High-resolution Imaging of Swarm Systems through seismological re-processing of earthquakeswarm episodes employing full waveform double-difference methods [HISS]
Geophysics
Final Report Abstract
This project developed new technologies targeting detection and localization of clustered earthquakes. Recordings of earthquake clusters can reach high data volumes, requiring efficient, scalable and automated modern techniques such as machine or deep learning for holistic data analysis. We developed a new technique designing and applying a deep convolutional neural network which allows to locate clustered events based on multi station full waveforms. After training the network the inference of earthquake locations takes approximately one millisecond per event. The precision of this approach depends on the quality of the dataset used for training. Benchmarking tests show that the precision of predicted hypocenter locations is similar to the training data set precision. 69 % of hypocenters deviate by less than 100 meters from their underlying reference locations. Furthermore, a new approach for event detection is evaluated based on the coherence of the waveforms with respect to the activation matrix of the first neural network layer. Further, we developed and evaluated a new method for focused analysis of attenuation within the source volume of clustered earthquakes. This method exploits spectral ratios of event couples sharing the greater part of their ray paths from the source volume to the receiver. Synthetic tests can reproduce the attenuation within the source volume. The application to recorded data shows a reduced attenuation of compressional waves within the source volume with respect to the surrounding material. The application to S phases fails due to the interference with p phase coda energy. The results are in line with a postulated fluid saturated and highly fractured source medium. Both approaches were developed and bench-marked using a new toolbox to generate large synthetic earthquake clusters and waveforms. Codes developed within this project have been developed under an open source policy and are available/accessible.
Publications
- (2017). Monitoring performance using synthetic data for induced microseismicity by hydrofracking at the Wysin site (Poland). Geophysical Journal International, 2017, 210: 42-55
López-Comino J. A., Cesca S., Kriegerowski, M., Heimann, S., Dahm, T., Mirek, J. and Lasocki, S.
(See online at https://doi.org/10.1093/gji/ggx148) - (2018). A Deep Convolutional Neural Network for Localization of Clustered Earthquakes Based on Multistation Full Waveforms. Seismological Research Letters, 90(2A):510516.
Kriegerowski, M., Petersen, G. M., Vasyura-Bathke, H., and Ohrnberger, M.
(See online at https://doi.org/10.1785/0220180320) - (2019). Event couple spectral ratio Q method for earthquake clusters: application to northwest Bohemia. Solid Earth, 10(1):317328
Kriegerowski, M., Cesca, S., Ohrnberger, M., Dahm, T., and Krüger, F.
(See online at https://doi.org/10.5194/se-10-317-2019) - A Python framework for efficient use of pre-computed Green's functions in seismological and other physical forward and inverse source problems, Solid Earth Discuss.
Heimann, S., Vasyura-Bathke, H., Sudhaus, H., Isken, M. P., Kriegerowski, M., Steinberg, A., and Dahm, T.
(See online at https://doi.org/10.5194/se-2019-85)
