GNGTS 2018 - 37° Convegno Nazionale

624 GNGTS 2018 S essione 3.2 The inversion was carried out using the genetic algorithms as the optimization method and using the two-grids strategy as described in Sajeva et al. 2016. The fine grid covers a model of 188 m in length and 260 m in depth and has a node spacing of dx=dz=2 m. The coarse grid consists of 20×3 nodes with a vertical spacing of 13 m and a horizontal spacing of 63 m, due to the limited horizontal velocity variation expected in the investigated area. The unknowns are the velocity values at the coarse grid nodes. The starting velocity model was derived from the smoothed and depth converted model (Dix equation) computed by the velocity analysis step. The parameters used for the genetic algorithms are chosen to allow a good exploration of the model space (the number of individuals in each population is 15 times the number of unknowns) and a fast rate of convergence (selection rate 80%). To reduce the possibility to be entrapped in a local minimum, 2 sub-populations are used along with 1.67% of mutation rate. The velocity model obtained after the GA inversion is illustrated in Fig. 3a., as can be expected, it shows mainly a flat layer stratification that well corresponds to the stack image shown in Fig. 1, with some minor lateral velocity variations. No wells are drilled, so to verify Fig. 2 - a) Raw shot gather after the array filtering; b) observed data for the shot in a); c) predicted data for the same shot. Fig. 3 - a) Velocity model estimated by the GA; b) CIGs after pre-stack depth migration.