The use of real-time monitoring data to evaluate major sources of airborne particulate matter
Introduction
A recently published study describes how historical SO2 pollution data can be used to estimate public exposure to black smoke and particulates (Hodgson et al., 2009). Such quantitative information is now commonly available in most urban centres along with other routine measurements of air quality such as PM2.5, PM10, NOx, CO and O3. These measurements are generally made to monitor compliance with European and National Air Quality standards and attract little attention unless threshold exceedances are observed. Such extensive datasets also have, potentially, an important use with regard to the quantification of localised, homogeneous and heterogeneous air pollution sources. The use of standard air quality data in this manner has been little explored, although a small number of studies have been reported both for gaseous pollutants and particulates (Bruno et al., 2001, Guo et al., 2004, Grover and Eatough, 2008).
A meaningful reduction in population exposure to ambient PM can only be achieved cost-effectively if the sources of ambient PM are identified. Identifying source contributions, also known as source apportionment, often employs multivariate data analysis and matrix factorisation in receptor modelling (Hopke, 1991, Hopke, 1997, Hopke, 2003b). Factor analysis strategies attempt to determine simple patterns in the relationships between the measured variables i.e. if a group of chemicals or species possess a common origin, they should show similar variation with time. The objective is to discover latent features of the independent variables that are not measured directly. In other words, factor analysis seeks to determine whether observed variables can be explained in terms of a reduced number of variables, called factors. After identifying proper factors, the temporal trend of each can be related to historical events or human activities.
The fundamental relationship between an emission source and receptor (the sampling location) can be expressed as follows (Hopke, 2003b):where G is an n × p matrix representing source contributions to the samples, FT is the transpose of an m × p matrix of source profiles, E is the matrix of residuals, n refers to the number of samples, p is the number of variables or species and m is the number of extracted components or source categories. Each sample is an observation in time, and so G describes the temporal variation of the source contributions. The overall dataset matrix is made up of samples (rows) and species (columns). A viable statistical solution to the expression is sought, the aim of which is to minimise the residuals in E.
A basic and common factorisation technique employed is Principal Component Analysis, or PCA (Nordtvedt et al., 1996, Hopke, 2003a), which attempts to express two or more variables by principal components, which are extracted as solutions to equation (1) above. PCA has been employed in a number of receptor modelling studies (Hosiokangas et al., 1999, Garcia et al., 2006, Viana et al., 2006) in Europe (Harrison et al., 1996, Harrison et al., 1997, Fernandez Espinosa et al., 2004, Salvador et al., 2004, Almeida et al., 2005, Pires et al., 2008) and Asia (Fung and Wong, 1995, Harrison et al., 1997, Fang et al., 2006, Ho et al., 2006). The relative contribution of each source can be quantified using absolute principal component scores (Thurston and Spengler, 1985, Artaxo et al., 1999, Vallius et al., 2003, Vallius et al., 2005) and the contribution of each component (i.e. source) to the total ambient aerosol mass can be obtained by regression against measured aerosol mass.
A more advanced method is termed Positive Matrix Factorisation, or PMF (Hopke, 2003a, Ramadan et al., 2003). It applies constrained maximisation of a weighted object function, related to equation (1) above which takes into account the uncertainties associated with each measurement. Hence the variables are weighted by a measure of trust in the individual measurements and adjusted to their detection limits; “bad” samples can be down-weighted or excluded and the selected solution is based on goodness of fit. Results are also constrained to the non-negative, because the result is not meaningful if the sources turn out to have negative contributions to measured concentrations at the receptor sites. This method has been successfully employed in many studies (Huang et al., 1999, Ramadan et al., 2000, Kim et al., 2003, Kim et al., 2007, Zhao and Hopke, 2004, Chung et al., 2005, Ogulei et al., 2005, Ogulei et al., 2006, Zabalza et al., 2006, Zhao and Hopke, 2006, Hwang and Hopke, 2007, Pere-Trepat et al., 2007, Shrivastava et al., 2007, Viana et al., 2007), including one carried out in Ireland (Huang et al., 2001). However, it is important to evaluate different receptor models on the same dataset because the strengths of the different techniques combine to provide a more robust solution (Viana et al., 2008).
One issue common to factor analysis techniques is that large datasets of pollutant measurement are required. Obtaining a dataset often involves prolonged collection periods followed by highly labour-intensive laboratory analyses for either organic species, inorganic species (metals and ions), or both. Model output improves for data collected at high temporal resolution, which also impacts on the levels of analytical labour required. For example, a high temporal resolution of the data (e.g. half-hourly or hourly records) allows for a resolution of sources of volatile organic compounds and organic aerosol such as local and transported vehicle exhaust emissions, biogenic emissions, photochemical reaction products, which typically exhibit different temporal variation within any given 24 h period, depending on factors such as human activity and sunlight (Jorquera and Rappenglück, 2004, Grover and Eatough, 2008). A temporal resolution in the dataset corresponding to one record for every 24 h period would not allow for separation of emissions with higher frequency variation. Studies therefore have to strike a balance between intensive sampling campaigns (with high temporal resolution of collection that sacrifice seasonal information) and long-term sampling campaigns with, say, daily resolution that can incorporate seasonal information at the expense of temporal resolution. Naturally, if sufficient financial and human resources are available, high temporal resolution can also be obtained over the long term, but these constraints mean that such studies focus on a limited number of well-supported and established research “super-sites” (Huang et al., 2001, Hwang and Hopke, 2007, Saarikoski et al., 2008). Many other important, but less resourced, locations throughout the world do not have the benefit of such long-term and extensive data bases.
The aim of this work is to show that better, more efficient and predictive use can be made of routinely obtained air quality data, if the right parameters are included in the monitoring network and appropriate source apportionment models are applied to real-time measurements. Two models are presented here, based on PCA and PMF of real-time data records of NO, O3, NO2, SO2, EC, OC, particulate SO42− and PM2.5 obtained in Cork, Ireland.
Section snippets
Sampling site
The sampling site was located at Tivoli Docks in the Port of Cork (51°54′5 N, 8°24′38 W), approximately 3 km east of the city centre. The Port of Cork is part of a large natural harbour, an outline of which is shown in Fig. 1. A berth for liquid bulk ships is located approximately 150 m to the south-west and berths for container ships are situated 400–600 m to the west-south-west (Healy et al., 2009). The prevailing winds are south-westerly.
Real-time measurements
The sampling station was a customised commercial
Dataset of real-time air quality measurements
Table 1 shows overall mean, monthly, weekday and weekend averages of the real-time measurements of air quality made in 2008. The median and mean values were calculated for the period that the instruments were fully functional. As a result, annual averages can only be reported for OC and EC, whilst for the other parameters, the values refer to the relevant periods listed in the table. In terms of monthly averages, OC and EC levels clearly decrease between June and August. Although the datasets
Conclusion
This study presents analyses of air quality data obtained in the Port of Cork during 2008. Eight main marker pollutants were measured and the results are discussed in the context of three different likely source categories: road transport, domestic solid fuel burning and a source group that includes domestic and industrial oil-fired boilers, regional sulfate emissions, electricity generation and incineration. Two source apportionment models have been presented that are based solely on the
Acknowledgements
The authors would like to thank the Port of Cork for providing a suitable sampling site for accommodating the instruments. We are grateful to the Irish Environmental Protection Agency, Cork City Council, Science Foundation Ireland and the Irish Research Council for Science, Engineering and Technology for funding.
References (54)
- et al.
Source apportionment of fine and coarse particulate matter in a sub-urban area at the Western European Coast
Atmospheric Environment
(2005) - et al.
Aerosol composition and source apportionment in Santiago de Chile
Nuclear Instruments & Methods in Physics Research Section B-Beam Interactions with Materials and Atoms
(1999) - et al.
The transport sector as a source of air pollution
Atmospheric Environment
(2001) - et al.
Ambient air particulate concentrations and metallic elements principal component analysis at Taichung Harbor (TH) and WuChi Traffic (WT) near Taiwan Strait during 2004-2005
Journal of Hazardous Materials
(2006) - et al.
Source characterisation of fine urban particles by multivariate analysis of trace metals speciation
Atmospheric Environment
(2004) - et al.
Apportionment of air pollution sources by receptor models in Hong Kong
Atmospheric Environment
(1995) - et al.
The impacts of combustion emissions on air quality and climate – from coal to biofuels and beyond
Atmospheric Environment
(2009) - et al.
Determination of PM2.5 sources using time-resolved integrated source and receptor models
Chemosphere
(2006) - et al.
Source contributions to ambient VOCs and CO at a rural site in eastern China
Atmospheric Environment
(2004) - et al.
Comparative receptor modelling study of airborne particulate pollutants in Birmingham (United Kingdom), Coimbra (Portugal) and Lahore (Pakistan)
Atmospheric Environment
(1997)
Characterisation of single particles from in-port ship emissions
Atmospheric Environment
Source apportionment of PM2.5 in urban area of Hong Kong
Journal of Hazardous Materials
Predicting black smoke levels from deposit gauge and SO2 data to estimate long term exposure in the United Kingdom, 1956–61
Atmospheric Environment
An introduction to receptor modelling
Chemometrics and Intelligent Laboratory Systems
The evolution of chemometrics
Analytica Chimica Acta
Effects of soil dust episodes and mixed fuel sources on source apportionment of PM10 particles in Kuopio, Finland
Atmospheric Environment
Sources and source variations for aerosol at Mace Head, Ireland
Atmospheric Environment
Testing and optimizing two factor-analysis techniques on aerosol at Narragansett, Rhode Island
Atmospheric Environment
Estimation of source apportionment and potential source locations Of PM2.5 at a west coastal IMPROVE site
Atmospheric Environment
Receptor modeling of ambient VOC at Santiago, Chile
Atmospheric Environment
Source identification of PM2.5 in an arid Northwest US City by positive matrix factorization
Atmospheric Research
Characteristics of the major chemical constituents of PM2.5 and smog events in Seoul, Korea in 2003 and 2004
Atmospheric Environment
Receptor modeling for multiple time resolved species: the Baltimore supersite
Atmospheric Environment
Source apportionment of Baltimore aerosol from combined size distribution and chemical composition data
Atmospheric Environment
Source apportionment of time and size resolved ambient particulate matter measured with a rotating DRUM impactor
Atmospheric Environment
Management of air quality monitoring using principal component and cluster analysis – Part I: SO2 and PM10
Atmospheric Environment
Comparison of positive matrix factorization and Multilinear Engine for the source apportionment of particulate pollutants
Chemometrics and Intelligent Laboratory Systems
Cited by (27)
Review of online source apportionment research based on observation for ambient particulate matter
2021, Science of the Total EnvironmentCharacterization of air quality and sources of fine particulate matter (PM<inf>2.5</inf>) in the City of Calgary, Canada
2018, Atmospheric Pollution ResearchAssessment of annual air pollution levels with PM1, PM2.5, PM10 and associated heavy metals in Algiers, Algeria
2018, Environmental PollutionCitation Excerpt :Throughout the study, a p value of <0.05 was considered to indicate statistical significance. Principal component analysis (PCA) has been widely used to identify the pollution sources observed at a receptor site (Amodio et al., 2010; Andriani et al., 2010; Hellebust et al., 2010; Pant and Harrison, 2012). The purpose of PCA is to reduce the number of variables which explain the total variance in the data: this can be achieved by creating new orthogonal and uncorrelated variables, called principal components (PCs) that are linear combinations of the original variables.
Characterization of air quality and fine particulate matter sources in the town of Hinton, Alberta
2018, Atmospheric Pollution ResearchCitation Excerpt :Given the fact that chemical composition of particulate matter data is not available for most air quality monitoring stations worldwide, numerous studies have been carried out to investigate sources and origins of ambient fine particulate matter, nanoparticles, particle size distribution using real-time concentrations of gaseous pollutants data and meteorological parameters (e.g., Yue et al., 2008; Thimmaiah et al., 2009; Hellebust et al., 2010; Sun et al., 2014; Khan et al., 2015; Al-Dabbous and Kumar, 2015; Sowlat et al., 2016). PCA has been used (Cheng et al., 2009; Hellebust et al., 2010; Parsons et al., 2013; Khan et al., 2015) to identify and characterize emission sources of PM2.5, particle size distribution and atmospheric mercury using gaseous pollutants data. However, compared to PCA, PMF provides a better fit to ambient data, offers non-negative source factors, error estimates and better data treatment including handling of missing values and values below the detection limit (Paatero and Tapper, 1994).