The use of wood for residential heating for the years from 2005 to 2016 is provided based on the responses to the Statistics Norway's Travel and Holiday Survey. Data on wood consumption are collected and officially reported by Statistics Norway at the county level. The survey gathers data in four quarterly surveys covering the preceding 12 months. The calculated consumption of wood is the average of five consecutive quarterly surveys. The survey contains 25 questions regarding wood burning for residential heating. The sampling pool of the survey is drawn at a nationwide level and is considered representative for all counties. Wood consumption in each of the three technology classes (open fireplace, stove produced before 1998 and stove produced after 1998) is available for each county, and these data are used in the MetVed model as input because finer resolution is not available.

Reported wood mass (kg) from questionnaires is recalculated to represent dry wood consumption by assuming an 18 % water content. In that way, wood consumption corresponds with emission factors that represent grams of pollutant per kilogram of dried wood. According to Statistics Norway, there are several elements of uncertainty in the wood consumption data, such as the survey sample size and the employed conversion factors (e.g. mass of bags of wood reported as volume). Statistics Norway concluded that the coefficient of variation of the total wood consumption is below 3 % based on an uncertainty study carried out in 2011 . The uncertainty is higher at the county level as the values are based on smaller samples. Large-scale production of fuelwood and sales of wood (estimated 70 % of consumption) are nearly exclusively birch, and also untraded wood will contain significant birch. As no other information is available, consumption is assumed to be birch, in accordance with emission factors.

The Norwegian emission factors are listed in Table  and were established by for three categories: (i) open fireplaces, (ii) stoves produced before 1998 and (iii) stoves produced after 1998. The emissions factors are determined by laboratory experiments following the Norwegian standard for testing enclosed wood heaters and smoke emissions and are considered as representative of real-world conditions. The particle sampling is carried out in a dilution tunnel in order to mimic the dilution and cooling effects when the smoke exits the chimney. In this way, the particle sampling also accounts for condensed matter. The Norwegian standard requires that the type of wood used is birch, so all emission factors are based on birch wood.

proposed two sets of emission factors depending on different firing conditions in Norwegian cities and rural areas. The reasoning behind the two emission factors is that users in the countryside fire their stove with low air intake during the night and other users do not . To our knowledge, there is not a solid and updated study that supports this assumption; thus we presume the same firing conditions as in large cities across all of Norway.

The data set containing the type and number of dwellings is obtained from Statistics Norway . The data set originates from the state tax agency registry (SERG), and that covers all households in Norway. It can be considered complete and up to date. The gridded version at 250 m of this registry maps the number of dwellings, the number of detached houses, duplexes, townhouses, the number of dwellings in apartment blocks and other dwellings. An example of the dwelling number data set (for the municipality of Stavanger) is shown in Fig. . The grid number of total houses (Fig. a) has the highest density of dwellings in downtown Stavanger. This is also where apartments (Fig. c) have their highest density, while detached houses (Fig. b) are much more uniformly spread through out the municipality. This illustrates a typical feature of most urban areas, where a city centre typically consists of mainly apartment blocks, whereas detached houses and duplexes are more prevalent in suburban areas. In Norway in 2017, there were around 2.5 million registered dwellings of which 50 % are detached houses, 9 % duplexes, 12 % townhouses, 24 % apartments and 5 % are classified as others. The spatial distribution of dwelling types differs across region and area. This is especially so for the apartment share, which ranges from 77 % in the Oslo municipality to 0 % in many rural municipalities.

In Norway, there are 620 fire stations divided among ∼300 fire department agencies. The fire and rescue agencies are responsible for inspecting and assessing all firing installations. These are dominated by residential heating installations, but the database also contains a small number of cooking appliances. The agencies carry out routine inspections as part of fire hazard safety procedures. During these inspections, information such as geographical location and type of installation is collected in a local database by each agency.

In this study, we contacted 270 fire and rescue agencies in Norway to gain access to their data set on firing installations. We successfully obtained the complete information from a total of 101 municipalities, covering almost half (1 million) of Norwegian residencies, including the five largest cities. The data set provides detailed information on all residential addresses inspected, including the geographical location of the pipe, the installation type, technology and model, and whether the technology can be classified as a clean burning technology (i.e. true/false).

The fire and rescue agency registry, if complete, should include all firing installations including those in residential households. There are however a few caveats to these data. Inspections are carried out continuously and over time, and the data obtained generally did not indicate when the inspection was done and some data could be obsolete. There is also a lack of uniform sampling method among the municipalities, and the type of data supplied was different for each fire departments system. For instance, there were 795 different housing types in the total fire agency data, which were filtered to each of the residential types of buildings and others. Classification of the installation ranged from a general description to a complete brand and model type, in all consisting of 890 different descriptions. Each one was placed in one of the three categories of wood burning installation or classified as non-wood based. It also appears that the coverage is not fully complete and must be considered partial even in the municipality where all fire agency data were provided. For example, the fire agency in the Oslo municipality informed us of continual dodging of inspections by house owners for various reasons, which results in an estimated 10 % unregistered firing installations (Oslo Fire Department, personal communication, 2018). As a result, the data are used as a statistical input to the MetVed model. Cross referencing wood consumption and firing installation statistics shows that there is a relatively good agreement with other existing data, and therefore it could also be used directly in the model. Figure d shows all the inspected installations in the Stavanger municipality with a density heat map Fig. e. The highest density of wood burning installations is obtained in the Stavanger city centre, but there are also other less densely populated areas where fireplace frequency per dwelling is much higher (e.g. Testa, which is north-west of the city centre).

The web-crawled data set is derived from the web-crawling program GoodOvening that scrapes data in a systematic way from a real state advertisement portal satisfying certain search criteria. It structures data obtained for further data analysis for more details see. The web-crawled data set is continually updated and consisted of 444 000 geo-positioned data points within Norway at the time the data was extracted for the MetVed model. Along with the geographical location, the web-crawling extracts data on the characteristics of each dwelling such as the type that they belong to (i.e. detached house, townhouse, duplex, apartment, etc.), the size and the type of the available energy system for residential heating (i.e. wood burning installation, district heating, heat pump, etc.). The web-crawling database is aggregated to the 250 m grid resolution and it allows us to establish, with a high level of detail, grids with low or high RWC potential. A zero RWC potential is assigned to those grids with dwellings with no wood-based heating installations. The highest potential would be to those grids with dwellings where a wood-based installation is the sole residential heating source. Between these, the potential is determined as a proportion of wood-based installation to other residential heating technologies.

Figure f shows the proportion of wood-based installations for residential heating relative to other listed technologies (e.g. district heating, heat pump, electricity) at 250 m resolution. The city centre shows a low proportion of wood-based technologies (i.e. from 0 % to 30 %) even though the same area shows the highest number of residential dwellings (Fig. a) and the highest density of wood-based technologies (Fig. b). The highest proportion of wood-based technologies is observed spread out in the areas outside the city centre. We see similar spatial patterns at the grid level for the proportion of wood-based heating technologies in and around other Norwegian cities.

The ENOVA energy labelling system was implemented in Norway in July 2010 and is a self-assessment report performed by owners of dwellings and buildings, or qualified experts in the case of new buildings. The ENOVA data set consists of about ∼650000 entries that include (i) the size of the dwelling, (ii) building type (e.g. apartment, small house, office building), (iii) building year, (iv) energy consumption of the dwelling (kWh), (v) primary and secondary heating installations, (vi) energy consumption per fuel and (vii) geographical information. While individual buildings are listed in the data, the data accessed were used on a municipal and postal code resolution due to data privacy. The information from these data supplements the two preceding data sets and also provides a relation between energy consumption for total heating, RWC and building size/type. Wood consumption input statistics to MetVed were used on the county level to correspond with wood consumption data.

Outdoor temperature is an important controlling factor for RWC because the main purpose of most RWC is the heating supply. The MetVed model uses observed temperatures as input meteorological data. Observed daily mean temperatures were obtained for the years from 2005 to 2016 for 57 official Norwegian temperature measurement stations through the eKlima database to form the basis for time variation of emissions for MetVed. For now we have ensured that urban areas have an observation point within them.

In addition to geospatial information of Norway, such as municipal and county administrative borders, we have ancillary data responses to questionnaires on wood consumption in Akershus/Oslo and Sarpsborg/Fredrikstad . Together with all of the individual answers of respondents to the questionnaires of the Statistics Norway Holiday Survey (where wood consumption is reported), they form the basis of establishing assumptions on variability between primary and supplementary wood heating habits.

To assess emissions, dispersion modelling has been used and the results have been compared with observations. To do this, measurement data of PM2.5 were taken from the Norwegian air quality network monitoring stations that report to the European air quality database (https://www.eea.europa.eu, last access: 14 June 2019). PM is measured by applying methods equivalent to the reference method (i.e. TEOM 1400A and Grimm-EDM180) that continuously monitor and log with a time resolution of 1 h. Measurements of BC in Oslo were obtained at two roadside monitoring stations in the winters 2014/2015 and 2015/2016. The measurements were performed with an Aethalometer (Magee Scientific) that measured aerosol light absorption at seven specific wavelengths at 1 min resolution . The results at the specific wavelengths are used to determine the contribution from traffic and wood burning to BC concentration based on the model established by . Benzo(a)pyrene (BaP) measurements were taken from the Norwegian monitoring network through active air sampling. The monthly B(a)P values are derived from the analysis of particle filters, which are collected with a 3 d frequency. The identification and quantification of B(a)P is carried out by GC/LRMS.

From an annual baseline consumption, MetVed estimates the hourly distribution of wood usage through a calendar year. The time variation of RWC activity is defined to be dependent on the heating demand defined by the outdoor temperature. Therefore the heating degree day (HDD) concept is used . MetVed looks up the geographically nearest meteorological station to obtain the outdoor temperature data set. Thereafter the HDD is calculated as 7HDD=max⁡0,TThreshold-T, where T is the outdoor daily average temperature in degree Celsius. Regarding TThreshold, we follow and use 15 ∘C (HDD15). Hourly resolution is obtained by coupling the daily consumption to the diurnal heating cycle in Norway. The time of day when RWC occurs is obtained by a survey covering dwellings that provides information regarding the use of the RWC installations both during the day and week . The obtained weekly and hourly activity shows a strong diurnal cycle, where RWC is higher after 17:00 CET. Hourly consumption is obtained by multiplying the daily HDD derived consumption with the hourly average reported activity, which is also weekday dependent. There is higher activity during weekends than weekdays . The consumption differences between weekdays and weekends are smaller than what is suggested by and . However, the diurnal pattern of concentrations they found for RWC is similar to the emissions profile derived based on the activity data, which assumes an emission profile that takes into account emissions of the later stages of the firing cycle e.g.. Fixed public holidays and Easter are treated as weekends.

The primary model output from MetVed is gridded emissions for Norway in the period of 2005–2016. Figure  shows the spatial distributions of emissions from RWC obtained with MetVed at a 250 m grid resolution in southern Norway and seven domains selected for the evaluation of urban emissions. The distribution of RWC emissions on the 250 m grid is concentrated where there are residential buildings. This is in cities, valleys and along the coast, and consequently they cover a small proportion of the surface area in Norway. Each grid cell has emissions relative to their proportion of the wood burning determined by the number and type of dwellings and available residential heating technology within it. In the urban domains, the lowest emissions (i.e. in the range 0–0.14 tyr-1 per grid) are obtained in the outskirts of the urban areas (blue grids in Fig. ), whereas the highest emissions (i.e. 0.30–1.50 tyr-1 per grid) are centred in the urban areas. The internal distribution within each urban domain varies among the cities, as it depends on the distribution characteristics of apartments and houses, as well as the availability of non-wood-based residential heating technologies.

Total annual MetVed emissions are closely linked to the total emissions in Norway reported to CLRTAP (Fig. a) as both are estimated using the same wood consumption data and emission factors. MetVed wood consumption is only for residential heating, and the difference with emissions reported to CLRTAP is that the latter includes consumption in cabins. The peak year of wood consumption is 2010, coinciding with an especially cold winter across Norway. Since 2010, consumption has gone down every year except in 2012. Before 2005, there was a general decline in reported emissions. This is mainly associated with a reduction in wood consumption and an increase in the share of newer technology ovens. From 2005 until 2016, consumption in open fireplaces is relatively constant, varying between 3 % and 5 %, whereas the share of consumption in new stoves has nearly doubled, from 34 % in 2005 to 62 % in 2016 Fig. b).

In 1998, regulations were put into place to reduce emission for newly sold stoves, which should not emit more than 12 gkg-1 of PM2.5. The bars in Fig. c show the assumed average EF of stoves sold in that year. Assuming ovens have equal usage, the share in a given year predicted by an exchange of stoves by sales is shown as a dashed yellow line. This EF fits very well with the derived emission factor for PM2.5 from both MetVed and from the CLRTAP emissions since 2005. Similarly as for PM2.5, all other compounds (Fig. d) except CH4 show a general decline in EFs since 2005. This decline in EFs is the main driver for the decreasing trend in emissions in Norway, as an increasing fraction of wood is consumed in new stoves. BC is the only EF that does not decrease uniformly. The reason is that the EF for open fireplaces is an order of magnitude higher than those for stoves, and the slight variability in the consumption estimates in open fireplaces drives the change.

The MetVed input data were analysed to evaluate both the share of stoves and their use. In the fire and rescue agency data, roughly 70 % indicate the age of the installation, of which only 34 % of residential stoves are noted to be newer than 1998. A recent survey carried out in Norway in 2016 shows around 64 % of the wood burning installations in 2016 are new stoves (Norstat 2016 survey, Norsk Varme, personal communication, 2018). Sales estimates based on this survey indicate that the transfer to cleaner technologies therefore will go on for the next several decades. Similarly, the increased efficiency of newer ovens from an estimated 50 % to 75 % for new ovens will act to reduce consumption. Thus, based on this, future emissions are expected to go down further. The red bars in Fig. b show an assumed share of existing stove technology assuming annual sales of new ovens to the residential sector of 67 500 units per year. This sale is derived from a fit to emission factors and will be influenced by consumption differences. Based on the Norstat 2016 survey, annual sales are close to 40 000 installations per year (Norsk Varme, personal communication, 2018). This difference indicates that there are large consumption differences and that on average a new installation may involve a higher consumption of fuelwood. Therefore, what the real effect of exchanging an old for a new installation will have on the overall consumption, and therefore on emissions, is still uncertain.

It is worth noting that the two stove technologies new and old are comprised of an assembly of stoves. Producers of stoves today claim to have significantly reduced PM2.5 emissions even further since 1998, to about 2 gkg-1 in 2016 (Norsk Varme, personal communication, 2018). Both Norwegian emissions reported to CLRTAP and MetVed assume constant EF for the new oven assembly. The dotted lines in Fig. c show the EF with a continuing reduction down to 2 gkg-1 in 2016, both for ovens sold in that year (light blue) and for the share of stoves in that year (green). Were EF based on this it would act to significantly reduce 2016 PM2.5 EF from 13.5 to 7.4 gkg-1 and thus nearly halving emissions.

Consumption of fuelwood follows the change in demand for heating energy. In Fig. a the total annual number for HDD15 at observation sites (57 meteorological stations) across the simulation domains in Norway is weighted by the total number of dwellings in the domain covered by each station. The MetVed annual consumption is from the CLRTAP (Fig. a), which is derived independently from the temperature. The linear fit to the period 2005 to 2016 indicates that consumption is reduced by ∼6 % per year, whereas a slight increase in HDD demand is observed over the same period (Fig. c). Residuals from both trends show that the temperature can explain about 63 % of the variance in consumption (r=0.79 Fig. b). HDD would therefore provide a good indicator for present and future annual consumption variations, but long-term trends depending on the physical properties of residences must also be considered. It is also important to note that the trend of decreasing consumption between 2005 and 2016 is not due to increasing average winter temperatures.

In the MetVed model, the concept of HDD is only used to distribute wood consumption within a year, and only to days with a heating demand. The model determines the total number of HDDs in a year over which the consumption is distributed. The choice of a threshold HDD, i.e. the coldest temperature where no heating is required, only influences the temperature sensitivity of consumption within a year. A lower HDD threshold would therefore only lead to higher emissions in winter and less during spring and autumn. In the cold year 2010, consumption increased by 25 %, and the number of HDDs with threshold 20∘ (HDD20) increased by 23 % relative to the 2005–2016 mean. In contrast, HDD15, HDD10 and HDD5 showed an increase of 36 %, 65 %, 135 %, respectively. Thus a lower threshold temperature will thus increase temperature sensitivity of emissions. From the residual consumption and HDD15 in Fig. b we derived the relation for the period 2005–2016: 8ΔC=0.77ΔHDD15, where ΔC is the change in consumption and ΔHDD15 the change in the HDD15, both unitless. The relation between the outdoor average temperature and consumption is 0.32 kg dwelling-1 HDD15-1.

As other factors also influence wood consumption, household energy consumption per square metre has been examined from the output of the Norwegian energy balance model (https://www.ssb.no/energi-og-industri/statistikker/husenergi/hvert-3-aar/2014-07-14, last access: 18 December 2018) and shown in Fig. c. An increase in the number of energy-efficient houses and a lower wood energy share reduce the annual wood energy demand (both by ∼1 % per year in Fig. c). report an energy efficiency of 15 %, 50 % and 75 % for open fireplaces, old stoves and new stoves, respectively, which are also used in the Norwegian energy balance model. Consumption per technology in 2012 gives a 61 % average stove efficiency, at a constant energy output. The effect of newer, more-efficient installations has influenced the decrease in consumption by 0.5 % per year (“oven efficiency” in Fig. c).

In 2012, the average reported dwellings in Norway received 16 % of their energy, about 3200 kWh, from wood. This is significantly higher than the average energy share suggested by the MetVed input data from ENOVA, where the annual average energy from wood is 960 kWh, although only 3 % reported using wood for heating. The reason for the lower reported wood consumption in the ENOVA data set is not clear but may be related to conscious under-reporting to achieve a better energy certificate by the dwelling owners. Birch wood, which is the most common fuelwood in Norway, has a dry (0 % moisture) energy content of 4.395 kWhkg-1 . To achieve the stated energy, an average Norwegian household would then in 2012 have to burn 1195 kg of dry birch wood. The same conversion of the ENOVA-reported consumption gives 7.2 kg of dry wood for an average Norwegian household. Based on the official wood consumption data, the total residential consumption in 2012 was 1460 kt, which equates to 589 kg of wood consumption per dwelling (800 kg per dwelling with a wood-based installation). Due to large differences in the input data, the housing type and size and energy dependencies calculated within MetVed are only based on the ENOVA-reported total energy consumption.

The main aim of MetVed was to improve the spatial and temporal distribution of emissions from RWC. Resulting emissions of PM2.5 within each of the urban domains in Fig.  from MetVed are shown with four other emission inventories available to do urban modelling in Norway in Fig. a. These domains contain the largest cities in Norway and together cover 44 % of Norwegian dwellings. With the exception of Norsk BeregningsVerktøy (NBV) emissions, all urban emission inventories are obtained by downscaling national emissions submitted to the CLRTAP (http://www.ceip.at). Though the year of reference varies between inventories, the magnitude of total national emission is comparable. The difference within these domains is therefore to a large extent determined by the method of spatially distributing the emissions.

Total emissions within each domain are computed by first-order conservative remapping emissions for each inventory to the domain e.g.. The European Monitoring and Evaluation Programme (EMEP) emissions are distributed on a 0.1∘×0.1∘ grid and represent the lowest urban emissions in all domains, except Oslo, having an order of magnitude lower emissions than the remaining domains. To our knowledge, detailed information about the method used to downscale the EMEP emissions is not publicly available, but the data are openly available at https://ceip.at. states that when the activity data used to estimate emissions are available at a higher geographical resolution than national, they are used to distribute emissions. In Norway, wood consumption is available at the county level; thus we assume that EMEP emissions from RWC are distributed at this level. This is consistent with the visualisation of emissions where the county administrative borders are visible, and emissions are widely distributed in the Norwegian geography, covering also unpopulated areas. The TNO_MACC_III emissions have a resolution of 0.125∘×0.625∘, similar to EMEP, but have much higher emissions within each urban domain than EMEP. TNO_MACC_III uses internal approaches based on population density and a function to describe proximity to wood as their downscaling method .

NWA emissions (NordicWelfAir, http://projects.au.dk/nordicwelfair/, last access: 16 June 2019) are on a 1 km grid and are based on scaling down wood consumption per technology from county level based on dwelling number at 250 m resolution. Different RWC activity is assumed for apartments and houses, and the Norwegian official emission factors (Table ) is used. Thereafter emissions are re-gridded to a 1 km resolution. NBV emissions, also on a 1 km grid, result from downscaling wood consumption per technology by dwelling number at the district level resolution also using the Norwegian official emission factors . The domain of Oslo in NBV is an exception. In this case emissions are reported in . In NBV, a multiplication scaling factor was derived for each urban domain individually based on the ratio of concentration levels obtained by atmospheric dispersion modelling to observed PM2.5 levels. These factors vary from 0.42 to 0.27, i.e, dividing the total emissions in these areas by a factor of 2.3–3.7. For the MetVed emission inventory, the total emissions within the area are dependent on all the input parameters detailed in Sect. .

The total emissions within each domain is for all inventories (with the exception of EMEP) closely related to the number of houses within each domain (Fig. b), either directly or indirectly through population. Note that the sizes of the domains vary somewhat and all have different total area (Fig. ), but they all cover a city and their area of influence. The dependency on dwelling or population density differs among the methods, and thus emissions at the urban scale vary accordingly. TNO_MACC_III results in the highest emissions at the urban scale followed by NWA emissions, but EMEP results in very low emissions at the same scale, as the downscaling approach distributes and diffuses emissions at the county level. In the MetVed model, the emissions are determined by the physical properties of the houses and their installed heating technologies within each domain (also done on the 250 m grid). In our study, we consider NBV emissions as a tuned emission data set, as they have been obtained by comparing results from dispersion modelling and observations, and then scaling emissions. The correction factors vary among the domains and therefore the methodology lacks consistency. However, MetVed emissions are produced at high resolution at the national level following a consistent methodology and relationships between variables that influence emissions. In addition, MetVed emissions at national levels equal the official emissions reported to the CLRTAP which gives it consistency as a national emission inventory. The results from the MetVed model at the urban scale are consistent with NBV for most of the domains. MetVed emissions within the urban domains have, for most of the domains, lower emissions than those derived as a result of downscaling approaches based on population (TNO_MACC_III) or dwelling density (NWA). The MetVed emissions are the most similar to the locally corrected NBV emissions. This lowering of urban emissions in MetVed relative to emissions from the downscaling approach is the result of taking into account the share of wood-based technologies from the web-crawled input data. In that way, we account for lower wood consumption when heating technologies other than wood-based technologies are available .

There are a number of dependencies in the consumption and wood installation statistics for these domains (Fig. b). For instance Trondheim, which has the fourth highest population, is located in the county with the second highest consumption per dwelling, and it has the second highest consumption per wood installation (after Grenland) of any domain. For NBV emissions, this leads to an observation-based emission correction of 0.33. In MetVed, Trondheim's high share of apartments (Fig. c) and low frequency of fireplaces (Fig. b) gives the lowest emissions for all of the MetVed domains, which shows that these properties act well to explain the reduced consumption. Dwelling size is an important factor for calculating consumption in MetVed, and it plays a significant role because apartments in the Trondheim city centre are generally small. The Oslo domain similarly has low emissions per dwelling, but in this case it is also driven by a low consumption per wood burning installation.

An increased prevalence of wood installations is positively correlated (R2=0.44) with higher consumption within the domains (Fig. c). This is a result of the MetVed calculations, where the combination of regional consumption statistics and the prevalence of wood installations is found by the web-crawled statistics. This acts to support assumptions made in previous studies, e.g. proximity to wood and rural consumption being higher than urban consumption, but is in MetVed represented in dwellings' heating installations and their usage rate. Additionally, MetVed considers the dwelling size which further increases a rural amplification of firewood use, as rural properties are on average larger.

The dispersion modelling in the seven domains of PM2.5 was compared to available measurements of total PM2.5 concentration. Comparisons must therefore include all anthropogenic emissions in the urban areas. The additional emission inventories include, along with RWC, emissions by shipping, off-road machinery, traffic exhaust and suspension of road dust. Urban emissions are estimated based on high-resolution input data that thereafter are aggregated to a 1 km grid and combined with time variation functions to result in emissions at 1 hkm2 resolution for more details see. Non-exhaust emissions associated with tyre and road wearing processes and the suspension of road dust were modelled by the vehicle abrasion and suspension model NORTRIP . Emission domains and the year of reference (2015) were chosen to be able to compare with modelled concentrations obtained based on NBV emissions . Similarly, measurement stations were selected to coincide with the existing model concentrations and assess a potential improvement when comparing with previous estimates.

With few exceptions, the air quality stations are classified as traffic stations and are operated by the road authorities. Few measurements are therefore ideally located with regards to a detailed evaluation of the spatial distribution of emissions from RWC within each domain. The main difference of the spatial distribution provided by MetVed and that by other methods not differentiating housing types (and size) and available residential heating technologies can best be seen through comparison between stations situated between different types of buildings. In Oslo for instance, only one monitoring station (Smestad) could be qualified as located in an area dominated by detached houses, together with similarly located monitoring stations Vaaland in Stavanger and Lensmannsdalen in Grenland. The remaining stations are in close proximity to apartments or a diversity of dwelling types, along with roads with large emissions. Together with the contributed uncertainty from other sources the differentiation on housing types is therefore hard to evaluate, and so the main focus is on evaluating total emissions within each domain.

Hourly correlation coefficients and biases are shown in Table . Compared to NBV emissions, which were calibrated with a weekly correction factor, the MetVed temporal emission profile generally improves correlation. The bias is not improved, because the MetVed emissions are generally higher, and so the overestimation by the model is increased. The Oslo comparison shows a strong overestimation, and a possible reason is that the MetVed approach has a limited effect in Oslo, as it is in itself a county, and so total consumption and emissions are direct results of the input wood consumption. In the Trondheim domain, the NBV emissions results from applying a factor of 0.33 to the dwelling density derived emissions , and MetVed produces similar results to NBV at the Elgeseter monitoring station. The consideration of dwelling type and size and the fireplace information accounted for in MetVed account similarly for lower emissions and concentrations in urban areas as the local scalings applied individually in NBV.

The annually averaged observed concentration at the sites is 8.0 µgm-3, and using MetVed emissions this is 8.7 µgm-3. Monthly observed concentrations at each station are shown in Fig.  as a black line, and the modelled concentrations are shown as bars, coloured by sector. Both model and observations show a pronounced seasonal cycle in PM2.5 concentrations with a winter peak primarily driven by the background concentration and RWC contributions, and the remaining sectors have a similar cycle but contribute less to the PM2.5 seasonality. Model concentrations, including RWC, have a winter to summer (i.e. DJF to JJA) ratio of ∼2, and the measurements have the winter (DJF) to summer (JJA) ratio of 1.45, indicating that the seasonality in model concentration is more than twice as strong. As wood burning for residential heating is an intensive winter activity in Norway, it is hard to envision that there should be no seasonality in RWC, and therefore the uncertainty on the seasonality of other contributing sectors must be considered. reported, based on simultaneous measurements in Oslo and a regional background site (Hurdal, 70 km NE of Oslo), similar PM1 values in summer at both sites, 7.6 and 7.7 µgm-3, respectively. However, in winter, PM1 concentrations in Oslo were similar to summer concentrations, i.e. 7.8 µgm-3, but at the regional background site they were 45 % lower than in summer, i.e. 4.3 µgm-3. Source apportionment also indicated that the elevated urban wintertime concentration consisted largely of organic mass from biomass burning. These results indicate a strong urban source rather than influence from regional background levels during winter.

The diurnal evening PM2.5 peak in winter (Fig. , bottom right) is also more pronounced in the model than that in the observations. While the difference between the summer and the winter diurnal pattern is also enhanced by meteorological conditions, the absolute difference between winter and summer (i.e. DFJ–JJA grey shape in Fig. , bottom right) is much less prominent than the diurnal winter contribution of RWC alone (orange line in Fig. , bottom right). The assessment of the diurnal variation also suggests that the total modelled influence of RWC on the air quality station PM2.5 is too strong.

In Oslo, three model simulations were used to assess the sensitivity of surface concentrations to emission altitude. In urban Oslo, the annual average RWC concentration of PM2.5 at 2 m was 4.41 µgm-3 when all RWC emissions were emitted in the surface layer (0–30 m). When apartment emissions were emitted in the second layer, the surface concentration was reduced to 3.76 µg-3, and when smaller buildings emit in the second layer, a further reduction to 3.19 µg-3 was observed. The wintertime diurnal cycle of PM2.5 has a more pronounced peak, and the seasonal cycle of PM2.5 similarly indicates a too-strong seasonal cycle obtained with all emission estimates, when compared to observations, and so the total contribution of RWC to PM2.5 concentrations seems to be overall overestimated. This is supported by the new emission inventory developed by for particulate emissions from RWC. When assessing organic carbon emissions, establish that while emissions in most of Europe are underestimated, in Norway they are overestimated.

During both winter 2014–2015 and 2015–2016, hourly concentrations of aerosol absorption were measured at two sites in Oslo (Smestad and RV4 Fig. ). The first period was at Smestad (17 December 2014–18 March 2015) and the second at RV4 (20 November 2015–30 May 2016). The decomposition of the absorbing aerosols includes BC from biomass burning (BCBB), which in Oslo (in winter) should be equivalent to BC from RWC. While both measurements were done in close proximity to larger trunk roads, Smestad is surrounded by a large area of detached houses, while the RV4 sampling site is characterised by apartment buildings and a hospital. Even though the number of dwellings in the area surrounding RV4 is higher with nearly 30 % more residences within 300 m, MetVed emissions in the grid of RV4 are 10 % lower, as the area is characterised by a high share of apartment buildings with low MetVed emissions.

The modelled year (2015) does not completely overlap with either of the two measurements (Fig. a), and therefore model data are shown for the temperature range (T<8 ∘C), which is the temperature range covered by the observations in Fig. b. Daily average BCBB concentrations at both sites show a similar dependence on temperature Fig. c, though overall lower concentration levels are observed in RV4, which is probably influenced by the measurements continuing further into spring. The difference between the two sites is somewhat smaller in the model. Model concentrations show a similar but weaker temperature dependence, suggesting that daily emissions could increase more than emissions obtained by HDD15.

The modelled diurnal variability in concentration (Fig. c) agrees well with BCBB. The diurnal time variation of wood consumption and the subsequent MetVed emissions is in agreement with the BCBB observations. The diurnal weekend emission profile is similar to the observed hourly concentration profile, and the total average model concentration fits (T<8 ∘C) well with the levels of measured BCBB.

For most of the air quality stations, dispersion model results show an overestimation of winter PM2.5 concentrations, and a similar underestimation of summer concentrations. While PM2.5 has many sources, B(a)P filter measurements offer a way to more directly investigate the RWC contribution.

Figure  shows monthly average concentrations between 2015 and 2017 at different urban sites, within the domains in Fig. , along with the annual average B(a)P concentration at Birkenes, a regional background air quality observatory. All urban measurement sites show the same B(a)P seasonality with the highest values in January. The most pronounced B(a)P profile is obtained in Lillehammer, followed by Oslo and Drammen, where measurements show similar monthly profiles and concentration levels, and Trondheim, where the lowest B(a)P levels are obtained. The B(a)P annual mean concentration in all urban areas varies from 0.18, 0.30 and 0.20 ng m-3 in Trondheim to 0.56, 0.61 and 0.68 ng m-3 in Lillehammer in 2015, 2016 and 2017, respectively. These levels are below the B(a)P European target value (i.e. 1 ng m-3) but above the reference value established by the World Health Organisation (i.e. 0.12 ng m-3). Unlike in Birkenes, where values represent regional background levels, B(a)P annual mean concentrations are below the WHO reference value, measured at 0.013, 0.010 and 0.011 ng m-3 in 2015, 2016 and 2017, respectively. Figure  shows in addition the monthly normalised HDD15, HDD10 and HDD5 obtained for Oslo (grey shades areas) based on outdoor temperatures for the same period. In MetVed, the monthly normalised HDD relates outdoor temperature with activity (i.e. wood consumption). Figure  shows the effect that different thresholds (i.e. 5, 10 or 15 ∘C) would have on monthly wood consumption and therefore emissions. A higher temperature threshold will increase activity into spring and autumn, and a lower temperature threshold will limit activity to wintertime. The overall B(a)P best fit to these profiles is obtained using HDD5, suggesting a more intense source during winter than what is obtained with HDD15. This is in agreement with the BC observations, which is the opposite of what is observed based on PM2.5 measurements.

The annual B(a)P average level at each site is indicative of total RWC activity in the area, and the levels are well in agreement with the spatial distribution of emissions in the MetVed model. Lillehammer, with the highest B(a)P levels, is the least densely populated area of those considered here but has high emissions from RWC due to high wood consumption in the region, and the area surrounding the station is made up of mainly detached houses with a high wood burning potential. Trondheim, with the lowest B(a)P levels, is a highly populated urban area located in a county with a high wood consumption. The proxies behind MetVed entail the lowest emissions compared with other urban areas (Fig. a) as a result of the high share of apartments with alternative residential heating sources other than wood based, which represent the observed B(a)P differences well.