EQcorrscan: About

EQcorrscan is an earthquake detection package written in Python for Windows, OS X and Linux, and distributed under the LGPL. The purpose of this package is to detect repeating and near-repeating earthquakes from continuous seismic data. Although this package is designed first-and-foremost to use data from a seismic network, it works equally well when using a single station.

The codes are designed for large-scale problems with multiple (hundreds to thousands of) templates, run over multiple days of data. However, it works equally well for simple problems, with few templates and not much data - you just don't see the benefit of the parallel processing so much.

At the heart of the routine, multiple templates are correlated with the same day of data in parallel. using the Python multiprocessing module. As such the parallel processing at this level is CPU bound (you can run as many templates in parallel as your machine has CPUs).

At the top level, multiple days can be run in parallel on multiple nodes, or machines. For this, I have taken advantage of the NeSi PAN HPC cluster.

This package has been tested on machines ranging from small dual-core laptops and desktops, to large (multi-thousand core) cluster computers. The codes scale well when increasing the processing power, resulting in the ability to use (tested) 600 templates, through 6.5 years of data in less than 10 hours clock time.

Design

EQcorrscan is freely available on github, and pypi. They also have an initial set of utilities that allow integration with Seisan, which was chosen to be the first integrated standard observatory software, simply due to familiarity. We have since integrated with Obspy Event classes to allow many event meta-data types to be handled. We also have routines for reading pick data from SAC files to generate templates. These utilities also have methods for:

We do not claim that these codes are the necasarily the best available, but we hope to ignite collaboration from observational seismologists (and ideally computer scientists) to help grow this project. Ideally we would like to see this being a useful test-base for people to develop and share methods.