European intercomparison for receptor models : preliminary results

Receptor models are commonly used to identify the sources of ambient particulate matter (PM) in Europe. However, the use of different tools and methodological approaches make it difficult to compare the results of different studies. In order to promote harmonization in this field an intercomparison exercise involving 16 expert groups was organized and evaluated by the JRC within the framework of FAIRMODE. The test DB consisted of 178 PM2.5 speciated samples deriving from two real-world re-arranged DBs (Lee & Hopke, 2006 JGR; Jaeckels et al., 2007 EST). Concentrations and uncertainties of 46 inorganic and organic species were used to characterize the PM. A total of 22 solutions obtained using the following models or model versions were reported for evaluation: EPA-PMF3 (8), PMF2 (6), EPA-CMB 8.2 (4), PCA (1), APCS (1), COPREM (1) and ME-2 (1). Although data in the DB were partially pre-processed, participants had to scrutinize the DB in order to identify, solve and report typical imperfections of real world DBs (e.g. missing values, values below detection limits, outliers and unusual uncertainty patterns). Participants reported the number and label of the identified sources, their contribution estimation (SCE) and uncertainty. In addition, the source/factor chemical profiles and the contribution of each source/factor in each sample were also requested. The match between observed and modelled PM mass in every sample was assessed using regression analysis, centered RMSE and bias. More than 75% of the solutions reconstructed the PM mass satisfactorily. The majority of participants reported between 7 and 11 source/factors. They were classified into 15 different source categories: biomass burning, gasoline, diesel, brakes, traffic, dust, sulphates, nitrates, Zn smelter, Cu metallurgy, Pb smelter, steel processing, industry & combustion, ship emissions, secondary. The inclusion of every source/factor into a category was checked by comparing its chemical profile and time trend with all the other members of the same category and with reference source profiles, when available. The SCEs of the different solutions were compared with a reference value obtained by robust analysis (standard ISO 5725-5). The acceptability criterion was set to 50% relative standard uncertainty. For more details see Karagulian & Belis (2012, IJEP accepted for publication). More than 90% of the 160 assessed source/factor contribution estimations met the acceptability criterion. Although no absolute reference values can be used in real-world DBs, a reasonable quantitative agreement in the SCEs is pointed out by these results. Indeed, the noise of experimental data, the variety of methodological approaches and the fact that the sampling site was completely unknown to participants (particularly crucial for those using CMB) shall be taken into account when interpreting the intercomparison outcome.