A friendly earthquake detector.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

6.6 KiB


A friendly earthquake detector.


First, install Pyrocko, then install Lassie:

git clone https://gitext.gfz-potsdam.de/heimann/lassie.git
cd lassie
sudo python setup.py install

Tutorial: detecting events in a regional network

This tutorial explains how to use Lassie to detect events in a regional network. The one-day test dataset from the Alentejo region in South Portugal can be downloaded from http://kinherd.org/lassie-example-alentejo.tar (235 MB). Data from 12 seismic stations from the DOCTAR experiment and from 5 permanent stations from regional networks [REFERENCE] are included. A region of roughly 100 x 100 km is covered by the available data.

Dataset preparation

# download data and unpack
wget http://kinherd.org/lassie-example-alentejo.tar
tar -xf lassie-example-alentejo.tar
cd lassie-example-alentejo
# data/                 - raw waveforms in mseed format
# stations.txt          - text file with station coordinates
# confirmed-events.txt  - catalog with some known events

To get a quick overview on the dataset, use the Snuffler application which is part of Pyrocko (tutorial).

snuffler --stations=stations.txt --events=confirmed-events.txt data/

Hints, when preparing your own dataset

  • NET.STA.LOC codes in waveform files must match exactly what is given in the stations file.
  • Waveform files and directory layout do not have to follow any specific convention, but with large files, data access can become less efficient. As a rule of thumb, use files smaller than 1MB each. To cut a dataset into a new chunksize, use the jackseis program which is part of Pyrocko, e.g. to cut a whole dataset into hour files use: jackseis INPUT_DIR --tinc=3600 --output-dir=OUTPUT_DIR

Running lassie from the command line

The Lassie earthquake detector can be run from the command line or by calling Lassie's library functions from a Python script. Here we will demonstrate the use of the lassie command line tool. To get brief help on available command line options, run lassie --help or lassie <subcommand> --help. Once dataset and configuration are ready, the command lassie scan <configfile> runs the main detector algorithm. But first it must be configured...

Lassie configuration

Lassie reads its configuration from a text file in the YAML format. To create an initial configuration, run

lassie init > config.yaml
# or filling in already the names of the stations file and the waveform directory:
lassie init --stations=stations.txt --data=data > config.yaml
%YAML 1.1
--- !lassie.Config

## Configuration file for Lassie, your friendly earthquake detector
## Receiver coordinates can be read from a stations file in Pyrocko format:
stations_path: 'stations.txt'

## Receivers can also be listed in the config file, lat/lon and carthesian 
## (x/y/z) = (North/East/Down) coordinates are supported and may be combined 
## (interpreted as reference + offset). Omitted values are treated as zero.
# receivers:
# - !lassie.Receiver
#   codes: ['', 'ACC13', '']
#   lat: 10.
#   lon: 12.
#   x: 2397.56
#   y: 7331.94
#   z: -404.1

## List of data directories. Lassie will recurse into subdirectories to find
## all contained waveform files.
- 'data'

## Processing time interval (default: use time interval of available data)
# tmin: '2012-02-06 04:20:00'
# tmax: '2012-02-06 04:30:00'

## Search grid; if not given here (if commented), a default grid will be chosen
# grid: !lassie.Carthesian3DGrid
#   lat: 38.7
#   lon: -7.9
#   xmin: -70e3  # in [m]
#   xmax: 70e3
#   ymin: -70e3
#   ymax: 70e3
#   zmin: 0.0
#   zmax: 0.0
#   dx: 2.5e3
#   dy: 2.5e3
#   dz: 2.5e3

## Size factor to use when automatically choosing a grid size
autogrid_radius_factor: 1.5

## Grid density factor used when automatically choosing a grid
autogrid_density_factor: 10.0

## Composition of image function
- !lassie.WavePacketIFC
  name: 'P'
  weight: 1.0
  fmin: 1.0
  fmax: 15.0
  fsmooth_factor: 0.1
  shifter: !lassie.VelocityShifter
    velocity: 3500.
- !lassie.WavePacketIFC
  name: 'S'
  weight: 1.0
  fmin: 1.0
  fmax: 10.0
  fsmooth_factor: 0.1
  shifter: !lassie.VelocityShifter
    velocity: 2500.

## Whether to divide image function frames by their mean value
sharpness_normalization: true

## Threshold on detector function
detector_threshold: 100.0

## Output filename for detections
detections_path: 'detections.txt'

Make sure, that the entries for data_paths and stations_path point to the respective directory and file.

An example event from the dataset is shown here:

example waveforms

Typical regional, shallow earthquakes appear in this dataset as clear P and S coda wave packets travelling with characteristic speeds, packet durations and frequency content through the network. To detect such events automatically, we can compose an image function with two contributions, one for each dominant phase. Such image function contributions (IFCs) can be defined under the image_function_contributions key in Lassie's configuration. Different types of IFCs are available and can be combined. Here, we use the WavePacketIFC which applies a normalized envelope based pre-processing to the waveforms before travel-time compensated stacking.

Running the detector

To test the configuration, select a short processing time span around one of the known events in the dataset. The time span can be restricted with the tmin and tmax entries in Lassie's configuration or with corresponding command line options --tmin='YYYY-MM-DD HH:MM:SS' and --tmax='YYYY-MM-DD HH:MM:SS.XXX', which override configuration settings.

Now run the detector with

lassie scan config.yaml

Scrutinize detections

After the detection has finished have a look at the results using Pyrocko's Snuffler:

lassie snuffle config.yaml

example snuffling

Snuffler opens loading waveform data together with station meta data and detections as event markers. Load a reference catalog to compare the detections to and scrutinize detection performance at different detection thresholds, interactively.