Metadata only
Author
Show all
Date
2022-12-15Type
- Other Conference Item
ETH Bibliography
yes
Altmetrics
Abstract
The continuously increasing availability of big datasets in seismology allows to enhance our understanding of earthquake related processes. However, manual processing of these datasets to determine earthquake locations and focal mechanisms is time consuming and limited to techniques that usually require signals well above the signal to noise ratio. In recent years machine learning models have been developed for routine tasks, such as picking phases and first motion polarities. We develop and test a convolutional neural network to compute first motion polarities for the Swiss seismic waveform archive, available from the Swiss Seismological Service (SED). We compare and evaluate three models: i) a model trained with a large data set from Southern California (SoCal); ii) a model trained with only Swiss data, i.e. with a significantly smaller data set; iii) the SoCal model, transfer learned with the Swiss data. We compare the predicted polarities from all three models with manually picked polarities for recent Swiss sequences (2019-2021). Our preliminary results with the SoCal model show that we achieve fairly high precision (>90%), but only moderate recall (80%). The re-trained models using the Swiss data will be able to accommodate both differences in the data itself between SoCal and Switzerland, as well as differences in manual picking practices. The ultimate goal is to implement the new model in near real time, especially in areas where the seismic network is dense and the seismicity rate is high, like in the Valais region. This work contributes towards a near automatic focal mechanism computation, and to track changes in fault orientation during energetic seismic sequences. Show more
Publication status
publishedExternal links
Journal / series
AGU Fall Meeting AbstractsPages / Article No.
Publisher
American Geophysical UnionEvent
Organisational unit
02818 - Schweiz. Erdbebendienst (SED) / Swiss Seismological Service (SED)
Notes
Conference lecture held on December 15, 2022.More
Show all metadata
ETH Bibliography
yes
Altmetrics