Exploration of subsurface faults by big data analysis of seismic waveforms using signal processing and machine learning

Abstract

Earthquakes occur reflecting the physical conditions of the seismogenic field. Through the properties of microearthquakes, we investigate the physical conditions of the seismogenic field in order to assess future earthquakes. Here, we study the geometry of inland active faults at depths from seismic data. The distribution of microearthquakes has been used as an indicator of subsurface fault geometry in a subjective way, for example, using cross-sections. To obtain three-dimensional fault geometries objectively, we developed methods to cluster microearthquakes on the same planes. In addition, we studied seismic later phases scattered and reflected from subsurface structural boundaries. As a basis of seismic data processing, we also developed a method using a variational autoencoder (VAE) to detect anomalies in seismic data due to the failure of seismometers. Applications of all these methods to observation data manifested the effectiveness of our developed methods.