← Back to Sessions

Using Machine Learning to Enhance Microseismicity Monitoring and Support Carbon Storage Initiatives in Oklahoma

Description: 

Induced seismicity in the southern midcontinent, primarily driven by human activities such as hydraulic fracturing and wastewater injection, presents significant challenges for monitoring due to the low signal-to-noise ratios of small-magnitude events. These conditions pose unique challenges to traditional analysis and detection efforts. To address this issue, we leveraged advanced machine learning techniques to enhance detection capabilities and improve seismic monitoring systems.
Using the PhaseNet framework, we adapted transfer learning to train models on a comprehensive, manually annotated dataset focused on microseismicity from Oklahoma and nearby regions. By applying fine-tuning and targeted preprocessing techniques, we demonstrated the effectiveness of this approach in detecting low-magnitude events. This work highlights the potential of transfer learning to better illuminate the complexities of anthropogenic seismicity.
We plan to publicly release this dataset, which spans from 2010 to 2024 and includes approximately 1.06 million traces of manually annotated seismic events. The dataset is predominantly events of magnitudes between 1 and 3 and depths ranging from 4 to 8 km. This makes it an ideal resource for microseismic research, benchmarking, and the development of advanced tools to support.
Looking forward, we aim to apply this machine learning-driven approach to support future carbon storage initiatives in Oklahoma. Effective microseismic detection and monitoring are critical for ensuring the safety and effectiveness of carbon sequestration efforts. Furthermore, this approach holds significant potential for monitoring other anthropogenic seismic events, advancing risk mitigation strategies, and contributing to the development of technologies needed for the energy transition.

Session: Seismology for the Energy Transition [Poster]

Type: Poster

Date: 4/16/2025

Presentation Time: 08:00 AM (local time)

Presenting Author: Hongyu

Student Presenter: No

Invited Presentation: 

Poster Number: 136

Authors