Comparison and Evaluation of Methods to Apportion Ambient PM 2 . 5 to Residential Wood Heating in Fairbanks , AK

Biomass burning for residential heating significantly contributes to ambient PM2.5 burdens in many areas, making source apportionment to wood heater emissions an important issue. This study compares and evaluates Chemical Mass Balance (CMB), levoglucosan analysis, and C analysis methods for apportionment. Results suggest that the CMB method appears to overestimate the contribution of residential wood heating in Fairbanks, perhaps due to non-representativeness of emissions source profiles. Carbon-14 analysis allows for apportionment to biomass sources, but must be corrected for noncarbon PM2.5 content. Levoglucosan analysis has the advantage of being relatively inexpensive, but there is considerable uncertainty in determining conversion factors to calculate wood smoke levels from measured levoglucosan concentrations. Conversion factors in the range of 9.1 to 13.3 are calculated from previously published and experimental mass fractions of levoglucosan in wood smoke PM2.5. Conversion factors in the range of 10.7 to 12.9 are determined from analysis of independent field measurements of C and levoglucosan in Fairbanks. The calculated and measured conversion factors are consistent and are similar to previously-reported values. The three apportionment methods (focused on residential wood smoke contributions) are complementary and collectively provide a means to evaluate or confirm apportionment results.


INTRODUCTION
Globally, biomass burning for residential heating plays an important role in both indoor and outdoor particulate matter (PM) exposures and associated adverse health effects.In mountainous regions of the northwestern United States, fine particulate matter less than 2.5 µm in diameter (PM 2.5 ) can be a major air pollutant of concern, with the majority of ambient PM 2.5 during the winter often resulting from residential wood combustion (Conner and Stevens, 1991;Rogge et al., 1998;Ward and Smith, 2005;Ward et al., 2006b).Reliable means to apportion PM 2.5 to various sources, including residential wood heating, is an essential first step in efforts to implement PM 2.5 control measures.
Exposure to elevated levels of PM 2.5 from wood combustion is known to cause adverse effects on human health (Smith et al., 2000;Naeher et al., 2007).These include chronic obstructive pulmonary diseases, asthma, lung cancer, tuberculosis, negative birth outcomes (e.g., low birth weight, stillbirth), eye disease (Ezzati and Kammen Daniel, 2002), and an elevated risk of lower respiratory tract infections (LRTI) among children (Kossove, 1982;Pandey et al., 1989;Collings et al., 1990;Armstrong and Campbell, 1991;Mishra, 2003).In vitro human immune cell exposures to wood smoke have been shown to induce significant increases in the pro-inflammatory response, with effects dependent on the stove type and combustion conditions (Kocbach et al., 2008;Boelling et al., 2012).
The United States Environmental Protection Agency (USEPA) has set National Ambient Air Quality Standards (NAAQS) to protect public health and welfare.NAAQS are reviewed every five years and are periodically revised to take into consideration advancements in the knowledge of adverse effects.Current Primary NAAQS for PM 2.5 include an annual standard of 12 µg m -3 (annual mean averaged over three years) and a 24-hour standard of 35 µg m -3 (annual 98 th percentile averaged over three years) (USEPA, 2012).Populated areas that exceed NAAQS are designated as nonattainment, requiring local and state environmental agencies to identify major PM 2.5 sources and then reducing emissions from these sources.
Several methods for quantitative and reliable source apportionment of PM 2.5 to residential wood stove use have been developed and implemented in recent years -each with significant advantages and limitations.Over the past several decades the Chemical Mass Balance (CMB) model has been commonly employed for comprehensive source apportionment (Vega et al., 1997;Paode et al., 1999;Chen et al., 2001;Ward, 2001;Li et al., 2003;Ward and Smith, 2005;Olson et al., 2008;Yin et al., 2010;Ward and Lange, 2010;Larson et al., 2011;Herich et al., 2014).CMB requires comprehensive chemical analysis of at least as many constituents of ambient PM 2.5 as there are potential sources, as well as reliable source profiles including the same constituents in PM 2.5 from all potential sources.CMB analysis identifies a linear combination of source profiles that best matches the composition of the ambient PM 2.5 .CMB assumes that the source profiles are representative of the sources impacting the measurement site and are constant throughout the sampling period.It is also assumed that the source profiles are independent and the chemical constituents do not react with one another.CMB is relatively expensive because it requires comprehensive chemical analysis of ambient and often source-specific PM 2.5 as well as careful application of the mass balance model.
Carbon-14 analysis, also referred to as radiocarbon analysis or carbon dating, has also been used extensively to apportion particulate matter carbon to biomass vs. fossil fuel sources (Szidat et al., 2004;Jordan et al., 2006;Ward et al., 2006b;Szidat et al., 2007;Gustafsson et al., 2009;Szidat, 2009;Ward and Lange, 2010;Buchholz et al., 2013).The 14 C content in emissions from the combustion of carbonaceous fuels mirrors that of the fuel.Carbon-14 is a cosmogenic isotope that is continually being produced in the upper atmosphere.Once produced, the carbon oxidizes to form CO 2 , and enters the terrestrial carbon cycle through several different avenues, including photosynthesis in plant life.The amount of 14 C found in a plant thus reflects the amount of 14 C present in the atmosphere when that plant grew.Over time after death, the 14 C fixed in an organic sample will decay, and if enough time is allowed to pass no 14 C will be detectable.Thus, analyzing the PM 2.5 samples for 14 C allows determination of the PM 2.5 emitted by biomass combustion ('modern' carbon -known 14 C) versus that emitted by fossil fuel combustion ('old' carbon -no 14 C), such as petroleum diesel exhaust or exhaust from coal burning sources.This approach is also relatively expensive due to the specialized instrumentation and expertise required, and the method is applicable to the carbon fraction of the particulate only.
Both inorganic and organic markers have been used in PM 2.5 source apportionment.Potassium is a recognized inorganic marker of PM 2.5 from wood combustion, and has been used to apportion PM 2.5 e.g., (Ward et al., 2006a;Caseiro et al., 2009).This is a relatively cost effective approach, but can over-estimate wood smoke contribution if additional sources of potassium are not considered or accounted for.
However, levoglucosan has not often been used for quantitative apportionment of PM 2.5 to residential wood burning because the mass fraction of levoglucosan in wood smoke PM must be known for it to be used as a quantitative tracer species.Several attempts have been made to quantitatively estimate wood smoke PM from levoglucosan or to measure and report conversion factors from levoglucosan to wood smoke PM.Hedberg et al. (2006) found that there was too much uncertainty or variation in the mass fraction of levoglucosan in wood smoke to allow quantitative estimates.Schmidl et al. (2008) and Caseiro et al. (2009) measured, reported and used a conversion factor of 10.7 to calculate wood smoke particulate from levoglucosan.Herich et al. (2014) compared results for multiple studies in alpine regions of Europe and found that wood smoke PM to levoglucosan ratios varied from 10.7 to 25.2.The positive matrix factorization (PMF) method has also been applied to develop quantitative conversion of levoglucosan to PM. Zhang et al. (2010a) used PMF to obtain a conversion factor of 18.3 for the southeastern US, while Piazzalunga et al. (2011) generated conversion factors of 10.4 using literature values and 16.9 using PMF in Italy.Others have used levoglucosan to quantitatively or semi-quantitatively apportion PM organic carbon to biomass burning in general (Zhang et al., 2010b;Sang et al., 2013;Zhu et al., 2015).Quantitative apportionment was limited in these latter studies by the lack of source-and region-specific mass fractions of levoglucosan in organic carbon from biomass combustion, requiring the use of an estimate calculated from published values.
An additional complication with the use of levoglucosan for quantitative apportionment is that it has been shown in laboratory studies to have limited stability in the presence of common photochemically-generated atmospheric free radicals (Hennigan et al., 2010;Hoffmann et al., 2010).This suggests that PM to levoglucosan ratios measured on "fresh" emissions from wood burning devices may not be representative of the values for aged wood smoke PM.However, levoglucosan is relatively stable in the winter months, which are the focus of this study (Hoffmann et al., 2010;Zhang et al., 2010a).Limited stability would thus be of greatest concern to efforts to use levoglucosan to sourceapportion PM that has been transported on a continental scale over a period of days, or in summer months when photochemical activity is relatively high.
In the current study we have used results from studies in Fairbanks Alaska PM 2.5 to obtain estimates for the quantitative levoglucosan to wood smoke PM 2,5 conversion ratio in that region.Fairbanks, Alaska has experienced elevated levels of PM 2.5 and was designated a nonattainment area due to frequent exceedance of 24 hour NAAQS standards during several heating seasons.Comprehensive chemical analyses and subsequent modeling of Fairbanks ambient PM 2.5 has included CMB, carbon-14, potassium, and levoglucosan analyses.These data present a rare opportunity to compare and evaluate these approaches to source apportionment in a single airshed.In the current study we evaluate and compare the 14 C, levoglucosan and CMB approaches to apportion PM 2.5 to residential wood combustion.We also utilize the correlations between levoglucosan and wood smoke PM 2.5 determined by the 14 C and CMB methods to gain estimates of field-relevant levoglucosan to wood smoke PM 2.5 conversion factors.

Methods
During  Twenty-four hour PM 2.5 sampling was conducted using a MetOne (Grants Pass, OR) Spiral Ambient Speciation Sampler (SASS) at each site.During each 24-hour sampling event, the SASS collected ~9.7 m 3 of air through Teflon, nylon, and two quartz filter media.In addition to traditional speciation analyses, one of the quartz filters was later analyzed for 14 C and chemical markers of wood smoke.Quality assurance and control procedures (USEPA, 2013) were followed throughout the sampling program.Following sampling, all filter samples were kept cold until their respective analyses described in the following sections.

PM 2.5 Speciation
Exposed SASS filter samples were analyzed by the Research Triangle Institute (RTI, Research Triangle Park, NC).From the Teflon filter, a gravimetric analysis (RTI, 2008) was initially performed followed by an elemental analysis (RTI, 2009d) using energy-dispersive X-ray fluorescence (EDXRF) where 33 elements were quantified.From the nylon filter, ions (including ammonium, potassium, sodium, nitrate, and sulfate) were measured by ion chromatography (IC) (RTI, 2009a, b).From the first quartz filter, Elemental Carbon and Organic Carbon (EC/OC) concentrations were quantified by Thermal Optical Transmittance following NIOSH protocol (RTI, 2009c).Following the analyses, sample results (including analyte concentrations and uncertainties) were provided for use in the CMB source apportionment model.

Carbon-14 ( 14 C) Analyses
For a random subset of the samples collected throughout the three-winter programs (and from each of the three sites), one half of the second quartz filter was analyzed for 14 C at the University of Arizona's (UA) Accelerator Mass Spectrometry Laboratory Facility.Carbon was extracted from each sample independently via combustion in an oxygen rich environment.The analysis yields the fraction of carbon in the PM 2.5 that is 14 C. Assuming no carbon on the blank filter and that the fraction of 14 C in old carbon is zero, the fraction of modern carbon can be calculated from (Ward and Lange, 2010): where F C14 (measured) is the measured fraction of 14 C and F C14 (modern) is the modern atmospheric fraction of 14 C. Several points need to be taken into account in assigning F C14 (modern).Nuclear testing in the 1950s and 1960s caused a significant increase of 14 C in the atmosphere, with a peak fraction of 1.85 parts per trillion (ppt) being reached in 1965.Since that time, the fraction has dropped to the present day level of approximately 1.075 ppt.With atmospheric concentrations changing yearly, the exact fraction for a piece of wood will be the integration of all 14 C incorporated over the period of growth.A complete survey of the wood harvesting methods, locations and average wood age, as well as the relative size and lifespan of the trees cut, was not possible to determine for this project.An examination of yearly atmospheric 14 C values over the last 130 years was conducted and decadal averages were calculated to obtain a high estimate of F C14 (modern) of 1.294 ppt, while the current value of 1.075 ppt was used as a low estimate of F C14 (modern).Using these values results in low and high estimates of %(modern).

Levoglucosan
Levoglucosan analysis was conducted using a previouslypublished method (Bergauff et al., 2008).This method was used for all samples except the Fairbanks source-specific filters.Because the PM loading of these filters was relatively high, it was necessary to adjust the filter fraction analyzed, amount of D-levoglucosan internal standard added, and the final dilution factors to obtain final analysis concentrations within the linear range of the original method.

Fairbanks Source-Specific Biomass PM 2.5
To support the CMB modeling, source emission testing was conducted by OMNI Environmental Services (Portland, OR).The goal of the OMNI testing was to generate emission profiles for the following types of heating appliances and fuel types commonly used in Fairbanks: pellet stoves, USEPA qualified wood stoves (birch, spruce), conventional wood stoves (birch, spruce), USEPA qualified hydroponic heaters (birch, spruce), non qualified outdoor hydroponic heaters (spruce, birch, wet stoker coal), oil burners (No. 1 fuel oil, No. 2 fuel oil), waste oil burning, coal stoves (dry stoker coal, wet stoker coal, wet lump coal, dry lump coal), and coal hydroponic heaters (wet stoker coal and coaltypical moisture).
Wood heating appliances were operated following USEPA method 28 (USEPA, 2014a), except that Alaskan cordwood was used in place of dimensional lumber.Particulate sampling was carried out in accordance with applicable portions of USEPA method 201A (USEPA, 2014b).The particulate sampling system relied on a cyclone head attachment on the sample probe in order to sample only particulate smaller than 2.5 microns in diameter (PM 2.5 ).The cyclone head was placed in the dilution tunnel and the sample flow was split into 5 branches, each with a respective filter (one Teflon, three quartz, and one glass fiber).The flow rate in each branch was individually controlled.From the Teflon filter, PM 2.5 mass, ions (potassium, sodium, ammonium, nitrate, and sulfate), and elements (33 in total) were quantified by the Research Triangle Institute.A single quartz filter sample for each of 41 trials with different devices, burn rates and wood species was shipped to the University of Montana for levoglucosan analysis.

Source Apportionment Modeling
The most recent version of the USEPA Chemical Mass Balance (CMB) computer model (Version 8.2) was utilized to apportion the sources of PM 2.5 in Fairbanks.The CMB receptor model (Hidy and Venkataraman, 1996;Friedlander, 1973;Cooper and Watson, 1980;Gordon, 1980, Watson, 1984;Watson et al., 1984;Gordon, 1988;Watson et al., 1990) is based on an effective-variance least squares method, and consists of a solution to linear equations that expresses each receptor chemical concentration as a linear sum of products of source fingerprint abundances and contributions.
A more complete description of the CMB modeling program (as well as experimental and analytical details) is provided in (Ward et al., 2012).Briefly, for each sample day, the CMB modeling process began by selecting from a combination of 91 sources and 43 chemical species (36 elements, 5 ions, OC and EC) in an effort to reconstruct the measured Fairbanks ambient PM 2.5 mass and chemical composition.Source profiles were either taken directly from the most recent version of SPECIATE 4.0 (USEPA, 2006) or from previous Missoula Valley CMB applications (Carlson, 1990;Schmidt, 1996;Ward and Smith, 2005).The types of source profiles included street sand and road dust, pure secondary source emissions, gasoline and diesel exhaust emissions, tire and brake wear, meat cooking, residential wood combustion, and other local sources/industry in Fairbanks.

Data Analysis
All results are reported as the means and 95% confidence intervals.Method intercomparisons are conducted using matched sample pairs.Distributions are compared using ttests with significance reported for the 95% confidence level.Results for wood smoke PM 2.5 by either CMB or 14 C analysis are plotted and regressed against measured levoglucosan PM 2.5 concentrations using Microsoft® Excel to obtain slopes and statistics.All plots and regressions are through the origin.

Chemical Mass Balance -Wood Smoke Contribution
The results of CMB source apportionment modeling are presented and discussed in detail elsewhere (Ward et al., 2012).Residential wood smoke was the major source of PM 2.5 identified by CMB modeling throughout the threewinter study in Fairbanks, contributing between 60% and 80% of the measured PM 2.5 at the three sites (see Table 1).The wood smoke source identified by the CMB model should be viewed as a general source predominantly composed of wood stove emissions.A source profile developed in Missoula, Montana in the late 1980s served as the best statistically fitting wood smoke profile for each of the three sites when conducting the Fairbanks CMB analyses.It should also be noted that many other residential wood combustion source profiles from the USEPA SPECIATE database gave good statistical fits throughout the computer modeling process for each of the sites.Note that the OMNI wood heating source profiles were not used for CMB modeling in this study.

Carbon-14
Throughout the 3-year program, OC concentrations averaged between 10.6 and 15.4 µg m -3 , with EC concentrations between 1.4 and 2.3 µg m -3 .PM 2.5 mass was composed of 43-58% OC and 6-9% EC, respectively, at each of the sites.The 14 C analyses return estimates of the fraction of total carbon attributable to biomass combustion.The CMB apportionment of Fairbanks PM 2.5 , however, suggests high non-carbon fractions, primarily of secondary sulfate.This causes some complications for quantitative apportionment of Fairbanks PM 2.5 using the 14 C method.To account for this we estimated the portion of total ambient PM 2.5 attributable to biomass from the mass fraction of biomass-generated carbon on the filter using Eq.(2): where PM 2.5 (% biomass) is the percent contribution of biomass burning PM 2.5 to the ambient PM 2.5 , X C,biomass is the mass fraction of carbon on the filter that originates from biomass combustion as determined by 14 C analysis, y C,biomass is the typical mass fraction of carbon in biomass emissions from an emission source profile, TC measured is the PM 2.5 total carbon (TC) concentration from the speciation sampler, and PM 2.5gravimetric is the gravimetric mass of PM 2.5 from the speciation sampler.The numerator in Eq. ( 2) represents the amount of biomass-generated carbon in the sample.That, divided by y C,biomass , yields an estimate for the total biomass PM 2.5 in the sample.Finally, division by PM 2.5gravimetric yields the biomass fraction.For y C,biomass , we used the measured average mass fraction of carbon (0.837) in the Fairbanksspecific wood stove emissions generated from the heating device combustion trials, which is not substantially different from the mass fractions generated from the Missoula wood smoke profile or the other USEPA wood smoke profiles from the SPECIATE database.
When using the values for fraction of modern carbon for each of the sample days, the percent wood smoke component of the PM 2.5 can be calculated.For the filter samples analyzed, ~32% to ~66% of the measured ambient PM 2.5 came from a new carbon source (in this case wood smoke).The % biomass PM 2.5 over all samples analyzed for 14 C are presented in Table 1.The results are reported as a minimum and maximum value based on high and low estimates of 14 C in modern biomass, respectively.

Levoglucosan
Detection of levoglucosan in the ambient PM 2.5 filter samples supports that wood smoke-related particles are present in the Fairbanks airshed.Table 2 presents averages (with 95% confidence intervals) for levoglucosan levels and PM 2.5 mass fractions by sampling site and year.The variability in these data reflects actual variations in levoglucosan concentrations and mass fractions as well as variations due to analytical reproducibility.Variability in the levoglucosan concentrations, expressed as relative 95% confidence intervals, is high, often exceeding 40%.This variation reflects that levoglucosan concentrations increase and diminish with PM 2.5 concentrations, which also vary significantly.Relative variations in levoglucosan as mass fractions of PM 2.5 are lower, and are typically 15% or less.Significant differences (Student t-test, p < 0.05) are observed between sampling sites, with the North Pole showing higher concentrations and mass fractions compared to the State Building and Peger Road sites.This is evidence that residential wood smoke accounts for a greater fraction of the PM 2.5 in the more rural North Pole location, consistent with the results of the CMB modeling.There are no significant differences or trends in levoglucosan concentrations or fractions for any given site as a function of heating season.A recent study (Caseiro et al., 2009) generated a quantitative apportionment of ambient PM 10 to biomass combustion in Austria by dividing the levoglucosan fraction of ambient PM 10 by the levoglucosan fraction of PM 10 from biomass combustion.The levoglucosan fraction of wood smoke was established by analysis of PM 10 from wood heaters burning wood species used in the region of Austria studied (Schmidl et al., 2008).The levoglucosan mass fraction is generally observed to vary between wood species (Fine et al., 2001(Fine et al., , 2002a(Fine et al., , b, 2004a, b;, b;Schmidl et al., 2008;Caseiro et al., 2009), so a representative value for the Austrian region was calculated as a weighted average based on a survey of the amount or fraction of each wood species consumed (Schmidl et al., 2008;Caseiro et al., 2009).The equation used to calculate a weighted conversion factor (CF) to convert levoglucosan mass fraction to biomass portion is as follows (Schmidl et al., 2008): where the fractional consumption (f n ) and mass fraction of levoglucosan in PM (L n ) is respective of each wood species (n) burned in the area of study.
A survey of wood fuel use in Fairbanks conducted by OMNI Environmental Services under contract to the Fairbanks North Star Borough found that residents used 43% aspen, 52% birch, and 6% spruce.Adaptation of the Eq. ( 3) to Fairbanks using survey data for wood species consumption yields: where L A , L B , and L S are the levoglucosan mass fractions for aspen, birch and spruce wood smoke respectively.We have investigated various approaches to calculating a conversion factor for Fairbanks using experimentally-generated and published levoglucosan fraction values for aspen, birch and spruce.
Fairbanks-specific levoglucosan mass fractions in PM 2.5 from biomass combustion were determined using PM 2.5 samples from the heating device combustion The results for levoglucosan fraction in PM 2.5 for these filters are presented in Table 3.In general, these results indicate a relatively low fraction of levoglucosan in the wood smoke (avg.= 3.7%) compared to published values (Fine et al., 2001(Fine et al., , 2002a(Fine et al., , b, 2004a, b;, b;Schmidl et al., 2008;Caseiro et al., 2009).No significant differences were observed in levoglucosan fraction across wood species based on Student ttest analysis, which is also not consistent with previous studies (Fine et al., 2001(Fine et al., , 2002a(Fine et al., , b, 2004a, b;, b;Schmidl et al., 2008;Caseiro et al., 2009).Significant differences were observed as a function of burner type and within burner types as a function of burn rate, as has been previously reported by (Jordan and Seen, 2005).
The results of this study demonstrate the substantial challenges in obtaining reliable experimental levoglucosan mass fraction results for use in quantitative source apportionment studies.Although this study was conducted by experienced contractors and followed approved USEPA methods and procedures, the results obtained deviate significantly from previously-published results.Due to cost and time limitations, the data are limited to single samples for each device, burn rate and wood species, and include results for spruce and birch, but not aspen.The average levoglucosan fractions reported in Table 3 are significantly lower than typical ambient PM 2.5 levoglucosan fractions at the North Pole site, implying unreasonable apportionments in excess of 100% wood smoke for this site and further eroding confidence in the experimental results.The stove burn rate is clearly an important factor, and burn rate data are difficult to collect in the field and are seldom available.
An alternative approach is to use levoglucosan mass fraction data reported in the literature.Significant variation in the published values complicates this approach (Hedberg et al., 2006;Herich et al., 2014).No published results specific for appliances and practices in Fairbanks are available, which may introduce significant error.Experimental levoglucosan fractions of PM 2.5 are reported in the literature for wood smoke from aspen, birch and spruce (Fine et al., 2004a).Other published results for levoglucosan fractions do not include the same species as those burned in Fairbanks and/or are for PM 10 rather than PM 2.5 (Fine et al., 2001(Fine et al., , 2002a(Fine et al., , b, 2004a, b;, b;Schmidl et al., 2008).The reported experimental levoglucosan fractions in each case are based on relatively few measurements, and their reliability is thus of concern.An additional concern is that the values are measured for "fresh" wood smoke PM and may not be valid for application to aged PM for which levoglucosan levels may be reduced via reaction with atmospheric free radicals (Hennigan et al., 2010;Hoffmann et al., 2010).
We used a combination of the experimental and published values for L A, L B and L s to establish a low and a high estimate of the conversion factor.Using only the most relevant published results (Fine et al., 2004a) gives a CF 1 = 9.01, which is used here as a lower limit.An upper limit CF was calculated using the average experimental values for L B and L S from Table 3 over all burn conditions and the published value of L A .The resulting CF 2 = 13.3 is strongly influenced (43%) by the published value for aspen.Finally, device type data by zip code was utilized together with wood species survey data to generate site-specific CF values weighted for both wood species and device type.These conversion factors, calculated using L B and L S from Table 3 and the published value for L A , ranged from 12.2-12.4.There was significant concern about these site-specific results because of the combined uncertainties in L values, wood species usage, and stove type usage.Because of this, and because they are bracketed by CF 1 and CF 2 , they were not used for additional calculations.Our values for CF 1 and CF 2 also bracket those reported by Caseiro et al. (Caseiro et al., 2009) and are at the low end of the range reported by Herich et al. (Herich et al., 2014).
Using CF 1 and CF 2 we calculated low and high estimates of the wood smoke percent contribution to ambient PM 2.5 in Fairbanks.Table 1 presents these results by site and season.The high end estimates are nearly 48% higher than the low end estimates, representing considerable uncertainty.

Overall Comparison
Table 1 presents the residential wood smoke apportionment results using all CMB, 14 C and levoglucosan analyses.Levoglucosan data were eliminated for a few low PM days with levoglucosan concentrations near or below the detection limit.No more than two data points were eliminated for any heating season.Levoglucosan and CMB results are each large datasets with considerable overlap.Carbon-14 results are from a relatively small random subset of samples.
Average apportionment results by site are presented in Fig. 1, with error bars indicating 95% confidence intervals.The results from 14 C and levoglucosan analysis are generally in good agreement.There are no significant differences between the average results at the State Building or Peger Road sites based on a Student t-test at p = 0.05.The lower levoglucosan estimate at the North Pole site shows a significant difference from the 14 C estimates, but the upper levoglucosan estimate does not.At the more centrally-located State Building and Peger Road sites, the CMB analysis suggests a significantly higher apportionment to wood smoke than do either the 14 C or levoglucosan approaches.The results converge at the North Pole site, where all three approaches suggest a higher proportion from biomass combustion.
Given the challenges in calculating a representative levoglucosan conversion factor from experimental or published data, we chose to use the results from different apportionment methods to obtain average CF values for Fairbanks.This was accomplished by analysis of observations in which paired data for both levoglucosan and either CMB or 14 C were available.

Levoglucosan vs. CMB
Using the subset of data for which both CMB and levoglucosan analyses were performed, the wood smoke PM 2.5 concentration estimated from CMB is plotted vs the measured levoglucosan levels in Fig. 2(a).Inspection of this plot suggests a different relationship between CMB and levoglucosan results at the State Building and Peger Rd sites compared to the North Pole site.Separate regression of the results at the three sites yields slopes of 15.12 ± 0.39 (r 2 = 0.96, F = 1470, n = 57), 23.3 ± 2.2 (r 2 = 0.89, F = 464, n = 58) and 19.8 ± 2.5 (r 2 = 0.84, F = 245, n = 46) at the North Pole, Peger Rd., and State Building sites, respectively.The slopes are estimates of the CF values assuming that CMB modeling provides an accurate estimate of wood smoke PM 2.5 .Each of these values is significantly higher than the calculated upper limit of CF 2 = 13.3.
Although no organic tracers (such as levoglucosan) were used in this CMB application to apportion the wood smoke component, we did consistently utilize 43 "common" chemical species including elemental potassium and the potassium ion.The ambient PM 2.5 in Fairbanks was heavily influenced by elevated concentrations of OC and EC, as well as high concentrations of sulfur and sulfate.A source profile developed in Missoula, Montana in the late 1980s served as the best statistically fitting wood smoke profile for the Fairbanks CMB.However, it is important to note this wood smoke profile was not representative of the specific types/models of wood burning devices used in Fairbanks, nor the sources of fuels (birch and spruce) that are typically combusted.We also hypothesize that the sulfur component was not correctly apportioned in this CMB application, and likely influences the amounts apportioned to other sources in our modeling (including wood combustion).We suspect there is a missing source (such as fuel oil combustion) that was not identified.This is supported by the especially high wood smoke proportions and CF values from CMB modeling for the State Building and Peger Road sites.Through the OMNI combustion trials described above, we have developed Fairbanks-specific profiles for various emissions and will report the results of the updated CMB modeling with those profiles in a subsequent publication.

Levoglucosan vs. Carbon-14
The wood smoke PM 2.5 mass contribution estimates from the 14 C data are plotted vs. ambient PM 2.5 levoglucosan concentrations on a sample by sample basis in Fig. 2(b).Results for 14 C presented in this figure are based on the arithmetic mean of the minimum and maximum contribution estimates.The plot suggests that the relationship is Fig. 1.Three year average (95% CI) values of wood smoke apportionment using 14 C analysis (horizontal pattern, minimum light and maximum dark), levoglucosan analysis (vertical pattern, low estimate light and high estimate dark), and CMB modeling (stippled pattern).independent of site and all of the data were pooled across sites for subsequent analyses.Fig. 2(b) demonstrates a high correlation between the levoglucosan and 14 C measures with a slope (CF) of 11.82 ± 0.67 (r 2 = 0.97, F = 1257, n = 40).This CF value is less than 10% higher than that published by Caseiro et al. (2009) for PM 10 (10.7).Regression analysis after eliminating the highest point (CF = 11.31 ± 0.62.r 2 = 0.97, F = 1374, n = 39) and four highest points (CF = 11.46 ± 0.70.r 2 = 0.97, F = 1121, n = 36) from North Pole yielded no significant differences in the CF values or regression statistics.Using all data and the minimum and maximum wood smoke PM 2.5 estimates from the 14 C data yielded CF = 10.72 ± 0.61 and 12.91 ± 0.74, respectively.Another approach is to calculate and average the ratios of wood smoke PM 2.5 to levoglucosan for each sample.Using minimum and maximum estimates for wood smoke PM 2.5 from the 14 C data yielded mean CF values of 11.45 ± 0.89 and 13.8 ± 1.1, respectively.These estimates are slightly higher than the regression slope estimates because of a non-zero regression intercept.All of these values are within the range of estimates of CF 1 = 9.01 and CF 2 = 13.3 calculated and presented above within 95% confidence, and fall within but at the lower end of the range published by (Herich et al., 2014).

CONCLUDING REMARKS
Source apportionment of ambient PM 2.5 to residential wood heater use is an important issue in many areas.Our results suggest that this can be done reliably using a variety of means.Each method has its strengths and limitations.Each relies, at least to some extent, on knowledge of local emission profiles.
In the current study, the CMB method appears to overestimate the contribution of residential wood heating.The method relies on the availability of representative emission profiles which may be difficult to obtain in areas like Fairbanks with unique climate and atmospheric chemistry.
The method is relatively expensive and time consuming, as it requires comprehensive chemical analysis of filter samples as well as considerable experience, time and effort with the model.
Apportionment through 14 C analysis is based on well understood and accepted principles, and thus allows good confidence in the results.The method only allows apportionment of the carbon in the PM, however.In situations like Fairbanks, where non-carbon species make up a significant fraction of the PM, the results must be corrected for the expected carbon content of biomass combustion emissions.This approach is also expensive as it utilizes costly and dedicated instrumentation.
The use of levoglucosan PM 2.5 mass fractions for apportionment has the advantage of being relatively inexpensive.The uncertainty in this approach is primarily in the conversion from levoglucosan fraction to wood smoke PM portion.The cost advantage is not real if location-specific conversion factors must be measured for each study because this is both expensive and, in our experience, not necessarily reliable.Efforts in this study indicate that an appropriate conversion factor lies in the range of 9.1 to 13.3.Using values in this range results in wood smoke apportionment consistent with that obtained from 14 C analysis.
To our knowledge, this is the first report with extensive results correlating levoglucosan levels to wood smoke PM 2.5 levels from 14 C analysis.These correlations result in conversion factors in the range of 10.7 to 12.9.These conversion factor values come from field data, and are as accurate and reliable as the 14 C apportionment results.Similarity of the measured values to those calculated from levoglucosan emission factors for fresh wood smoke PM also suggest a limited or negligible effect of levoglucosan instability on the conversion factor, at least during the relatively cold and dark winter months in Fairbanks and in a localized airshed where wood smoke PM 2.5 is continuously emitted.Previous studies have suggested conversion factors in the range of 10.7 to 25.2.The convergence of the current values and those for studies in different regions suggests that reasonable wood smoke apportionment estimates can be obtained using levoglucosan mass fractions and a conversion factor in the range of 9.1 to 13.3, and likely in the narrower range of 10.7 to 12.9.
CMB is intended to complement rather than replace other data analysis and modeling methods.Results from this study show that alternative approaches can be used to evaluate results from CMB modeling, as well as provide a less expensive alternative to source apportionment of residential wood smoke contributions.

Fig. 2 .
Fig. 2. (a) Wood smoke PM 2.5 as determined by CMB analysis vs ambient levoglucosan concentration in PM 2.5 .(b) Wood smoke PM 2.5 as determined by 14 C analysis vs ambient levoglucosan concentration in PM 2.5 . 14C wood smoke levels are the mean of minimum and maximum values.◊ State Building, □ North Pole, ○ Peger Road.

Table 1 .
Wood smoke (WS) contributions to ambient PM 2.5 as determined by 14 C analysis, levoglucosan (LG) analysis, and CMB modeling.

Table 2 .
Average levoglucosan (LG) mass concentrations and mass percentages in PM 2.5 for three sites over three years.

Table 3 .
Levoglucosan shares for various devices, fuels and burn rates.