hide
Free keywords:
-
Abstract:
The open-source, seismological software package SeisComP is widely used for seismic monitoring world-wide. Its automatic phase picking module consists of an STA/LTA-based P-wave detector augmented by an AIC onset picker. With proper configuration, it allows detection and accurate onset picking for a wide range of seismic signals. However, it cannot match the performance of experienced analysts. Especially broadband data are often challenging for phase pickers due to the enormous variety of the signals of interest. In order to optimize quality and number of automatic picks and reduce time-consuming manual revision, we chose to develop a machine-learning repicker module for SeisComP based on the SeisBench framework. SeisBench supports several deep-learning pickers and comes pre-trained for different benchmark datasets, one of which was contributed by GFZ Potsdam.The repicking module consists of several submodules that interact with both SeisComP and SeisBench via their Python interfaces. The current workflow is based on existing locations generated with a classical SeisComP setup. Shortly after event detection and preliminary location, our repicker (1) starts to repick previously picked onsets and (2) attempts to generate additional picks.Preliminary results are encouraging. The deep-learning repicker substantially improves pick quality over a large frequency range. The number of picks available per event has approximately doubled and the publication delay is often reduced, especially for small events. The total number of published events has increased by about 20 per cent.
