Then, the cluster analysis package included in the 4.9 version of HYSPLIT characterized the ensemble of 481 calculated four-day BTs terminating at a height of 750 m a.g.l. and recorded every hour during the study period. The 4-day BTs were utilized as input variables for the K-means clustering algorithm and the distances between trajectories were computed using simple Euclidean distance. Next, a cluster analysis algorithm was used to categorize the computed trajectories into groups of similar properties, the so-called clusters. The average BT of each cluster was then calculated from its trajectory members. Since accuracy in BT calculation decreases with distance and time (due to model assumptions and spatial and temporal Lapatinib of the meteorological data), 4-day trajectories were considered the most suitable option. Moreover, the clustering of BTs reduced errors associated with single trajectories. Finally, in order to confirm the long range transport contribution to PM locally measured and to localize the source emission, two statistical models were applied to 4-day-backtrajectories at 750 m.a.s.l. starting point: Potential Source Contribution Function (PSCF) and Concentration Weighted Trajectory (CWT). The first one depends on the frequency of passages over each grid cell for trajectories associated to concentrations above a specified quantile (90th for this study) therefore it calculates the probability that a source is located at latitude i and longitude j. The CWT computes a logarithmic mean of concentration at the receptor weighted by the residence time of the trajectory for each grid cell of the geographical domain ( Cheng et al., 2013, Fleming et al., 2012, Carslaw and Ropkins, 2012 and Wang et al., 2012). These statistical data analyses have been performed within the software environment R 3.0.2 ( R Development Core Team, 2013).