AI4EPS/PhaseNet

What it solves

PhaseNet is designed to automate the process of "picking" seismic arrival times—identifying the exact moment P-waves and S-waves from an earthquake reach a recording station. This is a critical step in earthquake detection and analysis that traditionally requires manual effort or simpler algorithms.

How it works

It uses a deep neural network to analyze raw seismic data and predict the probability of a seismic phase (P or S wave) occurring at each time point. The system supports multiple data formats (mseed, sac, hdf5, and numpy) and provides a pre-trained model for immediate use, as well as tools to train the model further on custom datasets.

Who it’s for

Seismologists and geophysicists who need high-resolution earthquake catalogs and automated, accurate seismic phase picking from raw data.

Highlights

• Multi-format support: Compatible with mseed, sac, hdf5, and numpy data formats.
• Pre-trained model: Includes a ready-to-use model for immediate seismic arrival time prediction.
• Batch processing: Supports batch prediction for large seismic datasets.
• Integration: Works as part of a larger earthquake detection workflow in the QuakeFlow project.