地震可能會對不同地區造成不同程度的破壞，因為地震對各地區所造成的地層震動程度可能不同，也就是震度不同。如果能提前預測某一區域的震度，就能讓人們有更多的反應時間，設備也可以做好適當準備。為此本研究提出了一種深度學習方法，應用於地震發生初期時，可預測某個區域的未來震度，此模型被稱為Ground shaking intensity prediction (GSIP)。GSIP使用地震剛發生時，各測站所接收到的數秒P波加速度波形，並學習隱藏在波形及其頻譜中的特徵。然後根據提取的特徵預測此測站區域的未來震度。傳統方法需要靠經驗調整參數或選擇閥值，但通常難以選擇適當的數值，而本研究所提出的模型只需使用來自單個測站的波形前三秒就可以預測震度，且不需要任何參數上的手動調整就可以直接使用於各個測站。本研究中用於模型訓練和驗證的波形資料來自於1991年至2020年的台灣、1996年至2021年的日本以及2005年至2020年的意大利。實驗證明GSIP在容許一級誤差的情況下，可在驗證集上的各震度等級上達到85%以上的準確性，在精確率與召回率的表現上也很好。另外使用2021年和2022年發生在台灣的近期事件來評估GSIP在真實事件上的表現，其結果證實了它能夠準確預測不同地區的震度。此外GSIP也有實際測試於台灣中央氣象局的即時地震預警系統，並能夠有效地預測各地區的震度與提供即時預警。

Earthquakes can cause varying degrees of destruction due to the varying intensity of shaking experienced in different regions. Therefore, accurate prediction of shaking intensity in a specific area ahead of time is essential for effective preparedness and response measures. In this study, advanced deep learning techniques are used to develop a robust model for early prediction of onsite ground shaking intensity at specific locations, namely ground shaking intensity prediction (GSIP). The proposed model uses the initial window of the acceleration waveform received by a seismometer after the arrival of the P-wave and learns the features hidden in the waveform and frequency spectrum. These learned features are used for predicting the intensity of the earthquake. Unlike traditional methods that require manual adjustments of certain criterion thresholds, the proposed model only uses the initial three seconds of the P-wave from a single station to predict intensity levels. It can be used at different locations without the need for calibrations. In the training and validation stage, the acceleration waveforms from 1991 to 2020 in Taiwan, 1996 to 2020 in Japan, and 2005 to 2020 in Italy were used. The experiment shows that GSIP can achieve more than 85 percent accuracy at each intensity level on the validation set under the condition of allowing one-level tolerance, and it also performs well in precision and recall. The effectiveness of GSIP was evaluated using real events that occurred in Taiwan in 2021 and 2022. The results demonstrated the capability of the model to accurately predict intensity levels in different regions. Moreover, GSIP is integrated into the earthquake early warning (EEW) system operated by the Central Weather Bureau (CWB) of Taiwan for testing, where it has demonstrated excellent performance in providing real-time early earthquake warnings.

[1] H. Kanamori, "Real-time seismology and earthquake damage mitigation," Annu. Rev. Earth Planet. Sci., vol. 33, pp. 195--214, 2005.
[2] C.-Y. Hsieh, W.-A. Chao, and Y.-M. Wu, "An examination of the threshold-based earthquake early warning approach using a low-cost seismic network," Seismological Research Letters, vol. 86, no. 6, pp. 1664--1667, 2015.
[3] Y.-M. Wu and H. Kanamori, "Rapid assessment of damage potential of earthquakes in Taiwan from the beginning of P waves," Bulletin of the Seismological Society of America, vol. 95, no. 3, pp. 1181--1185, 2005.
[4] National Research Institute for Earth Science and Disaster Resilience. NIED K-NET, KiK-net
[5] A. Michelini, S. Cianetti, S. Gaviano, C. Giunchi, and Jozinovi, "INSTANCE--the Italian seismic dataset for machine learning," Earth System Science Data, vol. 13, no. 12, pp. 5509--5544, 2021.
[6] D.-Y. Chen, T.-L. Lin, H.-C. Hsu, Y.-C. Hsu, and N.-C. Hsiao, "An approach to improve the performance of the earthquake early warning system for the 2018 Hualien earthquake in Taiwan," Terr. Atmos. Ocean. Sci, vol. 30, pp. 423-433, 2019.
[7] Y.-M. Wu and H. Kanamori, "Experiment on an onsite early warning method for the Taiwan early warning system," Bulletin of the Seismological Society of America, vol. 95, no. 1, pp. 347-353, 2005.
[8] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition," Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, 1998.
[9] A. Vaswani et al., "Attention is all you need," Advances in neural information processing systems, vol. 30, 2017.
[10] Y. Chen, G. Zhang, M. Bai, S. Zu, Z. Guan, and M. Zhang, "Automatic waveform classification and arrival picking based on convolutional neural network," Earth and Space Science, vol. 6, no. 7, pp. 1244--1261, 2019.
[11] W. Zhu and G. C. Beroza, "PhaseNet : a deep-neural-network-based seismic arrival-time picking method," Geophysical Journal International, vol. 216, no. 1, pp. 261--273, 2019.
[12] W. Zhu, K. S. Tai, S. M. Mousavi, P. Bailis, and G. C. Beroza, "An end-to-end earthquake detection method for joint phase picking and association using deep learning," Journal of Geophysical Research: Solid Earth, vol. 127, no. 3, p. e2021JB023283, 2022.
[13] T.-L. Chin, K.-Y. Chen, D.-Y. Chen, and D.-E. Lin, "Intelligent real-time earthquake detection by recurrent neural networks," IEEE Transactions on Geoscience and Remote Sensing, vol. 58, no. 8, pp. 5440--5449, 2020.
[14] M. Kriegerowski, G. M. Petersen, H. Vasyura-Bathke, and M. Ohrnberger, "A deep convolutional neural network for localization of clustered earthquakes based on multistation full waveforms," Seismological Research Letters, vol. 90, no. 2A, pp. 510--516, 2019.
[15] A. Lomax, A. Michelini, and Jozinovi, "An investigation of rapid earthquake characterization using single-station waveforms and a convolutional neural network," Seismological Research Letters, vol. 90, no. 2A, pp. 517--529, 2019.
[16] J. Mnchmeyer, D. Bindi, U. Leser, and F. Tilmann, "Earthquake magnitude and location estimation from real time seismic waveforms with a transformer network," Geophysical Journal International, vol. 226, no. 2, pp. 1086--1104, 2021.
[17] O. M. Saad, A. G. Hafez, and M. S. Soliman, "Deep learning approach for earthquake parameters classification in earthquake early warning system," IEEE Geoscience and Remote Sensing Letters, vol. 18, no. 7, pp. 1293--1297, 2020.
[18] M. P. A. van den Ende and J. P. Ampuero, "Automated seismic source characterization using deep graph neural networks," Geophysical Research Letters, vol. 47, no. 17, p. e2020GL088690, 2020.
[19] R. Otake, J. Kurima, H. Goto, and S. Sawada, "Deep learning model for spatial interpolation of real-time seismic intensity," Seismological Society of America, vol. 91, no. 6, pp. 3433--3443, 2020.
[20] A. Datta, D. J. Wu, W. Zhu, M. Cai, and W. L. Ellsworth, "DeepShake: Shaking intensity prediction using deep spatiotemporal RNNs for earthquake early warning," Seismological Society of America, vol. 93, no. 3, pp. 1636--1649, 2022.
[21] H. Kubo, T. Kunugi, W. Suzuki, S. Suzuki, and S. Aoi, "Hybrid predictor for ground-motion intensity with machine learning and conventional ground motion prediction equation," Scientific reports, vol. 10, no. 1, pp. 1--12, 2020.
[22] J. Song, J. Zhu, Y. Wang, and S. Li, "On-site alert-level earthquake early warning using machine-learning-based prediction equations," Geophysical Journal International, vol. 231, no. 2, pp. 786--800, 2022.
[23] Y.-J. Chiang, T.-L. Chin, and D.-Y. Chen, "Neural network-based strong motion prediction for on-site earthquake early warning," Sensors, vol. 22, no. 3, p. 704, 2022.
[24] S. M. Mousavi, W. L. Ellsworth, W. Zhu, L. Y. Chuang, and G. C. Beroza, "Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking," Nature communications, vol. 11, no. 1, p. 3952, 2020.
[25] T. P. a. M. G. a. M. Denolle, "Convolutional neural network for earthquake detection and location," Science Advances, vol. 4, no. 2, p. e1700578, 2018, doi: 10.1126/sciadv.1700578.
[26] T.-Y. Hsu et al., "Rapid on-site peak ground acceleration estimation based on support vector regression and P-wave features in Taiwan," Soil Dynamics and Earthquake Engineering, vol. 49, pp. 210--217, 2013.
[27] J. Zhu, S. Li, and J. Song, "Magnitude estimation for earthquake early warning with multiple parameter inputs and a support vector machine," Seismological Research Letters, vol. 93, no. 1, pp. 126--136, 2022.
[28] J. Mnchmeyer, D. Bindi, U. Leser, and F. Tilmann, "The transformer earthquake alerting model: A new versatile approach to earthquake early warning," Geophysical Journal International, vol. 225, no. 1, pp. 646--656, 2021.
[29] L. Ilya and H. Frank, "Decoupled Weight Decay Regularization," presented at the International Conference on Learning Representations, 2019.
[30] P. Brondi, M. Picozzi, A. Emolo, A. Zollo, and M. Mucciarelli, "Predicting the macroseismic intensity from early radiated P wave energy for on-site earthquake early warning in Italy," Journal of Geophysical Research: Solid Earth, vol. 120, no. 10, pp. 7174--7189, 2015.
[31] I. Spingos, F. Vallianatos, and G. Kaviris, "The scaling of PGA with IV2p and its potential for Earthquake Early Warning in Thessaly (Central Greece)," Bulletin of the Geological Society of Greece, vol. 58, pp. 179--199, 2021.
[32] J. W. Reed and R. P. Kassawara, "A criterion for determining exceedance of the operating basis earthquake," Nuclear Engineering and Design, vol. 123, no. 2-3, pp. 387--396, 1990.
[33] A. Arias, "A measure of earthquake intensity," Seismic design for nuclear plants, pp. 438--483, 1970.
[34] P.-L. Huang, T.-L. Lin, and Y.-M. Wu, "Application of τc* Pd in earthquake early warning," Geophysical Research Letters, vol. 42, no. 5, pp. 1403--1410, 2015.
[35] T.-Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollár, "Focal loss for dense object detection," in Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017, pp. 2980-2988.
[36] B. Zhou, Q. Cui, X.-S. Wei, and Z.-M. Chen, "BBN: Bilateral-Branch Network With Cumulative Learning for Long-Tailed Visual Recognition," in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (CVPR), 2020, pp. 9719-9728.