GNGTS 2014 - Atti del 33° Convegno Nazionale

To invert the k data Gentili and Gentile (2013) used the same grid and the same velocity values obtained by Bressan et al. (2012) in 3D tomographic inversion of velocity in the area (see Fig. 1). Fig. 1 also shows the earthquakes used in Gentili and Gentile (2013) work. �A cluster of seismicity can be easily detected in the northeastern part in correspondence with the two Kobarid 1998 and 2004 seismic sequences. Fig. 1 – Grid of the 3D Q S tomographic inversion. Diamonds: location of seismic stations used for the inversion. Circles: earthquakes used for Q S inversion. In order to obtain �������� �������� ����� ��� ��� �������� ����������� ����� �� �������� adequate starting model for the inverted parameters, since no previous knowledge for a valuation of the Q s field in the Friuli area was available, a genetic algorithm technique (Goldberg, 1989), the David L. Carroll GAFORTRAN code with MICRO-GA enabled (Carroll, 2001) was adopted, using a micropopulation of five individuals adopting as cost function the same SIMULPS2000 code minimizing the total weighted RMS. Reliability of our data. We must emphasize that this is just a first attempt to an attenuation tomography study in this area. Only 156 events could not be enough for a good inversion and, besides, large part of them are aftershocks of the magnitude 5.6April 12, 1998 Bovec earthquake, this causes a fairly strong non homogenous rays illumination. For these reasons a good estimate of the reliability of the results becomes essential for a acceptable geophysical interpretation. There are numerous tests that have been developed to assess the ray coverage. For understanding how our results approach a good reliability we used the following three tests: a) The model resolution matrix. In a non-uniform and then not evenly illuminated field, a single node much more resolved than his neighbors can lead to a smearing effect: his value conditions the values of the adjacent nodes. Each row of the resolution matrix is the averaging vector for the single parameter, describing its the dependence on all the others. The ideal resolution matrix should be the identity matrix, when the model parameters are perfectly resolved by the inversion.Agood resolved node has a high value in the averaging vector only near the node itself and close to zero elsewhere (compact averaging vector). Even if plotting each node of the grid can give an idea of the quality of the inversion, this choice is impractical when the grid contains hundreds of nodes. A rough estimate of the quality of a tomography corresponding to average relative measure of the density of seismic rays near a given each node is given by the derivatives weight sum (DWS). Michelini and McEvilly (1991) introduced the spread function of a node of the grid calculated from the resolution matrix. The well sampled nodes, characterized by high GNGTS 2014 S essione 1.2 147