Source apportionment of particulate matter in Europe: A review of methods and results

Abstract

European publications dealing with source apportionment (SA) of atmospheric particulate matter (PM) between 1987 and 2007 were reviewed in the present work, with a focus on methods and results. The main goal of this meta-analysis was to provide a review of the most commonly used SA methods in Europe, their comparability and results, and to evaluate current trends and identify possible gaps of the methods and future research directions. Our analysis showed that studies throughout Europe agree on the identification of four main source types (${\mathrm{PM}}_{10}$ and ${\mathrm{PM}}_{2.5}$): a vehicular source (traced by carbon/Fe/Ba/Zn/Cu), a crustal source (Al/Si/Ca/Fe), a sea-salt source (Na/Cl/Mg), and a mixed industrial/fuel-oil combustion ($\mathrm{V}/\mathrm{Ni}/{\mathrm{SO}}_4^{2-}$) and a secondary aerosol $({\mathrm{SO}}_4^{2-}/{\mathrm{NO}}_3^{-}/{\mathrm{NH}}_4^{+})$ source (the latter two probably representing the same source type). Their contributions to bulk PM levels varied widely at different monitoring sites, and showed clear spatial patterns in the cases of the crustal and sea-salt sources. Other specific sources such as biomass combustion or shipping emissions were rarely identified, even though they may contribute significantly to PM levels in specific locations.

Introduction

In the field of atmospheric sciences, source apportionment (SA) models aim to re-construct the impacts of emissions from different sources of atmospheric pollutants, e.g., particulate matter (PM), based on ambient data registered at monitoring sites (Bruinen de Bruin, Koistinen, Yli-Tuomi, Kephalopoulos, & Jantunen, 2006, Hopke and Song, 1997, Watson et al., 2002). There are three main groups of SA techniques:

• Methods based on the evaluation of monitoring data. Basic numerical data treatment is used to identify sources. Examples are: (1) correlation of wind direction with levels of measured components to identify source locations (Henry, Chang, & Spiegelman, 2002); (2) the correlation of gaseous pollutants with PM components to identify source associations; (3) subtraction of levels measured at regional background from those obtained at urban background and/or roadside levels to identify the contributions from the regional background, the city background and the monitored street (Lenschow et al., 2001), or (4) quantification of natural PM contributions (e.g., African dust) by subtracting PM levels at regional background sites from those at urban background locations for specific days (Escudero et al., 2007). The main advantage is the simplicity of the methods and the consequent low impact of mathematical artefacts due to data treatment.

• Methods based on emission inventories and/or dispersion models to simulate aerosol emission, formation, transport and deposition (Eldering and Cass, 1996, Visser et al., 2001). These models require detailed emission inventories that are not always available, and they are limited by the accuracy of emission inventories, especially when natural emissions are important. A significant advantage of these methods is that they may be used in scenario studies to evaluate the impact of emission abatement strategies on the anthropogenic contribution to ambient PM concentrations.

• Methods based on the statistical evaluation of PM chemical data acquired at receptor sites (receptor models). The fundamental principle of receptor modelling is that mass and species conservation can be assumed and a mass balance analysis can be used to identify and apportion sources of airborne PM in the atmosphere (Hopke et al., 2006). An overview of the wide range of statistical models and modelling approaches which are currently available in the literature is shown in Fig. 1. As shown in the graph, one of the main differences between models is the degree of knowledge required about the pollution sources prior to the application of receptor models. The two main extremes of receptor models are chemical mass balance (CMB) and multivariate models.

The CMB model assumes knowledge of the composition of the emissions for all relevant sources. If changes of the source profiles between the emitter and the receptor may be considered minimal, CMB can be regarded as the ideal receptor model. However, these requirements are almost never completely fulfilled, and thus, pure CMB approaches are often problematic. One important characteristic of CMB is that secondary aerosols must be included not as components of emission source profiles but as specific, single chemical compounds. This absence of mixture with other tracer elements is often regarded as a limitation, and may lead to misinterpretation of results.

Principal component/factor analysis (e.g., principal component analysis or PCA, positive matrix factorisation or PMF, UNMIX) attempts to apportion the sources on the basis of observations (internal correlations) at the receptor site alone. These are commonly used tools, because software to perform this type of analysis is widely available and detailed prior knowledge of the sources and source profiles is not required. The choice of the model dimension and the search for non-negative solutions by axis rotations can be based entirely on mathematical criteria. Nevertheless, it has been suggested that factor analysis attempts to get more information out of atmospheric data than is really there (Henry, 1987). Furthermore, it is a common problem that the resulting components or factors may represent mixtures of emission sources, as opposed to clearly independent source profiles. Source signatures that change with time are a limitation for this and other types of receptor models.

To combine the advantages and reduce the disadvantages of CMB and factor analysis hybrid models have been developed. Examples are confirmatory or target transformation factor analysis, which offer some control of the solutions by “fixing” or “freeing” specific parameters, set according to the theoretical expectation of the researcher (Gleser, 1997, Hopke, 1988). With a constrained physical receptor model (COPREM, Wåhlin, 2003), an initial profile matrix with the main characteristics of known sources as columns is used, and a priori knowledge about the character of the sources can be used to achieve a solution with a sufficient number of sources. The multilinear engine (ME, Paatero, 1999) can solve multilinear problems with the possibility of implementing many kinds of constraints using a script language. Like COPREM, the program allows choosing hybrid versions in the full range between PMF and CMB-type models, with the difference that individual data points can be properly weighed (not possible in any eigenvector analysis).

The main objective of receptor models is, therefore, to identify the possible sources of PM (if not assumed already from the source profiles) and to obtain data on their contributions to the bulk PM mass. Even human exposure to these pollution components has been evaluated to assess their health effects and risks (Hopke et al., 2006, Ilacqua et al., 2007, Watson et al., 2002). Furthermore, policy-makers require sound scientific knowledge of the PM sources and their contributions to atmospheric PM levels and associated health risks for the development and implementation of policies to protect human health and the environment. Thus the information provided by receptor models is key to the design of effective mitigation strategies on the local- and meso-scale.

Despite the widespread need for these data, there is little information available on receptor modelling results from different European countries, the type of models applied or the input data utilised. Furthermore, parameters such as time resolution, type of monitoring site and PM sampler, and analytical methods vary widely from one study to another. Consequently, general European-wide conclusions cannot yet be extracted. In addition to this lack of harmonisation in the published studies, a large amount of valuable material is currently available in the grey literature only.

The need to tackle this issue was detected by the COST Action 633 (“Particulate matter: Properties related to health effects”) and subsequently addressed by compiling a database of SA studies available in Europe, more specifically in COST633 member countries (http://www2.dmu.dk/atmosphericenvironment/COST633/). Both meta-data on publications and data on PM sources and source contributions were collected, with the aim to obtain an overview of the results and analyse their comparability. The issue of the comparability between receptor modelling results is considered essential if SA data are to be used for the design of multinational mitigation strategies, the assessment of human exposure to PM from specific sources, and investigations of long-term changes in exposure situations over Europe.