Comparison of methods for evaluation of wood smoke and estimation of UK ambient concentrations

. Airborne concentrations of the wood smoke tracers, levoglucosan and ﬁne potassium have been measured at urban and rural sites in the United Kingdom alongside measurements with a multi-wavelength aethalometer. The UK sites, and especially those in cities, show low ratios of levoglucosan to potassium in comparison to the majority of published data. It is concluded that there may be two distinct source types, one from wood stoves and ﬁreplaces with a high organic carbon content, best represented by levoglucosan, the other from larger, modern appliances with a very high burn-out efﬁciency, best represented by potassium. Based upon levoglucosan concentrations and a conversion factor of 11.2 from levoglucosan to wood smoke mass, average concentrations of wood smoke including winter and summer sampling periods are 0.23 µg m − 3 in Birmingham and 0.33 µg m − 3 in London, well below concentrations typical the


Introduction
The enforcement of legislation internationally is acting to reduce airborne concentrations of particulate matter, measured as PM 2.5 and PM 10 . In the European Union, member states have to meet an air quality objective of 25 µg m −3 for PM 2.5 as an annual mean by 2015, as well as an exposure reduction target to be met by 2020 whose magnitude depends upon the annual average concentration across qualifying sites in the years 2009-2011 (Harrison et al., 2012). While the contribution of components such as sulphate and nitrate to concentrations of PM 2.5 is readily determined, it is more difficult to quantify the contributions of primary sources such as road traffic to PM concentrations, and estimating the secondary organic aerosol contribution is especially problematic (Pio et al., 2011).
Amongst the sources most difficult to quantify by receptor modelling, and for which emissions inventory data are least reliable, wood smoke stands out as posing particular problems. Collecting activity data for emissions inventories is challenging and emission factors are highly variable depending upon the type of combustion appliance, or open combustion practice adopted (Kelz et al., 2010;Pio et al., 2008).
Since the estimation of emissions poses such large difficulties, most estimates of the source contribution are based upon receptor modelling methods. These may be based upon a CMB modelling approach using a range of compositional components (e.g. Chow et al., 2007), or more often upon individual source tracers. The majority of studies have measured anhydrosugars, of which levoglucosan is the predominant component. This method appears to be relatively specific to biomass combustion, but the ratio of levoglucosan to wood smoke mass is highly variable, depending upon the specific wood type being burnt and the combustion conditions (Puxbaum et al., 2007). Nonetheless, levoglucosan has been widely used to estimate contributions of wood smoke to organic carbon and to wood smoke mass (e.g. Puxbaum et al., 2007;Caseiro et al., 2009).
Fine particle soluble potassium has also been recommended as a wood smoke tracer (Chow et al., 2007). If it is to be used as a specific tracer, as opposed to part of a source profile in a CMB model, then it must be corrected for the contributions of sea salt and soil to airborne potassium (K). While the former can be achieved rather reliably in most situations based upon measurements of sodium, correction for the soil contribution is more difficult due to the inhomogeneity of soils and hence the difficulty of taking soil samples which give a relevant ratio of K/Ca to permit use of Ca as a soil tracer. Soil in the fine particle fraction may have travelled for significant distances and may hence differ substantially in composition dependent upon the source area. A third method for estimating wood smoke concentrations is based upon use of a dual, or multi-wavelength aethalometer (Sandradewi et al., 2008a, b). This method depends upon the assumption that elemental carbon arises from only two sources, road traffic and wood smoke, which may be differentiated from their Angstrom coefficients (i.e. the wavelength dependence of the optical absorption coefficient). This assumption may be correct in Swiss valleys under the atmospheric inversion conditions prevailing when the method was developed (Sandradewi et al., 2008a, b), but the applicability of the method under other circumstances has not been systematically explored.
There have been few studies of the source apportionment of particulate matter in the UK atmosphere. Yin et al. (2010) applied a CMB model to chemical data including a range of organic source tracers including levoglucosan. Their estimates of wood smoke contributions to annual mean PM 2.5 mass were only 0.07 and 0.06 µg m −3 respectively for urban and rural sites, with rather little difference between seasons. These estimates, measured in the UK West Midlands area, are in sharp contrast to data reported by Fuller et al. (2011) based upon levoglucosan concentrations measured in the London area showing estimated wood smoke concentrations of approximately 3 µg m −3 during winter. Even allowing for a divergence of a factor of approximately two in levoglucosan measurements between G. W. Fuller and ourselves revealed by an inter-laboratory comparison, there remains a very large discrepancy which seems unlikely to be explained by the different geographic areas in which measurements were made.
In this study, we report measurements of the wood smoke tracers levoglucosan and fine K at a number of UK locations. These are used to estimate concentrations of wood smoke, which are compared with estimates derived from simultaneous deployment of a multi-wavelength aethalometer.

Experimental
Four air sampling sites were used as follows: The site is about 3.5 km southwest of the centre of Birmingham which has a population of over one million and is part of a conurbation of 2.5 million population. The nearest anthropogenic sources are a nearby railway, and some moderately trafficked roads. There is little residential accommodation within 300 m. Samples were collected daily for about 8 days in each month. Data from this site have previously been used in a Chemical Mass Balance receptor model of PM 2.5 (Yin et al., 2010).

North Kensington, London (3 to 29 June 2010 and 16 February to 15 March 2011)
This is an urban background site in West London (51.52 • N; 0.21 • W). It is in the grounds of a school in a residential area, 7 km to the west of central London and hosts a station of the National Automatic Urban and Rural Network. The site is widely accepted as representative of air quality across a large part of London and the air pollution climate at this site has been analysed in detail by Bigi and Harrison (2010).

Marylebone Road, London (16 February to 15 March 2011)
This is on a heavily trafficked (ca. 80 000 vpd) six lane highway running through a street canyon in central London (51.52 • N; 0.16 • W). It shows a large increment in particle mass above neighbouring central urban background sites (Charron et al., 2007) and the particle sources influencing the site have been identified tentatively by Harrison et al. (2011).

Budbrooke, Warwick Sampling Site (19 November 2009 to 8 April 2010)
This is a rural location (52.17 • N; 1.38 • W), 55 km to the southeast of Birmingham and 4 km to the west of Warwick). The sampler was located in a clearing on the edge of an area of hardwood forest and was subject to local wood smoke emissions both from domestic wood stoves and from the  (λ = 370, 470, 520, 590, 660, 880 and 950 nm) was deployed for the measurement of black and brown carbon. The instrument was fitted with a PM 2.5 inlet and operates at 4 l min −1 . Data were corrected for loading effects according to the commonly used algorithm of Weingartner et al. (2003).

Potassium
PTFE filters were extracted into water and analysed by ion chromatography for potassium, sodium and calcium according to the methods described by Yin and Harrison (2008).
The concentrations of potassium needed to be corrected for the minor contributions of sea salt and soil to the fine fraction. The corrections were based on the method of Pio et al. (2008) which use sodium as a tracer for sea salt and calcium for a tracer of soil and assumes a mass ratio of 10 for K/Ca in wood smoke. In order to generate the corrections, the standard ratio of K/Na of 0.036 in sea salt was adopted and the soil calcium concentration was estimated by the sampling of local soils followed by sieving to collect a < 20 µm size fraction which was extracted with water and analysed according to the same methods as the air filters. Therefore, site specific corrections were adopted. The equation used to estimate the potassium arising from wood smoke is as follows: where the subscript "nss" = non-sea salt, [K/Ca] soil is the measured elemental ratio in soil and Ca nss is the measured Ca in air corrected for sea salt Ca calculated from the Na concentration. (Note: the subscript "ws" has been used to distinguish the results of chemical tracer studies from those derived from the aethalometer, designated "wb".)

Levoglucosan
Two procedures were used for the analysis of levoglucosan. Samples from the EROS and Budbrooke sampling sites were analysed according to the method of Zdrahal et al. (2002). Before extraction, an internal standard (methylbeta-D-xylopyranoside, C 6 H 12 0 5 ) was spiked evenly across the aerosol filter segment (1/4) to account for losses during sample extraction and concentration. The spiked filters were extracted three times, each time for 20 min with 30 ml of dichloromethane under ultrasonic agitation. The first extraction was performed under acidic conditions by addition of acetic acid (200 µl). The combined dichloromethane extracts were reduced with a rotary evaporator (400 hPa, 25 • C) to approximately 5-10 ml. Then the concentrated extracts were filtered through a PTFE filter and completely evaporated under a stream of nitrogen. Finally, the evaporated samples were re-dissolved in pyridine (200 µl) then kept in the 4 • C refrigerator until derivatization.
Aliquots of 50 µl of the sample solutions were taken out and derivatized by trimethylsilylation mixture (40 µl). After derivatization, a 10 µl recovery standard (1 ppm) was added in to the sample and total of 100 µl final sample was analysed by GC/MS. In this study, 1-phenyl dodecane (C 18 H 30 ) was applied as recovery standard. The addition of recovery standard was to compensate for injection volume effects and variations in the GC/MS detector response.
The derivatization process was carried out on batches of 10 samples using the trimethylsilylation mixture which comprised 99 % of N-methyl-N-trimethylsilyltrifluoroacetamide (MSTFA) and 1 % trimethylchlorosilane (TMCS). The trimethylsilylation mixture was freshly made before the derivatization process which was performed in a sealed vial for 60 min at 80 • C in a dry heater block. After derivatization, the sealed vials were opened to add the recovery standard, and then analysed immediately by GC/MS. Quantification was carried out by GC/MS and was based on an internal standard calibration procedure with appropriate recovery and blank corrections. A sample volume of 1.0 µl was injected into a split/splitless injector, operated in the splitless mode at a temperature of 250 • C. The carrier gas was helium at a pressure of 120 kPa. The temperature program was started at 45 • C for 4 min, a gradient of 20 • C min −1 was used up to 100 • C, followed by 10 min at this temperature, then the temperature increased to 315 • C at 5 • C min −1 and was held for 20 min. The m/z ratios for the ions used in the GC/MS analysis are 204, 217 and 333.

R. M. Harrison et al.: Comparison of methods for evaluation of wood smoke
Samples from London North Kensington and Marylebone Road were analysed according to the method of Yue and Fraser (2004) as described by Yin et al. (2010). This method uses derivatization with N, O-Bis (trimethylsilyl), trifluoroacetamide plus 10 % trimethylchlorosilane (BSTFA-TCMS) with subsequent GCMS analysis on an Agilent Technologies GCMS (GC-6890N plus MSD-5973N) fitted with a HP-5MS column (30 m, 0.25 mm diameter, 0.25 µm thickness). A levoglucosan-U 13 C6 isotopically used internal standard was utilised.
The two procedures for levoglucosan analysis were intercompared both by using a set of environmental samples and by analysis of levoglucosan in NIST SRM 1649B which contains 81.1 ± 6.1 µg g −1 levoglucosan. Both methods were acceptable with a slight over-estimation (up to 10 %) from the former and under-estimation (up to 20 %) from the latter. The precision of the method is <10 % r.s.d.

Aethalometer
Using the methodology of Sandradewi et al. (2008a, b) the contributions from traffic and wood burning to the total measured carbonaceous material, CM (organic matter + black carbon) can be apportioned using the different absorption coefficients measured for both sources at 470 nm and 950 nm by the aethalometer. By assuming that theÅngström exponent for both traffic and wood burning has a fixed value (1.0 and 2.0 respectively were used initially), simultaneous equations can be derived -and solved -to calculate an absorption coefficient for both wood smoke and traffic. Linear regression then correlates these values to a CM traffic and CM wb value and an intercept representing other carbonaceous aerosol sources which sum to give the total CM values.
The absorption coefficients b abs (470) and b abs (950) were derived from attenuation values ATN of the two different wavelengths (λ = 470 nm and 950 nm) measured across a quartz tape filter.
in which I o and I are both beam intensity before and after attenuation by the particle-laden quartz tape filter. Given the measured values of ATN for each of the wavelengths, the flow rate Q, the sampling time t and the filter spot size A = 1.67 cm 2 , values of aerosol attenuation coefficients b abs were calculated using For both wavelength (470 and 950 nm) channels, the b ATN values were "despiked" and smoothed using an in-house algorithm, before a correction due to Weingartner (2003) was applied to account for the "shadowing" effect of the particles as the filter becomes loaded.
The absorption coefficient b abs was wavelength (λ) dependent b abs ∝ λ −α and from this relation, theÅngström exponent α, can be derived. When comparing the ratio of the absorption by the wood smoke (which has a higher content of organic material) to the absorption by traffic particles (which have a higher of content of black carbon), both traffic and wood smoke have a stronger absorption at the shorter wavelengths, but wood smoke absorbs significantly more radiation than traffic at 470 nm compared to 950 nm. This is reflected by the typically higher absorption exponents for wood smoke compared to traffic. It follows that where the wood burning and traffic absorption components are b abs (470) wb and b abs (950) traf , respectively. Combining the above equations gives the following matrix: Having solved for the wood burning and traffic absorption components, b abs (470) wb and b abs (950) traf these are regressed against the total daily carbonaceous matter concentrations, CM.
where the first two terms corresponds to CM traffic and CM wb and C 3 is a residual, non-absorbing component.
Starting with values of α traff = 1.0 and α wb = 2.0 and a fixed value of C 1 = 260 000 µg cm −2 (Favez et al., 2010), initial temporal profiles of concentration did not look realistic, and optimization led to use of α traff = 1.05 and α wb = 2.0. Fixing C 1 = 260 000 µg m −2 , linear regression gave C 2 = 632 420 µg m −2 and C 3 = 1.331 µg m −3 . If C 1 was allowed to vary, the regression analysis gave similar values to these for all three C-variables. When winter (n = 131) and summer (n = 32) data were run in the regression analysis separately, the values of C 2 changed little, but C 3 was 0.619 µg m −3 in winter and 2.74 µg m −3 in summer.
During the course of this work, the Delta-C method was proposed by Wang et al. (2011) and data have also been analysed according to this method. Delta-C was simply taken as the difference between the black carbon (BC) values measured at 370 nm and 880 nm using the aethalometer and the in-built absorption coefficient values.
Measurements were also made of the elemental and organic carbon content of the samples using a Sunset Laboratory thermo-optical analyser and the EUSAAR 2 protocol (Cavalli et al., 2010).

Elms Road Observatory Site (EROS), Birmingham
This was the longest campaign, lasting in all for 11 months. The measured concentrations of levoglucosan are summarised in Fig. 1. This shows the expected seasonal variation with elevated concentrations from October to February when home heating is required and consistently very low concentrations of around 0.01 µg m −3 from March to September. Basic statistics and ratios of levoglucosan/K are summarised in Table 1 and the relationship between the variables appears in Fig. 2. No clear seasonal variation in ratio was evident. Calculation of wood smoke and traffic particle concentrations from the aethalometer data ( Fig. 3) using the optimised coefficients gave a diurnal profile of PM traffic as is normally observed for CO or NO x from traffic. The weekday diurnal pattern for estimated wood burning smoke (PM wb ) and particularly the different diurnal profile and higher concentrations seen on weekends are in line with expectations, but the average concentrations far exceed those measured by the other techniques (see below).

North Kensington
Data from this site were limited to two campaigns in summer and winter respectively, rather than sampling through the entire year. Basic statistics and ratios of levoglucosan/K appear in Table 1, and the relationship between the variables in Fig. 2. The summer and winter data are differentiated in Fig. 2 showing the clear seasonal dependence of concentrations, but no obvious systematic difference in levoglucosan/K ratios between seasons. The middle panels in Fig. 3 show diurnal variations in PM from traffic and wood smoke estimated from the aethalometer data. In this case, these conform to expectations in terms of the shape of both profiles, with a typical rush hour-related pattern to the traffic-related concentrations and a nocturnal elevation in the wood smoke, with a lesser diurnal variation at weekends when houses may be occupied throughout the entire day.
In the case of the North Kensington data, a mass of traffic particles was estimated from the aethalometer data and regressed upon the mass of black carbon (BC) directly measured by the aethalometer at 880 nm. This showed a very close relationship. The gradient is consistent with a primary OC : EC ratio in traffic emissions of 0.3-0.4 estimated by Harrison and Yin (2008) and Pio et al. (2011), together with an OM : OC ratio of around 1.5, which is high suggesting a modest over-estimation of the traffic particle mass. An organic aerosol : BC ratio measured at low concentrations in a rural tunnel (Chirico et al., 2011) is also consistent with the Pio et al. (2011) estimate of OC : EC in traffic emissions.

Marylebone Road, London
Air samples were collected over the same one month period as the winter campaign at North Kensington. The sites are separated by about 4 km and the difference between the two has often been used to infer a traffic contribution. A summary of the measured data appears in Table 1, showing concentrations of levoglucosan and K at Marylebone Road to be very similar in magnitude to those at North Kensington. Linear regression calculations using the Reduced Major Axis method between the sites gave the following relationships: for K ws MR = 0.86 NK − 0.02 µg m −3 r 2 = 0.85 for levoglucosan MR = 1.14 NK − 0.002 µg m −3 r 2 = 0.57 (13) [Note: the subscript "ws" has been used to distinguish the results of chemical tracer studies from those derived from the aethalometer, designated "wb". MR = Marylebone Road; NK = North Kensington]. These gradients of close to one and moderate to high r 2 coefficients strongly suggest a rather uniform distribution of wood smoke markers across central London. The lower r 2 coefficient for levoglucosan than for potassium is not readily explained, but may reflect greater localised emissions of the former than the latter marker, but this is hard to explain as the Marylebone Road site is surrounded largely by institutional Fig. 4. Ratio of levoglucosan: K ws at Budbrooke according to month of observation. and commercial buildings, rather than domestic, and this has the slightly higher levoglucosan concentrations.

Budbrooke, Warwick
This campaign ran from November 2009 to April 2010 and therefore included the season most impacted by wood smoke. As outlined above, the site was subject to the influences both of local and regional sources of wood smoke from home heating and from forestry activities. The latter involved burning of wood in the open air, and hence a wide ratio of levoglucosan/K ratios was expected, and was observed (see Table 1). Another consequence of the mixed sources was that the correlation between levoglucosan and K seen in Fig. 2 was weaker than for the other sites. At this site the levoglucosan/K ratio showed a strong seasonality (Fig. 4) not seen at the other sites. We interpret this as reflecting a greater contribution to wood smoke from open burning in the warmer months, with the low ratio in January dominated by the home heating emissions.
The analysis of aethalometer data for this site (see Fig. 3) showed profiles rather similar to North Kensington, but with much higher PM wb than PM traffic mass concentration.

Levoglucosan: potassium (levoglucosan/K ws ) ratios
The range of levoglucosan/K ratios (10-90 %ile) shown for all sites in Table 1 is considerable. Negative values arise from occasionally negative values of K ws , probably associated with advection of soils with a low K/Ca ratio, therefore causing an excessive correction. The seasonal difference for N. Kensington in the median, but not the mean is in the expected direction, with summer garden bonfires expected to have a higher levoglucosan/K ws ratio than winter emissions from wood burning stoves. However, this is an area of London with high density housing where garden bonfires are expected to be infrequent. Greater degradation of levoglucosan in summer than winter (Hennigan et al., 2010) would be expected to reduce levoglucosan/K ws ratios in summer. Taking the gradients of Fig. 2 as the best representa-tion of levoglucosan/K ws ratios, these were 0.74, 0.34, 0.18 and 0.15 for Budbrooke, EROS, Marylebone Road and North Kensington respectively, and fall well outside the range cited by Puxbaum et al. (2007) from published studies. Gradients were preferred to simple ratios of means as they are less subject to the effects of other unaccounted sources which appear as intercepts in the regression. The reported data from Puxbaum et al. (2007) range from ratios of 1.1 for quercus in a US fireplace (Schauer et al., 2001) to 50 for spruce in an Austrian wood stove (Schmidl, 2005), both for PM 10 . Ranges reported for PM 2.5 are narrower, from 33.3 for softwood in a US woodstove (Fine et al., 2004) to 6.25, for softwood in a US fireplace (Fine et al., 2001). Caseiro et al. (2009) also provided a review of ratios from the recent literature, ranging from 0.11 for PM 2.5 in a forest fire (Pio et al., 2008) to 200 from softwood burned in a woodstove (Schmidl et al., 2008). Chow et al. (2007) report chemical profiles of hardwood and softwood combustion in which levoglucosan/K ratios are 0.77 and 0.15 for hardwood and softwood, respectively, nicely spanning the values measured in this study.
There appears to be a huge variation in levoglucosan/K ratio according to the wood type. According to data reported by Goncalves et al. (2010) derived from combustion studies in an Austrian-design wood stove, levoglucosan/K ratios can vary from as low as 0.19 for acacia to 4.9 for eucalyptus with pinus and quercus in between, with no systematic variation between hardwoods and softwoods. Schmidl et al. (2008) measured a range of organic compounds and metals in emissions from an old type of domestic tiled stove (from 1994). The reported levoglucosan/K ratios ranged from 19.5 to 216 for the various wood types. However, these authors commented that the abundance of ionic species (including K) in their work was lower than in many previous studies due to samples being taken at the start and middle of the burn. This is likely to have not only biased the K data downwards, it will also have biased the levoglucosan data upwards leading to unreliable levoglucosan/K ratios if considering the entire combustion.
The ranking of levoglucosan/K ratios of Budbrooke > EROS > N. Kensington Marylebone Road is in the order anticipated if the contribution from smouldering combustion in open air sources leads to high levoglucosan/K ratios, given the nature of the activities local to the four sites. However, the absolute magnitude of the levoglucosan/K ws ratios is clearly out of line with earlier studies, and requires explanation. The form of the levoglucosan/K ws relationships revealed by Fig. 2 is not suggestive of any systematic over-estimation of K ws or under-estimation of levoglucosan as might be reflected in a large intercept. The concentrations of levoglucosan and K ws range from low in summer to high in winter as might be anticipated (Fig. 1), and relate inversely to temperature (Fig. 5).
Three explanations for low levoglucosan/K ws relationships appear plausible: a. the levoglucosan/K ws ratio is highly sensitive to combustion conditions with high temperature flaming combustion leading to a more complete burn-out of the organic carbon content and hence a low levoglucosan/K ws ratio. Levoglucosan as a proportion of wood smoke particles is reported to range from 43 to 309 mg g −1 (Puxbaum et al., 2007) in published studies, while according to Khalil and Rasmussen (2003), the abundance of K in wood smoke can range from 1.6-102 mg g −1 .
While these are not independent of one another, there still appears to be much scope for levoglucosan/K ratios well outside the range of those reported in the literature. Kelz et al. (2010) compared particulate matter emissions from modern state-of-the-art and older combustion appliances. The modern devices included a pellet boiler, wood chip boiler, logwood boiler, logwood stove and tiled stove, and the older devices comprised a logwood boiler and logwood stove. The particulate matter from the modern pellet boiler was almost 100 % inorganic, while the modern woodchip boiler and logwood boiler emitted particles with < 20 % carbonaceous content. Particulate emissions from the other four devices were predominantly carbonaceous with ca. 60-90 % elemental carbon and organic compounds. The emissions from the three modern devices with lowest carbon content contained 70-90 % alkali metal sulphates, chlorides or carbonates (Kelz et al., 2010) in which potassium dominated with only minor amounts of sodium present (T. Brunner, personal communication, 2011). Boman et al. (2004) reported that alkali metals dominate the mass of inorganic particulate matter from the combustion of pelletised biomass fuels. Potassium and chlorine were reported to be the dominant elements. There is a lack of information on the relative use of open fireplaces and wood burning stoves in UK homes, or on the types of wood burning stoves most in use. However, there has been a considerable increase in the use of wood as a domestic fuel in recent years, accompanying the rapid rise in the prices of natural gas and electricity. There may, therefore, be a preponderance of modern designs offering a more efficient combustion and lower levoglucosan/K ratios. There has also been a significant uptake of larger wood-burning heating systems for schools and office buildings, etc.
b. The atmospheric lifetimes of levoglucosan and K ws may be different. Levoglucosan has a potential for atmospheric degradation (Hennigan et al., 2010;Hoffman et al., 2010), whereas K does not. The lifetime of levoglucosan with respect to reaction with a typical summer concentration of hydroxyl radical is estimated as 0.7-2.2 days (Hennigan et al., 2010). As winter concentrations of hydroxyl are lower than those in summer, such degradation will be appreciably slower in winter. This alone therefore seems unlikely to influence the levoglucosan/K ratio greatly, even if the majority of wood smoke is advected from distant sources. Appreciable decomposition of levoglucosan would lead to higher levoglucosan/K ratios in winter than summer, for which Table 1 gives no clear support, with little systematic difference between the seasons, except at Budbrooke, where summer ratios are higher for reasons explained earlier in this paper. Another possible influence is from differential lifetimes relative to deposition. It is likely that levoglucosan and K co-exist in the same particles, but with a range of ratios which may depend upon particle size. No information is available on the respective size-associations, but as the sampling was of the PM 2.5 fraction, it seems likely that the atmospheric lifetime will be several days and any differences between levoglucosan and K will be a minor influence upon the airborne ratios.
c. There is an unrecognised source of K. To be significant, this would need to be larger in winter than summer and to correlate with levoglucosan. While this possibility cannot be entirely excluded, the probability is thought to be low. Wang et al. (2011) have recently shown that the enhanced absorption between 370 nm and 880 nm in the two wavelength aethalometer (referred to as Delta-C) correlates with wood smoke markers, although they do not propose a factor with which to calculate wood smoke masses. Delta-C is calculated using the conventional absorption coefficient for black carbon at both wavelengths and taking the difference in estimated black carbon concentration. The gradient observed by Wang et al. (2011) in their winter data (giving the greatest range of concentrations) was Delta-C/levoglucosan = 5.9. In our data, the ratio Delta-C/levoglucosan at the Budbrooke site was 4.5 (r 2 = 0.25) and at the North Kensington site (winter data) was 6.5 (r 2 = 0.25). Wang et al. (2011) also showed a relationship to potassium although this was not corrected for sea salt or soil. Our data for both sites showed a very poor relationship to K ws . Figure 6 shows the black carbon measurements at the two wavelengths together with the calculated values of Delta-C in the Budbrooke and North Kensington winter datasets. These diurnal profiles in Delta-C with the strong evening maximum look extremely plausible but it needs to be borne in mind that Delta-C is a surrogate for wood smoke and is not a direct measurement of wood smoke mass. At Budbrooke, the gradient of the Delta-C/K ws plot was 4.0 (r 2 = 0.2) and at North Kensington (winter) the two variables were uncorrelated (r 2 = 0.04). Wang et al. (2011) found a Delta-C/K slope of 6.2 (r 2 = 0.63) in their winter dataset, and lower values in spring and fall, broadly consistent with the ratio measured at Budbrooke. The generally low correlations between Delta-C and K ws seen in our data, and the fact that these fall below the values of the levoglucosan-K ws correlations is suggestive of an interference (possibly by SOA) in the Delta-C method. Such an interference could be due to u.v. absorbing organic compounds, including species known as humic-like substances (HULIS) and would affect the multi-wavelength aethalometer data analysis, as well as the simpler Delta-C method.

Estimation of wood smoke concentrations
From single marker data, it is very difficult to estimate the airborne wood smoke concentrations. The factor of 10.7 between levoglucosan and wood smoke mass relating to Aus-trian conditions reported by Schmidl et al. (2008) has been widely adopted, but as indicated above, this may be an over-estimate for the entire combustion process. Our levoglucosan/K ratios are entirely out of line with Schmidl et al. (2008) and hence we think it inappropriate to use these data directly. The relationship between levoglucosan and K ws is considerably less close at our sites than was reported by Caseiro et al. (2009) from measurements in Vienna, Graz and Salzburg. It is also lower than that reported by Saarikoski et al. (2008) between monosaccharide anhydrides and potassium in four central and northern European cities. Additionally, the relationship between Delta-C and levoglucosan is weaker than that observed by Wang et al. (2011) despite the fact that their data were not corrected for potassium from other sources. This gives rise to the question as to whether more than one source type is influencing the measured data. At Budbrooke, this is clearly the case with influences both of wood stoves and open burning of wood which would be expected to give different K ws /levoglucosan relationships. However, the situation is less clear for the other sites.
A survey of consumption of solid fuel in the UK and Ireland (Ecosolidfuel, 2008) has suggested the following split: open fireplace 11.5 %; closed fireplace 22.1 %; pallet stove 0.1 %; domestic boiler 66.3 %. However, the energy consumption in this form of around 75 PJ yr −1 is far smaller than the 322 PJ yr −1 used in non-domestic boilers with outputs of up to around 500 KW (R. Stewart, personal communication, 2011). The latter are more likely to be used in cities, where in the UK smoke control regulations require combustion of solid fuels in devices demonstrated to meet specific emission requirements. It consequently seems likely that the UK urban emissions affecting sites such as EROS and London North Kensington comprise a mixture of wood burning in open fireplaces and closed devices which are likely to be relatively rich in levoglucosan relative to potassium, and on the other hand, more advanced domestic boilers and nondomestic boilers having a much lower levoglucosan to potassium ratio as exemplified by the modern devices tested by Kelz et al. (2010). Consequently, it is difficult to propose a specific factor between either levoglucosan or K ws concentrations and the mass of wood smoke particles. Puxbaum et al. (2007), based on a review of literature data available at the time, propose use of the relationship: Biomass smoke OC = levoglucosan × 7.35 According to the studies reviewed by Puxbaum et al. (2007), OC represents between 51.3 and 100 % of PM from biomass combustion. Using the data for fireplaces and wood stoves presented by Puxbaum et al. (2007), the relationship between biomass smoke mass and levoglucosan can be derived and the various studies reviewed there give a range of 4.51 to 24.6 with a mean value of 11.2. If this value is adopted and applied to the measured levoglucosan data, the following average concentrations are derived: Correlations at Marylebone Road were measured only in one season, when they were very similar to those at North Kensington.
The literature is far less clear on the relationships between K ws and biomass smoke mass. However, given the very high ratios between K ws and levoglucosan, we think it likely that there is a separate source category of highly efficient larger biomass burning installations generating emissions with a very low organic content which would have little influence upon airborne levoglucosan. Judging from the work of Kelz et al. (2010), it is quite feasible for K ws to make up around 50 % of the mass of such aerosol which would imply an additional contribution of biomass burning to PM 2.5 at the urban locations with very low levoglucosan/K ws ratios as follows: If there are two distinct source types, i.e. on the one hand wood burning in relatively inefficient small appliances such as wood stoves and fireplaces, and on the other hand combustion in larger more efficient installations, then the two sets of concentrations should be approximately additive for a given location.

Comparison with earlier UK data
There are two sets of earlier data for the Birmingham, EROS site with which it is relevant to intercompare. Yin et al. (2010), using a Chemical Mass Balance (CMB) model, calculated source contribution estimates for wood smoke as a contributor to PM 2.5 as 0.07 µg m −3 in summer and 0.08 µg m −3 in winter in 2007/2008. The CMB profile used had a levoglucosan/OC ratio of 0.0292 but the final result implied a much larger ratio of levoglucosan to woodsmoke mass. In the light of our current data, the lack of an obvious seasonal variation is surprising but the sampling regime involved sampling for only five days of each month as opposed to daily in the current work and it is possible that the concentrations measured were not wholly representative of the overall period of sampling. Nonetheless, the two datasets are not strongly out of line in terms of the magnitude of concentration. More recent unpublished data also using the CMB method have shown higher concentrations at North Kensington of 0.23 µg m −3 (summer campaign) and 0.36 µg m −3 (winter campaign) which are far closer to the current dataset.
Further indirect evidence is available from the study of Heal et al. (2011) in which 14 C was measured in aerosol collected at the EROS site in 2007/08. In all, some 75 PM 2.5 samples were collected between June to September 2007 and January to May 2008 and a subset of 26 samples were analysed for their "fraction of modern carbon". The contemporary elemental carbon is interpreted as arising from biomass combustion and this again showed rather little seasonal variation which could once again be an artefact of the days selected. The average contemporary elemental carbon was only 2 % of total carbon equating to around 0.1 µg m −3 . If this is assumed to be 20 % of biomass smoke (Schmidl et al., 2008), this equates to a concentration of 0.5 µg m −3 of biomass smoke. The figure of EC being 20 % of biomass smoke is highly uncertain -for example Minguillón et al. (2011) assume a ratio of EC/OC = 0.3 in biomass smoke which implies an EC content of CM of below 14 %. The value of 0.5 µg m −3 of biomass smoke, although very uncertain due to the difficulty of achieving an accurate split between elemental and organic carbon in the samples and the uncertainty surrounding the %EC in biomass smoke, is not strongly out of line with the estimate for EROS derived from levoglucosan analysis above.
There are only two other UK datasets of which we are aware. One is from Fuller et al. (2011) who reported mean winter levoglucosan in London of 176 ng m −3 , almost four times the mean concentrations measured in our study in winter at North Kensington. They equate this to about 3 µg m −3 of wood smoke, which would cause a marked seasonal variation in PM 10 which is not observed. An analysis of data from North Kensington for 2001-2008 by Bigi and Harrison (2010) shows no seasonal influence upon PM 10 concentrations, despite a fall in traffic-generated gases in the summer months (CO and NO x ). The main components of PM 10 in London are nitrates, sulphates, elemental carbon and organic compounds, sodium chloride, calcium-rich dust and iron-rich (traffic associated) dust (Harrison et al., 2004). The elemental and primary organic compounds are largely trafficrelated and hence are expected to exhibit a higher concentration in winter, due to poorer dispersion, as for CO and NO x . Sodium chloride is also elevated in winter. Sulphate shows no seasonal trend in the UK, and nitrate is lower in summer due to ammonium nitrate vaporisation (R. M. Harrison and A. Charron, unpublished data). It therefore appears that in summer a lower primary carbonaceous emission component from traffic and lower sea salt component is compensated by a higher coarse particle component from the calcium and iron-rich dusts. However, a preliminary analysis of one year of PM 2.5 data for the UK (Harrison et al., 2012) does show a marked seasonality in PM 2.5 with higher winter concentrations. The extent to which this can be accounted for by a higher traffic contribution in winter and reduced nitrate in summer is yet unclear, and calls for more chemically speciated measurements. A second recently published dataset includes sites in Birmingham, although the sampling date is unspecified (Caseiro and Oliveira, 2012). Average fine fraction levoglucosan concentrations at a background site were 33 ng m −3 (winter) and 15 ng m −3 (summer) very close to our mean levels for EROS (31 and 10 ng m −3 , respectively).
Perhaps the most persuasive data are from Birmingham (where EROS is located), collected by Harrison and Yin (2008) who found that the OC/EC ratio for the urban increment (i.e. OC, urban minus rural/ EC, urban minus rural) was very close to that for the roadside traffic increment (i.e. OC, roadside minus urban/ EC, roadside minus urban), where the urban site was at a central urban background location. This demonstrates that carbonaceous aerosol emitted within the urban area had a composition close to that of traffic emissions and was not appreciably influenced by wood smoke, which has a much higher OC/EC ratio (close to 5, as opposed to 0.35 for traffic emissions).
Data from France provide an interesting comparator. Favez et al. (2009) sampling with an aethalometer in Paris concluded that carbonaceous aerosols originating from wood burning represented about 20 % of PM 2.5 over the winter period that was sampled. Absolute masses were not reported. Sciare et al. (2011) sampled at a suburban site 20 km southwest of the city of Paris over a 10-day winter period. Results from an aethalometer indicated mean concentrations of organic aerosol of 2.26 µg m −3 from wood burning, 0.23 µg m −3 from fossil fuel and 2.65 µg m −3 residual component from the aethalometer model, attributed to sec-ondary organic aerosol. In a study conducted in the Alpine city of Grenoble (France), Favez et al. (2010) compared CMB, aethalometer and AMS-PMF model results for wood burning organic aerosol, finding rather divergent estimate of 68 %, 61 % and 37 % of total organic aerosol from the three methods, respectively. Herich et al. (2011), sampling in Switzerland used the aethalometer to apportion black carbon to fossil fuel and wood-burning sources, but declined to apportion the organic matter (and hence wood smoke mass) because of the high standard errors shown in the data analysis, and the sensitivity of C 1 and C 2 to the chosen alpha values. In our work, the value of R 2 for the estimation of C 2 and C 3 in the aethalometer method was 0.43, also indicative of appreciable scatter. We therefore believe that the multi-wavelength aethalometer method is subject to substantial random errors due to treating C 3 as a constant (which manifestly it is not) and possibly also to systematic errors due to the likelihood of the presence of other UV-absorbing components. Coal smoke was considered as a contributor (Bond et al., 2002) but, like wood smoke, is unlikely to have major sources in London.

Conclusions
This work is suggesting that annual mean concentrations of biomass smoke at UK sites are very low at well below 1 µg m −3 and very much lower than some other estimates for the UK (Fuller et al., 2011). It is recognised that these concentrations are far below those measured elsewhere in Europe (e.g. Puxbaum et al., 2007;Bari et al., 2010) but there may be good reasons for this. Most modern urban dwellings in the UK no longer have fireplaces and the fitting of wood stoves is very difficult due to the lack of a flue and chimney. Most UK urban areas are subject to smoke control orders which prohibit the use of many kinds of solid fuel other than in devices designed to burn them smokelessly. There are very limited supplies of logs and wood products available within urban areas for purchase by domestic consumers for combustion. Consequently, burning of wood in domestic premises is largely limited to rural areas of the UK and the fact that levoglucosan concentrations correlate so strongly between the Marylebone Road and North Kensington sites may imply an advected rural source of wood smoke. While the datasets from Birmingham, EROS and London, North Kensington show a marked seasonal variation, the estimated wood smoke masses are insufficient to cause a significant perturbation in PM 2.5 or PM 10 concentrations such as would lead to a distinct seasonal pattern. Such a pattern in PM 10 is not observed at UK urban sites (see for example Bigi and Harrison (2010) for London, North Kensington), although it is for PM 2.5 with other probable explanations. The observation by Harrison and Yin (2008) that the urban increment in OC/EC above rural is consistent with the composition of traffic emissions also suggests a modest wood smoke contribution deriving from within the city.

R. M. Harrison et al.: Comparison of methods for evaluation of wood smoke
The multi-wavelength aethalometer calculations suggest higher concentrations but we discount these on the basis of their inconsistency with the levoglucosan data and the fact that the aethalometer method when applied to UK locations will be subject to interference from HULIS (humic-like substances) within secondary organic aerosol. Since this interference is variable, there is no correction for it and therefore the method may strongly over-estimate concentrations. The fact that the relationship between Delta-C and levoglucosan in our measurements is similar to that obtained by Wang et al. (2011) in the United States suggests that we can place trust in the levoglucosan data and ignore the more extreme concentration estimates deriving from application of the method of Sandradewi et al. (2008a, b) which was based originally upon data collected in a Swiss Valley where wood smoke and traffic emissions were expected to be the only significant sources of absorbing aerosol.