Resource information
The observation sites that make up air quality monitoring networks can have very different characteristics (topography, climatology, distance to emission sources, etc), which are partially described in the meta-information provided with data sets. At the scale of Europe, the description of the sites depends on the institute(s) in charge of the air quality monitoring in each country, and is based on specific criteria that can be sometimes rather subjective. The purpose of this study is to build an objective, homogeneous, and pollutant-specific classification of European air quality monitoring sites, primarily for the purpose of model verification and chemical data assimilation. Most studies that tackled this issue so far were based on limited data sets, and often took into account additional external data such as population density, emission estimates, or land cover maps. The present study demonstrates the feasibility of a classification only based on the past time series of measured pollutants. The underlying idea is that the true fingerprint of a given monitoring site lies within its past observation values. On each site to be categorized, eight indicators are defined to characterize each pollutant time series (O₃, NO₂, NO, SO₂, or PM₁₀) of the European AirBase and the French BDQA (Base de Données de Qualité de l’Air) reference sets of validated data over the period 2002–2009. A Linear Discriminant Analysis is used to best discriminate the rural and urban sites. After projection on the Fisher axis, ten classes are finally determined on the basis of fixed thresholds, for each molecule. The method is validated by cross-validation and by direct comparison with the existing meta-data. The link between the classes obtained and the meta-data is strongest with NO, NO₂, and PM₁₀. Across Europe, the classification exhibits interesting large-scale features: some contrasts between different regions depend on the pollutant considered. Comparing the classes obtained for different pollutants at the same site reveals an interesting consistency between the separate classifications. The robustness of the method is finally assessed by comparing the classifications obtained for two distinct subsets of years. The robustness – and thus the skill of the objective classification – is satisfying for all of the species, and is highest with NO and NO₂.
