WATCH VIDEOSeismic Inversion is a geophysical method to reconstruct the earth subsurface image by inverting seismic data observed via the multiple distributed sensors on the surface. As the data science rapidly reshapes the methodology used in many domains, we are working on applying data science to solve seismic inversion and interpretation problems. When the data science meets geophysics, it requires big training seismic data containing sufficient-enough features to let deep learning networks to learn and extract these features. Although researchers have shown promising results in seismic data interpretation with deep learning models, one of the many challenges researchers face is the availability of sufficient seismic datasets to meet the deep learning data requirements. We present a deep learning based model to generate the synthetic seismic velocity models using Generative Adversarial Networks(GAN) in this paper. We first describe the fundamentals of GAN and then present our experiments of training GANs using TensorFlow. Our results demonstrate the feasibility of generating synthetic seismic velocity models using GAN, which builds a foundation to enable data science to solve seismic inversion and interpretation.
