Source attribution in IEHIAS
- The text on this page is taken from an equivalent page of the IEHIAS-project.
Source attribution is the attempt to estimate the contribution of each major source to observed exposures to (or concentrations or doses of) pollutants. This is usually done by relating measurements of the pollutant from a sample of individuals or locations to putative sources, either by deductive methods or using statistical techniques. In contrast to exposure assessment, therefore, it tries to trace pollutants back through the causal chain, rather than to model their fate from point of release through to exposure and dose.
Source attribution provides a means of identifying where pollutants or other hazards have come from, and the proportion that has come from each source. Although this information is not always essential for health impact or risk assessment, it can be crucial for risk management. Most policy options control risks by reducing the exposures, and this is best achieved by restricting some or all of the sources. Source attribution thus allows actions to be targetted where they are likely to have most effect. It has been especially used in recent years to analyse and manage sources of air pollution (notably particulate matter).
Source attribution can, in principle, be undertaken for almost any pollutant (or other hazard). Attribution is likely to be most effective and reliable, however, where the pollutant has clearly definable sources, where the transport and transformation processes between source and measurement point are relatively simple and short, and where relevant information on the sources is available.
In simple situations, it can be successfully applied to a single measurement point and a single source. Where multiple sources and pathways occur, however, source attribution requires a relatively large sample of measurements, as well as detailed information on the characteristics of the sources. Where multiple pollutants are involved, this should include quantitative data on the composition of emissions and the pollutant mix at the measurement locations.
Inputs required to undertake source apportionment depend on the situation and the range of pollutants and sources of interest. In general, the more information available, the more reliable source apportionment can be, and the more sources and pathways that can be detected.
The primary starting point for any analysis is measurements of one or more pollutants at a number of points (in time or space). Measurements may relate to environmental concentrations, exposures or doses. Sufficient data are needed to identify spatial or temporal differences in the levels and/or composition of the pollutants.
In most cases, information is also required on characteristics of potential sources. This may include data on the location and intensity of possible sources, emission rates and/or the composition of emissions.
In addition, it is often helpful to have data at intermediate points in the source-exposure chain (e.g. in different micro-environments) in order to provide a check or control on inferences.
Outputs from source apportionment also vary depending on the scope and complexity of the analysis. Minimal outputs comprise identification of the major contributor(s) to the pollutants at the measurement locations. With more information, these contributions can be quantified (e.g. in terms of proportion). The most advanced analyses enable pathways and flows of pollutants from source to point of measurement to be reconstructed.
Source attribution is based on the principle that measured exposures to a pollutant or other hazard (or measured doses or environmental concentrations) are the result of a logical and definable process of release, transport and transformation. It further assumes that these processes leave a detectable signature in the measurements - for example, in terms of the spatial distribution, temporal patterns or pollutant mix. If information is available on the characteristics of possible sources, and if sufficient knowledge exists about the intervening processes, then it should be possible to deduce the origins of the pollutants and estimate the contributions from each source.
To do this with any degree of confidence requires that:
the emission profile of different sources is sufficiently distinct to enable them to be identified; that the transformations that occur after emission are known, or sufficiently consistent, so that they do not mask the imprint of the sources. Both these assumptions need to be carefully checked and evaluated before any analysis.
Several different methods can be used to attribute exposures to their sources. These can usefully be classified as logical techniques, regression analyses and formal source apportionment. The last of these can involve either statistical (empirical) methods or mechanistic (deterministic) methods.
Logical techniques use contrasts between measurements of the pollutant of concern in source-impacted (exposed) and non source-impacted (non-exposed) situations (e.g. locations, individuals, populations) to evaluate the influence of a source.
Measurements of ambient concentrations upstream of an industrial plant, for example, may show significantly higher concentrations than those at downstream locations, implying that the industry is a major source. Differences in concentrations of a contaminant in the presence of smoking, compared to its absence, would indicate tobacco smoke as a source. Logical deduction can similarly be used to interpret more complex situations.
In all these cases, calculation of the ratio between exposed and non-exposed samples provides a measure of the contribution of the specific source relative to background or reference levels. Logical techniques, however, can usually be applied only to assess the contribution of a single specified source on a specific pollutant.
For the attribution of a pollutant to multiple concurrent sources, statistical methods need to be applied. One of the most widely used approaches is regression analysis. This is done by relating measured levels of the pollutant (as the dependent variable) at a sample of locations or times to one or more indicators (as independent variables), selected to characterise the likely sources. The latter can include concentrations in upstream micro-environments and/or media, emissions or measures of source intensity or distribution.
The resulting models comprise regression equations which should be tested for model and coefficient significance, using conventional statistical methods. Sigificant variables in the equations can be interpreted evidence of a contribution from that source; the standardised slope coefficients of the various independent variables provides an indication of their relative contributions. For example, in a study of personal PM and soot exposures among non-ETS exposed elderly subjects with cardiovascular disease, regression analysis pointed to outdoor levels as the major determinant of personal and indoor levels. In addition, living near a busy road, time spent in traffic and cooking were all associated with increased exposures (Lanki et al. 2007).
When regression-based source apportionment is done using spatial data, the approach has similarities with the land use regression (LUR) methods often used for exposure assessment. However, regression analysis can also be done to analyse temporal data - e.g. to relate measured concentrations at a single point to time-varying activities or emissions of possible sources.
In both cases, regression models developed for the purpose of source apportionment can be used predictively, to show how concentrations (or exposures or doses) might change if source intensity or emissions are changed. In a study on traffic related air pollution, for example, regression modelling was used to relate annual average concentrations at a limited number of sites to traffic related variables (e.g. population density and traffic density). Results of the modelling were then used to estimate air concentrations at the home addresses of participants in a large cohort study on the risks of development of childhood asthma and other allergic disease (Brauer et al. 2003).
Formal apportionment techniques
Relatively straightforward techniques such as logical or regression analysis are not suitable for complex mixtures of exposures, involving multiple pollutants and multiple sources. In that case, more advanced techniques of source apportionment are needed.
A wide range of statistical techniques can be used for this purpose, including enrichment factors (EFs) , chemical mass balance (CMB), principal component analysis (PCA), factor analysis (FA), empirical orthogonal functions (EOF), multiple linear regression, neural networks, edge detection, cluster analysis and Fourier transformation time series analysis. For more details see the Watson et al. (2002). Some of these methods (e.g. neural networks, edge detection, and chemical evolution receptor models) are still at the development stage and are of greater interest to researchers than to those attempting to solve (air) pollution problems.
These various modelling techniques take two general forms: statistical (empirical) and deterministic (mechanistic) techniques. Statistical source apportionment methods (such as PCA) do not require a priori knowledge of the source emissions, but they can be applied only for a fairly large set of samples, not for a single sample. Deterministic techniques (such as CMB) can be applied to one sample — i.e. specifically and separately for one study subject or location — but to do so requires (usually quite demanding) prior analyses of the source emissions.
Brauer, M., Hoek, G., van Vliet, P., Meliefste, K., Fischer, P., Gehring, U., Heinrich, J., Cyrys, J., Bellander, T., Lewne, M., Brunekreef, B. 2003 Estimating long-term average particulate air pollution concentrations: application of traffic indicators and geographic information systems. Epidemiology 14, 228-239.
Bruinen de Bruin, Y., Koistinen, K., Yli-Tuomi, T., Kephalopoulos, S., Jantunen, M., 2006. Source apportionment techniques and marker substances available for identification of personal exposure, indoor and outdoor sources of chemicals. EC JRC: EUR 22349 EN.
Hildemann, L.M. 2002 Introduction to source apportionment methods. Air Toxics Exposure Workshop, San Francisco, California, June 25-27, 2002.
Hopke P.K (ed.) 1991 Receptor modelling for air quality management. Data handling in science and technology, Vol 7. Amsterdam: Elsevier
Hopke, P.K. 2003 Historical efforts at source apportionment. Health Effects Institute, Communication 10, Improving estimates of diesel and other emissions for epidemiologic studies. Proceedings of an HEI Workshop, Baltimore, Maryland, 2003.
Hopke, P.K., Ito, K., Mar, T., Christensen, W. F., Eatough, D.J., Henry, R.C., Kim, E., Laden, F., Lall, R.,Larson, T., Liu, H., Neas, L., Pinto, J., Stölzel, M., Suh, H., Paatero, P. and Thurston, G.D. 2006 PM source apportionment and health effects: 1. Intercomparison of source apportionment results. Journal of Exposure Science and Environmental Epidemiology 16, 275-286.
Lanki, T., Ahokas, A., Alm, S., Janssen, N.A.H., Hoek, G., De Hartog, J.J., Brunekreef, B, Pekkanen, J. 2007 Determinants of personal and indoor PM2.5 and absorbance among elderly subjects with coronary heart disease. Journal of Exposure Science and Environmental Epidemiology 17, 124–133.
Watson, J.G., Zhu, T., Chow, J.C., Engelbrecht, J., Fujita, E.M., Wilsone, W.E. 2002 Receptor modelling application framework for particle source apportionment. Chemosphere 49, 1093-1136.
- Positive Matrix Factorization
- Principal Components Analysis
- Chemical Mass Balance
- Source attribution database