Picking Datasets with SeisBench Picking Models

Installation

Download this repository to work in the project or install this project by running pip install git+https://github.com/JanisHe/seisbench_picking.git

Requirements

numpy, pandas, seisbench, obspy, joblib

Preparing Input Files

Stations

A .csv file that contains station information (note the file header!):

id	channel_code
FO.BETS.00	HH
RG.RITT.00	HH
RG.KUHL.00	EH
FO.OPS.00	BH
FO.OPS.00	HH

• id: ID of each station (network.station.location)
• channel_code: Channel code of each station. Note if you have different channel code, you can add a new line with a second channel code

Parameter File / Dictionary

In parfiles/parfile.yml you can find an example for settings. This parameter file is loaded in main, however instead of the parameter file the function main also takes a dictionary as input. Either the dictionary or the parameter file must contain the following parameters/keys:

• starttime: str of the start time, e.g. "2020-01-05"
• endtime: strof the end time, e.g "2020-01-10 05:00"
• sds_path: str of the pathname to read the waveform data from a SeisComp Data Structure (SDS)
• output_pathname: str where the picking results are saved
• workers: int of numbers of CPUs for parallelization
• picking:
  • picker: str which SeisBench picker is used
  • model: str which trained model is used for picking

More parameters, including optional ones are mentioned in the example parfile.

Start picking of dataset

To start the picking, you use the following code:

from seisbench_picking.core import main

parfile = "/Path/to/my/parfile.yml"  # or define a dictionary

# Dictionary instead of parfile
parameters = {
    "starttime": ...,
    "endtime": ...,
    ....
    "picking": {
        "picker": ...,
        "model": ...,
        ...
        }
    }

# Run main for parfile
main(parfile=parfile)
# Run main for parameter dict
main(parfile=parameters)

In case you get an ModuleNotFoundError, you can add the PythonPath to your script by

import sys
sys.path.append("/Path/to/my/SeisBench/picking/project")

Results

If you read in the settings from a parfile, the parfileand stations.csv file are copied to output_pathname. The picks will be saved in a single file called picks.csv, containing the following structure:

id	start_time	peak_time	end_time	peak_value	phase
TL.TL08.00	2020-04-01T00:06:05.720000Z	2020-04-01T00:06:05.840000Z	2020-04-01T00:06:05.960000Z	0.3475416	P
TL.TL08.00	2020-04-01T00:06:46.030000Z	2020-04-01T00:06:46.190000Z	2020-04-01T00:06:46.430000Z	0.605832	S
TL.TL08.00	2020-04-01T00:06:46.200000Z	2020-04-01T00:06:46.230000Z	2020-04-01T00:06:46.610000Z	0.33495384	P
TL.TL08.00	2020-04-01T00:06:50.450000Z	2020-04-01T00:06:50.450000Z	2020-04-01T00:06:50.510000Z	0.3267983	P

• id: ID of each station (network.station.location)
• start_time: Start time of the pick
• peak_time: Time of pick with the highest probability value (i.e. most probable arrival time)
• end_time: End time of the pick
• peak_value: Output probability from the loaded picker model
• phase: Phase type of the pick, either P or S

Additional notes

To keep memory low, it is recommended to pick only for single days. This means if you want to analyse a large dataset, use a for loop and define a new start- and end time in each iteration. Do not forget to copy your results from the output_pathname, otherwise your results will be overwritten in each iteration.

Next steps/Seismic Phase Association

One a file with all picks is created, the next step is to build from the picks an earthquake catalog by associating the picks. This is done by seismic phase associators. Here you can find a package which does the association using three different seismic phase associators.