from the International Energy Agency (IEA, 2019). Sector-and compoundspecific emission factors are typically derived from energy use and total emissions reported from other inventories, including 190 from the GAINS model (Klimont et al., 2017;IIASA, 2014;Amann et al., 2015), Speciated Pollutant Emission Wizard (SPEW) (Bond et al., 2007), and the US National Emissions Inventory (NEI) (NEI, 2013). For International Shipping, IEA activity data is supplemented with consumption data and EFs from the International Maritime Organization (IMO), as described in Hoesly et al. (2018) and its supplement. In contrast, default emissions (Em) for CEDS process sources are directly taken from other inventories, including from the EDGAR v4.3.2 global emission inventory Crippa et al., 2018). "Implied 195 emission factors" are then calculated for these process sources in Eq. (1) using global population data (UN, 2019(UN, , 2018 or pulp and paper consumption (FAOSTAT, 2015) as the primary activity drivers. For years without available emissions, default estimates for CEDS process sources are calculated in Eq. (1) from a linear interpolation of the "implied emission factors" and available activity data (A) for that year. Supplemental Sect. S2.1 and S2.2 provide additional details regarding the input datasets for activity drivers and emission factors used for both CEDS combustion and process source categories. 200 While CEDS Step 1 is designed to provide a complete set of historical emission estimates, CEDS Step 2 scales these total default emission estimates to existing, authoritative global, regional, and national-level inventories. As described in Hoesly et al. (2018), CEDS uses a "mosaic" scaling approach to retain detailed fuel-and sector-specific information across different inventories, while maintaining consistent methodology over space and time. The development and use of mosaic inventories has been recently increasing as they provide a means to utilize detailed local emissions, while harmonizing this 205 information across large regional or global scales (Li et al., 2017c;Janssens-Maenhout et al., 2015). The CEDS approach, however, differs from previous mosaic inventories, such as that developed for the HTAP project , as local and regional inventories in CEDSGBD-MAPS are used to scale sectoral emissions at the national-level, rather than merging together spatially distributed gridded estimates.
The first step in the scaling procedure is to derive a time series of scaling factors (SF) for each scaling inventory using 210 Eq.
(2), calculated as a function of chemical compound, country, sector, and fuel-type (where available). Due to persistent differences and uncertainties in the underlying activity data and sectoral definitions in each scaling inventory, CEDS emissions are scaled to total emissions within aggregate scaling sectors (and fuels, where applicable). These aggregate scaling groups are defined for each scaling inventory and are chosen to be broad in order to improve the overlap between CEDS emission estimates and those reported in other inventories. For example, the sum of CEDS emissions from working sectors 215 1A4a_Commercial-institutional, 1A4b_Residential, and 1A4c_agriculture-forestry-fishing are scaled to the aggregate After SFs are calculated in Eq.
(2), the second step in the scaling procedure is to extend these SFs forward and 220 backward in time to fill years with missing data. For these time periods, the nearest available SF is applied. If a particular sector or compound is not present in a scaling inventory, default CEDS estimates are not scaled. For BC and OC emissions, the default procedure in the CEDSv2019-12-23 system was to retain all default BC and OC emission estimates due to limited availability of historical BC and OC emissions. In the CEDSGBD-MAPS inventory, these species are now scaled to available regional and national-level inventories (further details in Sect. 2.2). For all other species, the CEDSGBD-MAPS system uses a 225 sequential scaling methodology where total default emissions for each country are first scaled to available global inventories (primarily EDGAR v4.3.2) and second scaled to regional and national-level inventories, many of which have been updated in this work (Sect. 2.2 and Table 3). This process results in final CEDSGBD-MAPS emissions that reflect the inventory last used to scale the emissions for that country (Fig. 2). Figure S2 provides a time series of implied emission factors after the scaling procedure for select sector-and fuel-combinations that dominant emissions of each compound in the top 15 emitting countries. 230 Sections 2.2 and S2.3 describe further details and updates to this scaling procedure.

CEDS
Step 3 extends the scaled emission estimates from 1970 back in time to 1750. This process is necessary as reported emission estimates and energy data are not typically reported with the same level of sectoral and fuel-type detail prior to 1970. Hoesly et al. (2018) provides a detailed description of this historical extension procedure, which is used to derive pre-1970 emissions in the CEDSHoesly inventory. The new CEDSGBD-MAPS inventory only reports more contemporary emissions 235 after 1970 and therefore, does not utilize this historical extension.

CEDS
Step 4 aggregates the scaled country-level CEDSGBD-MAPS emissions into 17 intermediate gridding sectors (defined in Table 2). In the CEDSv2019-12-23 system, Step 4 additionally aggregated sectoral emissions from all fuel-types.
In contrast, the CEDSGBD-MAPS system retains sectoral emissions from the combustion of total coal (hard coal + coal coke + brown coal), solid biofuel, and the sum of liquid oil (light oil + heavy oil + diesel oil) and natural gas, as well as all CEDS 240 process-level emissions ( Table 2). Sections 2.4 and 4.2.4 describe the CEDSGBD-MAPS fuel-specific emissions in further detail.

Lastly, CEDS
Step 5 uses normalized spatial distribution proxies to allocate annual country-level emission estimates on to a 0.50.5 global grid. Annual emissions from the 17 intermediate gridding sectors and four fuel groups are first distributed spatially using compound-, sector-, and year-specific spatial proxies, primarily from the gridded EDGAR v4.3.2 inventory. Supplemental Table S7 provides a complete list of sector-specific gridding proxies, with additional details specific 245 to the CEDSGBD-MAPS system in Sect. S2.5 and about the general CEDS gridding procedure in Feng et al. (2020). Second, gridded emission fluxes (units: kg m -2 s -1 ) are aggregated into 11 final sectors (Table 2) and distributed over 12 months using sectoral and spatially explicit monthly fractions from the ECLIPSE project (IIASA, 2015) and EDGAR inventory (international shipping only). Relative to CEDSv2019-12-23, the new CEDSGBD-MAPS inventory retains detailed sub-sector emissions from the aggregate RCO (now RCO-Residential, RC-Commercial, and RCO-Other) and TRA (now On-Road and Non-Road) 250 sectors, separate sectoral emissions from process sources, as well as combustion sources that utilize coal, solid biofuel, and the sum of liquid fuels and natural gas. Table 2 contains a complete breakdown of the definitions of CEDS working, intermediate gridding, and final sectors. Gridded total NMVOCs are additionally disaggregated into 25 VOC classes following sector-and country-specific VOC speciation maps from the RETRO project (HTAP2, 2013), which are different from those used in the recent EDGARv4.3.2 inventory (Huang et al., 2017). Similar to the gridding procedure, the same VOC speciation 255 and monthly distributions are applied to sectoral emissions associated with each fuel category.
Final products from the CEDSGBD-MAPS system include total annual emissions from 1970 -2017 for each country, as well as monthly global gridded (0.50.5) emission fluxes, both as a function of 11 final source sectors and four fuelcategories (total coal, solid biofuel, liquid fuel + natural gas, and remaining process sources). Section 5 provides additional details on the dataset availability and file formats. 260

Default Emission Scaling Procedure -CEDSGBD-MAPS Update Details
As described above, default emission estimates for each compound are scaled in CEDS Step 2 to existing authoritative inventories as a function of emission sector and fuel type (where available). In the scaling procedure, annual emissions and EFs for each country are first scaled to available global inventories, then to available regional and national-level inventories, assuming that the latter use local knowledge to derive more accurate regional estimates. Final CEDSGBD-MAPS emission totals 265 for each country therefore reflect the inventory last used to scale each compound and sector. Many of these inventories are updated annually and where available, have been updated in this work relative to the CEDSv2019-12-23 system (Table 3). For example, global CEDSGBD-MAPS combustion source emissions of NOx, total NMVOCs, CO, and NH3 are first scaled to EDGAR v4.3.2 country-level emissions as a means to incorporate additional country-specific information relative to default estimates derived using more regionally-aggregate EFs from GAINS. CEDSGBD-MAPS emissions from European countries are then scaled  (2018). Relative to the CEDSv2019-23-13 system, regional inventories have also been added to scale CEDSGBD-MAPS emissions from India and Africa as described below. Updates to additional regional scaling inventories, including South Korea, Japan, and other European and Asian countries are not available relative to those used in the CEDSv2019-12-23 system. Table 3 provides a complete list of the inventories used to scale CEDSGBD-MAPS default emissions, with additional details in Sect. S2.3. 280 Relative to the CEDSv2019-12-23 system, the CEDSGBD-MAPS system adds scaling inventories for two rapidly changing regions, Africa and India. First, CEDSGBD-MAPS emissions from Africa for select sectors are now scaled to the Diffuse and Inefficient Combustion Emissions in Africa (DICE-Africa) inventory from Marais and Wiedinmyer (2016). This inventory provides gridded (0.1  0.1) emissions for NOx (= NO + NO2), SO2, 25 speciated VOCs, NH3, CO, BC, and OC for 2006 and 2013 for select anthropogenic sectors and fuels. In this work, default CEDS emissions are scaled to total DICE-Africa 285 emissions from each country and later re-gridded in CEDS Step 5 using source-specific spatial proxies described in Sect. 2.1.
Following the CEDSv2019-12-23 scaling procedure (Supplemental Sect. S2.3), a set of aggregate scaling sectors and fuels are defined to ensure that CEDSGBD-MAPS emissions are scaled to emissions from consistent sectors and fuel types within the DICE-Africa inventory (Table S3). Briefly, CEDSGBD-MAPS 1A3b_Road and 1A4b_Residential emissions are scaled to DICE-Africa emissions from diesel and gasoline powered cars and motorcycles, as well as biomass and oil combustion associated with 290 residential charcoal, crop residue, fuelwood, and kerosene use. The DICE-Africa inventory also includes emission estimates from gas flares across Africa and ad-hoc oil refining in the Niger Delta, fuelwood use for charcoal production and other commercial enterprises, and gas and diesel use in residential generators. Marais and Wiedinmyer (2016) state that these particular sources are missing or not adequately captured in existing global inventories. Therefore, depending on the source sector and inventory details, they recommend that these emissions be added to existing global inventories for formal industry 295 and on-grid energy production in Africa (DICE-Africa, 2016). Due to uncertainties in the representation of these sectors in the default CEDS Africa emissions, these sources are not included in the scaling process here. Default CEDSGBD-MAPS emissions from the 1B2_fugitive_pert_gas (gas flaring) sector (derived from the ECLIPSE and EDGAR inventories) are larger than DICE-Africa gas flaring emissions in 2013, suggesting that this source may be accurately represented in the default CEDSGBD-MAPS estimates. As described in Sect. S2.3.2, however, residential generator and fuelwood use for charcoal production and 300 other commercial activities are not explicitly represented in CEDS and will be accounted for only to the extent that these sources are included in the underlying IEA activity data and EDGAR process emission estimates. In the event that the DICE-Africa emissions from these sources are missing in the default CEDS estimates, total 2013 CEDSGBD-MAPS emissions from Africa for each compound may be underestimated by up to 11% (Sect. S2.3, Table S5). These values range from 0.7% for SO2 to 11% for CO (Table S5)  (SMoG-India, 2019). Similar to DICE-Africa emissions, the final spatial distribution in the SMoG-India and CEDSGBD-MAPS inventories will differ as country-level emissions are scaled to country totals and spatially re-allocated using CEDS proxies in Step 5. SMoG-India emissions for each compound are available for 17 sectors and nine fuel types (coal, fuel oil, diesel, gasoline, kerosene, naptha, gas, biomass, process/fugitive). Similar to the DICE-Africa inventory, aggregate scaling groups have been defined to scale consistent sectors and fuels between inventories, as described in Sect. S2.3. Briefly default 315 CEDSGBD-MAPS emissions for 1A4c_Agriculture-forestry-fishing sector are scaled to the sum of SMoG-India emissions for agricultural pumps and tractors, 1A4b_Residential emissions are scaled to the sum of SMoG-India emissions from residential lighting, cooking, diesel generator use, space and water heating, 1A1a electricity and heat generation sectors are scaled to SMoG-India thermal power plant emissions, 1A3b road and rail sectors are scaled to the respective SMoG-India road and rail emissions, and CEDSGBD-MAPS industrial working sectors are allocated and scaled to four SMoG-India industrial sectors: light 320 industry (e.g., mining and chemical production), heavy industry (e.g., iron and steel production), informal industry (e.g., food production), and brick production. Calculated scaling factors for these sectors are held constant before and after 2015.  Figure S2 shows that after scaling, the implied emission factors of CO from oil and gas combustion in the on-road transport sector for four African countries range from 0.19-0.28 g g -1 , slightly smaller than the range of 0.029 -0.380 g g -1 used in the DICE-Africa inventory. Emissions from the residential/commercial sectors in Africa are generally lower in CEDS GBD-MAPS than in CEDS Hoesly due to both lower biofuel consumption and a lower assumed EF in the DICE-Africa inventory (Marais and Wiedinmyer, 2016). Residential BC and OC emission estimates are 340 also lower than those from GAINS (Klimont et al., 2017). The difference in biofuel consumption is due to different data sources. The DICE-Africa inventory uses residential wood fuel consumption estimates from the UN while CEDSHoesly uses data from the IEA. Both of these sources consist largely of estimates for African countries because there is little countryreported biofuel consumption data available. The estimation methodologies for both the UN and IEA estimates are not well documented, which adds to the uncertainty in these values (Sect. 4.2). After scaling, the implied EFs for residential biofuel 345 emissions of OC are ~0.001-0.002 g g -1 in three African countries ( Figure S2), within the range of EFs of 0.0007 -0.003 g g -1 implemented in the DICE-Africa inventory. Total CEDSGBD-MAPS emissions of NMVOCs are larger, primarily due to increased contributions from solvent use and the energy sector associated with changes in the EDGAR v4.3.2 inventory, while total emissions of CO, SO2, and NH3 are relatively consistent between the two CEDS versions.
For the India comparison, the right panel of Fig. 3 shows that total emissions of NOx, CO, SO2, NMVOCs, and OC 350 are lower in CEDSGBD-MAPS. Relative reductions in NOx emissions are largely associated with on-road transport. Scaled CEDSGBD-MAPS transport emissions are 5 Tg smaller than NOx emissions in CEDSHoesly, largely as a result of lower fuel consumption levels for gas, diesel, and CNG on-road vehicles used to develop SMoG-India estimates (Sadavarte and Venkataraman, 2014). Figure S2 shows that the implied emission factor for NOx emissions from oil & gas combustion in the on-road transport sector in India is ~0.015 g g -1 in 2015, which falls within the range of values of 0.0026 -0.046 g g -1 used for 355 various vehicles and fuel type in Venkataraman et al. (2018). Similarly, NOx transport emissions are also lower in CEDSGBD-MAPS relative to the EDGAR and GAINS inventories. Causes of other reductions are mixed. For example, lower emissions of SO2 and NMVOCs are largely associated with the energy sector, while reductions in the industry sector contribute to reduced CO emissions. For SO2, Figure S2 shows that the implied EF for coal combustion in the energy sector is ~0.004 g g -1 , slightly lower than the range of 0.0049 -0.0073 g g -1 used for the SMoG-India inventory. 360 To further examine the CEDSGBD-MAPS inventory in these regions, Fig. 4 compares final CEDSGBD-MAPS and CEDSHoesly emissions for India and Africa to total emissions from two widely used global inventories: GAINS (ECLIPSE v5a) and EDGAR (v4.3.2). First, Fig. 4 shows the percent difference between the CEDSGBD-MAPS inventory and the GAINS and EDGAR inventories on the y-axis, against the percent difference between the CEDSHoesly inventory and GAINS and EDGAR emissions on the x-axis. Percent differences are calculated from total emissions from Africa (left) and India (right) for the year 2012 for 365 the comparison with EDGAR and for 2010 for the comparison to GAINS (most recent years with overlapping data). The green shaded areas indicate regions where the updated CEDSGBD-MAPS inventory has improved agreement with EDGAR or GAINS relative to the CEDSHoesly inventory. This comparison shows that the additional scaling of CEDSGBD-MAPS emissions to the SMoG-India inventory generally improves agreement with both the EDGAR and GAINS inventories relative to CEDSHoesly for all species except black carbon (BC). Scaling to the DICE-Africa inventory generally improves CEDSGBD-MAPS agreement 370 with the EDGAR inventory but not with GAINS (except for OC). Further comparisons to these two inventories are discussed in Sect. 4. While uncertainties in emissions from these inventories are expected to be at least 20% for each compound (discussed in Sect. 3.3), this comparison provides an illustration of the changes between the two CEDS versions relative to two widely used global inventories.

Default BC & OC Scaling Procedure -CEDSGBD-MAPS Update Details 375
Relative to the CEDSv2019-12-23 system, the second largest change to the CEDSGBD-MAPS system is the added scaling of BC and OC emissions in CEDS Step 2. In the v2019-12-23 system, OC and BC were not scaled due to a lack of historical BC and OC emission estimates in regional and global inventories. Due to the focus of the CEDSGBD-MAPS inventory on more recent years, these two compounds are now scaled to available regional and country-level estimates (Table 3), following the same scaling procedure described above for the reactive gases. Unlike the reactive gases, however, BC and OC emissions are not 380 scaled to the global EDGAR v4.3.2 inventory due to the large reported uncertainties in this inventory (ranging from 46.8% to 153.2% ).
To examine the impact of the new BC and OC emissions scaling, in addition to the updated IEA energy consumption data, Fig TgC (16 Tg organic aerosol/ organic mass: organic carbon ratio of 1.1 -1.4), respectively (Bond et al., 2013). These also have 395 improved agreement with the CEDSGBD-MAPS estimates of BC and OC in 2000 relative to those in the CEDSHoesly inventory.
Lastly, we note plans for an upcoming update to the core CEDS system to improve historical trends in carbonaceous aerosol by incorporating reported inventory values for total PM2.5 and its ratio with BC and OC emissions.

Fuel Specific Emissions -CEDSGBD-MAPS Update Details
Prior to gridding, CEDSGBD-MAPS Step 4 combines total country-level emissions for each of the 52 working sectors and nine 400 fuel groups into 17 aggregate sectors and 4 fuel-groups: total coal (hard coal + brown coal + coal coke), solid biofuel, the sum of liquid fuels (heavy oil + light oil + diesel oil) and natural gas, and all remaining 'process' emissions (Table 2). In contrast, the CEDSv2019-12-23 system aggregates all fuel-specific emissions and reports inventory values as a function of sector only.
In CEDSGBD-MAPS, country-total emissions from these aggregate sectors and fuel groups are distributed across a 0.50.5 global grid using spatial gridding proxies, as discussed in Sect. 2.1 (Table S7). During gridding, the same spatial proxies are 405 applied to all fuel groups within each sector. In practice, this requires that the gridding procedure be repeated four times for each of the fuel groups. After gridding in CEDS Step 5, both annual country-total and gridded emission fluxes from each fuel group are aggregated to 11 final sectors. Figure S5 demonstrates the level of detail available in the new CEDSGBD-MAPS gridded emission inventory by illustrating global BC emissions in 2017 from 1) all source sectors, 2) the residential sector only, 3) residential biofuel-use only, and 4) residential coal-use only. Additional uncertainties associated with the CEDSGBD-MAPS fuel-410 specific emissions in both the country-total and annual gridded products are discussed further in Sect. 4.2.4

Results
The new CEDSGBD-MAPS inventory provides global emissions of NOx, SO2, NMVOCs, NH3, CO, OC, and BC for 11 anthropogenic sectors (agriculture, energy, industry, on-road, non-road transportation, residential, commercial, other, waste, solvents, international shipping) and four fuel groups (combustion of total coal, solid biofuel, and liquid fuels and natural gas, 415 and process sources) over the time period between 1970 -2017. Final country-level emissions are provided as annual time series in units of kilotons per year (kt yr -1 ) for each sector and fuel-type and include NOx as emissions of NO2. Final global gridded (0.5  0.5) emissions for each compound, sector, and fuel group have been converted to emission fluxes (kg m -2 s -2 ), distributed over 12 months, and represent NOx as NO to facilitate use in earth system models. Total NMVOCs in gridded products are additionally separated into 25 sub-VOC classes. Using a combination of updated energy consumption data and 420 scaling procedures, CEDSGBD-MAPS provides the most contemporary bottom-up global emission inventory to-date, and is the first inventory to report global emissions of multiple atmospheric pollutants from multiple fuel groups and sectors using consistent methodology. The following results section presents an overview of the CEDSGBD-MAPS emission inventory, with particular focus on emissions in 2017 and historical trends as a function of compound, sector, fuel type, and world region.
Section 4 compares these results to other global emission inventories and discusses the magnitudes and sources of inventory 425 uncertainties. Known issues in the inventory data at the time of submission are detailed in Sect. S4.  Table S8 show that 60% of NO x emissions are associated with the energy generation (22%), industry (15%), and on-road transportation (23%) sectors. These sectors have the largest contributions from emissions from coal combustion (> 46% for the energy and industry emissions) and the combined 440 combustion of liquid fuels (oil) and natural gas (with these two fuels accounting for 100% of NOx on-road emissions). Time series of regional contributions to global emissions in Fig. 8 additionally show that 50% of global 2017 NOx emissions are from the combined Other Asia/Pacific region (Table S9) (13 Tg), China (24 Tg), International Shipping (25 Tg). For global (86%: biofuel) sectors, 78% of SO2 emissions are from the energy generation (63%: coal) and industry (38% coal, 36% process, 445 25% oil + gas) sectors, 89% of NH3 emissions are from the agriculture (100%: process) and waste (100%: process) sectors, and emissions of NMVOCs have the largest single contribution (36%) from the energy sector, 99% of which are associated with CEDSGBD-MAPS process sources (Table 2). For carbonaceous aerosol in 2017, 58% of global BC emissions are from the residential (70%: biofuel) and on-road (100%: oil + gas) sectors, while 67% of global OC emissions are from the residential (92%: biofuel) and waste (100%: process) sectors. Fig. 8 shows that in 2017, China is the dominant source of global CO (144  450 Tg, 27% of global total), SO2 (12 Tg, 15% of global total), NH3 (12 Tg, 20% of global total), OC (2.7 TgC, 20% of global total), and BC (1.4 TgC, 24% of global total). In contrast, Africa is the dominant source of global NMVOCs in 2017 (48 TgC, 27% of global total) and International Shipping is the dominant source of global NOx emissions (25 Tg, 20% of global total).

Global Annual Total Emissions in 2017
As discussed above in Sect. 2 and below in Sect. 4.2.4, the distinction between CEDS combustion and process-level source categories for all species may result in the underrepresentation of emissions from combustion sources relative to those 455 from CEDS process-level sectors. As shown in Table 2, for example, some combustion emissions from the energy, industry, and waste sectors, such as fossil fuel fires and waste incineration are categorized as CEDS 'process-level' source categories (Table 2). These emissions are allocated to the final CEDS process category rather than the CEDS total coal, biofuel, or oil and gas categories.  Fig. S9-S12). Global CO emissions then decrease by 9% between 2012 and 2017, largely due to reductions in industrial coal, residential biofuel, and process energy sector emissions in China (S6, S9, S17-S18), associated with the implementation 470 of emission control strategies (reviewed in Zheng et al., 2018), as well as continued reductions in on-road transport emissions in North America and Europe (Fig. S7-S8). Similarly, global SO2 emissions decrease after peaking in 1979, largely due to emission control policies in the energy and industry sectors in North America and Europe (Fig. S7-S8). While simultaneous increases in emissions from coal use in the energy and industry sectors in China result in a brief increase in global SO2 emissions between 1999 and 2004 (Fig. 6, S9), global SO2 emissions decline by 32% between 2004 and 2017 due to the 475 implementation of stricter emission standards for the energy and industry sectors after 2010 in China (Zheng et al., 2018), as well as continued reductions in North America and Europe (Fig. S7-S8). Regional SO2 emission trends are particularly large with a factor of 9.5 decrease in total SO2 emissions in North America between 1973 and 2017, a factor of 6.9 decrease in Europe between 1979 and 2017, and a factor of 5.9 increase in China between 1970 and 2004, followed by a factor of 2.6 decrease after 2011 (Fig. 8). While China is the largest global contributor to SO2 emissions between 1994 and 2017, these 480 large regional reductions, coupled with increasing SO2 emissions in the Other Asia/Pacific region, African countries, and India ( Fig. 8), indicate that future global SO2 emissions will increasingly reflect activities in these other rapidly growing regions.

Historical Trends in Annual Global Emissions
In contrast to historical emissions of SO2 and CO, global emissions of NOx, BC, and OC peak later between 2011 and 2013. Global emissions then decrease by 7%, 9%, and 7%, respectively by 2017 (Fig. 6). These trends also reflect the sectorspecific regulations implemented in dominant source regions. For NOx for example, global emissions between 1970 and 2017 485 are dominated by the combustion of coal, oil, and gas in the on-road transportation, energy generation, industry, and international shipping sectors (Fig. 6, 8) (Zheng et al., 2018;Liu et al., 2015). Global emissions of NOx from waste combustion and agricultural activities also increased by a factor of 2 and 65%, respectively, between 1970 and 2017, also contributing to the offset of recent reductions in emissions from regulated combustion sources (Fig. 6). Similar to global NOx emissions, trends in historical BC and OC emissions reflect a balance between emission trends in North America, Europe and other world regions, with reduction between 2010 and 2017 largely 500 driven by reductions in emissions from China (Fig. 8, S9). In contrast to NOx emissions, however, BC and OC emissions are dominated by contributions from biofuel combustion in the residential sector, as well as on-road transportation, industry, and energy sectors for BC and the waste sector for global OC (Fig. 6). Though emissions of BC and OC have a higher level of uncertainty relative to other compounds (Sect. 4), emissions from African countries and the Other Asia/Pacific region experience growth in BC and OC emissions from these sectors. The exceptions are in China and India, both of which 505 experience a plateau or reduction in BC and OC emissions from the residential, energy (China only), industry, and on-road transportation sectors between 2010 and 2017. In India, reductions in BC and OC emissions from the residential and informal industry sectors are expected to continue under policies to switch to cleaner residential fuels and energy sources, while BC emissions from on-road transport may increase due to increased transport demand (Venkataraman et al., 2018). Similar to trends in SO2 emissions, increasing trends in total OC and BC emissions from Africa, India, Latin America, the Middle East, 510 and the Other Asia/Pacific region, coupled with large decreases in emissions from China, North America, and Europe (Fig. 8) indicate that global emissions will increasingly reflect activities in these rapidly growing regions.
Trends in historical emissions of NMVOCs and NH3 differ from other pollutants in that they continuously increase between 1970 and 2017. Global emissions of NH3 increase by 81% between 1970 and 2017 and are largely associated with emissions from agricultural practices (75% in 2017) and waste disposal and handling (14% in 2017) (Fig. 6, Table S8). Unlike 515 emissions from combustion sources, there are no largescale regulations outside of Europe targeting NH3 emissions from agricultural activities, such as livestock manure management. As a result, global agricultural emissions of NH3 increase between 1970 and 2017 by 82%, driven by increases in all regions other than Europe (Fig. 6, S6-S12). Similarly, global NH3 emissions from the waste sector increase by 77% between 1970 and 2017, driven by increases in Latin America, the Other Asia/Pacific region, Africa, and India ( Fig. S6-S12). Global emissions of NMVOCs increase by 40% between 1970 and 2017 520 and are largely associated with emissions from the on-road transport, residential, energy, industry, and solvent use sectors (Fig.   6). In contrast to other emitted pollutants, Africa is the largest global source of NMVOC emissions between 2010 and 2017, largely due to large contributions and continued increases in emissions from the residential (factor of 2.7) and energy (factor To provide a fuel-centric perspective of global historical emissions trends, Fig. 7 illustrates the contributions from the combustion of coal, solid biofuel, the sum of liquid fuel and natural gas, as well as all remaining CEDS 'process-level' sources (Table 2) to total global emissions between 1970 and 2017. Reductions discussed above between 2010 and 2017 for 535 global emissions of NOx, CO, SO2, BC and OC, are largely associated with reductions in coal combustion from the energy, industry, and residential sectors associated with emission control policies and residential fuel replacement in China, as well as coal-fired power plant reductions in North America and Europe (Fig. 7, S13, S17). Despite large reductions in emissions, China is still the single largest source of global emissions from coal combustion in 2017 (23-64% for each compound except NH3). Figure S17, however, also shows that emissions from coal combustion are simultaneously increasing in India, the Other 540  (Fig. 3) and combined, account for ~60% of the reduction in global NOx emissions, 23% of the reduction in global CO, and 14% of the reduction in global SO2.
The largest differences between these two inventories in India and Africa are the reduced NO x emissions from the transport sector, as well as reduced energy emissions of SO2 in India. Remaining differences between NOx and SO2 emissions in the 575 two CEDS inventories are largely associated with the updated China emission inventory from Zheng et al. (2018), which reports lower emissions in 2010 and 2012 than a previous version of the MEIC inventory that was used to scale China emissions in the CEDSHoesly inventory (Li et al., 2017c). These emission reductions are largely associated with the industrial and residential sectors in China and are partially offset by a simultaneous increase in transportation emissions of all compounds relative to CEDSHoesly. 580 For global emissions of NH3 and NMVOCs, these species remain relatively unchanged between the CEDSHoesly and CEDSGBD-MAPS inventories. In 2014 CEDSGBD-MAPS emissions are 5% higher than CEDSHoesly emissions for NMVOCs and 2% lower than CEDSHoesly global NH3 emissions. Emissions of NH3 remain relatively unchanged (within <2%) from dominant source regions, including India, Africa (Fig. 3), and China. In contrast, emissions of NMVOCs from Africa and China in the DICE-Africa and Zheng et al. (2018) scaling inventories are larger than those in the CEDSHoesly inventory. Global emissions 585 of NMVOCs are also higher in EDGARv4.3.2 inventory relative to the previous version used in the CEDSHoesly inventory.
NMVOCs are particularly large from the process energy sector emissions in Africa ( Figure S12), which primarily include fugitive emissions from oil and gas operations (Table 2). Default energy sector emissions from 'non-combustion' processes are taken from the EDGAR inventory and are not scaled to DICE-Africa inventory. Therefore, the large increase in these emissions in Africa relative to CEDSHoesly are largely driven by changes in the EDGAR v4.3.2 inventory, with emissions from 590 the 1B2_Fugitive_Fossil fuels sector, increasing for example by a factor of 5 in Nigeria between 2003 and 2017.
Global emissions of OC and BC have the largest differences between the two CEDS inventories, with CEDSGBD-MAPS emissions consistently smaller than CEDSHoesly emissions between 1970 and 2014. By 2014, CEDSGBD-MAPS emissions of BC and OC are 24 and 33% smaller than corresponding CEDSHoesly emissions. In the CEDSHoesly inventory, default emissions of BC and OC are not scaled and therefore these differences are largely associated with the added scaling inventories, discussed 595 in Sect. 2 and shown in Table 3. As shown in Fig. S3-S4, the added scaling of BC and OC emissions leads to a reduction in global CEDSGBD-MAPS emissions of OC in all scaled regions, and a reduction in BC emissions in all regions other than India.
In India, increases in industry and residential BC emissions from the SMoG-India scaling inventory result in a slight increase in BC emissions relative to the CEDSHoesly inventory (Fig. 3). Waste emissions of OC and BC are also reduced in the CEDSGBD-MAPS inventory due to updated assumptions for the fraction of waste burned (Sect. S1.1). As discussed in Hoesly et al. (2018)  Similar to the total global emissions, changes between the two CEDS versions for the national-level and 0.50.5 gridded products will also result from updates to the energy consumption data, scaling inventories (Section 2.2-2.3) and spatial 610 distribution proxies from EDGARv4.3.2 (Section 2.1). Time series of differences between the CEDSHoesly and CEDSGBD-MAPS inventories for 11 world regions are shown for each compound in Fig. S22. In recent years, Figure S22 shows that CEDSGBD-MAPS emissions are generally lower in each region, with the greatest differences in Africa, India and China. The relative changes in Africa and India are discussed previously in Section 2. For China, the CEDSGBD-MAPS emissions are generally lower than the CEDSHoesly estimates after the year 2010 as a result of the updated scaling inventory. Regional differences between inventories 615 are also greater for OC and BC emissions relative to other compounds due to the added scaling procedure discussed in Section 2. Differences in spatial distributions are not discussed here as changes represent differences in the spatial proxies, which are largely from updates to the EDGAR inventory.  Table S10.

Comparison to Other Global Inventories (EDGAR & GAINS)
The comparison in Fig. 6 shows that global emissions of all compounds in the CEDSGBD-MAPS inventory are 630 consistently larger than in the EDGAR v4.3.2 inventory . Global CEDS GBD-MAPS emissions of NO x , SO 2 , CO, and NMVOCs are at least 27% larger, while global emissions of NH3, BC, and OC are within 52%. Figure S23 indicates that differences in global BC and OC emissions are largely due to higher waste and residential and commercial emissions in the CEDS GBD-MAPS inventory. Figure 6, however also shows that the trends in global emissions are similar between EDGAR v4.3.2 and CEDSGBD-MAPS for most compounds. For example, between 1970 and 2012, global emissions of SO2, NH3, 635 NMVOCs, and BC peak in the same years. Global CO and NOx emissions both peak one year earlier in the CEDSGBD-MAPS inventory, but otherwise follow similar historical trends. Trends in OC emissions are the most different between the two inventories with a peak in emissions in 1988 in the EDGAR inventory, compared to 2012 in the CEDSGBD-MAPS inventory. A comparison of relative sectoral contributions in Fig. S23 shows that these differences in OC emissions are largely due to the residential and commercial sectors, which may be underestimated in the EDGAR v4.3.2 inventory relative to GAINS (Crippa 640 et al., 2018) and CEDSGBD-MAPS. Both inventories also show a net increase in global emissions of all compounds other than SO2 between 1970 and 2012. Global SO2 emissions follow a similar trend until 2007, after which, the emissions in CEDSGBD-MAPS decrease at a faster rate than in EDGAR v4.3.2. These differences are largely due to the energy sector, which increase between 2006 and 2012 in EDGAR, and decrease as a result of emission reductions in China in the CEDSGBD-MAPS inventory (Fig. S23). For all other compounds, the rate of increase in emissions between 1970 and 2012 is also slightly different between 645 the two inventories. For example, NH3 emissions in the CEDSGBD-MAPS inventory increase by 74% compared to a 139% increase in EDGAR. In contrast, BC and OC emissions increase at a faster rate in the CEDSGBD-MAPS inventory. Due to similar sources of uncertainty and the additional scaling of CEDSGBD-MAPS emissions to EDGAR (except for BC and OC), levels of uncertainty between the two inventories are expected to be similar, as discussed further in Sect. 4.2.
Similar to the comparison with EDGAR emissions, Fig. 6 also shows that global emissions in the CEDSGBD-MAPS 650 inventory are generally larger than emission estimates from the GAINS model, published as part of the ECLIPSE v5a inventory (referred to here as GAINS) (Klimont et al., 2017). Two exceptions are for SO2 emissions, which are up to 6% lower than GAINS in select years, and BC emissions, which are consistently 5-15% lower than GAINS for all years. While the sectoral definitions may slightly differ between these inventories, Fig. S24 shows that these differences are largely due to different  (Fig. 6). Sectoral contributions between the two inventories in Fig. S24 indicates that these differences are largely due differences in the energy, industry, and on-road transport emissions of NMVOCs. Uncertainties in the GAINS model have been previously estimated to fall between 10% and 30% in Europe for gas-phase species (Schöpp et al., 2005) and within the 665 uncertainty estimates for BC and OC of other global bottom-up inventories (Klimont et al., 2017;Bond et al., 2004), as discussed in the following section.

Uncertainties
The level and sources of uncertainty in the CEDSGBD-MAPS inventory are similar to those in the CEDSHoesly inventory, which are largely a function of uncertainty in the activity data, emission factors, and country-level inventories. As these uncertainties 670 have been previously discussed in Hoesly et al. (2018), we have not performed a formal uncertainty analysis here, but rather provide a brief summary of the sources of uncertainty associated with this work. We note plans for a robust uncertainty analysis in an upcoming release of the CEDS core system. While this section highlights many of the challenges associated with estimating comprehensive and accurate global bottom-up emission inventories, such inventories remain vital for their use in chemistry and climate models and for the development and evaluation of future control and mitigation strategies. 675

Uncertainties in Activity Data
As discussed in Section 2.1, CEDS default emissions from combustion sources are largely informed by fuel consumption data from the IEA 2019 World Energy Statistics Product (IEA, 2019). While this database provides energy consumption data as a function of detailed source sector and fuel-type for most countries, the IEA data is uncertain and includes breaks in time-series data that can lead to abrupt changes in the CEDSGBD-MAPS emissions for select sectors, fuels, and countries. For example, Fig.  680 S7 shows an order of magnitude decrease (0.1 TgC) in OC industrial emissions from North America between 1992 and 1993, which is driven by a break in IEA biofuel consumption data for the non-specified manufacturing industry sector (CEDS sector: 1A2g_Ind-Comb-other) in the United States. While the magnitude of this particular change is negligible on the global scale, this is not the case for all sectors. For example, as noted in Section S4, a known issue in the IEA data in China in the energy sector causes peaks in the associated NOx and SO2 CEDSGBD-MAPS emissions in 2004. These peak emissions may be over-685 estimated by up to 4 and 10 Tg, respectively, which is large enough to impact historical trends in both regional (

Uncertainties in Global Bottom-Up Inventories
Uncertainties in bottom-up emission inventories vary as a function of space, time, and compound, making total uncertainties 690 difficult to quantify. Default emission estimates in the CEDS system are subject to uncertainties in underlying activity data, such as IEA energy consumption data, as well as activity drivers for process-level emissions. Knowledge of accurate emission factors also drive inventory uncertainties as these are not often available for all sectors in countries with emerging economies, and are heavily dependent on the use, performance, and enforcement of control technologies within each sector and country (e.g., Zhang et al., 2009;. While improvements in data collection and reporting standards may decrease the 695 uncertainty in some underlying sources overtime, the most recent years of CEDSGBD-MAPS emissions are still subject to considerable uncertainty. For instance, the degree of local and national compliance with control measures is often variable or unknown (e.g., Zheng et al., 2018), recent activity and regional emissions data are often updated as new information becomes available, and emissions in generally more uncertain regions, including India and Africa are becoming an increasingly large fraction of global totals. Additionally, from a methodological standpoint, default CEDS emissions after 700 2010 also currently rely on the projection of emission factors from the GAINS EMF30 data release for sectors and countries where contemporary regional scaling inventories are not available. As the CEDS system uses a "mosaic" approach and incorporates information from other global and national-level inventories, the final CEDSGBD-MAPS emissions will also be subject to the same sources and levels of uncertainty as these external inventories. For example, as discussed in Sect. 2.1, default process-level emissions in CEDSGBD-MAPS are derived using 705 emissions from the EDGAR v4.3.2 inventory, with many countries additionally scaled to this inventory during Step 2. As reported and discussed in Crippa et al. (2018), EDGAR v4.3.2 emissions for 2012 at the regional level are estimated to have the smallest uncertainties for SO2, between 14.4% and 47.6%, with uncertainties of NOx between 17.2% and 69.4% (up to 124% for Brazil), CO between 25.9% and 123% (lower for industrialized countries), and NMVOCs between 32.7% and 148% (lower for industrialized countries). Emissions of NH3 are highly uncertain in all inventories (186% to 294% in EDGAR) due 710 to uncertainties in the reporting of agricultural statistics and emission factors that will depend on individual farming practices, biological processes, and environmental conditions (e.g., Paulot et al., 2014). As noted in Crippa et al. (2018) and Klimont et al. (2017), EDGAR v4.3.2 and GAINS uncertainty estimates for BC and OC fall within the factor of two range that has been previously estimated by the seminal work of Bond et al. (2004). While CEDSGBD-MAPS emissions are not scaled to EDGAR v4.3.2 BC and OC emissions, estimates are derived from similar sources and are therefore expected to be consistent with 715 uncertainties in both EDGAR and other global bottom-up inventories. It should also be noted that these reported uncertainty estimates from EDGAR only reflect the uncertainties associated with the emission estimation process and do not account for the potential of missing emissions sources or super-emitters within a given sector .
To evaluate and improve the accuracy of these bottom-up emission estimates, inventories are increasingly using information from high-resolution satellite retrievals, particularly for major cities, large area and natural sources, and large point 720 sources (e.g., Li et al., 2017a;McLinden et al., 2016;Streets et al., 2013;van der Werf et al., 2017;Beirle et al., 2011;McLinden et al., 2012;Lamsal et al., 2011;Zheng et al., 2019;Elguindi et al., 2020). For example, both the CEDSHoesly and CEDSGBD-MAPS inventories incorporate SO2 emission estimates derived using satellite retrievals in McLinden et al. (2016) to account for previously missing SO2 point sources in the CEDS 1B2_Fugitive-petr-and-gas sector (described further in the supplement of Hoesly et al. (2018)), with additional use of satellite data planned for a future CEDS core release. With the continued 725 advancement of satellite-retrievals, the development of source and sector-specific inventories, such as CEDSGBD-MAPS, will continue to provide new opportunities for the application of new satellite-based inventories, which will aid in the quantification of spatial and temporal emissions from distinct sources associated with specific sectors and fuel-types that may not be accurately estimated using conventional-bottom up approaches.

Uncertainties in Regional-Level Scaling Inventories 730
Similar to the CEDSHoesly inventory, the CEDSGBD-MAPS emissions will also reflect the uncertainties associated with the inventories used for the scaling procedure. The inventories with the largest impact on the CEDSGBD-MAPS emission uncertainties relative to the CEDSHoesly inventory will be those from China from Zheng et al. (2018), the DICE-Africa emission inventory from Marais and Wiedinmyer (2016), and the SMoG-India inventory from Venkataraman et al. (2018). While formal uncertainty analyses were not performed for all of these inventories, similar bottom-up methods used in these studies will 735 result in similar sources of uncertainties (activity and emission factors) as the global inventories. For example, Zheng et al. (2018) state that the largest sources of uncertainties are the accuracy and availability of underlying data (reviewed in Li et al. (2017b)) and that the levels of uncertainty for China emissions between 2010 and 2017 are expected to be similar to previous national-level bottom-up inventories derived using similar data sources and methodology, such as Zhao et al. (2011), Lu et al. (2011. Similar to global inventories, these previous regional studies estimate much lower levels of 740 uncertainty for SO2 and NOx ( 16% and -13 to +37% respectively) than for CO (70%) and OC and BC emissions (-43 to +258% and -43 to +208%, respectively). Some sectors in China and other regions are particularly uncertain, as discussed further below.
Regional and national inventories, however, have the added benefit of using local knowledge to reduce potential uncertainties in emission factors and missing emission sources. For example, Marais and Wiedinmyer (2016) note that the 745 DICE-Africa emissions are uncertain due to gaps in fuel consumption data, however, this inventory also includes sources frequently missing in global inventories such as widespread diesel/petrol generator use, kerosene use, and ad-hoc oil refining, and have used emission factors for on-road car and natural gas flaring that are more representative of the inefficient fuel combustion conditions in Africa (Marais and Wiedinmyer, 2016;Marais et al., 2019). As discussed in Sect. 2, the CEDSGBD-MAPS inventory may still underestimate total emissions from some of these sources (up to 11% in 2013; Sect. 2.2.3), but 750 otherwise will have uncertainties for total Africa emissions similar to the DICE-Africa inventory. For emissions in India, uncertainties also arise from missing fuel consumption data and the application of non-local or uncertain emission factors. Venkataraman et al. (2018), however, is one of the few studies to present a detailed uncertainty analysis of their inventory and use the propagation of source-specific activity data and emission factors to estimate that total emission uncertainties are smaller for SO2 (-20 to 24%), than for NOx (-65 to 125%) and NMVOCs (-44 to +66%). While uncertainties are not explicitly reported 755 for OC and BC emissions, Fig. 1 in Venkataraman et al. (2018) indicates that uncertainties in these emissions are between -60% to + 95%, consistent with BC and OC uncertainties reported in other bottom-up inventories. We also note the ongoing work to improve the accuracy of highly uncertain emission sectors in a future release of the SMoG-India inventory, through the CarbOnaceous AerosoL Emissions, Source apportionment and ClimatE impacts (COALESCE) project . 760 In addition to uncertainties in the scaling inventory emissions, uncertainties are also introduced by the CEDSGBD-MAPS scaling procedure. Uncertainties arise when mapping sectoral and fuel (when available) specific emissions between inventories (as discussed previously), as well as in the application of the calculated scaling factors outside the range of available scaling inventory years. For example, the implied CO EFs in Figure S2 highlight one case in China where the EFs for oil and gas combustion in the on-road transport sector peak in 1999 at a value over three times larger than EFs in all other top emitting 765 countries. For China specifically, the calculated scaling factors for the year 2010 (earliest scaling inventory year) are applied to emissions from all years prior, which was calculated as a value of ~1.58 for the on-road transport sector. The implied EF of ~1.8 g g -1 for this sector in 2003 ( Figure S2) suggests that the SF from 2010 may not be representative of emissions during this earlier time period. We do note, however, that the 1999 peak in total CO emissions in China ( Figure S9) is driven by the IEA energy data and is consistent with the CEDSHoesly inventory . In contrast, EFs from this sector in China 770 after the year 2010 agree with the magnitude and trends found in other countries, further indicating that the scaling factors are most appropriate for years with overlapping inventory data. Other similar examples include coal energy emissions of SO2 in Thailand ( Figure S2). In this case, the REAS scaling inventory spans the years 2000 -2008. The default EFs for the energy sector, however, independently decrease between 1997 and 2001. As a result, when the implied EF of 3.3 for the year 2000 is applied to all historical energy emissions, the implied EFs prior to 1997 become an order of magnitude larger than those in 775 nearly all other top emitting countries ( Figure S2). Overall, the applicability of the scaling factors to emissions in years outside the available scaling inventory years remain uncertain due to real historical changes in activity, fuel-use, and emissions mitigation strategies. These uncertainties, however, vary by compound and sector as, for example, there are no similar peaks in on-road emissions for compounds other than CO in China.
Though the inclusion of these regional inventories can improve the accuracy of the global CEDS system (particularly 780 during years with overlapping data), Hoesly et al. (2018) note that large uncertainties may still persist, even in developed countries with stringent reporting standards. In the US for example, it has been suggested that compared to the US National Emissions Inventory (US NEI), total NOx emissions from on-road and industrial sources in some regions may be overestimated by up to a factor of two (e.g., Travis et al., 2016). In addition, NH3 emissions in agricultural regions in winter may be underestimated by a factor of 1.6 to 4.4 (Moravek et al., 2019), and national and regional emissions of NMVOCs from oil and 785 gas extraction regions, solvents, and the use of personal care products may also be underestimated by up to a factor of 2 (McDonald et al., 2018;Ahmadov et al., 2015).

Uncertainties in Sectoral and Fuel Contributions
Emissions reported as a function of individual source sectors are typically considered to have higher levels of uncertainty than those reported as country totals, due to the cancelation of compounding errors (Schöpp et al., 2005). Source sectors with the 790 largest levels of uncertainty in CEDSGBD-MAPS estimates are generally consistent with other inventories, which include waste burning, residential emissions, and agricultural processes . This higher level of sectoral uncertainty is reflected in the relatively larger uncertainties discussed above in global emissions of OC, BC, and NH3 relative to other gasphase species. In general, uncertainties from these sources are larger due the difficulty in accurately tracking energy consumption statistics and uncertainties in the variability of source-specific emission factors, which will depend on local 795 operational and environmental conditions. For example, residential emission factors from heating and cooking vary depending on technology-used and operational conditions (e.g, Venkataraman et al., 2018;Carter et al., 2014;Jayarathne et al., 2018), while soil NOx emissions and NH3 from wastewater and agriculture result from biological processes that depend on local practices and environmental conditions (e.g., Chen et al., 2012;Paulot et al., 2014). While uncertainties are not always reported at the sectoral level, Venkataraman et al. (2018) do report that industry emissions of NOx and NMVOCs in the SMoG-India 800 inventory actually have larger uncertainties than those from the transportation, agriculture, and residential (NMVOCs only) sectors, while the relative uncertainties for SO2 emissions follow the opposite trend. For total fine particulate matter emissions, Venkataraman et al. (2018) estimate that the sectors with the largest uncertainties are the residential and industry emissions.
Similarly, Lei et al. (2011) estimate that BC and OC emissions from the residential sector in China have the largest inventory uncertainties, while Zhang et al. (2009) andZheng et al. (2018) also report relatively smaller uncertainties from power plants 805 and heavy industry in China due to known activity data, local emission factors, pollution control technologies, and direct emissions monitoring. Overall, the mosaic scaling procedure in the CEDS system will result in similar levels of uncertainties as these regional scaling inventories.
With the release of fuel-specific information in the CEDSGBD-MAPS inventory, additional uncertainties in the allocation of fuel types is expected. In this work, activity data at the detailed sector and fuel level are taken from the IEA World Energy 810 statistics (IEA, 2019) and are subject to the same sources of uncertainty. Emission factors for CEDS working sectors and fuels (Table S2) are derived from GAINS. In general, emissions from solid biofuel combustion are considered to be less certain than fossil fuel consumption due to large uncertainties in both fuel consumption and EFs, particularly in the residential and commercial sectors. For example, by combining information from EDGAR v4.3.2  and a recent TNO-RWC (Netherland Organization for Applied Scientific Research, Residential Wood Combustion) inventory from Denier van 815 der , Crippa et al. (2019) estimated that uncertainties in emissions from wood combustion in the residential sector in Europe are between 200 to 300% for OC, BC, and NH3. Crippa et al. (2019) also report that these uncertainties are largely driven by uncertainties in regional emission factors, as uncertainties in biofuel consumption are estimated to be between 38.9 and 59.5%. These uncertainties, however, are still larger than those estimated for fossil fuel consumption in many countries. As noted in Hoesly et al. (2018), increased levels of uncertainty in fossil fuel emissions are also expected in some 820 countries, including the consumption and emission factors related to coal combustion in China (e.g., Liu et al., 2015;Guan et al., 2012;Hong et al., 2017), which will have the largest impacts on CEDSGBD-MAPS emissions of NOx, SO2, and BC. Specific to the CEDSGBD-MAPS fuel inventory, additional uncertainties may arise from the potential underestimation of total coal, oil and gas, and biofuel emissions associated with fugitive emissions and gas flaring in the energy sector, as well as waste incineration in the waste sector. As discussed above and in Hoesly et al. (2018), fugitive emissions are highly uncertain. The degree of 825 underestimation in combustion-fuel contributions will be dependent on the fractional contribution of process level emissions in these sectors relative to those from coal, biofuel, and oil and gas combustion (Table S8). Additional uncertainties in the gridded fuel-specific products are discussed in the following section.

Uncertainties and limitations in gridded emission fluxes
As noted in Sect. 2.1, global gridded CEDSGBD-MAPS emission fluxes are provided to facilitate their use in earth system models. 830 Relative to the reported country-total emission files, additional uncertainties are introduced in the 0.50.5 global gridded CEDSGBD-MAPS emission fluxes through the use of source-specific spatial gridding proxies in CEDS Step 5. Historical spatial distributions within each country are largely based on normalized gridded emissions from the EDGAR v4.3.2 inventory. These spatial proxies are held constant after 2012, which serves to increase the uncertainties in spatial allocation in large countries in recent years. The magnitude of this uncertainty will depend on the specific compound and sector. For example, gridded 835 emissions from the energy sector will not reflect the closure or fuel-switching of individual coal-fired power stations after 2012. Changes in total country-level emissions from this sector and fuel-type, however, will be accurately reflected in the total country-level emission files. This source of uncertainty is also present in the CEDSHoesly inventory. An additional source of uncertainty in the gridded emissions is that the same spatial allocations are applied uniformly across emissions of all three fuel-types within each source sector. This may lead to additional uncertainties if, for example, emissions from the use of coal, 840 biofuel, and 'other' fuels within each sector are spatially distinct. These uncertainties, however, do not impact the final countrylevel CEDSGBD-MAPS products because they are not gridded.
Lastly, while CEDSGBD-MAPS emissions provide a global inventory of key atmospheric pollutants, this inventory does not include a complete set of sources or species required for GCM or CTM simulations of atmospheric chemical processes.
As noted in Sect. 2, neither CEDSHoesly nor CEDSGBD-MAPS estimates include emissions from large or small open fires, which 845 must be supplemented with additional open-burning inventories, such as the Global Fire Emissions Database (GFED, 2019;van der Werf et al., 2017) or Fire INventory from NCAR (FINN, 2018;Wiedinmyer et al., 2011). In addition, simulations of atmospheric chemistry require emissions from biogenic sources, typically supplied from inventories, such as the Model of Emissions of Gases and Aerosols from Nature (MEGAN, 2019;Guenther et al., 2012). Other sources to consider in atmospheric simulations include volcanic emissions, sea spray, and windblown dust. In addition, the CEDS system does not include dust 850 emissions from windblown and anthropogenic sources such as roads, combustion, or industrial process. Anthropogenic dust sources may contribute up to ~10% of total fine dust emissions in recent years and are important to consider when simulating concentrations of total atmospheric particulate matter (Philip et al., 2017). Lastly, the CEDSGBD-MAPS inventory also excludes emissions of greenhouse gases such as methane and carbon dioxide (CH4, CO2). These compounds were previously included through 2014 in the CEDSHoesly inventory. 855

Data availability
The source code for the CEDSGBD-MAPS system is available on GitHub (https://github.com/emcduffie/CEDS/tree/CEDS_GBD-MAPS and https://doi.org/10.5281/zenodo.3865670 (McDuffie et al., 2020a). To run the CEDS system, users are required to first purchase the proprietary energy consumption data from the IEA (World Energy Statistics; https://www.iea.org/subscribeto-data-services/world-energy-balances-and-statistics). The IEA is updated annually and provides the most comprehensive 860 global energy statistics available to-date. All additional input data are available on the CEDS GitHub repository. additionally available in the CMIP6 format. Note that NOx is in units of NO2 in this format. Additional file format details are in the README.txt file in the Zenodo repository (https://doi.org/10.5281/zenodo.3754964).
To provide an example of the products and file formats available for download from the full CEDSGBD-MAPS repository, we have also prepared an additional data 'snapshot' inventory that provides emissions in all three file formats described above, 875 for the 2014 -2015 time period (McDuffie et al., 2020b). The gridded data are provided as monthly averages for the Dec 2014 -Feb 2015 time period, while the annual data include total emissions from both 2014 and 2015. These data can be downloaded from https://doi.org/10.5281/zenodo.3833935, and are further described in the associated README.txt file.

Summary and Conclusions
We described the new CEDSGBD-MAPS global emission inventory for key atmospheric reactive gases and carbonaceous aerosol 880 from 11 anthropogenic emission sectors and four fuel types (total coal, solid biofuel, and liquid fuel and natural gas combustion and remaining process-level emissions) over the time period from 1970 -2017. The CEDSGBD-MAPS inventory was derived from an updated version of the Community Emissions Data System, which incorporates updated activity data for combustion and process-level emission sources, updated scaling inventories, the added scaling of BC and OC emissions, and adjustments to the aggregation and gridding procedures to enable the extension of emission estimates to 2017 while retaining sectoral and 885 fuel-type information. We incorporated new regional scaling inventories for India and Africa; as a result default CEDSGBD-MAPS emissions are now lower than previous CEDSHoesly estimates for all compounds in these regions other than NMVOCs in Africa and BC in India. These updates improve the agreement of CEDSGBD-MAPS Africa emissions with those from EDGAR v4.3.2, as well as the agreement of all India emissions other than BC with both the EDGAR (2012) and GAINS (2010) inventories. Scaling default BC and OC estimates reduces these global emissions by up to 21% and 28%, respectively, relative 890 to the CEDSHoesly inventory. This reduction improves CEDSGBD-MAPS agreement with both GAINS and EDGAR global estimates of BC and OC, particularly in recent years. The resulting CEDSGBD-MAPS inventory provides the most contemporary global emission inventory to-date for these key atmospheric pollutants and is the first to provide their global emissions as a function of both detailed source sector and fuel type. OC sources are from the residential biofuel and the waste sector. Outside of international shipping, China is the largest regional source of global emissions of all compounds other than NMVOCs. As emissions in North America, Europe, and China continue to decrease, global emissions of NOx, CO, SO2, BC, and OC will increasingly reflect emissions in rapidly growing regions 905 such as Africa, India, and countries throughout Asia, Latin America, and the Middle East. Lastly, in contrast to other compounds, global emissions of NMVOCs and NH3 continuously increase over the entire time period. These increases are predominantly due to increases in agricultural NH3 emissions in nearly all world regions, as well as NMVOCs from increased waste, energy sector, and solvent use emissions. In 2017, global emissions of these compounds have the largest regional contributions from India, China, and countries throughout Africa, Asia, and the Pacific. Due to similar bottom-up methodologies and the use of EDGAR v4.3.2 data in the CEDS system, country-level CEDSGBD-MAPS emissions are expected to have similar sources and magnitudes of uncertainty as those in the CEDSHoesly, EDGAR v4.3.2, GAINS, and scaling emission inventories. These inventories consistently predict the smallest uncertainties in emissions of SO2 and the largest for emissions of NH3, OC, and BC. The latter three compounds largely depend on accurate knowledge of activity data and emission factors for small scattered sources that vary by location, combustion technologies 925 used, and environmental conditions. Uncertainties in the sectoral and fuel allocations in CEDSGBD-MAPS emissions will also generally follow the uncertainties in the CEDSv2019-12-23 system and will largely depend on the accuracy of the fuel allocations for combustion sources in the underlying IEA activity data. Gridded CEDSGBD-MAPS emissions also have uncertainties associated with the accuracy of the normalized spatial emission distributions from EDGAR v4.3.2, which are equally applied to all the four fuel categories and are held constant after 2012. 930 Contemporary global emission estimates with detailed sector and fuel-specific information are vital for quantifying the anthropogenic sources of air pollution and mitigating the resulting impacts on human health, the environment, and society.
While bottom-up methods can provide sectoral-specific emission estimates, previous global inventories of multiple compounds and sources have lagged in time and do not provide fuel-specific emissions for multiple compounds at the global scale. To address this community need, the CEDSGBD-MAPS inventory utilizes the CEDS system (v2019-12-23) to provide emissions of 935 seven key atmospheric pollutants with detailed sectoral and fuel-type information, extended to the year 2017. Due to the direct and secondary contribution of these reactive gases and carbonaceous aerosol to ambient air pollution, contemporary gridded and country-level emissions with both sector and fuel-type information can provide new insights necessary to motivate and develop effective strategies for emission reductions and air pollution mitigation around the world. The CEDSGBD-MAPS source code is publicly available (https://github.com/emcduffie/CEDS/tree/CEDS_GBD-MAPS and https://doi.org/10.5281/zenodo. 940 3865670) and both country total and global gridded emissions from the 2020_v1 version of this dataset are publicly available at Zenodo with the following doi: https://doi.org/10.5281/zenodo.3754964.

Author Contributions
EEM prepared the manuscript with contributions from all co-authors. RVM, MB, and SSJ supervised the scientific content of 950 this publication. EEM led the development of the CEDSGBD-MAPS source code and CEDSGBD-MAPS dataset, with significant contributions from SSJ and PO, as well as supplemental data from KT, CV, and EAM.

Competing Interests
The authors declare that they have no conflict of interest.
Ainsworth   2. CEDS sector and fuel-type definitions. Aggregate sectors and fuel-types in the CEDSHoesly and CEDSGBD-MAPS inventories, as well as the system's intermediate gridding sectors, and detailed working sectors/fuel-types (consistent between CEDSHoesly and CEDSGBD-MAPS inventories). CEDS working sectors are methodologically treated as two different categories: combustion sectors (c) and 'process' sectors (p). As described in text, combustion sector emissions are calculated as a function of CEDS working fuels while process emissions assigned to the single 'process' fuel-type.

1490
indicating that the percent difference was calculated using EDGAR or GAINS, respectively.    Table S9).