Velocity Downward Continuation through Generative Models

Summary Velocity model building (VMB) should reflect the physics involved in wave propagation and honor the location of the recording surface in which we extract the information. Thus, we propose using a generative model to predict velocity from shallow to deep. The shallow velocity distribution acts as prior to predict the deep, and in our implementation, with using the seismic image and the well information as guide. An application on offshore data from North west Australia demonstrated the versatility of this approach in predicinting an accurate velocity model. With the generative nature of the process, we can also quantify the uncertainty, which was well in agreement with what we expected. We will share more examples in the presentation of this work.