Small Seismic Events in Oklahoma Detected and Located by Machine Learning-Based Models

A complete earthquake catalog is essential to understand earthquake nucleation and fault stress. Following the Gutenberg-Richter law, smaller, unseen seismic events dominate the earthquake catalog and are invaluable for revealing the fault state. The published earth-quake catalogs, however, typically miss a significant number of small earthquakes. Part of the reason is due to a limitation of conventional algorithms, which can hardly extract small signals from background noise in a reliable and efficient way. To address this challenge, we utilized a machine learning method and developed new models to detect and locate seis-mic events. These models are efficient in processing a large amount of seismic data and extracting small seismic events. We applied our method to seismic data in Oklahoma, United States, and detected similar to 14 times more earthquakes compared with the standard Oklahoma Geological Survey catalog. The rich information contained in the new catalog helps better understand the induced earthquakes in Oklahoma.

Automated summary

A complete earthquake catalog is crucial for understanding earthquake activity. Traditional algorithms often miss many small earthquakes, but new models developed through machine learning methods can efficiently detect and locate seismic events, providing a more comprehensive earthquake catalog. Applied in Oklahoma, the new catalog detected approximately 14 times more earthquakes compared to the standard catalog, thereby contributing to a better understanding of induced earthquakes in the region.