Exploring Machine Learning and Deep Learning Models for Seismic Event Classification in the Pacific Northwest

Akash Kharita, Marine A. Denolle, Alexander Hutko, Renate Hartog, & Stephen D. Malone

Published September 8, 2024, SCEC Contribution #14020, 2024 SCEC Annual Meeting Poster #059

The Pacific Northwest is a seismically active region that experiences a wide range of events, including small tectonic earthquakes, large megathrust events, low-frequency seismicity, mining explosions, quarry blasts, volcanic activity, and near-surface slope failures. Currently, analysts at the Pacific Northwest Seismic Network manually review automatically detected and located seismic events to classify them as earthquakes, explosions, or surface events. However, surface events, which are by-products of the automatic detection system, are not further analyzed to determine their specific types. Additionally, distinguishing between earthquakes and explosions can be challenging due to overlaps in their spatial, temporal, and magnitude characteristics.

To address these challenges, we have developed an event classification system that leverages machine learning and deep learning techniques, using a curated dataset of seismic events collected in the Pacific Northwest over the past 20 years. We explored various commonly used features, including physics-based, statistics-based, and scattering network features, and tested different models to identify those that provided the best performance. For each set of features and models, we tested different window lengths, frequency bands, and resampling rates, ultimately selecting three models for further exploration. To gain insights into the model's decision-making process, we computed the most important features and determined the minimum number of features needed to optimize computational efficiency. Our best-performing model is capable of classifying events with over 93% accuracy. We are currently working on implementing this classification system in real-time and properly aggregating the results from multiple stations.

Citation
Kharita, A., Denolle, M. A., Hutko, A., Hartog, R., & Malone, S. D. (2024, 09). Exploring Machine Learning and Deep Learning Models for Seismic Event Classification in the Pacific Northwest. Poster Presentation at 2024 SCEC Annual Meeting.


Related Projects & Working Groups
Seismology