C.A.Belis, D. Pernigotti, G.Pirovano, O.Favez, J.L.Jaffrezo, J.Kuenen, H.Denier van Der Gon, M.Reizer, V.Riffault, L.Y.Alleman, M.Almeida, F.Amato, A.Angyal, G.Argyropoulos, S.Bande, I.Beslic, J.-L.Besombes, M.C.Bove, P.Brotto, G.Calori, D.Cesari, C.Colombi, D.Contini, G.De Gennaro, A.Di Gilio, E.Diapouli, I.El Haddad, H.Elbern, K.Eleftheriadis, J.Ferreira, M. Garcia Vivanco, S.Gilardoni, B.Golly, S.Hellebust, P.K.Hopkea, Y.Izadmaneshia, H.Jorqueraa, K.Krajsek, R.Kranenburg, P.Lazzeria, F.Lenartza, F.Lucarelli, K.Maciejewska, A.Manders, M.Manousakas, M.Masiola, M.Mirceaa, D.Mooibroeka, S.Navaa, D.Oliveirac, M.Paglioney, M.Pandolfi, M.Perronea, E.Petraliaa, A.Pietrodangeloa, S.Pillona, P.Pokornaa, P.Prati, D.Salameh, C.Samara, L.Sameka, D.Saraga, S.Sauvage, M.Schaap, F.Scottoa, K.Sega, G.Sioura, R.Tauler, G.Vallia, R.Vecchia, E.Venturinia, M.Vesteniusa, A.Wakedd, E.Yuberoar
In this study, the performance of two types of source apportionment models was evaluated by assessing the results provided by 40 different groups in the framework of an intercomparison organised by FAIRMODE WG3 (Forum for air quality modelling in Europe, Working Group 3). The evaluation was based on two performance indicators: z-scores and the root mean square error weighted by the reference uncertainty (RMSEu), with pre-established acceptability criteria. By involving models based on completely different and independent input data, such as receptor models (RMs) and chemical transport models (CTMs), the intercomparison provided a unique opportunity for their cross-validation. In addition, comparing the CTM chemical profiles with those measured directly at the source contributed to corroborate the consistency of the tested model results. The most commonly used RM was the US EPA- PMF version 5. RMs showed very good performance for the overall dataset (91% of z-scores accepted) while more difficulties were observed with the source contribution time series (72% of RMSEu accepted). Industrial activities proved to be the most difficult sources to be quantified by RMs, with high variability in the estimated contributions. In the CTMs, the sum of computed source contributions was lower than the measured gravimetric PM10 mass concentrations. The performance tests pointed out the differences between the two CTM approaches used for source apportionment in this study: brute force (or emission reduction impact) and tagged species methods. The sources meeting the z-score and RMSEu acceptability criteria tests were 50% and 86%, respectively. The CTM source contributions to PM10 were in the majority of cases lower than the RM averages for the corresponding source. The CTMs and RMs source contributions for the overall dataset were more comparable (83% of the z-scores accepted) than their time series (successful RMSEu in the range 25% – 34%). The comparability between CTMs and RMs varied depending on the source: traffic/exhaust and industry were the source categories with the best results in the RMSEu tests while the most critical ones were soil dust and road dust. The differences between RMs and CTMs source reconstructions confirmed the importance of cross validating the results of these two families of models.