Training PhaseNet with SeisBench

Requirements

Follow the instructions to install SeisBench. Further you need to install

• pyyaml
• tqdm

Training a model

To train a new model, download or create your own data set in SeisBench data format. Please follow the instructions and tutorials for downloading or creating SeisBench datasets on https://github.com/seisbench/seisbench.

To train an own PhaseNet model, modify the parameter file pn_parfile.yml and start training by running python -u core/run_pn_parfile.py ./pn_parfile.yml

Applying your trained model

Once you have successfully trained your own PhaseNet model, you can apply your model to unseen seismic data:

import seisbench.models as sbm
from obspy import read

model = sbm.PhaseNet.load("path/to/my/PhaseNet/model")  # Load your PhaseNet model
stream = obspy.read()  # Load your seismic data
picks = model.classify(stream,
                       batch_size=256,
                       P_threshold=0.2,
                       S_threshold=0.2,
                       blinding=[250, 250],
                       overlap=0.75,
                       stacking="max")
print(picks.picks)

Now, you should have a list of picks. For details about the arguments of the method classify, please read the SeisBench docs.