GNGTS 2018 - 37° Convegno Nazionale

GNGTS 2018 S essione 3.2 623 185 blows are recorded in just 2 days of field work resulting in a profile length of approximately 180 m. Sampling interval is 0.5 ms and record length 1 s. A more detailed description of the acquisition layout is given in (Tognarelli and Stucchi, 2016). Processing sequence. The processing sequence applied is aimed at obtaining the best image of the shallow subsurface reflections in terms of signal-to-noise ratio (S/N) without bringing in unwanted artifacts. To this goal, we first spatially filtered the raw field data in the laboratory by Chebyshev weighted arrays to reduce the surface waves and air blast effects (Tognarelli and Stucchi, 2016). Successively, each step in the sequence is designed to increase the S/N ratio of the reflection events, carefully checking the output to ensure that no artifacts are introduced. In short, the processing procedure includes: band-pass filtering, refraction statics corrections, velocity analysis, deconvolution, residual statics computation, f-x deconvolution and stack. Fig. 1 shows the stack section obtained at the end of the described sequence. Fig. 1 - Stack section obtained at the end of the time processing sequence. FWI inversion. Due to the oscillating nature of the seismic wave propagation, the classical L 2 distance between predicted and observed data is characterized by many local minima that hinder the convergence of the inversion towards the true model corresponding to the global minimum. One approach that can be followed to reduce the non-linearity of the FWI and that allows a more robust inversion, is the application of some processing steps to the data, such as filtering and muting, that also allow to select the phases on the seismogram to invert (Tognarelli et al. , 2015). This is an important issue in the near-surface exploration because the source related noise contamination (ground-roll), together with the limited recorded offset, reduce the portion of the data that can be used in an acoustic FWI. We select 20 shot gathers along the profile, approximately one shot every 10 m, and for each record we design appropriate mute functions that include only the first arrivals. Trace envelope to recover some low frequencies, trace normalization and low-pass filtering (0-15 Hz) are then applied to the first arrivals of the observed and predicted data before computing the L 2 distance. Some attempts to infer the source wavelet from the real data are carried out. However, the use of the envelope in the inversion allows to relax the requirement of an accurate source wavelet, whose estimation is a very difficult task especially for near-surface seismic investigation. Figure 2. shows an example of a raw shot gather (Fig. 2a) and the corresponding observed (Fig. 2b) and predicted (Fig. 2c) data used to compute the misfit function.

RkJQdWJsaXNoZXIy MjQ4NzI=