Atmospheric Environment 75 (2013) 43E57

Atmospheric Environment 75 (2013) 43e57

Contents lists available at SciVerse ScienceDirect

Atmospheric Environment

journal homepage: www.elsevier.com/locate/atmosenv

A comparative analysis of two highly spatially resolved European atmospheric emission inventories

J. Ferreira a,*, M. Guevara b, J.M. Baldasano b,c, O. Tchepel a, M. Schaap d, A.I. Miranda a, C. Borrego a a CESAM & Department of Environment and Planning, University of Aveiro, 3810-193 Aveiro, Portugal b Earth Science Department, Barcelona Supercomputing Centre-Centro Nacional de Supercomputación (BSC-CNS), Jordi Girona 29, Ediﬁcio Nexus II, 08034 Barcelona, Spain c Environmental Modelling Laboratory, Technical University of Catalonia, Barcelona, Spain d TNO, Princetonlaan 6, 3584 CB Utrecht, The Netherlands highlights

Inter-comparative analysis of distinct spatial disaggregation methods of emission inventories. 2 EU emission inventories converted into 3 gridded datasets, under a common grid with 12 12 km2. Gridded emission inventories, well discretized and detailed, suitable for air quality modelling. Different databases and disaggregation methods lead to different spatial emission patterns. article info abstract

Article history: A reliable emissions inventory is highly important for air quality modelling applications, especially at Received 18 July 2012 regional or local scales, which require high resolutions. Consequently, higher resolution emission in- Received in revised form 20 March 2013 ventories have been developed that are suitable for regional air quality modelling. Accepted 27 March 2013 This research performs an inter-comparative analysis of different spatial disaggregation methodologies of atmospheric emission inventories. This study is based on two different European emission inventories Keywords: with different spatial resolutions: 1) the EMEP (European Monitoring and Evaluation Programme) in- European emission inventories ventory and 2) an emission inventory developed by the TNO (Netherlands Organisation for Applied Disaggregation methods Scientific Research). These two emission inventories were converted into three distinct gridded emission Inter-comparative analysis datasets as follows: (i) the EMEP emission inventory was disaggregated by area (EMEParea) and (ii) Spatial variability following a more complex methodology (HERMES-DIS e High-Elective Resolution Modelling Emissions System e DISaggregation module) to understand and evaluate the influence of different disaggregation methods; and (iii) the TNO gridded emissions, which are based on different emission data sources and different disaggregation methods. A predefined common grid with a spatial resolution of 12 12 km2 was used to compare the three datasets spatially. The inter-comparative analysis was performed by source sector (SNAP e Selected Nomenclature for Air Pollution) with emission totals for selected pollutants. It included the computation of difference maps (to focus on the spatial variability of emission differences) and a linear regression analysis to calculate the coefficients of determination and to quantitatively measure differences. From the spatial analysis, greater differences were found for residential/commercial combustion (SNAP02), solvent use (SNAP06) and road transport (SNAP07). These findings were related to the different spatial disaggregation that was conducted by the TNO and HERMES-DIS for the first two sectors and to the distinct data sources that were used by the TNO and HERMES-DIS for road transport. Regarding the regression analysis, the greatest correlation occurred between the EMEParea and HERMES-DIS because the latter is derived from the first, which does not occur for the TNO emissions. The greatest correlations were encountered for agriculture NH3 emissions, due to the common use of the CORINE Land Cover database for disaggregation. The point source emissions (energy industries, industrial processes, industrial combustion and extraction/distribution of fossil fuels) resulted in the lowest

* Corresponding author. E-mail address: [email protected] (J. Ferreira).

coefﬁcients of determination. The spatial variability of SOx differed among the emissions that were obtained from the different disaggregation methods. In conclusion, HERMES-DIS and TNO are two distinct emission inventories, both very well discretized and detailed, suitable for air quality modelling. However, the different databases and distinct disaggregation methodologies that were used certainly result in different spatial emission patterns. This fact should be considered when applying regional atmospheric chemical transport models. Future work will focus on the evaluation of air quality models performance and sensitivity to these spatial discrepancies in emission inventories. Air quality modelling will beneﬁt from the availability of appropriate resolution, consistent and reliable emission inventories. Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction (1 1km2) and temporal (1 h) resolution. The other module, HERMES-DIS is used for Europe, and performs a SNAP (Selected Under the EU (European Union) National Emission Ceilings Nomenclature for Air Pollution) sector-dependent spatial Directive and the UNECE (United Nations Economic Commission for (12 12 km2, and up to 1 km2) and temporal (1 h) disaggregation of Europe) Convention on Long-range Transboundary Air Pollution the original annual EMEP gridded emissions at a 50 50 km2 (LRTAP) and its protocols annual submission of air emissions in- resolution. ventory data are required at the national level. Such data are made In many cases, emission inventories clearly reflect the purpose publicly available. Similarly, while many environmental regulators that they have been designed for (i.e., serving for regulatory purposes within European countries have made industrial facility emissions such as the UNFCCC and UNECE CLRTAP EMEP inventories). Specif- data available at the national level, it is only in the last few years ically, compliance with international protocols drives the need for a that such data have been made publicly available in a coordinated pragmatic emissions accounting system. In contrast, inventories initiative at the European level (EMEP, 2007). such as the EDGAR emission database are bottom-up science driven The EMEP (European Monitoring and Evaluation Programme) emission compilations that are based on emissions factors and Centre on Emission Inventories and Projections (CEIP) has been generally provide openly available statistical information regarding assigned the task of collecting emissions and projections of acidi- activity rates (Van Aardenne et al., 2005). While the legal implica- fying air pollutants, heavy metals, particulate matter and photo- tions and validation of national submissions are important in the chemical oxidants from LRTAP Convention parties. In addition, the former inventory, the main objective of the latter inventory is to EMEP has been assigned the task of obtaining input data for long- provide comprehensive and consistent datasets for air quality range transport models, which estimate air pollution levels at the modelling. This difference is somewhat reflected in the sectorial European scale. Currently, the centre operates the UNECE/EMEP structure in which these inventories are compiled (Reis et al., 2009). emission database (WebDab), which contains information The spatial and temporal coverage of emissions for use in air regarding emissions and projections from all parties of the LRTAP quality models is important. On a global scale, a resolution of be- Convention in two separate datasets, including the official emis- tween 10 and 12 km is used to capture the general spatial distri- sions that are submitted by the parties, and the emissions used by butions of pollutants. However, for urban scale modelling, modellers (EMEP e CEIP, 2010; EEA, 2010). inventories with a grid spacing of less than 1 to 4 km potentially In addition to the inventory based on obligatory reporting of overlook vital distribution patterns, which result in mismatch be- national emissions, other emission inventories covering Europe are tween model results and observations. This mismatch occurs when available, including the CGEIC (http://www.ortech.ca/cgeic), RETRO the model fails to adequately represent street canyons and trans- (http://retro.enes.org), EDGAR (http://www.mnp.nl/edgar), TNO- port routes, which influences the spatial distributions of urban air GEMS (Visschedijk et al., 2007) and PAREST-MEGAPOLI (Denier pollution. van der Gon et al., 2010) inventories. These inventories are partly The geographical distribution of emissions within countries independent of the EMEP database, but maintain some of its fea- plays a larger role in explaining the differences between the in- tures. The ECCAD e GEIA database (Emissions of atmospheric ventories than in explaining the differences in the countries total Compounds & Compilation of Ancillary Data e Global Emission emissions. Very large differences were found between the contri- Inventory Activity) (http://eccad.sedoo.fr; http://www.geiacenter. butions of various sectors to the total emissions from each city. org) is a link to most of these emission inventories. The most These differences are related to the different methodologies that recent EDGARv4.2 database (EDGAR, 2011) provides global annual are used in inventory development (Butler et al., 2008). Emissions emissions data per country and on a grid with three different can be spatially distributed based on population (total, urban or spatial resolutions (up to 0.1 by 0.1 since 2005) for all relevant air rural), the locations of individual emitting facilities, or a combina- pollutants and GHGs. The E-PRTR (European Pollutant Release and tion of these factors. Butler et al. (2008) recommend the use of an Transfer Register database) has built diffusive emissions grid maps ensemble of inventories, placed more attention on the geographical based on officially submitted national emissions data at a resolu- distribution of emissions, and on the integration of local inventories tion of 5 5km2 that cover all EU27 states and EFTA countries into global emission inventories. (Theloke et al., 2011). The High-Elective Resolution Modelling A detailed evaluation of the GEMS-TNO emission inventory, Emission System (HERMES), developed between 2005 and 2006 by which was used for several modelling applications along with the the Barcelona Supercomputing Centre, is currently being used EMEP inventory, for 2003, showed that the annual differences be- within the CALIOPE operational air quality forecasting system for tween the two inventories were relatively small (typically of 10% or Europe and Spain. HERMES is divided into two main modules that less), although a few larger differences were obtained for specific can work together or separately, depending on the working domain sectors and/or pollutants (Simmons et al., 2010). (Baldasano et al., 2008). The first module, named HERMES-BOUP, Moreover, Reis et al. (2008a, 2009) investigated different NOx was specifically developed for Spain and uses a combination of emission inventories and highlighted some of the most relevant bottomeup approaches for estimating emissions at high spatial similarities and differences. One question that arose from this J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57 45 research was how far existing inventories are able to provide robust or forecasting modes. In the scope of European projects and initia- and comprehensive datasets for air quality modelling. Verification tives, applications are COST728 (http://cost728.dmi.dk/), CityDelta and validation of emissions through modelling exercises and in- (http://aqm.jrc.it/citydelta), and more recently, MACC (http://www. tercomparisons can only provide a meaningful contribution for gmes-atmosphere.eu) and AQMEII (http://ensemble2.jrc.ec.europa. closing existing gaps if the main input dataset (emissions) has eu/aqmeii). Examples of European air quality forecasting systems known uncertainty ranges. Another issue to be discussed is related include CALIOPE (http://www.bsc.es/caliope/), PREVAIR (http:// to which emission inventory is better for air pollution modelling, www.prevair.org), and RIU (http://www.eurad.uni-koeln.de/index_ the ‘official’ inventories that are compiled by national experts to e.html). Further information can be found in Zhang et al. (2012). fulfil reporting obligations or the inventories that are developed by Among these systems, the following widely used emission in- independent experts for modelling purposes. Research based inventories (e.g., Maes et al., 2009; Pouliot et al., 2012) were chosen for ventories can provide a valuable source for modelling activities inter-comparative analysis based on their coverage and considered because they are based on a transparent, harmonised methodolo- pollutants. These inventories differ in complexity, and have different gies that are adequate for their purpose. Furthermore, these in- bases and disaggregation methodologies: ventories can help identify potential gaps or inconsistencies in national ‘official’ inventories by stimulating scrutiny of missing - The EMEP emission inventory for 2008 on a 50 50 km2 grid sources. According to Reis et al. (2008b), the best option is to use was based on emission data delivered by Member-States, as both types of inventories for the purposes that they are compiled required by EU regulation (LRTAP Convention), and EMEP for, and perform intercomparison analysis. expert estimations (Co-operative Programme for Monitoring Methods to compare emission inventories need to be robust, and Evaluation of the Long-range Transmission of Air Pollut- easy to set in place, and efficient in their application. Graphical ants in Europe, http://www.emep.int)(UNECE, 2009). This in- approaches, which allow visual interpretation and judgement of ventory included anthropogenic and some natural emission the results, provide numerical indicators, and describe a certain sources (volcanoes in Italy and DMS marine fluxes) with a inventory quality, may be considered (Winiwarter et al., 2003). temporal and spatial resolution of 1 year and 50 km respec- Winiwarter et al. (2003) compiled and tested different methods, tively (Mareckova et al., 2009). including graphical methods (such as emission maps, scanning data - The TNO emission inventory was developed for Europe by TNO series, and scattergrams) and numerical methods (such as co- (Denier van der Gon et al., 2010) for the MACC European efficients of determination, the Moran coefficient, and the accept- Research Project for the base year 2005 and was updated for ability criterion) to compare derived gridded air emission 2007. This emission inventory is on a 1/8 1/16longitude- inventories), on the same spatial grid by using different techniques. latitude grid (approximately 7 8km2) that ranges between None of the applied methods were able to fully account for all of the 44.625 and 60.625 latitude (779 cells) and 29.75 and differences and agreements that potentially occur between 78.4375 longitude (842 cells). When possible, this inventory different inventories. Thus, a combination of methods should be was based on the official reported emissions from the Euro- applied based on respective needs. pean Environment Agency (EEA, 2008). In some cases, the in- After this overview of atmospheric emission inventories that ventory was completed with alternative emissions data from were developed to fit specific purposes, several questions arise the IIASA RAINS model or the TNO default values. regarding their applications for regional scale air quality modelling. First, which sectors are considered and what type of information or To make a spatial comparative analysis, the emission in- which databases are used? Should the choice of an inventory be ventories have been converted to a common 12 by 12 km2 reso- based on its coverage and resolution? Is it worth investing in com- lution horizontal grid, which ranges from approximately 16.0e plex disaggregation methodology to obtain a more realistic spatial 63.7 latitude and from 40.0e54.0 longitude (Fig. 1). This con- distribution of emissions? Is it important to characterise and eval- version was conducted for both cases while considering the area of uate the differences between the available inventories to infer their each inventory domain that intersects the common grid, as shown potential impacts on the modelled air pollutant concentrations? in Fig. 1. This paper attempts to answer these questions and aims to It is important to refer that the original EMEP dataset is also based perform a comparative analysis of two emission inventories that are on a downscaling process based on submitted emissions by country used for air quality modelling purposes. This comparative analysis is (http://www.ceip.at/fileadmin/inhalte/emep/pdf/ gridding_process. accomplished by spatially disaggregating emissions to a common grid pdf). This gridding process is a simple method that considers large resolution using different disaggregation methods. The intercom- point sources data and population data. However, the resolution is parison among the three different outputs was based on the two too coarse and so the downscaling parameters and principles un- emission inventories to 12 12 km2 per SNAP sector, and was per- derneath assume lower importance compared to the disaggregation formed for carbon monoxide (CO), nitrogen oxides (NOx), sulphur parameters under TNO or HERMES-DIS. oxides (SOx), ammonia (NH3), non-methane volatile organic com- The TNO inventory (TNO hereinafter) was adapted to the common pounds (NMVOC), and particulate matter (PM2.5 and PM10), based on grid with area aggregation (mass-conservative GIS-interpolation of graphical and numerical tools. Section 2 describes the emission in- TNO original emissions) and projection transformation to the ventories and the disaggregation methods that were used in this Lambert Conformal projection of the common grid. study. Section 3 presents the comparison approach and the methods For the EMEP emission inventory, the disaggregation into the that were used to analyse the results. Section 4 is devoted to the major resolution grid was conducted in two different ways as presentation and discussion of the obtained results, and the last sec- follows: tion summarises the conclusions and outlines several recommenda- tions regarding the use of European emission inventories for research. - By area in the study domain, the same way as TNO (EMEParea hereinafter); and 2. Studied European emission inventories - By applying a specific disaggregation methodology for each SNAP sector, which is currently being used in the European Air European gridded emission inventories have been applied in quality Forecasting System and is operationally performed by regional air quality modelling for individual use in either diagnostic the Barcelona Supercomputing Centre in the scope of the 46 J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57

Fig. 1. Emission inventory original grids within the common grid (Lambert Conformal projection) deﬁned in this study including a) the EMEP grid (50 50 km2 resolution) and b) the TNO grid (1/8 1/16 resolution).

CALIOPE system (http://www.bsc.es/caliope)(Pay et al., 2010) not to set the absolute levels. For CO, the contribution of each (HERMES-DIS hereinafter). source was calculated with a bottom-up emissions inventory that was created while considering transport and stationary combustion To understand the differences between the three outputs and sources (Denier van der Gon et al., 2010). their spatial distributions better, it is important to describe the The High-Elective Resolution Modelling Emissions System related spatial disaggregation methods that are applied in TNO (HERMES) uses up-to-date information and state-of-the-art (original version) and HERMES-DIS. methodologies for estimating emissions by sector-specific sources Table 1 lists the SNAP sectors ID, full name and the short name or by individual installations and stacks (Baldasano et al., 2008, that was adopted in this paper. Table 2 summarises the key infor- 2011; Pay et al., 2010; Baldasano et al., 2011). The second updated mation related to each SNAP sector for the HERMES-DIS and TNO version of the HERMES-DIS module that was used in the European emissions, and provides the type of sources that are considered in domain, performs SNAP sector-dependent spatial (12 12 km2, and each case and the elements of disaggregation that were used. up to 1 km2) and temporal (1 h) disaggregation of the original The TNO European emission database for NOx,SO2, NMVOC, annual EMEP gridded emissions for 2008 (as the reference year of CH4,NH3, CO and primary PM10 and PM2.5 for 2005 was created by the latest version) on a 50 50 km resolution. updating and improving the previous emissions inventory that was In both cases, TNO and HERMES-DIS, a Geographical Informa- developed for 2000 (Visschedijk and Denier van der Gon, 2005). tion System (GIS) combined with a variety of datasets, including This database was completed with alternative expert emissions proxy maps and collected and processed point source data, has data from the IIASA RAINS model (http://www.iiasa.ac.at/wrains) been used to obtain valid geographical information for spatially when gaps and/or errors were found. The SNAP-based emission distributing the emission data within a regular grid. Below, the input was subdivided into individual source contributions based on methodologies used for each SNAP sector are briefly described. IIASA’s RAINS model (Amann et al., 2005), which is a bottom-up TNO allocates nearly all of the emissions, from SNAP01 to point emission inventory for CH4,NH3,NOx, NMVOC, SO2, PM10 and sources, such as power plants, oil refineries and coke ovens. PM2.5. The detailed RAINS results were used to estimate how Regarding the first two elements, the use of the European Pollutant strong the contributing sources were in relation to each other, but Emission Register database (EPER) was combined with the World Electric Power Plants Database 2008 (WEPP, 2008; http://www. platts.com/) and the World Refinery Survey, 2006 (WRS, 2006). Table 1 This combination improved the geographical coverage of the SNAP (Selected Nomenclature for Air Pollution) sector IDs, full names and short names that were used in this paper. dataset and was helpful for assigning weight factors to each point source (i.e., power or crude capacity). SNAP Full name Short name Similarly, HERMES-DIS considered SNAP01 as a point source. sector Thus, the total emissions in each EMEP cell were disaggregated as a 1 Combustion in energy and Energy industries function of the number of industries that existed in each finer grid transformation industries cell. This disaggregation was conducted according to the European 2 Non-industrial combustion plants Residential/commercial combustion Pollutant Release and Transfer Register database, version 2.1 (E- 3 Combustion in manufacturing industry Industrial combustion PRTR v.2.1, 2010), which improved upon and replaced the EPER 4 Production processes Industrial processes database. In the case of EMEP cells without industries that are 5 Extraction and distribution of fossil Extraction/distribution linked to SNAP01, the corresponding emissions are disaggregated fuels and geothermal energy of fossil fuels 6 Solvent and other product use Solvent use according to the facility database developed for sectors SNAP 03 7 Road transport Road Transport and 04 (see below). 8 Other mobile sources and machinery Other mobile sources For TNO, all of the emissions under SNAP02 were distributed 9 Waste treatment and disposal Waste based on population density. A population map, which included 10 Agriculture Agriculture both urban and rural areas, was compiled from two related datasets 11 Other sources and sinks Other from the Centre for International Earth Science Information J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57 47

Table 2 Summary of the criteria used for the spatial disaggregation of HERMES-DIS and TNO emissions inventories by SNAP sector.

SNAP sector Type of sources considered Elements of spatial disaggregation

TNO HERMES-DIS TNO HERMES-DIS

SNAP01 Energy industries Point Point Power plants 1. Number of SNAP01 facilities per cell Oil refineries 2. Number of SNAP03_04 facilities per cell Coke ovens 3. Area cell balance SNAP02 Residential/commercial Area Area Population distribution 1. Fraction of urban land uses per cell combustion 2. Area cell balance SNAP03 Industrial combustion Point and Area Point SNAP03 facilities 1. Number of SNAP03_04 facilities per cell Population distribution 2. Area cell balance SNAP04 Industrial processes Point and Area Point SNAP04 facilities 1. Number of SNAP03_04 facilities per cell Population distribution 2. Area cell balance SNAP05 Extraction/distribution Point and Area Point SNAP05 facilities 1. Number of SNAP05 facilities/Ports/Airports per cell of fossil fuels Gas transport network 2. Number of petrol stations per cell Population distribution 3. Area cell balance SNAP06 Solvent use Area and Point Area Population distribution 1. Fraction of urban land uses per cell SNAP06 facilities 2. Area cell balance SNAP07 Road Transport Area (Linear) Area (Linear) Traffic flow map 1. Length of road type per cell (Interurban traffic) 2. Area cell balance Road Transport Area Area Population distribution 1. Fraction of urban land uses per cell (Urban traffic) 2. Area cell balance SNAP08 Other (non-road) Area and Point Area Sea port and airports 1. Fraction of port and airport land uses per cell transport Arable land uses 2. Fraction of agriculture and industrial land uses per cell Rail Network 3. Area cell balance Livestock distribution Population distribution Ship emissions Area Area EMEP gridded emissions Area cell balance based on the water Map of shipping traffic fraction of each cell on the North sea SNAP09 Waste treatment Area/Point Area Population distribution 1. Fraction of dump sites and industrial and SNAP09 facilities commercial units per cell 2. Area cell balance SNAP10 Agriculture Area Area Arable land uses 1. Fraction of agricultural land uses per cell Livestock distribution 2. Area cell balance

Network (CIESIN, http://sedac.ciesin.columbia.edu/gpw/). These plants). For SNAP05, the following additional information was datasets included the GPW version 3 (Gridded Population of the used: the main ports and airports in Europe, as reported by the World for 2005 in a longitude-latitude grid format with a resolu- Geographic Information System of the European Commission tion of 1/24th of a degree) and GRUMP (Global Rural-Urban Map- (GISCO) (in which the processes related to fossil fuel storage are Ò ping Project alpha version, for 2000 in a grid format with a important) and the petrol stations, provided by TeleAtlas Multinet resolution of 1/120th of a degree) datasets. 2010 (paid product provided by TomTom, http://www.tomtom. In the case of HERMES-DIS, the emissions were disaggregated com/en_gb/licensing/products/maps/multinet), where cars are using the CORINE Land Cover 2006 raster data, version 13, with a refuelled. This last disaggregation element has only been consid- resolution of 100 100 m2 (CLC06 hereinafter) (CLC, 2006). An ered in cells where the previous three types of data do not exist. assignment was conducted between the CLC06 land use types and This measure was taken due to the relatively minor importance of the SNAP sector, including all of the urban areas and the port and the last disaggregation element as an emissions source. airport areas due to the existence of non-industrial combustion Regarding SNAP06 (solvent use), in the case of TNO, a major plants in these sites. When CLC2006 v.13 was incomplete, CLC2000 portion of the emissions was distributed based on population v.13 was used for the UK and Greece. density, as in SNAP02 (residential/commercial combustion). How- TNO treated a large portion of the SNAP 03, 04 and 05 emissions as ever, minor parts were spatially assigned from the point source point sources. The geographical data include, among others, facilities information. In HERMES-DIS, the disaggregation was conducted as for cement production, non-ferrous metal production, the iron and in sector SNAP 02 because the main portion of emissions results steel industry, fossil fuel production and the chemical industry. A from domestic solvent use. combination of literature and branch organisation information was For the spatial distribution of road transport emissions (SNAP07), used, including the Iron & Steel Works of the World Directory 2007 TNO considers two available proxies, including the Trans-Tools (MBD, 2007) and the World Cement Directory 2002 (WCD, 2002). In traffic flow map and the population distribution (Trans-Tools, SNAP05, the emissions due to loss during transportation of natural 2006). Trans-Tools is an European transport network model that gas were assigned through the natural gas transport network that comprises passenger, freight and intermodal transport. Trans-Tools was developed within the European INOGATE project. has been manually expanded with GIS, based on the UNECE Euro- In contrast, HERMES-DIS considers all of these sectors as point pean road map, to include countries that were not previously fully sources and uses the E-PRTR v.2.1 database as in sector SNAP01. covered, such as Ukraine or Russia. The Trans-Tools focuses on inter- Based on the NACE economic activity description, two databases urban transport and only considers motorways and main roads. For were obtained. The first database includes both facilities that were short-range transport, the population is used. This method allocates related to SNAP03 and SNAP04 because the activities in these the fractions per substance and vehicle type per country. sectors can occur in the same facility (i.e., a chemical industry that The HERMES-DIS method first divided the road emissions of each has a combustion plant to generate its own energy). The second EMEP cell into urban and interurban. The spatial disaggregation of database includes plants related to SNAP05 (i.e., fuel/gas extraction interurban traffic emissions is conducted through GIS vectorised 48 J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57 road cartography of Europe that was developed by TeleAtlas Mul- country were obtained for the TNO and HERMES-DIS datasets. The Ò tinet 2010. This cartography includes six functional road classes, HERMES-DIS considers the same totals of EMEParea, since it is its from motorways to local roads of importance. The disaggregation basis. As a first approach, we found considerable differences be- criteria account not only for the road length that intersects each cell tween the inventories for all sectors. However, these discrepancies but also for the traffic volume, which is corrected for each type of were not only caused by spatial disaggregation but also by the road. The urban traffic emissions disaggregation is based on the area original inventories themselves. The base year was not always the of the cell that is occupied by urban land use from CLC06 (just as in same (2007 for TNO and 2008 for the EMEParea and HERMES-DIS). SNAP06). In addition, these methods used different source data and distinct TNO has regarded most of the emissions of sector SNAP08 as criteria regarding the spatial disaggregation of their emissions, as area sources. Sea harbours and airports have been the only point explained in Section 2. sources of information. Railway emissions have been distributed Fig. 2a illustrates the relative differences between the EMEP and equally across the European rail network throughout the country. TNO emission totals. Regarding the total emissions per country, the Moreover, emissions from mobile agriculture sources have been largest differences were observed for Italy, Spain, UK, Poland and disaggregated using arable land use as proxy data for. The premier France. Except for the UK, the TNO emissions were greater than datasets that were used include the CORINE land use dataset those of the HERMES-DIS. The reason for these differences varies by (supplemented by the PELINDA dataset (De Boer et al., 2000)) and country. For Spain, the majority of the differences were related to the global land cover dataset (Wilson and Henderson-Sellers, 1985) the SOx emissions from SNAP01. However, in Italy and the UK, the for two different areas outside EU27. Emissions from other mobile difference resulted from the CO emitted by road traffic (SNAP07). sources have been linked to population distribution. However, To visualise the relative contribution of each SNAP sector to the emissions that occur inland and in coastal waterways have been total emissions of each pollutant, the total EMEP and TNO emis- located in areas with similar descriptions. For the emissions from sions were plotted per sector and pollutant (Fig. 2b). sea shipping, gridded data of the sea shipping tracks from the EMEP For each pollutant, there are one or more activities (SNAP sec- were disaggregated by area for the TNO grid. For only the North Sea, tors) for which the contribution is often greater and are referred to the 50 50 km emissions were redistributed on a finer grid based as “key sectors”. The CO, PM10 and PM2.5 pollutants are mainly on a detailed map of shipping traffic(MARIN, 2003). emitted by residential/commercial combustion and road transport Regarding the HERMES-DIS method, emissions from SNAP08 (sectors SNAP02 and 07). NOx total emissions are mainly produced that occurred on land were remapped by using the port and air- from road transport and SOx from energy industries (SNAP01). ports areas from CLC06. When the EMEP cell was missing data, Solvent use and agriculture (SNAP06 and SNAP10) are the main agriculture and industry land uses were included, since off road contributors to NMVOC and NH3 emissions, respectively. Table 3 traffic in these areas was significant for this sector (agriculture and highlights the pollutants from each of the SNAP sectors that were industry mobile machinery). Conversely, the EMEP ship emissions selected for further analysis. PM10 and PM2.5 emissions (by sector) were remapped using area criteria that accounted for the fraction of had similar contributions to total emissions. Thus, only PM10 was water in each cell. considered further. SNAP sectors 03, 04 and 09 were assumed to be As in SNAP02 and SNAP06, TNO distributed most of the SNAP09 represented by CO emissions based on Fig. 2b. emissions based on population density with the previously described database. Conversely, the HERMES-DIS method used 3.2. Intercomparison methods different land use types from CLC06 v.13, including dump sites and industrial and commercial land uses. These sites usually have solid The prior analysis coarsely compared the emissions inventories waste treatment facilities. by country, sector, and pollutant. Next, the differences were iden- Finally, TNO spatially distributed all of the emissions from tified and the spatial distribution analysis was conducted at a SNAP10 among arable land and farm animals (livestock, including higher resolution (i.e., at grid level, by sector and by selected pol- chicken, poultry, cattle, pigs, and other animals). For the livestock lutants). After mapping these differences, other graphical and nu- distribution, the Gridded Livestock of the World (GLW) dataset merical approaches were used to refine and enrich the analysis. The from the Food and Agriculture Organisation (FAO) was used as graphical approaches applied in this study included the use of input. Similarly, the disaggregation in the HERMES-DIS was based emission maps, scanning data series, and scatter plots. The coeffi- on all agricultural land uses that were covered by CLC06. cient of determination was used as a metric in this analysis. For TNO, the disaggregation of residual emissions, which could not be allocated to a particular source, was conducted by using 3.2.1. Emission maps population density data. In contrast, a mass conservative area The gridded emissions of all pollutants from the 10 SNAP sectors interpolation was applied to the EMEP cells without spatial infor- for the three emission grids were graphically represented. The mation for HERMES-DIS (i.e., countries uncovered by the CLC06 differences among these grids were also quantified and plotted. The domain or E-PRTR). spatial representation of the emissions and the emission ranges allowed for the identification of the main visual similarities and 3. Inter-comparative analysis of emission inventories discrepancies, which were correlated to the spatial distribution of the respective driving variables, such as land use maps and emis- Prior analysis of the three spatially gridded emission inventories sion source locations. within the common grid was performed to identify differences among the datasets and to determine the most relevant sectors and 3.2.2. Line charts pollutants. In addition, this analysis was conducted to help define To visually represent the emissions with a scalable magnitude the most suitable method for the spatial intercomparison. rather than with a colour while maintaining the interrelationships between neighbouring cells, the emission numbers were displayed 3.1. Prior analysis on a line chart for each grid row (Goodwin et al., 1997). This data series scanning process was applied in the present work (i.e., by Emission totals were analysed by activity sector for EU27, as a producing line charts of the emissions vs cell number, and by first approach. The total emissions by sector and pollutant for each allowing a grid scan from the lower left corner to the right, West to J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57 49

Fig. 2. (a) Emission differences between the EMEP (2008) and the TNO (2007) datasets for all of the SNAP sectors, and (b) the total EMEP and TNO emissions per SNAP sector and primary pollutant.

East, and line by line from South to North). In this way, the absolute performed, including the representation of regression lines and the emission values from the different datasets could be directly calculation of the coefﬁcients of determination (r2). In this case compared for each grid cell (Winiwarter et al., 2003). (regular grid), all cells were given the same weight.

3.2.3. Linear regressions 4. Results and discussion The scatter between pairs of datasets was graphically represented by plotting the emission values for corresponding grid cells The three gridded emission datasets presented in Section 2 were e in an x y plot. In addition, a classical comparison of data pairs was compared based on the methods described in Section 3. This analysis was performed by SNAP sector for each of the pollutants in Table 3 the inventories. The following subsection will present the results Pollutant shares (%) per SNAP sector and the set of considered pollutants (grey that were obtained for the total pollutant emissions and for the shading) per SNAP sector for the analysis and presentation of results. pollutants that are denoted in Table 2 for each SNAP sector. SNAP sectors Pollutants

CO NOx NMVOC SOx NH3 PM10 PM2.5 4.1. Total emissions 1 Combustion in 3% 20% 1% 61% 0% 6% 5% energy industries The emission totals were compared for each pollutant by grid 2 Non industrial 30% 7% 11% 10% 0% 29% 37% cells of 12 12 km2. First, a data scanning was conducted to combustion plants identify the main differences between the two more refined grid- 3 Combustion in 12% 11% 1% 16% 0% 8% 9% manufacturing industries ded inventories, the HERMES-DIS and the TNO. These discrepancies 4 Production processes 11% 3% 10% 7% 1% 11% 9% were investigated in more detail through differences mapping and 5 Fossil fuel production 0% 0% 6% 1% 0% 6% 5% by using the EMEParea inventory as the base because it is the 6 Solvent and product use 0% 0% 41% 0% 0% 3% 2% simplest and coarsest inventory. In addition, the spatial correlations 7 Road Transport 34% 41% 16% 0% 2% 14% 16% fi 8 Other (non-road) 7% 16% 5% 5% 0% 8% 11% between the ner (HERMES-DIS and TNO) and base (EMEParea) transport inventories were analysed. 9 Waste treatment and 2% 0% 1% 0% 2% 2% 3% disposal 4.1.1. Data series scanning and spatial differences 10 Agriculture 2% 2% 6% 0% 94% 13% 5% The grid data from each of the three emission datasets was Totals UUU UUU U plotted against the cell number to obtain a line graph of absolute 50 J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57 emissions vs cell number. These line charts can be interpreted as 4.2. Emissions by SNAP sector the spatial variation of emissions from the southwest to the northeast of the domain, since the grid cell ID ranges from 0 (the The methodology that was followed for the inter-comparative lower left corner of the domain) to 191,999 (the top right cell of the analysis of emission totals was applied for each of the 10 SNAP study area) (cell ID ¼ 1 is east of cell ID ¼ 0, and so on). The latitude sectors. This methodology included data series scanning, the increases from left to right, and lines are “coloured” (grey scale) computation and mapping of emission differences and regression according to longitude and highlight the meridional variation. For analysis. all pollutants, the finer disaggregated inventories were more detailed in space and (except for NH3) exhibited greater emissions 4.2.1. Data series scanning and spatial differences than those of the EMEParea. These disaggregated inventories To simplify the analysis, the data scanning and maps are pre- reached values, for CO and PM10, that were one order of magnitude sented and discussed by SNAP sector. In addition, this analysis fo- greater than those in the EMEParea. The EMEParea emission line cuses on the sectors that contribute most to the total emissions of charts are not presented. Although these line charts clearly show the most relevant pollutants, based on previous analysis. Therefore, the coarser data details, they are not beneficial for the comparative only sectors SNAP01 (energy industries) (relevant for SOx and NOx), analysis of the disaggregation methodologies. Between the two SNAP 02 (residential/commercial combustion) (CO and PM10), more detailed inventories, the TNO inventory presents maximum SNAP06 (solvent use) (NMVOC), SNAP07 (road transport) (CO and emissions that are two times greater than those of the HERMES-DIS NOx) and SNAP10 (agriculture) (NH3) will be addressed. for most pollutants. Fig. 3 provides examples of HERMES-DIS and Figs. 4 and 5 contain maps of the differences that were ob- TNO line graphs for the CO, NOx, PM10 and PM2.5 pollutants. The tained for the same sectors and pollutants by subtracting the TNO origins of these pollutants are not related to a single sector, but emissions from those of the HERMES-DIS and the EMEParea rather to different sectors. These plots show different magnitudes emissions from those of the HERMES-DIS (positive values indicate in space for all of the pollutants shown. The observed discrepancies that the HERMES-DIS was greater than the TNO or EMEParea, can be better characterised and further investigated by mapping respectively). In the first case (HERMES-DIS vs TNO), only the the differences between the EMEParea and HERMES-DIS data and EMEP database reports emissions in the North Africa common the HERMES-DIS and TNO data. Total emission maps that show domain, which implies that all of the maps will show positive these differences were generated for all analysed pollutants. values in this zone. However, based on these figures, it was concluded that the spatial As stated previously, the EMEParea and HERMES-DIS results variability evaluation of emissions should be conducted by sector. were derived from the original EMEP inventory, whose emissions Thus, the maps related to total emission differences are not pre- were already spatially allocated on a 50 50 km grid. Conversely, sented here. the TNO results are based on total emission values per country, which are subsequently disaggregated with an accurate method. In 4.1.2. Regression relationships some cases, this difference could cause different results regardless The spatial qualitative analysis of the three gridded emission of the disaggregation methodology and information used in each datasets provided a clear picture of the differences that were particular case. encountered and was used to understand the impacts of the Sector SNAP01 (energy industries) comprises large point sour- different disaggregation methodologies with distinct input data on ces, with a specific location, emitting high quantities of NOx and SOx the spatial variability of emissions. However, a quantitative analysis and thus, gridded emissions, originally on 50 50 km grid cells could be useful for determining the overall extent of the spatial (EMEParea, not presented), do not represent the actual spatial correlation between the emission inventories. pattern of this sector. Globally, TNO emissions are much greater Scatter plots for all of the pollutants were produced to investi- than HERMES-DIS emissions for both NOx and SOx. Differences gate the spatial correlations between EMEParea and HERMES-DIS, were noticeable between the HERMES-DIS and TNO emissions over EMEParea and TNO, and HERMES-DIS and TNO. In addition, linear the entire area and especially in the northeastern part of the fits of the scatter plot data were determined. Table 4 summarises domain where peaks were found in different locations. In most the coefficients of determination (r2) that were obtained from the cases, these cells belong to Russia and Ukraine, where E-PRTR 2008 linear regressions. v.2.1 has no information. Thus, EMEP emissions are disaggregated As expected, the greatest correlations for all pollutants were throughout the area. found between EMEParea and HERMES-DIS, which indicated that The difference maps for SOx (Fig. 4.1) confirm that emissions (besides the differences identified in the maps) these two in- from this sector were assigned to different cells within the two ventories had greater spatial agreement, which was true since disaggregation methodologies. In addition, the greatest negative EMEParea and HERMES-DIS have the same basis. differences are located, in most cases, in Eastern Europe. These The spatial distribution of NH3 exhibited the highest correla- differences are caused by two elements that are related to the tions for all three of the studied cases, which indicated that it had temporality and coverage of the databases that were used in each fewer spatial differences among the inventories. This pollutant is case (see Section 2, Table 2). While TNO complements the EPER emitted by SNAP sector 10 activities and land use was the main 2004 information with the WEPP database, the results of the variable that was considered for emission disaggregation in both HERMES-DIS method are derived from the E-PRTR 2008 v.2.1 HERMES-DIS and TNO. Moreover, the land use type of “agriculture” database, which only covers the EU27 domain. Hence, some of the was not zero in most grid cells. facilities that are included in the TNO inventory may not be in use. The poor correlations between HERMES-DIS and TNO empha- This fact, along with the higher coverage of the TNO inventory, sise the fact that these are two distinct emission inventories. which includes areas such as the Balkan countries, explains most of Although both inventories are discrete, detailed, and suitable for air the observed differences. In other cases, shown in the maps as quality modelling, they may lead to different air pollution con- positive and negative differences in contiguous cells, these differ- centration patterns. However, additional detailed analysis, by SNAP ences could result from the readjustment of the original TNO grid to sector, is needed to verify this overview regarding the impact of the common grid. distinct disaggregation methodologies on the spatial variability of CO and PM10 emissions from residential/commercial com- air pollutant emissions. bustion had similar patterns between the two individual data .Frer ta./AmshrcEvrnet7 21)43 (2013) 75 Environment Atmospheric / al. et Ferreira J. e 57

Fig. 3. Line charts for the emission inventory data series, HERMES-DIS and TNO, for CO, NOx, PM10 and PM2.5 emissions from all SNAP sectors. Grid cells ranged from 0 (the southwest corner of the domain) to 191,999 (the northeastern cell of the study area, cell ID ¼ 1 is east of cell ID ¼ 0, and so on). The latitude increases from left to right. The “coloured” (grey scale) lines highlight the meridional variations. 51 52 J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57

Table 4 series. However, very different spatial variability was verified Coefficients of determination obtained by the three computed types of linear spatial between the HERMES-DIS and TNO methods. The maximum TNO correlations, including EMEParea vs HERMES-DIS, EMEParea vs TNO and HERMES- and HERMES-DIS emissions were generated in different areas of DIS vs TNO. the study domain. The TNO emissions were four times higher fi 2 Coef cients of determination (r ) than the HERMES-DIS emissions for both pollutants. These results

CO NOx NMVOC NH3 PM10 PM2.5 SOx suggest that the TNO emissions are more concentrated in speciﬁc EMEParea vs 0.29 0.45 0.52 0.81 0.53 0.48 0.28 cells, and the HERMES-DIS emissions are distributed more HERMES-DIS homogenously. The maps (Fig. 4.2) support this hypothesis. In and EMEParea vs TNO 0.17 0.15 0.25 0.42 0.09 0.09 0.02 near urban areas, such as Barcelona, Paris or London, negative HERMES-DIS vs TNO 0.14 0.21 0.33 0.42 0.10 0.10 0.04 values were observed. In contrast, HERMES-DIS had greater

Fig. 4. Differences maps of (HERMES-DIS e EMEParea) (a) and (HERMES-DIS e TNO) (b), for SOx emissions from SNAP01 (1) for CO emissions from SNAP02 (2), and for NOx emissions from SNAP07 (3) per 12 12 km2 grid cell. J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57 53

Fig. 5. Differences maps of (HERMES-DIS e EMEParea) (a) and HERMES-DIS -TNO) (b) for the NMVOC emissions from SNAP06 (1) and the NH3 emissions from SNAP10 (2) per 12 12 km2 grid cell.

emissions in rural areas. These differences are likely caused by the main highways. The two following facts explain these differences: different disaggregation criteria that were used in each case (see the detailed road network information that is available in the Section 2, Table 2). For example, the TNO methodology uses the HERMES-DIS model, which can assign emissions to a greater number of inhabitants per cell, while the HERMES-DIS method- number of cells, and the correction factor assigned to each road ology uses the percentage of urban land use in each cell. The type in the HERMES-DIS model, which significantly differs from the homogenous differences shown in Bosnia, Herzegovina and Serbia traffic intensity in the TNO. and Montenegro are caused by the GPWv3 input data in this area, For cells situated in the greater urban areas, positive and which has a low spatial resolution due to the few administrative negative results exist, depending on the city. For example, in units used for the population distribution (Bak and Yetman, Madrid or Barcelona, the TNO reports greater emissions. However, 2004). Considering these results and the variability of the emis- the opposite result occurs for Paris. This variation could be caused sion data input for EMEP and TNO, in greater urban areas the by the percentage of total traffic emissions that are assigned to population disaggregation element is more important than the urban traffic. In the HERMES-DIS case, these fractions are the same urban land use element. in every EMEP cell (see Section 2). However, in the TNO case, the Road traffic (SNAP07) emissions are presented in Fig. 4.3 (dif- assumption changes from country to country. ferences maps). For this sector, the magnitudes of the CO and NOx NMVOCs are the major pollutants that are emitted from SNAP06 emissions were similar for the two gridded inventories. For all data activities. The TNO emissions are roughly two times greater than series, the greatest emissions were recorded for the high longitude/ the HERMES-DIS emissions per grid cell. In addition, the HERMES- latitude areas. From Fig. 4.3, the areas that correspond to the largest DIS spatial distribution pattern is denser, which results from the positive differences are located in Russia. As stated in Section 2 assignment of greater emissions in more areas than in the TNO (Table 2), the HERMES-DIS and TNO inventories for road traffic emissions. Regarding the calculated spatial differences (Fig. 5), the were based on distinct data sources. TeleAtlas includes a more results suggest a similar pattern to the one observed in sector detailed road network than the Traffic Flow map from the TransTool SNAP02. In this case, TNO emissions were more concentrated project because it contains secondary and local roads along with where large cities are located. However, in the remaining areas, the the motorways and main roads. majority of the differences were positive. For Poland, these high In the western European countries, such as Spain, France or negative values occurred throughout the entire territory due to the Germany, the TNO emissions are greater in the cells that cover the different emission data sources (the TNO inventory reports higher 54 J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57

Fig. 6. Scatter plots of the three linear regressions performed (EMEParea vs HERMES-DIS, EMEParea vs TNO and HERMES-DIS vs TNO) for SNAP01 (SOx), SNAP02 (CO), SNAP06 (NMVOC), SNAP07 (NOx) and SNAP10 (NH3) (as examples). J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57 55 values). As in SNAP02, the spatial disaggregation was conducted by and industrial combustion sectors where Large Point Sources in- population distribution (TNO) and urban land use (HERMES-DIS), fluence the results. which reinforced the idea that the differences in the maps were As previously determined from the total emissions analysis, a mainly caused by the different data sources. greater correlation occurred in the EMEParea vs HERMES-DIS SNAP sector 10 is mainly represented by NH3 emissions from regression because the HERMES-DIS emissions were based on the agriculture. Although the driving parameter for disaggregation EMEParea emission, which does not occur for the TNO emissions. (CLC) is the same for both HERMES-DIS and TNO, considerable For sectors SNAP02 and 06, the coefficients of determination differences regarding the spatial distributions of NH3 emissions between the HERMES-DIS and EMEParea were very similar occurred. Nevertheless, the specific agriculture land uses selected because the land use types in each of these cases were practically from the CLC in each case may differ. As previously mentioned the same (see Section 2). Nevertheless, the HERMES-DIS vs TNO (Section 2, Table 2), TNO has also considered farm animal (live- regression had a greater correlation in SNAP06. In both sectors, stock) distribution as a disaggregation variable that helps justify the TNO used population distribution as a default proxy to distinguish verified differences. The highly defined resolution of the results between urban and rural areas. The inclusion of port and airport in Russia made it possible to detect complementary land cover areas along with urban land uses in the HERMES-DIS disaggrega- datasets used in the TNO to cover the entire model domain (see tion methodology for sector SNAP02 may cause this lower corre- Section 2). lation because these two land uses are not associated with population distribution. 4.2.2. Regression relations Coefficients of determination were not calculated for the road Three types of spatial correlations were calculated for each SNAP transport (SNAP07) regressions because the multi linear pattern (a sector and respective pollutants displayed in Table 3 to comple- single value from EMEParea corresponds to a set of values in the ment the comparative analysis. Scatter plots for EMEParea vs other two datasets) of the scatter plot indicated that a single co- HERMES-DIS, EMEParea vs TNO and HERMES-DIS vs TNO were efficient was not accurate. This behaviour was explained by the made and some examples are presented in Fig. 6. The respective linear shape of the SNAP07 emission sources, which is not observed coefficients of determination that were obtained are plotted in in the spatial distribution of the EMEParea emissions because they Fig. 7. originated from a 50 50 km cell size (a resolution that is too For energy industries (SNAP01) and for the three types of coarse to differentiate the line sources). However, for all pollutants, regression analyses, the plotted points are scattered, which in- the HERMES-DIS vs TNO had high coefficients of determination (see dicates that the spatial variability of SOx differs among the emis- Fig. 7). This result can be explained by the similarity of the disag- sions that were determined with different disaggregation methods. gregation elements that were used in both methodologies, which This finding is supported by the low coefficients of determination mainly consisted of traffic road networks with varying degrees of that were obtained (Fig. 7). Emissions from energy industries, in- detail. dustrial combustion, industrial processes and the extraction/dis- As previously noted, the greatest correlations were encoun- tribution of fossil fuels (SNAP01, 03, 04 and 05) resulted in the tered for SNAP10 NH3 emissions due to the use of the CLC data- lowest coefficients. All activities in these sectors were considered as base for disaggregation by both emission inventories. The point sources. Thus, their total disaggregated emissions were coefficients of determination were farther from unity (as dis- concentrated in only a few cells. If the location of the point ele- cussed previously regarding the spatial differences) because the ments used in each methodology differs, then the difference be- TNO uses additional information to spatially disaggregate the tween the results becomes substantial, especially for the energy emissions.

Fig. 7. Coefﬁcients of determination obtained for the three linear regressions performed (EMEParea vs HERMES-DIS, EMEParea vs TNO and HERMES-DIS vs TNO) for the pollutants analysed by SNAP sector (according to Table 3). 56 J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57

5. Conclusions Acknowledgements

This paper focused on the analysis of two disaggregated emis- The authors are thankful to Hugo Denier van der Gon for his sion inventories with high resolution. These inventories are support regarding the TNO emission inventory. currently available in Europe and are being used in regional air The CRUP and the Ministerio de Ciencia e Innovación of Spain quality modelling applications. The first one is the HERMES-DIS are also acknowledged for supporting the Integrated Action E 122- emission inventory, which is a gridded emission dataset derived 10/PT2009-0029. from the EMEP 2008 annual emission inventory by applying spatial In addition, the authors thank the Portuguese ‘Ministério da disaggregation methods to each activity sector. The second emis- Ciência, da Tecnologia e do Ensino Superior’ for financially sup- sion inventory was developed by TNO for 2007 for modelling porting the EMOSAT research project (PTDC/CTE-ATM/103253/ purposes based on the official emissions reported by country and 2008, FCOMP-01-0124-FEDER-009305) and the postdoc grant for J. was disaggregated by sector. To perform a comparative analysis of Ferreira (SFRH/BPD/40620/2007). the different disaggregation methods used in each case, a common grid of 12 12 km2 was considered. Hence, the original EMEP Appendix A. Supplementary data gridded inventory was converted to the common grid with a simple area disaggregation to compare with the other two datasets and to Supplementary data related to this article can be found at http:// evaluate the importance of the detailed disaggregation procedure dx.doi.org/10.1016/j.atmosenv.2013.03.052. and its potential impacts on air quality modelling studies. The comparative analysis was performed with qualitative (line charts and differences mapping) and quantitative (linear regres- References sion) tools. This analysis focused on total emissions and on emis- Amann, M., Bertok, I., Cabala, R., Cofala, J., Heyes, C., Gyarfas, F., Klimont, Z., sions by activity sector, following the European Environmental Schöpp, W., Wagner, F., 2005. A Further Emission Control Scenario for the Clean Agency’s Selected Nomenclature for Air Pollution (SNAP). Air For Europe (CAFE) Programme. International Institute for Applied Systems The analysis of emission totals showed that the TNO dataset Analysis (IIASA). presented greater maximum emissions then HERMES-DIS dataset Bak, D., Yetman, G., 2004. Gridded Population of the World. Methodological Documentation. In: The Global Distribution of Population: Evaluating the Gains for most of the considered pollutants. Spatial differences between in Resolution Refinement. the two gridded datasets were also encountered and were further Baldasano, J.M., Güereca, L.P., López, E., Gassó, S., Jimenez-Guerrero, P., 2008. investigated by SNAP sector. Development of a high-resolution (1 km _ 1 km, 1 h) emission model for Spain: the high-elective resolution modelling emission system (hermes). Atmospheric The spatial analysis of the results by SNAP sector indicated that, Environment 42, 7215e7233. for residential/commercial combustion (SNAP02) and solvent use Baldasano, J.M., Pay, M.T., Jorba, O., Gassó, S., Jiménez-Guerrero, P., 2011. An annual (SNAP06), the TNO and HERMES-DIS maximum emissions were assessment of air quality with the CALIOPE modelling system over Spain. Sci- ence of the Total Environment 409, 2163e2178. generated in different areas of the domain. Thus, the TNO emissions Butler, T.M., Lawrence, M.G., Gurjar, B.R., van Aardenne, J., Schultz, M., Lelieveld, J., of CO and PM10 were four times greater than the HERMES-DIS 2008. The representation of emissions from megacities in global emission in- emissions. These results suggest that TNO emissions are more ventories. Atmospheric Environment 42, 703e719. http://dx.doi.org/10.1016/ j.atmosenv.2007.09.060. concentrated in the greater urban areas, while HERMES-DIS emis- CLC (CORINE Land Cover), 2006. v.13. European Environmental Agency. http://www. sions are distributed more homogenously. This difference is related eea.europa.eu/data-and-maps/data/corine-land-cover-2006-raster. to the spatial disaggregation that was conducted through popula- De Boer, M., de Vente, J., Mücher, C., Nijenhuis, W., Thunnissen, H., 2000. PELCOM INDAVOR, Instrument for Detecting Land Cover Change for Europe. NPRS-2 tion distribution (TNO) and urban land use (HERMES-DIS). project no. 4.2/DE-03, BCRS report, Delft. Regarding road transport (SNAP07), the HERMES-DIS and TNO Denier van der Gon, H., Visschedijk, A., Van de Brugh, H., Dröge, R., 2010. A High emissions were more detailed and had greater spatial variability Resolution European Emission Data Base for the Year 2005. In: A Contribution “ ” e than the EMEParea emissions data. The differences that occurred to UBA-Projekt: Strategien zur Verminderung der Feinstaubbelastung PAREST: Partikelreduktionsstrategien e Particle Reduction Strategies. TNO between the first two inventories were caused by their distinct data report. TNO-034-UT-2010e01895_RPT-ML. June 2010. sources. E-PRTR v.2.1, 2010. The European Pollutant Release and Transfer Register. http:// Regarding the regression analysis, greater correlations were prtr.ec.europa.eu/Home.aspx. fi EDGAR, 2011. Emissions Database for Global Atmospheric Research (EDGARv4.2). veri ed for the EMEParea vs HERMES-DIS regression because the http://edgar.jrc.ec.europa.eu/ (last accessed 20.11.12.). latter is based on the first, which does not occur for the TNO EEA, 2008. Aggregated and Gap Filled Air Emission Data, Source: the European emissions. The highest correlations were encountered for the Topic Centre on Air and Climate Change (date of delivery 06 Sep 2007). http:// www.eea.europa.eu/data-and-maps/data/eea-aggregated-and-gap-filled- SNAP10 NH3 emissions, due to the common use of the CLC data- airemission-data-1. base for disaggregation. In addition, the SNAP01, 03, 04 and 05 EEA, 2010. European Union Emission Inventory Report 1990e2008 Under the emissions (from point sources), led to the lowest coefficients of UNECE Convention on Long-range Transboundary Air Pollution (LRTAP). EEA Technical report No 7/2010. http://www.eea.europa.eu/publications/european- determination. The spatial variability of SOx differed among the union-emission-inventory-report. emissions that were obtained from the different disaggregation EMEP, 2007. EMEP/CORINAIR Emission Inventory Guidebook e 2007. EEA Technical methods. Report No 16/2007. European Environment Agency, Copenhagen. http://reports. eea.europa.eu/EMEPCORINAIR5/en/. In conclusion, HERMES-DIS and TNO are two distinct emission EMEP e CEIP, 2010. Emissions Used in EMEP Models. Centre on Emission Inventories inventories, both very detailed and suitable for air quality model- and Projections. http://www.ceip.at/emission-data-webdab/emissions-used-in- ling. However, the different databases and disaggregation methods emep-models/. that are used to generate different emissions regarding both Goodwin, J., Eggleston, H.S., Stedman, J., 1997. Atmospheric Emissions Inventory 1995: Detailed Spatial Emission Estimates and Method, AEAT- magnitude and spatial variability certainly lead to different air 1835yRAMPy20090001. AEA Technology, NETCEN, Culham, UK. pollution concentration patterns. This fact should be considered Maes, J., Vliegen, J., Van de Vel, K., Janssen, S., Deutsch, F., De Ridder, K., Mensink, C., when using these datasets in regional atmospheric models. Future 2009. Spatial surrogates for the disaggregation of CORINAIR emission inventories. Atmospheric Environment. ISSN: 1352-2310 43 (6), 1246e1254. work will focus on the evaluation of air quality models performance http://dx.doi.org/10.1016/j.atmosenv.2008.11.040. and sensitivity to these spatial discrepancies in emission in- Mareckova, K., Wankmueller, R., Anderl, M., Poupa, S., Wieser, M., 2009. Inventory ventories. Air quality modelling would benefit from the availability Review 2009. Technical Report CEIP 1/2009. In: Emission Data Reported Under the LRTAP Convention and NEC Directive. Stage 1 and 2 Review. Status of of consistent and reliable emissions inventories with appropriate Gridded Data and LPS Data, ISBN 978-3-99004-038-6. http://www.ceip.at/ spatial and temporal resolution. fileadmin/inhalte/emep/pdf/2009/Inventory_Review_2009.pdf. J. Ferreira et al. / Atmospheric Environment 75 (2013) 43e57 57

MARIN, 2003. Emissieberekening routegebonden scheepvaartverkeer op NCP, op Distribution of the Diffuse Emissions in Europe. European Commision. http:// basis van aangepaste bronbestanden en emissiefactoren. MARIN Rapport Nr. www.eea.europa.eu/data-and-maps/data/european-pollutant-release-and- 18717.620/1 (in Dutch). transfer-register-e-prtr-regulation-art-8-diffuse-air-data. MBD, 2007. Metal Bulletin Directories. CD-ROM, seventh ed. Iron & Steel Works of Trans-Tools, 2006. Tools for Transport Forecasting and Scenario Testing, EU FP6 the World, London. Project. http://ec.europa.eu/research/fp6/ssp/trans_tools_en.htm. Pay, M.T., Piot, M., Jorba, O., Gassó, S., Gonçalves, M., Basart, S., Dabdub, D., Jiménez- UNECE, 2009. Present State of Emission Data, ECE/EB.AIR/97. Guerrero, P., Baldasano, J.M., 2010. A full year evaluation of the CALIOPE-EU air Van Aardenne, J.A., Dentener, F.J., Olivier, G.J., Peters, J.A.H.W., 2005. The EDGAR 3.2 quality modelling system over Europe for 2004. Atmospheric Environment. ISSN: Fast Track 2000 Dataset (32FT2000), p. 12. 1352-2310 44 (27), 3322e3342. http://dx.doi.org/10.1016/j.atmosenv.2010.05.040. Visschedijk, A.H.J., Denier van der Gon, H.A.C., 2005. Gridded European Anthro- Pouliot, G., Pierce, T., Van der Gon, H.D., Schaap, M., Moran, M., Nopmongcol, U., pogenic Emission Data for NOx,SO2, NMVOC, NH3, CO, PM10, PM2.5 and CH4 for 2012. Comparing emission inventories and model-ready emission datasets the Year 2000. TNO B&O-A Rapport 2005/106. between Europe and North America for the AQMEII project. Atmospheric Visschedijk, A.J.H., Zandveld, P.Y.J., Denier van der Gon, H.A.C.A., 2007. High Reso- Environment. ISSN: 1352-2310 53, 4e14. http://dx.doi.org/10.1016/j.atmosenv. lution Gridded European Emission Database for the EU Integrate Project GEMS. 2011.12.041. TNO-report 2007-A-R0233/B. Reis, S., Sutton, M., Zhang, M., Pinder, R., Lijie, G., 2009. Reactive nitrogen in at- WCD, 2002. World Cement Directory 11th Edition, 2002 (WCD 2002) CEMBUREAU. mospheric emission inventories. Atmospheric Chemistry & Physics 9, 7657e The European Cement Association, Brussels. 7677. WEPP, 2008. World Electric Power Plants Database. (Available from: http://www. Reis, S., Zhang, M., Lijie, G., Sutton, M., 2008a. Reactive and non-reactive species of platts.com/). nitrogen in atmospheric emission inventories e a review. In: 17th Annual In- Wilson, M.F., Henderson-Sellers, A., 1985. A global archive of land cover and soils ternational Emission Inventory Conference ‘Inventory Evolution e Portal to data for use in general circulation climate models. Journal of Climate 5, 119e Improved Air Quality’. Portland, Oregon, USA. June 2e5. 143. Reis, S., Pfeiffer, H., Scholz, Y., Theloke, J., 2008b. Temporal and spatial distribution of Winiwarter, W., Dore, C., Hayman, G., Vlachogiannis, D., Gounaris, N., Bartzis, J., carbon emissions. In: Dolman, H. (Ed.), The Continental-scale Greenhouse Gas Ekstrand, S., Tamponi, M., Maffeis, G., 2003. Methods for comparing gridded Balance of Europe. Springer Publishers, p. 390. inventories of atmospheric emissions e application for Milan province, Italy Simmons, A., Granier, C., Rayner, P., Schultz, M., Boucher, O., Engelen, R., Peuch, V.- and the Greater Athens Area, Greece. The Science of the Total Environment 303, H., Eskes, H., Rouïl, L., Tarrason, L., Chevallier, F., Flemming, J., Morcrette, J.-J., 231e243. Schulz, M., January 2010. A Monitoring and Forecasting System for Atmospheric WRS (World Reﬁnery Survey) 2006. Oil & Gas Journal, Online Research Centre Composition. Final report of the GEMS project. In: FP6 Research Project e http://www.pennenergy.com/index/research-and_data/oil-and_gas.html. Global and Regional Earth-system Monitoring Using Satellite and in Situ Data. Zhang, Y., Seigneur, C., Bocquet, M., Mallet, V., Baklanov, A., 2012. Real-time air Theloke, J., Thiruchittampalam, B., Orlikova, S., Uzbasich, M., Gauger, T., 2011. Diffuse quality forecasting, Part I: history, techniques, and current status. Atmospheric Air Emissions in E-PRTR e Methodology Development for the Spatial Environment 60, 632e655.