Comparing machine-learning based picking algorithms in a subduction setting

As the number of seismic stations and experiments greatly increases due to ever greater availability of instrumentation, automated data processing becomes more and more necessary and important. Machine Learning (ML)methods are becoming widespread in seismology, with programsthat identify signals and patterns or extract features that can eventually improve our understanding of ongoing physical processes. We here focus on comparing and testing a selection of currently available methods for machine-learning-based seismic event detection and arrival time picking, performing a comparative study of the two autopickers EQTransformer and GPD with seismic data from the IPOC deployment in Northern Chile within the open-source Seisbench framework.

As a small benchmark dataset, we chose a random day for which we handpicked all visually discernible events on the 16 IPOC stations, which led to 200 events from 450 extracted, comprising 1493 P and 1163 S-phases. These events cover a large range of hypocentral depths (surface to >200 km) as well as magnitudes (<1.5 to 4.5).

We present first results from the application of the two autopickers EQTransformer and GPD, which have been shown to be most suitable for our type of dataset in a recent study by Münchmeyer et al. (2021), to IPOC data. We use our small benchmark dataset to evaluate detection rate (missed events, false detections) as well as picking accuracy (residuals to handpicks), and also investigate the effect of using different training datasets.

The present study is the first step towards the design of an automated workflow that comprises event detection and phase picking, phase association and event location and will be used to evaluate subduction zone microseismicity in different locations.