Identifying drivers for the intra-urban spatial variability of airborne particulate matter components and their interrelationships

(cid:1) Peripatetic measurements of UFP, BC and PM 2.5 were taken in an urban environment. (cid:1) Spatial variability in UFP and BC were much larger than in PM 2.5 . (cid:1) UFP and BC were signi ﬁ cantly correlated with traf ﬁ c counts, while PM 2.5 was not. (cid:1) PM 2.5 variability was largely determined by synoptic meteorological in ﬂ uences. The aim of this work was to compare the variability in an urban area of ﬁ ne particles (PM 2.5 ), ultra ﬁ ne particles (UFP) and black carbon (BC) and to evaluate the relationship between each particle metric and potential factors (local traf ﬁ c, street topography and synoptic meteorology) contributing to the vari- ability. Concentrations of the three particle metrics were quanti ﬁ ed using portable monitors through a combination of mobile and static measurements in the city of Edinburgh, UK. The spatial variability of UFP and BC was large, of similar magnitude and about 3 times higher than the spatial variability of PM 0.5-2.5 (the PM size fraction actually quanti ﬁ ed in this work). Highest inter-daily variability was observed for PM 0.5-2.5 , which was approximately 2 times higher than inter-daily variability of BC and UFP. Elevated concentrations of UFP and BC were observed along streets with high traf ﬁ c volumes whereas PM 0.5-2.5 showed less variation between streets and a footpath without road traf ﬁ c. Both BC and UFP were signi ﬁ cantly correlated with traf ﬁ c counts, while no signi ﬁ cant correlation between PM 0.5-2.5 and traf ﬁ c counts was observed. BC was signi ﬁ cantly correlated with UFP, with signi ﬁ cantly different regression slopes between working days and non-working days implying that the increased number of diesel powered heavy goods vehicles during working days contributed more to BC than to UFP. It is concluded that variations in BC and UFP concentrations were mainly determined by the nearby traf ﬁ c count and varying background concentrations between days, while variation in PM 0.5-2.5 concentration was mainly associated with regional sources. These ﬁ ndings imply the need for different policies for managing human exposure to these different particle components: control of much BC and UFP appears to be manageable at local scale by restricting traf ﬁ c emissions; however, abatement of PM 2.5 requires a more strategic approach, in cooperation with other regions and countries on emissions control to curb long- range transport of PM 2.5 precursors.


Introduction
Evidence continues to accumulate of the adverse health impacts of PM 2.5 , the mass concentration of airborne particulate matter (PM) with an aerodynamic diameter of less than 2.5 mm (WHO, 2013). However, metrics of other characteristics of ambient PM including numbers of ultrafine particles (UFP, particles of diameter <100 nm) and black carbon (BC) concentrations are emerging as important in terms of their association with health effects (Heal et al., 2012). A relevant issue is the extent to which UFP and BC concentrations vary within populated areas, since a shortcoming in many epidemiological studies is assumption of homogenous exposure within the study area. This might be plausible for pollutants with less spatial variability but could result in significant bias in exposure-response relationships for highly spatially variable pollutants (Hoek et al., 2002). In this context, variables related to the contribution of emissions from major roads (e.g. traffic intensity, or distance to the road) are commonly identified as significant predictors for a range of traffic-related air pollutants in many studies applying land-use regression models (Hoek et al., 2008), the validity of which might be influenced by the underlying causes of the variability of different pollutants. Thus one of the aims of this work was to evaluate the extent to which potential factors affect the spatiotemporal variability of ambient BC, UFP and PM 2.5 in an urban area. These factors include local traffic, street topography and synoptic meteorology which, although recognised in the literature, have rarely been compared in terms of their influences on different metrics.
The three airborne particle metrics are closely related to traffic in urban environments (HEI, 2010;Kassomenos et al., 2014;Sandradewi et al., 2008) but, for PM 2.5 in particular, synopticscale meteorology also affects the dispersion and long-range transport of secondary particles (Pinto et al., 2004). UFP variability is also subject to high intensity secondary formation associated with strong solar radiation (Reche et al., 2011). Street canyons, which are ubiquitous in many urban environments, introduce complex dispersion characteristics that further increase the spatial variability of BC and UFP (Peters et al., 2014;Rakowska et al., 2014). One limitation of inter-and intra-urban studies of airborne particle concentrations is that the fixed-site measurements on which they are based are rarely sufficient in number, and thus in spatial coverage, to explore the variability of exposure to particles at street level. One way to monitor particle concentration at high spatiotemporal resolution is by use of mobile monitoring instruments, which has good prospects for wider application in the assessment of human exposure to air pollution in the future (Steinle et al., 2013).
Both UFP and BC have received increasing interest in recent studies (Patton et al., 2014;Ruths et al., 2014), as they can be considered markers of a range of traffic-related particulate pollutants. Therefore the relationships of UFP and BC with traffic volume and composition need to be understood in order to correctly assign exposure to traffic pollution in health studies. In the UK and many other countries, PM 2.5 and PM 10 are the only two regulated PM metrics (AQEG, 2012). Given the increasing evidence for the harmfulness of UFP (WHO, 2013) with its ability to penetrate deep into the airways (Knibbs et al., 2011), investigation on the relationship between UFP and PM 2.5 can provide insight on the extent to which current policy can effectively protect human health. Thus another aim of this work was to investigate the inter-relationships between the different metrics of PM and their relationships with traffic.
In this work, pairs of portable instruments were used to measure PM 0.5-2.5 (used here as a measure of PM 2.5 ), UFP number and BC concentrations within the city of Edinburgh (Scotland) in two series of measurement campaigns in winter and in spring. Analyses of data from a combination of mobile and stationary measurements were used to evaluate possible causes of the variations in the concentrations of the different PM metrics.
In the winter campaign, between December 2013 and January 2014, the measurements were conducted three times on Mondays and once on Sunday primarily near roadside by walking between and pausing at designated sites (Fig. 1a). The sites were selected to cover potential hotspots, urban background sites (at least 130 m from the nearest major road) and different street topographies (open or built-up) over an area of about 6 km 2 . In a typical walk, the measurements started at around 10:00 a.m. and proceeded through the designated sites to the final location roughly in the order from south to north and from east to west (Fig. 1a). At each site a 5-min static measurement was conducted, during which the number and type of vehicles (car, van, heavy goods vehicle and bus) passing the observer were recorded. Throughout each walk, measurements for each pollutant were taken in parallel (with duplicate instruments) on both sides of the road. To evaluate the duplicate precision, intercomparison between the pairs of instruments was conducted in a separate trial on Mon 3rd Feb 2014 from 10:00 a.m. to 12:00 p.m. by walking through the same route but with the duplicate instruments carried by one person. The weather conditions on this day were similar to other measurement days (Supplementary Information  Table S1). Traffic characteristics are assumed to be similar to other Monday measurements.
In the spring campaign, between April and May 2014, measurements were taken around a park area (~1 km 2 ), referred to locally as 'the Meadows', focusing on understanding the contributions from traffic-related sources and local background sources to BC, UFP and PM 0.5-2.5 on typical urban streets (Fig. 1b). The road on the south edge of the Meadows had an annual average daily flow of 13,272 vehicles in 2013 (Dft, 2015). The pollution level associated with traffic was monitored by walking along a route surrounding the Meadows. To measure temporal variation in the background concentrations, a duplicate set of instruments was located at a static site inside the route circuit during the collection of the mobile measurements. This background location had perpendicular distance between 160 and 480 m to the three sides of the triangle route (Fig. 1b). The measurements were conducted on one Sunday and five weekdays. On each day the mobile measurements started together with the static measurements at the background site and proceeded in the directions indicated in Fig. 1b. Two trips were carried out in the morning (~9e10 a.m.) and early afternoon (~1e2 p.m.) during each day, except for adverse weather conditions on one of the weekday afternoons. Other incomplete sets of measurements during each day were due to instrument faults. The route of the mobile measurements was divided into five segments with different street topographies and traffic densities, as labelled on Fig. 1b. The total traffic passing the observer in the direction of the route was counted for each segment. Traffic flow in the opposite direction of the route is assumed to be similar. The duplicate instruments were compared against each other during the last four measurement days by co-location for at least 20 min either at the static background site or near a busy roadside (Fig. 1b). The inclusion of both background and roadside sites was to cover a range of concentrations for the evaluation of duplicate instrument precision.

Instrumentation
The AE51 determines BC concentration from absorption of 880 nm laser light by particles continuously collected on a glassfibre filter. The CPC measures particle number concentrations of particles between 0.01 and 1 mm in diameter by using laser light scattering after condensing particles with super-saturated isopropanol vapour. Although UFP is usually defined as particles smaller than 100 nm, since the number concentration is dominated by ultrafine particles the measurement from the CPC can be considered to represent UFP number concentration. The Dylos measures the particle number concentration using laser light scattering technique in two size ranges, >0.5 and >2.5 mm. Only particles between 0.5 and 2.5 mm in diameter were included in this study and are thus referred to as PM 0.5-2.5 . The use of the terminology PM 2.5 elsewhere in this paper refers to the mass of all particles <2.5 mm as defined for air quality standards.
In the winter campaign the time bases for the AE51, CPC and Dylos were 1 min, 1 s and 1 min, respectively. Because of the shorter duration of a trip in the spring campaign the resolution for the AE51 was increased to 30 s to ensure sufficient data points for each segment. The Optimised Noise-reduction Averaging (ONA) algorithm was used to reduce the noise in the data recorded by the AE51 (Hagler et al., 2011). The ONA algorithm conducts adaptive timeaveraging of the BC data, with the incremental light attenuation (DATN) through the instrument's internal filter determining the time window of averaging. The DATN thresholds were set at 0.01 and 0.05 for winter and spring measurements, respectively, as a result of the different proportions of clean background areas and sampling resolutions in two campaigns. Negative values recorded after the smoothing were omitted from further analyses (consisting of~5% of the whole data set), which mostly occurred when the measured concentrations were <100 ng/m 3 .
Major axis (MA) regression analysis was carried out to test the equivalence between duplicate instruments, assuming that the uncertainties in the duplicate instruments are similar (Warton et al., 2006). A statistical summary of instrument intercomparison results is given in SI Table S2. Correlations between duplicate instruments were highly significant, however the 95% confidence interval of the slopes for AE51 and CPC in spring campaign and Dylos in both campaigns did not encompass unity. Therefore corrections based on the slope and intercept from the MA regression analyses were applied to the corresponding instruments to allow comparison between duplicate instruments.

Additional data
The average PM 2.5 concentrations measured by a TEOM-FDMS instrument at the UK national network monitoring station St. Leonards (55.945589 N, 3.182186 W) located 600 m to the east of the Meadows during each set of measurements are summarised in SI Fig. S1. The TEOM-FDMS is a reference-equivalent instrument for quantifying gravimetric PM 2.5 for statutory purposes. Meteorological data for the period of measurement for each day in both campaigns and for the period when instrument inter-comparisons were carried out in the winter campaign were obtained from a weather station on the rooftop of a seven storey building located 3 km to the south of the Meadows (55.92 N, 3.17 W) and are summarised in Table S1.

Data analyses
Reduced major axis (RMA) regression analysis was used to investigate the correlation between different pollutants, since the magnitudes of the data values and their uncertainties are not the same for different instruments (Ayers, 2001). Given the skewed nature of pollutant distributions, the non-parametric Man-neWhitney U test and the KruskaleWallis test were used to determine whether median concentrations differed significantly between two and more than two samples, respectively. Data analyses were performed mainly using R software (R Core Team, 2014) and sometimes Microsoft Excel. Back trajectory data was imported from pre-calculated trajectory data using the HYSPLIT trajectory model via the "importTraj" function in openair (Carslaw and Ropkins, 2012), an R package for air quality data analysis.

Results and discussion
3.1. Spatiotemporal variability of BC, UFP and PM 0.5-2.5 Distributions of BC, UFP and PM 0.5-2.5 concentrations in the winter measurements are summarised in Fig. 2. The distributions of  (Table S2). Side of road is defined with respect to the walking direction in the mobile measurements. The UFP concentrations in (b) are 1 min averages of the raw 1 s data. pollutant concentrations are highly skewed, especially for BC and UFP, so the interquartile range (IQR) is used to illustrate the variation in distribution. Statistical summaries of the median and IQR for measurements on both sides of the road are listed in Table 1a. BC and UFP concentrations in spring campaign and PM 0.5-2.5 concentrations in both campaigns were corrected according to the MA regression analyses results in Table S2 to allow comparison between duplicate instruments. The ratio between IQR and median is used here as a metric of the spatiotemporal variability of each pollutant during each measurement trip. Table 1a shows that BC had the highest variability in the winter campaign (average IQR/ median ratio of 1.34), which is about twice as high as for the metric with lowest variability, PM 0.5-2.5 (average IQR/median ratio of 0.56). The difference in concentrations measured on two sides of the road was not significant for BC and UFP as the median on one side of the road was always within the IQR of the other side (Fig. 2). PM 0.5-2.5 Table 1 The median, IQR and ratio between IQR and median on each day for BC, UFP and PM 0.5-2.5 for (a) winter and (b) summer campaign. Left and right side of the road is defined with respect to the walking direction in the mobile measurements. Adjustment derived from MA regression analyses of instrument co-deployments (Table S1) was applied to one set of PM 0.5-2.5 data in both campaigns, and one set of BC and UFP data in the spring campaign.  (Table S2). Fig. S2 suggests that the relationship between the duplicate Dylos instruments might follow a different linear relationship between different days as indicated by the deviation of scatter points on one day from the main trend line. Due to lack of inter-comparison between the Dylos instruments during each measurement day, it is difficult to quantify the genuine difference in PM 0.5-2.5 concentrations between the two sides of road in Fig. 2c. The discrepancies in the relationship between duplicate CPC and AE51 units on different days were considered to be less significant than the Dylos, since the correlation coefficients between CPC and AE51 units (r > 0.97) in the mobile winter inter-comparison were much higher and the gradients (0.98 and 0.97, respectively) were much closer to unity than the Dylos equivalents (r ¼ 0.90, gradient ¼ 0.44) ( Table S2). Distributions of BC, UFP and PM 0.5-2.5 concentrations in the spring measurements are summarised in Fig. 3. Jittered points are plotted for UFP to reveal the overlapping of large numbers of data points. In order to discern the details of the boxes in Fig. 3a and b, BC and UFP concentrations greater than 15,000 ng/m 3 and 100,000 cm -3 , respectively, are not included. However, extremely high concentrations were observed, extending to 50,000 ng/m 3 for BC and 400,000 cm -3 for UFP. Elevated BC concentrations were observed on streets with traffic (route segments 1e4) compared with the footpath through an area of urban park (route segment 0). The median concentrations for each route segment also varied between different times of a day. Only on Thu 10th Apr 2014 were BC concentrations not significantly different between morning and noon sessions (ManneWhitney U test). BC concentration on Sun 6th Apr 2014 (median ¼ 251 ng/m 3 ) was significantly lower than on weekdays (median ¼ 1281 ng/m 3 ) (one-tailed ManneWhitney U test, P < 0.01), which is similar to findings in the winter  Fig. 2. Additional jitter points are plotted in Fig. 2b to reveal the extent of data in the outliers.

measurements.
The large range of outliers in the UFP distributions (Fig. 3b) implied that UFP was greatly influenced by the emissions of nearby traffic. Despite the fact that the outlier measurements on each street were usually a factor 3 higher than their median, the difference in the medians between streets was relatively small, suggesting that elevated levels due to occurrence of exhaust plumes only transiently affected the UFP concentration. Using ratios between median concentrations on streets and on the footpath it was calculated that, on average, a pedestrian was exposed to 1.8 times higher concentration on streets than on the footpath for UFP, and 2.3 times higher for BC. The markedly high UFP level on Fri 18th Apr 2014 at noon was associated with the lowest wind speed, highest solar flux (Table S1) and relatively high O 3 concentration of~70 mg/ m 3 (recorded at the St. Leonards UK national network monitoring station) compared to other days. All these conditions are likely to have promoted secondary particle formation from photochemical reactions (Reche et al., 2011).
The distribution of PM 0.5-2.5 concentrations (Fig. 3c) exhibited a different pattern compared with the BC and UFP concentrations. The range of concentration on each street was smaller, and the relative variation in medians between streets within the same day was lower. Only on Sunday and Wednesday 1 were PM 0.5-2.5 concentrations highly significantly different between footpath and the streets (ManneWhitney U test, P < 0.01), indicating that local traffic is frequently not a dominant contributor to PM 0.5-2.5 concentrations in the locations where measurements were made. Possible trafficrelated sources to PM in this size range would be tyre and brake wear and road abrasion, but these are considered to contribute relatively little to PM 2.5 (AQEG, 2012). Significant variation in PM 0.5-2.5 concentrations was noticed between all spring working sampling days (KruskaleWallis test, c 2 ¼ 297, df ¼ 3, P < 0.01), which implies that PM 0.5-2.5 is more influenced by the regional sources and meteorological conditions on a particular day rather than the local traffic emissions. This is consistent with observations in other cities, e.g. relatively low contribution (13%) to PM 2.5 from local traffic was also found in Paris (Skyllakou et al., 2014). Despite the Dylos not measuring particles smaller than 0.5 mm, PM 0.5-2.5 variation between days was in very good relative agreement with that of the PM 2.5 concentration measured by TEOM-FDMS at the St Leonard's national network urban background site (SI Fig. S1), i.e. highest PM 2.5 levels on Mon 27th Jan 2014 in winter, and on Wed 23rd Apr 2014 and Wed 30 th Apr 2014 in spring, and lower on Sundays in both campaigns.
Assuming that UFP represents contribution from local traffic sources, the ratio between median UFP and PM 0.5-2.5 approximates the relative contribution of local sources to PM 2.5 concentration during that day. However caution should be made in this interpretation since any photochemical new particle formation (e.g. Fri 18th Apr 2014 in spring campaign) may lead to overestimation of the relative level of local contribution compared to days when such an event is unfavourable. In the winter measurements, the highest UFP/PM 0.5-2.5 ratio (3999), was associated with relatively low PM 2.5 concentration (8 mg/m 3 ) on Sunday (Fig. 4). In the spring measurements, the highest ratio was on Fri 18th Apr 2014 (3715), which was about ten times the ratio on Wed 23rd Apr 2014 (349), despite the fact the PM 2.5 concentration on Fri 18th Apr 2014 (~9 mg/m 3 ) was only a third of that on Wed 23rd Apr 2014 (~27 mg/m 3 ). It is noted that in both campaigns highest local contributions did not coincide with the highest PM 2.5 concentrations. This observation again indicates a strong regional component in PM 2.5 observed in urban areas, as also noted by AQEG (2012).
Four-day air-mass back trajectories arriving at Edinburgh at 09:00 a.m. and 12:00 p.m. on each measurement day are plotted in SI Fig. S3. The trajectories associated with the highest PM 2.5 concentrations in the winter and spring campaigns originated from North America and northern continental Europe, respectively. However, the highest PM 2.5 concentration in winter coincided with a relatively large contribution from local sources (UFP/PM 0.5-2.5 ratio of 2682) suggesting that the high PM 2.5 on that particular day may be mainly due to local emission sources. In contrast, the southeasterly trajectory originating from northern continental Europe (Fig. S3) was associated with high PM 2.5 concentrations and low local contributions. It is notable that the PM 2.5 concentrations on the days associated with relatively high local emissions were close to the Scottish annual air quality objective threshold concentration for PM 2.5 of 12 mg/m 3 to be achieved by 2020 (AQEG, 2012), and that the two days with PM 2.5 concentrations breaching the limit (Weds 23rd and 30 th Apr 2014 in spring, Fig. S1) had the largest regional contribution. In the UK context, transboundary import of inorganic aerosol components resulting in high particle concentrations was reported in a modelling study (Vieno et al., 2014).
The evidence presented here highlights that the effective management of PM 2.5 in the UK may require international scale cooperation in the control of emission sources. The good agreement between Dylos and TEOM-FDMS, together with the above analysis on the potential sources of high PM 2.5 episodes during each campaign, confirm that the Dylos is capable of measuring elevated PM 2.5 arising from both regional and local influences (Steinle et al., 2015). However, this analysis does not reveal the contribution from local traffic emission at street level. This is in line with the findings of Price et al. (2014) who showed that number concentration of particles greater than 262 nm was more closely related to meteorological conditions whilst UFP was more closely associated with traffic variables. Nevertheless this work presents a novel approach to apportioning PM 2.5 to local and regional sources by comparing the relationship between UFP and PM 2.5 with the help of backtrajectory analysis.
The IQR/median ratio was used to represent the variability of each pollutant in the spring measurements (Table 1b), in which the IQR/median ratio from the static measurements quantifies only the temporal variation between different times of a day, and the ratio in the mobile measurement reflects both spatial and temporal variation between different streets (with the former expected to be greater than the latter). Consistent with the observations in the winter measurements, the spatial variations of BC and UFP (average mobile IQR/median ratios of 1.05 and 1.08, respectively) were much larger than spatial variations of PM 0.5-2.5 (average mobile IQR/ Fig. 4. Ratios between median UFP and PM 0.5-2.5 concentrations and average PM 2.5 concentration measured by TEOM-FDMS on each day. The ratios in the spring campaign were calculated from the static measurements. The same instruments were used to calculate the ratios in the winter campaign for the purpose of comparison between the ratios in the two campaigns. median ratio of 0.29). The within-day variations of BC and UFP (range of static IQR/median ratios 0.14e2.92 and 0.26e1.08, respectively) were also generally larger than equivalent within-day variations of PM 0.5-2.5 (range of static IQR/median ratios 0.06e0.33) but were of varying magnitudes. This implies that changes in local traffic counts and atmospheric conditions within a day have more pronounced effects on variations in BC and UFP than on variations in PM 0.5-2.5 . However PM 0.5-2.5 varied more between days (static IQR/median ratio for the whole spring campaign of 3.27) than BC and UFP (equivalent ratios of 1.77 and 2.19, respectively). Collectively this evidence suggests that the variability of all three PM metrics is subject to the varying background concentrations. For BC and UFP in particular, the geographical locations, namely proximity to trafficked roads, also contribute a significant part of the spatial variability. Contrary to another study (Sullivan and Pryor, 2014), which found that the spatial variability of PM 2.5 (defined as the relative standard deviation for the mobile measurements on different routes) was 2e3 times greater than the sub-daily temporal variability (defined as the RSD for the measurements when stationary), results from this study suggest that the spatial variability of PM 0.5-2.5 was of similar magnitude to the sub-daily temporal variability. Possible explanations of this discrepancy include the larger geographical area in the former study, and the potential for bias from a few extremely high concentrations when using RSD rather than IQR as an indicator for variability.

Pollutant concentration in relation to street topography and traffic counts
To understand the relationship between traffic counts and the pollutant concentration, reduced major axis (RMA) regression analyses were conducted between the mean of the concentrations on both sides of the road at each spot measurement in the winter campaign and traffic counts. Fig. 5 shows the scatter plots of BC and UFP against traffic counts grouped by the classification of each site. Correlations of 5-min average BC and UFP concentrations with traffic counts were moderate and highly significant (r ¼ 0.56 and 0.39, respectively, P < 0.01, n ¼ 72). PM 0.5-2.5 was not significantly correlated with traffic counts (r ¼ 0.17, n ¼ 72).
Ratios between mobile and static measurements in the spring campaign were calculated for each timestamp to represent the elevation in the pollutant concentration due to traffic. The distribution of the ratios for BC and UFP were grouped by streets and plotted in Fig. 6. Only ratios in the range 0e30 are shown in Fig. 6 to avoid extreme values skewing the plots. Median mobile/static ratios are higher on Sun 6th Apr, Thu 10th Apr and Wed 7th May 2014 than on other days, an observation more pronounced for BC than for UFP. These days coincide with the days with higher wind speed (Table S1), which explains the greater contrast between roadside and background concentrations as the high wind speed facilitates the mixing of clean air at the background site while the immediate roadside concentration still stays at relatively high level. A summary of the median values of the mobile/static measurement ratio for each street and the traffic density on each street are tabulated in Table 2. The data show a common pattern in the mobile/static ratios for BC and UFP: streets 2 and 3 had the highest median mobile/ static ratios. Considering that the traffic density in streets 2 and 3 were not the highest among all the streets, the difference in street topography is the most likely explanation of the greater mobile/ static ratios. Streets 2 and 3 are mostly characterised by street canyons with aspect ratios in the range of 0.6e1.3, whereas streets 1 and 4 are beside the Meadows urban park with an open terrain. As a result, dispersion in streets 2 and 3 is likely to be reduced compared to dispersion in streets 1 and 4. Similar results were reported in a study in Hong Kong characterised by high-rise buildings, where notably elevated BC and UFP concentrations were observed in deep street canyons compared to an open road although the traffic flow were significantly lower in street canyons (Rakowska et al., 2014).
The correlation of averaged BC and UFP concentration with traffic counts for each route segment was calculated in the spring campaign. BC was again significantly correlated with traffic counts (r ¼ 0.45, P < 0.01, n ¼ 50), but UFP was not (r ¼ 0.25, n ¼ 48, P ¼ 0.09). The lower correlation in spring compared to winter campaign could be attributed to the fact that traffic counting on one side of the street in spring campaign may not have represented total traffic composition sufficiently. The low correlation between traffic count and UFP may have been caused by the secondary particle formation mentioned in Section 3.1. This interpretation was supported by the increased correlation when the data from Friday was excluded (r ¼ 0.66, P < 0.01, n ¼ 39). Despite the moderate to high correlation coefficients of BC and UFP with traffic counts found in this study, mixed conclusions have been drawn in the literature for the relationships between traffic and BC or UFP as a result of characteristics of specific measurement sites, the consistency of traffic flow and the formation/transformation of particles governed by environmental conditions (Kumar et al., 2008;Peters et al., 2014;Price et al., 2014;Rakowska et al., 2014). Further investigation of relationships between BC, UFP and traffic would provide beneficial information to inform the potential use of nearby traffic flow data to predict BC and UFP concentrations. Street topography effects on BC and UFP concentrations are similarly important for further investigation.

Correlation between BC and UFP
The correlations between BC and UFP were evaluated using RMA regression. The results summarised in Table 3 were calculated from 1-min averaged concentrations with BC as the y variable and UFP as the x variable.
The correlations between BC and UFP during each week ranged between 0.49 and 0.77 and in all cases were significant (P < 0.01). During working and non-working days, slope coefficients were not significantly different between measurements taken on different sides of the road in the winter campaign or between mobile and static measurements in the spring campaign. The agreement of the BC/UFP relationship between roadsides or between busy roads and local background suggests that BC and UFP concentrations vary similarly as they disperse away from traffic sources. The slopes on non-working days were significantly lower than the slopes on working days, indicating a decrease in BC/UFP ratios on non-work days. This variation in the BC/UFP ratios was not only observed in the measurements taken near road but also in the static background measurements during the spring campaign. The reason for this variation in BC/UFP relationship is likely due to the decrease in heavy goods vehicles (HGV) on the road during the weekend as HGV had the lowest share on weekends (3.2%) compared with weekdays (~6.8%). This observation suggests that HGV contribute relatively more to BC than to UFP concentrations. Therefore a policy targeting on reduction of UFP may not be effective if it only focuses on restricting HGV. On the other hand policies aimed at reduction of BC should focus on controlling HGV emissions, which has direct implications for instance for the design and implementation of Low Emission Zones in urban areas. Considering that recent epidemiological studies suggested that BC was more strongly associated with harmful particle substances than was the PM 2.5 mass (Grahame et al., 2014;WHO, 2013), traffic control strategy targeting at the HGV may effectively contribute to  Table S2. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) alleviation of population health burdens arising from urban air pollution.

Study limitations
This paper describes design and interpretation of a pilot study to investigate potential drivers for the spatiotemporal variability of different particulate matter components in the urban environment. It is recognised that it may be limited in terms of giving exhaustive consideration and quantification of each individual driver on pollutant concentration, particularly the complex influence of local street topography. The absence of summer measurements could mean that the influence of photochemical particle formation may be underestimated (Ma and Birmili, 2015). Overall, however, this study has demonstrated that careful experimental design and data interpretation of short-term mobile measurements can identify the important drivers governing different metrics of PM concentration, and which also support findings in literature that use more sophisticated long term measurements.

Conclusions
Distributions of BC, UFP and PM 0.5-2.5 concentrations in the urban environment and the relationship between concentrations and potential influences (local traffic, street topography and synoptic meteorology) were studied through a combination of mobile and static measurements in the south of the city of Edinburgh, UK.
BC and UFP exhibited a high spatial variability in the urban environment, roughly 3 times greater than that of PM 0.5-2.5 ; however PM 0.5-2.5 had the highest inter-daily variability. Very high BC and UFP concentrations were frequently measured at roadside and sometimes exceeded 3 times the street median concentration. However the median BC and UFP concentrations measured on the streets were not greatly influenced by the frequent spikes and were on average approximately double the median measured on the footpath. PM 0.5-2.5 did not show a consistent elevated concentration in the streets in comparison with the footpath. Both geographical locations and varying background concentrations had important effects on the BC and UFP concentrations. Variation in PM 0.5-2.5 concentrations was largely influenced by regional sources, although local sources also contributed to a lesser extent.
Both BC and UFP were highly correlated with traffic counts. PM 0.5-2.5 showed very low correlation with the traffic counts. Significant difference was found between the slopes of BC vs. UFP regression analyses between working days and non-working days. During non-working days the reduction of HGV flows results in a decrease in the BC/UFP ratio, suggesting that HGV may contribute more to BC than to UFP concentrations. Therefore control of HGVs may be effective in reducing the negative health effects associated with BC. PM 2.5 was observed in a few instances to have a large background component, the variability of which may not be well accounted for by only using local sources as indicators. Therefore international cooperation in control of emission sources is important for PM 2.5 management in the UK. Furthermore, this study indicates that the proximity to the road as a primary indicator for the assessment of exposure to air pollutants is highly dependent on the pollutant being considered. Table 3 RMA regression analyses for 1-min averaged BC and UFP concentrations. The shading in the table represents the data collection period (grey: working days; white: nonworking days). Left and right side of the road is defined with respect to the walking direction in the mobile measurements. ** indicates correlation at >99% significance.