Support Vector Machine-Based On-Site Prediction for China Seismic Instrumental Intensity from P-Wave Features

• Published in: Pure and applied geophysics Vol. 180; no. 10; pp. 3495 - 3515
• Main Authors: Hou, Baorui, Li, Shanyou, Song, Jindong
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.10.2023; Springer Nature B.V
• Subjects: Early warning systems; Earth and Environmental Science; Earth Sciences; Earthquakes; Geophysics/Geodesy; Ground motion; Lead time; Modelling; Onsite; P waves; Performance evaluation; Seismic activity; Support vector machines; Windows (intervals); China; China seismic instrumental intensity scale; peak ground velocity; support vector machine; P-wave features; Earthquake early warning
• ISSN: 0033-4553; 1420-9136
• DOI: 10.1007/s00024-023-03335-6

The China seismic instrumental intensity can be used to measure the level of destruction and serve as the foundation of earthquake early warning (EEW) systems. To indirectly develop the instrumental intensity estimation and its application to EEW, we estimated the on-site filtered peak ground motion velocity (PGV) of the intensity using a support vector machine (SVM)-based model with eight P-wave features at a 3-s time window. Alert thresholds were set when the PGV was ≥ 8.18 cm/s (VII on the instrumental intensity scale). Compared with two linear estimation models (IV2 and P d ), the mean absolute error (MAE) and standard deviation of the error of the SVM estimation model were less, 0.241 and 0.298, respectively, with better performance on the PGV estimation. To evaluate the feasibility of transforming the SVM estimation for EEW by issuing alerts based on the intensity scale, we used the accuracy, precision, recall, F1 score, and false-negative rate (FNR) as evaluation metrics, achieving values of 99.62%, 95.68%, 79.90%, 87.08%, and 20.10%, respectively, using 11,970 records. We also provided the ratio, maximum, and average of the true positives to evaluate the lead time performance. Meanwhile, we used six earthquakes to evaluate the performance of our approach in detail. The approach performed well on EEW applications by issuing alerts based on the China seismic instrumental intensity. The analysis of the feature importance and data balance strategy can provide the basis for improving the performance of the SVM-based PGV estimation model.