Recently, improvements were made in the spatial distribution of European emission data, as well as in completeness of country emissions in Europe (Pouliot et al., 2012; Kuenen et al., 2011). The spatial distribution used in the present study is a 1/8∘×1/16∘ longitude–latitude grid. The area domain is Europe from -10 to +60∘ Long and +35 to +70∘ Lat (excluding Kazakhstan and the African continent, but including Turkey). The set of gridding tools used in this study is described in Denier van der Gon et al. (2010). The exception is residential wood combustion for which a new distribution map has been compiled (see Sect. 2.3.1). For gridding a distinction is made between point and area sources. Point source emissions are distributed according to location, capacity and fuel type (when applicable). Area sources are distributed using distribution maps of proxy data such as population density. For a detailed description of the gridding we refer to Denier van der Gon et al. (2010). The point sources and area sources used to distribute the emissions for individual source categories are presented in the Supplementary material Table S1. The emission inventory database provides the emissions at a detailed level of about 200 sub-source categories. Each subcategory was spatially distributed using the most appropriate proxy map and then aggregated to Standardized Nomenclature for Air Pollutants (SNAP) level 1 source categories (Table 1).

Wood, woody biomass and wood pellets are extensively used as fuel in European households. However, reliable fuel wood statistics are difficult to obtain because fuel wood is often non-commercial and falls outside the economic administration. Therefore, fuel wood consumption has been notoriously underestimated in the past. Since combustion of wood is a key source of EC and OC we improved the available wood usage data through a stepwise approach. Specific wood use by country (GJ person-1) was primarily taken from GAINS. Estimates from the International Energy Agency (IEA, 2008) were used when GAINS data were lacking. By comparing industrial and residential use of fuel wood in GAINS we conclude that only the residential use is important on a European scale; industry and power generation both consume less than 1 % of the total amount of wood used annually in Europe (IEA, 2008; GAINS, 2009). Moreover, combustion appliances in the residential sector have much higher PM emission factors per unit of fuel. Therefore, our focus is on residential combustion of wood and we neglect its minor use as a fuel in industrial combustion or power generation here.

Grouping the available statistical data resulted in five country cluster averages, based on geographical location and tradition, with wood use varying between 1.6 and 8.6 GJ person-1 (Fig. 2). The observed differences between countries and country clusters can be related to the availability of local sources of fuel wood. We define “wood availability” by the geographical intersection (arithmetic product) of population and local fuel wood sources, modelled by overlaying a map of gridded population on 1/16∘×1/8∘ resolution (from the Center for International Earth Science Information Network, CIESIN; for details see Denier van der Gon et al., 2010) with gridded land cover of woodlands (combination of land-cover maps described by Denier van der Gon et al., 2010) on a coarser 0.25∘×0.5∘ resolution and subsequent summing by country. Based on analysis of reported specific wood use versus expected specific wood use (given a country's estimated average wood availability and cluster average specific consumption) we singled out and corrected wood use for countries where reported usage data substantially differed from expected wood use. In addition, we made default usage estimates based on country cluster averages and wood availability for countries with no reported wood use. Ultimately for the Slovak Republic, Cyprus, Malta, Russia, Ukraine, Moldova and the Caucasus states, specific wood consumption was based on our own estimates of specific consumption instead of using GAINS/IEA data. For documentation of the underlying assumptions we refer to Visschedijk et al. (2009). The estimated residential fuel wood use by country is presented in Fig. 2. Total wood use in UNECE Europe after reviewing the activity data and gap filling was about 20 % higher than the old data set.

Various types of appliances are used in Europe for residential wood combustion and this has a significant impact on the EC/OC and PM emissions. In this study we adopt the split in appliance types given by Klimont et al. (2002) and Kupiainen and Klimont (2007) who distinguished seven appliance types and provided relative shares of their use in different countries. In terms of emission of particulate matter these technologies were ranked:

Fireplace > Conventional stove > Newer domestic stoves and manual single house boilers > Automatic single house boilers and 50–100 kW medium boilers > 1–50 MW Medium boilers.

Especially the fraction of fireplaces and conventional stoves has important implications for the PM/EC/OC emission because of the corresponding relatively high emission factors (Kupiainen and Klimont, 2007). For countries within our domain where no ratios between different appliances were given by Klimont et al. (2002) we used values for neighbouring or comparable countries (see Table S2). For several Eastern European countries the wood usage of fireplaces was reported as 0 %, we adjusted this by assuming 5 % application in fireplaces (the country cluster average). From the activity data for fuel wood consumption by appliance type by country, it is evident that Western European countries with a relatively high use of fuel wood also have the highest market penetration of more modern combustion equipment.

The activity data described earlier, in combination with the adjusted allocation of wood by appliance type were used to develop a revised RWC emission inventory by selecting emission factors for each appliance type, independent of the country (Table 2). This is a first-order approach because it neglects the importance of combustion conditions and “cultural” differences in how to burn wood. Nevertheless it leads to a more transparent and comparable emission inventory.

Emission factors for wood combustion vary widely even for the same appliance type. This is partly due to the influence of combustion type, fuel parameters and different operation conditions. However, another important factor is the different sampling and measurement protocols or techniques. Nussbaumer et al. (2008a, b) made a detailed survey and review of the various emission factors in use in Europe, also in relation to the type of measurement technique. A total of 17 institutions from seven countries (Austria, Denmark, Germany, Norway, The Netherlands, Sweden and Switzerland) participated in the survey and contributed data to the questionnaire. In addition, data for national emission factors were reported or gathered from the literature.

Nussbaumer et al. (2008a, b) describe various sampling methods and the respective emission factors. The most important are filter measurements, measuring only solid particles (SP), and dilution tunnel (DT) measurements, measuring solid particles and condensable organics (or semivolatile organics). An example of the latter is the Norwegian standard NS 3058-2 which samples filterable particles in a dilution tunnel with a filter holder gas temperature at less than 35 ∘C and at small dilution ratios (DR) of the order 10. Due to the cooling, condensable organic material in the hot flue gas condenses on the filter or the solid particles. The impact of the choice of SP or DT emission factors is large, as illustrated in detail in Table 2. For example, for conventional woodstoves, one of the most important categories in Europe, the average solid particle emission factor is 150 gGJ-1 (range 49–650) whilst the average of the dilution tunnel measurements, that include both solid and condensable particles, is 800 gGJ-1 (range 290–1932). This implies a factor of 5 difference between the absolute PM emissions depending on the choice to use an SP- or DT-based emission factor. National emission factors, used in official reporting, show a considerable range, even if they are of the same type (DT or SP), as is reflected in the range presented in Table 2 and documented in detail in Nussbaumer et al. (2008a, b). In the TNO-newRWC emission inventory, the average DT emission factors were used for the respective appliance types (Table 2); for all other EC and OC emissions sources the EUCAARI emission values (Visschedijk et al., 2009; Kulmala et al., 2011) remained unchanged; in Fig. 1 only the sector SNAP 2-wood is different. The result was a revised inventory with a consistent approach for residential wood combustion, independent of individual country emission factor choices used for official reporting. A detailed example is presented in Sect. 4.3.

It should be noted that we revised the primary PM10 emissions to be used in the CTMs but as emission factors depend on burner type, operation and sampling method, also the secondary organic aerosol (SOA) produced from the emissions of different types of burners under different conditions will differ (Grieshop et al., 2009; Heringa et al., 2011). These studies showed that the volatility distribution of the organic emissions can vary substantially, both between different fuel and burner types and between different operation conditions/practices. To use a single volatility distribution for organic aerosol emissions for all types of residential biomass combustion as is done here, is a simplification.

The EMEP MSC-W chemical transport model used in this work is based on the 3-D CTM of Berge and Jakobsen (1998), extended with photo-oxidant, inorganic and organic aerosol chemistry (Simpson et al., 2012; Bergström et al., 2012). The model domain used in this study covers the whole of Europe, and includes a large part of the North Atlantic and Arctic areas, with a horizontal resolution of 50km×50km (at latitude 60∘ N). The model includes 20 vertical layers, using terrain-following coordinates; the lowest layer is about 90 m thick. Meteorological fields are derived from the ECMWF-IFS model (European Centre for Medium Range Weather Forecasting Integrated Forecasting System, http://www.ecmwf.int/en/research/modelling-and-prediction).

The most recent version of the EMEP MSC-W model includes an organic aerosol scheme that uses the volatility basis set (VBS) approach (Donahue et al., 2009; Robinson et al., 2007) described in Sect. 3.3. An extensive sensitivity analysis of this model has been presented by Bergström et al. (2012). In the present study we used an OA scheme with a nine-bin VBS for the primary OA (POA), including semivolatile and intermediate volatility (IVOC) gases (see Sect. 3.3 and Table 3). The IVOCs are missing in traditional OA and VOC emission inventories and for the standard emission scenario (referred to as EUCAARI) the total emissions of semivolatile POA and IVOCs were assumed to amount to 2.5 times the POA inventory (based on Shrivastava et al., 2008) – that is, an IVOC mass of 1.5 times the POA emissions was added to the total emission input in the model. For the EMEP model simulations that used the revised RWC emissions, with emission factors based on dilution tunnel measurements, a slightly different emission split was applied for the RWC POA. We assumed that the DT methodology captures a larger fraction of the total semivolatile POA and IVOC emissions than traditional inventories (48 % for the new DT emissions, compared to 40 % for the EUCAARI emissions); the same volatility distribution of the OA emission was used in both cases but for the revised RWC emission inventory total emissions are assumed to be 2.1 times the inventory (compared to the factor 2.5 for EUCAARI emissions).

The EMEP inputs used in the present study are based on Bergström et al. (2012) with a few updates. The most important changes are the following:

The background concentration of organic aerosol is set to 0.4 µgm-3. Bergström et al. (2012) used a higher OA background concentration (1 µgm-3) but found that this led to overestimations of OA at many sites during some periods.

PMCAMx (Fountoukis et al., 2011, 2013; Skyllakou et al., 2014) uses the framework of the CAMx air quality model (ENVIRON, 2003) describing the processes of horizontal and vertical advection, horizontal and vertical dispersion, wet and dry deposition, gas-, aqueous- and aerosol-phase chemistry. For the aerosol processes, three detailed aerosol modules are used. The approach of Fahey and Pandis (2001) is used for the simulation of aqueous-phase chemistry. The inorganic aerosol growth is described in Gaydos et al. (2003) and Koo et al. (2003). These aerosol modules use a sectional approach to dynamically track the size evolution of the aerosol mass across 10 size sections ranging from 40 nm to 40 µm. The aerosol species modelled include sulfate, nitrate, ammonium, sodium, chloride, potassium, calcium, magnesium, elemental carbon, primary and secondary organics. The chemical mechanism used in the gas-phase chemistry is based on the SAPRC99 mechanism (ENVIRON, 2003). The version of SAPRC99 used here includes 211 reactions of 56 gases and 18 radicals and has five lumped alkanes, two olefins, two aromatics, isoprene, a lumped monoterpene species, and a lumped sesquiterpene species. In the current version of the model, primary organic aerosol in PMCAMx is assumed to be semivolatile using the VBS scheme (see Sect. 3.3). For the inorganics a bulk equilibrium approach is used in which equilibrium is assumed between the bulk inorganic aerosol and gas phase. At a given time step the amount of each species partitioned between the gas and aerosol phase is determined by applying the multicomponent aerosol thermodynamic equilibrium model ISORROPIA-II (Fountoukis and Nenes, 2007) and is then distributed over the aerosol size sections by using weighting factors for each size section based on their surface area (Pandis et al., 1993).

The PMCAMx modelling domain covers a 5400×5832 km2 region in Europe with 36×36 km grid resolution and 14 vertical layers covering approximately 7 km with a surface layer thickness of 55 m. PMCAMx was set to perform simulations on a rotated polar stereographic map projection. The necessary inputs to the model include horizontal wind components, vertical diffusivity, temperature, pressure, water vapour, clouds and rainfall all created with the meteorological model WRF (Weather Research and Forecasting). The biogenic emissions were produced by MEGAN (Model of Emissions of Gases and Aerosols from Nature) (Guenther et al., 2006). A marine aerosol emission model (O'Dowd et al., 2008) was also used for the estimation of mass fluxes for both accumulation and coarse mode including an organic fine-mode aerosol fraction. The model was successfully evaluated against hourly aerosol mass spectrometry (AMS) data from various stations in Fountoukis et al. (2011) during a photochemically intense period. The model was also used to assess the importance of horizontal grid resolution and the use of high-resolution emissions on the predicted fine PM in a European Megacity (Fountoukis et al., 2013).

Both the EMEP MSC-W and PMCAMx models employ the volatility basis set (VBS) framework (Donahue et al., 2009; Robinson et al., 2007), using methods similar to those of Lane et al. (2008) and Shrivastava et al. (2008). Secondary organic aerosol (SOA) production from VOCs (Lane et al., 2008) is simulated using four semivolatile surrogate SOA products for each VOC, with four volatility bins (effective saturation concentrations C∗ of 1, 10, 100 and 1000 µgm-3 at 298 K). The models treat all organic aerosol species in the gas phase (primary and secondary) as chemically reactive. Each reaction with OH radical is assumed to decrease the volatility of the vapour material by a factor of 10 with a small net increase in mass (7.5 %) to account for added oxygen (Robinson et al., 2007). The primary OA emissions in the models were distributed by volatility (Table 3) using the volatility distributions of Shrivastava et al. (2008). This distribution was derived by fitting gas particle partitioning data for diesel exhaust and wood smoke (Lipsky and Robinson, 2006; Shrivastava et al., 2006).

Primary organic gases, formed due to evaporation of semivolatile POA in the atmosphere, are assumed to react with OH radicals reducing their volatility. Chemical aging through gas-phase oxidation of primary OA vapours is modelled using a rate constant of k=4×10-11 cm3molec-1s-1 (Atkinson and Arey, 2003).

We constructed a revised bottom-up RWC emission inventory for Europe that also accounted for condensable organic material. Since EC is present as solid particles the change between the EUCAARI and TNO-newRWC for EC is rather small in most countries (Fig. 1; for a detailed discussion of the EC emissions see Genberg et al., 2013). However, for OC the change is large as most OC emissions are condensable PM (Fig. 1). The revised RWC inventory changed the total primary OC emissions from wood combustion significantly but not in the same way for all countries. This is illustrated in a more detailed comparison for selected countries in Fig. 3. Note that in this figure the sum of all anthropogenic OC sources are shown but the difference between the two bars for each country is only due to the new assessment of the wood-burning OC emissions.

The spatial distribution, using the new distribution proxy described in Sect. 2.3.1, of the European anthropogenic OC emissions is presented in Fig. 4. Important population centres in countries with high wood consumption can be easily recognized such as Oslo, Paris and Moscow, despite the fact that rural homes were allocated a relatively higher wood use. However, a distinct rural pattern can also be seen with for example high emissions in the eastern part of France. To give a spatial impression of the changes in emissions upon using the TNO-newRWC inventory we calculated the ratio of the revised emissions over the EUCAARI emissions (Fig. 5). The TNO-newRWC emissions are higher than the EUCAARI emissions in all countries except Norway. This is discussed in detail in Sect. 4.3.

The methodology of the EUCAARI inventory for RWC was not a consistent bottom-up inventory but partly involved country-specific data and expert judgements as explained in Sect. 2.1. A direct emission factor comparison with the emission factors used in the revised inventory (Table 2) is not possible. For more details on the different emission factors used in the EUCAARI inventory we refer to the underlying PM emission factors from Kupiainen and Klimont (2007) and more specifically to Table 2 in their paper. Here ranges of the PM emission factors that were used are presented. These ranges originate partly from different methods that countries use and cannot be directly compared with the SP and DT emission factors used in this study as no detailed description of measurement methodology is given. Another feature of Fig. 5 is that the relative increase is a fixed number for each country. This is due to the fact that we have an estimate for the appliance types (Table 2) and their associated wood use for each country but no information how these appliance types are distributed within the country. Hence, within a country the fraction of each appliance type per grid cell is constant. This may not be realistic as, for example, more affluent regions may have more modern wood stoves. However, we are not aware of reliable and documented data on this subject.

The two CTMs were run with the two different RWC emission inventories. Figure 6 shows the modelled OA concentrations over Europe with the EUCAARI emission inventory, and the relative change in OA when using the revised RWC inventory instead. For the EUCAARI late winter/early spring AMS campaign (25 February–23 March 2009) PMCAMx predicts mean OA concentrations of 1–2 µgm-3 in most of Europe and above 2 µgm-3 in western France, Portugal, a fairly large region in western Russia and parts of the Baltic countries, and around some cities (e.g. Oslo, Lyon and Barcelona). The peak concentration, above 6 µgm-3, is predicted in the Paris area. The EMEP MSC-W model predicts higher OA concentrations than PMCAMx in most of the modelling domain, above 2 µgm-3 in most of southern and eastern Europe, and a remarkably high concentration in the Oslo region in southern Norway (above 4 µgm-3; a considerably higher level than anywhere in the neighbouring Nordic countries). However, in Paris and western France, the British Isles, Moscow, and the northernmost part of the domain, EMEP gives lower concentrations than PMCAMx.

Both models show a significant increase in OA in large parts of Europe with the revised RWC inventory (and a substantial decrease in southern Norway). The response to changing the RWC emissions is larger in the EMEP model than in PMCAMx. For the February–March period, the EMEP model predicts more than 50 % higher total OA concentrations with the revised RWC inventory than the old one in large parts of central and southern Europe, and Latvia, Lithuania and southern Finland. The impacts of changing the RWC emissions are much more “local” (and, especially for southern Europe, smaller) for PMCAMx than for EMEP (at least for the areas with increasing emissions); this indicates that the wood burning OA is removed more rapidly in PMCAMx than in the EMEP MSC-W model.

One of the models (EMEP MSC-W) was run for an extended period to evaluate the revised emission inventory against long-term measurements of OC at selected sites in Europe during the 3-year period 2007–2009. The revised RWC inventory has much larger emissions of OC in Finland, Sweden, Germany and the Netherlands compared to the EUCAARI inventory (Fig. 3). For this reason we focus on available observations in these countries, where relatively large changes in modelled total OC are expected. RWC has a strong seasonal variation with the largest emissions during the cold seasons. In the EMEP MSC-W model about 80 % of the RWC emissions occur during the winter half-year, November–April. During the warmer months other sources of OC are more important (especially secondary organic aerosol from oxidized biogenic VOC). It is difficult to model the boundary layer height accurately in large-scale CTMs and there is a risk that the models will underestimate PM from residential wood combustion (and other local, low stack-height sources) during calm, cold nights with strong temperature inversions. However, using the EMEP MSC-W model we find essentially no correlation between either mixing height or temperature and model bias for OC (see Supplementary information).

The modelled OC is in better agreement with observations at most sites when using the revised RWC emissions compared to the EUCAARI emissions (Fig. 7, Table 4 and Supplement Fig. S1). The model–measurement correlation is higher with the revised inventory, at all of the five investigated sites in the countries with large increases in RWC emissions (the average correlation coefficient, r, for the winter half-year data, is 0.69 with the EUCAARI inventory and 0.75 with the revised RWC inventory); the mean absolute error (MAE) of the model OC is lower with the revised inventory, except in Hyytiälä, where it is slightly higher (the average relative MAE for the five sites, for November–April data, decreases from 52 to 42 % when updating the RWC emission inventory).

The Norwegian site Birkenes was also included in the comparison (Fig. S1); Birkenes is situated in the southernmost part of Norway and it is to a large extent influenced by long-range transport from other parts of Europe. The long-term average (cold-season) model results and statistics for this site are very similar with the old and revised RWC emission inventories; the reduced Norwegian emissions are balanced by increased emissions in the neighbouring countries.

The revised emission inventory leads to higher correlation between modelled and measured total OC concentrations and lower model MAE also when comparing to full-year data (Supplement Table S3). Since the measurements of OC are not source-specific, improved model predictions can be caused by the increased RWC-emissions compensating for other missing OC emissions in the model. However, the fact that both winter and full-year results are improved, in combination with the strong seasonal variation of the RWC emissions, is an indication that the revised emissions are indeed more realistic than the old ones.

Figure 8 shows a comparison of predicted (PMCAMx) vs. Observed (AMS) PM1 OA concentrations (Crippa et al., 2014) from seven measurement sites in Europe during the EUCAARI winter 2009 campaign (25 February–23 March). Even for this short-term period the revised emission inventory improves the model performance against the AMS data. The slope of the linear fit increases from 0.51 to 0.82 with the use of the revised inventory for RWC emissions. The average fractional bias is substantially reduced (from -0.3 to 0.1) with the use of the revised RWC inventory. The model predicts 35 to 85 % higher OA concentrations as a result of the update in the RWC emission inventory.

In order to test the revised RWC emission inventory in more detail we performed a case study using data from published source apportionment studies for the two neighbouring countries Norway and Sweden. Genberg et al. (2011) measured levoglucosan (LG) levels at Vavihill, in southern Sweden, during the period April 2008–April 2009. We compare the model-calculated bbOC (OC from wood burning, including both residential combustion and wildfires) for both inventories to the observed LG concentrations in Fig. 9. The amount of LG that is emitted during wood combustion varies (see e.g. Genberg et al., 2011, and references therein); here we assume that bbOC lies in the range 5.5 to 14 times the LG concentration (as in Szidat et al., 2009). Observations at Vavihill show a clear seasonal variation with high concentrations of LG during the winter and, mostly, low concentrations during summer. Modelled bbOC with the EUCAARI emissions underestimates the winter observations severely. Using the TNO-newRWC emission inventory improves the model results a lot; there is still a tendency to underestimate bbOC but not as badly as with the older inventory. Modelled total OC is also improved compared to measurements (Table 4 and Fig. S1).

We also compared predicted OC from wood burning to source apportionment data from winter campaigns in and near Gothenburg in Sweden (Szidat et al., 2009) and in the Oslo region in Norway (Yttri et al., 2011). A more extensive comparison to these campaigns (including other sources and both summer and winter periods) was done by Bergström et al. (2012); in the present study we focused on the wood-burning part and the impact of the choice of RWC emission inventory. The results are shown in Fig. 10. For the Norwegian sites (Oslo and Hurdal, 70 km NE of Oslo) modelling with the EUCAARI emission inventory led to large overestimations of wood-burning OC; results were clearly improved for both sites when the TNO-newRWC inventory was used. Model results for the Swedish campaign were quite different, with underestimated wood-burning OC with the EUCAARI inventory while the results with the revised RWC inventory were within the 10–90 percentile of the source apportionment estimate, based on the measurements, for the rural background site Råö, but still somewhat underestimated for the urban background site (Gothenburg). The revised RWC inventory also improves model results for wood-burning EC at these sites as shown by Genberg et al. (2013).

Although the two source apportionment campaigns were relatively short, and limited to two regions, the model improvement using the revised RWC emission inventory is consistent with the findings for modelling total OC over longer time periods. These results further support the need to update and harmonize the official estimates of wood-burning emissions in Europe.