The GAINS model is developed by IIASA to explore emission control strategies for air pollutants and greenhouse gases by modelling the impact of possible measures. Originally developed for Europe, it now covers most regions of the world. The model is used to underpin policies such as the UNECE Gothenburg Protocol (UNECE, 2012) and the EU National Emission Ceilings Directive (European Commission, 2016). For this inventory, the most recent emission data from the model were taken as they were incorporated in the CEP_post2014_CLE scenario updated in December 2018 (IIASA, 2018). This scenario takes into account historical emission data up to 2015, and future emissions for 5-yearly intervals (2020, 2025, 2030) were modelled based on the latest available information on activity data and control measures available at the time. The emission data were obtained at the level of detailed source categories and fuels for each country for 5-yearly intervals (2005, 2010, 2015, 2020). Linear interpolation was used to estimate emissions for each of the years in between. To estimate emissions prior to 2005, an earlier GAINS dataset which was used in the TNO_MACC-II inventory (Kuenen et al., 2014) and includes the year 2000 was used to extrapolate the trend backwards in time until 2000.

The GAINS sector and fuel classifications were converted to our own sector and fuel definitions (see Supplement: Table S1 for definition, Table S4 for the links). For industrial combustion, an additional step was needed since the GAINS sector classification has most industrial combustion aggregated to one industrial combustion sector. To split these over various industrial sectors, a specific bottom-up emission inventory was set up for the industrial sectors. This bottom-up inventory uses energy consumption from the International Energy Agency (IEA) energy statistics combined with default emission factors to calculate emissions per pollutant, country, year and industrial sector. The share of each sector in this bottom-up inventory was then used to disaggregate the GAINS industrial combustion emissions over the different industrial sectors as identified in this inventory (see Table S1).

CH4 emissions for 2000–2017 (based on reporting year 2019) were obtained from the national inventory submissions to the UNFCCC (UNFCCC, 2019). Emission data at the CRF level were extracted from the CRF tables and combined into a single database. For categories 1A1–1A4, which concern emissions from the combustion of fuels, emission data were collected for each individual fuel. Subsequently, the CRF sectors were converted to our own sector definitions (see Table S3).

Officially reported emissions (reporting year 2019) were obtained from CEIP for the years 2000–2017 (CEIP, 2019), containing sectoral emissions reported under the EMEP reporting requirements, for all the countries for which data were available and for all air pollutants included in the scope of this inventory. Reporting follows the NFR (Nomenclature for Reporting) structure, which was converted to our own sector definitions as developed for this inventory (see Table S2). Whereas the UNFCCC emissions from combustion activities and IIASA GAINS emissions are available per main fuel type, the EMEP emissions are not. Therefore, the relative distribution of fuels for each pollutant, year, country and sector from the GAINS emissions dataset was used to add the fuel split to the dataset. Where for a specific combination, no emission was available from the GAINS dataset for the same pollutant, year, sector and country, an average fuel split was used which was calculated by taking the GAINS data for the sum of all years between 2000 and 2017. Ultimately, if this average split was also not available, default fuel splits were calculated for each pollutant and sector based on the total emissions from GAINS for the pollutant and sector (in all countries and all years).

As a next step, the reported data were quality-checked to decide the cases in which these are fit for use. In the quality checks, it was taken into account that the greenhouse gas emissions from Annex I parties have been reviewed on an annual basis for many years. For air pollutants, this annual review cycle has been in place since 2017 as part of the National Emission Ceilings Directive (NEC Directive), thus covering only the EU Member States. While this annual review cycle has improved reporting by countries over time, shortcomings are still identified for EU Member States (IIASA, 2019). For non-EU countries, the completeness and quality of the inventory data differ significantly between countries, and for some countries no national emission inventory is available at all (EMEP, 2020). All in all, a thorough check of completeness and accuracy is key before using reported emission inventory data for air quality assessment. Therefore, as a starting point the reported data from each EU Member State (including the United Kingdom and also including Iceland, Norway and Switzerland) were used, whereas reported emission data from other countries were not used. Hereafter these 31 countries are referred to as the EU+ countries. While for all of these other countries (consisting of Turkey, Balkan countries and former Soviet Union countries) in some cases reported emission data are submitted, no consistent emission inventory for air pollutants is available on an annual basis. Therefore GAINS emissions were used for these countries. And while for some of these countries GHG emissions are being reported to the UNFCCC, for consistency reasons GAINS data were also used for the CH4. Figure 2 shows the data sources for each European country.

Next, a quality check is performed for the 31 countries for which reported data were used. These checks focussed on completeness (emissions reported for each GNFR sector where emissions are expected to occur), time series consistency (trend in reported emissions as expected, no missing years) and the distribution of emissions over sectors to identify possible missing sectors. Compared to the assessment of reported data for the earlier TNO_MACC-II inventory (Kuenen et al., 2014), the number of gaps and inconsistencies was significantly smaller, and no major issues of such nature were identified. For three countries (Romania, Malta and Lithuania), emissions prior to 2005 were found to be incomplete and/or inconsistent with later years. This is likely related to the fact that under the NEC Directive, 2005 is the base year, and there is relatively little attention for earlier years. For these three countries, the 2000–2004 emissions were replaced with an extrapolation of the 2005 emissions based on the trend in GAINS emissions per GNFR category.

At the same time, larger inconsistencies were identified for agricultural emissions. For NMVOCs from animal husbandry and manure application, a methodology to estimate emissions was only recently included in the EMEP/EEA Guidebook (EEA, 2019a), which led to inconsistent and incomplete reporting by countries. Given that these emissions are also not included in the GAINS model, it was decided to leave this source out of the CAMS-REG-v4 inventory. For NOx from agriculture, reporting is also found to be inconsistent between countries. In addition, one of the main sources of agricultural NOx emissions is soil NOx, and many air quality models have separate modules to calculate soil NOx internally. To avoid double-counting, it was therefore decided to exclude NOx emissions from agriculture.

Minor issues that were found include mainly gaps or outliers in a specific year for a specific sector. These were identified by looking at year-to-year changes for each country and pollutant at the level of GNFR categories (aggregated sectors) and taking those cases where the average importance of the GNFR category in the national total was above 3 %, and at the same time there was a change of more than 5 % from one year to the next in the time series. Where this was based on reported data, the underlying detailed sector data were checked, and gaps or other errors were identified. These were fixed using interpolation and/or extrapolation or by keeping emissions constant from the previous or next year (based on a manual case-by-case assessment).

Finally, some other modifications were made to the dataset to make it consistent and fit for purpose for the spatial distribution:

Emissions of NMVOCs from natural gas production and distribution systems (NFR category 1B2b) were split into production, high-pressure distribution and low-pressure distribution based on the relative contribution of these subsectors in GAINS to total emissions of CH4.

For public power and heat plants, three different datasets were combined. E-PRTR (the European Pollutant Release and Transfer Register) collects facility-level emission data for all EU Member States, with informing the European citizens about pollutant transfers and releases in their area as main a goal. The data have been reported on an annual basis since 2007 (before that 3-yearly since 2001) and are publicly available (EEA, 2019b). In addition to E-PRTR, EU Member States are required to report information (fuel consumption and emissions) at the level of individual installations (stacks) of large combustion plants annually. This requirement follows from the Large Combustion Plants (LCP) Directive (European Commission, 2001), superseded by the Industrial Emissions Directive (European Commission, 2010). Thirdly, a commercial dataset on power plants known as Platts-WEPP was used, which records characteristics of power plants worldwide, such as the technology type, fuel type, capacity, etc. (Platts, 2017).

In processing the E-PRTR- and LCP-reported emission data at the facility level, a sector designator was used to select the facilities corresponding to the public power and heat sector. Matching and linking of identical facilities in the LCP and E-PRTR datasets was partly possible using a joint “national ID” field. Linking was then manually completed based on similarities in facility name and location, and each individual facility was assigned a new and unique identifier. Emissions of NOx, SO2 and dust reported to the LCP dataset were supplemented with an estimate of CO2 emissions based on reported fuel use and default CO2 emission factors (Eggleston et al., 2006). While CO2 is not explicitly part of this work, it has been included in the point source database since it is relatively well covered and has a relatively low uncertainty and could therefore be used as an indicator for gap-filling.

Emissions reported to the E-PRTR and LCP datasets were then combined by facility, pollutant and year using the new joint ID field. Since the scope of the E-PRTR is the most complete

LCP reporting is only for facilities with a thermal capacity > 50 MW thermal and covers emissions of NOx, SO2 and dust.

The facility-level emissions were then split by fuel type. This was done by calculating a proxy emission value using the LCP-reported fuel input by fuel type and country-, fuel- and pollutant-specific emission factors from the IIASA GAINS model at the country-, fuel- and pollutant-specific level. The relative contribution of each fuel type was then applied to the reported emission value to distribute it to the various fuel types. Where no fuel input data were available, the unit fuel type from the Platts-WEPP dataset was used instead to assign the emission value to one or multiple fuel types. Finally, for the remaining facilities the fuel type used was searched for online to fill the gaps.

Several checks were then performed to compare the total emissions by country with the reported UNFCCC and EMEP sector totals for public power and heat production. This check led to the identification of unrealistically high emission values for some facilities, where there had evidently been an error in reporting. In these cases, the erroneous value was removed (and then gap-filled following the routine described above) or lowered in case of a likely unit error (e.g. factor 10 too high).

The final step was the assignment of point source and area source emissions to GNFR A. In the case that the reported sector-level emissions were higher than the processed facility-level emissions, the remainder is assigned as area source emissions under GNFR A, representing the smaller facilities which are below the threshold for reporting. The processed facility-level emissions were then assigned as point source emissions under GNFR A. For some countries and years, the total facility-level emissions were higher than the reported sector total. In that case the emissions for all facilities were scaled down to arrive at the country total emission for the sector power plants. For the point source emissions, the facility coordinates included in the E-PRTR dataset were used for spatial distribution of the emissions. For some facilities, coordinates were added or corrected manually when they were found to be incorrect or missing. For the area source emissions, the CORINE dataset was used to spatially allocate emissions to areas with industrial activity (see Sect. 2.3.5).

For industrial point sources, similar to the power plants the main sources of information for the EU+ were E-PRTR, combined with various online sources and one commercial industrial directory, and the TNO point source database for the other countries in Europe (see Sect. 2.3.3). For industrial emissions in particular the E-PRTR registry is the most complete and best available database for the EU+, but it nonetheless frequently contains facilities that have an incorrect sector code or erroneous or missing emission data. Compared to power plants, industrial point sources in E-PRTR are much more numerous and diverse in type, and as a consequence a somewhat less detailed approach had to be followed to correct any pressing deficiencies in E-PRTR.

For oil refineries and integrated iron and steel plants, external lists of all existing plants in the EU+, including operational status, were consulted to extract all corresponding records from E-PRTR (years 2001, 2004 and annually from 2007 onwards) regardless of the E-PRTR sector code. Any missing facilities were added. Complete lists of existing and operational oil refineries in the EU+ were available from, for instance, the OGJ Worldwide Refining Survey or online directories such as the Wiki site “A barrel full” (A barrel full, 2021). For integrated iron and steel plants, the extensive commercial Plantfacts Capacity Database was consulted to extract any facilities operational after the year 2000 from E-PRTR and to complete any missing plants. The Plantfacts Capacity Database is an online source (World Steel Dynamics, 2020) which was initially compiled and maintained by the German Steel Institute VDEh.

Next, based on E-PRTR sector classification, the following types of industrial facilities were selected from E-PRTR (including data from the European Pollutant Emission Register (EPER), E-PRTR's predecessor, for the years 2001 and 2004):

coal mines

The basic aim for industrial point sources was to compile a complete emission time series for plants that appeared to have been operational in the period 2000–2017 but also to take plant closures, production stops and emission data below the reporting thresholds into account. For oil refineries and integrated iron and steel plants, any missing (but expected) CO2, NOx and SO2 data were estimated, in addition to NMVOC emission data for refineries specifically and CO and PM data for iron and steel plants. For all other industrial activities the basic assumption used in gap-filling was that all operational plants must have emission data for each relevant pollutant for each reporting year. So principally all emission data missing in this sense have been gap-filled for each facility by assuming the average of the emissions reported by that facility for other years unless

emission data reported for other years was ever close to the threshold (missing emission data are below threshold);

An independent point source database was developed for the spatial distribution of point source emissions in earlier European inventories (Denier Van Der Gon et al., 2010; Kuenen et al., 2014). Since this point source dataset is only available for the year 2005, it is only used for specific sectors where no point source data could be extracted from E-PRTR.

Airports have been included as point sources in this dataset. The contribution of each airport to the country total for this sector was calculated based on flight statistics per airport (Eurostat, 2019). First, a split was made at the country level between passenger and freight traffic using the number of flights in the country as a whole. Thereafter, freight traffic was distributed to individual airports using the tonnage of freight, and passenger traffic was distributed using the number of passengers per airport. This was done on an annual basis to allow for changes in time (e.g. opening of a new airport in a different location). Eurostat data were only available from 2003; for 2000–2002 the distribution is based on the Eurostat statistics as of 2003.

For domestic waste water treatment, an EEA dataset on urban waste water treatment plants was used (EEA, 2014), which includes plant coordinates. This dataset provides the flow and capacity or urban waste water treatment plants in all European countries, from which their share in total emissions was inferred.

The LandScan Global dataset (Oak Ridge National Laboratory, 2017) for population density was obtained for the years 2005, 2010 and 2015 at high spatial resolution (∼ 1 km). A country mask was combined with this dataset to allocate each grid cell to a specific country. For cells that include a border between countries the relative area of each country was used to split up population for the two (or more) countries.

Based on the population of each grid cell, an additional qualification was made whether the cell was allocated as urban or rural. The definition of urban and rural areas however differs significantly between different regions of the world and between countries. To ensure consistency across the domain, a fixed value of 250 inhabitants per square kilometre was chosen. Above this value the cell classifies as urban; below it classifies as rural. This results in around 75 % of the people living in urban areas, which corresponds well to the urban population percentage for the EU as published elsewhere (World Bank, 2018).

As a final step the data were converted to a resolution of 0.05∘ × 0.1∘ for the entire domain, separately for urban, rural and total population.

The CORINE Land Cover dataset (Copernicus Land Monitoring Service, 2016) was obtained from Copernicus Land Monitoring. This dataset has a resolution of approximately 100 m and for each grid cell the main use type is given. These high-resolution grid cells were aggregated to 0.05∘ × 0.1∘, and for each of these larger grid cells the fraction of different main use types was calculated by adding up the number of grid cells with this main use type and normalising the totals per 0.05∘ × 0.1∘ grid cell.

The dataset identifies around 45 different use classes. For this work, three different proxies were extracted from this dataset:

industrial area (taken as the sum of “industrial or commercial units”, “port areas” and “construction sites”)

For road transportation, shape files from Open Transport Map (OTM; OpenTransportMap, 2017) and Open Street Map (OSM; Open Street Map, 2017) were obtained for the entire European domain.

While OSM contains the road network across Europe, OTM adds the traffic volumes to these and divides the roads into different classes (main roads and from first-class to fifth-class roads). The two datasets were merged, which resulted in a European-wide map of roads, each with an associated intensity. This intensity is directly used as a proxy for the emissions, not taking into account more detailed parameters like vehicle speed, traffic jams, etc. However, especially for smaller roads in many cases the traffic volume was not available, and traffic volume had to be estimated for these cases. To do this, first a relation between the traffic volume and the population density in each grid cell was determined based on all the grid cells with a known traffic volume per country and per road class. Then this relation was used to estimate the traffic volume where this was not available, based on the population. Since smaller roads are expected to contain mostly local traffic, this approach is expected to represent reality reasonably well.

The traffic intensity map was classified per country, per vehicle type and per road type. The latter refers to the separation between highway (assumed equivalent to main roads) and non-highway emissions. Thereafter, the non-highway emissions were split between urban and rural by means of overlaying the traffic intensity map with a population map (which includes rural and urban shares; see Sect. 2.3.4). This way, the fact that on average the road in urban areas will have higher emissions per vehicle kilometer (vkm) is taken into account in a generic way. Using this approach, 18 different traffic intensity maps were created (6 vehicle types, 3 road types) for Europe as a whole. As a final step, the resulting dataset was combined with a country mask to introduce the country codes similar to the population and land use distributions, and the map was subsequently normalised with the country total traffic intensities to create the final proxy map.

For agriculture, a number of specific agricultural proxies have been used:

Gridded livestock of the world (FAO, 2010), available from the UN Food and Agricultural Organisation (FAO), has been obtained per animal type and converted to a 0.05∘ × 0.1∘ resolution.

Finally, for high-pressure gas distribution network and for rail transport, specific maps are available to spatially distribute emissions. These are similar to those used in the earlier TNO_MACC inventory (Kuenen et al., 2014).

Shipping emissions include shipping both at sea and on rivers. For the main air pollutants, the contribution of inland shipping at the European scale is limited, typically 3 %–4 % for PM and NOx and < 0.5 % for SO2. Nevertheless, in major rivers and near large harbours, inland shipping emissions may be important for local air quality. As an example of the output from the STEAM model, Fig. 6 shows the distribution of shipping emissions in the North Sea and Benelux. This illustrates major shipping routes at sea, and high emissions in and close to the major ports (Rotterdam, Amsterdam, Antwerp). Further transport along the main rivers is shown on the inland shipping map (centre), which shows higher emissions along the inland waterways in the Netherlands, Belgium and Germany.

As explained in Sect. 2.4, gridded emissions from the STEAM model were only available for 2014 onwards. For earlier years, scaling factors were developed for the shipping emissions to estimate emissions in the year 2000–2013 by sea, taking into account environmental control measures such as the Sulfur Emission Control Areas (SECAs). This is illustrated in Fig. 7, where implementation of SECAs on the North Sea in 2007 and consecutive sulfur reductions in 2010 and 2015 are clearly visible in the trend of SO2 emissions.

CAMS-REG-v4.2 does not only add new years to the inventory compared to earlier CAMS-REG versions, it also provides updated emissions for the entire time series back to 2000. To illustrate the difference with earlier versions, Table 3 shows the relative change between total emissions for the sum of all EU+ countries for the same year with the two most widely used predecessors. TNO_MACC-III is an extension of the TNO_MACC-II database (Kuenen et al., 2014), covering the years 2000–2011. CAMS-REG-v2 is an earlier version of the current CAMS-REG-v4 dataset, covering the years 2000–2015. Differences between the same years in the different inventories are partly related to methodological changes in the CAMS-REG inventory, but most can be explained by recalculations of officially reported emissions of air pollutants by each country. For TNO_MACC-III, emission data from reporting year 2013 are used as the basis; for CAMS-REG-v2 this concerns emission data as reported in 2017, while CAMS-REG-v4.2 builds on emission data as reported in 2019. Table 3 shows that PM and NH3 emissions are typically higher in CAMS-REG-v4.2 compared to TNO_MACC-III, while NOx, SO2 and CO emissions are typically lower.

One of the significant changes between this dataset and its predecessors is the approach to AWB as discussed in Sect. 2.5. Table 3 shows that CAMS-REG-v4.2 for incomplete combustion-related species like CO and PM2.5 is 6 % lower that its immediate predecessor CAMS-REG-v2.2.1. While this is the net sum of various sources being adjusted downward and upward, an important contribution comes from the revised AWB estimate based on earth observation data. Over the entire European domain AWB now contributes 3.1 % and 3.3 % to total CO and PM2.5 emission, respectively. This used to be 8.2 % and 11.2 %, respectively, in CAMS-REG-v2.2.1.

Figure 8 shows the trends in emissions for the EU+ countries from the most widely used earlier versions: TNO_MACC-III and CAMS-REG-v2. It illustrates that the difference is not static in time and may change from year to year. Most of the differences can be explained by changes in reporting, which are in turn related to improved understanding of emissions and improved guidance for emission estimation.

Figure 10 shows a country-specific comparison for PM2.5 emissions from GNFR C (small combustion) in TNO_MACC-III, CAMS-REG-v2 and CAMS-REG-v4 for the EU+ countries where reported data are used as the basis, which implies that these are representative of reporting of PM2.5 emissions from this source category in different reporting years (2013, 2017 and 2019, respectively). The figure shows significant changes for some countries (e.g. EST, LTU, GBR, ESP, ITA, ROU) but only very small changes for others (e.g. DNK, NOR, DEU, FRA, SVK). These differences are significant and related to the inclusion of condensables in the emission inventories (which is further discussed in Sect. 4).

EDGAR (Emissions Database for Global Atmospheric Research) (Crippa et al., 2018, 2020) is a widely used global emission inventory which uses a bottom-up approach for all sectors based on activity data (energy statistics, industrial production, etc.) combined with emission factors, developed independently from the national inventories from individual countries. Table 4 shows a comparison between the results from this inventory and EDGAR v5.0 (Crippa et al., 2020) for the year 2015. The comparison is made separately for the EU+ countries (where CAMS-REG is largely based on reported data from the countries) and non-EU countries (where CAMS-REG is largely based on the GAINS emissions). The Russian Federation is excluded from this analysis since EDGAR covers the entire country, while CAMS-REG only includes the European part of Russia.

Total emissions from EDGAR and CAMS-REG at the European scale differ considerably between pollutants, as illustrated in Table 4, especially for non-EU countries. However, also for the EU+ countries significant differences are seen for CH4, NH3, NMVOCs and SO2 in particular. In most cases EDGAR emissions are higher, except for PM in non-EU countries, where CAMS-REG-v4.2 provides higher emissions.

Figure 10 shows the difference between CAMS-REG-v4 and EDGAR-v5.0 on a GNFR sector level for NOx and NMVOCs. To make this comparison, EDGAR emissions (which use the IPCC classification) were converted to GNFR sector classification. The match is not always perfect, which is illustrated by the figure for NMVOCs, where the EDGAR-v5.0 emissions for GNFR category B (industry) also include emissions from solvents (GNFR E), which explains the large discrepancy there. Also emissions from GNFR I (off-road) are partly included in other sectors, in particular in GNFR C. Key differences are seen for GNFR D (fugitives) and GNFR J (waste), where EDGAR includes significantly higher emissions, but also for agriculture livestock (GNFR K) there is a discrepancy since NMVOC emissions from this source are not included in CAMS-REG (see Sect. 2.2.4).

For NOx, the comparison shows a large difference for shipping (GNFR G), where EDGAR emissions are significantly higher. This may be related to the allocation of shipping emissions between countries and sea regions as this comparison excludes international shipping. On the other hand, EDGAR emissions are significantly lower for GNFR I (off-road and other transportation), which is partly related to the sector allocation issue. Another difference is seen in GNFR L (other agriculture), which could be related to the inclusion of soil NOx in EDGAR, which is not included in CAMS-REG-v4 (see Sect. 2.2.4).

Figure 11 shows the emission trends between 2000 and 2015 in both CAMS-REG-v4.2 and EDGAR-v5.0 for five selected pollutants. It is shown that generally the trends are comparable in both datasets, but the downward trend in CAMS-REG is stronger for each of these pollutants compared to EDGAR. This difference in trend is most notable for NMVOCs, SO2 and PM.

For a comparison of previous versions of the CAMS-REG and TNO-MACC data to the inventories made by the US-EPA we refer to Pouliot et al. (2015).