Structure-guided 3D anisotropic tomography: a case study Jessé Carvalho Costa (UFPA), Eduardo Filpo* (PETROBRAS), Luiz Alberto Santos (PETROBRAS) and Etory Feller Sperandio (PETROBRAS) Copyright 2019, SBGf - Sociedade Brasileira de Geofísica This paper was prepared for presentation during the 16 th International Congress of the Geophysical Society held in Rio de Janeiro, Brazil, 19 to 22 August 2019. Contents of this paper were reviewed by the Technical Committee of the 16 th International Congress of the Brazilian Geophysical Society ˙ Ideas and concepts of the text are authors’ responsibility and do not necessarily represent any position of the SBGf, its officers or members. Electronic reproduction or storage of any part of this paper for commercial purposes without the written consent of the Brazilian Geophysical Society is prohibited. Abstract Macro-velocity model building remains a challenge for seismic imaging. Currently, seismic datasets with large offsets and multi-azimuth usually require anisotropic models to produce a consistent 3D image of subsurface. The larger number of parameters to represent anisotropic models increases the ambiguity in 3D tomography. We present a successful case study illustrating the importance of using carefully designed conditioning of common image point gathers (CIP), dense RMO events picking and interactive QC tools, combined with 3D anisotropic structure guided tomography in order to reduce ambiguity and estimate a geologically conformable velocity model. Introduction Imaging seismic data with large offsets and multi-azimuths accurately demands 3D anisotropic tomography. The methodology for tomography in anisotropic media is well established in the literature (Barbosa et al., 2008; Bakulin et al., 2010; Zhou et al., 2011; Wang and Tsvankin, 2013), however, its implementation is not trivial. A carefully designed 3D tomography workflow still is an important and robust processing tool for 3D seismic imaging. We present a case study where two implementations of tomography produced noticeable differences in the velocity model and the resulting 3D seismic image. The baseline velocity model for our study is VTI and was produced by a service company after five iterations of 3D anisotropic tomography. The interpreter was not satisfied with the results and suggested further investigation for reducing the RMO through isotropic iterations. In order to evaluate the interpreter conjecture we used an in house implementation of 3D anisotropic structure-oriented tomography. The main features of our tomography implementation consist of a raytracing implementation for arbitrary Hamiltonians, model representation using nonuniform B-splines of variable order, structure-guided preconditioning (Hale, 2009) and multi-scale iterations strategy. Also critically important was a robust implementation of dense event picking and interactive QC. Using this approach we validated the interpreter conjecture with a velocity model that reduced the RMO in all CIP compared with the baseline result for the whole offset range. Consequently, the resulted image is better focused. Surprisingly, the isotropic velocity model, structurally conformable, was more consistent with the dataset than the baseline anisotropic model. Methodology The anisotropic 3D tomography follows closely the methodology for raytracing and computation of Frechét derivatives presented by Barbosa et al. (2008). Given an Hamiltonian H (x, p; C)= 0, defined by an eikonal equation for each position x and slowness vector p, for a set of material parameters represented by C. For example, to model qP TTI anisotropy we use the anelliptical approximation (Schoenberg and de Hoop, 2000): H (x, p)= 1 2  C ε (x)(p · p − (p · ν (x)) 2 )+ C 0 (x)(p · ν (x)) 2 − C 0 (x)( C ε (x) − C δ (x)(p · p − (p · ν (x)) 2 )(p · ν (x)) 2 − 1]= 0 , (1) where C 0 represents the square of propagation velocity along the symmetry axis, C ε ≡ C 0 (1 + 2ε ) and C δ ≡ C 0 (1 + 2δ ); ε and δ are the Thomsen’s parameters (Thomsen, 1986); ν (x) indicates the direction of the symmetry axis. The residual moveout at CIP gathers relative to a reference offset h 0 , i.e., δ z rmo ≡ z(h) − z(h 0 ), can be modeled by the fundamental equation of tomography in the migrated domain: z(h) − z(h 0 )= − 1 ‖p s (h)+ p r (h)‖ n · e z   T s (h) 0 ∂ H ∂ C δ Cdτ +  T r (h) 0 ∂ H ∂ C δ Cdτ  + 1 ‖p s (h 0 )+ p r (h 0 )‖ n · e z   T s (h 0 ) 0 ∂ H ∂ C δ Cdτ +  T r (h 0 ) 0 ∂ H ∂ C δ Cdτ  . (2) Equation (2) determines a linear relationship between the event residual moveout and the perturbations of velocity model parameters δ C around a current reference model C. In order to build a linear system using this equation we need to compute pairs of specular rays from each picked event to the surface constrained to fit the offsets, h and h 0 , and azimuth; p s and p r are the initial slowness vector at CIP depth for the ray branch to source and to receiver, respectively; T s and T r represent traveltimes for each branch from the event coordinates to the surface for a defined offset and azimuth, n is the direction normal to a possible reflector and, e z the vertical direction. The model parameters are specified using B-splines interpolation (De Boor et al., 1978; Piegl and Tiller, 2012). The value of each parameter at a given position, C n (x, y , z) for n ∈{1, ..., N}, is uniquely defined by corresponding Sixteenth International Congress of the Brazilian Geophysical Society