Application of Deep Learning to Seismology – Estimation of Focal Mechanism Solutions using Automatically Picked P-wave First-Motion Polarities –

Deep learning which has been rapidly developed is now introduced to various fields including seismology. So, it is now time to measure the ability of deep learning. It seems that deep learning works well for processing seismic data, and in fact, deep learning is often applied to the picking of seismic data.
In this seminar, I will introduce an application study to focal mechanism estimations of numerous microearthquakes (Uchide, GJI, 2020). One of the keys to assess the rupture process of future earthquakes and understand the tectonics is the stress field. Therefore, I estimated focal mechanisms, as clues of the stress estimation, of ~ 110 thousand inland microearthquakes that occurred shallower than 20 km. I applied deep learning for picking the P-wave first-motion polarities of 2.3 million seismograms. The spatial distributions of P- and T-axes of obtained focal mechanisms imply the trend in the stress field at the regional scale in addition to local anomalies. The focal mechanism distribution covers the almost entire area of the Japan Islands. This study enables us to estimate the crustal stress field in detail in a wider area.

# The talk will be given in Japanese, while the slides will be written in English. Questions in Japanese or English would be welcome.