Results of the European Intercomparison exercise for Receptor Models 2011 - 2012. Part I

The quantification of pollution sources contributions to ambient atmospheric pollutants is a key element for the development of any effective air quality management policy. Source apportionment is explicitly or implicitly needed for the implementation of the Directives on Air Quality (Directive 2008/50/EC and 2004/107/EC, hereon AQD). Pollution source information is required, for instance in: identifying exceedances due to natural sources or to road salting and sanding, preparing air quality plans, quantifying transboundary pollution, and in demonstrating eligibility for postponement of PM10 and NO2 limit value attainment (COM/2008/403). In order to have a better understanding about the comparability and performance of different source apportionment methodologies an intercomparison exercise (IE) has been organized by the European Commission's Joint Research Centre (JRC). This IE is part of a JRC initiative for the harmonization of source apportionment with receptor models that has been launched in collaboration with the European networks in the field of air quality FAIRMODE (modelling) and AQUILA (measurements). It was organized to fill a gap in the knowledge about the quantitative assessment of source apportion model performances. Facing such a challenging task was possible thanks to the collaboration of many European experts in the field that accepted to participate. The intercomparison exercise (IE) main objective was to assess whether the estimations of source contributions in terms of mass (ng/m3) compared with a reference value are consistent with a quality standard expressed as maximum accepted uncertainty. A database was distributed to the participants, including information on air pollutant concentration, their uncertainties and the emission inventory information. Due to the lack of a specific methodology to assess receptor model performances in IEs, the organizers developed a battery of tests, partially based on existing international standards, and defined quality criteria (more details in Karagulian & Belis, 2012). In the overall evaluation were also considered a) the ability of models to reconstruct the measured PM mass, and b) the capacity of models to identify the number of sources. These two tests are, however, to be considered a complement of the main performance test. The test to assess models' performance was divided in two stages: a) a preliminary stage aiming at assessing the similarity of the factor/source profiles reported by participants, mainly based on their fingerprints and their uncertainties, and b) a second stage targeted at evaluating whether the bias in the quantification of the solutions is consistent with the established quality standards. The preliminary test was passed by a 90% of the tested factor/sources. APCS and COPREM were the models with the highest rate of rejected profiles (44% and 33% respectively). Of the 167 scores (z-scores) calculated in the final performance test, 144 (86%) complied with the 50% standard uncertainty quality criterion. Only 7% of the factor/source profiles were rated as unsatisfactory while 6% were ranked as questionable. Concerning the subordinate tests, the majority of the solutions reproduced the PM mass in an acceptable manner, however a number of solutions presented either an overestimation or an underestimation. The average number of factors/sources identified by participants was 9. Nevertheless, this value varied considerably between solutions. The CMB type models presented an average of 8.3 sources per solution while the factor analysis type models average was 9.2 factors per solution. These values are in good agreement with the 10 sources identified in a previous study on the same database (Lee et al., 2006). As a whole the IE results indicate a good general agreement between the performances of the different participants and models. Participants demonstrated good skills in dealing with complex real-world data. The next step of the IE consists in the use of a synthetic database containing known source contributions for the evaluation of the solutions.