今週のうなぎセミナーについてお知らせいたします。

Here is information of the Unagi-seminar(May, 9).

************** Seminar on Seismology IV A, C /地震学ゼミナールIV A, C (Unagi Seminar) **************

科目：地震学ゼミナールIV A, C / Seminar on Seismology IV A, C（修士・博士）
日時：2024年 5月 9日 (木) 13:30~
場所：京都大学 防災研究所 本館E-232D

Date and Time：2024-05-09, 13:30~
Place：Uji Campus Main Building E232D

---------- ---------- ---------- ----------

Speaker（発表者）1:　Masaru Morinaga

Title（題目）Jp:　地震動予測を広範化する拡張PLUM法の開発 ― 2023年トルコ・シリア地震のケーススタディ
Title（題目）En:　The Extended PLUM Method to Expand the Reach of the Ground Motion Estimation: a Case Study of the 2023 Turkiye-Syria Earthquake

Abstract（要旨）:
The Japanese Early Earthquake Warning system uses two methods. One of these is the PLUM (Propagation of Local Undamped Motion) method, which estimates seismic intensity without estimating epicenter or seismic magnitude, only using neighborhood intensity observations. Therefore, this method especially works for large earthquakes with large fault areas. In the current method, observation stations within 30 km of each target point are used for estimation, which requires a denser observation network than 30 km intervals. Many other countries do not facilitate such high-density observation. Therefore, we design a new scheme of the PLUM method that accommodates sparser observation networks.
In this study, we focused on the 2023 Turkiye-Syria Earthquake (Mw 7.8), which was caused by a long fault; therefore, the PLUM method is potentially profitable for such a case. This earthquake occurred at 01:17:34 UTC on 6 February 2023, the epicenter was located at 37.226 °N, 37.014°E and the depth was 10.0 km. We used available 249 data of seismometers in the time range from 01:17:05 to 01:26:20. For this earthquake, firstly, we experimentally applied the existing scheme of the PLUM method, and secondly, we installed an improved scheme.
In the existing scheme of the PLUM method, by using seismic intensities recorded at observation stations in radius R = 30 km around the target point and ground characteristics of the region, we calculate intensities of ground motion propagated from each observation station to the target point, and then take the maximum value as the predicted seismic intensity. However, in this case, 74 stations (27 %) have no neighbor station in a radius of 30 km, and 87 (32 %) have only one, which means lack of coverage. Therefore, we initially simulate with R varying from 10 km to 90 km with a step of 5 km. Besides, as for seismic intensity, we applied to these data a calculation method of JMA Real-time Seismic Intensity, same as the Japanese earthquake early warning system.
Subsequently, we devised an extended scheme of the PLUM method in which we flexibly select the reference area without fixing R: the scheme utilizing the Delaunay Triangulation. Firstly, we enumerate all observation stations in radius R1 around the target point. Secondly, in order of closeness, we enumerate other observation stations that are in radius R2 (> R1) around the target point and share a Delaunay Edge with the target point - until we count N or more stations and the Max Gap Angle < θ. Finally, by using seismic intensities recorded at these stations, we predict seismic intensities likewise.
The extended selection method, which we propose, has the feature that some usable neighbor stations around each target station are ensured. Moreover, in comparison with the existing scheme, we obtained a longer median leading time and less error between observed and predicted seismic intensity. Also, we found a trade-off between prediction error and leading time.
The extended selection method, allowing the distance from the target station to change, can optionally set the number of usable neighbor stations. By utilizing this characteristic, we can choose to emphasize prediction error or leading time depending on the purpose. Overall, the extended selection method (utilizing Delaunay Triangulation) brought better results than the existing selection method (fixing distance).

* * * * * * * * * * * * * *

Speaker（発表者）2:　Yanhan Chen

Title（題目）:　Deep Learning detects Low Frequency Earthquakes from OBS observations in Guerrero Seismic Gap

Abstract（要旨）:
Low frequency earthquakes (LFEs) occur largely during slow slip events at subduction interfaces. They affect the stress state of the seismogenic zone and potentially link to larger ordinary earthquakes. Detecting and analyzing low frequency earthquakes is crucial for a better understanding of the subduction process. As LFEs are weak seismic signals, it is challenging to detect them using the traditional regular earthquake detection method. The traditional useful method for identifying low frequency earthquakes is matched filter techniques, however, it often has low computational efficiencies and totally relies on the waveform templates. In recent years, deep learning algorithms have made good progress in the application of seismology due to the emergence of convolutional neural network (CNN), and similarly, deep learning has been attempted to be invoked in the study of slow earthquakes detection. Although there is a gap in recognition accuracy compared to regular earthquakes, it has advantages in terms of the number of detections and location and recognition without relying on templates.
Here, I plan to carry out using deep learning method to detect LFEs in the Mexican subduction zone. Previously, I used the matched filter method to detect LFEs and identified about 243 LFEs. Based on this result, I will employ the U-Net neural network structure, firstly, applying the trained model in Parkfield and Mexico identification of the LFEs to Mexico subduction zone LFEs detection. Then, based on the evaluation of the parameters associated with the detection results, an attempt will be conducted to retrain the model using the LFEs dataset detected by the previous matched filter method, and the retrained model is then used for the detection of LFEs.
By combining the relatively comprehensive catalog of LFEs detected by the matched filter method and the deep learning method, it may be possible to better help understand the mechanism of slow earthquakes in the Mexican subduction zone.

---------- ---------- ---------- ----------