Fast and Accurate Seismic Tomography via Deep Learning

Abstract

This chapter presents a novel convolutional neural network (CNN)-based approach to seismic tomography, which is widely used in velocity model building (VMB). VMB is a key step in geophysical exploration where a model of the subsurface is needed, such as in hydrocarbon exploration for the Oil & Gas industry. The VMB main product is an initial model of the subsurface that is subsequently used in seismic imaging and interpretation workflows. Existing solutions rely on numerical solutions of wave equations, and requires highly demanding computation and the resources of domain experts. In contrast, we propose and implement a novel 3D CNN solution that bypasses these demanding steps, directly producing an accurate subsurface model from recorded seismic data. The resulting predictive model maps relationships between the data space and the final earth model space. The subsurface models are reconstructed within seconds, namely, orders of magnitude faster than existing solutions. Reconstructed models are free of human biases since no initial model or numerical technique tuning is required. This chapter is a significant extension of previous published material and provides a detailed explanation of the seismic tomography problem, and of the previously unpublished 3D CNN architecture, training workflows and comparisons to state-of-the-art.