Size-Resolved Respiratory Tract Deposition of Sub-Micrometer Aerosol Particles in a Residential Area with Wintertime Wood Combustion

Particle size distributions and hygroscopic growth were studied in a town in Sweden with extensive emissions from wood combustion. The average deposited fraction of particle number, surface area and volume dose in the human respiratory tract was estimated using the data set, as well as the typical deposition pattern of the two dominant particle source types: wood combustion and traffic exhaust. As far as we know, this is the first report on the deposited fraction and hygroscopicity of ambient particles from domestic wood combustion in the literature. The use of PM2.5 as a substitute for the deposited dose was also tested. Source/receptor modeling and the hygroscopicity measurements showed that wood combustion and traffic exhaust are dominant sources, and that these particles have a low water uptake. Number fractions of 38 and 69% of the wood combustion and traffic particles, respectively, were deposited in the respiratory tract, and 53% of the particles were deposited as an average for the whole period. The deposited fraction of the surface area and volume dose was also calculated for wood combustion particles, with the result being 22% for both parameters. The results also revealed that the PM2.5 average over more than 10 hours correlated well (r > 0.80) with the deposited surface area and volume dose. This means that PM2.5 can be used as proxy for the deposited dose when examining health effect relationships during short-term exposure studies if the averaging time is sufficient, while a PM2.5 proxy is not recommended for shorter averaging times.


INTRODUCTION
Residential wood combustion has been associated with adverse health effects on a societal level in epidemiological studies as well as with more limited number of subjects in chamber exposure studies (Boman et al., 2003;Barregard, et al., 2006;Sehlstedt et al., 2010).There is, however, insufficient evidence available to decide whether particles from wood combustion are more or less toxic than other combustion-derived particles and ambient fine particles in general (Naeher et al., 2007).To understand better the toxicity of the particles it is vital to determine the amount, or "dose", that actually enters the body.The main route of exposure is deposition in the respiratory tract during breathing.
The dose is often not linearly related to particle concentration as might be expected as a first approximation, since the respiratory tract deposition could vary substantially between particles from different sources.This is because deposition probability is not only a function of concentration, but also of parameters such as particle size distribution, water uptake, density, breathing pattern and lung morphology.For example, the particle surface area deposited dose of ambient traffic exhaust particles is three times higher than for laboratory generated wood combustion particles for an equal inhaled mass (Löndahl et al., 2009).
There is a lack of data on respiratory tract deposition of ambient particles.Two studies describe measurements of lung deposition in urban environments (Montoya et al., 2004;Löndahl et al., 2009).However, neither provide information about how deposition is related to particle concentrations as function of time, which is important for dose estimates in exposure-response time-series studies.In a recent study, Ham et al. (2011) showed that filter measurements of toxic PM components have a potential to improve the correlation with adverse health effects compared to PM 2.5 .Furthermore, Kelly et al. (2011) have developed a lung deposition model of the particle number size distribution that can be used in epidemiological studies.
The objective of this work is to two-fold.First a unique data set of two months measurements of particle number size distributions and hygroscopic growth is presented for particles in Lycksele, a town in northern Sweden, with extensive domestic wood combustion where polluted air is often trapped in an inversion-like situation.Second, this data set is used to estimate the size-resolved respiratory tract deposition of sub-micrometer particles.The particle classification is made according to mobility diameter (using Differential Mobility Analyzers) which has been shown to well describe the deposition size-pattern in the respiratory tract for particles < 400 nm, regardless of particle shape (Rissler et al., 2012).In the cold season, when measurements were carried out, most particles originate from either traffic or residential wood combustion, while the contribution from long-range transport is small compared to most other populated places.Thus, the air pollution in Lycksele provides an opportunity to find correlations between deposited dose and emissions from two major sources of aerosol.The particle origin is derived by source/receptor modeling.Since PM 2.5 was measured in Lycksele in parallel with the other particle monitoring instruments, it is also investigated how surface area, volume and mass deposited fraction correlates with this PM fraction.It is discussed whether PM exposure levels is sufficient for prediction of lung dose in epidemiological time series studies or if there is potential for improvement of associations between air pollution and health effects by adjustment of data for lung deposition probability.

Site Description and Measurements
The measurement campaign was conducted between January and March 2002 in the residential area of Forsdala (6454.75'N,1871.75'E,240 m. a.s.l.) in Lycksele, a town in northern Sweden with a population of 8500.The central residential areas of Lycksele, alike Forsdala, are confined in a valley with a large through flowing river, which often gives stagnant meteorological conditions especially during cold days with PM 10 surpassing 30 μg/m 3 .At the time of measurements, Forsdala had 101 local fireplaces (mostly conventional wood-burning stoves) and 13 nonenvironmentally approved old boilers without a heat storage tank.The latter are in general worse pollutants than other types of devices (Johansson et al., 2004).Other heating systems in the area were wood boilers connected to heat storage tanks, multifuel boilers, and electrical radiators.
A trailer housed the instruments with a PM 10 inlet (Commercial Rupprecht and Patashnik) on the roof placed about 4 m above the ground.Air was drawn through the inlet (16.7 L/min), and a fraction of 5 L/min was directed to a differential mobility particle sizer (DMPS) used for the particle number size distribution measurements.The large temperature difference between indoor and outdoor air enabled a determination of the dry particle number size distribution from 3.2 to 930 nm diameter in 38 mobility channels with a 10-minute time resolution (see Rissler et al., 2004).Another 1 L/min of the inlet flow was directed to a hygroscopicity tandem-differential mobility analyzer (H-TDMA), which was used to measure the hygroscopic diameter growth at 90% relative humidity (RH) of the initially dry particles between 20 and 265 nm diameter with a 1hour time resolution (details in Rissler et al., 2004).A tapered element oscillating microbalance (TEOM, series 1400a, Rupprecht and Patashnik) was used to measure PM 2.5 and PM 10 alternatively by shifting between a PM 2.5 and a PM 10 inlet every 30 minutes.

Hygroscopic Growth Rate at 99.5% Relative Humidity
To determine how the particles grow due to water absorption in the respiratory tract, the diameter growth factor at the RH prevailing in the lungs, i.e., 99.5% (Anselm et al., 1990;Löndahl et al., 2006), was calculated from the hygroscopic growth measured with the H-TDMA at 90% RH.The calculation previously explained by Rissler et al. (2006) and Rissler et al. (2010) is based on classical Köhler theory and can be found in the Appendix together with details about the H-TDMA measurements.Extrapolation from 90% RH to super saturation has been shown to be reasonably accurate in previous studies (Rissler et al., 2010), and could therefore be assumed to be reliable also at 99.5% RH.Furthermore, in one respiratory tract deposition study of NaCl it was shown that the deposition pattern agrees well if NaCl is assumed to be at equilibrium at 99.5% RH (Löndahl et al., 2006).
The calculation was done in the following way: First, measured hygroscopic growth data at 90% RH was used to determine the number of soluble entities (ions or nondissociating molecules) in the aerosol particles as a function of size.For particles > 265 nm the same ion density was used as for the 265 nm particles.Second, this information was used to extrapolate hygroscopic growth to 99.5% RH.
The H-TDMA data also gives information about the external mixture of particles.For each size, one or two separate growth factors are identified.In this way, "less hygroscopic" particles could be separated from "more hygroscopic" particles with a higher content of soluble compounds."Less hygroscopic" and "nearly hydrophobic" particles from the definition of Swietlicki et al. (2008) are lumped into only one class denoted as "less hygroscopic" in this paper, while the "more hygroscopic" particles follow the same convention with a growth factor higher than 1.33 at 90% RH as in the Swietlicki et al. paper.

Calculation of the Deposited Fraction in the Respiratory Tract
The deposited number concentration of particles in the respiratory tract as function of dry particle diameter was calculated employing the following three-step process: 1. Hourly averages of the dry particle number size distribution during the entire winter campaign in Lycksele were acquired from the DMPS data.2. Wet number size distributions, when humidified to 99.5% RH, were calculated for each of the two hygroscopic growth factor modes of particles for each hour-averaged data point, by means of DMPS data and H-TDMA data as explained in the previous chapter.3. The size-dependent probability of deposition in the respiratory tract for each data point and each of the two growth factor groups was then estimated by means of the probabilities of deposition of dry hydrophobic particles of the same sizes as the humidified particles.
The deposition fraction for dry hydrophobic particles (i.e., growth factor equal to 1) was taken from the literature as described in the next paragraph This approach has been thoroughly described in previous papers (Löndahl et al., 2006(Löndahl et al., , 2008(Löndahl et al., , 2009)).The literature data for dry hydrophobic particles has been taken from the Multiple Part Particle Deposition (MPPD) model (version 1.0, Chemical Industry Institute of Toxicology, Research Triangle Park, NC), with inspiratory fraction of 0.5, particle density of 1000 kg/m 3 , nasal breathing, tidal volume of 0.625l, and a breathing frequency of 12 breaths per minute.The reason for choosing this model is that the respiratory deposition of sub-200 nm particle diameter resembles that of the deposition previously measured with the RESPI instrument of hydrophobic particles during spontaneous breathing (Löndahl et al., 2007).For instance the MPPD model fits better with these empirical data than the ICRP model (ICRP, 1995), which is also frequently used.However, the MPPD model still predicts a slightly lower deposition than measured with RESPI for particles below 70 nm.
Size dependent deposited surface area and volume fraction was calculated from the particle number size distribution data assuming that the particles are spherical.The motivation and potential errors introduced due to this assumption is discussed in the section "Respiratory tract deposition for the surface area and volume concentrations" below.

Source/Receptor Modeling
A source/receptor model was used to calculate source contributions to the size-resolved particle number concentration in Lycksele.The source profile input to the model consisted of log-normally fitted particle number size distribution data.There were two source profiles extracted: 1) a traffic tail pipe exhaust profile, and 2) a profile which consisted of two sources merged, wood combustion and long range transport (background), since these sources could not be separated from each other in the source profile.More information is provided in the Appendix.

Overview and Source Contributions
The source/receptor analysis below shows that traffic vehicle exhaust and wood combustion sources contribute roughly equally to particle number concentrations in Lycksele over the period studied.The background contribution is small, however still significant.
There are in total 538 hourly averaged data points when there is data available in parallel from the DMPS and H-TDMA.For this subset of data, the average particle number concentration is 3590 cm -3 .Of these, traffic vehicle exhaust contributes with 1510 particles cm -3 , while the contribution from the combined wood combustion with background source is 2080 cm -3 .
In an attempt to separate the background from the combined wood combustion/background source, the total particle number concentration from the two Finnish background sites Hyytiälä (61°51'N 24°17'E, 181 m a.s.l.) and Pallas (67°58'N, 24°07'E, 560 m a.s.l.) was used.The average background concentration in Lycksele was estimated as the average of Hyytiälä and Pallas concentrations.This yielded an average wood combustion and background contribution in Lycksele of 1540 and 540 cm -3 , respectively.See Asmi et al. (2011) for an overview of size distribution data from Pallas and Hyytiälä.The use of the stations Hyytiälä and Pallas can be justified by the Asmi et al. (2011) paper.It shows that there is a smooth transition towards lower particle number concentrations for four measurement sites including Hyytiälä and Pallas along a south-northerly transition line from southern Sweden to northern Finland.
In the forthcoming, the wood combustion and background sources are not separated from each other for individual data points.Often the air mass over Lycksele is not the same as over Hyytiälä and Pallas, since these sites are both at a distance of about 500 km.However, the contributions calculated above should be stable, since they are averaged over a long period.
The hourly averaged contributions from the two sources, traffic and wood combustion with background, to the particle number size distribution as function of local time can be seen in Fig. 1.Traffic particles are emitted mainly during the morning rush hour after 06:00, and the afternoon rush hour between 15:00 and 17:00.The contribution from wood combustion is also strong in the morning when people light their boilers and stoves as they wake up.However, the strongest peak is found during evening, when people again burn more wood for heating purposes.The wood combustion with background particles are dominated by a mode around 65 nm in diameter, while the traffic particles are dominated by a mode around 20 nm in diameter.

Hygroscopic Growth
H-TDMA data described in more detail in the following show that the hygroscopic growth factor is generally low for the Lycksele particles, especially for the smaller size range.
In each H-TDMA scan, two growth factor classes were allowed: denoted "less hygroscopic" and "more hygroscopic" particles, which are defined as outlined in the Methods section.On average, most particles with dry diameters below 30 nm belong to the less hygroscopic group.With increasing size, the fraction of particles in the more hygroscopic group increases, and at sizes above 200 nm diameter, almost half of the particles belong to the more hygroscopic group.The growth factor at 99.5% RH increases from about 1.3 to about 4.0 with increasing size between 10 and 900 nm diameter for the more hygroscopic aerosol group, while it increases from about 1.0 to 1.8 for the less hygroscopic group.See Fig. 2(a).
To be able to estimate the hygroscopicity and respiratory tract deposition of traffic exhaust and wood combustion particles separately, typical representative hours of each source needs to be identified.As mentioned in the methods section, it is not possible to separate wood combustion particles from background particles for individual data points.However, there are a few data points when the contribution from the background can anyhow be neglected.Namely, by selecting occasions with concentrations higher than 10000 cm -3 , 0% contribution from traffic according to the source apportionment, and the period between 19:00 and 03:00 local time, it should yield a number of data points where wood combustion is the dominant source.This criterion was fulfilled for 5 hours.Correspondingly, occasions with 100% contribution from traffic defines typical traffic size distributions, which yielded 13 hours in total.
It is expected that the more hygroscopic fraction increases with size for the average particle population in Lycksele as shown in Fig. 2(a), since the larger particles partly originate from relatively hygroscopic background particles.During the periods dominated by traffic or wood combustion, most particles are expected to be less hygroscopic independent of particle size.This is indeed the case for the 5 hours dominated by wood combustion (Fig. 2(c)).However, for the 13 hours with traffic particles, the more hygroscopic aerosol fraction is approaching 40% for sizes larger than 200 nm diameter (Fig. 2(b)).The reason for the hygroscopic particles at larger sizes that were observed for the traffic source may be that the 13 hours supposedly dominated by traffic emissions were mixed with background particles.These particles appear in a size range where the traffic contribution is normally minor.Thus, also a small total fraction of background particles may still comprise a substantial fraction of the larger particles.
Growth factor parameterizations of H-TDMA data are provided in the Appendix.

Respiratory Tract Deposition
The respiratory tract deposition was calculated by number, surface area and volume of the particles.All these three dose measures have potential importance for health effects.The mass concentration, which might be calculated from the measured volume concentration, has been the standard in air quality guidelines and is the particle quantity used in most epidemiological studies.For soluble particles, mass is the only of the three characteristics that is of relevance since the soluble species dissolve immediately after deposition in the lung lining fluid.For non-soluble particles, the particle surface area is attracting increasing interest as several studies have shown a close correlation with the toxicological response -a result that may be explained by the interactions at the interface between the particles and the exposed tissue or proteins body fluids (e.g., Stoeger et al., 2006).This particle property is however not always easy to attain for irregular particles (like soot agglomerates) (Messing et al., 2012), or particles containing soluble material with an insoluble core.Number concentration is sometimes more readily measured and has been indicated to be directly associated with health outcomes (Wittmaack, 2007).

Respiratory Tract Deposition for the Number Concentrations
In summary, roughly 2/3 of the number of traffic exhaust particles in Lycksele are deposited in the respiratory tract, while only about 2/5 of the wood combustion particles are deposited.The derivation of these values follows below.
Fig. 3(a) shows the average deposition probability as function of dry particle diameter as estimated from the MPPD model adjusted for hygroscopicity of the particles.Compared to the hydrophobic particles according to the MPPD curve, the deposition probability minimum of the less hygroscopic and the more hygroscopic particles is shifted from about 450 nm diameter to 220 and 120 nm diameter respectively.This shift in the deposition probability gives the average deposited size distribution as shown in Fig. 4(a).According to the figure, a large part of the deposition takes place around 20 nm diameter, where most particles deposit in the alveolar region of the respiratory tract as shown in Fig. 3(b).Fig. 3(c) is showing the tracheobronchial deposition for comparison.The average fraction of particles deposited in the human respiratory tract is 53%, and 84% of the deposited particles belong to the less hygroscopic group.
The deposited fraction is also calculated for the wood combustion and traffic particles separately.By comparing Figs.4(a) and 4(b) it can be seen that the wood combustion size distribution peak concentration is positioned at larger sizes, or in other words closer to the deposition minimum in the MPPD curve in comparison to the average size distribution.Hence, the deposition probability is lower than for the entire period.Vice versa is true for the traffic distribution seen in Fig. 4(c) due to a high concentration of particles deposited in the alveolar tract for particles smaller than 30 nm diameter.Thus, the resulting deposited number size distribution is determined by both the number size distribution and the hygroscopic growth at 99.5% RH (i.e., by the wet number size distribution).
The 538 individual hours for the deposited versus the measured particle number concentration can be seen in Fig. 5 together with the relative contribution from the wood combustion with background source.A linear regression line for the 5 wood combustion hours and 13 traffic hours from Figs. 4(b) and 4(c) respectively is also shown in the graph.Most of the data points fall between these lines as expected if the particles have their origin partly from wood combustion and partly from traffic.There is a minor contribution also from the background particles, especially in the smallest concentration range.
The results for the deposited number fraction for the traffic particles based on linear regression of the 13 traffic hours in Fig. 5 is given in Table 1, where it can be seen that 69% of the traffic particles are deposited in the respiratory tract.
For the size range 10 to 600 nm, the deposition fraction becomes 61%.This is the same diameter interval as used in the study with the RESPI instrument by Löndahl et al. (2009) where a deposition fraction of 68% was found for nine volunteers exposed to curbside particles in Copenhagen, Denmark.Hence, these deposition patterns are similar, which is not a coincidence, since both size distributions and In the same manner, linear regression shows that 38% of the wood combustion particles are deposited in the respiratory tract (Table 1).This value is higher than the deposited fraction measured by the RESPI in the laboratory for both the experiments with efficient as well as low temperature combustion in the study by Löndahl et al. (2008) (21% and 23% respectively).The reason for the higher value for Lycksele is mainly because there is a larger fraction of smaller particles than in the laboratory, and hence the alveolar deposition is higher.In addition, there is a difference in hygroscopic growth, which is dependent on combustion efficiency and may vary from hydrophobic to salt dominated hygroscopic particles (Rissler et al., 2005).
A separation has also been made for alveolar and tracheobronchial deposited fraction for health effect purposes, and it shows that the alveolar and tracheobronchial deposited particle fraction of wood combustion particles is smaller than for traffic particles (Table 1).This is the expected outcome due to the shape of the traffic and wood combustion size distributions, and the hygroscopic growth.

Respiratory Tract Deposition for the Surface Area and Volume Concentrations
The analysis described in detail below shows that about 1/5 of the wood combustion surface area as well as volume are deposited in the respiratory tract, while roughly 1/3 of the traffic exhaust surface area and volume are deposited, although the latter value is more uncertain.
The size-dependent average surface area and volume size distributions and the deposited counterpart can be viewed together with the average number and deposited number size distributions in Fig. 6.The deposition probability by surface area and volume is lower than for number, since The fraction of deposited surface area and volume for traffic particles is 31% and 32% respectively, while it is 22% for the wood combustion particles for both of the parameters (Table 1).These deposition fractions for the traffic exhaust particles are similar to those measured by the RESPI system at a street with heavy traffic in Copenhagen (35% and 28% for the surface area and volume, respectively, Löndahl et al., 2009).The wood combustion deposition is also similar to efficient (23% and 24% respectively) as well as inefficient wood combustion (23% and 24% respectively) as measured with the RESPI system (Löndahl et al., 2008).While the size distribution and hygroscopic pattern in Lycksele and Copenhagen are similar for the traffic particles, the similarity between Lycksele and RESPI wood combustion experiments is coincidental since the size distribution and the hygroscopicity are not alike between Lycksele and the RESPI data from wood combustion.
The wood combustion with background source contributes 79% and 80% on average to the surface area and volume, respectively.Since it is difficult to identify a traffic source profile in the source/receptor modeling for Lycksele which is completely free of influence from other sources in the sub-900 nm diameter range, it suggests that a portion of the traffic particle surface area and volume is mixed with Table 1.The number, surface area and volume respiratory tract deposited particle fraction of 3.2-930 nm diameter particles for the traffic (13 hours) and wood combustion source (5 hours).Calculations are based on linear regression of the data points of the deposited part of the size distribution versus the measured size distribution (uncertainty margins for Lycksele data are within the 98% confidence interval of the regression).Results are also provided for tracheobronchial and alveolar deposition.wood combustion or background particles.Hence, it is likely that the contribution from the wood combustion with background source is even higher than 79% and 80%, and that the deposited fraction of traffic particles for surface area and volume has considerable error margins, an error which is difficult to assess based on the available data.
The estimations of volume and surface area are based on the assumption of spherical particles.Since freshly emitted soot particles, in several studies, have been shown to be highly agglomerated and thus far from spherical (Park et al., 2003;Rissler et al., 2012) such assumption could result in a severe overestimation of the deposited particulate volume and an underestimation of surface area when calculated from mobility number size distributions (as measured by a mobility particle sizer type of particle number size distribution instrument) (Rissler et al., 2012).However, in the case of the particles found in Lycksele during the campaign there are several factors that indicates that the assumption of spherical particles is the best assumption to be made.Generally, as agglomerate particles are aged and processed in the atmosphere, secondary material is condensed unto the particles, filling external voids and making the surface hydrophilic.A hydrophilic surface, in turn, leads to particle water adsorption and further restructuring of the agglomerates into more spherical shapes at RHs typically found in the ambient (Pagels et al., 2009).The particles found in Lycksele are most likely not pure soot.For the traffic generated particles there is a large contribution to the particulate matter, not only from primary soot emissions, but also from secondary organic material originating mainly from gasoline engines.This material will likely condense onto already existing particles, as discussed previously.Furthermore, the dominating part of particles originating from biomass combustion in Lycksele were most likely formed under poor combustion resulting in particles containing a high fraction organic compounds (Bølling et al., 2009), which typically do not form agglomerates.Furthermore, from looking at the hygroscopic growth data (averaged over the entire period) of particles with sizes closest to the volume mode peak (~250 nm) nearly 50% of the particles belong to the more hygroscopic group of particles.These are certainly not agglomerates.Also the group of "less hygroscopic particles" does have a slight content of water soluble material (Table 2) leading to a water uptake and likely a restructuring due to the surface tension.Thus, even if it is likely that some of the very fresh diesel soot particles might be highly agglomerated, the dominating part of the ambient particles found during winter time in Lycksele, especially those of sizes that are most relevant for the volume size distribution, are compact particles rather than agglomerates.However, one should still bear in mind that there is a risk of overestimating particle volume assuming spherical particles.Note that the estimation of deposited number concentration is not affected by this assumption.

Can PM Be Used as a Proxy for the Deposited Dose?
When the averaging time is longer than 10 hours, PM 2.5 is highly correlated with the deposited fraction of surface area or volume (Pearson's product-moment correlation coefficient, r 2 > 0.80).
This indicates that PM 2.5 could be used as a proxy for deposited dose in time-series exposure health effect studies, at least when averaging times are sufficient.
On short time scales the particle size distribution and particle chemistry show large variability.Thus, for similar total mass concentrations (e.g., PM 2.5 ) the deposited dose of inhaled particles may be varying.For instance, a fresh aerosol dominated by ultrafine hydrophobic particles have a different probability to deposit in the lungs compared to an aged aerosol with larger more hygroscopic particles even if the mass concentrations are equal.When data are averaged over long periods the temporal variations in the particle characteristics are smoothed out.Thereby correlations between PM 2.5 and deposited dose increase with averaging time.
Hence, for 1 hour averages of the dose calculated from size distribution and hygroscopicity, the correlation with PM 2.5 was much lower (r 2 ~0.55), suggesting that in health effect studies, where exposure on 1-hour basis (or a few hours) is important for acute health outcomes, PM 2.5 would no longer be a good measure for the deposited dose.Similar exposure concentrations by particle mass will result in different amounts of particles deposited after inhalation.This is the expected outcome when individual aerosol sources dominate the contribution to PM during different hours.Namely, for similar PM masses, deposited surface area and volume will not be similar for different aerosol sources due to the different shapes of the particle surface area or volume size distributions.
There may also be experimental reasons for a decrease in correlation between deposited dose and PM 2.5 .PM 2.5 is not averaged over the same time interval as DMPS and H-TDMA data (PM 2.5 is measured only during the first halfhour of the hour measurements, while the H-TDMA data is taken nearly from a full hour of measurements).The inlets for the H-TDMA and DMPS are 7 m away from the PM 2.5 / PM 10 inlets.Hence, the monitoring instruments were not necessarily measuring in the same pollution plumes, which might have decreased correlation between PM 2.5 and deposited dose.

CONCLUSIONS
Unique data of particle size distribution and hygroscopic aerosol measurements in a residential area in a Swedish Table 2.The HTDMA results where κ R (mol/m 3 ) is used representing the hygroscopic growth of each hygroscopic group of particles.Also the fraction of particles belonging to each group (AF) is presented.Aditionally, κ P , defined according to Petters and Kreidenweis (2007)  Using the measurements and the MPPD model of respiratory tract deposition, it is possible to determine the respiratory deposited fraction of particle number, surface area and volume concentrations.The analysis showed that the number deposition depends strongly on source, and about 2/3 of the traffic particles and about 2/5 of the wood combustion particles are deposited.Only 1/5 of the particle surface area and volume are deposited for wood combustion particles.A higher; however also a more uncertain value of 1/3 has been estimated for the traffic particles.
The correlation between PM 2.5 and deposited surface area and volume is high for averages longer than 10 hours.This indicates that PM 2.5 can be used as a very good proxy for the deposited surface area and volume in short-term health-effect epidemiological studies if a persistent daylong exposure to high particle concentrations is important for health effects.Conversely, on a 1-hour basis, the correlation is lower.In a situation, where a 1-hour or a few hours high concentration peak is important for health outcomes, the deposited dose might not be properly reflected by PM 2.5 concentrations, and a better dose estimate is needed.

ACKNOWLEDGMENTS
The Swedish Energy Agency is acknowledged for financial support of the project and the colleagues of the Environmental Administration of Lycksele for their cooperation in Lycksele.Heikki Lihavainen is acknowledged for providing Finnish data, and Jingchuan Zhou is acknowledged for taking part in the experimental work and analysis of data.Anna Oudin is finally acknowledged for the discussions about short-term health effect studies.

Hygroscopic Growth Rates at 99.5% RH
According to Köhler (1936) theory assuming volume additivity (V d + V w = V sol ) and ideally behaving solutions, the hygroscopic growth factor (Gf), defined as the wet diameter of the particle divided by the dry diameter (D/D p ), is described by: 3 1 1 where V d , V w , and V sol are the dry, water and total solute volume of the particle respectively, a w water activity, M w and ρ w the molecular weight and density of water.κ R represents the effective number of moles of soluble entities per dry particle volume unit, a concept introduced by Rissler et al. (2006).Note that the Gf is determined solely by κ R , a w , and some constants.a w is, according to Köhler theory related to RH as: where C k is the Kelvin curvature correction factor given by 4 exp σ s is the surface tension of the solution and R and T the ideal gas constant and temperature respectively.Estimating Gf from κ R -values does include some iteration since C k depends on wet diameter D.
From the hygroscopic growth at 90% RH, the effective number of soluble entities per volume dry unit, κ R , was estimated according to Eqs. (1-3).Then, these equations and the value of κ R , were used to calculate the Gf at 99.5% RH.
The H-TDMA stepped from 20 nm up to 265 nm (20, 35, 50, 73, 109, 166 and 265 nm), performing 1 or 2 scan cycles every hour.Only those data when there were five or more size scans available each hour was used in the analysis, and a time resolution of 1 hour was chosen for the final data.After estimating κ R for the sizes measured by the H-TMDA, κ R was interpolated for sizes in between the measured ones, and for smaller and larger particles (< 20 nm and > 265 nm) κ R was kept fixed according to the GF of the lowest and highest measured diameter.
The κ R values for each of the size intervals and for the wood combustion with background and traffic exhaust particles as well as the average Lycksele particle population is presented in Table 2. Additionally, the κ P values from the alternative Kappa-Petters formulation (Petters and Kreidenweis, 2007) are given in the same Table.

Source/Receptor Modeling
There are two source/receptor model results on particle number size distributions from Lycksele.The first one is for the current data from 2002 using the COPREM model (Wåhlin, 2003).The second one comes from PMF modeling (Paatero and Tapper, 1994) on size distribution data measured during 2006 (Krecl et al., 2008).The choice has been to stick with the COPREM model in the current paper for the 2002 data, since it gives the opportunity to compare with PMF model results on the 2006 data.Details about how the source receptor modeling was prepared for the current data are described in Kristensson (2005).A shortened explanation follows below.
The COPREM model diverged when the 38 size channels in the particle number size distribution were used as source profile parameters.To be able to run COPREM, the number of parameters describing the different source profiles had to be decreased to only one, namely the ratio of the number concentration of larger sized particles and the number concentration of smaller sized particles.Hence, the size distribution in Lycksele was fitted with 5 log-normal modes with non-linear least-squares fitting as described by Kristensson (2005), where the number concentration for the modes 1 and 2 (3-35 nm diameter in geometric mean diameter) belonged to the group of the smaller sized particles, and where the concentration in modes 3, 4, and 5 (40-850 nm diameter in geometric mean diameter) belonged to the larger sized particles.
Full constraints were used in the COPREM source receptor modeling.The traffic source profile was calculated as the average of the ratio of the number concentration of larger to smaller particles during occasions when the temperature was relatively high, and when then the concentration of traffic particles in mode no. 2 was high.Conversely, the wood combustion and background source profile was selected from periods with low temperatures, during nighttime when there was little traffic, and when the meteorological stability was high allowing for high concentrations of wood combustion particles.The values for the ratio between large and small particles for the traffic and wood combustion source profiles became 0.222 ± 0.081 and 2.816 ± 0.977 respectively.
The average particle number size distributions that gave these source profiles can be viewed in Fig. 7 and lognormal fitting parameters of modes 1 to 5 describing these distributions can be viewed in Table 3.The size distribution typical of traffic vehicle exhaust peaks around 20 nm diameter (mode 2) with a second strong peak around 8 nm (mode 1), while the wood combustion with background size distribution peaks around 65 nm (mode 3) with two additional distinct peaks around 25 (mode 2) and 150 nm (mode 4) respectively.
The traffic and wood combustion size distributions are similar to the Lycksele 2006 measurements (Fig. 7, Krecl et al., 2008).The traffic source distribution is also similar to the distribution measured in a road tunnel in Sweden (Fig. 7 (a), Kristensson et al., 2004), and the wood combustion size distribution is similar to the size distribution measured in the chimney of a conventional wood stove (Fig. 7(b), Hedberg et al., 2002).
The same source contribution was used for the surface area and volume concentrations.While this is satisfactory for number concentrations, it unfortunately disqualifies any additional sources that are potentially important for especially the volume source contributions, like road dust.tunnel, 1999tunnel, , Kristensson et al., 2004)), with laboratory measurements from 1999 in a conventional wood stove (Hedberg et al., 2002).The size distribution is normalized with respect to the number concentration of particles between 28 and 900 nm diameter (the range of the 2006 Lycksele DMPS measurements).

Fig. 1 .
Fig. 1.The contribution to the particle number size distribution from (a) traffic particles, and (b) wood combustion with background particles as function of local time.Data is averaged over each hour interval.

Fig. 2 .
Fig. 2. The aerosol fraction (AF) of the more and less hygroscopic groups, and the growth factor (GF) at 99.5% relative humidity as function of dry particle diameter.Average growth factors are shown for (a) all particles, (b) typical traffic particles (13 data hours), and (c) typical wood combustion particles (denoted Wood, 5 data hours).

Fig. 3 .
Fig. 3. (a) The average total deposition curve for the more and less hygroscopic modes as well as for the hydrophobic particles from the MPPD model as function of dry particle size.(b) Correspondingly for the average alveolar, and (c) for the average tracheobronchial deposition.

Fig. 4 .
Fig. 4. (a) The average particle number size distribution measured with the DMPS and the calculated deposited average size distribution together with the average deposited less and more hygroscopic size distribution respectively (The sum of the less and more hygroscopic deposited distribution equals the average deposited distribution).(b) Correspondingly for 100% wood combustion contribution (5 hours), and (c) for 100 % traffic contribution (13 hours).

Fig. 5 .
Fig. 5. (a) The relation between the measured total particle number concentration and the calculated deposited number concentration in the respiratory tract.The color code shows the relative contribution in percent to the number concentration from wood combustion with background particles.Inserted are the regression lines for the 13 traffic (denoted Traffic in the graph) and 5 wood combustion hours (denoted Wood comb.)taken from Figs. 4(c) and 4(b) respectively.(b) A zoom-in plot of the lower concentration range.

Fig. 6 .
Fig. 6.The average particle number size distribution, surface area and volume size distribution as measured by the DMPS system and the corresponding deposited part.The volume size distribution has been multiplied by a factor 20 to enhance visibility in the graph.The surface area and volume have been calculated assuming spherical particles.

Fig. 7 .
Fig. 7.The relative particle number size distributions resulting from the source/receptor (a) traffic source profile and (b) the wood combustion with background source profile.Data from source/receptor runs during 2002 (this study), and from 2006 (also Lycksele,Krecl et al., 2008).Comparison with measurements in a road tunnel in Sweden (Söderleds tunnel, 1999,Kristensson et al., 2004), with laboratory measurements from 1999 in a conventional wood stove(Hedberg et al., 2002).The size distribution is normalized with respect to the number concentration of particles between 28 and 900 nm diameter (the range of the 2006 Lycksele DMPS measurements).
is given.κ R and κ P are related as κ R = (ρ w /M w )κ P .

Table 3 .
(Krecl et al., 2008)arameters for the size distribution in Lycksele from 2002 data (this study), and 2006 data(Krecl et al., 2008)used in the source/receptor modeling.The log-normal fitting is for the traffic size distribution and the wood combustion with background distribution with a relative concentration scale.The Krecl et al. wood with background data is taken from the source profiles no. 2 and 3 in the paper.The size distribution data is for diameters larger than 28 nm, hence no results for mode no. 1 in the log-normal fitting exist in theKrecl et al. paper.