The aim of this challenge is to design and train a neural network for earthquake wave detection at Grillo seismic sensors. If you are successful, your algorithm will become part of OpenEEW earthquake detection/determination algorithms.

How to start

All you need for the challenge is contained in a folder available in AWS S3 storage. You can find it in here:

<https://grillo-priojects.s3.amazonaws.com/lse_challenge.zip>

Start by downloading and unzipping the folder. Navigate into the repository and install dependencies in requirements.txt. It is very basic, besides common packages Numpy and Scipy it will also install seismology package named Obspy.

pip install -r requirements.txt

This should be enough for the start. You will need more packages depending on the design of your solution.

Dataset

A network of about 20 Grillo instruments has been monitoring the southwest coast of Mexico since 2017. You are given segments with and without earthquake P-waves (a.k.a. signal and noise). Your task will be to train a model that can distinguish between them.

• Signal: Roughly 1,223 earthquake records from Grillo stations. Each record comes from a single seismic station, has 3 independent channels (x, y, z) and is 2000 data samples long. The records are centered at earthquake P-wave, which starts always at data sample 1,000.
• Noise: Randomly selected noise segments from the data and non-earthquake disturbances. There are 17,850 noise segments. Noise segments are also 2,000 data samples long.

Data is located in data folder in the repository. All data is in MSEED format, which is a standard binary seismological format. The easiest way to read-in MSEED data is using Obspy package for seismology.

Getting to know MSEED format and OBSPY package

To familiarize yourself with data, MSEED format and the Obspy package, navigate to codes directory and open obspy_tutorial.py. In depth tutorials on Obspy can be found at here.

• As always, we need to start with importing the package or required modules by:

import obspy

• Read MSEED file. To read-in a MSEED file mseed from test_data directory, we type:

st = obspy.read("test_data/2017_12_15_23_13_48_sta021.mseed")

The data is now stored in st variable, which is an instance of 'Stream' class of Obspy. Stream is a collection (list) of 'Traces', while each trace represents a single component (x,y,z) of a single seismic sensor.

Our MSEED file contained data from a single sensor, therefore the Stream should contain three Traces.