Efficient Full Waveform Inversion by using Plane-wave Transformation

Abstract

Full waveform inversion (FWI) can provide accurate information about the velocities of subsurface media, and advances in computer science have made the technique popular in seismic data processing. However, a FWI still involves high computational costs and is sensitive to noise included in data. Additionally, critical problems can occur when a two-dimensional (2-D) FWI is conducted using three-dimensional (3-D) data acquired in the field to save computational time, because 2-D and 3-D wave equations have different Green’s functions. A plane-wave transformation method was suggested to mitigate these problems, and acoustic/elastic FWIs using plane-wave gathers have been developed. Because plane-wave gathers transformed from shot gathers are used as input data in inversions, the computational cost can be significantly reduced and efficient FWI is possible for a massive data set, because far fewer plane-wave gathers are involved, compared with common shot gathers. Moreover, there is no difference in Green’s functions between observed and predicted data because plane-wave data are always followed by a one-dimensional Green’s function, regardless of the dimension of the wave equation. As a result, 2-D FWI using plane-wave data can be implemented for 3-D data. To verify the validity of this method, the proposed 2-D/3-D acoustic FWIs were applied to the SEG/EAGE Overthrust model data set. The data set obtained from the 3-D model was successfully inverted by the proposed 2-D plane-wave FWI and the 3-D FWI, which used a smoothed version of the 2-D inversion result as the initial model. The results demonstrated that the velocity information was accurate, and the computational time was dramatically reduced, compared with that of the conventional method. To verify that the plane-wave inversion method can be expanded to an elastic case, a multiparameter inversion was carried out using the Mamousi-2 data set. High-resolution images of S-velocity and density, as well as P-velocity, were obtained for noise-free data. Additionally, the plane-wave approach provided clearer images than the conventional method for noise-added data, because it reduced the influence of random noise through a stacking process.