concentration, composition and sources of particulate ...€¦ · concentration, composition and...

Concentration, composition and sources of particulate matter in the Johnstone’s Hill Tunnel, Auckland

Authors Perry Davy

Bill Trompetter

Andreas Markwitz

GNS Science Consultancy Report 2010/296 November 2011 FINAL

Project Number: 624W1012

DISCLAIMER

This report has been prepared by the Institute of Geological and Nuclear Sciences Limited (GNS Science) exclusively for and under contract to the New Zealand Transport Agency. Unless otherwise agreed in writing by GNS Science, GNS Science accepts no responsibility for any use of, or reliance on any contents of this Report by any person other than the New Zealand Transport Agency and shall not be liable to any person other than the New Zealand Transport Agency, on any ground, for any loss, damage or expense arising from such use or reliance.

BIBLIOGRAPHIC REFERENCE

Davy, P.K..; Trompetter, W.J.; Markwitz, A. 2011. Concentration, composition and sources of particulate matter in the Johnstone’s Hill Tunnel, Auckland, GNS Science Consultancy Report 2010/296. 60p.

November 2011 Confidential 2011

GNS Science Consultancy Report 2010/296 i

CONTENTS

EXECUTIVE SUMMARY ........................................................................................................ IV

1.0 INTRODUCTION .......................................................................................................... 1

1.1 Background ................................................................................................................... 1 1.2 Scope ............................................................................................................................ 2 1.3 Objectives ...................................................................................................................... 2 1.3 Report structure ............................................................................................................. 3

2.0 SAMPLE COLLECTION AND ANALYTICAL METHODOLOGY ................................ 4

2.1 Particulate matter sampling ........................................................................................... 4 2.2 Elemental analysis of airborne particles ........................................................................ 5 2.3 Receptor modelling, data analysis and data reporting .................................................. 5

3.0 MONITORING SITE AND PM10 SAMPLING RESULTS .............................................. 8

3.1 Site description .............................................................................................................. 8 3.2 Air particulate matter sampling ...................................................................................... 9 3.3 In-tunnel sampling location and set-up ....................................................................... 10 3.4 Johnstone’s Hill tunnel conceptual receptor model ..................................................... 11 3.5 Local meteorology ....................................................................................................... 12 3.6 PM10 concentrations .................................................................................................... 12

4.0 RECEPTOR MODELLING OF PM10 IN JOHNSTONE’S HILL TUNNEL .................. 15

4.1 Analysis of PM10 .......................................................................................................... 15 4.2 Composition of PM2.5 and PM10-2.5 ............................................................................... 15 4.3 Source contributions to PM10 in Johnstone’s Hill tunnel .............................................. 18 4.4 Weekend and weekday variations in sources of PM10 ................................................ 24

5.0 DISCUSSION OF RECEPTOR MODELLING RESULTS .......................................... 26

5.1 Light duty vehicles ....................................................................................................... 26 5.2 Heavy commercial vehicles ......................................................................................... 27 5.3 Smoky vehicles ............................................................................................................ 29 5.4 Road dust .................................................................................................................... 30 5.5 Biomass burning .......................................................................................................... 30 5.6 Marine aerosol ............................................................................................................. 30

6.0 SUMMARY OF THE PM10 MONITORING AND RECEPTOR MODELLING ............. 32

7.0 GLOSSARY ............................................................................................................... 34

8.0 REFERENCES ........................................................................................................... 35

APPENDICES ........................................................................................................................ 40

APPENDIX 1 ANALYSIS METHODOLOGY ............................................................ 41

A.1.1 Ion Beam Analysis ....................................................................................................... 41 A.1.1.1 Particle Induced X-Ray Emission .................................................... 42 A.1.1.2 Particle Induced Gamma-Ray Emission ......................................... 43 A.1.1.3 Particle Elastic Scattering Analysis ................................................. 44 A.1.1.4 IBA Data reporting ........................................................................... 45 A.1.1.5 Limits of Detection for elements determined by IBA ....................... 46

A.1.2 Black carbon measurements ....................................................................................... 47 A.1.3 Receptor Modelling Methodology ................................................................................ 48

A.1.3.1 Principal Components Analysis ...................................................... 48 A.1.3.1.1 PCA model outline ....................................................................... 48 A.1.3.1.2 Limitations of PCA ....................................................................... 50

A.1.3.2 Positive Matrix Factorisation ........................................................... 50 A.1.3.2.1 PMF model outline ....................................................................... 51 A.1.3.2.2 PMF model used .......................................................................... 52 A.1.3.2.3 PMF model inputs ........................................................................ 52


GNS Science Consultancy Report 2010/296 ii

A.1.4 Dataset Quality Assurance .......................................................................................... 54 A.1.4.1 Mass reconstruction and mass closure ........................................... 54 A.1.4.2 Dataset preparation ......................................................................... 56

A.1.4.2.1 PMF data matrix population ......................................................... 56 A.1.4.2.2 PMF uncertainty matrix population .............................................. 56

APPENDIX 2 Statistics and Diagnostics for Receptor Modelling ...................... 58

FIGURES

Figure 1 Average source contributions to PM10 in Johnstone’s Hill northbound tunnel for June 2010 ............ iv Figure 2 Average hourly motor vehicle contributions to PM10 in Johnstone’s Hill tunnel for June 2010 ........... v Figure 3.1 Map showing location of Johnstone’s Hill twin tunnels monitoring site () (Source: Wises

Maps www.wises.co.nz). ....................................................................................................................... 8 Figure 3.2 Aerial view of Johnstone’s Hill tunnel (----) (Source: Google Maps 2010). ............................................. 9 Figure 3.3 Schematic of Johnstone’s Hill tunnel showing the location ( ▄ ) of PM10 monitoring

equipment in the northbound tunnel ................................................................................................. 10 Figure 3.4 Pillar box in the north-bound Johnstone’s Hill tunnel ....................................................................... 11 Figure 3.4 Wind rose for April 2010 - July 2010 (data supplied by Watercare Services Limited) ...................... 12 Figure 3.5 PM10 (BAM) concentrations in Johnstone’s Hill (northbound) tunnel: a) 15 minute average; b)

1-hour average; c) 24-hour average ................................................................................................ 14 Figure 4.1 Box and whisker plot of PM2.5 elemental concentrations in Johnstone’s Hill tunnel ......................... 17 Figure 4.2 Box and whisker plot of PM10-2.5 elemental concentrations in Johnstone’s Hill tunnel ...................... 18 Figure 4.3 Source profiles and elemental concentrations for PM10 in Johnstone’s Hill tunnel ........................... 20 Figure 4.4 Average (3-hourly) source contributions to PM10 in Johnstone’s Hill tunnel for June 2010 .............. 22 Figure 4.5 Temporal variation of source contributions to PM10 mass in Johnstone’s Hill tunnel ....................... 23 Figure 4.6 Aggregate* time-series of vehicle source contributions to PM10 mass in Johnstone’s Hill

tunnel (*light duty vehicles, heavy commercial vehicles, smoky vehicles plus road dust) ................ 24 Figure 4.7a Weekend and weekday variation of source contributions to PM10 mass in Johnstone’s Hill

tunnel ............................................................................................................................................... 25 Figure 4.7b Weekend and weekday variation of vehicle counts in Johnstone’s Hill tunnel for light duty

(L1) and heavy commercial vehicles (HCV) respectively ................................................................. 25 Figure 5.1 3-hour average light duty vehicle numbers for June 2010 ................................................................ 27 Figure 5.2a 3-hour average heavy commercial vehicle numbers for June 2010 (same scale as for Figure

5.1) ................................................................................................................................................... 27 Figure 5.2b 3-hour average heavy commercial vehicle numbers for June 2010 (expanded scale) ..................... 28 Figure 5.3 External wind speed and tunnel air speed compared to heavy commercial vehicle PM10

contributions for June 2010 in Johnstone’s Hill tunnel ..................................................................... 28 Figure 5.4 Ambient air temperature inside in Johnstone’s Hill tunnel from April to July 2010 ........................... 29 Figure 5.5 External wind speed compared to marine aerosol PM10 contributions for June 2010 in

Johnstone’s Hill Tunnel .................................................................................................................... 31 Figure A1.1 Air particulate matter analysis chamber with its associated particle detectors ................................. 41 Figure A1.2 Schematic of the typical IBA experimental set-up at GNS. .............................................................. 42 Figure A1.3 Typical PIXE spectrum for an aerosol sample analysed by PIXE. ................................................... 43 Figure A1.4 Typical PIGE spectrum for an aerosol sample analysed by PIGE. .................................................. 44 Figure A1.5 PESA spectrum for an aerosol sample showing the hydrogen peak at 1.250 MeV. ........................ 45 Figure A1.6 Elemental limits of detection routinely achieved at the GNS IBA facility for air filters with

PIXE. ................................................................................................................................................ 46 Figure A2.1 Scatterplot matrix of Johnstone’s Hill tunnel PM2.5 elemental data .................................................. 58 Figure A2.2 Scatterplot matrix of Johnstone’s Hill tunnel PM10-2.5 elemental data ............................................... 59 Figure A2.3 Plot of PM10 reconstructed mass versus gravimetric mass for Johnstone’s Hill tunnel PM

samples ............................................................................................................................................ 60 Figure A2.4 Scree plot of PCA Eigenvectors for PM10 in Johnstone’s Hill tunnel used to help identify the

number of different sources ............................................................................................................. 62

TABLES Table 2.1 Standards, guidelines and targets for airborne particulate matter in New Zealand ............................ 6 Table 2.2 Interim NZTA In-Tunnel Air Quality (Visibility) Guidelines .................................................................. 7 Table 3.1 Statistics for PM10 concentrations in Johnstone’s Hill Tunnel ........................................................... 12 Table 4.1 Elemental analysis results for 3-hourly PM2.5 samples in Johnstone’s Hill tunnel

(220 samples). LOD = limit of detection ........................................................................................... 16 Table 4.2 Elemental analysis results for 3-hourly PM10-2.5 in Johnstone’s Hill tunnel (220 samples) LOD

= limit of detection ............................................................................................................................ 17


GNS Science Consultancy Report 2010/296 iii

Table 4.3 Elemental composition of source profiles and contributions to 3-hourly average PM10 in Johnstone’s Hill tunnel ..................................................................................................................... 19

Table A1.1 Proton scattering energies of various elements for a 2.5 MeV proton beam .................................... 45 Table A2.1 Reconstructed mass components and mass closure for PM10 in Johnstone’s Hill tunnel ................ 60 Table A2.2 PCA factor loadings matrix for PM10 in Johnstone’s Hill tunnel ........................................................ 61 Table A2.3 PMF diagnostics for Johnstone’s Hill tunnel (prior to any rotation) .................................................. 63


GNS Science Consultancy Report 2010/296 iv

EXECUTIVE SUMMARY

The NZ Transport Agency (NZTA) manages the New Zealand state highway system, which now includes six major road tunnels, and is currently reviewing the management of these existing tunnels to assist in the design of future upgrades and to set policy to cover new tunnels. As part of the review, the NZTA has embarked on a programme of detailed air quality monitoring for each tunnel. The first campaign was undertaken for the Mount Victoria and the Terrace tunnels in Wellington during 2008. The second campaign focussed on Johnstone’s Hill twin tunnels on State Highway 1 north of Auckland from March to July 2010 and measured a range of traffic and air quality parameters in and around the tunnel.

This report presents the results of the particulate matter monitoring component conducted by GNS Science in the Johnstone’s Hill northbound tunnel and examines the concentrations of fine particulate matter, the elemental composition of those particles and the application of receptor modelling techniques to determine the various sources of particulate matter present in the tunnel.

Fine particulate (PM10) was measured continuously using a beta attenuation monitor located 30 metres in from the exit of the northbound tunnel between 29 April and 14 July 2010. Recorded concentrations were compared against guidelines and standards for short-term and long-term exposure. Source apportionment was carried out at the same location using a Streaker sampling system. This collected discrete three-hourly particulate samples throughout June 2010 which were subsequently analysed for their composition to establish the main sources influencing in tunnel particulate concentrations.

The source apportionment study of PM10 in the Johnstone’s Hill tunnel derived six primary sources that contributed to particulate matter concentrations – light duty vehicles, heavy commercial vehicles, smoky vehicles, re-suspended road dust, biomass (wood) burning, and marine aerosol (sea salt). The results showed that vehicle emissions (exhaust and brake dust) and road dust were the predominant sources of PM10 in the tunnel. On occasion, sources of PM10 from outside the tunnel, mainly marine aerosol (seasalt) had a significant influence on particulate matter concentrations in the tunnel. Figure 1 presents the average source contributions to PM10.

Figure 1 Average source contributions to PM10 in Johnstone’s Hill northbound tunnel for June 2010

Motor vehicle sources of PM10 were found to be dominated by heavy commercial vehicle emissions and re-suspended road dust. Heavy commercial vehicles were responsible for

Light duty vehicles

7%

Heavy commercial

vehicles25%

Smoky vehicles

7%

Road dust28%

Biomass burning

4%

Marine aerosol29%

0.8

3.0

0.8

3.3

0.4

3.4

0

2

4

6

8

Lig

ht

du

tyv

ehic

les

He

avy

com

mer

cial

veh

icle

s

Sm

oky

ve

hic

les

Ro

ad d

ust

Bio

mas

s b

urn

ing

Mar

ine

aero

solP

M10

mas

s co

ntr

ibu

tio

n (g

m-3

)


GNS Science Consultancy Report 2010/296 v

66% of the vehicle related PM10 (i.e. excluding the road dust component) and most of the nitrogen oxides concentrations (see Chapter 4.3). Heavy commercial vehicle numbers only constituted 10% of the total traffic volume passing through the tunnel. Light duty vehicles (90% of traffic volume), mainly petrol fuelled, contributed 17% of the vehicle related PM10, and were responsible for the majority of carbon monoxide present in the tunnel (see Chapter 4.3). A separate motor vehicle source was attributed to smoky vehicles due to a distinct zinc signature. This source was considered to be due to poorly tuned or old vehicles that burn lubricating (crankcase) oils and two stroke engines for which lubricating oils are part of fuel formulation.

The re-suspended road dust component was found to comprise 42% of the total vehicle related PM10 concentrations in the tunnel. The road dust source demonstrated a regular diurnal pattern as all vehicles passing through the tunnel re-entrain dusts on the road surface and this explains the regularity seen in PM10 from this source. Figure 2 shows the average daily variation in PM10 concentration attributed to motor vehicle sources together with the total PM10 concentration. Peak PM10 concentrations associated with motor vehicles occur during the middle of the day, and are aligned with peak heavy commercial vehicle traffic through the tunnel.

Figure 2 Average hourly motor vehicle contributions to PM10 in Johnstone’s Hill tunnel for June 2010

Overall, in-tunnel PM10 concentrations (maximum 32 g m−3 24-hour average) did not exceed the New Zealand National Environmental Standard (50 g m−3 24-hour average). However, the New Zealand National Environmental Standard relates to a 24-hour exposure period and does not apply to tunnels. Exposure in the tunnel will be of a much shorter duration (i.e. minutes rather than hours) for which there are no readily available air quality criteria for the protection of vehicle occupants and workers. From a health and safety perspective minimising risk is probably most important.

This study showed that good ventilation (in this case natural forcing by outside wind speeds discussed in Section 5.2) was effective in keeping in-tunnel PM10 concentrations attributed to motor vehicle exhaust emissions low and that road dust is effectively suppressed by surface wetting. For tunnel maintenance and design purposes, road surface composition and cleaning can minimise the generation and re-suspension of surface dusts.

0

5

10

15

20

25

00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00

PM

10co

nce

ntr

atio

n (g

m-3

)

Hour of Day

Heavy commercial vehicles

Light duty vehicles

Smoky vehicles

Road dust

Total PM10


GNS Science Consultancy Report 2010/296 vi


GNS Science Consultancy Report 2010/296 1

1.0 INTRODUCTION

This report presents the results of particulate matter monitoring in the north-bound tunnel of the twin Johnstone’s Hill road tunnels on State Highway 1 (SH1) north of Auckland. The monitoring programme was conducted over the period from 29 March 2010 until 14 July 2010 on behalf of the New Zealand Transport Agency (NZTA)

NZTA manages the New Zealand state highway system, which includes five major road tunnels, and is currently reviewing the management of these existing tunnels to assist in the design of future upgrades and to set policy to cover new tunnels. As part of the review, the NZTA has embarked on a programme of detailed air quality monitoring for each tunnel. The Johnstone’s Hill tunnel study was the second detailed road tunnel air quality monitoring campaign, the first study having been undertaken in the Mt Victoria and Terrace tunnels in Wellington during 2008.

This report presents the results of the particulate matter monitoring component conducted by GNS Science and examines the concentrations of fine particulate matter in the tunnel, the elemental composition of those particles and applied receptor modelling techniques to determine the various sources of particulate matter present in the tunnel.

1.1 Background

The state highway network consists of 10,894 kilometres of state highways, which account for around half of the 36 billion vehicle kilometres travelled each year, including six major road tunnels:

the Terrace and Mt Victoria tunnels in Wellington;

the Lyttelton tunnel near Christchurch;

the Homer tunnel which provides access to Milford Sound;

the Johnstone’s Hill twin tunnels northwest of Orewa;

the Victoria Park tunnel in central Auckland.

Motor vehicles emit a range of air pollutants such as fine particulates (PM10), nitrogen dioxide (NO2), carbon monoxide (CO), and volatile organic compounds, which are associated with adverse human health effects. In the case of tunnels, vehicle emissions impact both in-tunnel air quality and external (ambient) air quality, the latter via stacks or portals depending on the ventilation employed. Concentrations of traffic-related contaminants are likely to be many times higher inside as opposed to outside the tunnel.

NZTA is committed to minimising the environmental impacts of the state highway network in its environmental plan. Specific objectives relating to air quality include:

A1 Understand the contribution of vehicle traffic to air quality



A2 Ensure new state highway projects do not directly cause national environmental standards for ambient air quality to be exceeded

A3 Contribute to reducing emissions where the state highway network is a significant source of exceedances of national ambient air quality standards

NZTA has developed interim guidelines for in-tunnel air quality to ensure that tunnel air quality meets safety criteria for human health and visibility. External (ambient) air quality is monitored against existing national and international ambient air quality guidelines or standards designed to offer an acceptable minimum level of human health protection.

1.2 Scope

This report covers the particulate monitoring (source apportionment) component of a broader tunnel monitoring campaign undertaken at the Johnstone’s Hill twin tunnels between March and July 2010.

The Johnstone’s Hill tunnels were opened in January 2009 and are comprised of two identical but separate uni-directional tunnels. The tunnels are 15 metres apart and are built to carry two lanes each, however, the northbound tunnel has only one lane in operation as the highway north of the tunnel is currently only single lane. Each tunnel has longitudinal ventilation, which is triggered when in-tunnel levels of CO reach 30 ppm. The Johnstone’s Hill tunnels are part of a toll road which offers a more direct route for vehicles travelling on State Highway 1 between the Orewa turnoff and Puhoi.

1.3 Objectives

The main objectives of the particulate monitoring undertaken by GNS Science were to:

Monitor the concentrations of PM10 in the tunnel over a period of three months;

Determine the elemental composition of PM10 in the tunnel;

Identify the sources contributing to PM10 concentrations in the tunnel.

This report describes the sampling, the results and the outcomes according to these objectives. GNS Science examined the concentrations of fine particulate matter in the tunnel, the elemental composition of those particles and applied receptor modelling techniques to determine the various sources of particulate matter present in the tunnel. This information is critical to ensuring that ventilation in the tunnel is adequate and that aspects of traffic management and tunnel maintenance are appropriate for the effective control of particulate matter pollution in the tunnel.

Recorded PM10 concentrations were also compared to ambient air quality standards and guidelines.



1.3 Report structure

The report is structured as follows:

Chapter 2 provides a brief overview of the analytical techniques and methodology used for PM10 monitoring, sample collection, elemental determination and the receptor modelling analysis. Appendix 1 contains a detailed description of the analytical methodology.

Chapter 3 describes the Johnstone’s Hill tunnel air quality monitoring site, monitoring system set-up, PM10 monitoring results and local meteorology.

Chapter 4 presents elemental concentrations in PM10 and the receptor modelling results for PM10, including temporal variations.

Chapter 5 discusses the results and implications for air quality management in the tunnel

Chapter 6 provides concluding remarks and recommendations where appropriate



2.0 SAMPLE COLLECTION AND ANALYTICAL METHODOLOGY

Elemental concentrations in the particle samples were determined by ion beam analysis (IBA) techniques at the New Zealand Ion Beam Analysis Facility at Gracefield, Lower Hutt. IBA is a mature analytical technique that provides a non-destructive determination of multi-elemental concentrations present in a sample. Using elemental concentrations coupled with appropriate statistical techniques and purpose-designed mathematical models, the sources contributing to each ambient sample can be identified. In general the more ambient samples that are included in the analysis the more robust the receptor modelling results.

2.1 Particulate matter sampling

Continuous monitoring of particulate matter (PM10) in the tunnel was conducted using a beta attenuation monitor (BAM) which measures the mass of particles collected on a filter tape by the attenuation of high-energy electrons passing through it. For this study a MetOne E-BAM system was used (http://www.metone.com/documents/E-BAM_Datasheet_Rev_Aug09.pdf). The MetOne E-BAM is a self-contained and highly portable system capable of providing short-term PM10 concentrations (e.g. 15 min averages), although for comparative analyses a minimum of 1-hour averages are recommended in order to minimise analytical noise.

Particulate matter samples for elemental determination were collected using a Streaker sampling system that is a clockwise rotating filter sampler (Figure 2.1) designed to collect high temporal resolution PM2.5 and PM10-2.5 samples. Each sample of particulate matter is collected as a discrete sample with a rectangular area approximately 3 mm x 8 mm. The differing darkness of the bands represents the variation in air particulate matter concentrations for each discrete time period (in this case hourly). For the Johnstone’s Hill Tunnel study, hourly and three-hourly resolution samples were collected. The Streaker sampler and sampling methodology are described in (Formenti, Annegarn et al. 1997; Filippi, Prati et al. 1999; D'Alessandro, Lucarelli et al. 2004; D'Alessandro, Nava et al. 2004; Trompetter, Davy et al. 2007).



Figure 2.1 (Left) the Streaker sampler system; (Right) Typical Streaker sample filter showing hourly particle samples collected as discrete bands

All particulate matter sampling and systems maintenance at the sampling site was carried out by GNS Science and all records of equipment, flow rates and sampling methodologies used for the particulate matter sampling regime are available for review.

2.2 Elemental analysis of airborne particles

Ion Beam Analysis (IBA) was used to measure the elemental concentrations of particulate matter on the size-resolved filter samples from the tunnel monitoring site. IBA is based on the measurement of X-rays and -rays characteristic and particles of an element produced by the ion-atom interaction using high-energy protons in the two to five million electron volt (MeV) range. IBA is a mature and well developed science with many research groups around the world using IBA in a variety of routine analytical applications including analysis of atmospheric aerosol (Maenhaut and Malmqvist 2001; Trompetter, Markwitz et al. 2005). IBA techniques do not require sample preparation and are fast, non-destructive and sensitive (Cohen 1999; Maenhaut and Malmqvist 2001; Trompetter, Markwitz et al. 2005).

A detailed description of the methodology for the elemental analysis of particulate matter samples is provided in Appendix 1.

2.3 Receptor modelling, data analysis and data reporting

Measuring the mass concentration of particulate matter (e.g. as PM10) provides little or no information on the identity of the contributing sources. Airborne particles are composed of many elements and compounds from various sources. Receptor modelling provides a means to determine the relative mass contribution of sources that impact significantly on the total mass of air particulate matter samples collected at a monitoring site.



Receptor modelling is a multivariate factor analysis methodology for determining the sources contributing to air particulate matter and is based on the assumption that chemical species (e.g. elements) from the same source are correlated and that grouped correlations (factors) represent an emission source of particulate matter. Receptor modelling provides a per-sample estimate of the mass contribution to particulate matter concentrations from each source identified as impacting on a monitoring site. A detailed description of the methodology for the receptor modelling is provided in Appendix 1.

The results of receptor modelling have been reported in a manner that provides as much information as possible on the relative contributions of sources to particulate matter concentrations so that it may be used for tunnel air quality management and tunnel maintenance. The data have been analysed to provide the following outputs:

1. mass of elemental species apportioned to each source;

2. source elemental profiles;

3. average particulate matter mass apportioned to each source;

4. temporal variation of source mass contributions (time-series plots);

5. analysis of source contributions during peak particulate matter pollution events.

Table 2.1 presents the relevant standards, guidelines and targets for air particulate matter pollution. Note that these standards and guidelines apply to ambient concentrations and for averaging times that do not typically apply to the short-term exposure periods that vehicle occupants or workers may be exposed to in a tunnel. It was estimated that for vehicles travelling at the speed limit (80 km/h) through the Johnstone’s Hill tunnel, it would take approximately 14 seconds to traverse the 300 m tunnel length.

Table 2.1 Standards, guidelines and targets for airborne particulate matter in New Zealand

Particle size Averaging

Time

Ambient Air Quality

Guideline

MfE* ‘Acceptable’ air quality category

National Environmental

Standard

NES

(Allowable Exceedences per annum)

PM10

24 hour 50g m-3 33g m-3 50g m-3 1

Annual 20g m-3 13g m-3

PM2.5 24 hour 25g m-3 17g m-3

*Ministry for the Environment air quality categories taken from Ministry for the Environment, October 1997. Environmental Performance Indicators: Proposals for Air, Fresh Water and Land

NZTA has developed interim guidelines for in-tunnel air quality to ensure that tunnel air quality meets safety criteria for human health and visibility as presented in Table 2.2. The guideline provides a surrogate measure for particulate matter and is primarily intended to



manage potential road safety issues inside tunnels by ensuring adequate visibility is maintained in front of vehicles.

Table 2.2 Interim NZTA In-Tunnel Air Quality (Visibility) Guidelines

Traffic Situation

Guideline Averaging time

Extinction Coefficient

K/m

Transmission

(Beam Length: 100m)%

Fluid peak traffic

50 – 100 km/h 0.005 60

Daily congested traffic, standstill on all lanes

0.007 50

Exceptional congested traffic, Standstill on all lanes

0.009 40

Planned maintenance work in a tunnel under traffic

0.003 75

Closing of the tunnel 0.012 30

The interim visibility guidelines adopted by NZTA is based on the World Road Association (PIARC) recommendations.



3.0 MONITORING SITE AND PM10 SAMPLING RESULTS

3.1 Site description

Of the six tunnels that NZTA currently manages in the State Highway network, the Johnstone’s Hill twin tunnels are the most recent. The twin tunnels were opened in January 2009 to improve travel times on the Northern Motorway between the Orewa turnoff and Puhoi by between 10 and 30 minutes. As a toll road, they form part of SH1 and bypass the seaside townships of Orewa, Hatfield’s Beach, Waiwera, and the Wenderholm Regional Park.

The tunnels are two identical semi-circular uni-directional tubes, approximately 12 m wide, 9 m high and 380 m long. They are built to carry two lanes each, plus a shoulder and an emergency pathway. Currently, the southbound tunnel has two lanes open but the northbound tunnel has only one lane open due to the merging of the traffic into a single lane immediately after the tunnel. The intention is to open the northbound tunnel to two lanes of traffic once the Puhoi to Wellsford double-laning project has been completed (scheduled for potential completion post 2015).

Based on the traffic volumes for the ten months (late January to late November 2009) the annual average daily traffic (AADT) for the Johnstone’s Hill tunnel in 2009 is estimated to be approximately 12,742 vehicles per day, with the proportion of heavy commercial vehicles (HCVs) at 10.2%. Figure 3.1 shows the site location on a map of the region.

Figure 3.1 Map showing location of Johnstone’s Hill twin tunnels monitoring site () (Source: Wises Maps www.wises.co.nz).



The Johnstone’s Hill tunnel is located in hill country close to the coastal zone with the SH1 roadway passing over estuarine inlets either side of the tunnel portals. Figure 3.2 provides an aerial photo of the Johnstone’s Hill tunnel site and its immediate environs.

Figure 3.2 Aerial view of Johnstone’s Hill tunnel (----) (Source: Google Maps 2010).

3.2 Air particulate matter sampling

Airborne particles less than 10 microns in aerodynamic cross-section (PM10) were collected in two size fractions for analysis in this study;

PM2.5 for particles 2.5 microns and less in diameter; and

PM10-2.5 for particles between 2.5 microns and 10 microns in diameter

Continuous monitoring of PM10 was carried out using a beta attenuation monitor (BAM). The initial sampling program design was to run three Streaker samplers in series to cover all hours of the week as the site could only be visited for filter changes on a weekly basis. However, problems with the sequential sampling system electronics in the tunnel environs resulted in only the first sample set (60 hours) being collected in each weekly sequence. In order to cover all days of the week a separate Streaker sampler system was run at a three-hour sample period for all of June 2010.



3.3 In-tunnel sampling location and set-up

The particulate matter sampling equipment was contained in a pillar box (1500 mm high x 800 mm wide x 600 mm deep) that was bolted to the western wall of the northbound tunnel approximately 25 metres in from the northern portal. Figure 3.3 presents a schematic diagram of the tunnel layout and location of the monitoring equipment.

Figure 3.3 Schematic of Johnstone’s Hill tunnel showing the location ( ▄ ) of PM10 monitoring equipment in the northbound tunnel

Figure 3.4 shows the pillar box in place, bolted to the wall of the tunnel, with PM10 inlet for the BAM on top alongside the Streaker samples

N



Figure 3.4 Pillar box in the north-bound Johnstone’s Hill tunnel

3.4 Johnstone’s Hill tunnel conceptual receptor model

An important part of the receptor modelling process is to formulate a conceptual model of the receptor site. This means understanding and identifying the major sources that may influence ambient particulate matter concentrations at the site. For the Johnstone’s Hill tunnel site the initial conceptual model includes emission sources in the tunnel:

Motor vehicles – all motor vehicles passing through the tunnel will contribute particulate matter to tunnel air from tail pipe emissions and the abrasion of tyre and road surfaces;

Road dust - road surface dust may be re-suspended into tunnel air by the turbulent passage of motor vehicles.

Sources that originate from outside the tunnel may also contribute to tunnel particle loadings due their presence in tunnel intake air, these include:

Marine aerosol (sea salt); Secondary particulate matter resulting from atmospheric gas-to-particle conversion

processes (sulphate and nitrate species, organic particle species); Wind-blown soil (crustal matter); Combustion sources such as biomass burning or emissions from vehicles on other

roads in the vicinity.

Another category of emission sources that may contribute are those considered as ‘one-off’ emission sources:

Short-term road works, tunnel maintenance and demolition/construction activities.

The variety of sources described above can be recognised and accounted for by appropriate data analysis methods, such as temporal variations and receptor modelling itself.



3.5 Local meteorology

A meteorological station operated by Watercare Services Limited (WSL) was located adjacent to Fowler Access Road which runs along the ridge of Johnstone’s Hill above the tunnel portals. The predominant wind directions during the monitoring period in the Johnstone’s Hill area were from the northwesterly and southeasterly quarters as shown by the wind rose in Figures 3.4.

Figure 3.4 Wind rose for April 2010 - July 2010 (data supplied by Watercare Services Limited)

It is likely that the meteorology at the monitoring site is defined and constrained by local topography.

3.6 PM10 concentrations

PM10 was continuously monitored in the north-bound tunnel using a MetOne E-BAM Beta-particle Attenuation Monitor (BAM) operated in accordance with AS/NZS 3580.9.11.2008. Figure 3.5 presents the BAM PM10 monitoring results for the monitoring period in the tunnel (29 April 2010 – 14 July 2010) and Table 3.1 contains statistics for PM10 concentrations.

Table 3.1 Statistics for PM10 concentrations in Johnstone’s Hill Tunnel

NORTH

SOUTH

WEST EAST

4%

8%

12%

16%

20%

WIND SPEED (m/s)

>= 10.0

7.0 - 10.0

5.0 - 7.0

3.0 - 5.0

2.0 - 3.0

1.0 - 2.0

Calms: 26.42%



15- minute average PM10 1-hour average PM10 24-hour average PM10

Average 12.3 13.8 13.5

Maximum 304.9 96. 9 32.3

Minimum 0.0 0.0 5.2

Median 11.1 13.0 13.1

Standard deviation 11.0 7.6 4.7

A peak in PM10 concentrations was evident in the 15-minute and 1-hour averages around midday on 16 May 2010. The site was not visited on that day and instrument records suggest there was no other reason to invalidate the data. Consultation with the Auckland Motorways Alliance (AMA) regarding tunnel activities on that day, indicate that no dust generating maintenance or roadworks in the tunnel environs were being undertaken. Furthermore, examination of CO, NOx and traffic data (vehicle numbers and vehicle speeds) did not suggest that the concentrations were due to vehicle (tailpipe) related emissions. It is therefore surmised that the particulate matter may have been due to dust generated by a vehicle passing through the tunnel such as a poorly covered load of topsoil or potentially some incident nearby the tunnel environs such as a bushfire, however, these are speculative associations.

0

50

100

150

200

250

300

350

31/03/2010 14/04/2010 28/04/2010 12/05/2010 26/05/2010 09/06/2010 23/06/2010 07/07/2010

PM

10 1

5 m

inu

te a

ver

ag

e (g

m-3

)

a)



Figure 3.5 PM10 (BAM) concentrations in Johnstone’s Hill (northbound) tunnel: a) 15 minute average; b) 1-hour average; c) 24-hour average

Figure 3.5 shows that PM10 concentrations in the tunnel followed a regular diurnal pattern and that baseline concentrations increased during the monitoring period (most evident in the 1-hour and 24-hour plots).

0

20

40

60

80

100

31/03/2010 14/04/2010 28/04/2010 12/05/2010 26/05/2010 09/06/2010 23/06/2010 07/07/2010

PM

10 1

-ho

ur

aver

age

( g

m-3

)

b)

0

10

20

30

40

50

31/03/2010 14/04/2010 28/04/2010 12/05/2010 26/05/2010 09/06/2010 23/06/2010 07/07/2010

PM

10 2

4-h

ou

r av

era

ge

(g

-3

)

c)



4.0 RECEPTOR MODELLING OF PM10 IN JOHNSTONE’S HILL TUNNEL

4.1 Analysis of PM10

Samples of PM10 from the tunnel site were collected as two size fractions (PM2.5 and PM10-2.5) with a Streaker sampler (as described in Section 3.2) operating on a 3-hourly sample period for all of June 2010.

PM10 concentrations were derived from the BAM data and elemental concentrations provided by IBA with black carbon (BC) by light reflectance as described in Chapter 2 and Appendix 1. Receptor modelling was performed using two multivariate analysis approaches;

1. Principal components analysis (PCA) was used to characterise the p variables (elemental concentrations) in sample X in terms of a small number m of common factors F in order to examine relationships between the variables and provide an initial estimate of the number of factors (contributing particulate matter sources) present;

2. Positive matrix factorisation (PMF) Positive matrix factorisation (PMF) is a linear least-squares approach to factor analysis for environmental data where factors (sources) are constrained to have non-negative species concentrations, no sample can have a negative source contribution and error estimates for each observed data point are used as point-by-point weights in order to mimic real-world conditions. PMF was applied the multivariate data from Johnstone’s Hill tunnel to provide source profiles (elemental composition of particulate matter from each of the sources) and apportion per-sample PM10 mass concentrations to the various sources.

Each source identified was rationalised against geochemical principals and other research published in the literature. Only those solutions that could be related to physical sources were considered acceptable. Appendix 1 provides a detailed description of the receptor modelling methodology as applied to the Johnstone’s Hill tunnel data.

4.2 Composition of PM2.5 and PM10-2.5

Elemental concentrations for PM2.5 collected in the tunnel are presented in Table 4.1 with a box and whisker plot of the elemental concentrations shown in Figure 4.1. The data in Table 4.1 show that carbonaceous aerosols (represented by BC) account for the bulk of measured PM2.5 mass. It is likely that particulate organic compounds from the incomplete combustion of fuels also contribute to PM2.5 mass. For example the analysis of organic carbon (OC) content associated with organic compounds and elemental carbon (EC - equivalent to BC or soot) in the Mt Victoria Tunnel, Wellington found that OC and EC comprised approximately 50% and 34% of the fine particle mass respectively (Ancelet, Davy et al. 2011). Studies elsewhere have shown that OC is 3-5 times the EC concentration in tunnel environs with the ratio dependent on the relative numbers of light duty (petrol) and heavy commercial (diesel) vehicles using the tunnel (Schauer, Lough et al. 2006).

Interestingly Table 4.1 also shows that sodium (Na) and chlorine (Cl) are important elemental



components of PM2.5 and signal a marine aerosol source contribution.

Table 4.1 Elemental analysis results for 3-hourly PM2.5 samples in Johnstone’s Hill tunnel (220 samples). LOD = limit of detection

Species Average (ng m-3)

Max (ng m-3)

Min (ng m-3)

Median (ng m-3)

StdDev (ng m-3)

Average LOD

(ng m-3) #>LOD

BC 3839 12304 29 3422 2688 329 209

Na 297 1033 0 258 204 360 68

Mg 95 150 35 98 24 36 218

Al 67 113 16 69 20 21 218

Si 107 221 36 110 33 15 220

P 5 36 0 0 8 35 7

S 123 306 34 120 44 17 220

Cl 400 2207 16 262 399 13 220

K 25 124 0 21 20 11 172

Ca 30 150 1 27 17 9 207

Ti 2 24 0 1 3 10 11

V 1 8 0 0 2 8 2

Cr 5 26 0 4 5 5 96

Mn 1 7 0 0 1 6 13

Fe 58 254 4 49 44 4 220

Co 1 9 0 0 2 7 9

Ni 1 6 0 1 2 5 19

Cu 4 27 0 3 4 5 64

Zn 6 38 0 4 7 7 73

As 1 14 0 0 2 13 2

Se 1 14 0 0 3 15 0

Br 9 54 0 2 12 24 33

I 9 59 0 4 11 35 10

Ba 5 26 0 2 6 31 0

Pb 1 21 0 0 4 33 0

BC Na Mg Al Si S Cl K Ca Ti Mn Fe Cu Zn As Pb

1

10

100

1000

10000

100000

ng

m-3



Figure 4.1 Box and whisker plot of PM2.5 elemental concentrations in Johnstone’s Hill tunnel

Elemental concentrations for PM10-2.5 collected in the tunnel are presented in Table 4.2 with a box and whisker plot of the elemental concentrations shown in Figure 4.2.

Table 4.2 Elemental analysis results for 3-hourly PM10-2.5 in Johnstone’s Hill tunnel (220 samples) LOD = limit of detection

Species Average (ng m-3)

Max (ng m-3)

Min (ng m-3)

Median (ng m-3)

StdDev (ng m-3)

Average LOD

(ng m-3) #>LOD

BC 192 748 27 174 108 195 97

Na 1637 8100 34 1132 1540 272 193

Mg 98 362 22 90 51 42 209

Al 81 949 8 64 78 22 201

Si 240 1495 44 192 187 16 220

P 53 166 0 50 27 27 188

S 218 957 12 173 170 21 218

Cl 2746 13308 29 1957 2394 14 220

K 104 400 11 90 67 11 220

Ca 207 705 36 183 119 10 220

Ti 7 44 0 3 9 11 55

V 1 7 0 0 2 13 1

Cr 2 20 0 1 3 10 24

Mn 3 18 0 2 3 8 21

Fe 184 784 5 135 163 5 220

Co 1 12 0 0 2 11 5

Ni 2 21 0 1 3 6 32

Cu 10 41 0 8 9 5 144

Zn 12 499 0 7 35 7 120

As 1 35 0 0 3 13 2

Se 1 38 0 0 4 14 1

Br 6 41 0 0 9 22 20

I 8 65 0 0 13 45 6

Ba 13 103 0 9 16 37 21

Pb 3 41 0 0 7 35 2

It is evident from Table 4.2 that Na and Cl are also major components of coarse particulate mass with significant contributions from crustal matter elements of aluminium (Al), silicon (Si), calcium (Ca) and iron (Fe).



Figure 4.2 Box and whisker plot of PM10-2.5 elemental concentrations in Johnstone’s Hill tunnel

4.3 Source contributions to PM10 in Johnstone’s Hill tunnel

Six primary source contributors to PM10 in the tunnel were determined from the PMF receptor modelling analysis of Streaker 3-hourly elemental composition and are presented in Table 4.3. Both coarse and fine elemental mass was included in the same PMF analysis as the apportionment was to PM10. Note that coarse fraction elements (i.e. those present in PM10-2.5) are identified as such in Table 4.2. For CO the units are in mg m−3 (x 100) and for NOx the units are ppm (x100) (gaseous data was supplied by Watercare Services Limited). The L1 (wheel base <5.5 m) and HCV (wheel base >5.5 m) represent light duty vehicle and heavy commercial vehicle numbers respectively (unpublished data supplied by Auckland Motorway Alliance (AMA)). Table 4.2 represents the summary results for a reiterative process that examines the effect of each species on the PMF receptor modelling using the modelling diagnostics presented in Appendix 2.2. Species that were poorly modelled (slope, r2 < 0.7) have been removed from the analyses unless considered vital for source identification. The source contributors identified by PMF were found on average to explain 83 % of PM10 gravimetric mass. Figure 4.3 presents the source profiles extracted from the PMF analysis of tunnel PM10 data.

BC Na Mg Al Si S Cl K Ca Ti Mn Fe Cu Zn As Pb

1

10

100

1000

10000

ng

m-3



Table 4.3 Elemental composition of source profiles and contributions to 3-hourly average PM10 in Johnstone’s Hill tunnel

Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6

Species Light duty

vehicles

Heavy commercial

vehicles

Smoky vehicles

Road dust

Biomass burning

Marine aerosol

PM10 g m−3 average 0.8 3.0 0.8 3.3 0.4 3.4

maximum 4.9 13.9 8.7 7.0 2.7 20.2

Na PM10-2.5 (ng m-3) 26.3 0.0 55.8 17.5 90.1 1386.9

Mg PM10-2.5 (ng m-3) 0.0 5.5 8.4 28.6 14.8 41.9

Al PM10-2.5 (ng m-3) 3.5 9.3 27.9 31.9 5.9 0.0

Si PM10-2.5 (ng m-3) 16.9 26.2 107.3 50.9 9.0 30.6

S PM10-2.5 (ng m-3) 10.4 2.8 16.9 2.5 21.3 161.0

Cl PM10-2.5 (ng m-3) 29.9 31.8 80.6 0.0 143.8 2451.5

K PM10-2.5 (ng m-3) 2.7 8.7 4.5 2.6 28.7 57.7

Ca PM10-2.5 (ng m-3) 12.7 12.6 46.9 26.1 23.9 81.8

Fe PM10-2.5 (ng m-3) 82.4 11.2 46.9 0.0 20.1 25.5

Cu PM10-2.5 (ng m-3) 5.1 1.7 1.4 0.0 1.3 0.5

Zn PM10-2.5 (ng m-3) 4.2 0.0 3.8 0.0 1.9 1.0

BC PM2.5 (ng m-3) 913.1 2298.7 205.7 34.2 438.6 38.2

Na PM2.5 (ng m-3) 0.0 0.0 17.0 77.4 2.6 179.2

Mg PM2.5 (ng m-3) 2.5 15.5 6.0 54.9 0.0 15.9

Al PM2.5 (ng m-3) 2.9 13.4 4.5 38.6 0.3 8.3

Si PM2.5 (ng m-3) 7.7 23.7 7.1 51.4 0.0 16.8

S PM2.5 (ng m-3) 12.6 21.7 2.0 40.6 8.8 36.0

Cl PM2.5 (ng m-3) 4.5 21.2 5.1 0.0 0.0 359.6

K PM2.5 (ng m-3) 6.3 3.8 0.0 8.7 6.5 5.8

Ca PM2.5 (ng m-3) 4.9 6.5 3.4 8.3 0.8 8.1

Fe PM2.5 (ng m-3) 27.8 16.9 1.9 9.9 0.0 1.8

Zn PM2.5 (ng m-3) 1.3 0.3 1.4 0.0 0.2 0.0

CO (mg m-3 x 100)1 16.9 10.1 6.5 6.9 0.2 0.0

NOx (ppm x 100)2 3.5 5.8 1.4 0.0 0.9 0.0

L1 (# of vehicles)3 5.5 0.4 4.9 1.3 0.0 2.0

HCV (# of vehicles)4 0.6 2.1 1.2 0.5 0.0 0.4 1CO = carbon monoxide concentrations expressed as a 3-hourly average 2NOx = nitrogen oxides concentrations expressed as a 3-hourly average 3L1 = number of vehicles with wheel base <5.5 metres expressed as a 3-hourly average 4HCV = number of vehicles with wheel base >5.5 metres expressed as a 3-hourly average



Figure 4.3 Source profiles and elemental concentrations for PM10 in Johnstone’s Hill tunnel

1

10

100

1000

10000

NaC MgC AlC SiC S_C ClC KC CaC FeC CuC ZnC BC Na Mg Al Si S Cl K Ca Fe Zn Cog Nox L1 HCV

Co

nce

ntr

atio

n (

ng

m-3

)

Element

Light duty vehicles

1

10

100

1000

10000


Co

nce

ntr

atio

n (

ng

m-3

)

Element


1

10

100

1000

10000


Co

nce

ntr

atio

n (

ng

m-3

)

Element

Smoky vehicles

1

10

100

1000

10000


Co

nce

ntr

atio

n (

ng

m-3

)

Element

Road dust

1

10

100

1000

10000


Co

nce

ntr

atio

n (

ng

m-3

)

Element

Marine aerosol

1

10

100

1000

10000


Co

nce

ntr

atio

n (

ng

m-3

)

Element

Biomass burning



Sources identified are described briefly as follows, more detail description and analysis is provided in Chapter 5:

The first source profile has been attributed to emissions from light duty vehicles due to the association of PM2.5 BC, Ca, Fe, the majority of CO and some association with L1 (wheel base <5.5m). The light duty fleet is primarily petrol fuelled (84 %) which produces significantly more CO than those vehicles powered by diesel (15.5% (MOT 2010). This source also accounts for most of the copper containing particulate matter which is most likely due to the wear of brake pads.

The second source was attributed to heavy commercial vehicles (trucks and buses) as this source accounts for majority of the PM2.5 BC, Ca, Fe and NOx plus a distinct association with the number of HCVs. Heavy commercial vehicles are almost exclusively powered by diesel.

The smoky vehicles source was a minor contributor to PM10 but has a distinct BC and Zn component which suggests that the particulate matter was produced from engine emissions that include a significant component of lubricating oil combustion. This can be produced by two stroke engines or four stroke engines in poor mechanical condition.

The road dust source consisted primarily of aluminium, silicon and other crustal matter components which can be attributed to the mechanical abrasion of road surfaces by vehicle tyres. Road dust is constantly being re-suspended by the passage of motor vehicles through the tunnel (less so if the road surface is wet (Amato, Querol et al. 2009)).

The fifth source has been attributed to biomass burning due to the association of PM2.5 BC and K, a known marker for the combustion of wood and other vegetative matter (Li, Posfai et al. 2003; Niemi, Tervahattu et al. 2004; Reisen, Meyer et al. 2011).

The sixth source is due to marine aerosol (sea salt) as the pairing of both fine and coarse Na and Cl is distinct along with other elements found in sea water.

Figure 4.4 presents the relative source contributions to PM10 concentrations in Johnstone’s Hill Tunnel. Also included in Figure 4.4 are the standard deviations in mass contributions for each of the sources indicating the variability in average mass contributions over the monitoring period.



Figure 4.4 Average (3-hourly) source contributions to PM10 in Johnstone’s Hill tunnel for June 2010

Figure 4.4 shows that the primary sources of PM10 in the Johnstone’s Hill northbound tunnel were vehicle associated emissions (dominated by heavy commercial vehicles and road dust), and marine aerosol from external ambient air.

Peak concentrations of PM10 in the tunnel and the sources responsible are of interest for air quality management and the protection of human health. Figure 4.5 presents time-series plots for the 3-hourly contributions from each source. Peak motor vehicle source contributions are dominated by heavy commercial vehicle emissions and the road dust source shows a regular diurnal pattern of contributions throughout the monitoring period.

An aggregate plot of all motor vehicle source contributions (light duty vehicles, heavy commercial vehicles, smoky vehicles plus road dust) as presented in Figure 4.6 shows that motor vehicle emissions are the primary source of PM10 in the tunnel and a diurnal pattern in concentrations is clearly evident. It can be seen that marine aerosol had the highest individual source contributions during the monitoring period. Concentrations of marine aerosol in ambient air are known to reach elevated levels in the Auckland region and are dependent on prevailing meteorological conditions (Davy, Trompetter et al. 2009). Further discussion of each of the sources is provided in Chapter 5.

Light duty vehicles

7%

Heavy commercial

vehicles25%

Smoky vehicles

7%

Road dust28%

Biomass burning

4%

Marine aerosol29%

0.8

3.0

0.8

3.3

0.4

3.4

0

2

4

6

8

Lig

ht

du

tyv

ehic

les

He

avy

com

mer

cial

veh

icle

s

Sm

oky

ve

hic

les

Ro

ad d

ust

Bio

mas

s b

urn

ing

Mar

ine

aero

solP

M10

mas

s co

ntr

ibu

tio

n (g

m-3

)



Figure 4.5 Temporal variation of source contributions to PM10 mass in Johnstone’s Hill tunnel

0

5

10

15

20

01-Jun 06-Jun 11-Jun 16-Jun 21-Jun 26-Jun 01-Jul

PM

10

co

nce

ntr

atio

n (g

m-3

)

Smoky vehicles

0

5

10

15

20


PM

10c

on

cen

trat

ion

(g

m-3

)

Road dust

0

5

10

15

20


PM

10

co

nce

ntr

ati

on

(g

m-3

)

Biomass burning

0

5

10

15

20


PM

10co

nc

entr

atio

n (g

m-3

)

Marine aerosol

0

5

10

15

20


PM

10co

nc

entr

atio

n (g

m-3

)

Light duty vehicles

0

5

10

15

20


PM

10

con

cen

tra

tio

n (g

m-3

)




Figure 4.6 Aggregate* time-series of vehicle source contributions to PM10 mass in Johnstone’s Hill tunnel (*light duty vehicles, heavy commercial vehicles, smoky vehicles plus road dust)

4.4 Weekend and weekday variations in sources of PM10

Mass contributions to PM10 from the different sources were divided into weekday (Monday-Friday) and weekend (Saturday and Sunday) categories to examine any differences in relative contributions. Figure 4.7a presents a plot of weekday and weekend mass contributions for each category. Note that there were only four weekends plus Queens Birthday Monday (nine days total) included in the analysis (62 samples) and 23 week days (156 samples). The light duty vehicle and road dust sources showed no significant difference in weekend and weekday source activity whereas the heavy commercial vehicle and smoky vehicles sources show a bias towards higher contributions on weekdays. The biomass burning and marine aerosol source contributions are more likely dependent on meteorological conditions outside the tunnel environs.

The weekend/weekday split derived from the receptor modelling for PM10 attributed to light duty and heavy duty vehicles is consistent with the vehicle count data as presented in Figure 4.7b.

0

5

10

15

20


PM

10 c

on

ce

ntr

ati

on

(g

m-3

)



Figure 4.7a Weekend and weekday variation of source contributions to PM10 mass in Johnstone’s Hill tunnel

Figure 4.7b Weekend and weekday variation of vehicle counts in Johnstone’s Hill tunnel for light duty (L1) and heavy commercial vehicles (HCV) respectively

0

2

4

6

8

Lig

ht

du

tyv

ehic

les

He

avy

com

mer

cial

veh

icle

s

Sm

oky

veh

icle

s

Ro

ad

du

st

Bio

ma

ss b

urn

ing

Mar

ine

aero

sol

PM

10m

ass

con

trib

uti

on

(g

m-3

)Weekday

Weekend

0

10

20

30

40

50

60

70

L1

HC

V

Ave

rag

e n

um

ber

of

veh

icle

co

un

ts

Weekday

Weekend



5.0 DISCUSSION OF RECEPTOR MODELLING RESULTS

Six distinct sources were extracted from the source apportionment of PM10 in Johnstone’s Hill tunnel. The following sections examine each source in turn and rationalise the source contributions and temporal patterns observed with vehicle numbers, meteorological influences and known characteristics for each source from other studies.

5.1 Light duty vehicles

The New Zealand light duty vehicle fleet is primarily petrol-fuelled with the diesel vehicle proportion currently at 15.5 % (MOT 2010). The source of PM10 attributed to light duty vehicles (defined here as a wheelbase less than 5.5 m) is dominated by PM2.5 BC and the majority of apportioned CO (see Figure 4.3) from exhaust emissions. It is likely that the light duty vehicle source profile represents petrol vehicle emissions only, since light duty diesel engines have similar combustion characteristics and emissions to heavy commercial diesel vehicles.

The majority of coarse particle Cu was also associated with the light duty vehicle source and this is considered to be due to the covariant emission of brake dust as vehicles pass through the tunnel. Brake pads contain significant amounts of Cu along with Fe, Zn, barium (Ba) and antimony (Sb), since the phase out of the asbestos component used as the abrasive medium (Salma and Maenhaut 2006). Studies elsewhere have found a similar covariance of brake dust with the tailpipe emissions and others have separated a brake dust source or found it included as part of the road dust component (Gertler, Abu-Allaban et al. 2001; Gillies, Gertler et al. 2001; Gertler, Gilliss et al. 2002; Abu-Allaban, Gillies et al. 2003; Lough, Schauer et al. 2005; Schauer, Lough et al. 2006; Amato, Pandolfi et al. 2009; Amato, Pandolfi et al. 2009).

While the contribution from light duty vehicles is low (16 % of vehicle related emissions - not including road dust) the number of light duty vehicles passing through the tunnel represents 90 % of all traffic. Figure 5.1 presents the 3-hour average time-series for light duty vehicle numbers through the northbound Johnstone’s Hill tunnel (unpublished data supplied by Auckland Motorway Alliance (AMA)).

Several patterns are evident in light duty vehicle numbers passing through the tunnel. The first is a diurnal pattern that peaks in the afternoon representing commuter traffic leaving Auckland and minima occur in the early hours of the morning. Vehicle numbers were also slightly higher on Fridays and during the weekend. Peak vehicle numbers during the monitoring period were on Friday 4th and Saturday 5th June as this was the Queens Birthday long weekend and reflects vacationers heading north for the holiday period.



Figure 5.1 3-hour average light duty vehicle numbers for June 2010

5.2 Heavy commercial vehicles

Heavy commercial vehicles on New Zealand roads are essentially all diesel powered (MOT 2010). The heavy commercial vehicle source profile contains most of the PM2.5 BC mass and apportioned NOx. The estimated contribution to PM10 in the tunnel represents 67 % of the mass concentrations attributed to tailpipe emissions (i.e. heavy commercial, light duty and smoky vehicle sources) yet heavy commercial vehicles represent only 10 % of vehicle numbers using the tunnel. Figures 5.2a and 5.2b present the three-hourly average heavy commercial vehicle numbers for the northbound tunnel during the monitoring period. Note that Figure 5.2a presents the HCV data on the same scale as that of Figure 5.1 to illustrate the difference in the number of vehicles for each vehicle class using the tunnel (unpublished data supplied by Auckland Motorway Alliance (AMA)).

Figure 5.2a 3-hour average heavy commercial vehicle numbers for June 2010 (same scale as for Figure 5.1)

0

50

100

150

200

250


Nu

mb

er

of

veh

icle

s (

3-h

ou

r av

era

ge)

0

50

100

150

200

250


Nu

mb

er

of

veh

icle

s (

3-h

ou

r a

vera

ge)



Figure 5.2b 3-hour average heavy commercial vehicle numbers for June 2010 (expanded scale)

Heavy commercial vehicle numbers were significantly higher during weekdays than weekends as they represent normal working week road freight and transport activity. Peak numbers were through the middle of the day with minima in the early hours of the morning. Interestingly there was a maximum in heavy commercial vehicle numbers on the Friday (4th June) before Queens Birthday weekend. The higher heavy commercial vehicle numbers during the week compared to the weekend are mirrored by the higher weekday PM10 contributions from this source as presented in Figure 4.7.

PM10 contributions from the heavy commercial vehicle source show a diurnal pattern but not the regularity expected from the pattern in vehicle numbers. The controlling factor appears to be tunnel ventilation rate. As the Johnstone’s Hill tunnels are relatively short and are aligned north-south in line with the predominant wind directions (see Figure 3.4), external wind speeds have a significant influence on particulate matter concentrations in the tunnel. Figure 5.3 presents a plot of external wind speed and the heavy commercial vehicle source contributions to PM10.

Figure 5.3 External wind speed and tunnel air speed compared to heavy commercial vehicle PM10 contributions for June 2010 in Johnstone’s Hill tunnel

0

5

10

15

20

25

30


Nu

mb

er o

f ve

hic

les

(3-h

ou

r av

erag

e)

0

2

4

6

8

10

0

5

10

15

20


Win

d s

pee

d (

m s

-1)

PM

10co

nce

ntr

atio

n (g

m-3

)


External wind Speed

Tunnel airspeed



Figure 5.3 shows that tunnel ventilation as measured by the in-tunnel air speed was strongly influenced by external wind speeds, which in turn affect PM10 concentrations due to vehicles within the tunnel (i.e. this is a ‘natural’ ventilation effect). Vehicle related PM10 is anti-correlated with tunnel ventilation rates (i.e. the same applies to light duty, heavy commercial and smoky vehicle sources). This result is obvious but lends weight to the robustness of source apportionment results.

The other factor that appeared to affect concentrations of PM10 in the tunnel from motor vehicle sources was ambient temperature. All motor vehicle sources show a rise in contributions from tailpipe emissions from the beginning of the monitoring period, particularly from middle of June onwards. The rise was also obvious in the BAM PM10 concentrations presented in Figure 3.5 which showed a steady increase in baseline from the beginning of the monitoring period (April – July). The effect was also evident in the BC concentration time-series which was independent of PM10 measurements. Ambient temperature measured in the tunnel for the entire monitoring period (April-July 2011) is presented in Figure 5.4.

Figure 5.4 Ambient air temperature inside in Johnstone’s Hill tunnel from April to July 2010

The decrease in ambient temperature probably results in increased tailpipe emissions from cooler engine operating conditions, though cold calm meteorological conditions (i.e. decreased tunnel ventilation) may also have played a part. It is unlikely that the increase in emissions was due to cold-start engine conditions as vehicles using the tunnel would have already driven some distance on the motorway since the nearest on-ramp (Orewa) is 5 km to the south of the tunnel. Receptor modelling studies in Auckland have shown that that there is a rise in motor vehicle related PM2.5 and PM10 during winter months (Davy, Trompetter et al. 2009). This has been ascribed to both a rise in engine emissions, which may include cold-start emissions under urban driving conditions, and cool calm meteorological conditions resulting in poorer dispersion.

5.3 Smoky vehicles

The minor source of PM10 that has been attributed to smoky vehicles is intermittent and appears to be higher during weekdays as seen in Figure 4.7. The smoky vehicle profile contains a distinct BC and Zn signature which has been observed in Auckland urban receptor modelling studies, separate from the general motor vehicle source contributions

0

5

10

15

20

25

29-Mar 12-Apr 26-Apr 10-May 24-May 07-Jun 21-Jun 05-Jul

Tu

nn

el

air

te

mp

era

ture

(oC

)



(Davy, Trompetter et al. 2009). The results suggest that the smoky vehicle contribution is due to old or poorly-tuned vehicles burning a significant amount of lubricating oil (which is high in Zn-containing additives). Two stroke engines (where the lubricating oil is added directly to the fuel) may also contribute but particulate matter concentrations from the smoky vehicle source in the Auckland urban studies suggest that they are too high for the two stroke vehicle fleet (primarily motorbikes) to be entirely responsible.

5.4 Road dust

The road dust source demonstrated a regular diurnal pattern and appears to be composed entirely of crustal matter elements indicating that it was due to the abrasion of the road surface by tyres and crustal matter brought in by vehicles, although some may be due soil particles in external air. All vehicles passing through the tunnel re-entrain dusts on the road surface and this explains the regularity seen in the diurnal pattern. The road dust source does not appear to be particularly affected by tunnel ventilation rates probably because the sheer number of vehicles passing through the tunnel constantly re-entrains road surface dusts. Studies of road dust sources have found similar patterns and that a significant quantity of OC was also associated with the mineral dust content due to abrasion of tyres (Schauer, Lough et al. 2006; Amato, Pandolfi et al. 2009). All studies describe the road surface as asphaltic though no specific formulation of compositional data was supplied. The Johnstone’s Hill tunnel road surface (including the highway approach) is asphaltic.

Wet road surfaces do suppress road dust re-suspension and a distinct minima was observed on 21 June 2010 (see Figure 4.5) (Amato, Querol et al. 2009). Analysis of synoptic weather maps (http://www.bom.gov.au/australia/charts/archive/index.shtml) for that period indicate that a series of cold fronts passed over the Auckland region. Rainfall records from an Auckland Council monitoring site at Orewa 5 km to the south show significant rainfall during the period (from http://maps.auckland.govt.nz/aucklandregionviewer). The wet weather conditions support the resulting low road dust PM10 contributions at the time.

5.5 Biomass burning

The pairing of PM2.5 BC and K in the source profile attributed to biomass burning is indicative of particulate matter arising from the combustion of vegetative matter. The source contributions were relatively minor and are likely to be due to local solid fuel (wood burning) fires used for domestic heating. A cold snap around the middle of June (http://tvnz.co.nz/national-news/cold-snap-hits-new-zealand-3582251) appears to have resulted in a rise in biomass burning contributions. Peak concentrations for the biomass burning source were found to occur mid to late evening, a pattern observed in many parts of New Zealand for domestic fire related PM10 (Trompetter, Davy et al. 2010).

5.6 Marine aerosol

Due to New Zealand’s oceanic location, marine aerosol can be a significant contributor to PM10. For example, a PM10 exceedence event at Orewa, on the coast not far from the Johnstone’s Hill tunnel, was largely attributed to marine aerosol (Davy, Trompetter et al. 2009). Source apportionment studies elsewhere in New Zealand have found marine aerosol to be a primary contributor to the coarse fraction and also significantly influence PM10



concentrations (Wilton, Smith et al. 2007; Davy, Trompetter et al. 2008; Davy, Trompetter et al. 2009; Davy, Trompetter et al. 2009; Davy, Trompetter et al. 2010). It is therefore not surprising that marine aerosol concentrations can be significant inside the tunnel due to the influx of ambient air. The concentrations of marine aerosol in ambient air are largely influenced by prevailing meteorological conditions and the origin of the air mass over a particular location.

For the Johnstone’s Hill tunnel study it was found that marine aerosol contributions to PM10 were significant on occasion (up to 20 g m−3). Figure 5.5 presents a plot of marine aerosol PM10 concentrations overlaid with external wind speed.

Figure 5.5 External wind speed compared to marine aerosol PM10 contributions for June 2010 in Johnstone’s Hill Tunnel

It can be seen that marine aerosol peaks in the tunnel during high external wind speeds, opposite to that found for vehicle related emissions where higher ventilation rates reduce tailpipe emission contributions. The most significant marine aerosol event was over several days of stormy weather from the 19th to 23rd June, the same conditions that resulted in suppression of road dust described in Section 5.4. Another marine aerosol event of interest was during Queens Birthday weekend when a series of fronts and a low pressure system passed over the region. High winds and rain suppressed road dust and resulted in low contributions from motor vehicle tailpipe emissions even though traffic volumes were the highest for the monitoring period.

0

5

10

15

20


PM

10 c

on

ce

ntr

ati

on

(

g m

-3)

0

2

4

6

8

10

Win

d s

pe

ed

(m

s-1

)

Marine aerosol PM10

Wind Speed



6.0 SUMMARY OF THE PM10 MONITORING AND RECEPTOR MODELLING

A source apportionment study of PM10 in the Johnstone’s Hill tunnel derived six primary sources that contributed to particulate matter concentrations. Three-hourly samples of size resolved particulate matter (PM2.5 and PM10-2.5) were collected over the entire month of June 2010. Ion beam analysis of the particulate matter samples to determine elemental concentrations was carried out at the New Zealand Ion Beam Analysis Facility in Lower Hutt. Source apportionment of PM10 elemental concentrations showed that motor vehicle tailpipe emissions, brake dust and road dust were the predominant sources of PM10 in the tunnel. On occasion sources of PM10 from outside the tunnel, mainly marine aerosol (seasalt) can have a significant influence on particulate matter concentrations in the tunnel.

PM10 composition was dominated by carbonaceous matter from the incomplete combustion of fuels and trace metals such as zinc and copper associated with motor vehicle emissions and mechanical wear of surfaces (primarily brake systems). It should be kept in mind that some elements that have been associated with motor vehicle emissions were below the detection limits of the analysis technique used here, (for example barium and antimony). Particulate matter from motor vehicle emission also include products of incomplete combustion such as toxic organic compounds (e.g. polyaromatic hydrocarbons and oxygenated or nitrogenated heterocyclic compounds).

Motor vehicle sources of PM10 were found to be dominated by heavy commercial vehicle emissions and re-suspended road dust. Heavy commercial vehicles were responsible for 66% of the vehicle exhaust related PM10, mostly in the PM2.5 size range, while heavy commercial vehicle numbers were only 10% of the total traffic volume passing through the tunnel. Light duty vehicles, mainly petrol fuelled, contributed 17% of the exhaust related PM10, (primarily as PM2.5) but were found to have a significant brake dust component associated, probably due to the sheer number of light duty vehicles using the tunnel. Light duty vehicles (90% of traffic volumes) were responsible for the majority of carbon monoxide present in the tunnel, whereas most of the nitrogen oxides were associated with heavy vehicle emissions. A separate motor vehicle source was attributed to smoky vehicles due to a distinct zinc signature. This source was considered to be due to poorly tuned or old vehicles that tend to burn lubricating oils (which contain added zinc as part of their formulation). Two stroke engines also burn lubricating oils but the relative contribution from the smoky vehicle source (10 % of vehicle related PM10) is probably too high for the corresponding proportion of two strokes in the vehicle fleet.

Overall, in-tunnel PM10 concentrations (maximum 32 g m−3 24-hour average) did not exceed the New Zealand National Environmental Standard (50 g m−3 24-hour average). However, the New Zealand National Environmental Standard relates to a 24-hour exposure period and does not apply to tunnels. Exposure to air quality in the tunnel will be short-term (approximately 14s at 80 km/h in the 300 m tunnel) for which there are no readily available air quality criteria for the protection of vehicle occupants and tunnel workers. From a health and safety perspective minimising risk is probably most important. The study showed that good ventilation (in this case natural forcing by external winds) was effective in keeping in-tunnel PM10 concentrations attributed to motor vehicle exhaust emissions low and that road dust is effectively suppressed by surface wetting. For tunnel maintenance and design



purposes, road surface composition and cleaning can minimise the generation and re-suspension of surface dusts.



7.0 GLOSSARY

AAQG ambient air quality guidelines AMA Auckland Motorways Alliance BC black carbon CO carbon monoxide CPF conditional probability function analysis EC elemental carbon GC/MS gas chromatography / mass spectroscopy GNS Geological and Nuclear Sciences Limited GUI graphical user interface HCV heavy commercial vehicle IBA Ion Beam Analysis LOD limit of detection ME2 multi-linear engine MLR multiple linear regression NES New Zealand National Environmental Standard NOX nitrogen oxides nss-Sulphate non-seasalt sulphate NZTA New Zealand Transport Agency OC organic carbon OMH organic mass from hydrogen PAH(s) polyaromatic hydrocarbon(s) PCA principal components analysis PESA particle elastic scattering analysis PIARC World Road Association PIGE proton induced gamma-ray emission PIXE proton induced X-ray emission PM particulate matter PM10 particulate matter less than 10 microns PM10-2.5 particulate matter between 10 and 2.5 microns PM2.5 particulate matter less than 2.5 microns PMF positive matrix factorisation QA/QC quality assurance - quality control RBS Rutherford back-scattering RCM reconstructed mass S/N signal-to-noise ratio SVOC semi-volatile organic compounds TC total carbon TSP total suspended particulate matter USEPA United States Environment Protection Agency VOCS volatile organic compounds WHO World Health Organisation



8.0 REFERENCES

Abu-Allaban, M., J. A. Gillies, et al. (2003). "Tailpipe, resuspended road dust, and brake-wear emission factors from on-road vehicles." Atmos. Environ. 37(37): 5283-5293.

AMA (2010). Traffic data for the Johnstone's Hill Northbound Tunnel collected by the Auckland Motorway Alliance between late March and early 2010.

Amato, F., M. Pandolfi, et al. (2009). "Quantifying road dust resuspension in urban environment by Multilinear Engine: A comparison with PMF2." Atmospheric Environment 43(17): 2770-2780.

Amato, F., M. Pandolfi, et al. (2009). "Spatial and chemical patterns of PM10 in road dust deposited in urban environment." Atmospheric Environment 43(9): 1650-1659.

Amato, F., X. Querol, et al. (2009). "Evaluating urban PM<sub>10</sub> pollution benefit induced by street cleaning activities." Atmospheric Environment 43(29): 4472-4480.

Ancelet, T. A., P. K. Davy, et al. (2011). "Carbonaceous aerosols in an urban tunnel." Submitted.

Begum, B. A., P. K. Hopke, et al. (2005). "Source identification of fine particles in Washington, DC, by expanded factor analysis modeling." Environ. Sci. Technol. 39(4): 1129-1137.

Brown, S. G. and H. R. Hafner (2005). Multivariate Receptor Modelling Workbook. Research Triangle Park, NC, USEPA.

Cahill, T. A., R. A. Eldred, et al. (1989). "Indirect measurement of hydrocarbon aerosols across the United States by nonsulfate hydrogen-remaining gravimetric mass correlations." Aerosol Sci. Technol. 10(2): 421-429.

Chueinta, W., P. K. Hopke, et al. (2000). "Investigation of sources of atmospheric aerosol at urban and suburban residential areas in Thailand by positive matrix factorization." Atmos. Environ. 34(20): 3319-3329.

Cohen, D. (1998). "Characterisation of atmospheric fine particles using IBA techniques." Nucl. Inst. Meth. Phys. Res. B 136-138: 14-22.

Cohen, D., G. Bailey, et al. (1996). "Elemental analysis by PIXE and other IBA techniques and their application to source fingerprinting of atmospheric fine particle pollution." Nucl. Inst. Meth. Phys. Res. B 109/110: 218-226.

Cohen, D., G. Taha, et al. (2000). The measurement and sources of fine particle elemental carbon at several key sites in NSW over the past eight years. 15th Clean Air Conference, Sydney, Clean air Society of Australia and New Zealand.

Cohen, D. D. (1999). "Accelerator based ion beam techniques for trace element aerosol analysis." Advances in Environmental, Industrial and Process Control Technologies 1(Elemental Analysis of Airborne Particles): 139-196.



D'Alessandro, A., F. Lucarelli, et al. (2004). "A summertime investigation on urban PM fine and coarse fractions using hourly elemental concentration data series." Nuovo Cimento della Societa Italiana di Fisica C 27(1): 17-28.

D'Alessandro, A., S. Nava, et al. (2004). "PIXE and ToF-SIMS analysis of streaker samplers filters." Nuclear Instruments and Methods in Physics Research, Section B: Beam Interactions with Materials and Atoms 222(1-2): 261-269.

Davy, P., K. (2007). Composition and Sources of Aerosol in the Wellington Region of New Zealand. PhD Thesis. School of Chemical and Physical Sciences. Wellington, Victoria University of Wellington: 429 pages.

Davy, P., K., W. Trompetter, et al. (2009). Source apportionment of airborne particles at Wainuiomata, Lower Hutt. Wellington, GNS Science.

Davy, P., K., W. Trompetter, et al. (2009). Source apportionment of airborne particles in the Auckland region: 2008 Update. Wellington, GNS Science.

Davy, P., K., W. J. Trompetter, et al. (2007). Source apportionment of airborne particles in the Auckland region. Wellington, GNS Science.

Davy, P., K., W. J. Trompetter, et al. (2008). Source apportionment of airborne particles at Seaview, Lower Hutt. Wellington, GNS Science.

Davy, P. K., W. J. Trompetter, et al. (2010). Source apportionment of PM10 at Tahunanui, Nelson, GNS Science.

Eberly, S. (2005). EPA PMF 1.1 User’s Guide, USEPA.

Filippi, E., P. Prati, et al. (1999). "Hourly measurement of particulate concentrations with streaker samplers and optical methods." Nuclear Instruments and Methods in Physics Research, Section B: Beam Interactions with Materials and Atoms 150(1-4): 370-374.

Fine, P. M., G. R. Cass, et al. (2001). "Chemical characterization of fine particle emissions from fireplace combustion of woods grown in the northeastern United States." Environ. Sci. Technol. 35(13): 2665-2675.

Formenti, P., H. J. Annegarn, et al. (1997). "Source apportionment by receptor models in a study of Geneva aerosol via streaker sampling and PIXE analysis." Physica Medica 13(3): 101-109.

Gertler, A. W., M. Abu-Allaban, et al. (2001). "Measurements of mobile source particulate emissions in a highway tunnel." Int. J. Vehicle Desgn. 27(1-4): 86-93.

Gertler, A. W., J. A. Gilliss, et al. (2002). "Emissions from diesel and gasoline engines measured in highway tunnels." Research Report - Health Effects Institute 107: i-iii, 1-94.

Gillies, J. A., A. W. Gertler, et al. (2001). "On-road particulate matter ( PM2.5 and PM10) emissions in the Sepulveda Tunnel, Los Angeles, California." Environ. Sci. Technol. 35(6): 1054-1063.

Hopke, P. K. (1999). "An introduction to source receptor modeling." Adv. Environ. Indust.



Process Control Technol. 1(Elemental Analysis of Airborne Particles): 273-315.

Hopke, P. K. (Undated). A guide to positive matrix factorisation. Potsdam, Clarkson University.

Hopke, P. K., Y. L. Xie, et al. (1999). "Mixed multiway analysis of airborne particle composition data." J. Chemomet. 13(3-4): 343-352.

Horvath, H. (1993). "Atmospheric Light Absorption - A Review." Atmos. Environ, 27A: 293-317.

Horvath, H. (1997). "Experimental calibration for aerosol light absorbtion measurements using the integrating plate method - Summary of the data." Aerosol Science 28: 2885-2887.

Jacobson, M. C., H. C. Hansson, et al. (2000). "Organic atmospheric aerosols: review and state of the science." Reviews of Geophysics 38(2): 267-294.

Jeong, C.-H., P. K. Hopke, et al. (2004). "The comparison between thermal-optical transmittance elemental carbon and Aethalometer black carbon measured at multiple monitoring sites." Atmos. Environ. 38(31): 5193.

Kavouras, I. G., P. Koutrakis, et al. (2001). "Source apportionment of PM10 and PM2.5 in five Chilean cities using factor analysis." J. Air Waste Manag. Assoc. 51(3): 451-464.

Kim, E., P. K. Hopke, et al. (2003). "Source identification of Atlanta aerosol by positive matrix factorization." J. Air Waste Manage. Assoc. 53(6): 731-739.

Kim, E., P. K. Hopke, et al. (2004). "Factor analysis of Seattle fine particles." Aerosol Sci. Technol. 38(7): 724-738.

Lee, E., C. K. Chan, et al. (1999). "Application of positive matrix factorization in source apportionment of particulate pollutants in Hong Kong." Atmos. Environ, 33(19): 3201-3212.

Lee, J. H., Y. Yoshida, et al. (2002). "Identification of sources contributing to Mid-Atlantic regional aerosol." J. Air Waste Manag. Assoc. 52(10): 1186-1205.

Li, J., M. Posfai, et al. (2003). "Individual aerosol particles from biomass burning in southern Africa: 2. Compositions and aging of inorganic particles." J. Geophys. Res. Atmos. 108(D13): SAF 20/21-SAF 20/12 8484.

Lough, G. C., J. J. Schauer, et al. (2005). "Emissions of metals associated with motor vehicle roadways." Environmental Science and Technology 39(3): 826-836.

Maenhaut, W. and J. Cafmeyer (1987). "Particle induced x-ray emission analysis and multivariate techniques: an application to the study of the sources of respirable atmospheric particles in Ghent, Belgium." Journal of Trace and Microprobe Techniques 5(2-3): 135-158.

Maenhaut, W. and K. Malmqvist, G. (2001). Particle Induced X-ray Emission Analysis. Handbook of x-ray spectrometry. R. V. Grieken. Antwerp, Marcel Dekker Inc.



Malm, W. C., J. F. Sisler, et al. (1994). "Spatial and seasonal trends in particle concentration and optical extinction in the United States." J. Geophys. Res. Atmos. 99(D1): 1347-1370.

Maxwell, J. A., J. L. Cambell, et al. (1989). "The Guelph PIXE software package." Nucl. Instr. And Meth. B 43: 218.

Maxwell, J. A., W. J. Teesdale, et al. (1995). "The Guelph PIXE software package II." Nucl. Instr. And Meth. B 95: 407.

Niemi, J. V., H. Tervahattu, et al. (2004). "Characterization and source identification of a fine particle episode in Finland." Atmos. Environ. 38(30): 5003-5012.

Paatero, P. (1997). "Least squares formulation of robust non-negative factor analysis." Chemom. Intell. Lab. Syst. 18: 183-194.

Paatero, P. (2000). PMF User's Guide. Helsinki, University of Helsinki.

Paatero, P. and P. K. Hopke (2002). "Utilizing wind direction and wind speed as independent variables in multilinear receptor modeling studies." Chemometrics and Intelligent Laboratory Systems 60(1-2): 25-41.

Paatero, P. and P. K. Hopke (2003). "Discarding or downweighting high-noise variables in factor analytic models." Analytica Chimica Acta 490(1-2): 277-289.

Paatero, P., P. K. Hopke, et al. (2005). "A graphical diagnostic method for assessing the rotation in factor analytical models of atmospheric pollution." Atmospheric Environment 39(1): 193-201.

Paatero, P., P. K. Hopke, et al. (2002). "Understanding and controlling rotations in factor analytic models." Chemometrics and Intelligent Laboratory Systems 60(1-2): 253-264.

Ramadan, Z., B. Eickhout, et al. (2003). "Comparison of Positive Matrix Factorization and Multilinear Engine for the source apportionment of particulate pollutants." Chemomet. Intellig. Lab. Syst. 66(1): 15-28.

Reisen, F., C. P. Meyer, et al. (2011). "Impact of smoke from biomass burning on air quality in rural communities in southern Australia." Atmospheric Environment 45(24): 3944-3953.

Salma, I., X. Chi, et al. (2004). "Elemental and organic carbon in urban canyon and background environments in Budapest, Hungary." Atmos. Environ. 38(1): 27-36.

Salma, I. and W. Maenhaut (2006). "Changes in elemental composition and mass of atmospheric aerosol pollution between 1996 and 2002 in a Central European city." Environmental Pollution 143(3): 479-488.

Schauer, J. J., G. C. Lough, et al. (2006). "Characterisation of metals emitted from motor vehicles. Research Report 133. Health Effects Institute, Boston."

Scott, A. J. (2006). Source Apportionment and Chemical Characterisation of Airborne Fine Particulate Matter in Christchurch, New Zealand. PhD Thesis, University of Canterbury.



Song, X. H., A. V. Polissar, et al. (2001). "Sources of fine particle composition in the northeastern US." Atmospheric Environment 35(31): 5277-5286.

Swietlicki, E., S. Puri, et al. (1996). "Urban air pollution source apportionment using a combination of aerosol and gas monitoring techniques." Atmospheric Environment 30(15): 2795-2809.

Thurston, G. D. and J. D. Spengler (1985). "A quantitative assessment of source contributions to inhalable particulate matter pollution in metropolitan Boston." Atmos. Environ. 19(1): 9-25.

Trompetter, W., P. K. Davy, et al. (2007). Ambient air particulate source activity measured with hourly resolution. Proc. 14th International Union of Air Pollution Prevention and Environmental Protection Associations (IUAPPA) World Congress 2007, 18th Clean Air Society of Australia and New Zealand (CASANZ) Conf., Brisbane, QLD.

Trompetter, W., A. Markwitz, et al. (2005). "Air particulate research capability at the New Zealand Ion Beam Analysis Facility using PIXE and IBA Techniques." International Journal of PIXE 15(3&4): 249-255.

Trompetter, W. J. (2004). Ion Beam Analysis results of air particulate filters from the Wellington Regional Council. Wellington, Geological and Nuclear Sciences Limited.

Trompetter, W. J. and P. K. Davy (2005). Air Particulate Research Capability at the New Zealand Ion Beam Analysis facility using PIXE and IBA techniques. BioPIXE 5, Wellington, New Zealand.

Trompetter, W. J., P. K. Davy, et al. (2010). "Influence of environmental conditions on carbonaceous particle concentrations within New Zealand." Journal of Aerosol Science 41(1): 134-142.

Watson, J. G., T. Zhu, et al. (2002). "Receptor modeling application framework for particle source apportionment." Chemosphere 49(9): 1093-1136.

Wilton, E., J. Smith, et al. (2007). Source identification and apportionment of PM10 and PM2.5 in Hastings and Auckland, NIWA.



APPENDICES



APPENDIX 1 ANALYSIS METHODOLOGY

A.1.1 Ion Beam Analysis

Ion Beam Analysis (IBA) was used to measure the elemental concentrations of particulate matter on the size-resolved filter samples from the Johnstone’s Hill tunnel monitoring site shown in Figure 2.1. IBA is based on the measurement of X-rays and -rays characteristic and particles of an element produced by the ion-atom interaction using high-energy protons in the two to five million electron volt (MeV) range. IBA is a mature and well developed science with many research groups around the world using IBA in a variety of routine analytical applications including analysis of atmospheric aerosol (Maenhaut and Malmqvist 2001; Trompetter, Markwitz et al. 2005). IBA techniques do not require sample preparation and are fast, non-destructive and sensitive (Cohen 1999; Maenhaut and Malmqvist 2001; Trompetter, Markwitz et al. 2005).

IBA measurements for this study were carried out at the New Zealand IBA facility operated by GNS. Figure A1.1 shows the air particulate matter analysis chamber with its associated X-ray, -ray and particle detectors for Proton Induced X-ray Emission (PIXE), Proton Induced Gamma-ray Emission (PIGE), Proton Elastic Scattering Analysis (PESA) and Rutherford Back-Scatter (RBS) measurements.

Figure A1.1 Air particulate matter analysis chamber with its associated particle detectors

The following sections provide a generalised overview of the IBA techniques used for elemental analysis and the analytical set-up at GNS (Cohen, Bailey et al. 1996; Cohen 1998; Trompetter 2004; Trompetter and Davy 2005). Figure A1.2 presents a schematic diagram of the typical experimental setup for IBA of air particulate filters at GNS.



Figure A1.2 Schematic of the typical IBA experimental set-up at GNS.

A.1.1.1 Particle Induced X-Ray Emission

Particle induced X-ray emission (PIXE), is used to determine elemental concentrations heavier than neon by exposing the filter samples to a proton beam accelerated to 2.5 million volts (MV) from the GNS 3 MV van-de-Graaff accelerator. When high energy protons interact with atoms in the sample, characteristic X-rays (from each element) are emitted by ion-electron processes. These X-rays are recorded in an energy spectrum. While all elements heavier than boron emit K X-rays, the production of them become too few to satisfactorily measure elements heavier than strontium. Elements heavier than strontium are detected via their lower energy L X-rays. The X-rays are detected by means of a Si(Li) detector and the pulses from the detector are amplified and recorded in a pulse height analyser. In practice sensitivities are further improved for the lighter elements by using two X-ray detectors, one for light element X-rays and the other for heavier element X-rays, each with different filtering and collimation. Figure A1.3 shows an example of a PIXE spectrum for airborne particles collected on a filter and analysed at the GNS IBA facility.

Particle beam 1H+

RBS particle detector

Target filter

apertures

ion energy 2.5 MeV

X-ray detector 1

X-ray detector 2

135 o

X-ray filter 1

X-ray filter 2

45 o 135 o

-ray detector

PESA particle detector

45 o



Figure A1.3 Typical PIXE spectrum for an aerosol sample analysed by PIXE.

As the PIXE spectrum consists of many peaks from different elements (and a Bremsstrahlung background), some of them overlapping, the spectrum is analysed with quantitative X-ray analysis software. In the case of this study, Gupix Software was used to perform the deconvolution with high accuracy (Maxwell, Cambell et al. 1989; Maxwell, Teesdale et al. 1995). The number of pulses (counts) in each peak for a given element is used by the Gupix software to calculate the concentration of that element. The background and neighbouring elements determine the statistical error and the limit of detection. Note, that Gupix provides a specific statistical error and limit of detection (LOD) for each element in any filter, which is essential for source apportionment studies.

Typically 20-25 elements from Mg to Pb are routinely determined above their respective LODs. Sodium (and fluorine) was determined by both PIXE and PIGE (see next section). Specific experimental details, where appropriate, are given in the results and analysis section for each site.

A.1.1.2 Particle Induced Gamma-Ray Emission

Particle Induced Gamma-Ray Emission (PIGE) refers to -rays produced when an incident beam of protons interacts with the nuclei of an element in the sample (filter). During the de-excitation process, nuclei emit -ray photons of characteristic energies specific to each element. Typical elements measured with -ray are:

Element nuclear reaction gamma ray energy (keV)

Sodium 23Na(p,αγ)20Ne 440, 1634

Fluorine 19F(p,αγ)16O 197, 6129

Gamma rays are higher in energy than X-rays and are detected with a germanium detector. Measurements of a light element such as sodium can be measured more accurately using PIGE because the -rays are not attenuated to the same extent in the filter matrix or the

1 2 3 4 5 6 7 8 9 1010

100

1000

10000

100000

Cl esc.

Mg

Ti

K

Al

Na

S

Ca

Fe

Cl

Fe K

Si

Co

unts

Energy (keV)



detector material, a problem in the measurement of low energy X-rays of sodium. Figure A1.4 shows a typical PIGE spectrum.

Figure A1.4 Typical PIGE spectrum for an aerosol sample analysed by PIGE.

A.1.1.3 Particle Elastic Scattering Analysis

Particle Elastic Scattering Analysis (PESA) allows hydrogen to be measured quantitatively in air particulate matter collected on a filter providing the filter material contains no or little hydrogen atoms, e.g. Teflon filters. Note that Teflon contains fluorine that introduces a significant background in the X-ray spectra which increases the limits of detection (LODs) of PIXE. Hydrogen can be detected by measuring the elastically scattered protons in a forward direction for a proton beam passing through the air particulate matter filter. At a forward scattering angle of 45º, the protons are elastically scattered from hydrogen with 50 % of the initial proton energy (i.e. for an incident beam of 2.50 MeV the energy of protons scattered off hydrogen is 1.25 MeV) which is much less energy than the energy of the protons scattered from the other heavier elements in the filter. Thus, in the PESA spectrum of a sample filter, a peak corresponding to protons elastically scattered from hydrogen occurs separated from the protons elastically scattered from the other atoms in the air particulate matter filter. The air particulate matter filter is thin enough for this measurement when the hydrogen PESA peak is separated from the noise at the low end of the spectrum and from protons elastically scattered from heavier atoms at the high energy end of the spectrum. For Teflon filters analysed with a 2.5 MeV proton beam, proton scattering energies for PESA are shown in Table A1.1 and Figure A1.5 presents a typical PESA spectrum.

50 100 150 200 250 300 350 400 450 500 550 600

100

1000

10000

100000

511 keV electron-positron

annihilation

23Na(p,)20Ne

19F(p,)16O

Yie

ld (

Co

unts

)

Energy (keV)



Table A1.1 Proton scattering energies of various elements for a 2.5 MeV proton beam

Element Energy detected at 45º

forward angle (MeV)

H 1.250

C 2.380

O 2.410

F 2.424

Fe 2.474

Figure A1.5 PESA spectrum for an aerosol sample showing the hydrogen peak at 1.250 MeV.

As PESA and IBA measurements in general, are conducted in high vacuum (residual gas pressure better than 106 mbar), free water vapour and VOCs are volatilised before analysis, only bound hydrogen is detected (e.g. SVOCs and ammonium ions) (Cohen 1999).

A.1.1.4 IBA Data reporting

Most filters used to collect particulate matter samples for IBA analysis are sufficiently thin that the ion beam penetrates the entire depth producing a quantitative analysis of elements present. Due to the thin nature of the air particulate matter filters, the concentrations reported from the IBA analyses are therefore in aerial density units (ng cm2) and the total concentration of each element on the filters is calculated by multiplying with the exposed area of the filter. Typically the exposed area is 0.016 cm2 for filters collected with the Streaker sampler used in this study. For example, to convert from Cl (ng cm2) into Cl (ng m3) for filter samples, the equation is:

0 500 1000 1500 2000 2500

100

1000

10000

100000Proton's scattered

from elements heavier than Hydrogen

Hydrogen "peak"

Energy (keV)

Cou

nts

Hydrogen peak



Cl (ng m-3) = 0.016(cm2) Cl (ng cm-2) / Vol(m3) (A.1)

A.1.1.5 Limits of Detection for elements determined by IBA

The exact limits of detection for reporting the concentration of each element depends on a number of factors such as:

the method of detection; filter composition; sample composition; the detector resolution; spectral interference from other elements.

In order to determine the concentration of each element the background is subtracted and peak areas fitted and calculated. The background occurs through energy loss, scattering and interactions of the ion beam as it passes through the filter material or from -rays produced in the target and scattered in the detector system (Cohen 1999). The peaks of elements in spectra that have interferences or backgrounds from other elements present in the air particulate matter, or filter matrix itself, will have higher limits of detection. Choice of filter material is an important consideration with respect to elements of interest as is avoiding sources of contamination. The GNS IBA laboratory routinely runs filter blanks to correct for filter derived analytical artefacts as part of their QA/QC procedures. Figure A1.6 shows the LODs typically achieved by PIXE for each element at the GNS IBA facility.

Figure A1.6 Elemental limits of detection routinely achieved at the GNS IBA facility for air filters with

PIXE.

10 20 30 40 50 60 70 80 90 100

1

10

100

1000

10000

lower LOD withdetector 2

L-line X-raysK-line X-rays

lower LOD withdetector 1

Element atomic number

Ave

rage

det

ect

ion

limit

(ng/

cm2 )

detector1 detector 2



All IBA elemental concentrations determined in this work were accompanied by their respective LODs. The use of elemental LODs is important in receptor modelling applications and is discussed further in Section A.4.2.

A.1.2 Black carbon measurements

Black carbon (BC) has been studied extensively but it is still not clear to what degree it is elemental carbon (EC (or graphitic) C(0)) or high molecular weight refractory weight organic species or a combination of both (Jacobson, Hansson et al. 2000). Current literature suggests that BC is likely a combination of both, and that for combustion sources such as petrol and diesel fuelled vehicles and biomass combustion (wood burning, coal burning), EC and organic carbon compounds (OC) are the principal aerosol components emitted (Jacobson, Hansson et al. 2000; Fine, Cass et al. 2001; Watson, Zhu et al. 2002; Salma, Chi et al. 2004).

Determination of carbon (soot) on filters was performed by light reflection to provide the BC concentration. The absorption and reflection of visible light on particles in the atmosphere or collected on filters is dependent on the particle concentration, density, refractive index and size. For atmospheric particles, BC is the most highly absorbing component in the visible light spectrum with very much smaller components coming from soils, sulphates and nitrate (Horvath 1993; Horvath 1997). Hence, to the first order it can be assumed that all the absorption on atmospheric filters is due to BC. The main sources of atmospheric BC are anthropogenic combustion sources and include biomass burning, motor vehicles and industrial emissions (Cohen, Taha et al. 2000). Cohen and co-workers found that BC is typically 10 – 40 % of the fine mass (PM2.5) fraction in many urban areas of Australia.

When measuring BC by light reflection/transmission, light from a light source is transmitted through a filter onto a photocell. The amount of light absorption is proportional to the amount of black carbon present and provides a value that is a measure of the black carbon on the filter. Conversion of the absorbance value to an atmospheric concentration value of BC requires the use of an empirically derived equation (Cohen, Taha et al. 2000):

BC (g cm-2) = (100/2(F)) ln[R0/R] (A.2)

where:

is the mass absorbent coefficient for BC (m2 g-1) at a given wavelength; F is a correction factor to account for other absorbing factors such as sulphates,

nitrates, shadowing and filter loading. These effects are generally assumed to be negligible and F is set at 1.00;

R0, R are the pre- and post- reflection intensity measurements respectively.

Black carbon was measured at GNS using the M43D Digital Smoke Stain Reflectometer. The following equation (from Willy Maenhaut, Institute for Nuclear Sciences, University of Gent Proeftuinstraat 86, B-9000 GENT, Belgium) is used for obtaining BC from reflectance measurements on Nucleopore polycarbonate filters or Pall Life Sciences Teflon filters:

BC (g cm-2) = [1000 LOG(Rblank/Rsample) + 2.39] / 45.8 (A.3)

where:



Rblank: the average reflectance for a series of blank filters; R_blank is close (but not identical) to 100; GNS always use the same blank filter for adjusting to 100.

Rsample: the reflectance for a filter sample (normally lower than 100).

With: 2.39 and 45.8 constants derived using a series of 100 Nucleopore polycarbonate filter samples which served as secondary standards; the BC loading (in g cm-2) for these samples had been determined by Prof. Dr. M.O. Andreae (Max Planck Institute of Chemistry, Mainz, Germany) relative to standards that were prepared by collecting burning acetylene soot on filters and determining the mass concentration gravimetrically (Trompetter 2004).

A.1.3 Receptor Modelling Methodology

Two receptor modelling approaches were combined for this study to provide a robust understanding of the number of primary sources and relative source contributions to airborne particles in the Johstone’s Hill tunnel. Principal Components Analysis (PCA) was used to provide an initial indication of the number of sources that may be contributing to the sample. Positive Matrix Factorisation (PMF) was used to apportion mass contributions and determine relative uncertainties and closeness of fit of the model to the data. This approach is used successfully for source apportionment studies by researchers in New Zealand and overseas.

A.1.3.1 Principal Components Analysis

PCA is a multivariate technique for identifying patterns in a data matrix and expressing these patterns in such a way as to highlight their similarities and differences. PCA has commonly been used on its own for receptor modelling studies (Thurston and Spengler 1985; Maenhaut and Cafmeyer 1987; Swietlicki, Puri et al. 1996; Kavouras, Koutrakis et al. 2001). Multivariate models as applied to receptor modelling of airborne particles are based on the assumption that chemical species from the same source are correlated and that grouped correlations represent an emission source. PCA extracts a number of factors (representing sources) from a dataset of chemical species, identifies the chemical profile of each source and indicates source significance on a per sample basis using factor scores. It is important to note that PCA is based on statistical associations (variance and covariance) in the dataset rather than being derived from physical or chemical principles. An important aspect of source apportionment of airborne particles is recognising the physical (i.e. the ‘real world’) significance of the results obtained by factor analysis of the dataset. This can be greatly assisted if specific source profiles are available.

A.1.3.1.1 PCA model outline

Understanding the mathematics behind the calculations in PCA helps with interpretation. The goal of a factor analysis is to characterise the p variables (elemental concentrations) in sample X in terms of a small number m of common factors F, which impact all the variables and a set of errors or specific factors , which only affect a single X variable i.e. the unexplained variance not fitted by the PCA model (STATGRAPHICS Manual – Rev. 1/10/2005, Statpoint Inc.). In matrix notation this can be written:

X = AF (A.4)

Where



X is the m x n pollutant concentration data matrix and

A is the m x p matrix containing the p source profiles and

F is the p x n matrix of source contributions.

The receptor model attempts to estimate A and F so that X is reproduced with sufficient accuracy (Swietlicki, Puri et al. 1996).

In practice, PCA results are calculated using an Eigenvector analysis of a correlation or covariance matrix. Firstly the data is standardised by subtracting the mean so that each pollutant species has equal weighting and from this the correlation matrix is calculated. There are n Eigenvectors of a correlation matrix with n x n dimensions and each of these Eigenvectors is orthogonal to the other no matter how many dimensions there are, therefore the data can then be expressed in terms of these Eigenvectors – PCA uses unit Eigenvectors (i.e. their length = 1).

The physical interpretation of Eigenvectors is that they provide information about patterns in the data. Eigenvalues are a scalar multiplier of Eigenvectors and the Eigenvector with the highest Eigenvalue is the principle component of the dataset. In PCA the Eigenvectors are ordered by Eigenvalue – this provides the components (factors) in order of significance and there are as many factors as variables (elemental species) in the original dataset. In practice we are only interested in the factors with the highest Eigenvalues i.e. those that explain most of the variance within the dataset and lesser ones are ignored as they are generally due to noise (unexplained variance) in the original dataset (Hopke 1999). The exact cut-off point is not scientifically defined and largely dependent on user experience, the principle guiding factors being:

the physical significance of the factors in terms of identifiable sources; the number of variables included in the original data matrix; the natural variance in the data due to measurement error or analytical noise (i.e.

concentrations of elements near the detection limit will have a higher level of uncertainty or variance).

Each factor contains all the elements (or other analyte species) used in the original dataset and alongside each is a factor loading. Species that have been determined to vary together through the dataset will have a high factor loading with the exact figure indicating how strongly it is associated with that factor (Swietlicki, Puri et al. 1996). Quite often the initial factor extraction does not give easily interpretable results and so the factor loadings may be rotated (another feature of PCA software) to provide factors that can be named and interpreted more easily. Essentially, rotation makes the large loadings larger than before and the smaller loadings smaller. There are number of different types of rotation possible with most software packages (varimax, quartimax, orthomax etc.) with varimax the most commonly used (Hopke 1999).

Several diagnostic outputs from PCA were found to be useful in the interpretation of the results as follows:

Estimated communality



Provides an indication (number between 0 and 1) how well each element was modelled by the extracted factors, a value closer to 1 indicating good correlation.

Percent variance explained

Indicates percent of variance in the dataset explained by each factor, or cumulative percent explained indicates how much of the total variance was explained by the extracted factors.

All PCA results have been expressed with the estimated communality and percent variance explained along with a Scree plot of the Eigenvalues for each factor. A number of statistical packages will perform PCA, for this study STATGRAPHICS was used.

A.1.3.1.2 Limitations of PCA

There are several constraints to PCA that need to be kept in mind when analysing the results (Paatero, Hopke et al. 2005):

1. The extracted factors are not necessarily physically related to the real world (negative contributions are possible) and the physical significance of rotation is unknown;

2. PCA is a least squares fit weighted by implicit unrealistic errors for the variables in the data matrix and do not necessarily reflect the errors in sampling and analysis;

3. PCA cannot account for missing or below detection limit data that commonly occur in environmental measurements;

4. PCA does not consider limits of detection and uncertainties.

Another problem with PCA is that it does not directly apportion the mass contributions of sources, a primary goal in receptor studies. PCA has been adapted using some simple matrix algebra to develop a receptor modelling method called absolute PCA (APCA) where the relative contribution of each factor to elemental concentrations can be estimated through the factor scores, however, it is still not an absolute score as the factor scores still have a standard deviation of 1 (Maenhaut and Cafmeyer 1987), nor does APCA address the issues outlined in points 1-3 above. Nevertheless, PCA is a useful method for examining the number of significant sources that may be contributing ambient particle concentrations at a receptor (monitoring) site and assists with modelling and interpretation of more robust receptor models such as Positive Matrix Factorisation as discussed in the next section.

A.1.3.2 Positive Matrix Factorisation

Positive matrix factorisation (PMF) is a linear least-squares approach to factor analysis and was designed to overcome the receptor modelling problems associated with PCA as outlined in the previous section (Paatero, Hopke et al. 2005). With PMF, sources are constrained to have non-negative species concentrations, no sample can have a negative source contribution and error estimates for each observed data point are used as point-by-point weights. This feature is a distinct advantage, in that it can accommodate missing and below detection limit data that is a common feature of environmental monitoring results (Song, Polissar et al. 2001). In fact, the signal to noise ratio for an individual elemental measurement can have a significant influence on a receptor model and modelling results. For the weakest (closest to detection limit) species the variance may be entirely due to noise



(Paatero and Hopke 2002). Paatero and Hopke strongly suggest down-weighting or discarding noisy variables (particularly for PCA) that are always below their detection limit or species that have a lot of error in their measurements relative to the magnitude of their concentrations (Paatero and Hopke 2003). The distinct advantage of PMF is that mass concentrations can be included in the model and the results are directly interpretable as mass contributions from each factor (source) i.e. the analyses and modelling is conducted in real space rather than standardised space as in PCA.

A.1.3.2.1 PMF model outline

The mathematical basis for PMF is described in detail by Paatero (Paatero 1997; Paatero 2000) and Hopke (Hopke Undated). An overview is provided to highlight the differences to the PCA approach and how a combination of the two models assists greatly with the elucidation of sources from the elemental composition of airborne particles.

PMF uses a weighted least-squares fit with the known error estimates of measured elemental concentrations used to derive the weights. In matrix notation:

X = GF + E (A.5)

where:

X is the known n x m matrix of m measured elemental species in n samples; G is an n x p matrix of source contributions to the samples; F is a p x m matrix of source compositions (source profiles). E is a residual matrix – the difference between measurement X and model Y

E can be defined as a function of factors G and F:

Where:

i = 1,……,n elements j = 1,……,m samples k = 1,…...,p sources

PMF constrains all elements of G and F to be non-negative, meaning that elements cannot have negative concentrations and samples cannot have negative source contributions as in real space. The task of PMF is to minimise the function Q such that:

eij = xij – yij = xij –k = 1

p

gik fkj

Q(E) = j = 1

m

(eik / kj)2

i = 1

n

(A.6)



Where ij is the error estimate for xij. Another advantage of PMF is the ability to handle extreme values typical of air pollutant concentrations as well as true outliers that would normally skew PCA. In either case, such high values would have significant influence on the solution (commonly referred to as leverage) (Hopke Undated). PMF has been successfully applied to receptor modelling studies in a number of countries around the world (Hopke, Xie et al. 1999; Lee, Chan et al. 1999; Chueinta, Hopke et al. 2000; Song, Polissar et al. 2001; Lee, Yoshida et al. 2002; Kim, Hopke et al. 2003; Jeong, Hopke et al. 2004; Kim, Hopke et al. 2004; Begum, Hopke et al. 2005) including New Zealand (Scott 2006; Davy 2007; Davy, Trompetter et al. 2007; Davy, Trompetter et al. 2008; Davy, Trompetter et al. 2009; Davy, Trompetter et al. 2009).

A.1.3.2.2 PMF model used

Two programs have been written to implement different algorithms for solving the least squares PMF problem, these are PMF2 and the Multilinear Engine (ME-2) (Hopke, Xie et al. 1999; Ramadan, Eickhout et al. 2003). In effect, the ME-2 program provides a more flexible framework than PMF2 for controlling the solutions of the factor analysis with the ability of imposing explicit external constraints.

This study used both the PMF2 program and a recently released software version (EPA.PMF3.1) with a graphical user interface (GUI) based on the ME-2 program. Both PMF2 and EPA.PMF programmes can be operated in a robust mode, meaning that “outliers” are not allowed to overly influence the fitting of the contributions and profiles (Eberly 2005). The user specifies two input files, one file with the concentrations and one with the uncertainties associated with those concentrations. The methodology for developing an uncertainty matrix associated with the elemental concentrations for this work is discussed in Section A.4.2.

A.1.3.2.3 PMF model inputs

The PMF programs provide the user with a number of choices in model parameters that can influence the final solution. Two parameters, the ‘signal-to-noise ratio’ and the ‘species category’ are of particular importance and are described below.

Signal-to-noise ratio - this is a useful diagnostic statistic estimated from the input data and uncertainty files using the following calculation:

Where xij and ij are the respective concentration and uncertainty of the ith element in the jth sample. Smaller signal to noise ratios indicate that the measured elemental concentrations are generally near the detection limit and the user should consider whether to include that species in the receptor model or at least be strongly down weighted (Paatero and Hopke 2003). The signal-to-noise ratios (S/N ratio) for each element are reported alongside other statistical data in the results section.

Species category - this enables the user to specify whether the elemental species should

i = 1

n

(ij)2(1/2) /

i = 1

n

(xij)2 (A.7)



be considered:

Strong – whereby the element is generally present in concentrations well above the LOD (high signal to noise ratio) and the uncertainty matrix is a reasonable representation of the errors.

Weak – where the element may be present in concentrations near the LOD (low signal to noise ratio); there is doubt about some of the measurements and/or the error estimates; or the elemental species is only detected some of the time. If ‘Weak’ is chosen EPA.PMF increases the user-provided uncertainties for that variable by a factor of 3.

Bad – that variable is excluded from the model run.

For this work, an element with concentrations at least 3 times above the LOD, a high signal to noise ratio (> 2) and present in all samples was considered ‘Strong’. Variables were labelled as weak if their concentrations were generally low, had a low signal to noise ratio, were only present in a few samples or there was a lower level of confidence in their measurement. Mass concentration gravimetric measurements and BC were also down weighted as ‘Weak’ due to their concentrations generally being several orders of magnitude above other species which can have the tendency to ‘pull’ the model. Paatero and Hopke recommend that such variables be down weighted and that it doesn’t particularly affect the model fitting if those variables are from real sources (Paatero and Hopke 2003). What does affect the model severely is if a dubious variable is over-weighted. Elements that had a low signal to noise ratio (< 0.2), or had mostly missing (zero) values, or doubtful for any reason, were labelled as ‘Bad’ and were subsequently not included in the analyses.

If the model is appropriate for the data and if the uncertainties specified are truly reflective of the uncertainties in the data, then Q (according to Eberly) should be approximately equal to the number of data points in the concentration data set (Eberly 2005):

Theoretical Q = # samples x # species measured (A.8)

However, a slightly different approach to calculating the Theoretical Q value was recommended by (Brown and Hafner 2005), which takes into account the degrees of freedom in the PMF model and the additional constraints in place for each model run. This theoretical Q calculation Qth is given as:

Qth = (# samples x # good species)+[(# samples x # weak species)/3] - (# samples x factors estimated) (2.9)

Both approaches have been taken into account for this study and it is likely that the actual value lies somewhere between the two.

In PMF, it is assumed that only the xij’s are known and that the goal is to estimate the contributions (gik) and the factors (or profiles) (fkj). It is assumed that the contributions and mass fractions are all non-negative, hence the “constrained” part of the least-squares. Additionally, EPA.PMF allows the user to say how much uncertainty there is in each xij. Species-days with lots of uncertainty are not allowed to influence the estimation of the contributions and profiles as much as those with small uncertainty, hence the “weighted” part of the least squares and the advantage of this approach over PCA.



Diagnostic outputs from the PMF models were used to guide the appropriateness of the number of factors generated and how well the receptor modelling was accounting for the input data. Where necessary, initial solutions have been ‘rotated’ to provide a better separation of factors (sources) that were considered physically reasonable (Paatero, Hopke et al. 2002). Each PMF model run reported in this study is accompanied by the modelling statistics along with comments where appropriate.

A.1.4 Dataset Quality Assurance

Quality assurance of sample elemental datasets is vital so that any dubious samples, measurements and outliers are removed as these will invariably affect the results of receptor modelling. In general, the larger the dataset used for receptor modelling, the more robust the analysis. The following sections describe the methodology used to check data integrity and provide a quality assurance process that ensured that the data being used in subsequent factor analysis was as robust as possible.

A.1.4.1 Mass reconstruction and mass closure

Once the sample analysis for the range of analytes has been carried out, it is important to check that total measured mass does not exceed gravimetric mass (Cohen 1999). Ideally, when elemental analysis and organic compound analysis has been undertaken on the same sample one can reconstruct the mass using the following general equation for ambient samples as a first approximation (Cahill, Eldred et al. 1989; Malm, Sisler et al. 1994; Cohen 1999):

(A.10)

where:

The reconstructed mass (RCM) is based on the fact that the six composite variables or ‘pseudo’ sources given in equation A.10 are generally the major contributors to fine and coarse particle mass and are based on geochemical principles and constraints. The Soil factor contains elements predominantly found crustal matter (Al, Si, Ca, Fe, Ti) and includes a multiplier to correct for oxygen content and an additional multiplier of 1.16 to correct for the fact that three major oxide contributors (MgO, K2O, Na2O), carbonate and bound water are excluded from the equation. Organic carbon concentrations (OC) were estimated by using equation A.11 where PESA was used to determine the hydrogen concentration on filters. In this case total hydrogen on the filter was assumed to be comprised mainly of H from organic

[Soil] = 2.20[Al] + 2.49[Si] + 1.63[Ca] + 2.42[Fe] + 1.94[Ti]

[OC] = Σ[Concentrations of organic compounds]

[BC] = Concentration of black carbon (soot)

[Smoke] = [K] 0.6[Fe]

[Seasalt] = 2.54[Na]

[Sulphate] = 4.125[S]

Reconstructed mass = [Soil] + [OC] + [BC] + [Smoke] + [Sulphate] + [Seasalt]



material and ammonium sulphate (assuming sulphate is in fully neutralised form) and therefore organic content (designated OMH) was calculated from total H by the following equation (Malm, Sisler et al. 1994; Cohen 1999):

[OMH] = 11([H] – 0.25[S]) (A.11)

Equation A.11 assumes that average particulate organic matter is composed of 11% H, 71% C, and 20 % O by weight. Where a measure of [OC] was not available it was assumed that it composed part of the ‘remaining mass’ (the difference between RCM and gravimetric mass) that includes water and nitrates as major components (Cahill, Eldred et al. 1989).

[BC] is the concentration of black carbon, measured in this case by light reflectance/absorbance. [Smoke] represents K not included as part of crustal matter and tends to be an indicator of biomass burning.

[Seasalt] represents the marine aerosol contribution and assumes that the NaCl weight is 2.54 times the Na concentration. Na is used as it is well known that Cl can be volatilised from aerosol or from filters in the presence of acidic aerosol, particularly in the fine fraction via the following reactions (Lee, Chan et al. 1999):

NaCl(p) + HNO3(ag) NaNO3(p) + HCL(g) (A.12)

2NaCl(p) + H2SO4(ag) Na2SO4(p) + 2HCL(g) (A.13)

Alternatively, where Cl loss is likely to be minimal, such as in the coarse fraction or for both size fractions near coastal locations and relatively clean air in the absence of acid aerosol then the reciprocal calculation of [Seasalt] = 1.65[Cl] can be substituted, particularly where Na concentrations are uncertain.

Most fine sulphate particles are the result of oxidation of SO2 gas to sulphate particles in the atmosphere (Malm, Sisler et al. 1994). It is assumed that sulphate is present in fully neutralised form as ammonium sulphate. [Sulphate] therefore represents the ammonium sulphate contribution to aerosol mass with the multiplicative factor of 4.125[S] to account for ammonium ion and oxygen mass (i.e. (NH4)2SO4 = ((14 + 4)2 + 32 + (16x4)/32).

Additionally the sulphate component not associated with seasalt can be calculated by equation A.14 (Cohen 1999):

Non-seasalt sulphate (NSS-Sulphate) = 4.125 ([Stot] - 0.0543[Cl]) (A.14)

Where the sulphur concentrations contributed by seasalt is inferred from the chlorine concentrations i.e. [S/Cl]seasalt = 0.0543 and the factor of 4.125 assumes that the sulphate has been fully neutralised and is generally present as (NH4)2SO4 (Cahill, Eldred et al. 1990; Malm, Sisler et al. 1994; Cohen 1999).

The RCM and mass closure calculations using the pseudo-source and pseudo-element approach are a useful way to examine initial relationships in the data and how the measured mass of species in samples compares to gravimetric mass. Note that some scatter is possible as not all aerosols are necessarily measured and accounted for such as all OC, ammonium species, nitrates and unbound water.



As a quality assurance mechanism, those samples for which RCM exceeded gravimetric mass or where gravimetric mass was significantly higher than RCM were examined closely to assess gravimetric mass and IBA data. Where there was significant doubt either way, those samples were excluded from the receptor modelling analysis. The reconstructed mass calculations and pseudo source estimations are presented in the appendices at the end of this report.

A.1.4.2 Dataset preparation

Careful preparation of a dataset is required as serious errors in data analysis and receptor modelling results can be caused by erroneous individual data values. The general methodology followed for dataset preparation was as recommended by (Brown and Hafner 2005). For this study all data was checked for consistency with the following parameters:

1. Individual sample collection validation; 2. Gravimetric mass validation; 3. Analysis of RCM versus gravimetric mass to ensure RCM < gravimetric; 4. Identification of unusual values including noticeably extreme values and values that

normally track with other species (e.g. Al and Si) but deviate in one or two samples. Scatter plots and time series plots were used to identify unusual values. One-off events such as fireworks displays, forest fires or vegetative burn-offs may affect a receptor model as it is forced to find a profile that matches only that day;

5. Species were included in a dataset if at least 70 % of data was above the LOD and signal-to-noise ratios were checked to ensure data had sufficient variability. Important tracers of a source where less than 70 % of data was above the LOD were included but model runs with and without the data were used to assess the effect;

6. For PCA, % errors and signal-to-noise ratios were used as a guide as to whether a species was too ‘noisy’ to include in an analysis.

In practise during data analyses the above steps were a reiterative process of cross checking as issues were identified and corrected for, or certain data excluded and the effects of this were then studied.

A.1.4.2.1 PMF data matrix population

The following steps were followed to produce a final dataset for use in the PMF receptor model (Brown and Hafner 2005).

Below detection limit data: For given values, the reported concentration used and the corresponding uncertainty checked to ensure it had a high value.

Missing data: Substituted with the dataset median value for that species.

A.1.4.2.2 PMF uncertainty matrix population

Uncertainties can have a large effect on model results so that they must be carefully compiled. The effect of underestimating uncertainties can be severe while overestimating uncertainties does not do too much harm (Paatero and Hopke 2003).



Uncertainties for data: Data was multiplied by % fit error provided by IBA analysis to produce an uncertainty in ng m-3.

Below detection limit data: Below detection limit data was generally provided with a high % fit error and this was used to produce an uncertainty in ng m-3. Zero data was given a corresponding uncertainty value of 4LOD.

Missing data: Uncertainty was calculated as 4 median value over the entire species dataset.

BC: Due to the high mass values for BC, the uncertainties were generated by multiplying mass values by a factor of four to down-weight the variable.

PM gravimetric mass: Uncertainty given as 4 mass value to down-weight the variable.

Reiterative model runs were used to examine the effect of including species with high uncertainties or low concentrations. In general it was found that the initial uncertainty estimations were sufficient and that adjusting the ‘additional modelling uncertainty’ function accommodated any issues with modelled variables such as those with residuals outside ± 3 standard deviations.



APPENDIX 2 Statistics and Diagnostics for Receptor Modelling

Scatterplot matrix of elemental data

Figures A.2.1 and A.2.2 present a scatterplot matrix of PM2.5 and PM10-2.5 elemental concentrations in Johnstone’s Hill tunnel respectively.

Figure A2.1 Scatterplot matrix of Johnstone’s Hill tunnel PM2.5 elemental data

PM10

BC

Na

Mg

Al

Si

S

Cl

K

Ca

Ti

Fe

Cu

Zn

Ba



Figure A2.2 Scatterplot matrix of Johnstone’s Hill tunnel PM10-2.5 elemental data

Mass reconstruction

Table A2.1 presents the reconstructed mass components (RCM) and mass closure (Total RCM) for the Johnstone’s Hill tunnel PM10 elemental data using the methodology described in Appendix 1.

PM10

BC

Na

Mg

Al

Si

S

Cl

K

Ca

Ti

Fe

Cu

Zn

Ba



Table A2.1 Reconstructed mass components and mass closure for PM10 in Johnstone’s Hill tunnel

Component of PM10 Average mass

g m−3

Percentage of PM10 mass

Soil 1.1 8

Seasalt 2.0 14

Sulphate 0.8 6

nss-Sulphate 0.5 4

Smoke 0.001 0.09

BC 3.9 29

Total RCM 7.8 60

Figure A2.3 presents a plot of reconstructed mass versus PM10 gravimetric mass

Figure A2.3 Plot of PM10 reconstructed mass versus gravimetric mass for Johnstone’s Hill tunnel PM samples

The mass reconstruction analysis assists with data quality assurance and identification of outliers that may skew the factor analysis and determination of source mass contributions.

0

10

20

30

40

0 10 20 30 40

PM

10R

ec

on

stru

cte

d M

as

s (g

/m3 )

PM10 BAM Mass (g/m3)



Principal components analysis (PCA)

The PCA provides relationships in the data where the factors extracted often represent sources. The number of factors is chosen so that noisy variables (near or below detection limits) generally appear in a factor of their own. Elements that are strongly associated (highly correlated) with other elements (i.e. representing a particular source) tend to appear in the same factor together no matter how many factors are chosen. Table A2.2 presents the PCA for the PM10 data from Johnstone’s Hill tunnel and Figure A2.4 shows the associated scree plot.

Table A2.2 PCA factor loadings matrix for PM10 in Johnstone’s Hill tunnel

Factor

1 Factor

2 Factor

3 Factor

4 Factor

5 Factor

6 Factor

7 Factor

8 Factor

9 Factor

10 Estimated

communality

Marine Vehicles Crustal Fine

crustal Fine

marine Biomass

NaC 0.97 -0.01 0.04 -0.02 0.08 0.02 -0.05 -0.01 0.03 0.03 0.96

MgC 0.84 0.15 0.06 0.15 -0.02 0.12 0.16 0.03 0.15 0.07 0.83

AlC -0.10 0.17 0.33 0.19 -0.02 0.00 0.87 0.08 0.10 0.08 0.96

SiC 0.09 0.39 0.78 0.22 0.03 -0.02 0.28 0.05 0.14 -0.08 0.93

S_C 0.97 0.07 0.16 0.01 0.07 0.03 -0.04 0.00 0.07 0.04 0.98

ClC 0.97 0.01 0.04 -0.03 0.13 0.01 -0.08 0.00 0.02 0.01 0.96

KC 0.89 0.16 0.21 0.04 -0.07 0.16 -0.05 -0.01 0.10 0.03 0.91

CaC 0.64 0.29 0.60 0.16 -0.01 0.01 0.15 0.03 0.13 -0.02 0.93

TiC 0.01 0.19 0.87 0.16 -0.04 0.14 0.07 0.02 0.10 0.17 0.88

MnC 0.32 0.08 0.22 0.01 0.06 0.09 0.10 -0.01 0.90 0.01 0.99

FeC 0.26 0.50 0.66 0.19 -0.12 0.09 0.15 0.04 0.09 0.31 0.94

CuC 0.17 0.34 0.53 0.14 -0.01 -0.01 0.13 0.02 0.02 0.68 0.92

ZnC 0.01 0.01 0.05 0.07 0.01 0.03 0.06 0.99 0.00 0.01 1.00

BC 0.10 0.76 0.40 0.22 0.03 0.01 0.08 0.02 0.04 0.16 0.84

Na 0.66 -0.06 -0.11 0.03 0.63 0.08 -0.01 0.01 0.09 0.02 0.86

Mg 0.07 0.07 0.14 0.89 -0.03 0.09 0.07 0.04 -0.05 0.09 0.86

Al -0.06 0.33 0.16 0.85 -0.03 0.04 0.10 0.05 0.05 0.01 0.88

Si 0.13 0.55 0.26 0.62 0.17 -0.06 0.07 0.02 0.07 -0.04 0.82

S 0.60 0.44 0.07 0.28 0.29 0.25 0.00 -0.02 0.13 0.18 0.85

Cl 0.74 0.02 -0.12 -0.12 0.57 0.10 -0.06 0.03 0.06 -0.06 0.92

K 0.29 0.27 0.12 0.08 0.11 0.86 -0.02 0.04 0.08 0.01 0.94

Ca 0.36 0.57 0.41 0.15 0.43 0.17 0.02 0.03 0.03 -0.08 0.88

Fe 0.08 0.78 0.36 0.24 0.03 0.14 0.01 0.03 0.02 0.25 0.89

Zn -0.01 0.71 0.10 0.17 -0.23 0.40 0.20 -0.03 0.05 -0.10 0.81



Figure A2.4 Scree plot of PCA Eigenvectors for PM10 in Johnstone’s Hill tunnel used to help identify the number of different sources

Scree Plot

0 5 10 15 20 25

Factor

0

2

4

6

8

10

Eig

enva

lue



Positive matrix factorisation

The PMF diagnostics for PM10 in Johnstone’s Hill tunnel are presented in Table A2.3.

Table A2.3 PMF diagnostics for Johnstone’s Hill tunnel (prior to any rotation)

Species Intercept Slope RMSE r2

PM10 4.8 0.83 3.5 0.3

NaC 40.83 0.88 66.47 0.97

MgC 9.41 0.69 5.33 0.8

AlC 2.38 0.9 4.16 0.92

SiC 2.82 0.95 6.92 0.98

S_C 1.74 0.96 5.61 0.99

ClC -8.37 1.01 51.3 1

KC 1.03 0.97 3.79 0.96

CaC 3.57 0.92 9.22 0.93

FeC 1.36 0.97 7.63 0.98

CuC 0.71 0.75 1.61 0.59

ZnC 1.96 0.44 2.45 0.3

BC 0.22 1 341.9 0.98

Na 67.56 0.71 89.02 0.72

Mg 28.71 0.7 14.21 0.59

Al 15.21 0.78 7.74 0.8

Si 22.3 0.78 12.15 0.82

S 21.04 0.82 14.79 0.86

Cl -3.11 1.01 19.48 1

K 19.91 0.44 9.53 0.47

Ca 11.87 0.67 6.86 0.73

Fe 7.76 0.86 11.11 0.92

Zn 1.63 0.27 2.9 0.29

COg 0.14 0.46 0.22 0.39

NOx 0 0.82 0.05 0.6

L1 7.91 0.09 10.13 0.13

HCV 1.92 0.35 2.58 0.31

Q Robust 906.652

Q True 906.646

Q Theoretical 3960

Number of bootstrapped factors mapped to original factor 1 : 200





Number of bootstrapped factors mapped to original factor 6: 157

Number of bootstrapped factors mapped to no original factor

concentration, composition and sources of particulate ...€¦ · concentration, composition and...

Documents