GNGTS 2013 - Atti del 32° Convegno Nazionale

rectilinearity, the ratio between transverse and total energy, have been computed. Finally, the CFsw time series is built, as the product, at each time along the seismogram, of three squared polarization attributes multiplied by a weighting function depending on the modulus of the radial components. The reading of the first S-arrival time was finally carried out on the CFsw functions and used for the subsequent analysis of refined re-picking. The refined re-picking technique proposed by Rowe et al. (2002) has been applied to obtain highly accurate phase arrival time readings. The technique was applied to both the hand-picked first P- and S-phase arrival-times and to S-picks obtained from the polarization analysis. All waveforms from the analyzed seismic events have been preliminarily organized in common receiver gathers. For each pair of traces recorded at the same station, the waveform similarity is evaluated by using a cross-correlation function applied to a window bracketing the reference P- or S-picking. The choice of the appropriate time window length for the cross-correlation is assessed through preliminary tests, by studying the distribution of cross- correlation coefficients for variable window widths. The optimal time window corresponded to the window which allowed to have the largest number of waveform couples with a cross- correlation value greater than 0.5. This criterion leaded to the choice time window lengths of 0.5 and 1.0 seconds for the P- and S-phases, respectively. The waveform catalog is then divided in clusters, based on the similarity as assessed by the cross-correlation coefficient. The traces which are not classified according to the clustering criterion are discarded from the analysis and all the waveforms belonging to a cluster are cross-correlated again. At this step, the result is a table of time shifts that must be added or subtracted to all the waveforms belonging to the same cluster. The optimal pick adjustements are finally determined through an inversion method applied to the estimated delay times, using an iterative conjugate gradient technique (Aster and Rowe, 200; Rowe et al. , 2002). The uncertainty on refined picking measurements are therefore assigned as the standard deviation, after estimation via a Monte Carlo sampling technique (Tarantola, 2005). Fig. 1 – Epicentral map showing routine locations of the studied earthquakes. Triangles are INGV (in green) and ISNet (blue) stations. 7 GNGTS 2013 S essione 1.1