Deep learning prediction of measured earthquake waveforms from synthetic data

• DOI: 10.5194/egusphere-egu24-12357
• Publication Date: 2024-03-08T23:32:02Z
• Authors: Alexander Bauer; Jan Walda; Conny Hammer
• Keywords: Earthquake prediction

Seismic waveforms of teleseismic earthquakes are highly complex since they a superposition numerous phases that correspond to different wave types and propagation paths. In addition, measured contain noise contributions from the surroundings measuring station. The regional distribution seismological stations is often relatively sparse, in particular regions with low seismic hazard such as Northern Germany. However, detailed knowledge wavefield generated by large can be crucial for precise measurements or experiments carried out instance field particle physics, where wavefields considered noise. While synthetic cataloged computed any point on Earth’s surface, based simplified Earth model. As first step towards prediction dense region sparsely distributed stations, we propose train convolutional neural network (CNN) predict their counterparts. For purpose, compute past years IRIS synthetics engine (Syngine) use corresponding actual Germany labels. Subsequently, test performance trained events not part training data. promising results suggest able largely translate more ones, indicating means overcome lack complexity model underlying waveform computation paving way large-scale earthquakes.