Deep Learning Model for Seismic Phase Picking

This repository contains the test dataset and trained model files used to evaluate a deep learning model developed for automated seismic phase picking in northwestern Türkiye. The dataset includes around 28,000 labeled three-component seismograms (30 seconds long, sampled at 100 Hz), each annotated with manually picked P- and S-wave arrivals. These waveforms were recorded between 2008 and 2022 by KOERI and regional seismic networks, and were preprocessed with 1–45 Hz bandpass filtering and amplitude normalization. In addition to the data, this archive provides the best 4 model checkpoints selected after training, which were used in the final ensemble evaluation. A Python script for running ensemble inference is also included, making it easier to reproduce our results or adapt the code for new datasets.