Accelerating machine learning development and deployment in seismology through standardisation

Jannes Münchmeyer
Helmholtz Centre Potsdam – GFZ German Research Centre for Geosciences Potsdam, Germany

In recent years, machine learning algorithms have been presented for several seismological tasks, e.g., event detection, phase picking, magnitude estimation, or early warning. In many cases, these approaches considerably outperform traditional methods. However, as few benchmark datasets are available, comparing the performance of different models is difficult. Furthermore, there is no standard interface for applying machine learning models, in particular, for practitioners without strong machine learning experience. Both of these issues are detrimental to the progress of machine learning in seismology.

Here, we present SeisBench – A framework for machine learning in seismology. SeisBench is an open source collection of high-quality datasets and models. By offering interfaces for both machine learning researchers and seismological practitioners, it aims to bridge the gap between these groups. We present examples, how SeisBench can be used to facilitate seismic workflows, such as event picking or real-time ground motion estimation.