Signal Processing and Machine Learning for Monitoring the Subsurface Earth

Abstract:

The talk will present various uses of machine learning and signal processing techniques to better characterize and monitor the subsurface earth. In the first case study, unsupervised manifold-approximation method is used to process high-dimensional micro-seismic data measured by geophones. Second case study achieves accurate mapping of the spatiotemporal evolution of cracks and fractures embedded in materials using multipoint wave-transmission measurements. In the third case study, a large geological model (geomodel) containing numerous cells is massively compressed to extract low- dimensional representations of the subsurface earth.