Seismic Phase Picking at Regional Distances

Albert L. Aguilar, & Gregory C. Beroza

Submitted September 10, 2023, SCEC Contribution #13278, 2023 SCEC Annual Meeting Poster #012

We present SKYNET, a set of deep learning models, combining convolutional and recurrent neural networks for picking seismic phase arrivals for earthquakes at distances up to 20 degrees of source-receiver separation. Our models were trained on the Curated Regional Earthquake Waveforms Dataset, which consists of over 2 million 5 minute, three component waveforms sampled at 100 Hz, with both P and S arrivals labeled for each example. We employed both supervised and semi-supervised learning to train our models, complemented with Feature Engineering and Multiview Learning.

We apply our models to earthquake monitoring in the Mendocino Triple Junction and Northern Chile, both places rich in offshore seismicity, where the closest land instrumentation is typically more than 100 km away from the epicenters.

Aguilar, A. L., & Beroza, G. C. (2023, 09). Seismic Phase Picking at Regional Distances. Poster Presentation at 2023 SCEC Annual Meeting.

Related Projects & Working Groups