The application of hierarchical cluster analysis and non-negative matrix factorization to European atmospheric monitoring site classification

• Published in: Atmospheric research Vol. 138; pp. 30 - 40
• Main Authors: Malley, Christopher S., Braban, Christine F., Heal, Mathew R.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.03.2014; Elsevier
• Subjects: Analysis methods; anthropogenic activities; Applied sciences; Atmospheric pollution; Atmospherics; Classification; Cluster analysis; Clustering; Clusters; data collection; EMEP monitoring sites; Exact sciences and technology; Monitoring; Non-negative matrix factorisation; Ozone; Pollution; United Kingdom; United Kingdom; Europe; British Isles; EMEP monitoring sites; Cluster analysis; Ozone; Non-negative matrix factorisation; Receptor model; Photochemical oxidants; Cluster analysis (statistics); Hierarchical classification; Air quality; Matrix factorization; Surveillance; Air pollution; Measuring network
• ISSN: 0169-8095; 1873-2895
• DOI: 10.1016/j.atmosres.2013.10.019

The effective classification of atmospheric monitoring sites within a network allows conclusions from measurements to be extrapolated beyond the confines of the site itself and applied to larger areas or populations. This is especially important for the European EMEP ‘supersites’ because these are relatively few in number yet are subject to much investment in composition monitoring capability. Here, the representativeness of the two UK EMEP supersites, Auchencorth and Harwell, was evaluated using the hierarchical cluster analysis (HCA) of all available EMEP monitoring sites based on measured ozone concentration datasets for the period 1991–2010. A novel feature was to apply non-negative matrix factorization (NMF) to order the sites within the HCA dendrograms according to the relative anthropogenic influence on ozone. The ordered dendrograms enabled UK sites to be placed more precisely in a European context. For 2007–2010, all 19 UK EMEP sites were assigned to two of the site classification clusters, with 17 of the sites grouping closely with each other in each cluster. Auchencorth clustered with the sites characterised by less modification of hemispheric background ozone levels, whilst Harwell grouped with the sites showing a more polluted regime. A similar grouping of sites occurred between 1991 and 2010, with relatively closer clustering of Polluted UK sites compared with Remote UK sites due to the larger, transboundary spatial domain for which the Remote UK sites are representative. This tight clustering of the majority of the other UK ozone monitoring sites with either one of the supersites, shows that UK background ozone conditions are well represented by Auchencorth and Harwell, and gives confidence that more extensive chemical climatologies developed for the two supersites will have wider geographical relevance. •Classification of 154 EMEP sites between 1991 and 2010 based on ozone variation•Hierarchical cluster analysis reveals four major European ozone regimes.•Non-negative matrix factorization evaluated each site's anthropogenic influence.•The two UK EMEP supersites are representative of the UK's two major ozone regimes.•Auchencorth and Harwell grouped with “Remote” and “Polluted” sites respectively.