← Back to Sessions

Picking Regional Earthquake Waveforms With Neural Networks

Description: 

We present a set of deep learning models trained to perform seismic phase picking at regional distances. These models use raw seismometer data, as well as a combination of raw data and featured engineered data. Also, we present a multiview learning approach that takes advantage of having multiple instruments at the same location, like a seismometer and an accelerometer. The training of these models was done on a quality controlled dataset of over 1.5 million, 5 minute long three component waveforms that contain both P and S labeled arrivals. Our base model handles the first arriving P and S waves, whereas other models handle both the first and secondary arrivals, like picking Pn and Pg, Sn and Sg simultaneously. We show the performance of these models on train/test splits along with deploying the models to continuous data to find new uncataloged earthquakes.

Session: Network Seismology: Recent Developments, Challenges and Lessons Learned - I

Type: Oral

Date: 5/1/2024

Presentation Time: 02:45 PM (local time)

Presenting Author: Albert

Student Presenter: Yes

Invited Presentation: 

Authors