The North Kensington site is located within the grounds of a school in a residential area (51∘31′15′′ N, 0∘12′49′′ W) and is classified as urban background. Central London lies 7 km to the east and a major road is located approximately 500 m to the south of the site. A more complete description of the site and air pollution climate is available in Bigi and Harrison (2010); the site is considered representative of the background air quality within much of London. This site is part of the Automatic Urban and Rural Network (AURN), with further details on the instrumentation and sampling procedures available in Sect. 2.2.

This is a rural site located in the Detling show grounds, approximately 45 km south-east of central London (51∘18′7′′ N, 0∘35′22′′ E) and is bordered by fields. To the south of the site, approximately 150 m, lies a busy road; the closest village (Detling) is to the south-west (approximately 1.5 km). The winter campaign site, approximately 800 m to the south of the AURN site, was not co-located with the long-term AURN monitoring site (Mohr et al., 2013).

The second rural site was located approximately 80 km west of central London within the Harwell Science Centre (51∘34′16′′ N, 1∘19′31′′ W) and is part of the AURN. The closest village, Chilton, and the nearest main road (A34) are about 2 km to the east of the site. The site is surrounded by fields and is considered as representative of a rural location (Charron et al., 2007).

The elevated site utilised in this study was the BT tower (51∘31′17.31′′ N; 0∘8′20.12′′ W) located at 60 Cleveland St. in London. The main platform for the aethalometer was a well-ventilated balcony (level T35, 160 m above ground level), and the gas monitors were sampled from a height of approximately 180 m above ground level (CO and NOx). Further description of the BT site can be found in Harrison et al. (2012b).

During the winter campaign, filter samples were collected for the chemical analysis of selected biomass burning tracers. Due to various logistical reasons, the sampling regimes for the different filter and instrumental analysis utilised in this paper were not all run at the same time or over the same sampling interval and the filter sampling and analysis methods are summarised in Table 1. The levoglucosan sampling was similar at NK and Harwell but differed at Detling in terms of the size fraction collected and sampling start time (Table 1). A Digitel DHA-80 sampler was used at the NK and Harwell sites to collect PM2.5 particles on 150 mm diameter quartz fibre filter circles over a 24 h sampling period. These filters were analysed for levoglucosan, total OC, EC, K+ and 14C (the last only applies for NK). At Detling aerosols were collected on pre-baked quartz fibre filters (PALL 2500QAT-UP Tissuquartz) using a high-volume sampler (CAV-A/MSb) with a PM1 inlet, a flow rate of 500 L min-1 and a sampling time of 24 h. These filters were analysed for EC, OC and 14C in the total carbon (TC) fraction. Separate filter samples with a PM10 inlet were collected for levoglucosan analysis.

The different size fraction analysed for levoglucosan at Detling compared to the other sites is unlikely to affect the results as most of the particle-associated levoglucosan is generally within the PM2.5 (Saarikoski et al., 2008; Wagener et al., 2012; Yttri et al., 2005). Potassium samples were collected at Detling as a part of this project but failed quality control and therefore were not included in the current work.

In addition to the filter samples collected, measurements of BC were conducted using a Magee Scientific seven-wavelength aethalometer (model AE-31) at NK and Detling during the winter campaign. At NK, the seven-wavelength aethalometer sampled continuously with a 5 min sampling interval, while at Detling the 7W AE sampled at a different sampling interval of 2 min. Furthermore, at NK the 7W AE also sampled during the ClearfLo summer IOP (21 July 2012–23 August 2012) with a 5 min sampling interval.

As Harwell, NK and Detling are part of the AURN, BC and PM2.5 concentrations were obtained from instruments in this network for the winter campaign. PM2.5 mass concentrations were determined by the Filter Dynamic Measurement System method at an hourly interval and were subjected to the AURN network quality control procedures (Ricardo-AEA, 2013). In addition to the 7W AE, data from a two-wavelength aethalometer (Magee Scientific, model AE 22) were also collected at 15 min intervals at each AURN site.

At the BT tower, measurements of BC were performed by a 7W AE (Magee Scientific, model AE-30), while concentrations of NO and NO2 were analysed by chemiluminescence (Ecophysics, model 780TR fitted with photolytic blue light converter (BLC) for NO2), all at a 5 min sampling interval. Calibration was carried out using a 5 ppm NO in nitrogen gas standard, diluted to 50 ppb in NOx free air. Calibration of the BLC conversion efficiency was done using gas phase titration of the NO calibration gas with ozone. An Aero-Laser fast carbon monoxide analyser (model AL5002) at a 10 Hz sampling rate was used at the BT tower to measure the concentration of CO. As 7W AE data were only available for limited periods at NK compared to BT tower, two-wavelength aethalometer data from the AURN network were also obtained at NK, along with concentrations of CO, NO and NO2 from 1 January 2012 to 30 June 2013.

Meteorological data were obtained from the Met Office for three sites that were judged to be the most suitable based upon relative location to the sampling sites. For NK and BT tower, wind data from Heathrow airport station were used as this site will not be affected by large buildings. The Met Office stations Benson and Gravesend were used for Harwell and Detling respectively, which were the closest stations geographically with wind data. Benson is approximately 16 km to the east of Harwell and Gravesend about 20 km north-west of Detling. The meteorological data obtained from the Met Office had a 1 h time resolution. The approximate times of sunrise and sunset throughout the winter campaign were 07:40 and 16:45 UT (and local time).

During the ClearfLo summer and winter IOPs, a scanning Doppler lidar (Halo Photonics Streamline) was used for monitoring the boundary layer depth and structure, with the locations shown in Fig. 1. The lidar operates at 1.5 µm wavelength, integrated signals are outputted every 3.6 s and the vertical resolution is 30 m. For the first part of the winter IOP period (6–11 January 2012) the lidar was operating on the rooftop of the Westminster City Council's building (WCC, 18 m a.g.l., 51∘31′17′′ N, 0∘09′40′′ W) before it was moved to the engineering building at Imperial College London (ICL, 33 m a.g.l., 51∘29′55′′ N, 0∘10′29′′ W), where it was operating until 8 February 2012. For the summer IOP period an identical lidar was operating at NK (ground level, 51∘31′15′′ N, 00∘12′49′′ W) with a gate resolution of 18 m. The mixing height was detected as the level where the vertical velocity variance measured with the lidar drops below a threshold value, following the method described in Barlow et al. (2014) and originally proposed by Barlow et al. (2011). The minimum observable height is different for each location and depends on the lidar gate and the height that the instrument is placed (above ground level) at each site. Thus it is 124 m for WCC, 139 m for ICL and 63 m for NK; all heights are measured above ground level. Long-term meteorological measurements were conducted at the top of BT tower (190.8 m a.g.l.), employing an instrumentation platform equipped with an eddy covariance system (Gill Instruments R3-50 sonic anemometer, and Licor Li-7500 Hygrometer), a net radiometer (CNR4, Kipp & Zonen) and a weather station (Vaisala WXT520).

The analytical method applied is similar to that described in Yin et al. (2010) and Wagener et al. (2012) but modified for the detection of levoglucosan. In brief, one-quarter of the quartz fibre filter sample was spiked with an internal standard (IS), methyl beta-D-xylopyranoside (Sigma-Aldrich Ltd) and extracted with dichloromethane and methanol under mild sonication at room temperature. The combined extracts were filtered and concentrated down to 50 µL. One aliquot of the extract was evaporated until nearly dry, derivatized by the addition of N,O-bis(trimethylsilyl)trifluoroacetamide plus 1 % trimethylchlorosilane (BSTFA + 1 % TMCS) and pyridine at 70 ∘C for 1 h and finally cooled in a desiccator. The quantification was based on the IS and a six-point authentic standard calibration curve, with selected ion monitoring mode on an Agilent GC-MS instrument. The ions monitored were m/z 204 and 217 for the IS and m/z 204, 217 and 333 for levoglucosan.

The quartz fiber filter samples from NK, Harwell and Detling were also analysed for OC and EC by a Sunset thermal–optical carbon analyser (Sunset Laboratory Inc.). The protocol used was the EUSAAR 2 protocol (Cavalli et al., 2010), in which organic carbon is removed during an initial non-oxidising temperature ramp from 200 to 650 ∘C under a helium atmosphere, and then a second temperature ramp from 500 to 850 ∘C is initialised with the carrier gas switched to a helium/oxygen mixture for removing elemental carbon and organic carbon pyrolysis products.

Sample preparation for the 14C analysis was performed with two different techniques. First, TC was isolated before the 14C analysis using the THEODORE system (see Szidat et al., 2004, for more details). In brief, TC is oxidised to CO2 from filter punches with at 650∘C for 12 min in an oxygen stream. The evolving CO2 is separated from interfering reaction gases, cryo-trapped and sealed in glass ampoules for 14C measurements. Second, samples were also prepared following the approach described in Fahrni et al., 2010). Several small filter pieces are put in quartz tubes together with ∼ 0.25 g of copper oxide (small rods for elemental analysis, Fluka, Switzerland), evacuated, sealed and combusted for 4 h at 800∘C in a muffle furnace (model LE 14/11/B150, Nabertherm, Germany).

The 14C measurements were carried out using the MIni radio CArbon DAting System (MICADAS; Synal et al., 2007) at the Swiss Federal Institute of Technology (ETH) Zürich using a gas ion source (Ruff et al., 2007; Wacker et al., 2013), allowing the 14C analysis of the gaseous CO2 without prior graphitisation (Ruff et al., 2010). Results from the 14C measurement are expressed as fraction of modern (fM), denoting the ratio of the 14C / 12C content of the samples related to the 14C / 12C ratio in 1950, defined as 0.95 times the value of the standard SRM4990B (Stuiver and Polach, 1977). The fM values were corrected for δ13C fractionation (Wacker et al., 2010) and for the 14C decay between 1950 and the year of measurement. Furthermore, a mass depending blank correction was applied following an isotopic mass balance as presented by Zapf et al. (2013). The obtained blank mass and fM of TC were 0.53 ± 0.18 and 0.50 ± 0.17 µg cm-2 respectively. To account for the excess 14C from the thermonuclear weapon tests in the late 1950s and early 1960s, the blank corrected fM values were converted into non-fossil fractions using a reference value (fNF,ref=1.082±0.04) representing the modern 14C content of biogenic and biomass burning aerosols during the sampling period compared to 1950 before the bomb testing (see e.g. Zotter et. al., 2014a).

Each filter of PM2.5 was equally divided. The percent modern carbon (fM) in TC was derived from one-half of the filter sample through the following two-step combustion, similar to that described in Heal et al. (2011) and using the same quality assurance procedures. Filters were first combusted in high purity O2 at 375 ∘C for 1 h and then combusted in high purity O2 at 800 ∘C for 4 h, with collection of the evolved CO2 from each combustion. In both cases the CO2 was subject to offline clean up, graphitisation and AMS determination of fM. The fM (TC) was calculated by mass balance of the two subsamples. This was converted to fraction contemporary carbon by dividing by a value that assumes an average age for the material comprising the contemporary carbon. Here a divisor of 1.08 was used, as discussed in Heal et al. (2011).

Small pieces of the quartz filters collected at Harwell and NK were extracted with distilled deionised water (5 mL) by undergoing 40 min mechanical agitation. The resulting solutions were filtered with 32 mm Acrodisc syringe filter (0.2 µm pore size) and analysed for K+ with a Dionex DX500 system, which uses a GP-40 gradient pump, but run with isocratic elution using 15.4 mM MSA (CH3SO3H) as eluent. The sample concentrations were calibrated with a series of mixed standards of known concentration (0.01–10 ppm).

A number of analytical methods were employed for apportioning the contribution of fossil fuel and biomass burning emissions to the EC and OC fractions. All methods assume that these two sources, biomass burning and fossil fuel emissions, are the sole sources of either EC or BC. The first method, referred to as the tracer method throughout this paper, is described in detail in Gelencsér et al. (2007). In this method, the primary OC concentration from biomass burning (OCbb) is estimated based on the concentration of levoglucosan and the ratio of levoglucosan to OC measured for biomass burning emissions from the literature. The EC concentration from biomass burning (ECbb) is calculated from the OCbb using previously observed OC / EC ratio for biomass burning emissions and the remainder of EC is apportioned to EC from fossil fuel emissions (ECff). Primary OC from fossil fuel emissions (OCff) is estimated using previously observed OC / EC ratios for vehicle emissions. The remaining OC is from biogenic, cooking (both primary and secondary) or other secondary sources that can include secondary organic aerosols formed from fossil fuel combustion; this fraction is referred to as OCSOA. One limitation of the tracer method is the assumption of constant source ratios, which may not have been the case throughout the current work. Furthermore, the value of these source ratios for vehicle emissions and biomass burning (both levoglucosan / OC and OC / EC) are dependent on the combustion conditions and fuel type. As a consequence, a wide range of ratios have been previously reported in the literature for vehicle emissions (e.g. El Haddad et al., 2013; Gelencsér et al., 2007; Pio et al., 2011) and for biomass burning (e.g. Gelencsér et al., 2007; Schmidl et al., 2008), and thus there is uncertainty regarding the correct source ratios for emissions at the sites. In this study, we have initially applied the average source ratios for Europe as reported in Gelencsér et al. (2007). As a result of the uncertainty regarding the correct source ratios for biomass burning and vehicle emissions, a sensitivity study was carried out using a range of source ratios from the literature to assess the variability in the source apportionment.

Further source apportionment of the total carbon (sum of EC and OC concentration) was carried out using the radioisotope of carbon (14C) concentrations as described in Heal at al. (2011). In this method, the TC is apportioned based on whether carbon had a fossil (TCff) or contemporary (non-fossil, TCnf) origin and will be referred to as the 14C method throughout this paper. For comparison to the tracer method, the sum of ECff and OCff is considered equal to the TCff while the TCnf is equivalent to the sum of OCSOA, OCbb and ECbb. This approach ignores the fact that some secondary OC is of fossil origin, although most published work indicates that the larger proportion is from biogenic (non-fossil) precursors. Secondary OC of fossil origin has been estimated to be 25 ± 13 % of secondary OC levels for several European cities in summer (Zotter et al., 2014a, and references therein). Recent results in the UK found that fossil EC and fossil OC comprised 27 and 20 % of TC respectively (Heal et al., 2011). Using the average primary vehicle emissions OC / EC ratio of 0.58 (Gelencsér et al., 2007) the fraction of fossil OC from primary vehicle emissions was 15.7 % of TC. This implies that fossil secondary OC was only around 4 % of TC and therefore any differences in the calculated TCnf, assuming that the OCSOA was totally non-fossil, would be within the uncertainties associated with the tracer method. As part of the sensitivity study reported in Sect. 3.3.1, inclusion of a 20 % fossil fraction of secondary OC generated results which were still consistent with the 14C data and within the variability of individual days. Furthermore, recent work by Charron et al. (2013) suggests a large biogenic contribution to SOA at southern UK sites.

The seven-wavelength aethalometer provides a real-time optical measurement of the light absorption (babs) with seven different wavelengths (370, 470, 520, 590, 660, 880 and 950 nm) from which the equivalent black carbon concentration can be deduced (Petzold et al., 2013). Aethalometers suffer from various systematic errors e.g. multiple scattering by the filter fibres (scattering effect) and increased absorption by light-absorbing particles accumulating in the filter (shadowing or loading effect) (Bond et al., 1999; Liousse et al., 1993; Petzold et al., 1997). BC concentrations from the 2W AE and 7W AE were corrected for loading and scattering effects according to the procedure described in Weingartner et al. (2003). In the Weingartner correction a value of 1.2 was used for the loading correction, and the multiple scattering correction constant (C) was determined based on comparison to results from a multi-angle absorption photometer (MAAP) at 630 nm, as described in Collaud Coen et al. (2010).The results from this comparison determined a C value of 3.336 and 3.095 for NK and Detling respectively. As there were no MAAP data available at Harwell or the BT tower, the average of the C values determined at NK and Detling were applied to all 2W and 7W AE data to enable comparison between the sites. The mass absorption cross section, which is needed to convert the light absorption coefficient to BC, was determined by plotting EC concentration (from thermal–optical analysis) against the corrected 880 nm babs at Detling, NK and Harwell. As there were no EC data at BT tower, the mass absorption cross section determined for NK was applied to the BT tower 7W AE data. The 2W AE mass absorption cross section values were calculated to be 7.3, 8.1 and 7.3 m2 g-1 for Harwell, NK and Detling respectively while a mass absorption cross section of 7.4 m2 g-1 was found for the 7W AE at both NK and Detling.

Measurements from the 7W AE data were apportioned into the contribution from wood burning (BCwb) and traffic (BCtr) according to Sandradewi et al. (2008a, b), referred to as the aethalometer model. Aethalometer absorption values at 470 and 950 nm were used to determine BCwb and BCtr respectively, and an absorption Ångstrom exponent of 2 for wood smoke (αwb) and 1 (αtr) for traffic was used. As the UK AURN monitoring network employs 2W AEs that measure at 370 and 880 nm, UVPM (also referred to as delta-C) concentrations were also determined as UVPM has been proposed as a marker for wood smoke. UVPM concentration was calculated for both the 2W and 7W AE data by the difference of BC equivalent concentrations at 370 and 880 nm (Wang et al., 2011a, b). Furthermore for 7W AE data sets at the BT tower and NK, α for the total ambient absorbing particles was calculated using the 470 and 950 nm wavelength data according to the procedure described in Harrison et al. (2012a).

Using the tracer method, contributions of biomass burning (ECbb and OCbb) and fossil fuel (ECff and OCff) to the EC and OC fractions were estimated first using the average of source ratios reported for Europe in Gelencsér et al. (2007). These applied source ratios for biomass burning were 0.136 for the ratio of levoglucosan to OC and an OC / EC ratio of 6, while for vehicle emissions an OC / EC ratio of 0.58 was used. This average OC / EC ratio for vehicle emissions is notably similar to that recently determined for vehicle emissions in London (0.63) (Pant et al., 2014). The average estimated concentrations using these source ratios are shown in Fig. 4 for the three sites, with the results of the sensitivity study discussed later in this section. As expected, the concentrations of ECff and OCff were higher at NK (p < 0.05), as it is an urban site with higher traffic influence compared to the rural sites. At all three sites, ECff accounted for the majority of the EC fraction, indicating that the main source of EC was traffic emissions. Overall, the OCbb and ECbb concentrations were similar across the sites, but OCbb accounted for a higher average percentage of the total OC: 30 ± 16 and 23 ± 17 % at Harwell and Detling respectively compared to NK (16 ± 7 %). At all three sites according to the tracer method, approximately 40 % of the OC fraction was accounted for by the sum of biomass burning and fossil fuel emissions (Fig. 4). The remaining fraction, OCSOA, which includes both primary biogenic, cooking and secondary organic aerosols therefore, exceeds the sum of primary emitted biomass burning and vehicle emissions.

A more direct analysis for apportioning the fossil and non-fossil fractions of TC is the 14C method, which was applied to samples from NK and Detling, with the results given in Table 3. At NK, the percentage of TCff was higher compared to that observed at Detling (Table 3, with the daily TCnf fraction shown in Fig. S5, in the Supplement), showing the greater influence of traffic emissions at NK. As shown in Table 3, a very good agreement was observed between the percentage of contemporary (TCnf) and fossil (TCff) carbon as determined by both the tracer and 14C method, indicating that both methods were consistent in the TC apportionment. The high proportions of TCnf in PM2.5 in the range ∼ 50–70 % derived in this work, even for urban background (Table 3), are consistent with findings from elsewhere in Europe and worldwide (Heal, 2014; Hodzic et al., 2010).

To examine whether the ratios used to apportion the TCnf and TCff were reasonable, a sensitivity study was carried out using a range of source ratios for biomass burning and vehicle emissions from the literature. Gelencsér et al. (2007) proposed minimum and maximum source ratios based on the ranges in the published data and these have been employed to estimate the uncertainty for the apportionment between contemporary and fossil carbon by the tracer method. It is worth noting that Minguillón et al. (2011) used a similar range of levoglucosan / OC ratios to assess the uncertainty of their source apportionment. Pio et al. (2011) examined PM2.5 OC / EC ratios from vehicle emissions and concluded that a range of 0.3–0.4 was typical of fresh vehicle emissions, and the lower value has also been applied to assess the uncertainty in the apportionment of vehicle emissions. Harrison et al. (2012a) recently proposed a mean levoglucosan / OC ratio of 0.09 for wood burning based on the literature, which has also been applied in this sensitivity study. Ambient ratios for biomass burning were recently determined using the non-fossil fraction of OC and EC during winter in Switzerland (Zotter et al., 2014b), and the average ratios (0.098 and 8.2 for levoglucosan / OC and OC / EC respectively) have also been included in the sensitivity analysis.

The results of this sensitivity study are presented in Table 4, along with the values for the ratios used to apportion the contributions of fossil and non-fossil sources. The calculated ECff, ECbb, OCbb, OCff and OCSOA concentrations for different ratios can be found in the Supplement, Table S1, while the results in Table 4 are shown in graphical form in Fig. S6. Overall, comparison between the calculated percentage TCnf and TCff in Table 4 exhibit a wide range of values and highlights the importance of selecting the correct source ratio. The closest match to the 14C results for TCnf and TCff was found using the mean levoglucosan / OC ratio proposed by Harrison et al. (2012a) (Table 4, E), the average Swiss ambient biomass burning ratios (Table 4, G) and the average source ratios from Gelencsér et al. (2007) (Table 3). Therefore, it appears that with the application of average source ratios from the literature the split between contemporary and fossil carbon was reasonably estimated, though these ratios may not always accurately represent the sources (Pio et al., 2011). Inclusion of fossil secondary OC, not considered in Table 4, will also give alternative possible fits to the data.

Diurnal cycles of the BC concentrations at Harwell, NK and Detling during the winter campaign measured by 2W AEs are shown in Fig. 5, with the cycles for the 7W AE given in Fig. S7, in the Supplement. Similar cycles and concentrations were observed by the 2W AE and 7W AE at NK while at Detling the diurnal cycles were slightly different, which may have been due the instruments not being co-located. The classic bimodal cycle, coinciding with peak traffic times, was observed at both sites, indicating that traffic was the likely main source of BC at NK and Detling. At Harwell the diurnal cycle of BC was also bimodal with a similar morning peak, but the peak in concentration in the evening was observed at a later time (20:00) more associated with domestic heating. This difference in the diurnal cycles of BC suggests that local sources of biomass burning may be more influential at Harwell than the other two sites.

UVPM, defined as the difference in aethalometer-determined concentrations for the 370 and 880 nm wavelength, has been proposed as a marker for biomass burning (Wang et al., 2011a). Hence the diurnal cycles of UVPM for both the 2W AE and 7W AE were calculated and are shown in Fig. 6. All of the five cycles presented in Fig. 6 are broadly consistent with the diurnal cycles expected for wood smoke from domestic heating, with a peak in concentration in the evening. At Detling, the observed UVPM cycle is similar to the observed diurnal cycle for other biomass burning markers at Detling, including a biomass burning (BBOA) factor derived from Positive Matrix Factorization (PMF) analysis of Aerosol Mass Spectrometer data (Mohr et al., 2013). The timing of the evening concentration peak varied between the sites, with the observed concentration peak occurring earlier at the rural sites (18:00–19:00 and at 20:00 at Detling and Harwell respectively) while at NK the peak appeared later (22:00). This may point to differing source influences: the earlier peak points to more local emissions while the observed later peak at NK and to a lesser extent at the rural sites is suggestive of a regional influence. In addition to the evening concentration peak observed as expected in the UVPM cycle at NK for both instruments, there was also a peak at around 07:00, suggesting a traffic influence in the observed UVPM concentrations at NK. Furthermore, the UVPM and levoglucosan concentrations were found to be correlated at all sites (r2 between 0.8 and 0.9). Overall, based on the diurnal cycles for UVPM presented in Fig. 6, it suggests that UVPM may be a useful qualitative marker for wood smoke in areas where traffic emissions are not dominant, in agreement with Harrison et al. (2013).

Another approach to apportioning the contributions of wood smoke and traffic emission to the observed BC concentrations is the two-component aethalometer model (Sandradewi et al., 2008a, b), which was applied to the 7W AE data. The concentrations of BCtr at both sites were well correlated with NOx (r2 of 0.92 and 0.65 for NK and Detling respectively) and demonstrated higher correlations than between BCwb and NOx (r2 of 0.57 and 0.45 for NK and Detling respectively). At Detling, the concentrations of BCwb recorded a higher correlation with levoglucosan concentrations (r2 of 0.69) compared to BCtr (r2 of 0.41), as expected. A similar trend was also observed at NK, with BCwb demonstrating a higher correlation with levoglucosan (r2 of 0.95) than BCtr (r2 of 0.80). The high correlation observed between BCtr and levoglucosan at NK is surprising and may be due to dominance of traffic emissions in London.

Notable differences were also observed in the diurnal cycles of the BCwb and BCtr concentrations at NK and Detling as shown in Fig. 7. The diurnal cycles were as expected for BCtr at both sites, exhibiting a bimodal cycle as for traffic emissions. However, at Detling the BCtr peak in the morning is broader and the evening peak earlier than would normally be expected for a local traffic source; however the NOx diurnal cycle also demonstrated an evening peak at the same time. Differences were observed in the BCwb diurnal cycles at Detling and NK (Fig. 7). At Detling, a peak in concentration centred on 20:00 was observed, which is typical for emissions from domestic heating. Meanwhile, the flat diurnal cycle in BCwb concentrations observed at NK is suggestive of regional rather than local sources. Overall, the estimated concentrations of wood smoke differed between the methods and these differences are examined in the next section.

Overall as shown in Fig. 4 and Table 5, the main source of both BC and EC was traffic emissions as opposed to biomass burning; however, the relative contributions of each source varied depending on the method employed and applied source ratio. ECbb accounted for on average 24 ± 16, 7 ± 2 and 14 ± 12 % of the total EC at Harwell, NK and Detling respectively. For the organic fraction, the relative contribution of biomass burning also varied and was generally higher than that from fossil fuel emissions at the rural sites but not at NK (Fig. 4). Using the average conversion factor for levoglucosan to wood smoke particle mass of 11.2 adopted by Harrison et al. (2012a), the mean concentrations of wood smoke mass were calculated to be 1.0, 0.87 and 0.78 µg m-3 for Harwell, NK and Detling respectively. From the tracer method, the total biomass burning contribution, taken as the sum of the ECbb and OMbb (OCbb converted to organic matter by conversion factor of 1.6; Aiken et al., 2008), was 1.2, 1.0 and 0.88 µg m-3 for Harwell, NK and Detling respectively. The wood smoke mass concentrations as determined by both the levoglucosan conversion factor and the tracer method were notably similar.

The two-component aethalometer model determined the contribution of wood burning to be an average of 15 ± 12 and 30 ± 13 % of the BC at NK and Detling respectively. Fuller et al. (2014) found that during December and January of 2009–2011, wood burning accounted for an average of 23 % of BC at NK, higher than the average found in the current work at NK though within the associated variability. The difference is probably attributable to interannual variations in concentration. In this study, the BCwb followed the expected trend of higher contributions at the rural compared to the urban site, as also observed for ECbb. However, the percentage contributions determined by the aethalometer model were higher than that observed for the tracer model for ECbb (Table 5). The assumptions in the tracer method and aethalometer model are constant source composition and α respectively. However, the α for wood burning is highly variable, ranging at least from 1.4 to 2.2 and has been shown to be dependent on variables such as the fuel and amount of atmospheric aging (Saleh et al., 2013; Kirchstetter et al., 2004). Therefore the applied α for wood burning may not accurately represent the biomass burning aerosols across the sites in this study, possibly resulting in higher calculated contributions. Similarly, the source composition for biomass burning (OC / ECbb and levoglucosan / OC) is also variable and has been shown to have wide range, which influences the calculated ECbb percentage contributions, ranging from 3 to 32 and 6 to 50 % of the total EC at NK and Detling respectively (Tables 5 and S1). From Table 5, the closest match for ECbb to the BCwb percentage contributions was found using the mean levoglucosan / OC and OC / ECveh and minimum OC / ECbb (Table 4, F). However, this combination of source ratios did not match 14C source apportionment results as well as other combinations (see Table 4). There were no combinations in the sensitivity analysis in Table 4 that were able to closely match both the 14C and aethalometer model source apportionment and this may be due to the presence of a small fraction of fossil secondary OC. Just as for αwb, these source ratios may not have been selected appropriately for the study and thus the differences between the aethalometer model and tracer method may be due to the uncertainties in the applied source ratios.

There have been a number of diagnostic ratios utilised to differentiate between the sources of biomass burning including the ratio of levoglucosan to both OC and K+ (Harrison et al., 2012a; Minguillón et al., 2011; Reche et al., 2012). While there are a number of sources of OC in an urban environment in addition to biomass burning (e.g. vehicle emissions and cooking), observed correlations between levoglucosan and OC in wood smoke for different fuels suggest that its ratio can be used to determine the source (Sullivan et al., 2008). Average values for the diagnostic ratios, levoglucosan / OC and levoglucosan / K+ at each site are given in Table 6. At the three sites the average ratio of levoglucosan / OC was similar and is comparable to the characteristic ratio determined for long-range transport of biomass burning in Spain (0.05 ± 0.01) (Reche et al., 2012). This, therefore, suggests that the major source was regional biomass burning at all of the sites.

Furthermore, the observed high correlations between levoglucosan and OC at all the sites (r2 of 0.71–0.88), along with the majority of the OC being found to be secondary in origin in Sect. 3.3.1, suggest a common regional source, possibly including a contribution from mainland Europe (Charron et al., 2013) for both levoglucosan and hence for OCbb.

The levoglucosan / K+ ratios were determined by gradient to reduce the impact of other sources of K+, such as airborne sea salt and crustal material particles. A higher gradient was observed at Harwell (1.1 ± 0.14) than at NK (0.73 ± 0.09). Traditional fires (e.g. fireplaces, woodstoves and garden bonfires) have been shown to have a higher levoglucosan / K+ ratio compared to more modern heaters due their higher burn efficiency (Harrison et al., 2012a; Heringa et al., 2011; Khalil and Rasmussen, 2003). The results are suggestive of the use of more efficient devices in NK, consistent with it being located in a smoke control area where the use of fireplaces and woodstoves is controlled.

At all of the sites, the concentrations of levoglucosan and K+ showed some temperature dependence, with the lowest concentrations observed during the warmer periods. However, above about 5 ∘C the concentrations did not appear to be as dependent on temperature and thus the levoglucosan / K+ was largely unaffected by temperature. Higher concentrations during weekends compared to weekdays are likely to be indicative of discretionary or recreational wood burning (Fuller et al., 2014); to determine if this influenced the observed ratios, weekend and weekday levoglucosan / K+ gradients were compared. These plots are shown in the Supplement Fig. S8, with the weekday and weekend gradients similar at Harwell. In contrast, at NK, there was a difference in the observed gradients for the weekend (1.35) compared to weekday (0.57). The higher gradient on the weekend suggests more influence from recreational burning in fireplaces during the weekend in London compared to the rural sites, where fireplaces may be utilised more throughout the week as a means of domestic heating. However, the small number of weekend samples (n=8) limits the conclusions that can be drawn in this regard.

Overall, the levoglucosan / K+ ratios are low compared to most published data directly measured at the sources (e.g. fireplaces) (Caseiro et al., 2009; Puxbaum et al., 2007) but NK and Harwell were within the range reported by Harrison et al. (2012a) for ambient measurements in the UK. NK had a higher levoglucosan / K+ (0.73 ± 0.01) in the current study compared to the ratio of 0.15 observed by Harrison et al. (2012a). The higher ratio observed in the present work may be due to more influence from recreational burning at NK due to the colder winter. Previous work in London also found higher concentrations of wood burning tracers on weekends, which was attributed to secondary or recreational domestic heating (Fuller et al., 2014) supporting the trends observed in the current work for the levoglucosan / K+ ratio. Thus, based on the diagnostic ratios it appears that there was a contribution from local sources dependent somewhat on ambient temperature, on top of a regional background. Therefore, in the next section CPF analysis was applied to the BC data sets at Harwell, NK and Detling during the winter campaign to analyse whether local or regional sources were the largest contributors to biomass smoke.

CPF analysis was applied to the aethalometer data sets from the winter campaign in order to find the direction of highest contributions at each site to aid the source identification. In Fig. 8, CPF analysis of the BC concentrations is presented as polar plots and indicates likely main sources at each site. At NK the highest BC contributions were associated with low wind speeds (Fig. 8), which suggests that local emissions are the main source. This, along with the diurnal cycles for BC (Fig. 5), further indicates that vehicle emissions were the main source of BC. Figure 8 also shows high contributions to the east of Harwell: in this direction lie the village of Chilton and the A34 (both 2 km). Charron et al. (2013) observed a similar trend for EC concentrations and attributed it to both local traffic emissions from the A34 and also regional sources from mainland Europe. Studies have shown that the levels of BC and vehicle emissions diminish to background levels within 150 m of a road (Massoli et al., 2012; Zhu et al., 2004), and thus the A34 was probably too far away to have a strong influence. Therefore, there was probably a regional background of BC, with contributions from local domestic heating (Fig. 8). A regional influence was also evident at Detling, with the highest contributions to the BC concentrations associated with high wind speeds from the east, in agreement with that observed in Harwell and probably influenced by advection from mainland Europe (Charron et al., 2013). High contributions were also associated with low wind speeds in Detling (Fig. 8), indicative of an additional local source of BC along with the regional source.

Polar plots of the mean concentrations of BCwb and BCtr concentrations were calculated in order to further analyse whether the aethalometer model could distinguish between the two sources. At Detling different source directions were observed for the two components; however, at NK the BCwb and BCtr polar plots were similar (Fig. 9). In Fig. 9, the polar plot for BCtrat NK showed a slight trend in direction of the closest road (< 100 m to the east) and further indicates local traffic emissions as the source. In contrast, at Detling the BCtr polar plot (Fig. 9) indicated little influence from the closest road, which was 150 m to the south and thus unlikely to have a large influence (Massoli et al., 2012; Zhu et al., 2004). Rather, the highest concentrations were in the direction of central London, suggesting that emissions from London may be a larger source of BCtr at Detling. Different source contributions between NK and Detling were also observed for BCwb (Fig. 9). From Fig. 9, the BCwb polar plots indicated that local sources were the main source of BCwb at North Kensington, whereas there were both local and regional influences at Detling. The high concentrations of BCwb at Detling (Fig. 9) at low wind speeds to the south/south-west may be indicative of emissions from nearby villages. The highest concentrations were at high wind speeds to the east at Detling (Fig. 9), which is further indication of a strong regional biomass burning source of BC at the sites.

In Fig. 13b, while both aethalometer channels follow the same temporal trends at NK, the increased concentration at 370 nm relative to the 880 nm channel at around 20:00 suggests a change in contributing sources. This timing is consistent with increased contributions from biomass burning for domestic heating, while traffic emissions would be decreasing with reduced traffic flows. The time series of the calculated α values further demonstrates the influence of wood smoke at NK (Fig. 14). Between 20:00 and 21:00, α increased from 1.08 to 1.41, which is indicative of a change in the dominant emission source from traffic (α=1) to wood smoke, which has a highly variable α ranging from 1.38 to 2 (Saleh et al., 2013; Sandradewi et al., 2008b). Furthermore, the re-emergence of traffic emission as the main source can be seen at 06:00 (when traffic counts starts to rise) as α dropped to 1.16 (Fig. 14).

The average α during the night was 1.32 ± 0.05, consistent with a significant biomass burning component of BC at NK. Similar temporal trends in the aethalometer concentrations at 370 and 880 nm were also observed at NK during the other winter de-coupling events (e.g. Fig. S15), while during summer de-coupling events (e.g. Figs. S17 and S18) there was little difference observed between the concentrations at 370 and 880 nm.

An increased concentration at 370 nm compared to 880 nm was also observed at the BT tower in Fig. 13, although the level of enhancement was not a large as at NK. The flat temporal trends observed at the BT tower by the aethalometer during this de-coupling period suggest regional/background levels rather than the influence of local sources. A similar flat temporal trend in BC concentrations was generally observed across the other de-coupling events (e.g. Figs. S16–S18) though without the higher concentrations at 370 nm. Elevated concentrations at 370 nm relative to the 880 nm channel were only observed during the winter de-coupling periods (Figs. 13 and S15), with an average increased concentration at 370 nm of 20–40 % during these de-coupled times. With little brown carbon observed during the summer/spring, it points to a seasonal effect due to biomass burning on a regional scale.

In Fig. 14, the constancy of α at the BT tower, with limited changes during this de-coupled period, suggests a well-mixed background of light-absorbing aerosols with a relatively uniform composition. To examine this further, the frequency distributions of α at the BT tower were calculated for the identified de-coupling periods and compared to the remaining nights without evidence of de-coupling (Fig. 15). Similar distributions of α were found with and without de-coupling from surface emissions, pointing to uniform background particles that were difficult to distinguish from local sources of BC using UV wavelengths. Although most of the identified events were during summer and spring, little seasonal variation in the α distributions at the BT tower was observed, which suggests that regional aerosols over the whole sampling period had a similar value of α.

It should be noted that recent studies have demonstrated that aethalometers may have difficulties distinguishing brown carbon when its levels are low compared to BC, but when there are high concentrations, for example from biomass burning, aethalometers can detect these high levels (Liu et al., 2013; Rizzo et al., 2011). Thus in the current work, apart from winter when there was a substantial amount of biomass burning from domestic heating, the levels of regional brown carbon may have been too low relative to BC for detection using an aethalometer.