contribution of residential wood combustion and other …...p. krecl et al.: wood combustion...

Atmos. Chem. Phys., 8, 3639–3653, 2008www.atmos-chem-phys.net/8/3639/2008/© Author(s) 2008. This work is distributed underthe Creative Commons Attribution 3.0 License.

AtmosphericChemistry

and Physics

Contribution of residential wood combustion and other sources tohourly winter aerosol in Northern Sweden determined by positivematrix factorization

P. Krecl, E. Hedberg Larsson, J. Str̈om, and C. Johansson

Department of Applied Environmental Science, Atmospheric Science Unit, Stockholm University, Stockholm, Sweden

Received: 11 January 2008 – Published in Atmos. Chem. Phys. Discuss.: 19 March 2008Revised: 15 May 2008 – Accepted: 8 June 2008 – Published: 10 July 2008

Abstract. The combined effect of residential wood combus-tion (RWC) emissions with stable atmospheric conditions,which frequently occurs in Northern Sweden during win-tertime, can deteriorate the air quality even in small towns.To estimate the contribution of RWC to the total atmo-spheric aerosol loading, positive matrix factorization (PMF)was applied to hourly mean particle number size distribu-tions measured in a residential area in Lycksele during winter2005/2006. The sources were identified based on the par-ticle number size distribution profiles of the PMF factors,the diurnal contributions patterns estimated by PMF for bothweekends and weekdays, and correlation of the modeled par-ticle number concentration per factor with measured aerosolmass concentrations (PM10, PM1, and light-absorbing car-bon MLAC). Through these analyses, the factors were iden-tified as local traffic (factor 1), local RWC (factor 2), andlocal RWC plus long-range transport (LRT) of aerosols (fac-tor 3). In some occasions, the PMF model could not separatethe contributions of local RWC from background concentra-tions since their particle number size distributions partiallyoverlapped. As a consequence, we report the contribution ofRWC as a range of values, being the minimum determinedby factor 2 and the possible maximum as the contributionsof both factors 2 and 3. A multiple linear regression (MLR)of observed PM10, PM1, total particle number, and MLACconcentrations is carried out to determine the source contri-bution to these aerosol variables. The results reveal RWC isan important source of atmospheric particles in the size range25–606 nm (44–57%), PM10 (36–82%), PM1 (31–83%), andMLAC (40–76%) mass concentrations in the winter season.

Correspondence to:P. Krecl([email protected])

The contribution from RWC is especially large on weekendsbetween 18:00 LT and midnight whereas local traffic emis-sions show similar contributions every day.

1 Introduction

Recently, renewed attention has been paid to residentialwood combustion (RWC) as a substantial source of airborneparticulate matter (PM) in regions with cold climate. Studieson the impact of RWC on air quality have been conducted inseveral countries, like Sweden (Hedberg et al., 2006; Kreclet al., 2007; Krecl et al., 2008), Norway (Kocbach et al.,2005), Denmark (Glasius et al., 2006), USA (Gorin et al.,2006), New Zealand (Wang and Shooter, 2002) and Aus-tralia (Keywood et al., 2000). In Sweden, the main en-ergy sources for residential heating in 2005 were electric-ity (∼40%), combined firewood and electricity (21%), fol-lowed by exclusively bio-fuel combustion (11%) (StatisticsSweden, 2005). The energy output from bio-fuels increasedby about 70% between 2000 and 2005 in the whole coun-try. Approximately 61% of the residential wood boilers havelow combustion efficiency and their emissions can be sev-eral times larger than modern installations (Johansson et al.,2004). The combined effect of these small scale emissionswith stable atmospheric conditions during wintertime, whichoccur frequently in Northern Sweden, can deteriorate the airquality even in small towns (Krecl et al., 2007; Krecl et al.,2008).

To implement effective strategies to control PM emissionsand assess health effects due to poor air quality, source appor-tionment of atmospheric aerosol is needed in areas with highPM concentrations. Different techniques, such as unique

Published by Copernicus Publications on behalf of the European Geosciences Union.

http://creativecommons.org/licenses/by/3.0/

3640 P. Krecl et al.: Wood combustion contribution to winter aerosol in Sweden

emission source tracers, air quality dispersion modeling orsource-receptor modeling, can be employed to estimate thecontribution of the sources. A number of elemental andmolecular tracers (e.g., potassium and chlorine, methyl chlo-ride (Khalil and Rasmussen, 2003), and levoglucosan; Hed-berg et al., 2006) have been used to identify and quantifywood smoke. However, the reliability of some of these trac-ers often suffers from high emission variability and lackof uniqueness. In contrast to these markers, radiocarbon(14C) measurements provide an unambiguous source appor-tionment of contemporary and fossil fuel derived carbona-ceous aerosol since it retains its identity throughout any at-mospheric chemical change (Reddy et al., 2002). In the at-mosphere, high temporal resolution measurements of manyof these tracers are not possible due to the necessity of largesampling volumes to detect the concentrations accurately.On the other hand, atmospheric dispersion modeling can pro-vide spatial and temporally resolved source contributions butcan be difficult to perform accurately since detailed quanti-tative information of the emissions and meteorology is re-quired, and in some cases also aerosol chemical transfor-mation and removal processes might be considered. Thus,dispersion model calculations need to be validated. Briefly,aerosol source-receptor modeling quantifies the impact ofvarious relevant sources to the concentrations measured ata certain site (the receptor). Among source-receptor mod-els, positive matrix factorization (PMF) has been extensivelyused for source apportionment of particle composition data,where the goal is to determine the sources that contributeto PM samples (e.g., Hedberg et al., 2005; Hedberg et al.,2006). Lately, PMF has been applied to particle size distri-bution data to estimate possible sources from model identi-fied particle size distributions (Kim et al., 2004; Zhou et al.,2004). Continuous aerosol size distribution measurementscan provide very large data sets with high temporal resolu-tion, which is relevant for source apportionment calculations.

In order to characterize the urban aerosol during the woodburning season in Northern Sweden, a field campaign wasconducted in a residential area in winter 2005/2006. In thisstudy, hourly mean particle number size distributions are an-alyzed using the PMF method to obtain the factor profilesand identify the emission sources. Then a multiple linear re-gression (MLR) of observed PM10, PM1, particle number,and light-absorbing carbon concentrations is carried out todetermine the source contribution to these aerosol variableson an hourly basis and for the whole measurement period.

2 Methodology

2.1 Aerosol measurements

The field campaign was carried out in the town of Lyck-sele (64.55◦N, 18.72◦E, 240 m a.s.l., population 8600). Thereceptor site was placed in Forsdala where RWC is com-

mon and local traffic within the area is limited (the clos-est major road is located 200 m from the site,∼3000 vehi-cles/day). Particles were sized and counted in the diameterrange 25–606 nm with a differential mobility particle sizer(DMPS) composed of a custom-built differential mobilityanalyzer (DMA, Vienna type) and a condensation particlecounter (CPC, TSI 3760, TSI Inc., USA). Particle numberconcentration was calculated from particle number size dis-tribution and is denoted N25−606, where the subindices in-dicate the lower and upper bin limit of particle diameters.Total PM10 mass concentrations were provided by a FilterDynamics Measurement System (FDMS, series 8500 Rup-precht and Patashnick Inc.) whereas total PM1 mass con-centrations were measured with a Tapered Element Oscillat-ing Microbalance (TEOM 1400a, Rupprecht and PatashnickInc., USA). No other correction than the TEOM inbuilt cor-rection (1.3TEOM+3) was applied to the PM1 mass concen-trations. A commercial Aethalometer (series 8100, MageeScientific Inc.) operated with a PM1 sample inlet measuredthe light-absorbing carbon mass concentration MLAC . Thereader is referred to Krecl et al. (2007) and Krecl et al. (2008)for more operational details on these aerosol measurements.Additionally, PM10 mass concentrations were measured witha TEOM 1400ab (Rupprecht and Patashnick Inc., USA) atVindeln station. This is a background monitoring station ofthe Cooperative Program for Monitoring and Evaluation ofthe Long-Range Transmissions of Air Pollutants in Europe(EMEP), situated in a forest∼65 km southeast of Lycksele.All measurements from 31 January to 9 March 2006 were av-eraged on an hourly basis considering a minimum data avail-ability of 75% per hour.

2.2 Positive matrix factorization

PMF is a powerful multivariate least-squares technique thatconstraints the solution to be non-negative and takes intoaccount the uncertainty of the observed data (Paatero andTapper, 1994). This method relies on the time-invarianceof the source profiles and, thus, requires the emission par-ticle size distributions to be stable in the atmosphere be-tween the sources and the receptor site. According to Zhou etal. (2004), after some initial size distribution changes in thevicinity of the sources (due to coagulation and dry deposi-tion), it is reasonable to expect that particle size distributionswill become relatively stable when sampling at some appro-priate distance from the emission sources.

The basic source-receptor model in matrix form is:

X = G.F+E, (1)

whereX is the matrix of observed particle number size dis-tributions,G andF are, respectively, the source contributionsand particle number size distribution profiles of the sourcesthat are unknown and are estimated from the analysis, andE

Atmos. Chem. Phys., 8, 3639–3653, 2008 www.atmos-chem-phys.net/8/3639/2008/

P. Krecl et al.: Wood combustion contribution to winter aerosol in Sweden 3641

is the residual matrix (observed – estimated). Equation 1 canalso be expressed in the element form as:

xij =

p∑k=1

gik·fkj+eij , (2)

wherexij is the particle number concentration of size in-terval j measured on samplei, p is the number of factorscontributing to the samples,fkj is the concentration of sizebin j from thekth factor,gik is the relative contribution offactork to samplei, andeij is the residual value for the sizebin j measured on the samplei. For a givenp, values offkjandgik are adjusted using a least-square method (with theconstraint thatfkj andgik values are non-negative), until aminimumQ value is found:

Q =

n∑i=1

m∑j=1

(eij

σij

)2. (3)

σij is the uncertainty of the particle number concentration ofsize binj in samplei, n is the number of samples, andm isthe number of size intervals.

Following Hedberg et al. (2006), a multiple linear regres-sion model of the g-factors onto the measured concentra-tions of each aerosol variable (i.e. N25−606, MLAC , PM10,and PM1) was performed. The MLR model assumes that theconcentrations of each aerosol variable can be expressed as alinear function of the g-factors and determines the regressioncoefficients and their confidence intervals (95%). The sourcecontributions to each aerosol variable in then estimated basedon the calculated regressions coefficients.

Reff et al. (2007) recommended documenting all of theprocedural details used in the PMF application in order toobtain source apportionment results that are of known qual-ity. The next sections describe the data preparation, selectionof model parameters, and diverse tests on the PMF runs. Asummary of the methodological details chosen for this PMFanalysis is shown in Table 1.

2.2.1 Data preparation

A total of 769 hourly mean particle number size distribu-tions, each with 18 size intervals, were used in this studyafter discarding faulty scans. There are several sources ofmeasurement errors for DMPS particle number size distribu-tions. Errors due to particle counting might arise from theCPC detection efficiency, problems in the CPC optics, andlarge flow rate fluctuations in the CPC. Neither large CPCflow rate variations nor problems associated to the CPC op-tics were observed during the Lycksele campaign. Accord-ing to Wiedensohler et al. (1997), the particle detection effi-ciency for the CPC TSI-3760 operated at 1.5 l min−1 is 90%at 25 nm and rapidly increases for larger particle diameters.Another error source is related to the particle sizing, being

Table 1. Summary of the PMF methodological details used in thisstudy.

PMF parameters Selected option

Number of factors (p) 3PMF mode robustOutliers distance (α) 4Fpeak [−1.6:−1.2] in steps of 0.1Error model σij=Cij + C3 max(

∣∣xij ∣∣ , ∣∣yij ∣∣),with Cij=0 andC3=0.25

Missing data Samples were omittedModel uncertainty 25% samples randomly removed,

300 runs

the fluctuations of flow rate in the DMA the most impor-tant effect in this experiment. The sheath flow rate varia-tion was 1–2% during the campaign, producing a 2–3% er-ror in particle size calculations. Particle losses in the sys-tem could also lead to measurement errors. In this study,we discard losses produced by diffusion and impaction be-cause of the size range covered by the DMPS. Particle resi-dence times in the DMPS system are very short compared tocoagulation time scales and hence losses due to coagulationare negligible. The inversion algorithm included a correctionfor doubly-charged aerosol particles and a triangular-shapetransfer function was implemented. The fraction of triply-charged particles is lower than 8% at 600 nm where usu-ally low particle concentrations are measured. As a result,if triply-charged particles at 600 nm are wrongly assigned toa smaller size bin their effect might be negligible.

2.2.2 Selection of PMF parameters

Number of factors

Different factor numbers were tested and a 3-factor modeladequately fitted the data with the most meaningful results.When 4 factors were included in the analysis, no more rele-vant sources could be identified.

Rotations

As other factor analysis techniques, PMF suffers from rota-tional indeterminacy of the solution as extensively discussedby Paatero et al. (2002).F peak is the model parameter thatcontrols the rotation in PMF by adding and/or subtracting therows and columns of F and G matrices from each other de-pending on the sign of theF peakvalue. Diverse methods havebeen proposed to adjustF peakto obtain the most meaningfulresults (Paatero et al., 2002; Paatero et al., 2005). Usually,PMF is run several times with differentF peak values to de-termine the range within which theQ value remains stable(Paatero et al., 2002). Figure 1a shows theQ values ob-tained when a 3-factor PMF model was run forF peakvalues

www.atmos-chem-phys.net/8/3639/2008/ Atmos. Chem. Phys., 8, 3639–3653, 2008


−2 −1.5 −1 −0.5 0 0.5 1 1.5 23500

4000

4500

5000

Fpeak [−]

Q r

obus

t [−

]

(a)

−2 −1.5 −1 −0.5 0 0.5 1 1.5 20

0.2

0.4

0.6

0.8

1

Fpeak [−]

R2

[−]

(b)g1−g2

g3−g1

g3−g2

−2 −1.5 −1 −0.5 0 0.5 1 1.5 23500

4000

4500

5000

Fpeak [−]

Q r

obus

t [−

]

(a)

−2 −1.5 −1 −0.5 0 0.5 1 1.5 20

0.2

0.4

0.6

0.8

1

Fpeak [−]

R2

[−]

(b)g1−g2

g3−g1

g3−g2

Fig. 1. (a) Fpeak values versusQ robust values for PMF. (b) Co-efficient of determination (R2) among g-factors (g1, g2, g3) versusFpeakvalues. PMF was run using three factors,C3=0.25, andα=4.[–] denotes the variable is dimensionless.

between−2 and +2 in steps of 0.1. Based on Paatero etal. (2005), the coefficients of determinationR2 among thethree g-factors are plotted as a function ofF peak in Fig. 1b.The PMF solutions forF peak ≥ 0 resulted in negative re-gression coefficients when the factor contributions were re-gressed onto PM1 mass concentrations and, thus, these so-lutions are considered physically invalid. It can be observedthat the statistical independence between pairs of g-factorsincreases for decreasingF peakvalues (allR2 were lower than0.1 for F peak ≤ −1.2). Thus, the selection ofF peak is re-

stricted to values smaller than−1.1. Another approach toreduce the rotational ambiguity is to use some a priori in-formation that helps constraining the solution. In our case,we employ PM10 mass concentration measured at a back-ground site (Vindeln). Figure 2 shows the temporal series ofthe modeled PM10 contribution per factor forF peak= − 0.1(maximum valid value) andF peak= − 1.4 (panels a, b, andc) together with observed PM10mass concentrations at Vin-deln (panel c). As will be discussed in Sect. 3.1, factor 1can be interpreted mostly as the contribution from local traf-fic, factor 2 as local RWC, whereas factor 3 is a combinationof two sources: local RWC and long-range transport (LRT).The largest difference in PM10 mass concentration when run-ning PMF forF peak= − 0.1 and−1.4 is found for factors 2and 3. When PMF is run forF peak= − 1.4, a larger con-tribution of factor 2 (mostly local RWC) to PM10 is foundwhereas the local RWC contribution to factor 3 is reducedand the correlation with Vindeln background measurementsincreases. ForF peak< − 1.6, the contribution of factor 3 toPM10 is mostly below the observed background contribution.As a result of these tests, we selected a range ofF peak val-ues between−1.6 and−1.2. For the sake of completeness,the time series of modeled aerosol variables MLAC , N25−606,and PM1 together with the observed data are presented in theappendix (Figs. A1, A2, A3).

Error model

A dynamical error model was chosen for this study and PMFuncertainties are calculated at each iteration step of the pro-gram by using the formula shown in Table 1. In this expres-sion, the uncertaintyσ ij is derived from the measurement er-ror Cij , the constantC3 and the maximum value betweenthe observedxij and the modeledyij . C3 is included to ac-count for some source profile variation and, in this way, pro-vides the fitting more flexibility to accommodate this vari-ability. As previously shown, the DMPS measurement errorwas quite small and, hence, we decided to setCij to 0 andinclude all the uncertainty input in theC3 constant. In orderto obtain small scaled residuals,C3 was set up to 0.25.

Robust mode and outliers

PMF was run in the robust mode to reduce the influence ofatypical measurements in the dataset. In this mode, the un-certainties of measurements for which the scaled residualsare larger than the outlier threshold distanceα are increasedto diminish their influence on the PMF solution. As sug-gested by Paatero (2000), standardα values of 2, 4 (defaultvalue), and 8 were tested in this study. The test was carriedout with a 3-factor model,F peak= − 1.4, andC3=0.25. Nodifference between the solutions usingα=4 andα=8 wasobserved since the scaled residuals for both runs lay in therange [−2, 3.4]. Small differences in f and g-factors werefound when PMF was run withα=2 compared to the default



0

10

20

30Traffic(a)

Fpeak=−0.1Fpeak=−1.4

0

25

50 RWC(b)

0

25

50

P

M10

[μg

m−

3 ]

RWC/LRT(c) Vindeln meas.

31−Jan 04−Feb 08−Feb 12−Feb 16−Feb 20−Feb 24−Feb 28−Feb 04−Mar 08−Mar0

35

70

(d) Lycksele meas.

Total mod.

Fig. 2. Time series of modeled PM10 contribution for each factor forFpeak=−0.1 and−1.4 (a), (b), (c), together with total modeled andmeasured PM10 concentrations(d). PM10 mass concentration at Vindeln (rural background site) is also shown in panel c.

α value. Emulated aerosol concentrations (N25−606,MLAC ,PM10, and PM1) were compared when running PMF withα=2 andα=4. The largest mean difference between aerosolconcentrations (25%) was observed for the contribution offactor 2 to PM1 mass concentrations. This indicates, oncemore, that PM1 is the most sensitive aerosol variable in re-lation to PMF initialization parameters in this study. As aresult, PMF was run withα=4 in this work.

2.2.3 Tests on PMF runs

Global minimum

Least-squares can yield multiple solutions depending on theinitial starting point for each entry in theF andG matrices.PMF was run 5 times with different seed values to ensurethat a global minimum has been reached. The setup of theinitialization PMF parameters was:F peak= − 1.4, C3=0.25,3 factors andα=4. The same output values (f andg-factors,andQ) were obtained for all the runs. Hence, we concludethat the PMF solution consistently converges and a globalminimum was found for this particular setup of model pa-rameters.

Goodness of model fit

Two methods were employed to assess the adequacy of thePMF fit to the measurements. First, the distribution ofscaled residuals was examined when using a 3-factor model(F peak=−1.4,C3=0.25, andα=4). The residual concentra-tions are normally distributed and no structured features wereidentified. Second, the modeled aerosol concentrations werecompared to the measurements. Time series of observedand total modeled concentrations of PM10, PM1, MLAC , andN25−606 are displayed in Figs. 2d, A1d, A2d and A3d, re-spectively. Figure 3 shows scatter plots and least-squares lin-ear regressions between modeled and observed PM10, PM1,MLAC and N25−606 concentrations. The 95% confidence in-tervals for the slope and intercept were also calculated andincluded in the linear regression equation. As expected, thehighestR2 and slope close to 1 was obtained for N25−606since PMF was run on particle number size distributions. Thevariability of the measured MLAC , is very well predicted bythe model (R2=0.85) whereas for PM10, and PM1 the coef-ficients of determination are 0.75 and 0.72, respectively. Theintercept seen in Fig. 3b (PM1 linear regression) is signifi-cantly different from zero at 95% confidence interval. Thismight suggest the inbuilt correction for the loss of volatilematerial applied by the TEOM instrument is not adequatefor the measurements carried out in this Lycksele campaign.



0 20 40 600

20

40

60

PM10

, FDMS [μg m−3]

PM

10, P

MF

−M

LR [μ

g m

−3 ]

y=0.86x(±0.04) + 0.6(±0.6)R2=0.75n=741

1:1

(a)

0 10 20 30 400

10

20

30

40

PM1, TEOM [μg m−3]

PM

1, P

MF

−M

LR [μ

g m

−3 ]

y=0.97x(±0.04) −1.4(±0.5)R2=0.72n=736

1:1

(b)

0 2 4 6 80

2

4

6

8

MLAC

, Aeth. [μg m−3]

MLA

C, P

MF

−M

LR [μ

g m

−3 ]

y=0.9x(±0.03) + 0.06(±0.05)R2=0.85n=748

1:1

(c)

0 1 2 3

x 104

0

0.5

1

1.5

2

2.5

3

3.5x 10

4

N25−606

, DMPS [cm−3]

N25

−60

6, P

MF

−M

LR [c

m−

3 ]

y=1x(±0.004) −22(±28)R2=1n=769

1:1

(d)

Fig. 3. Scattergram of predicted (PMF-MLR) vs. measured concentrations of: (a) PM10, (b) PM1, (c) MLAC , and (d) N25−606. The 95%confidence intervals for the slope and intercept are included in the linear regression equation (in parenthesis). The solid line represents theleast squares line regression and the dotted line indicates the identity line (1:1). The number of samplesn and coefficient of determinationR2 are also shown.

Hedberg et al. (2006) reported a similar problem when per-forming PMF on PM2.5 mass concentrations measured witha TEOM series 1400 in a previous winter campaign in Lyck-sele.

Model uncertainties

Following Hedberg et al. (2005), in order to estimate themodel uncertainties 25% of the original samples were ran-domly removed and then PMF was run 300 times on thesenew datasets (always with 3 factors,C3=0.25, α=4, andF peak= − 1.4). Figure 4 presents the mean and standarddeviation of the factor profiles. The difference between themean factor profiles after removing 25% of the samples, andthe factor profiles when all samples (using the same modelinitialization values) are included lies between one standarddeviation values. As a result, the PMF solution is consideredstable.

3 Results and discussion

3.1 Source identification

The sources were identified based on the particle number sizedistribution profiles of the PMF factors (Fig. 5), the diur-nal contributions patterns estimated by PMF for both week-ends (WE) and weekdays (WD) (Fig. 6), and correlation ofthe modeled N values with measured aerosol concentrations(PM10, PM1, and MLAC). Time series of particle numbercontributions for each factor are presented in the appendix(Fig. A3). Figure 5 displays the calculated factor profileswhen the PMF model was run withF peak values rangingfrom −1.6 to −1.2 in steps of 0.1. The left panels showmean modeled particle number size distributions per factorwhereas normalized f-factor profiles (mean± standard de-viation) to the total number of particles per factor are pre-sented in Fig. 5d. This right panel highlights the contributionof each size bin to the total number particle concentration perfactor.



101

102

103

0

1000

2000

3000

4000

5000

Dp [nm]

dN/d

LogD

p [c

m−

3 ]

Traffic

RWC

RWC/LRT

Fig. 4. Mean f-factors and 1STD error bars as a result of randomlyremoving 25% of the original samples. PMF was run 300 timesusing three factors,C3=0.25,α=4, andFpeak= − 1.4.

Factor 1 has a peak at particle diameterDp ∼28 nm(Fig. 5a) and shows a very well defined daily pattern dur-ing weekdays and weekends (Fig. 6a). The strong diurnalvariation might suggest that these particles are produced inthe immediate vicinity of the receptor site. The origin of thisfactor is likely to be local traffic emissions. The shape ofthis modeled profile is similar to the shape of particle num-ber size distributions measured at a street canyon (Gidhagenet al., 2004) and road tunnel (Kristensson et al., 2004) sitesin Stockholm where traffic is mainly dominated by gasolinevehicles. The peak number concentration in Stockholm wasobserved atDp ∼20 nm for both studies, which could not beobserved in our case since DMPS measurements started at25 nm. In Sweden, only 5% of passenger cars are diesel ve-hicles and heavy-duty vehicles comprise 5% of the total vehi-cle fleet (SIKA, 2006). The weak correlation (R = 0.37) wefound between modeled N25−606 for factor 1 and measuredMLAC could be explained by the dominant gasoline vehiclesemissions since gasoline vehicles produce lower MLAC thandiesel engines (Burtscher, 2000; Gillies and Gertler, 2000).

Factor 2 is strongly associated with the light absorbing car-bon content of fine aerosols as shown by the high correlation(R=0.76) between modeled N25−606 (attributed to factor 2)and the observed MLAC . Kim et al. (2004) also found a sim-ilar correlation between modeled particle size distributionsattributed to RWC and light absorption coefficients in Seat-tle (USA) during wintertime. In Lycksele, particle numberconcentrations are significantly higher on weekends than onweekdays after 13:00 LT (unpaired t-test, 95% confidence in-terval), reaching mean concentrations of∼1×104 cm−3 from21:00 LT to midnight (Fig. 6b). Several studies (e.g., Hueglinet al., 1997; Hedberg et al., 2002; Johansson et al., 2004;Boman, 2005) have shown that particle size distributionsfrom wood combustion under controlled laboratory condi-

0

2000

4000 Traffic(a)

0

2000

4000

dN/d

LogD

p [c

m−

3 ]

RWC(b)

101

102

103

0

500

1000RWC/LRT(c)

Dp [nm]

101

102

103

0

0.05

0.1

0.15

0.2

Dp [nm]

[dN

/dLo

gDp]

/Nt [

−]

(d)

0

2000

4000 Traffic(a)

0

2000

4000

dN/d

LogD

p [c

m−

3 ]

RWC(b)

101

102

103

0

500

1000RWC/LRT(c)

Dp [nm]

101

102

103

0

0.05

0.1

0.15

0.2

Dp [nm]

[dN

/dLo

gDp]

/Nt [

−]

(d)

Fig. 5. Left panels: Absolute mean f-factors expressed in [cm−3].Right panel: Normalized f-factor profiles (mean± standard devia-tion) to the total number of particles per factor. PMF was run withFpeak values from –1.6 to –1.2,C3=0.25, andα=4. [–] denotesthe variable is dimensionless.

tions vary in shape, peak concentration value and mode di-ameter depending on a number of factors such as the com-bustion phase (i.e. ignition, intermediate, and smoldering),appliance type (e.g. wood stove, boiler, fireplace), type andamount of wood, and wood moisture content. Despite of thisbroad variation, a consistent conclusion is that RWC emitsparticles mainly in the size range 60–300 nm. As shown inFig. 5b, factor 2 peaks at∼70 nm and its shape is similarto the shape of particle number size distributions measuredin winter field campaigns (Kristensson, 2005; Hering et al.,2007) when wood burning was an important particle emis-sion source. Local RWC emissions are suggested to be thesource of this factor.



0

2000

4000

6000(a)

0

4000

8000

12000

N25

−60

6 [c

m−

3 ]

(b)

weekendweekday

0 6 12 18 230

500

1000

1500

2000 (c)

Time of day (start)

Fig. 6. Comparison of model resolved hourly contributions toN25−606 between weekends and weekdays.(a) Traffic. (b) RWC.(c) RWC/LRT. PMF 3-factor model was run withFpeak= − 1.4,C3=0.25, andα=4.

Factor 3 particle number size distribution is depicted inFig. 5c and peaks at∼160 nm. We suggest this factor couldbe a combination of two sources: local wood combustion andlong-range transport of particles. Tunved et al. (2003) foundthat particle number size distributions are typically bimodal(mean mode diameters at∼56 nm and 160–190 nm) duringwinter in five Nordic background stations (covering latitudes58◦N–68◦N). They observed a gradient in the mean daily in-tegral number concentration from∼2000 cm−3 at the south-ernmost station to values


0

4000

8000

dN/d

LogD

p [c

m−

3 ] (a)

0

25

50

75

100

Cum

. num

ber

cont

rib. [

%]

(b)

Traffic

RWC

RWC/LRT

Measured

101

102

103

0

25

50

75

100

Cum

. vol

. con

trib

. [%

]

(c)

Dp [nm]

Fig. 7. Cumulative contribution per factor as a function of particlediameter and mean measured particle size distribution.(a) Absoluteparticle number concentrations.(b) Relative number concentration.(c) Relative volume concentration considering spherical particles.PMF 3-factor model was run withFpeak= − 1.4, C3=0.25, andα=4.

3). For example, the contribution of factor 1 for the small-est particles is quite large (77%) and decreases to 50% forDp=100 nm. Considering all particle diameters now, factor1 has the largest contribution at the smallest sizes and thendecreases maintaining a nearly constant particle emission of44% for Dp>190 nm. Factor 2 provides 20% of the parti-cles at the smallest diameter and its contribution increases to∼45% and remains constant for larger particles. Factor 3 hasa low overall contribution to particle number concentrationsthat slightly increases when increasing the particle diameter.The same interpretation applies to Fig. 7c but now consider-ing volume fractions instead of number fractions. Factor 2has the largest factor contribution to volume (mass) concen-tration for 80


0

2

4

6

8

10P

M10

[μg

m−

3 ](a)

WD WE Total0

2

4

6

8

10

PM

1 [μ

g m

−3 ]

(b)

WD WE Total

Traffic

RWC

RWC/LRT

0

0.5

1

1.5

MLA

C [μ

g m

−3 ]

(c)

WD WE Total0

1

2

3

4

5

N25

−60

6 [1

03 c

m−

3 ]

(d)

WD WE Total

Fig. 8. Summary of the mean contribution of the three factors to PM10, PM1, MLAC , and N25−606 for weekdays (WD), weekends (WE) andthe whole campaign (Total), in the period 31 January – 9 March 2006. PMF was run withFpeak= − 1.4, C3=0.25, andα=4.

contribution to particle number concentration was calculatedas the average of measurements at two background stations(Hyytiälä and Pallas) for the winter 2002. Hedberg etal. (2006) apportioned 70% of PM2.5 mass concentration tolocal RWC when performing a PMF analysis on inorganiccompounds and using mainly the abundance of K and Zn toidentify this combustion source. This fraction apportionedby Hedberg et al. (2006) lies within the local RWC contri-bution intervals we estimated for PM10 and PM1 mass con-centrations and might suggest the local RWC contributionis closer to the possible maximum fraction (very low back-ground contribution to factor 3) than to the minimum possi-ble value (factor 3 is all LRT). Finally, our apportionment ofMLAC can be roughly compared to the radiocarbon analysisresults of total carbon reported by Sheesley et al. (20081)when sampling in Lycksele from 23 January to 8 March

1 Sheesley, R. J., Kruså, M., Krecl, P., Johansson, C., andGustaffson,̈O.: Source apportionment of elevated wintertime PAHsin a northern Swedish town by compound specific radiocarbon anal-ysis, Environ. Sci. Technol., submitted, 2008.

2006. Total carbon is defined as the carbon remaining af-ter removal of the inorganic carbonates by acid treatmentand, thus, includes elemental and organic carbon. If we as-sume that the fossil total carbon fraction absorbs light in thesame amount as the modern total carbon fraction does, thenthe apportionment found with radiocarbon analysis might ap-ply to the MLAC data. To correctly interpret and comparethese results, the reader has to bear in mind that14C anal-ysis provides the apportionment of modern and fossil car-bonaceous aerosol but does not provide information on thelocation of the emission sources (local or LRT). The aver-age contribution of fossil total carbon (attributed to trafficemissions) was 24% which coincides with the mean frac-tion of MLAC we attributed to local traffic emissions. Theother 76% was mostly attributed to wood combustion sincebiogenic emissions and combustion of grass fires and incin-eration of household vegetable waste were not observed inthe area. Another study in the Nordic region (Glasius et al.,2006) reported two main sources of ambient aerosol whenperforming COPREM model calculations on PM2.5 massconcentrations measured in a residential area in Denmark



0 6 12 18 230

10

20

30P

M10

[μg

m−

3 ]

WEEKDAYS

(a)

0 6 12 18 230

10

20

30

WEEKENDS

(b)

0 6 12 18 230

10

20

30

PM

1 [μ

g m

−3 ]

(c) Traffic

RWC

RWC/LRT

Measured

0 6 12 18 230

10

20

30 (d)

0 6 12 18 230

1

2

3

4

5

MLA

C [μ

g m

−3 ] (e)

0 6 12 18 230

1

2

3

4

5

(f)

0 6 12 18 230

5000

10000

15000

N25

−60

6 [c

m−

3 ]

Time of day (start)

(g)

0 6 12 18 230

5000

10000

15000

Time of day (start)

(h)

Fig. 9. Model resolved mean diurnal contributions per factor to PM10, PM1, MLAC and N25−606 for weekdays (left panels) and weekends(right panels). The daily mean measured concentrations are also shown with black lines. PMF 3-factor model was run withFpeak= −1.4,C3 = 0.25, andα = 4.

in winter 2003/2004. The largest contribution to PM2.5 wasassigned to long-range transport (mean 10.65µg m−3) withadditional and episodically contributions from RWC (mean4.60µg m−3). The regional traffic was found to be of minorimportance accounting only for 0.83µg m−3 of PM2.5 con-centrations. As discussed by Tunved et al. (2003), there is agradient in background aerosol concentrations in the Nordicregion with highest concentrations in the South and decreas-

ing towards the North. Thus, it is expected to have a largerabsolute contribution of LRT to mass concentrations in Den-mark compared to Northern Sweden. Local RWC influenceon aerosol concentrations might be lower in Denmark sincewinters are milder than in Northern Sweden. This is associ-ated to lower heat demand (lower emissions) and more unsta-ble lower atmosphere (favors the vertical dispersion of pollu-tants).



Figure 9 displays the diurnal contribution of the modeledfactors to the PM10, PM1, MLAC , and N25−606concentrationsfor weekdays and weekends. Krecl et al. (2008) showed thatmeasured mean aerosol concentrations were statistically sig-nificantly higher on WE than on WD after 12:00 LT at 95%confidence level when analyzing the same data set employedin this study. For all aerosol variables, the contribution oflocal traffic (factor 1) on weekdays shows a peak concentra-tion at 09:00–10:00 LT and a second and smaller maximum at20:00–21:00 LT. During weekdays, local traffic emissions in-crease the aerosol concentrations during the morning reach-ing a peak value∼11:00–12:00 LT while in the evening itseffect is smaller and more diffuse. As shown before, themean contribution of local traffic to the aerosol concentrationis similar on weekdays and weekends. Then this large differ-ence between weekends and weekdays for all aerosol vari-ables is attributed by PMF to local RWC as shown in Fig. 9.On weekends, the contribution of local RWC to atmosphericaerosol is largest between 18:00 and midnight even whenconsidering the minimum contribution (only factor 2). Thisresult is in agreement with the findings by Kim et al. (2004)when applying PMF analysis on particle volume distributionsmeasured in Seattle (USA) during the winter season. Theyidentified a local RWC particle emission source based on thedistinct diurnal pattern of modeled volume concentrations onweekends compared to weekdays. In Seattle, daytime contri-butions of this factor during weekends were lower than thoseof weekdays and the highest concentrations were observedon weekends between 21:00 LT and midnight.

4 Conclusions

This work demonstrates that is possible to estimate the emis-sion sources of atmospheric aerosols applying PMF analy-sis on particle size distributions in a wood smoke-impacted

residential area. Hence, this methodology could be appliedto other urban sites where RWC is a substantial source ofaerosol particles and particle size distribution emissions ofvehicle exhaust and RWC present characteristic modes andshapes that can be properly separated and identified by PMF.The high-temporal resolution of the source apportionment al-lows studying in detail the diurnal variation of source con-tributions to ambient aerosol and also provides a better es-timation of the time periods when the inhabitants are moreexposed to harmful aerosol concentrations. Although thePMF factors were attributed to certain emission sources, theymight still be influenced by other unknown sources or amongthemselves. This PMF source-receptor modeling should becomplemented with chemical speciation analysis to provide amore precise source apportionment in relation to local RWCdue to the overlapping of sources profiles between RWC andLRT at this receptor site. In the Nordic region, where LRTcan have a large influence on particle concentrations, DMPSmeasurements at rural background sites close to the recep-tor site of interest should be conducted in future field cam-paigns. The results obtained in Lycksele could be gener-alized to other cities/towns in Northern Sweden with simi-lar topography and meteorological conditions during winter.However, to obtain a better estimation of the contribution ofthe different emission sources to atmospheric aerosols at an-other location, a source apportionment study should be car-ried out in the area of interest.

Appendix

The time series of the modeled aerosol variables for each fac-tor together with the observed data are presented in Figs. A1,A2, and A3.



0

10

20

Traffic(a)

0

20

40 RWC(b)

0

10

20

30

P

M1

[μg

m−

3 ]

RWC/LRT(c)


25

50

(d) Measured

Total mod.

Fig. A1. Time series of modeled PM1 contribution for each factor(a), (b), (c), together with total modeled and measured PM1 concentrations(d). PMF was run withFpeak= − 1.4, C3=0.25, andα=4.

0

2

4

6Traffic(a)

0

4

8RWC(b)

0

2

4

MLA

C [μ

g m

−3 ]

RWC/LRT(c)


5

10

(d) Measured

Total mod.

Fig. A2. Times series of modeled MLAC contribution for each factor(a), (b), (c), together with total modeled and measured MLAC concen-trations(d). PMF was run withFpeak= − 1.4, C3=0.25, andα=4.



0

20000

40000Traffic(a)

0

20000

40000 RWC(b)

0

2000

4000

6000

N

25−

606

[cm

−3 ]

RWC/LRT(c)


20000

40000

(d Measured

Total mod.

Fig. A3. Time series of modeled N25−606 contribution for each factor(a), (b), (c), together with total modeled and measured N25−606concentrations(d). PMF was run withFpeak= − 1.4, C3=0.25, andα=4.

Acknowledgements.Financial support from the Swedish EnergyAgency and the Swedish Environmental Protection Agency is ac-knowledged. We thank T. Hennig for calculating the error in parti-cle sizing due to flow rate fluctuations in the DMA and for usefuldiscussions on DMPS measurement errors.

Edited by: K. Lehtinen

References

Boman, C.: Particulate and gaseous emissions from residentialbiomass combustion, Ph.D. Thesis, Energy Technology andThermal Process Chemistry, Umeå University, Ume̊a, Sweden,available at:http://www.diva-portal.org/umu/theses/index.xsql?lang=en, 2005, last access: 25 November 2007.

Burtscher, H.: Comparison of particle emissions from differentcombustion systems, J. Aerosol Sci., 31, S620–S621, 2000.

Gidhagen, L., Johansson, C., Langner, J., and Olivares, G.: Simu-lation of NOx and ultrafine particles in a street canyon in Stock-holm, Sweden, Atmos. Environ., 38, 2029–2044, 2004.

Gillies, J. A. and Gertler, A. W.: Comparison and evaluation ofchemically speciated mobile source PM2.5 particulate matterprofiles, J. Air Waste Ma., 1459–1480, 2000.

Glasius, M., Ketzel, M., W̊ahlin, P., Jensen, B., Mønster, J.,Berkowicz, R., and Palmgren, F.: Impact of wood combustionon particle levels in a residential area in Denmark, Atmos. Envi-ron., 40, 7115–7124, 2006.

Gorin, C. A., Collet, J. L., and Herckes, P.: Wood smoke contri-bution to winter aerosol in Fresno, CA, J. Air Waste Ma., 56,1584–1590, 2006.

Hedberg, E., Kristensson, A., Ohlsson, M., Johansson, C., Johans-son, P.Å., Swietlicki, E., Vesely, E., Wideqvist, U., and Wester-holm, R.: Chemical and physical characterization of emissionsfrom birch wood combustion in a wood stove, Atmos. Environ.,36, 4823–4837, 2002.

Hedberg, E., Gidhagen, L., and Johansson, C.: Source contributionsto PM10 and arsenic concentrations in Central Chile using posi-tive matrix factorization, Atmos. Environ., 39, 549–561, 2005.

Hedberg, E., Johansson, C., Johansson, L., Swietlicki, E., andBrorstr̈om-Lund́en, E.: Is levoglucosan a suitable quantitativetracer for wood burning?: comparison with receptor modelingon trace elements in Lycksele, Sweden, J. Air Waste Ma., 56,1669–1678, 2006.

Hering, S. V., Kreisberg, N. M., Stolzenburg, M. R., and Lewis, G.S.: Comparison of particle size distributions at urban and agricul-tural sites in California’s San Joaquin Valley, Aerosol Sci. Tech.,41, 86–96, 2007.

Hueglin, C., Gaegauf, C., K̈unzel, S., and Burtscher, H.: Character-ization of wood combustion particles: morphology, mobility, andphotoelectric activity, Environ. Sci. Technol., 31, 3439–3447,1997.

Johansson, L. S., Leckner, B., Gustavsson, L., Cooper, D., Tullin,C., and Potter, A.: Emission characteristics of modern and old-type residential boilers fired with wood logs and wood pellets,Atmos. Environ., 38, 4183–4195, 2004.

Keywood, M. D., Ayers, G. P., Gras, J. L., Gillet, R. W., and Co-hen, D.: Size distribution and sources of aerosol in Launceston,Australia, during winter 1997, J. Air Waste Ma., 50, 418–427,2000.

Khalil, M. A. K. and Rasmussen R. A.: Tracers of wood smoke,Atmos. Environ., 37, 1211–1222, 2003.

Kim, E., Hopke, P. K., Larson, T. V., and Covert, D. S.: Analysis


http://www.diva-portal.org/umu/theses/index.xsql?lang=enhttp://www.diva-portal.org/umu/theses/index.xsql?lang=en


of ambient particle size distributions using Unmix and positivematrix factorization, Environ. Sci. Technol., 38, 202–209, 2004.

Kocbach, A., Johansen, B. V., Schwarze, P. E., and Namork, E.:Analytical electron microscopy of combustion particles: a com-parison of vehicle exhaust and residential wood smoke, Sci. TotalEnviron., 346, 231–243, 2005.

Krecl, P., Str̈om, J., and Johansson, C.: Carbon content of atmo-spheric aerosols in a residential area during the wood combustionseason in Sweden, Atmos. Environ., 41, 6974–6985, 2007.

Krecl, P., Str̈om, J., and Johansson, C.: Diurnal variation of atmo-spheric aerosol during the wood combustion season in NorthernSweden, Atmos. Environ., 42, 4113–4125, 2008.

Kristensson, A., Johansson, C., Westerholm, R., Swietlicki, E., Gid-hagen, L., Wideqvist, U., and Vesely, V.: Real-world traffic emis-sion factors of gases and particles measured in a road tunnel inStockholm, Sweden, Atmos. Environ., 38, 657–673, 2004.

Kristensson, A.: Aerosol particle sources affecting the Swedishair quality at urban and rural level, Ph.D. thesis, LundInstitute of Technology, Lund University, Sweden, avail-able at:http://theses.lub.lu.se/postgrad//search.tkl?field\query1=pubid\&query1=tec936\&recordformat=display, last access: 5November 2007, 2005.

Paatero, P. and Tapper, U.: Positive matrix factorization: a non-negative factor model with optimal utilization of error estimatesof data values, Environmetrics, 5, 111–126, 1994.

Paatero, P.: User’s guide for positive matrix factorization programsPMF2 and PMF3, 2: reference, 2000.

Paatero, P., Hopke, P. K., Song, X.-H., and Ramadan, Z.: Un-derstanding and controlling rotations in factor analytic models,Chemometr. Intell. Lab., 60, 253–264, 2002.

Paatero, P., Hopke, P. K., Begum, B. A., and Biswas, S. K.: Agraphical diagnostic method for assessing the rotation in factoranalytical models of atmospheric pollution, Atmos. Environ., 39,193–201, 2005.

Reddy, C. M., Pearson, A., Xu, L., McNichol, A. P., Benner, B. A.,Wise, S. A., Klouda, G. A., Currie, L. A., and Eglinton, T. I.:Radiocarbon as a tool to apportion the sources of polycyclic aro-matic hydrocarbons and black carbon in environmental samples,Environ. Sci. Technol., 36, 1774–1782, 2002.

Reff, A., Eberly, S. I., and Bhave, P. V.: Receptor modeling of am-bient particulate matter data using positive matrix factorization:review of existing methods, J. Air Waste Ma., 57, 146–154, 2007.

SIKA: Vehicles in counties and municipalities at the turn ofthe year 2005/2006,http://www.sika-institute.se/Templates/FileInfo.aspx?filepath=/Doclib/Import/101/ss20064.pdf, lastaccess: 17 March 2008, 2006.

Statistics Sweden: Energy statistics for one- to two-dwelling build-ings in 2005, Statistiska Meddelanden EN16SM0601,http://www.scb.se/templates/Publikation 178027.asp, last access: 8November 2007, 2005.

Tunved, P., Hansson, H. C., Kulmala, M., Aalto, P., Viisanen, Y.,Karlsson, H., Kristensson, A., Swietlicki, E., Dal Maso, M.,Ström, J., and Komppula, M.: One year boundary layer aerosolsize distribution data from five Nordic background stations, At-mos. Chem. Phys., 3, 2183–2205, 2003,http://www.atmos-chem-phys.net/3/2183/2003/.

Wang, H. and Shooter, D.: Coarse-fine and day-night differ-ences of water-soluble ions in atmospheric aerosols collected inChristchurch and Auckland, New Zealand, Atmos. Environ., 36,3519–3529, 2002.

Wiedensohler, A., Orsini, D., Covert, D. S., Coffmann, D., Cantrell,W., Havlicek, M., Brechtel, F. J., Russell, L. M., Weber, R. J.,Gras, J., Hudson, J. G., and Litchy, M.: Intercomparison study ofthe size-dependent counting efficiency of 26 condensation parti-cle counters, Aerosol Sci. Tech., 27, 224–242, 1997.

Zhou, L., Kim, E., Hopke, P. K., Stainer, C. O., and Pandis, S.:Advanced factor analysis on Pittsburgh particle size-distributiondata, Aerosol Sci. Tech., 38(S1), 118–132, 2004.

http://theses.lub.lu.se/postgrad//search.tkl?field query1=pubid&query1=tec_936&recordformat=displayhttp://theses.lub.lu.se/postgrad//search.tkl?field query1=pubid&query1=tec_936&recordformat=displayhttp://www.sika-institute.se/Templates/FileInfo.aspx?filepath=/Doclib/Import/101/ss_2006_4.pdfhttp://www.sika-institute.se/Templates/FileInfo.aspx?filepath=/Doclib/Import/101/ss_2006_4.pdfhttp://www.scb.se/templates/Publikation____178027.asphttp://www.scb.se/templates/Publikation____178027.asphttp://www.atmos-chem-phys.net/3/2183/2003/

contribution of residential wood combustion and other …...p. krecl et al.: wood combustion...

Documents