The Helsinki metropolitan area (HMA) includes four cities – Helsinki, Espoo, Vantaa, and Kauniainen. The total population of the HMA is approximately 1.1 million, while the population of Helsinki is about 0.63 million. The contributions of the different emission source categories to the total combustion emissions of PM2.5 in HMA in 2015 were 39 % from small-scale wood combustion, 31 % from energy production and other stationary sources, 28 % for vehicular traffic, and 2 % for harbors, according to Kaski et al. (2016a, b).

The center of Helsinki is located on a peninsula that is surrounded by the Baltic Sea; the main detached-house areas are situated to the west, east, and north from the city center (Fig. 1). The annual mean temperature in Helsinki is 5.9 ∘C. However, the seasonal variation in temperatures is substantial; the monthly mean minimum and maximum is -4.7 ∘C in February and 17.8 ∘C in July. Heating in the HMA is mainly based on an extensive district heating system that has only a minor impact on air quality. The reason is that this heat is mainly obtained from energy plants burning fossil fuels; most of these plants also have very high stacks. However, fireplaces and sauna stoves are commonly used in suburban detached houses.

The measurement sites used for this study (see Table 1 and Fig. 1) in the HMA were as follows: eight sites were in detached-house areas (2008–2015), one was in a urban background (2007–2015), and three were situated within street canyons (2007, 2010 and 2015). The monitoring height at all these stations was approximately 4 m. The Kallio urban background station is situated in a sports field in the city center; its distance from the closest street with a traffic volume of 6300 vehicles day-1 is approximately 80 m. The stations in the detached-house areas, numbers 1–8 (Vartiokylä, Itä-Hakkila, Päiväkumpu, Kattilalaakso, Kauniainen, Tapanila, Ruskeasanta, and Lintuvaara), were situated in suburban areas with relatively lower traffic volumes. In these eight areas, the measurement sites were surrounded by detached houses. Wood combustion appliances were being used in 90 % of the detached houses in the HMA (see Sect. 2.3). The street canyon stations (Unioninkatu, Töölöntulli, and Mäkelänkatu) were on busy streets with high traffic volumes and nitrogen oxide (NOx) concentrations (Table 1). Average traffic density at Unioninkatu was 12 800 vehicles weekday-1 (7 % heavy traffic), at Töölöntulli it was 44 000 vehicles weekday-1 (10 % heavy traffic), and at Mäkelänkatu 28 000 vehicles weekday-1 (9 % heavy traffic).

Data from the stations in the HMA were compared with data from the rural and remote stations. Rural station 1 of Virolahti (60∘31′ N, 27∘40′ E; 5 m a.s.l) in eastern Finland is located at a distance of 160 km from Helsinki in a rural district on the coast of the Gulf of Finland (Vestenius et al., 2011). Rural station 2 is situated a distance of 210 km north from Helsinki in Hyytiälä (61∘51′ N, 24∘17′ E; 181 m a.s.l) and is a boreal forest site that is part of the SMEAR network, SMEAR II (Station for Measuring Ecosystem–Atmosphere Relationships) in southern Finland (Hari and Kulmala, 2005). The remote station of Pallas is situated in Matorova, northern Finland (68∘00′ N, 24∘14′ E; 306 m a.s.l.), a 900 km distance from Helsinki. Pallas is located in the subarctic region in the northernmost limit of the northern boreal forest zone (Hatakka et al., 2003).

The daily PM10 samples were collected on polytetrafluoroethylene (PTFE) filters (Fluoropore membrane filters, 3.0 µm, Ø 47 mm, Merck Millipore, Germany) every 2–4th day by MicroPNS – low-volume samplers. The flow rate used was 38 L min-1; the average total collected volume for the 24 h samples was 55 m3. For the analysis samples were usually pooled together as monthly samples, soxhlet extracted with dichloromethane, dried with sodium sulfate, concentrated to 1 mL, and cleaned using Florisil solid-phase extraction (SPE) cartridges. Afterwards the concentrations of BaP were analyzed using gas chromatograph-mass spectrometers (GC-MSs, Agilent 6890N and 5973). For chromatographic separation, the J&W DB 5 ms column (50 m × 0.25 mm i.d., film thickness 0.25 µm) and 5 m pre-column (Agilent FS) were used. Helium (99.9996 %) was used as a carrier gas with a flow of 1 mL min-1. The temperature program started at 60 ∘C with a 3 min hold, followed by an increase of 8 ∘C min-1 to 290 ∘C and 20 ∘C min-1 to 320 ∘C with a hold of 5 min. Deuterated PAH compounds (phenantrene-d12, chrysene-d12, perylene-d12, and dibenzo[a,h]anthracene-d14, Dr. Ehrenstorfer) were used as internal standards and added to an extraction solvent before extraction. External standards (EPA 610 Polynuclear aromatic hydrocarbons Mix, Supelco) with five different concentration levels were used. In the analysis of BaP, the ISO 12884 (2000) and EN 15549 (2008) standards were followed. Measurement uncertainty was calculated from the partial uncertainties by following the standard EN15549 (2008) for the target value (1 ng m-3) and lower concentration (0.2 ng m-3); that value was found to be 11 and 30 %, respectively. The method has been previously described in detail by Vestenius et al. (2011).

The levoglucosan concentrations were determined from daily PM10 samples taken from the urban background and detached-house areas in 2012 and from monthly means in 2011. Samples were collected on PTFE filters on the same days and at same sites as the BaP samples, extracted with 5 mL of Milli-Q water with an internal standard and analyzed using high-performance anion exchange chromatography mass spectrometry (Dionex ICS-3000) as described by Saarnio et al. (2010). The used column system consisted of a Carbopac^™ PA10 guard (2 mm i.d. × 250 mm length) and analytical (2 mm i.d. × 250 mm length) columns. The eluent was produced by a potassium hydroxide eluent generator (EGC II KOH). The standard solutions were prepared by dissolving a weighed amount of solid levoglucosan (purity 99 %, Acros Organics, NJ, USA) into Milli-Q water. Carbon-13-labeled levoglucosan in dimethyl sulfoxide (100 µg mL-1, purity 98 %, Cambridge Isotope Laboratories) was used as the internal standard.

Emissions of BaP that originated from small-scale wood combustion were evaluated in the HMA, including the spatial and temporal variation in those emissions. We previously modeled the BaP concentrations that originated from local vehicular traffic in this area (Douros and Moussiopoulos, 2014). In this study, the traffic emissions of BaP were calculated from PM2.5 exhaust emissions by a scaling factor 0.000031, which was based on traffic emissions used in LOTOS-EUROS model. According to national emission inventory for traffic exhaust emissions, the PM2.5 exhaust emission for HMA was 138 tonnes in 2014 and 258 tonnes in 2008 (www.lipasto.vtt.fi, Mäkelä & Auvinen, 2009). The BaP emission from vehicular traffic is thus less than 5 % of the emission from wood combustion. Vehicular traffic has also been shown to have a minor influence on the total emissions of BaP on European scale (Guerreiro et al., 2015). In addition, there are no potential industrial or other local sources of BaP in the HMA (Soares et al., 2014).

A novel wood combustion emission inventory was compiled for the HMA in 2014 for the following pollutants: particles (PM1, PM2.5 and PM10), NOx, non-methane volatile organic compounds (NMVOC), carbon monoxide (CO), black carbon (BC), and BaP (Kaski et al., 2016a). The BaP emissions that originate from wood combustion depend on numerous factors, such as the type and construction of a fireplace, the operating procedures and habits, and the quality, processing, and storage of the wood being used. For example, in modern fireplaces the BaP emission factors are much lower than in conventional heaters or in sauna stoves (Tissari et al., 2007). Emissions can also be lowered by using a sufficient supply of air (e.g., by controlling batch size) and using clean dry wood (Tissari et al., 2009). However, the detailed emission inventories of wood combustion are scarce internationally; even information on the amount of wood and the kinds of appliances used can be insufficient (Pastorello et al., 2011).

The amount of wood combusted, and the procedures and habits for that combustion were estimated using a questionnaire distributed in the HMA (Kaski et al., 2016a). The goal of the questionnaire was to gather quantitative information on the amount of wood combusted and its combustion characteristics to evaluate its impacts on air quality. The questionnaire was sent to 2500 inhabitants in detached-house areas. The response rate was 35 %. A stratified sampling procedure was used to ensure the representativity of the replies. The stratification procedure included the following three parameters of these houses: spatial distribution in the HMA, house construction year, and primary heating method. In the HMA, residential wood combustion is not common as a primary heating method, but it is much more frequently used as a method for supplementary heating. According to official statistics by Statistics of Finland and the Helsinki Region Environmental Services Authority (HSY), the total number of detached and semidetached houses in the HMA in 2014 was about 69 000. Most (52 %) of the detached and semidetached houses were primarily heated with electricity. The shares of other primary heating methods were 22 % oil or gas combustion, 18 % district heating, 4 % geothermal heating, 2 % wood combustion, and 3 % unknown heating method (Statistics of Finland).

Based on the questionnaire (Kaski et al., 2016a), wood combustion was used in approximately 90 % of the detached and semidetached houses in the HMA. However, wood combustion was seldom used as a primary heating method in these detached and semidetached houses (only in approximately 2 % of the houses); however, it was much more commonly used as a supplementary heating method, and as fuel for the sauna stoves. The annual average amount of wood burned per house was 1.52 (±1.91 standard deviation, SD) solid cubic meters. Most of the wood was used in heat-storing masonry heaters (0.72 ± 1.19 solid m3 house-1) and sauna stoves (0.31 ± 0.77 solid m3 house-1); only a minor amount was burned in boilers (0.09 ± 0.95 solid m3 house-1). In addition, several other fireplace types used wood as a fuel (0.40 ± 1.08 solid m3 house-1; e.g., open fireplaces, ovens, and stoves).

In this current study, the emission factors for different types of fireplaces were adopted from the literature (Tissari et al., 2007; Todorović et al., 2007; Hytönen et al., 2009; Lamberg et al., 2011). The BaP emissions from wood combustion were calculated using the following emission factors: 809, 68, and 102 µg MJ-1, for sauna stoves, boilers, and other fireplace types (e.g., heat-storing masonry heaters, open fireplaces, ovens, stoves), respectively (Tissari et al., 2007; Todorović et al., 2007; Hytönen et al., 2009; Lamberg et al., 2011). The total BaP emissions from wood combustion were estimated to be 196 kg in the HMA in 2014. The shares of BaP emissions were 2 % for primary heating boilers, 67 % for sauna stoves, and 31 % for other fireplace types. The pollutant emissions from sauna stoves are very high since their combustion conditions are usually very poor compared to other fireplace types (e.g., Savolahti et al., 2016 and references therein). Unfortunately, only a limited number of studies provide BaP emission factors for the typical types of fireplaces used in Finland. Therefore, it is very important to undertake new combustion experiment studies to achieve more robust knowledge on the BaP emission factors for sauna stoves and various other fireplace types as well as additional for different burning conditions.

Karvosenoja et al. (2008) presented an uncertainty evaluation for the PM2.5 emissions of residential wood combustion in Finland. This uncertainty was estimated to be relatively low for the amounts of wood burned in different fireplace types (±25 % at confidence limits of 95 %), whereas it was much higher for the emission factors (-54 % lower and +88 % higher at confidence limits of 95 %, respectively). The background information and the methods of evaluation were similar to those found in Karvosenoja et al. (2008) for the emission inventory of this study. Therefore, we estimated that the uncertainties regarding the amounts of wood used are similar to those estimated by Karvosenoja et al. (2008). However, the uncertainties regarding the emission factors of BaP from wood combustion may be higher than those evaluated by Karvosenoja (2008) for PM2.5. The reason is that estimates of the PM2.5 emission factors from wood combustion are available from several studies (e.g., Karvosenoja et al., 2008; Savolahti et al., 2016), whereas the estimates for BaP are much more scarce.

The total amounts of wood burned and its allocation to different fireplace types depends on the primary heating method for a house (Kaski et al., 2016a). The spatial distribution of emissions for dispersion modeling was calculated using the following annual BaP emissions estimates for houses that were using different primary heating methods: 2.5, 3.7, 2.0, 4.1, 3.9, and 3.1 g house-1 for electricity, thermal, district, oil, wood, and unknown heating methods, respectively (Kaski et al., 2016a). The previously mentioned values include the total emissions from all fireplace types in a house, including sauna stoves. For the spatial allocation of these emissions, the geographical location and primary heating method information on all 69 000 detached and semidetached houses of the HMA was available from the regional basic register (SePe and SeutuCD) provided by the HSY. The spatial distribution of houses and the BaP emissions are presented in Fig. 1.

The temporal patterns (month, week day, time of day) of the emissions for three different fireplace categories (sauna stoves, boilers, and other fireplaces) were estimated based on the information gathered from the questionnaires (Kaski et al., 2016a; Gröndahl et al., 2011). Unfortunately, we could not extract sufficient information from these questionnaires to model quantitatively the influence meteorological variables, such as temperature, on the emissions. Clearly, during cold periods more wood is used as additional heating, but it was not possible to model this amount quantitatively based on the results of the questionnaire. Therefore, we used the average emission patterns for different months, weekdays, and times of day, instead of modeling the emissions based on the actual variation of meteorological parameters.

The atmospheric dispersion of BaP emissions was evaluated using the Urban Dispersion Modeling system developed at the Finnish Meteorological Institute (UDM-FMI). This system includes various local-scale dispersion models and a meteorological pre-processor (MPP-FMI, Karppinen et al., 1998, 2000a). The dispersion modeling of UDM-FMI is based on multiple sourced Gaussian plume equations for various stationary source categories (point, area, and volume sources). For the selected calculation grid, this system was used to compute an hourly time series of concentrations. The modeling system has been evaluated by Karppinen et al. (2000b).

Meteorological input data needed by the dispersion model were evaluated using the meteorological pre-processing model MPP-FMI, based on the energy budget method. The model utilizes meteorological synoptic and sounding observations, and its output consists of an hourly time series of relevant atmospheric turbulence parameters and the atmospheric boundary layer height. We used a combination of synoptic observations from the stations at Helsinki–Vantaa (15 km north of the city center) and Helsinki–Harmaja (on an island 7 km south of the city center), and sounding observations from Jokioinen (90 km northwest of Helsinki). The predicted meteorological parameters will vary for each hour of the year, but for each hour, the same value is applied to the whole spatial domain.

In this study, we evaluated the dispersion of BaP that originated from domestic wood combustion. Emissions were uniformly distributed in squares of the size 100 m × 100 m, and the model was applied to calculate the dispersion that originated from these area sources. The altitude of the releases for domestic wood combustion was assumed to be equal to 7.5 m, including the initial plume rise. This altitude value was based on the average heights of the detached and semidetached houses with the study domain and their chimneys and an estimated average plume rise (Karvosenoja et al., 2010).

The dispersion was separately computed for three different emission source categories: sauna stoves, boilers, and other fireplaces. The diurnal, weekly, and monthly variations within the emission inventory were applied to each source category. In the dispersion modeling, BaP was treated as an inert substance, i.e., it was assumed to follow atmospheric diffusion, and no chemical or physical transformation was assumed to take place within the urban timescales. We also did not allow for the dry or wet deposition of BaP. The concentrations were computed for the years 2008, 2011, 2013, and 2014 for a receptor grid with a horizontal grid spacing of 100 m × 100 m. The influence of terrain on the atmospheric dispersion is parameterized simply as surface roughness.

The spatial distributions of annual average BaP concentrations in the HMA that originated from wood combustion were predicted for 4 years: 2008, 2011, 2013, and 2014. The results for 2 years of these years, 2008 and 2011, are presented in Fig. 6a–b. The results for these 2 years were selected as they represent both the lowest and highest annual average concentrations, due to the differences in meteorological conditions.

The influence of the regional and long-range transported background was not allowed for in the values of these submitted figures. In this way, the influence of local emission sources can be visualized more clearly. The same emission values were applied for all years; however, the influence of the hourly variation in meteorological factors on atmospheric dispersion was taken into account for only all the target years. The predicted differences between the BaP concentrations were, therefore, due solely to the inter-annual variability in the meteorological conditions.

As expected, the spatial variation of the pollution distribution (Fig. 6) was closely associated with the corresponding variation of the emissions (Fig. 1b), which in turn was closely associated with the density of the detached and semidetached houses (Fig. 1a). The highest concentrations occurred in the detached-house areas where the highest annual average concentrations (in the selected calculation grid with horizontal spacing of 100 m × 100 m) were 1.0 and 1.3 ng m-3 for 2008 and 2011, respectively. In 2011, the concentrations were clearly higher than they were in 2008. In the center of Helsinki, however, the annual average concentration from local wood combustion was below 0.2 ng m-3.

To compare the predicted to the measured concentrations, a regional background concentration of 0.135 ng m-3 was added to the concentrations computed from local residential combustion. This value was the median of the measured BaP concentrations at the regional background station of Hyytiälä (RB2) in southern Finland in 2009–2014. This regional background value was assumed to be a constant both in time and throughout the entire urban area. Clearly, the actual hourly values of the regional background depend, e.g., on wind direction and other meteorological parameters. As shown in Fig. 2, the values at RB2 varied between 0.09 and 0.25. A fairly high temporal variation in the regional background would also be expected for Helsinki.

The predicted and observed annual average concentrations are presented in Fig. 7. Only those days with measurements were taken into account. The predicted concentrations agreed fairly well with the measured concentrations for four detached-house areas (DH2, DH3, DH6, and DH7). For two detached house areas (DH1 and DH5), agreement was relatively worse. In the case of the urban background site (UB), the agreement varied from year to year. For most stations and years (except for DH2 in 2008 and DH3 in 2011), the computed concentration was higher than the observed value.

One probable reason for the disagreements between modeled and observed concentrations was the inaccurate description of the temporal variation of emissions. The treatment of the temporal variation of emissions in the model is based on temporal variation coefficients (monthly, weekly, and daily). However, the model did not take into account the influence of the daily or inter-annual variation of meteorological conditions (especially that of the ambient temperature) on the amount of wood combustion (although it does take into account the influence of meteorology on the dispersion conditions). Furthermore, the uncertainty of emission factors can be substantial. Such uncertainties are partly caused by a scarcity of experimental studies regarding the emissions of various fireplace types, especially for the stoves of saunas (Sect. 2.3). Such uncertainties are also partially caused by the wide variation of emission factors in terms of the quality and processing of fuels, the quality and structure of heaters, and combustion techniques and procedures (e.g., Ozgen et al., 2014; Tissari et al., 2007, 2009; Savolahti et al., 2016).

The model also does not take into account the reactivity of BaP in the air. Heterogeneous reactions of BaP on particle surfaces may have an effect on concentrations especially in the summer (Keyte et al., 2013). However, the detailed chemical transformation equations of these reactions are still insufficiently known; it is also expected that most of the BaP molecules are located inside the bulk particles and may, therefore, not be accessible for these reactions. Therefore, BaP is expected to be removed mainly by the dry and wet deposition of particles; the degradation of BaP through chemical reactions, however, is expected to have a relatively smaller effect on the measured BaP concentrations.

Another factor that causes differences between the modeled and the observed concentrations is the spatial resolution of the modeling; the emission sources were assumed to be located in grid squares with a size of 100 m × 100 m. However, measurements represent specific spatial points; especially in the case of small-scale combustion, so the measured values may be influenced by very local distributions of sources and other features.

The regional background concentration was based on the measured values at the s Hyytiälä station in southern Finland. However, this site represented more continental conditions compared with the HMA, and thus, it may not have been sufficiently representative of the study domain.