Machine Learning Enhanced Seismic Monitoring at 100 Km and 10 m Scales

Abstract

Rapid and accurate monitoring of seismicity is crucial for both reservoir management and risk mitigation. It is important to monitor not only seismic events due to hydraulic fracturing but also naturally occurring background seismicity. Depending on the size of the reservoir, monitoring of background and induced seismicity usually deals with different spatial scales. Thanks to an expanded station coverage and continuous improvements on seismic sensors, a large amount of data has been collected for both natural and induced seismicity. Traditional seismic data processing techniques can provide rapid and automatic seismic event catalogs or more accurate results but at a higher time and labor cost. To obtain accurate seismicity catalogs rapidly and extract valuable information from large amounts of data effectively, we developed a machine learning enhanced seismic monitoring workflow that combines cutting-edge machine learning techniques and advanced seismic data processing algorithms. The workflow is suitable for the monitoring of both natural and induced seismicity at dramatically different spatial scales. To demonstrate this capability, we applied the workflow to the Oklahoma region as an example for 100 km scale sites and to the experiment 1 site of the EGS Collab project with a length scale of 10 m. Our workflow not only produces high-precision seismicity catalogs but also images the 3D subsurface structure with high resolution. We will compare our results against those from traditional techniques.