← Back to Gallery

A Comparison of Machine Learning and Array-Beamforming Methods in Detecting Microearthquakes Near Cushing, Oklahoma, Using a Dense Nodal Array

Midcontinent seismic activity in the United States has sharply increased from background levels in the last 15 years. Oklahoma has been at the center stage for this increase due to the production of oil and gas and the associated injection of fracking fluids and wastewater disposal. While there were notable large earthquakes, such as the M5 Cushing and Pawnee events, there are a vast number of smaller events; detecting and cataloging these events can provide valuable insight into the evolution of the subsurface, leading to better industry practices and a more robust understanding of the impact that fluid injection has on fault activations. Here, two methods are employed to detect small earthquakes using continuous waveform with a dense nodal array near Cushing, Oklahoma. Array-beamforming detection requires no prior knowledge of the region's waveform or local velocity model, which detects events by comparing inter-station waveform similarity. LOC-FLOW is a machine learning-based workflow combining several algorithms to detect and locate earthquakes. Both methods have been shown to detect more events than are contained in previous catalogs and, if automated, can be a hands-off approach to creating catalogs of high resolution. The merits and drawbacks of each method must be understood to apply them most appropriately. The array-beamforming method detected thousands of events compared to the machine learning method's 112 events for the same period. However, array-beamforming is susceptible to detecting non-tectonic events such as passing traffic, perhaps owing to its roadside deployment. Event location remains variable between earthquakes detected by both methods, with an average offset of 3.18 km between the 43 co-detected events. Fine-tuning processing parameters will improve the beamforming event detection to remove false detections, such as increasing the maximum inter-station distance for which waveform similarities are compared, increasing the duration of the time window over which waveforms are compared, or considering different sub-arrays. Further characterization of detected events will be performed.

Session: Seismic Monitoring, Modelling and Management Needed for Geothermal Energy and Geologic Carbon Storage [Poster Session]

Type: Poster

Room: Exhibit Hall

Date: 5/2/2024

Presentation Time: 08:00 AM (local time)

Presenting Author: Braden Hoefer

Student Presenter: Yes

Additional Authors