GNGTS 2014 - Atti del 33° Convegno Nazionale

DWS, generally correspond to low level of spread function. There is scatter on the trend because the averaging vector depends on the geometry of the rays in its surrounding and not just on the density of rays. For this reason, estimates of the resolution, based on DWS only, supply only qualitative results (Toomey and Foulger, 1989). Toomey and Foulger (1989) analyzed several plots depicting the elements of the averaging vector in the three- dimensional space of the studied volume. This allows to estimate a good threshold value for the spread function under which we consider the resolution acceptable. We followed Toomey and Foulger (1989) approach and estimated a threshold value of 7 for our spread function. The layers at 4, 6 and 8 km depth, characterized by wide central zones with spread function value less or equal 7, are the most reliable ones. Shallower and deeper layers are unevenly illuminated by the seismic ray paths; there are only smaller areas where the smearing effect is not prevalent: at 0 and 2 km only the southernmost zone and below 8 km the westernmost one. b) Checkerboard sensitivity test (Zelt, 1998). It is based on the capability to reconstruct the Q S field on synthetic data characterized by the same distribution of earthquakes and recording stations. The checkerboard test consists in alternating adjacent nodes of high and low anomalies on a uniform Q m field in evaluating the capability to reconstruct the field. In particular, we chose two different Q S values Q 1 and Q 2 , where Q 1 = Q m + a and Q 2 = Q m - a , Q m is the mean of all the Q S values and a is a parameter. The synthetic k were estimated keeping the hypocenters fixed. We performed the checkerboard test with Q m =200 and a = 80, obtaining in such a way a checkerboard with variations of ±40%. We repeated the test adding to synthetic k a random Gaussian noise with sigma equal to 2.5% of the data value, compatible with the errors in k value estimation. In the layers deeper than 2 km the checkerboard test conformity with the 7 spread function isoline is clear. c) Shurr et al. (2003) and De Siena et al. (2010) approach: starting from the same distribution of earthquakes and recording stations of real data, a synthetic model with anomalies comparable in size and amplitude with those seen in the inversion results, but different geometry has been developed, checking the capability of the method to reconstruct the output size and intensity of anomalies. This test allows to explore the true resolving capability in zones of maximum interest. In this test, the amplitudes and anomaly shapes are recovered well in areas of high ray coverage and poorly when ray coverages decreases. We simulated a model with the same values of Q S of the output of the genetic algorithm, on which we superimposed two high Q S anomalies in the western Fig. 2 – Q S synthetic anomaly test: (a) synthetic model for two cross sections (b) recovered model for sections in (a) after adding a random Gaussian noise with sigma equal to 2.5% of the data value. 148 GNGTS 2014 S essione 1.2