Machine Learning Based Seismic Catalog for Puerto Rico Subduction Zone

This repository contains supplemental material for the following abstract presentation:

Seismotectonics of the Puerto Rico Subduction Zone: Leveraging Machine Learning Analysis of Ocean Bottom Seismometers, Seismological Society of America Annual Meeting,  2025, Asiye Aziz Zanjani, Heather R. DeShon, Seismological Research Letters, 96(2B)

Repository Contents

1. Relocated-ML-Catalog.txt

Machine learning-based seismic catalog developed using the PickBlue model and PhaseNet on the SeisBench platform with probability threshold of 0.2. Events were relocated using the open-source XCORLOC package with differential times from waveform cross-correlation. The catalog contains 4,969 events with the following header format:

YEAR, MONTH, DAY, HOUR, MINUTE, SECONDS, LATITUDE, LONGITUDE, DEPTH (km), NUMBER OF EVENT PAIRS, NUMBER OF DIFFERENTIAL TIMES, X ERROR (m), Y ERROR (m), Z ERROR (m), ORIGIN TIME ERROR (ms), rms_diff

2. histograms.png

Histograms displaying location uncertainties (latitude, longitude, depth, and origin time errors), as well as the distribution of event pairs and differential times from cross-correlation.

3. velocity_model.png

Plot of the simple 1-D velocity model used for travel time calculations in the relocation process.

4. rotating_3d_seismicity.mp4

3D visualization of the relocated seismic catalog.