Publication - Research and analysis

Air quality study: assessing variations in roadside air quality with sampling height

Published: 14 Aug 2015
Environment and Forestry Directorate
Part of:
Environment and climate change

A mobile air quality monitoring study commissioned to assess variations in roadside air quality.

3 Methodology

3.1 Monitoring Equipment

3.1.1 Mobile

Table 3.1 below details the monitoring equipment used for the Glasgow study. For PM 2.5, both automatic and gravimetric sampling was carried out using the Lighthouse IAQ 3016 and Harvard-PEMs sampler, respectively. Measurement of BC was carried using the Magee MicroAeth AE51 and UFP using the Philips NanoTracer. All four particulate samplers are handheld and are principally designed for personal exposure measurements.

In addition to the particulate analysers, measurements of NO 2, CO 2 and benzene were made using AQMesh, COZIR optical sensor and pumped tube sampling, respectively. Meteorology measurements of wind speed/direction, temperature, humidity, pressure and rainfall were made using the Lufft WS600. GPS measurements were recorded using the Garmin 110 GPS watch, which provided location and speed data every 20 s. Lastly, the sampling route was filmed allowing pollution events to be further investigated in addition to enhancing data visualisation.

All data were logged on a web-based logging system (web logger) or on the sampler's internal memory with integrated GPS and 3G telemetry.

Table 3.1 Mobile Monitoring Equipment


Instrument type


Particulate matter (PM)

Lighthouse IAQ 3016 PM analyser

This analyser provides fast response particle counts of 6 size bins from 10 microns down to 500 nm. It uses proprietary algorithms to estimate PM 10, PM 2.5 and PM 1 concentrations.

PM 2.5

Harvard-PEMS + BGI pump (Personal Exposure Monitor)

Ambient air is drawn through a filter at a constant flow rate of 4 l min -1. The filter is weighed pre and post-sampling and a mass concentration derived from the weight of material collected and the volume of air sampled.

Black Carbon

Magee AE51 MicroAeth

The spectrometer used in this analyser is identical to the spectrometer used in the Aethalometer used in the UK national research network. The MicroAeth is used as a transfer calibration standard for the research network analysers

Ultrafine Particles

Philips NanoTracer

This analyser detects and counts particles between 10 nm and 300 nm in size.


Lufft WS600

This portable met sensor makes high resolution measurements of wind speed, wind direction, temperature, pressure, relative humidity and rainfall.


Garmin 110

This watch accurately records position and time.

NO 2, SO 2 and CO


The sampler measured gaseous pollutant concentrations at 1-minute resolution.


Roadhawk Recorder

This camera records standard definition video at 15 frames per second.

The monitoring equipment was mounted on a purpose-modified industrial trolley, which was pushed around the chosen routes. The equipment was mounted at two fixed heights: 1.68 m, to reflect the typical breathing height of a Scottish adult (The Scottish Health Survey, 2009); and 0.80 m to reflect the breathing height of a child in a buggy. All data were recorded using high resolution 1-minute averaging. Figure 3.1 shows a schematic of the mobile monitoring trolley and equipment.

Figure 3.1 Mobile Monitoring Trolley Schematic

Figure 3.1  Mobile Monitoring Trolley Schematic

3.1.2 Fixed

Table 3.2 lists the measurement techniques utilised at 5 fixed monitoring sites that were used for co-location exercises (details provided in Section 3.2). Measurements of NO 2 are carried out using the EU reference method using chemiluminescence [2] . Measurements of PM 10 and PM 2.5 are carried out using a Filter Dynamics Measurement System [3] ( FDMS) at Glasgow Townhead, Glasgow Kerbside and Fife Dunfermline. This measurement technique has been assessed as equivalent to the EU reference method. Measurements of PM 10 concentrations were carried out using a Tapered Element Oscillating Microbalance [4] ( TEOM) at Perth Atholl Street and Waltham Crooked Billet. The TEOM has been assessed as not equivalent to the EU reference method however, these data can be corrected using the Volatile Correction Model (Green et al, 2008). This correction has been assessed as providing reference equivalent PM 10 data from TEOMs.

Table 3.2 Measurement Techniques - Fixed Sites

Colocation Site

NO 2 Measurement Technique

PM 10 Measurement Technique

PM 2.5 Measurement Technique

Glasgow Townhead




Glasgow Kerbside




Fife Dunfermline




Perth Atholl Street




Perth Atholl Street




3.2 Co-location Exercises

A number of colocation exercises were carried out to characterise both the intra-sampler responses (e.g. between Low and High samplers of the same type) and to assess the samplers against fixed point measurement methods for PM 10, PM 2.5 and NO 2.

Table 3.3 details the colocation exercises carried out as part of this study. A total of 8 exercises were carried out between February and September 2014. Glasgow Townhead, Glasgow Kerbside and London Marylebone Road monitoring sites are part of the UK Automatic Urban and Rural Network ( AURN - Note that a co-location of UFP only was carried out at the London Marylebone Road monitoring site. However, no comparison could be made to the on-site Scanning Mobility Particle Sizer Spectrometer ( SMPS) due to a fault with this analyser. In addition to the on-site colocations carried out at fixed monitoring sites, a mobile colocation exercise was carried out along Hope St, Glasgow on 13/08/2014.

A range of urban environments were used (urban background, roadside and kerbside) to try and maximise the range of pollutant concentrations measured by each sampler. In the case of the Hope Street mobile co-location exercise, all samplers were set up to sample ambient air at 0.8 m. A further 4 co-location exercises were carried out as part of parallel studies and are detailed in Table 3.4. Fife Dunfermline and Perth Atholl Street both form part of the Scottish Air Quality Database ( SAQD) network ( and Waltham Crooked Billet forms part a local authority monitoring site in London (

Table 3.3 Colocation Exercises as Part of this Study

Colocation Site

Site Type

Coordinates (Lat/Long)


Glasgow Townhead

Urban Background

55.865782, -4.243631


Glasgow Kerbside


55.859170, -4.258889


Glasgow Kerbside


55.859170, -4.258889


Hope Street, Glasgow




Glasgow Kerbside


55.859170, -4.258889


London Marylebone Road ( UFP)


51.522530, -0.154611


Glasgow Townhead

Urban Background

55.865782, -4.243631


Glasgow Kerbside


55.859170, -4.258889


Table 3.4 Colocation exercises as Part of Other Studies

Colocation Site

Site Type

Coordinates (Lat/Long)


Fife Dunfermline


56.073830, -3.448620


Perth Atholl Street


56.399327, -3.434182


Waltham Crooked Billet


51.601728, -0.016442


3.3 Mobile Monitoring Exercises

A total of 8 mobile monitoring exercises were carried out between March and August 2014, with Table 3.5 listing the day and dates of the exercises. All exercises were carried out using the same predetermine route, shown in Figure 3.2. The route was designed to incorporate as many urban microenvironments as possible (Table 3.6), such as street canyons, pedestrianised zones and busy streets. The route was approximately 2.6 miles in distance with one lap of the route taking approximately 1 hr 15 min to travel around. All exercises were carried out, where possible, on days with no rain and low wind speeds.

The mobile exercises were carried out between 0700 and 1900 during six weekdays (2 x Fridays) and 0900 to 1600 during two weekend days to ensure that pollutant concentrations during rush hour periods were recorded. This resulted in between five and ten loops of the route being completed during each exercise with a total 185 miles over 82 hours being covered as part of the study.

Table 3.5 Mobile Monitoring Exercises - 2014




Glasgow City Centre

















Figure 3.2 Mobile Monitoring Route - Glasgow City Centre

Figure 3.2 Mobile Monitoring Route – Glasgow City Centre

Table 3.6 Mobile Monitoring Route Details - Glasgow City Centre

Street name

Description of street on route

Approximate Length of street within study route (miles)

Hope Street

Busy urban canyon orientation south to north (partially restricted to buses and taxes)


Sauchiehall Street

Urban pedestrian precinct orientation west to east


Buchanan Street

Urban pedestrian precinct orientation north to south


St Vincent Street

Busy urban canyon orientation west to east


George Square

Busy urban street orientation west to east


George Street

Busy urban canyon orientation west to east


Montrose street

Busy urban canyon orientation north to south


Ingram Street

Busy urban street orientation west to east


High street

Busy urban street orientation north to south



Busy urban street orientation north to south


Clyde Street/Broomielaw

Busy Urban street orientation east to west


Oswald Street

Busy urban street orientation north to south


3.4 Quality Assurance/ Quality Control ( QA/ QC)

3.4.1 Fixed Monitoring Sites

Data from monitoring sites included within both the Automatic Urban and Rural Network and the Scottish Air Quality Database are underpinned by a comprehensive Quality Assurance/Quality Control regime. These QA/ QC procedures include:

Intercalibration and Audit procedures

The audit and intercalibration procedures adopted by Ricardo-AEA rely upon the principle that a set of recently certified gas cylinders (called "audit gas") is taken to all the stations in a monitoring network. This gas is certified at the Ricardo-AEA Gas Calibration Laboratory. At each station, analyser response to audit gas is recorded to check if the expected concentration (i.e. the certified value for the cylinder) is obtained. The analyser response to audit gas is obtained using calibration factors obtained from the site operator. The audit procedure checks the validity of the provisional data, the correct overall operation of the analyser and the reliability of calibrations undertaken routinely at that station. These site audit procedures are compliant with the requirements of the CEN standard methods of measurement and are used throughout the UK AURN network.

The results of the audit exercises form an integral part of the data management system and are fed directly into the data ratification process. After the audit exercise, data from all the stations visited are traceable to recently calibrated UKAS accredited gas calibration standards (the audit gas).

UKAS Accreditation

Ricardo-AEA holds UKAS accreditation to ISO 17025 for the on-site calibration of the gas analysers (NO X, CO, SO 2, O 3), for flow rate checks on particulate (PM 10) analysers and for the determination of the spring constant, k 0, for the TEOM analyser. ISO17025 accreditation provides complete confidence that the analyser calibration factors are traceable to national metrology standards, that the calibration methods are sufficient and fit for purpose, and that the uncertainties are appropriate for data reporting purposes.

Ricardo-AEA also holds ISO17025 accreditation for laboratory certification of NO, NO 2, CO and SO 2 gas cylinders.

Ratification tasks and output

When ratifying data the following are closely examined:

  • Issues that have been flagged up automatically by the software
  • Zero and sensitivity factors used on each day
  • General review of the result to make sure that there are no other anomalies.

Ratified Data Checking

Once the data have been initially ratified proforma reports are produced and passed to the data checker. The role of the data checker is to:

  • Assess if there are any station problems if not the data can be marked as ratified
  • Return the station to the data ratifier if there are any issues requiring further action by the data ratifier
  • Forward the report to the project Quality Circle if there are data quality issues which require a group discussion to resolve.

Following the Quality Circle meeting the data are then corrected if required and uploaded as ratified to the database and web site.

3.4.2 Mobile Monitoring Equipment

A similar QA/ QC approach to that that is described above was used for the mobile monitoring data with the following differences:


Data from AQMesh were uploaded real-time to a web-based server via 3G telemetry. Data acquisition was checked before, during and after each mobile exercise. No calibrations could be carried out in the traditional sense as this is not possible due to interferences and the algorithms used in data processing. Therefore, the performance of the AQMesh pods needed to be "characterised" at fixed sites by carrying out co-location exercises. The co-location exercises were used to correct NO and NO 2 data ( Appendix D). Once corrected, the mobile data were compared to the fixed site data and spurious data (e.g. negative data spikes) were removed.

For CO and SO 2 data no co-locations comparing mobile to fixed monitoring data were carried out. Data were adjusted for offset using the minimum measured CO and SO 2 concentrations as a reference.

AE51 Micro-Aethelometer

BC data were recorded on the sampler's internal memory and downloaded and validated at the end of each monitoring exercise. The flow rate of the samplers were set to 100 ml min -1, which was checked using a calibrated BIOS Defender flowmeter [5] at the beginning of each exercise. In addition, a new filter was used for each monitoring exercise. Negative data points were removed during final data ratification.

A potential further adjustment of AE51 data may have been required had the relationship between the attenuation of light, filter loading and the calculated mass concentration not being linear. As a result, there is potential for the AE51 to underestimate black carbon concentrations as the filter loading increases (Virrkula et al, 2007). However, it was found that no such correction was necessary. Appendix C details the algorithm and the analysis used to determine whether the BC data required this correction.


Patriculate Matter data were recorded on the sampler's internal memory and downloaded and checked at the end of each monitoring exercise. The flow rate of the samplers were set to 4 l min -1, which was checked using a calibrated BIOS Defender flowmeter at the beginning of each exercise. A further zero check was carried out using a disposable filter unit attached to the sample inlet to check that PM concentrations dropped to zero.

No calibrations could be carried out, however, the colocation exercises were used to correct PM 2.5 and PM 10 data ( Appendix E).


UFP data were recorded on the sampler internal memory and downloaded and checked at the end of each monitoring exercise. The flow rate of the samplers were set to 300 ml min -1, which was checked using a calibrated BIOS Defender flowmeter at the beginning of each exercise. Negative data were removed during final data ratification.


Ambient air was drawn through a filter at a constant flow rate of 4 l min -1, which was checked every 30 - 60 mins during the monitoring exercises. The flowrate, start time and finish time were recorded. Pall Corporation Teflo™ PTFE Membrane filters with plastic ring (Pore = 2 μm, Dia. = 37 mm) were used. All filters were weighed in line with BS EN 12341:2014 [6] .

The filters were uniquely identified and conditioned in the weighing room at 19 °C to 21 °C and 45 % RH to 50 % RH for ≥ 48 hr pre and post-sampling.

Benzene Pumped Sampling

Ambient air was sampled through analytical desorption tubes ( ATD) at a constant flow rate of 250 ml min -1, which was checked every 30 - 60 mins during the monitoring exercises. The flowrate, start time and finish time were recorded. Environmental Scientifics Group ( ESG) were used to supply and analyse the samples. Standard preparation and sample measurement was carried out according to UKAS accredited method ASC/ SOP/236 Issue 2 (

3.5 Data Analyses

A number of analysis software packages have been utilised for this study:

  • Openair (Carslaw, Ropkins, 2012)
  • Microsoft Excel 2013
  • ArcGIS

The following analyses have been carried out:

  • Regression analyses.
  • Outlier Testing.
  • Uncertainty calculation.
  • Summary statistics.
  • Time-series plots.
  • Wind rose/polar plots.
  • Contour plots summarising the spatial distribution of 1-minute average pollutant concentrations.

3.5.1 Regression Analyses

Regression analysis was used to investigate the relationships between monitors sampling at 0.80 m (LOW) and 1.68 m (HIGH), the relationships between mobile and fixed monitoring, and cross-pollutant relationships. In order to take account of the uncertainty associated with each component (x and y) of the analysis, orthogonal regression was used as opposed to linear regression, which only takes into account of the variation in the y-axis.

Three key statistics are provided in the regression model: the slope (b) and intercept (a) of the regression line and the coefficient of determination (r 2). Both the slope and intercept provide a relationship of the form y = a+bx where x and y are the data pairs, where x is always the recorded measurements from the LOW samplers and y from the HIGH samplers. The value of r 2 is given between 0 - 1 with a value closer to 1 indicating a stronger correlation and more accurate model between the two datasets under investigation. For this study, the following strength of correlation are defined:

  • 0 - 0.2 = weak, slight
  • 0.2 - 0.4 = mild/modest
  • 0.4 - 0.6 = moderate
  • 0.6 - 0.8 = moderately strong
  • 0.8 - 1.0 = strong

Appendix A details the calculations used to carry out orthogonal regression.

3.5.2 Grubbs' Outlier Test

The resultant regression model from the comparison of two datasets may consist of outliers and although these outliers are valid data they may unduly influence the regression model. As a result, Grubbs' outlier test was used to identify and remove potential outliers.

For the purposes of this study and due to the large numbers of data points collected, Grubb's' outlier test was used iteratively until no more outliers were identified. This resulted in data rejection of between 0% and 7.6% - Appendix B details Grubbs' Outlier Test

3.5.3 Uncertainty Analyses

A number of uncertainty calculations were used within this study:

  • The between-sampler uncertainty ( U bs).
  • The Mean Absolute Error ( MAE).
  • The expanded relative uncertainty ( W).
  • Uncertainty in slope (u b) and intercept (u a).

The between-sampler uncertainty provides an indication of the error between two measurements taken by two samplers during the colocations. The MAE is the error in the measured values as compared to the regression model and has been reported for the pollutant vs height results. The relative expanded uncertainty has been calculated using the fixed monitoring versus mobile monitoring samplers colocations. In this case charts plotting the calculated W versus the measured concentration have been provided for NO 2, NO, PM 10 and PM 2.5. The standard CEN/ TS 16450:2013 [7] was used as a guide.

The calculated uncertainty in slope and intercept are provided in all regression analyses and Appendix A details the uncertainty calculations used.

3.5.4 Statistics

Openair uses R (, a computing language and environment for statistical computing and graphics. The Openair project ( provides analytical tools for use with air quality data and can be used to quickly analyse large amounts of data.

Summary statistics are provided in tabular and chart form and consist of the minimum, quartiles, mean and maximum.

3.5.5 Geographical Information System

ArcGIS 10 was used to create the contour plots measured of 1-minute average pollutant concentrations at each of the four study areas using inverse distance weighted ( IDW) interpolation. This provides a visualisation of the spatial variation of pollutant concentrations throughout study route.

Please note that the concentrations shown in the contour plots are indicative as they are the result of a geo-processing routine in ArcGIS. Also, the pollution gradients can be seen to be wider than the roads, which are an artefact of the GPS accuracy and the GIS processing. When interpreting the plots all concentrations shown are from roadside measurements and indicate potential exposure a pedestrian might experience when walking along the road.

Figure 3.3 shows the GPS points recorded during all mobile sampling exercises. On average, three GPS measurements were recorded every minute. Each point was then assigned a pollutant concentration e.g. three GPS points within the same minute would be assigned the same associated 1-minute average pollutant concentration. A total of 33,753 points were recorded during the 8 mobile monitoring exercises.

All map data within this report Contains Ordnance Survey data © Crown copyright and database right 2015

Figure 3.3 Mobile Monitoring Sampling Points

Figure 3.3 Mobile Monitoring Sampling Points
Contains Ordnance Survey data © Crown copyright and database right 2015

3.6 Equipment Issues

A number of issues with equipment were encountered during the monitoring programme. This section details these problems with associated impacts regarding data quality and analyses.

3.6.1 AQ Mesh

Overall, during the study, the AQMesh samplers performed well. However, it was not possible to use the samplers for one mobile monitoring exercise on 23/06/2014 due to an aerial fault. Although this reduced the number of 1-minute average data available for analyses by approximately 600, over 3700 data points were recorded. It is therefore thought that this loss of data has had minimal effect on the final results.

In addition, the NO 2 sensor used within the AQMesh is known to suffer from cross sensitivity with ozone (O 3), resulting in an overestimation of NO 2 concentrations at higher O 3 concentrations. Accurate measurements of O 3 are required to correct for this and are available at the Glasgow Townhead monitoring site. However, the Townhead site is located in an urban background location and therefore O 3 concentrations measured there will at times differ considerably from concentrations measured at a roadside location. This has added uncertainty to the regression analysis between AQMesh and fixed site (chemiluminescence) data ( Appendix D) due to a lack of accurate O 3 measurements within the Glasgow town centre.

3.6.2 Meteorological Measurements

A temporary fault with the meteorological mast bracket resulted in an inability to measure microsite weather conditions during one mobile monitoring exercise on 10/04/2014. This resulted in the loss of approximately 600 1-minute average measurements. However, a total of 3193 measurements were taken and therefore it is considered that this loss has had a minimal effect on the final results.

3.6.3 COZIR

The COZIR CO 2 sensor response was reset at the beginning of each monitoring exercise. This was done by assuming that CO 2 concentrations were close to 500 ppb prior to commencement of each exercise. Unfortunately, it was found that the sensor reset was not consistent between the two sensors (0.80 m and 1.68 m) and therefore no consistent relationship can be inferred between the two sensors from the intra-sampler regression results (shown in Table 3.7 and Figure 3.4). As a result, no further analyses were carried out using the CO 2 data.

Table 3.7 Intra-Sampler Orthogonal Regression Results - CO 2




1 minute

Orthogonal Regression

n bs

u bs

r 2

Slope (b) ± u b

Intercept (a) ± u a

Colocations Exercises













































All Exercises

All Data










Figure 3.4 Intra-Sampler Orthogonal Regression Lines - CO 2 ( g m -3)

Figure 3.4 Intra-Sampler Orthogonal Regression Lines – CO2 (g m-3)

3.6.4 PM 2.5 Gravimetric

The results from gravimetric sampling of PM 2.5 using the Harvard-PEMS samplers are detailed in Table 4.3.3. Inconsistencies were found in the results from mobile monitoring exercises. In order to attempt to identify and rectify the problem, flow measurements were taken at ½ hourly intervals during the monitoring exercises and was adjusted back to 4 l min -1 where necessary. No significant fluctuations in flow rate were recorded however, and a possible cause was attributed to the use of a power supply without a voltage regulator causing fluctuations in the flowrate. Although analyses have been carried out using these data, the fault has increased the uncertainty of the results.