PhaseNet: A Deep-Neural-Network-Based Seismic Arrival Time Picking Method

1. Install miniconda and requirements

Download PhaseNet repository

git clone https://github.com/wayneweiqiang/PhaseNet.git
cd PhaseNet

Install to default environment

conda env update -f=env.yml -n base

Install to "phasenet" virtual envirionment

conda env create -f env.yml
conda activate phasenet

2. Pre-trained model

Located in directory: model/190703-214543

3. Related papers

Zhu, Weiqiang, and Gregory C. Beroza. "PhaseNet: A Deep-Neural-Network-Based Seismic Arrival Time Picking Method." arXiv preprint arXiv:1803.03211 (2018).
Liu, Min, et al. "Rapid characterization of the July 2019 Ridgecrest, California, earthquake sequence from raw seismic data using machine‐learning phase picker." Geophysical Research Letters 47.4 (2020): e2019GL086189.
Park, Yongsoo, et al. "Machine‐learning‐based analysis of the Guy‐Greenbrier, Arkansas earthquakes: A tale of two sequences." Geophysical Research Letters 47.6 (2020): e2020GL087032.
Chai, Chengping, et al. "Using a deep neural network and transfer learning to bridge scales for seismic phase picking." Geophysical Research Letters 47.16 (2020): e2020GL088651.
Tan, Yen Joe, et al. "Machine‐Learning‐Based High‐Resolution Earthquake Catalog Reveals How Complex Fault Structures Were Activated during the 2016–2017 Central Italy Sequence." The Seismic Record 1.1 (2021): 11-19.

4. Batch prediction

See examples in the notebook: example_batch_prediction.ipynb

PhaseNet currently supports four data formats: mseed, sac, hdf5, and numpy. The test data can be downloaded here:

wget https://github.com/wayneweiqiang/PhaseNet/releases/download/test_data/test_data.zip
unzip test_data.zip

For mseed format:

python phasenet/predict.py --model=model/190703-214543 --data_list=test_data/mseed.csv --data_dir=test_data/mseed --format=mseed --amplitude --response_xml=test_data/stations.xml --batch_size=1 --sampling_rate=100 --plot_figure

python phasenet/predict.py --model=model/190703-214543 --data_list=test_data/mseed2.csv --data_dir=test_data/mseed --format=mseed --amplitude --response_xml=test_data/stations.xml --batch_size=1 --sampling_rate=100 --plot_figure

For sac format:

python phasenet/predict.py --model=model/190703-214543 --data_list=test_data/sac.csv --data_dir=test_data/sac --format=sac --batch_size=1 --plot_figure

For numpy format:

python phasenet/predict.py --model=model/190703-214543 --data_list=test_data/npz.csv --data_dir=test_data/npz --format=numpy --plot_figure

For hdf5 format:

python phasenet/predict.py --model=model/190703-214543 --hdf5_file=test_data/data.h5 --hdf5_group=data --format=hdf5 --plot_figure

For a seismic array (used by QuakeFlow):

python phasenet/predict.py --model=model/190703-214543 --data_list=test_data/mseed_array.csv --data_dir=test_data/mseed_array --stations=test_data/stations.json  --format=mseed_array --amplitude

Notes:

The reason for using "--batch_size=1" is because the mseed or sac files usually are not the same length. If you want to use a larger batch size for a good prediction speed, you need to cut the data to the same length.
Remove the "--plot_figure" argument for large datasets, because plotting can be very slow.

Optional arguments:

usage: predict.py [-h] [--batch_size BATCH_SIZE] [--model_dir MODEL_DIR]
                  [--data_dir DATA_DIR] [--data_list DATA_LIST]
                  [--hdf5_file HDF5_FILE] [--hdf5_group HDF5_GROUP]
                  [--result_dir RESULT_DIR] [--result_fname RESULT_FNAME]
                  [--min_p_prob MIN_P_PROB] [--min_s_prob MIN_S_PROB]
                  [--mpd MPD] [--amplitude] [--format FORMAT]
                  [--s3_url S3_URL] [--stations STATIONS] [--plot_figure]
                  [--save_prob]

optional arguments:
  -h, --help            show this help message and exit
  --batch_size BATCH_SIZE
                        batch size
  --model_dir MODEL_DIR
                        Checkpoint directory (default: None)
  --data_dir DATA_DIR   Input file directory
  --data_list DATA_LIST
                        Input csv file
  --hdf5_file HDF5_FILE
                        Input hdf5 file
  --hdf5_group HDF5_GROUP
                        data group name in hdf5 file
  --result_dir RESULT_DIR
                        Output directory
  --result_fname RESULT_FNAME
                        Output file
  --min_p_prob MIN_P_PROB
                        Probability threshold for P pick
  --min_s_prob MIN_S_PROB
                        Probability threshold for S pick
  --mpd MPD             Minimum peak distance
  --amplitude           if return amplitude value
  --format FORMAT       input format
  --stations STATIONS   seismic station info
  --plot_figure         If plot figure for test
  --save_prob           If save result for test

The output picks are saved to "results/picks.csv" on default

file_name	begin_time	station_id	phase_index	phase_time	phase_score	phase_amp	phase_type
2020-10-01T00:00*	2020-10-01T00:00:00.003	CI.BOM..HH	14734	2020-10-01T00:02:27.343	0.708	2.4998866231208325e-14	P
2020-10-01T00:00*	2020-10-01T00:00:00.003	CI.BOM..HH	15487	2020-10-01T00:02:34.873	0.416	2.4998866231208325e-14	S
2020-10-01T00:00*	2020-10-01T00:00:00.003	CI.COA..HH	319	2020-10-01T00:00:03.193	0.762	3.708662269972206e-14	P

Notes:

The phase_index means which data point is the pick in the original sequence. So phase_time = begin_time + phase_index / sampling rate. The default sampling_rate is 100Hz

5. QuakeFlow example

A complete earthquake detection workflow can be found in the QuakeFlow project.

6. Interactive example

See details in the notebook: example_interactive.ipynb

7. Training

Download a small sample dataset:

wget https://github.com/wayneweiqiang/PhaseNet/releases/download/test_data/test_data.zip
unzip test_data.zip

Start training from the pre-trained model

python phasenet/train.py  --model_dir=model/190703-214543/ --train_dir=test_data/npz --train_list=test_data/npz.csv  --plot_figure --epochs=10 --batch_size=10

Check results in the log folder

happylifedev001/PhaseNet