A Data-Driven Technique for Building the Initial Velocity Model for Refraction Tomography

Abstract

Near-surface characterization is an important part of seismic data processing, especially with land seismic data. The conventional approaches rely on refracted waves and estimate the compressional velocity models from the tomography of the first-break traveltimes (Glushchenko et al., 2012, Speziali et al., 2014). Despite its strong ability to image the subsurface, seismic tomography is a non-unique inverse problem (Kanlı, 2009, Mantovani et al. 2013). Because most inverse geophysical problems are non-unique, each problem must be studied to determine what type of non-uniqueness applies and, thus, determine what type of a-priori information is necessary to find a realistic solution (Ivanov et al., 2005). There are several ways to incorporate the available a-priori information in the inverse problem; one of them is the definition of the initial model, which is the starting point of the inversion process. In this work, we present a data-driven approach that derives the initial velocity model for a refraction tomography workflow in an automated fashion, thus trying to minimize the amount of subjectivity that influences the starting model definition (Osypov, 2001). We demonstrate the technique by mean of a synthetic, but realistic, 3D example.