Kerb and urban increment of highly time-resolved trace elements in PM 10 , PM 2 . 5 and PM 1 . 0 winter aerosol in London during ClearfLo 2012

. Ambient concentrations of trace elements with 2 h time resolution were measured in PM 10 − 2 . 5 , PM 2 . 5 − 1 . 0 and PM 1 . 0 − 0 . 3 size ranges at kerbside, urban background and rural sites in London during winter 2012. Samples were collected using rotating drum impactors (RDIs) and subsequently analysed with synchrotron radiation-induced X-ray ﬂuorescence spectrometry (SR-XRF). Quantiﬁcation of kerb and urban increments (deﬁned as kerb-to-urban and urban-to-rural concentration ratios, respectively), and assessment of diurnal and weekly variability provided insight into sources governing urban air quality and the effects of urban micro-environments on human exposure. Trafﬁc-related elements yielded the highest kerb increments, with values in the range of 10.4 to 16.6 for SW winds (3.3–6.9 for NE) observed for elements inﬂuenced by brake wear (e.g. Cu, Sb, Ba) and 5.7 to 8.2 for SW (2.6–3.0 for NE) for other trafﬁc-related processes (e.g. Cr, Fe, Zn). Kerb increments for these elements were highest in the PM 10 − 2 . 5


Introduction
Ambient particulate matter (PM) has long been recognized to have a detrimental effect on public health in urban areas (e.g.Dockery and Pope, 1994).Of particular interest are particles with an aerodynamic diameter less than 10 µm (PM 10 ) as these particles can penetrate deeply into the lungs (Franklin et al., 2008;Zhou et al., 2011).Reche et al. (2012) reported even higher toxicity to human cells for the PM 2.5−1.0 than for the PM 10−2.5 fraction.Particle toxicity is known to vary significantly with PM composition and emission sources (Kelly and Fussell, 2012), with identified toxic constituents including soluble secondary inorganic particles, elemental and organic carbon, and especially metals.Effective mitigation strategies therefore require detailed, size-dependent characterization of particle composition and emission sources.
In addition to their direct effects on human health, metals and trace elements are of importance because their high source specificity and atmospheric stability make them effective tracers for source apportionment.In Europe, four main source types in PM 10 are commonly identified: vehicles (with tracers including e.g.Fe, Ba, Zn, Cu), crustal materials (e.g.Al, Si, Ca, Fe), sea salt (mainly Na, Cl, Mg) and mixed industrial/fuel-oil combustion (mainly V, Ni, S) and secondary aerosol (mainly S) (Putaud et al., 2010;Viana et al., 2008).The contribution of mineral dust and sea salt in most urban areas is larger in PM 10 than in PM 2.5 (Harrison et al., 2001;Weijers et al., 2011).Emissions from vehicle exhaust, industry and secondary aerosol are predominantly emitted and formed as PM 1.0 or in PM 2.5 (Bukowiecki et al., 2010;Harrison et al., 2011;Richard et al., 2011).Several of these sources have been directly linked to adverse health effects.For example, the largest aerosol source of human toxicity in Barcelona was attributed to traffic activities (encompassing vehicle emissions, road dust and secondary nitrate), with fuel oil combustion and industrial emissions also contributing to increased cancer risk (Reche et al., 2012).Turoczi et al. (2012) observed higher toxicity from direct emissions (e.g. from traffic) than from photochemically processed aerosol.
The Clean Air for London project (ClearfLo; www.clearflo.ac.uk) is a multinational effort to elucidate the processes driving poor air quality in London, implemented through comprehensive measurements of particle-and gasphase composition, and meteorological parameters (Bohnenstengel et al., 2014).ClearfLo builds upon recent modelling and monitoring studies in London (Arnold et al., 2004;Bohnenstengel et al., 2011Bohnenstengel et al., , 2013;;Harrison et al., 2012a;Mavrogianni et al., 2011).Despite improved air quality, PM 10 concentrations are not decreasing, resulting in frequent exceedances of the daily PM 10 limit (Harrison et al., 2008).Such exceedances are caused by complex interactions of regional and local emission sources, together with meteorological factors such as wind speed, air mass origin and daily cycles of the atmospheric boundary layer (Charron and Harrison, 2005;Harrison and Jones, 2005;Jones et al., 2010).Currently, emissions by industrial sources and stationary combustion are modest, while traffic is thought to contribute up to 80 % of the total PM 10 in London, compared to less than 20 % for the entire UK, according to emission inventories between 1970 and 2001 (Dore et al., 2003).
The spatial density of emission sources found in typical urban environments leads to elevated particle concentrations compared to nearby rural locations.As an example, buildings may influence local meteorology by restricting air circulation (street canyon effect), producing human exposures that are orders of magnitude higher than those predicted by regional dispersion models (Zhou and Levy, 2008).This provides both acute exposure risk and increased long-term exposure for those passing through regularly, thereby producing a non-negligible impact on public health.To assess the impact of such micro-environments, we here investigate London trace element concentrations in terms of increments, defined as the concentration ratios between an environment of interest and a reference site (Charron et al., 2007;Lenschow et al., 2001).
Only a few studies have investigated trace elements through simultaneous measurements at multiple sites.Harrison et al. (2012b) reported increments of kerbside-to-urban background sites in London for non-size-segregated aerosol with a time resolution of 1 to 4 days.Theodosi et al. (2011) found that at urban and suburban sites in Athens and a regional site in Finokalia, Greece, crustal elements dominate coarse particles (PM 10−2.5 ), whereas anthropogenic sources such as fossil fuel combustion were confined to fine particles (V, Ni and Pb have > 70 % of their mass in PM 1.0 ).Bukowiecki et al. (2009aBukowiecki et al. ( , 2010) ) examined trace elements in PM 10−2.5 , PM 2.5−1.0 and PM 1.0−0.1 aerosol at street canyon and urban background sites in Zürich, Switzerland, and found increasing increments (note: 1 means no increment) with particle size from about 1.2 (fine mode) to 2.4 (coarse mode) (averaged over all elements).All these studies report increments close to 1 for elements originating from regional sources such as sea salt and Saharan dust, while local, especially traffic-related, sources yield increments around 2 for resuspension-related elements and between 3 and 5 for traffic-related elements.Additionally, the 1 h time resolution used by Bukowiecki et al. (2009aBukowiecki et al. ( , 2010) ) enabled identification of enhanced increments for resuspension-and wearingrelated elements like Si and Sb during peak traffic flows.
There is a need for more high time-resolved sizesegregated increment analyses to assess the exposure to trace elements from emission sources within urban areas under varying meteorological conditions.Here we present sizesegregated measurements of aerosol trace elements with 2 h time resolution performed simultaneously at kerbside and urban background sites in London, and a rural site outside London.We assess the effects of urban micro-environments on human exposure to particulate pollutants through the quantification of urban and kerb increments.These exposures are further investigated in terms of contributing emission sources, diurnal and weekly variability, local wind patterns and regional transport effects.

Measurement campaigns
The ClearfLo project was a measurement programme in and around London lasting 2 years (2011-2012) and including two month-long intensive observation periods (IOPs) in 2012 (Bohnenstengel et al., 2014).This paper focuses on the winter IOP lasting from 6 January to 11 February 2012.Measurements took place at three sampling sites located at or near permanent air quality measurement stations in the Automatic Urban and Rural Network (AURN): a kerbside site close to a very busy road, an urban background site in a residential area, and a rural background site away from direct emission sources (see Fig. 1).
The urban background sampling site was at the grounds of the Sion Manning Secondary School in North Kensington (NK, lat 51 • 31 21 N, long 0 • 12 49 W).NK is situated within a highly trafficked suburban area of London (Bigi and Harrison, 2010;Harrison et al., 2012a).During the ClearfLo IOPs this site served as the main measurement site and was upgraded with a full suite of particle-and gas-phase instruments, and instruments to measure meteorological parameters (Bohnenstengel et al., 2014).The kerbside site was located at Marylebone Road (MR, lat 51 • 31 21 N, long 0 • 09 17 W) about 4.1 km to the east of NK (Charron and Harrison, 2005;Harrison et al., 2011).This site is located at the southern side of a street canyon, with an axis running approximately 260 to 80 • .Measurements took place at 1 m from a busy six-lane road with a traffic flow of approximately 73 000 vehicles per day of which 15 % consist of heavy-duty vehicles.Braking and stationary vehicle queues are frequent at the site due to a heavily used pedestrian light-controlled crossing (65 m west of MR) and a signal-controlled junction (200 m west of MR).The rural site at the Kent Showgrounds at Detling (DE, lat 51 • 18 07 N, long 0 • 35 22 E) was approximately 45 km to the southeast of London downtown on a plateau at 200 m a.s.l.surrounded by fields and villages, and was close to the permanent measurement station of Kent and Medway Air Quality Monitoring Network.The site provides excellent opportunities to compare the urban and kerbside air pollution with the rural background pollution levels (Bohnenstengel et al., 2014;Mohr et al., 2013).A busy road with ∼ 160 000 vehicles per day is located approximately 150 m south of DE.Meteorological parameters were measured at DE and at the British Telecom (BT) Tower (lat 51 • 31 17 N, long 0 • 08 20 W), ∼ 0.5 km east of MR (Harrison et al., 2012a tails).A detailed description of the RDI can be found in Bukowiecki et al. (2005Bukowiecki et al. ( , 2009c) ) and Richard et al. (2010).
In short, aerosols are sampled through an inlet that removes all particles with aerodynamic diameter d > 10 µm at a flow rate of 1 m 3 h −1 .The particles are size segregated in three size ranges based on d -PM 10−2.5 (coarse), PM 2.5−1.0 (intermediate) and PM 1.0−0.3(fine) -by passing sequentially through three rectangular nozzles of decreasing size.Particle deposition occurs via impaction on 6 µm thick polypropylene (PP) foils mounted on aluminium wheels and coated with Apiezon to minimize particle bouncing effects.After the last impaction stage a backup filter samples all remaining particles before the air passes through a pump.After each 2 h sampling interval the three wheels rotate stepwise to a blank section of the foil before a new sampling interval takes place.The small-size collection limit of the fine fraction was previously estimated at 100 nm (Bukowiecki et al., 2009c;Richard et al., 2010).However, new laboratory measurements of the RDI collection efficiency indicate an instrument-dependent (i.e. based on the machining of the specific nozzle) small-end cut point of approximately 290-410 nm (see Supplement A for details).This results in sampling of a smaller size range (PM 1.0−0.3 ) than the PM 1.0−0.1 range reported in previous studies, and influences the measured concentrations of elements with significant mass near this cut point (S, K and Pb).

SR-XRF analysis
Trace element analysis on the RDI samples was performed with synchrotron radiation-induced X-ray fluorescence spectrometry (SR-XRF) at the X05DA beamline (Flechsig et al., 2009)  The fluorescence counts per element were calibrated to the element mass concentration using multi-element standards, where each standard consisted of a set of preselected elements in five different concentrations ranging between 0.05 and 0.4 µg cm −2 .The absolute element concentrations in these standards were determined with inductively coupled plasma-optical emission spectroscopy (ICP-OES).The absolute calibration factor for the SR-XRF system was referenced to Fe and determined from the linear relation between the SR-XRF response and the ICP-OES measurements.Because the fluorescence yield increases with atomic number Z, a relative calibration curve was constructed as follows: for each element present in the standards and having a detectable Kα 1 line, an absolute calibration factor was determined as for Fe, and a dimensionless relative response factor was calculated as the ratio of this absolute factor to that of Fe.These relative response factors were plotted as a function of line energy and a polynomial curve was fit to obtain response factors by interpolation for elements not present in the standard.In total 25 elements were quantified (Na, Mg, Al, Si, P, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Ni, Cu, Zn, Br, Sr, Zr, Mo, Sn, Sb, Ba, Pb).Although some of these elements are not found in trace amounts in aerosols, for simplicity we refer here to trace elements for all analysed elements.Details of the methodology can be found elsewhere (Bukowiecki et al., 2005(Bukowiecki et al., , 2008;;Richard et al., 2010), with the following significant changes (see Supplement B for further details): 1.At SLS, we used an e2v SiriusSD detector (SiriusSD-30133LE-IS) and in-house built vacuum chamber to extend the measured range of elements down to Na and Mg.c Attenuation factors and a (µm −1 ; a = 2/3 • µ • ρ with µ the mass attenuation coefficient (cm 2 g −1 ) and ρ the particle mass density (g cm −3 )) according to Eq. ( 4) in Formenti et al. (2010).d Total correction factor defined as ratios AF calibration standard / AF sample.Self-absorption effects are neglected in the PM 1.0−0.3samples; therefore these samples are only corrected for AF calibration standards.e Ca is corrected based on the average AF of calibration standards 1 and 2, and a calibration standard used to calibrate the elements Ti to Zn at SLS (Ca present in all three standards).Data for this third calibration standard are: particle size of 7.0 ± 2 µm, average density of 2.37 g cm −3 and AF and a for Ca of 0.89 and 0.03 µm −1 , respectively.

Reference
time for RDI samples and reference standards, thereby reducing uncertainties in absolute and relative calibrations.
3. Data were processed with the Spectral Analysis for Multiple Instruments toolkit for XRF (SAMI-XRF) developed in-house within the IGOR Pro software environment (Wavemetrics, Inc., Portland, OR, USA).SAMI handles spectral fitting, quantification of associated uncertainties, and calculation and application of calibration parameters.
XRF is sensitive to self-attenuation of fluorescence radiation in the sample and depends on the sample composition and density, as well as particle layer thickness or particle size.The PM sample thickness of the coarse and intermediate fractions was maximally 0.7-1.5 µm at a maximum concentration of 10 µg m −3 total PM mass for each sample.For these fractions, self-absorption therefore mainly occurs within the individual particles (geometric mean of 5 and 1.6 µm for PM 10−2.5 and PM 2.5−1.0 fractions, respectively).For the fine fraction the PM layer is several micrometres thick, resulting in absorption inside the PM layer.However, this layer is mainly composed of species not resolved by SR-XRF (H, C, N, O).Furthermore, most mass of the lightest elements (Na-Ca) is restricted to the coarse and in-termediate fractions (except for S and K).We therefore neglect self-absorption effects in the fine fraction samples.The calculated layer thickness of the dried calibration solution on the calibration standards is negligible at 3-60 nm, but the particle size of the dried droplets shows a geometric mean volume size distribution of 9 ± 5 µm and is therefore relevant for self-attenuation.Attenuation factors (AFs) were calculated for the calibration standards as well as for the coarse and intermediate fraction samples, as a function of density, mass attenuation coefficient and particle size, according to a simple attenuation model (Table 2; Formenti et al., 2010).For ambient samples and calibration standards, the attenuation length or penetration depth of X-rays for the heavier elements (i.e.above Ca) is greater than 33 µm, resulting in the near-complete excitation of these elements and correction factors below 3 %.Self-absorption correction factors for Na-Ca are non-negligible, as discussed below and in Table 2.The attenuation of the calibration standards is taken into account for all samples, and additional corrections are applied to the coarse and intermediate samples.Calzolai et al. (2010) found comparable self-absorption effects for samples of different composition, total loading and sampling site.Because the elemental composition and particle size distribution of each sample are unknown, we assume a uniform correction for each element within a given size fraction.

Other measurements
Here a short description is given of relevant particle-and gas-phase instruments deployed at MR, NK and DE during the winter IOP.Daily PM 10 filter samples (midnight to midnight) were collected at MR and NK using Partisol 2025 samplers (Thermo Scientific, Inc.).The filters were digested in a 1 : 2 mixture of perchloric and hydrofluoric acid, and subsequently analysed by ICP-mass spectrometry (ICP-MS, calibration with NIST standards) for the determination of Na, Al, Ca, Ti, V, Mn, Fe, Ni, Cu, Zn, Sr, Mo, Sb, Ba and Pb.Additionally, Mg, K and Sn were available at NK. Highresolution time-of-flight aerosol mass spectrometers (HR-ToF-AMS, Aerodyne Research, Inc., Billerica, MA, USA) were deployed at MR (5 min time resolution), NK (5 min resolution every 30 min) and DE (2 min resolution) to characterize the non-refractory submicron aerosol components (DeCarlo et al., 2006).PM 10 mass concentrations were measured at all three sites with FDMS-TEOM (Filter Dynamics Measurement System Tapered Element Oscillating Microbalances; Thermo Scientific, Inc.) with a 1 h time resolution.NO x measurements at MR and NK were performed with a NO x chemiluminescent analyser with a single chamber and a single detector (API, A Series, model M200A; 15 min resolution).At DE, NO was determined with a Thermo Scientific 42i analyser and NO 2 with an Aerodyne CAPS-NO 2 (SN 1002) and an Aerodyne QCL-76-D.These NO and NO 2 measurements were summed together to obtain NO x (1 min resolution).Equivalent black carbon (EBC) was measured with a two-wavelength Aethalometer (λ = 370 and 880 nm, model AE22, Magee Scientific) at MR and a seven-wavelength Aethalometer (λ = 370-950 nm, model AE31, Magee Scientific) at NK and DE (5 min resolution), with a 2.5 µm cyclone at MR and DE and a 3.5 µm cyclone at NK. Traffic counts by vehicle group at MR from road sensors (number of vehicles per 15 min) were available as well.Wind direction and wind speed data for MR and NK were taken from the BT Tower (30 min resolution) where anemometers were placed to the top of an open lattice scaffolding tower of 18 m height on top of the main structure (190.8 m a.g.l.; Wood et al., 2010), whereas local data were used at DE (1 min resolution).Air mass origins were analysed with back trajectory simulations using the UK Met Office's Numerical Atmospheric Modelling Environment (NAME) dispersion model (Jones et al., 2007).

Data intercomparison and uncertainty
Here we compare RDI-SR-XRF data with independent filter data (24 h PM 10 trace element data analysed with ICP-MS; roughly 9 % uncertainty at a 95 % confidence interval) for 18 elements collected at MR and NK (no filter data were available at DE).For this comparison, the three size ranges of the RDI were summed up to total PM 10 and averaged to the filter collection period.Details of the intercomparison results can be found in Supplement C. In short, the majority of the elements (Al, Ca, Ti, Mn, Fe, Cu, Zn, Sr, Sb, Ba) agree within approximately ±50 % with Pearson's R > 0.78.Na and Mg agree as well, but have higher uncertainties due to self-absorption corrections.For the other elements, disagreement can be attributed to low or unknown filter sample extraction efficiencies (Ni, Mo) and differences in the particle size range sampled by the two measurement techniques (K, V, Sn, Pb).However, all elements are retained in the ensuing analysis as (1) they yield internally consistent results, as described in the following sections; (2) the ensuing analysis relies on relative changes/ratios per element across sites and is therefore not affected by a systematic bias in absolute magnitude.
The agreement between XRF and filter measurements in the present study compares favourably with that obtained in previous intercomparisons of trace element measurement techniques.Comparison of RDI-SR-XRF with daily element concentrations from a high-volume sampler followed by subsequent analysis using laboratory-based wavelength dispersive XRF (Bukowiecki et al., 2005) and by ICP-OES and ICP-MS (Richard et al., 2010) yielded slopes between 0.7 and 1.6 (except for S and K) with Pearson's R > 0.5.The spread/biases in these intercomparisons are not necessarily due to SR-XRF issues, as can be seen from a comparison by Salcedo et al. (2012) of ICP with proton-induced X-ray emission (PIXE) and AMS trace element measurements.Agreement between ICP and PIXE data was in the same range as between either method and the AMS data, with slopes ranging between 0.06 and 0.93 with Pearson's R from about 0.3 to 0.7.
Estimated uncertainties (per size fraction) and detection limits for each measured element are given in Supplement Table S3.A brief overview is presented here: 1. RDI sampling: the fluctuations in the flow rate are negligible within 5 % (Richard et al., 2010) and the uncertainties in the size cut-off are discussed in Supplement A.
2. SR-XRF accuracy: uncertainties in the absolute and relative calibrations affect absolute/fractional concentrations, but cancel out for relative changes/ratios, because all samples were measured under the same calibration conditions.
3. Issues such as imperfect flatness of the sample foils and detector dead-time corrections (Richard et al., 2010) reduce measurement precision but affect all elements with the same scaling factor.correction uncertainties.Sample inhomogeneity was assessed by Bukowiecki et al. (2009c) and found to contribute ±20 % uncertainty.
For most elements, except Mn and the lightest elements, sample inhomogeneity is the largest source of uncertainty.Mn is affected by spectral analysis uncertainties due to peak overlap with Fe, which is present in much higher concentrations.Therefore, a small bias in the energy calibration as function of detector channel leads to a large change in the peak area of Mn.Self-absorption effects are a significant source of uncertainty for the lightest elements (Na-Ca), but the good comparisons to the filter data suggest that the corrections lead to reasonable results.All data points lie well above their element detection limits, resulting in negligible uncertainties from the signal strength.In addition, RDI-SR-XRF measurements (both absolute/fractional and relative/ratio) are affected by atmospheric variability.This variability is likely the predominant source of the data spread evident in Table 3 and the following analyses.

Trace element concentrations
During the ClearfLo winter IOP total mass concentrations of the analysed trace elements ranged from less than 0.1 µg m −3 to ∼ 10 µg m −3 .Typically, concentrations were highest at MR and lower at NK and DE, and decreased with particle size.An overview of the obtained trace element concentrations as a function of size and site is given in Table 3.Note that S is not a trace element, but is commonly reported in trace element studies and is a good tracer for regional transport.Among the analysed trace elements, highest concentrations at MR were found for Na (28 %), Cl (25 %) and Fe (22 %).At NK this was the case for Na (39 %), Cl (29 %) and Fe (11 %) and at DE Na (40 %), Cl (28 %) and Fe (8 %).Total analysed mass measured by the RDI-SR-XRF (trace elements +S) contributed on average 14 % to the total PM 10 mass (from FDMS-TEOM) of 32 (5-74) µg m −3 at MR (not extrapolated to the corresponding oxides), 10 % to the mass of 23 (1.4-63) µg m −3 at NK and 7.4 % to the mass of 17 (0.5-58) µg m −3 at DE.
A comparison between the contributions of coarse, intermediate and fine fractions to the total PM 10 mass of each trace element is shown in Fig. 2 for the three sites.Trace elements at MR are dominated by the coarse fraction.Analysis in the following sections and previous measurements at this site (Charron and Harrison, 2005) suggest this is caused by large contributions of resuspension-and traffic-related mechanical abrasion processes, which primarily contribute to the coarse fraction.For all elements at this site, except S, Br and Pb, the coarse fraction contributes more than 50 %.Mass fractions of intermediate mode elements to total PM 10 are rather constant with contributions ranging from 11 to 27 %.The fine fraction contributes up to 50 % of total mass for S,  K, Zn, Br and Pb; for other elements fine contributions are less than 20 %; S, K, Zn, Br and Pb are typically dominated by the fine fraction with known sources including heavy oil combustion (S, K, Zn; Lucarelli et al., 2000), traffic exhaust (Br, Pb; Formenti et al., 1996), industrial processes (Zn, Pb; Moffet et al., 2008), and secondary sulfate and wood combustion (S, K, Pb; Richard et al., 2011).For most elements, particle mass contributions of the smaller size fractions are more important as one moves from kerbside to urban background to rural sites (Fig. 2).The relatively large fine fraction contribution at DE is probably caused by the absence of local traffic which results in lower contributions of resuspension-and traffic-related processes to total element concentrations.A different behaviour is observed for Cr, Ni and Mo with on average 80 % of their mass at DE in the coarse fraction, compared to 73 % at MR and 60 % at NK.The time series of these coarse mode species are very spiky, are slightly enhanced with SW winds, but are not collocated with measurements of BC and AMS species, suggesting emissions from a local industrial source, potentially from stainless steel production (Querol et al., 2007;Witt et al., 2010) near DE rather than regional transport.
Comparing the contributions of groups of elements to total trace element concentrations at the sites provides an overview of local and regional sources affecting London; a detailed source apportionment study will be the subject of a future paper.Na, Mg and Cl are typical sea salt elements and contribute around 66 % to the total PM 10 trace element mass at all three sites, indicating that the air pollutant levels caused by elements are dominated by natural emission sources being transported to London.Mineral dust elements (Al, Si, Ca, Ti) mainly brought into the air via resuspension contribute on average 13 % at MR, NK and DE.For some specific brake wear elements (Cu, Sb, Ba) these contributions

Urban increment
The urban increment compares the trace element concentrations at the urban background site to the concentrations at the rural site, and is calculated here as the ratio of concentrations at NK to DE. Figure 3 shows the mean, median and 25-75th percentile urban increment ratios for the coarse, intermediate and fine fractions per element.Most elements, except Ni and coarse mode Cr, are enriched at the urban background site by factors between 1.0 and 6.5 (median ratios Coarse mode Zr exhibits low concentrations at DE, where the median value actually falls below detection limit, though discrete events above detection limit also exist.For this reason, the median-based urban increment is not plotted, while the mean ratio is driven by several large concentration peaks at NK, resulting in a large mean ratio of 21.In the case of Cl, a large spread in the urban increment values is seen for all three size ranges.Cl is likely depleted relative to other sea salt elements like Na and Mg (throughout the campaign Cl concentrations fall to 0, whereas Na and Mg concentrations remain positive) due to replacement by nitrate, and the extent of such depletion is greater in small particles (Nolte et al., 2008).At DE, Cl depletion seems apparent at all size ranges, whereas at MR depletion mainly takes place in the PM 1.0−0.3fraction.NK shows Cl depletion especially in the PM 1.0−0.3fraction, but to some extent also in intermediate mode particles.
For ease of discussion, we empirically group elements based on similar urban increment values.Mn, Fe, Cu, Zn, Zr, Mo, Sn, Sb and Ba show urban increments on average of 3.5 in the coarse, 3.1 in the intermediate and 2.0 in the fine fraction (Fig. 3).These have been identified as traffic-related elements by e.g.Amato et al. (2011), Bukowiecki et al. (2010), Minguillón et al. (2014), Richard et al. (2011) and Viana et al. (2008).Zr has also been linked to mineral dust (Moreno et al., 2013).We can understand that from analysing the enrichment factors (EFs) of these elements.EF is a measure of the enrichment of elements relative to the upper continental crust (UCC) and is defined as ppm metal in the sample divided by ppm metal in UCC with Si as reference material (UCC from Wedepohl, 1995).Zr is the only element in this traffic group that is depleted in the atmosphere relative to their UCC concentrations, but with concentrations at NK higher than at DE.Most other elements clearly indicate anthropogenic origin with EF > 10.Dependent on the method, Zr can be either grouped with traffic-related elements or with dust elements.The urban increments are similar to that of NO x , where concentrations at NK were on average a factor of 4.9 higher than at DE (the mean concentration at NK was 68 ppb; at DE, 14 ppb).EBC, a marker for both traffic and wood burning emissions, had an urban increment of only 1.1 (concentration at NK 757 ng m −3 , at DE 633 ng m −3 ), likely due to local wood burning emissions around DE (Mohr et al., 2013).Al, Si, Ca, Ti and Sr as markers for mineral dust (e.g.Amato et al., 2009;Lin et al., 2005;Lucarelli et al., 2000) show a factor 2.0 higher concentrations at NK relative to DE in the coarse, 1.9 in the intermediate and 1.6 in the fine fraction (EF < 10).These results indicate that moving from rural to urban backgrounds yields a larger relative increase in traffic than in mineral dust elements.Surprisingly, sea salt elements (Na, Mg, Cl) show higher concentrations at NK than at DE of up to a factor of 2 for the coarse mode, despite the expected dominance of regional over lo-cal sources.This highlights the potential importance of sea or road salt resuspension by traffic.Similar urban increment values for traffic-related, resuspension and sea salt elements have been observed by Lee et al. (1994) for particles below a few micrometres.Theodosi et al. ( 2011) also found higher increments (> 2) for trace elements in PM 10 aerosol from local anthropogenic sources like fossil fuel combustion (V, Ni, Cd) and traffic (Cu), relative to long-range transported Saharan dust (Fe, Mn) with increments close to 1.However, our study suggests that the non-size-resolved increment values reported in the cited studies do not fully capture the urban/rural differences.
The influence of regional transport by anthropogenically produced elements (Fig. 3) is seen by the low urban increments between 1.1 and 1.8 for P, S, K, Zn, Br, Sn and Pb in PM 1.0−0.3(EF > 22) and of 1.6 for total PM 10 mass (concentration at NK 23 µg m −3 , at DE 17 µg m −3 ).The concentrations of the main components in PM 10 (sulfate, nitrate and secondary organic compounds) within an urban area are mostly influenced by regional transport, as found in London during the REPARTEE project (Harrison et al., 2012a) and in Paris during the MEGAPOLI project (Crippa et al., 2013;Freutel et al., 2013), resulting in low increments for total PM 10 mass.Similar urban increment values (1.3 to 1.8) for 1 and 24 h total PM 2.5 mass concentrations were reported across many sites in the UK (Harrison et al., 2012c).

Kerb increment
While the urban increment investigates the effect of diffuse emission sources on particle concentrations, the kerb increment investigates an urban micro-environment, specifically the local effects of roadside emissions and activities.Here, the kerb increment is calculated as the ratio of concentrations at MR to NK.However, observed concentrations at MR strongly depend on wind direction, because the road runs from approximately 260 to 80 • and the street canyon with the surrounding buildings and intersections creates a complex wind circulation system (Balogun et al., 2010).Since the measurement station is located at the southern side of the canyon, measurements during time periods with winds from the south are influenced by on-road emissions on top of the urban background pollution.Higher concentrations were observed with SSE winds, i.e. perpendicular to the direction of the road by e.g.Balogun et al. (2010), Charron and Harrison (2005) and Harrison et al. (2012b).
In this study, the RDI-SR-XRF data was split into four equally spaced wind direction sectors based on wind direction data: N (315-45 • ), E (45-135 • ), S (135-225 • ) and W (225-315 • ). Figure 4 shows size-resolved trace element concentrations per wind sector normalized to the global median concentration for each element at MR.As expected, winds from the south yield the highest concentrations, whereas northern winds yield the lowest, independent of size fraction.West and east winds are parallel to the street canyon Traffic-related and some other anthropogenically related elements (V, Cr, Mn, Fe, Ni, Cu, Zn, Zr, Mo, Sn, Sb, Ba) show the strongest wind direction dependency with up to a factor of 2-3 higher concentrations during S relative to N winds for the three size fractions (Fig. 4).A factor of 1.5-2 is obtained for resuspended dust elements.Harrison et al. (2012b) found a ratio of 2 for Fe (as tracer for brake wear) and 1.2 for Al (as tracer for mineral dust) for SW versus NE winds for particles between 2 and 3 µm.However, they were limited by their time resolution of several days, resulting in potentially substantial wind direction variations during each measurement, which would blur the different conditions and yield reduced ratios.

Atmos
Other elements show only minor correlations with wind direction (Fig. 4), indicating more influence from regional transport, instead of being locally affected by traffic.Only fine mode S, K and Br seem to be enriched with winds from the east, potentially related to long-range transport from the European continent.
Local wind direction has a greatly reduced effect at urban background and rural sites.At NK, the element concentrations are only subject to high concentration outliers for E winds (Fig. S4), potentially caused by the transport of pollutants from downtown London, or by lower wind speeds occurring with E winds resulting in reduced dilution and increased concentrations of traffic pollutants (e.g.NO x ) throughout the city (Fig. S6).The rural site hardly shows wind direction dependent concentrations (Fig. S5-6).Interpretation of data from the E sector is unclear due to the low number of data points (45 out of 318 data points).Only data from the N sector show enhanced concentrations for several elements correlating with higher wind speeds and back trajectories consistent with transport from continental Europe.
To simplify reporting of the kerb increment and facilitate comparison with previous studies (e.g.Harrison et al., 2012b), we combined the south/west sectors and the north/east sectors into SW (135-315 • ) and NE (315-135 • ) sectors.To eliminate meteorological and/or regional transport effects, this segregation is performed at both MR and NK.The kerb increment is then calculated as the ratio of MR to NK and shown in Fig. 5 (Fig. S7 shows the increments for the four individual sectors).As with the urban increment, we focus on the ratio of the medians at MR and NK to reduce the effects of outliers.Two features become directly visible: the kerb increment is much higher for coarse than for intermediate and fine particles, and kerb increments are much higher for SW than for NE wind conditions.With the latter, kerb increments are on average 2.7, 1.6 and 1.7 for coarse, intermediate and fine mode particles, respectively.This significant enhancement is likely due to recirculation of particles within the street canyon following their resuspension and/or emission by traffic.However, these increments are much smaller than those observed in the SW sector, where enhancements relative to NK of 6.7, 3.3 and 3.1 (coarse, intermediate, fine) are observed.These results indicate the existence of microenvironments within the street canyon dependent on wind direction.
As in the previous discussion, we again group elements by kerb increment (Fig. 5).The first group consists of Cu, Zr, Mo, Sn, Sb and Ba and yields the highest increments in the coarse mode ranging from 10.4 to 16.6 in the SW sector (3.3-6.9 for NE).These elements are typically associated with brake wear (e.g.Bukowiecki et al., 2009b;Harrison et al., 2012b), and are much higher than the increments of 4.1 to 4.4 reported by Harrison et al. (2012b)  particles < 21 µm.They assigned Fe, Cu, Sb and Ba to brake wear, but in the current study Fe has a significantly lower kerb increment than other brake wear tracers, suggesting a significant alternative source.When combining all size fractions and ignoring wind direction influences, increments in this study are about 4.9, and more similar to previous studies.The discrepancies between the kerb increments obtained using these two calculation strategies highlight the difficulties in characterizing human exposure to locally generated pollutants in urban environments, as the detailed topography and microscale meteorology greatly alter particle concentrations, and the effects are size dependent.Amato et al. (2011) calculated road side increments in Barcelona for trace elements in PM 10 with a 1 h time resolution and found increments for brake wear elements of only 1.7 (based on Fe, Cu, Sb, Cr, Sn).These low increments are probably due to the reduced dispersion in Barcelona caused by a complex topography, resulting in high urban background levels.
The second group consists of V, Cr, Mn, Fe, Ni, Zn and Pb with increments of 5.7-8.2(PM 10−2.5 ) in the SW sector (2.6-3.0 for NE) (Fig. 5).V and Ni are typically assigned to industrial sources and heavy-oil combustion (e.g.Mazzei et al., 2007;Viana et al., 2008), Zn is usually associated with tyre wear (e.g.Harrison et al., 2012b;Lin et al., 2005), and the other elements are commonly associated with trafficrelated emissions (e.g.Amato et al., 2013;Bukowiecki et al., 2009a;Richard et al., 2011).We label this group as anthropogenically influenced (ANTH).The EF of V, Cr and Ni are much lower than those of the other elements in this group (4 vs. > 10), indicating at least to some extent different source origins.These kerb increments are similar to the ones for NO x of 8.5 for SW and 2.4 for NE, confirming the anthropogenic influence (traffic and other sources) on these elements.The high braking frequency at MR due to congested traffic probably resulted in increased kerb increments of brake wear relative to ANTH elements that are also influenced by local traffic and other sources around NK. Increments of these ANTH elements are higher than previously reported values of 1.8-4.5 for studies with low time resolution and non-size-segregated particles (Boogaard et al., 2011;Janssen et al., 1997).The high increments presented here might be caused by street canyon effects, trapping pollutants emitted at street level and preventing dilution to the urban background.The enhanced kerb increments for brake wear relative to ANTH elements are apparent in all three size fractions, although increments become more similar towards smaller sizes with a factor 1.7 between both element groups in the coarse, 1.5 in the intermediate and 1.4 in the fine mode.Both groups show the additional information gained with size-segregated aerosol, where exposure to trace elements in the street canyon relative to the urban background increases with particle size, either caused by increased traffic-related emissions with particle size or by more efficient transport of submicron particles from street sites to the urban background.Furthermore, the highly time-resolved element mea-surements presented here enabled us to resolve the systematic, wind direction dependent variability in kerb increments.
The third group is associated with mineral dust (Al, Si, Ca, Ti, Sr) with coarse mode increments of 3.4-5.4for SW winds (1.7-2.3 for NE) (Fig. 5).These elements are brought into the air both by traffic-induced resuspension and transport from other locations.This second process increases both urban background and kerbside concentrations, and thus reduces kerb increments relative to direct traffic-related elements.Lower kerb increments for mineral dust than trafficrelated elements are generally observed in increment studies (Amato et al., 2011;Boogaard et al., 2011;Bukowiecki et al., 2009b;Harrison et al., 2012b), although the dust increments found in this study are larger than most reported increments (typically 1-2).As in the traffic-related groups, increments increase with particle size, indicating enhanced human exposure at the street side of particles above 1 µm.
Na, Mg and Cl (sea salt) form the fourth group and yield kerb increments of 1.0 to 2.7, independent of size fraction but with slightly enhanced ratios with SW compared to NE winds (Fig. 5).Similar increments were observed for total PM 10 mass.As discussed for urban increments, even though these elements have regional sources, they are influenced by resuspension processes within the urban area which are enhanced at kerbside sites.
The remaining elements (P, S, K, Br) can be grouped together.In the coarse mode, these elements yield increments similar to the mineral dust group, indicating that this group is influenced by resuspension processes in the street canyon (Fig. 5).However, especially in the fine mode, increments around 1 were found, consistent with regional transport dominating over local emission sources.

Temporal trends in trace element concentrations
In contrast to traditional trace element measurements, the RDI-SR-XRF enables measurement of element concentrations with high time resolution (2 h in this work).This enables investigation of diurnal cycles, which are useful both for source discrimination and in determining the processes contributing to elevated PM levels.We also discuss weekly cycles, which can be useful in distinguishing emissions from heavy-duty and passenger vehicles (HDV and LDV); HDV numbers typically diminish during the weekend.Back trajectory analysis aids source discrimination by understanding regional transport influences by different air mass origin.
Here we discuss the temporal trends of trace elements in five groups based on expected sources and the increment analyses in Sect.4.2, in order of increasing local influence: regional background, sea salt, mineral dust, traffic-related and brake wear.
Figures 6 and 7 show size-segregated median diurnal and weekly cycles, respectively, for five elements representative of the classes mentioned above: Na (sea salt), Si (mineral dust), S (regional background), Fe (traffic-related) and Sb (brake wear) at the three sites.Because of the wind direction effect evident at MR, diurnal cycles at all three sites are shown for SW and NE winds.Wind direction analyses are not incorporated into the weekly cycles because the month-long campaign provided insufficient data points for meaningful division.This also means that weekly cycles are subject to influences by mesoscale events.For example, sea salt shows no clear weekly cycle, except for a peak on Fridays in intermediate and fine fractions coinciding with westerly winds, which coincidentally occurred more frequently on Fridays than on other days.Except for such events, regionally dominated elements tend to display flat, featureless diurnal/weekly cycles, while elements dominated by recurring local processes (e.g.traffic patterns) show interpretable features.Diurnal and weekly cycles of all other elements can be found in Supplement Figs.S8-9.For comparison, diurnal and weekly cycles of NO x and total PM 10 mass at all sites, and of traffic flow at MR are shown in Fig. 8.The time series of these species were averaged to the RDI collection times before obtaining the cycles.BC diurnal and weekly cycles (not shown) are very similar to those of NO x .

Regional influences
Elements dominated by regional sources (P, S, K, Br) occur mainly in the fine fraction and are similar to total PM 10 mass in showing no obvious diurnal and weekly patterns.This interpretation is consistent with the urban/kerb increment anal-ysis discussed in Sect.4.2.Weekly patterns suggest fine Zn and Pb are also dominated by regional transport (Fig. S9).P, S and K have been identified as tracers for mixed wood combustion and secondary sulfate (Amato et al., 2011;Richard et al., 2011), whereas Hammond et al. (2008) have identified S, K and Pb from mixed secondary sulfate and coal combustion.Br is usually associated with sea salt (Lee et al., 1994;Mazzei et al., 2007) or traffic emissions (Gotschi et al., 2005;Lee et al., 1994), but Maenhaut (1996) has also found Br, together with S, K, Pb and other elements in biomass burning.In this study, the diurnal cycle of fine Br is different from the Na, Mg and Cl cycles, but more similar to K. The Br is thus likely more associated with wood burning than with other sources.
The time series of fine S, K, Zn, Pb at NK (very similar at MR and DE) are explored in relation to total PM 10 mass, wind direction and air mass origin, and compared to representative elements from the other emission groups (coarse Na, Si, S, Sb; Fig. 9).Air mass origin was studied with back trajectories simulated for three case study periods (marine, European mainland and locally influenced) using the NAME model (Jones et al., 2007).Particles are released into the model atmosphere from the measurement location and their origin is tracked using meteorological fields from the Unified Model, a numerical weather prediction model.Each particle carries mass of one or more pollutant species and evolves by various physical and chemical processes during 24 h pre- ceding arrival at NK. Potential emission source regions can be highlighted along the pathway to the measurement site at 0-100 m above ground.Under marine air mass origin (case A, 18-24 January, Fig. 9) with strong W winds the concentrations of the fine mode elements are fairly low, whereas sea salt concentrations are enhanced (see Na in Fig. 9).Although the air mass has also passed over Ireland and the Midlands, the influence of these rather sparsely populated regions on pollution levels seems small.This is confirmed by low total PM 10 mass and NO x concentrations.Enhanced fine fraction and total PM 10 mass concentrations (latter not shown) occur during northeasterlies with high wind speeds from the European mainland (case B) bringing in pollutants through regional transport.
During this episode, both the urban background and rural site observed the highest concentrations for these trace elements of the entire campaign.Traffic-influenced species were not enhanced during this pollution episode.Elevated concentrations of all trace elements, NO x and PM 10 mass occurred only during a local pollution episode of roughly 3 days caused by local air mass stagnation over London and the southeastern UK (case C).The very high concentrations observed in case B through regional transport from the European mainland were identified as the main reason for PM 10 limit exceedances at urban background sites in London by Charron et al. (2007), while exceedances were much less frequent under marine-influenced air as represented by case A in this study.

Sea salt
The sea salt group yields comparable, rather flat diurnal cycles for fine and intermediate mode Na, Mg and Cl, and coarse mode Na and Cl (Na in Fig. 6 Fig. S8), and no obvious weekly patterns (Na in Fig. 7; others in Supplement Fig. S9).This indicates that the regional transport of sea salt is probably the main source of Na, Mg and Cl, as seen in case A in Fig. 9. Interestingly, although coarse mode sea salt exhibits no obvious temporal trend, the urban and kerb increments indicate additional source contributions besides regional transported sea salt.The urban increment might be caused by the natural sea salt gradient observed in the UK, with reducing concentrations from west to east (Fowler and Smith, 2000), while the kerb increment could be the result of road salt resuspension in addition to sea salt resuspension.Coarse mode Mg originates probably both from mineral dust and sea salt, because at MR with SW winds Mg correlates with Al and Si temporal trends, while with NE winds Mg correlates better with Na and Cl.

Mineral dust and traffic
Both mineral dust and traffic-related elements are strongly influenced by traffic patterns at MR, which are shown in Fig. 8 as the number of vehicles per 2 h split in LDV and HDV (shorter/longer than 5.2 m).HDV numbers peak in the morning, whereas LDV numbers peak in the evening when the flow of traffic leaves the urban area, consistent with Harrison et al. (2012b).A single peak during midday in the weekend compared to a double peak at weekdays is observed for LDV; HDV numbers show a similar pattern during weekdays, but with a reduced maximum on Saturday and a small The element diurnal (Fig. 6 for Si, Fe and Sb; Supplement Fig. S8 for others) and weekly (Fig. 7 for Si, Fe and Sb; Supplement Fig. S9 for others) cycles yield highest concentrations at MR and lower concentrations at NK and DE, consistent with observed urban and kerb increments.More importantly, and only retrievable with high time-resolved data, concentrations are higher during the day than at night, with night time concentrations at MR and NK similar to median urban background and rural concentrations, respectively, demonstrating the effects of local traffic and enhanced human exposure during daytime.Weekdays yield stronger increments than weekends and closely follow NO x and HDV traffic patterns (Fig. 8), indicating the strong influence of these vehicles on element concentrations.This confirms observations by Charron et al. (2007), who stated that PM 10 limit exceedances at MR are more likely to occur on weekdays, in combination with large regional contributions from the European mainland with easterly winds.Similarly, Barmpadimos et al. (2011) found strong weekly cycles for PM 10−2.5 and PM 2.5 mass concentrations in Switzerland over a 7-12 year period, with higher concentrations on weekdays and lowest on Sundays.
In the street canyon with SW winds, all coarse mode elements (including dust elements) except Na and Cl exhibit a double peak in the diurnal cycles, closely following the flow of traffic and confirming that traffic-related processes such as braking and resuspension dominate the concentration of most elements.With NE winds, source discrimination is possible between mineral dust (Si in Fig. 6) and traffic-related elements (Fe and Sb in Fig. 6).Mineral dust yields a strong maximum between 08:00 and 14:00 LT, and continued high concentrations throughout the day, while the traffic-related group yields a reduced double peak relative to SW winds.The increase in dust concentrations coincides with the start of traffic flows at 06:00 LT resulting in resuspension of particles within the street canyon.However, concentrations decrease before traffic flows reduce, possibly as a result of increased mixing and dilution during boundary layer growth.At NK diurnal and weekly patterns of the dust and traffic groups yield similar variability but reduced concentrations relative to MR, which suggests increased human exposure during day time and weekdays and confirms that traffic dominates urban background element concentrations in London (see Dore et al., 2003).At DE, freshly emitted pollutants from London and other cities in the southeastern UK have been diluted and mixed with other pollutants during their transport to the rural background, resulting in no obvious diurnal and weekly patterns independent of size range.
The kerb increments at MR under SW winds were divided into two traffic-related groups: brake wear and other trafficrelated elements.However, the diurnal and weekly cycles of all these elements correlate well and no obvious split into two groups is seen.Apparently, both groups are co-emitted as a single group under comparable vehicle fleet and/or set of driving conditions, at least on a 2 h timescale, but in different ratios at MR and NK.The ratio of these two element classes for SW to NE wind sectors at MR is almost 2, with the lack of difference between these classes supporting co-emission.In a future paper we will further explore the diurnal variability of emission sources at both sites with statistical analyses based on the Multilinear Engine (Canonaco et al., 2013;Paatero, 1999).

Conclusions
Aerosol trace element composition was measured at kerbside, urban background and rural sites in the European megacity of London during winter 2012.Sampling with rotating drum impactors (RDIs) and subsequent measurements with synchrotron radiation-induced X-ray fluorescence spectrometry (SR-XRF) yielded trace element mass concentrations in PM 10−2.5 , PM 2.5−1.0 and PM 1.0−0.3aerosol with a 2 h time resolution.Total median element mass concentrations of 4.1, 2.1 and 1.0 µg m −3 were found at kerbside, urban background and rural sites, respectively, being 14 % of total PM 10 mass (highest at kerbside; lowest at rural site), neglecting the corresponding oxides.The contribution of emission sources to coarse fraction elements was on average largest at kerbside (65 %) and reduced for urban background (52 %) and rural sites (49 %).
Urban and kerb increments were defined as the concentration ratios of urban background to rural, and kerbsideto-urban background, respectively, and the kerb increments were further explored as a function of wind direction.The group with the largest kerb increments consisted of elements typically associated with brake wear (Cu, Zr, Mo, Sn, Sb, Ba).The second largest kerb increments were observed for anthropogenically influenced elements typically assigned to non-brake wear traffic emissions (Cr, Mn, Fe, Zn, Pb) but also V and Ni.This could indicate either a traffic source for these elements or a similar kerbside-to-urban emission gradient.Kerb increments were larger for the brake wear group and under SW winds due to local street canyon effects, with coarse fraction increments between 10.4 and 16.6 for SW winds (3.3-6.9 for NE winds) against increments for the anthropogenically influenced group between 5.7 and 8.2 for SW winds (2.6-3.0 for NE winds).The kerb increments for all these elements in the PM 10−2.5 size fraction are roughly twice that of the PM 1.0−0.3fraction.Urban increments (no distinction between both groups) were around 3.0.In addition to direct emissions, traffic-related processes influence the concentrations of other elements by resuspension, with mineral dust (Al, Si, Ca, Ti, Sr) increments of 1.3-3.ments during periods of heavy traffic, whereas regionally influenced elements (fine P, S, K, Zn, Br, Pb) showed no enhancements.All traffic-related elements at the kerbside site yielded temporal patterns similar to variations in heavy duty vehicle numbers as opposed to total vehicle numbers, and resulted in enhanced exposure to elements during day time and weekdays.Traffic-related processes therefore exhibit a dominant influence on air quality at the kerbside and urban background sites, and should be the main focus of health effect studies and mitigation strategies.With technological improvements for the reduction of traffic exhaust emissions, the traffic contribution to coarse PM is becoming more important as shown by decreasing PM 2.5 mass trends with no significant changes of coarse PM (Barmpadimos et al., 2012).
Trace element and total PM 10 mass concentrations are also affected by mesoscale meteorology, increasing with the transport of air masses from the European mainland.Under these conditions, coarse and intermediate fraction trace elements are hardly affected, but fine fraction elements showed elevated concentrations.Trace element concentrations in London are therefore influenced by both local and regional sources, with coarse and intermediate fractions dominated by anthropogenic activities (particularly traffic-induced resuspension and wearing processes), whereas fine fractions are significantly influenced by regional processes.
These observations highlight both the strong influence of regional factors on overall air quality, as well as the need for detailed characterization of urban micro-environments for accurate assessment of human exposure to airborne particulates and the associated health risks.
The Supplement related to this article is available online at doi:10.5194/acp-15-2367-2015-supplement.

Figure 1 .
Figure 1.Map of southeastern UK.Indicated are the sampling sites MR (kerbside site Marylebone Road), NK (urban background site North Kensington), DE (rural site Detling), and the elevated BT Tower site for meteorological measurements (adapted from Google Maps).
at the Swiss Light Source (SLS) at Paul Scherrer Institute (PSI), Villigen PSI, Switzerland and at Beamline L at Hamburger Synchrotronstrahlungslabor (HASYLAB) at Deutsches Elektronen-Synchrotron (DESY), Hamburg, Germany (beamline dismantled November 2012).The samples with the deposited particles were placed directly into the Xray beam.Irradiation of the samples took place at a 45 • angle for 30 s.The light spot of the incoming beam was ∼ 140 by 70 µm at SLS (monochromatic excitation at 10.5 keV, in vacuum) and ∼ 80 by 150 µm at HASYLAB (polychromatic excitation, in air).Fluorescence light produced by the elements in the samples was detected by energy-dispersive detectors (silicon drift detector at SLS, nitrogen cooled Si(Li)detector at HASYLAB) at a 90 • angle relative to the incoming beam.At SLS Kα lines of the elements with atomic number Z = 11-30 (Na-Zn) were measured, and at HASYLAB Kα lines of the elements with Z = 22-56 (Ti-Ba) and Lα lines of Z = 82 (Pb).
Si P S Cl K Ca Ti V Cr Mn Fe Ni Cu Zn Br Sr Zr Mo Sn Sb Ba Pb 1

Figure 3 .
Figure 3. Mean, median and 25-75th percentile urban increment values for trace elements at NK relative to DE for PM 10−2.5 (top), PM 2.5−1.0 (middle) and PM 1.0−0.3(bottom).Note that the median of Zr in PM 10−2.5 is below detection limit.

Figure 4 .
Figure 4. Mean, median and 25-75th percentile trace element concentrations at MR split in four wind direction sectors (N, E, S, W) normalized to the global median concentration per element for PM 10−2.5 (top), PM 2.5−1.0 (middle) and PM 1.0−0.3(bottom).See Sect.4.2.2 for the definition of the wind direction sectors.

Figure 5 .
Figure 5. Mean, median and 25-75th percentile kerb increment values for trace elements at MR relative to NK for PM 10−2.5 (top), PM 2.5−1.0 (middle) and PM 1.0−0.3(bottom) split into SW and NE wind sectors.See Sect.4.2.2 for the definition of the wind direction sectors.

Figure 9 .
Figure 9. Top panel: time series of (top left axis) PM 1.0−0.3S, K, Zn and Pb concentrations at NK and (top right axis) wind direction from BT Tower, time series of (bottom left axis) PM 10−2.5 Na, Si, S and Sb concentrations at NK and (bottom right axis) total PM 10 mass concentration at NK; bottom panel: three NK footprints simulated with the NAME model corresponding to the vertical lines (A, B, C) indicated in the top panel.Trajectories are simulated for particles released from NK and followed back at 0-100 m a.g.l. for the previous 24 h at: (a) 23 January 2012 09:00 LT, (b) 31 January 2012 21:00 LT, (c) 6 February 2012 18:00 LT; particle concentrations increase from blue to red.
3.The highly time-resolved data enabled studying diurnal patterns.The cycles of mineral dust elements and coarse Na, Mg and Cl both indicate major concentration enhance-

Table 2 .
Self-absorption correction factors.Average density of the calibration standards and of ambient aerosol.The composition of calibration standard 1 is Na 3.76 Al 3.76 P 3.76 Cl 3.76 Ca 3.76 CoN 8 O 24 , of calibration standard 2 Mg 3.76 Si 3.76 S 3.76 K 3.76 Ca 3.76 CoN 7 O 21 , and of ambient samples C 39 H 29 N 10 O 18 S 3 Fe. b AF for Na to Ca are 0.52-0.87,0.30-0.84and 0.22-0.82for coarse, intermediate and fine fraction samples, respectively.

1
Mean conc.per size range to mean total PM10 conc.