Code related to the submitted paper Deep learning models for regional phase detection on seismic stations in Northern Europe and the European Arctic.
Tested setup for installation of required packages :
conda create -n test python=3.10.12
conda activate test
pip install -r requirements.txt
To train a model for phase detection run :
python train.py
All training and model parameters can be changed in config.yaml.
This is a simply training example on a cpu. It is recommend to train on gpu. To train on gpu you have to adapt the call of train.py for example using docker.
Due to limited space on GitHub, this example uses a dummy training data set with only a few events (tf/data). Currently, to train you own model, you must generate your own data. The data structure can be explored by loading the dummy training data files.
We have uploaded parts of the full data (events in the NORSAR catalogue recorded at station ARA0 in years 2001--2022) to Zenodo. The models used in the paper are also too large for GitHub and are uploaded to Zenodo as well.
Data and model are available here:
To pick arrivals in continuous data of station ARA0 first download the models, store them in tf/output/models, adjust pred_config.yaml, and then run:
python predict.py
Andreas Köhler - andreas.kohler@norsar.no - ORCID
Erik B. Myklebust - ORCID
Project Link: https://github.com/NorwegianSeismicArray/tphasenet
- Models are built with TensorFLow
- ARCES waveform data are available via the Norwegian EIDA node
- Reviewed seismic event bulletins from which the input data labels were obtained are available from the Finish National Seismic Network and NORSAR