What the study found
An open-source, machine learning–based workflow could automatically detect and locate microseismic events in a hydraulic fracturing monitoring dataset from northern Montney, British Columbia. The authors report that the resulting catalog identified about 55% of a reference industrial catalog and showed consistent spatiotemporal patterns and moment magnitude (Mw 4.6) detection capability.
Why the authors say this matters
The study suggests that automatic microseismic monitoring is important for understanding fracture growth, reservoir behavior, fault activation, and seismic hazards linked to hydraulic fracturing. The authors conclude that their package-based workflow may work sufficiently well compared with proprietary commercial software and could be a viable open-source alternative for HF operators and regulatory agencies.
What the researchers tested
The researchers evaluated multiple pretrained deep learning pickers on geophone data recorded between 10 August and 10 September 2015, during an induced earthquake sequence that included a Mw 4.6 event. They selected PhaseNet, increased the input sampling rate from 100 to 500 Hz to address domain shift, and then applied automatic phase association and event location algorithms to build an event catalog.
What worked and what didn't
Increasing the sampling rate helped PhaseNet better identify low signal-to-noise ratio phases, where signal is weak relative to background noise. The resulting deep learning catalog successfully matched about 55% of the industrial reference catalog, and the authors say missed events were mainly true low-SNR arrivals or visually noise-like segments rather than events hidden by overlapping signals. The study also reports that the detection gap largely reflects differences between single-station classifiers and network-level brightness methods.
What to keep in mind
The abstract does not provide detailed quantitative performance metrics beyond the ~55% catalog match. It also describes the work as a controlled domain-transfer benchmark on one geophone dataset and one induced earthquake sequence, so the available summary does not establish how broadly the approach performs across other settings.
Key points
- The workflow automatically detected and located microseismic events in a hydraulic fracturing monitoring dataset.
- PhaseNet was the selected deep learning picker after testing multiple pretrained models.
- Raising the input sampling rate from 100 to 500 Hz improved identification of low signal-to-noise ratio phases.
- The resulting catalog identified about 55% of a reference industrial catalog.
- The authors say the workflow may be a viable open-source alternative to proprietary commercial software.
Disclosure
- Research title:
- Open-source machine learning detected microseismic events in Montney sequence
- Image credit:
- Photo by Google DeepMind on Pexels
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.