Magnitude estimation is a vital task within earthquake early warning (EEW) systems (EEWSs). To improve the magnitude determination accuracy after P-wave arrival, we introduce an advanced magnitude prediction model that uses a deep convolutional neural network for earthquake magnitude estimation (DCNN-M). In this paper, we use the inland strong-motion data obtained from the Japan Kyoshin Network (K-NET) to calculate the input parameters of the DCNN-M model. The DCNN-M model uses 12 parameters extracted from 3 s of seismic data recorded after P-wave arrival as the input, four convolutional layers, four pooling layers, four batch normalization layers, three fully connected layers, the Adam optimizer, and an output. Our results show that the standard deviation of the magnitude estimation error of the DCNN-M model is 0.31, which is significantly less than the values of 1.56 and 0.42 for the τc method and Pd method, respectively. In addition, the magnitude prediction error of the DCNN-M model is not affected by variations in the epicentral distance. The DCNN-M model has considerable potential application in EEWSs in Japan.

Accurately estimating the magnitude within the initial seconds after the P-wave arrival is of great significance in earthquake early warning (EEW). Over the past few decades, single-parameter approaches such as the τc and Pd methods have been applied to EEW magnitude estimation studies considering the first 3 s after the P-wave onset. However, these methods present considerable scatter and are affected by the signal-to-noise ratio (SNR) and epicentral distance. In this study, using Japanese K-NET strong-motion data, we propose a machine-learning method comprising multiple parameter inputs, namely, the support vector machine magnitude estimation (SVM-M) model, to determine earthquake magnitudes and resolve the aforementioned problems. Our results using a single seismological station record show that the standard deviation of the magnitude prediction errors of the SVM-M model is 0.297, which is less than those of the τc (1.637) and Pd (0.425) methods. The magnitudes estimated by the SVM-M model within 3 s after the P-wave arrival are not obviously affected by the SNR or epicentral distance, and not overestimated for MJMA≤5. In addition, in an offline EEW application, the magnitude estimation error of the SVM-M model gradually decreases with increasing time after the first station is triggered, and the underestimation of event magnitudes for 6.5≤MJMA gradually improves. These results demonstrate that the proposed SVM-M model can robustly estimate earthquake magnitudes and has potential for EEW.

PurposeUsing the strong motion data of K-net in Japan, the continuous magnitude prediction method based on support vector machine (SVM) was studied.Design/methodology/approachIn the range of 0.5–10.0 s after the P-wave arrival, the prediction time window was established at an interval of 0.5 s. 12 P-wave characteristic parameters were selected as the model input parameters to construct the earthquake early warning (EEW) magnitude prediction model (SVM-HRM) for high-speed railway based on SVM.FindingsThe magnitude prediction results of the SVM-HRM model were compared with the traditional magnitude prediction model and the high-speed railway EEW current norm. Results show that at the 3.0 s time window, the magnitude prediction error of the SVM-HRM model is obviously smaller than that of the traditional τc method and Pd method. The overestimation of small earthquakes is obviously improved, and the construction of the model is not affected by epicenter distance, so it has generalization performance. For earthquake events with the magnitude range of 3–5, the single station realization rate of the SVM-HRM model reaches 95% at 0.5 s after the arrival of P-wave, which is better than the first alarm realization rate norm required by “The Test Method of EEW and Monitoring System for High-Speed Railway.” For earthquake events with magnitudes ranging from 3 to 5, 5 to 7 and 7 to 8, the single station realization rate of the SVM-HRM model is at 0.5 s, 1.5 s and 0.5 s after the P-wave arrival, respectively, which is better than the realization rate norm of multiple stations.Originality/valueAt the latest, 1.5 s after the P-wave arrival, the SVM-HRM model can issue the first earthquake alarm that meets the norm of magnitude prediction realization rate, which meets the accuracy and continuity requirements of high-speed railway EEW magnitude prediction.

The traditional magnitude estimation method, which establishes a linear relationship between a single warning parameter and the magnitude, exhibits considerable scatter and underestimation. In addition, the extraction of features from raw waveforms by a deep learning network is a black box. To provide a more robust magnitude estimation and to construct a deep learning network with an interpretable input, in light of deep learning and earthquake rupture physics, we have established a magnitude estimation network model (MEANet) via the physics-based features time series, an attention mechanism, and neural networks. We use events with 4 ≤  M ≤ 7.5 that occur in Japan and the Sichuan-Yunnan region, China, to train and validate MEANet, and then use MEANet to test additional events. Our results find that MEANet has a more robust magnitude estimation than the traditional τ c and P d methods, with a standard deviation of error of ±0.25 magnitude units at a single station with a 3 s P-wave time window. Within 10 s after the first station is triggered, based on the weighted average of the triggered stations, MEANet provides robust magnitude estimation without underestimation for events with 4 ≤  M ≤ 7.5. Our finding implies that the final magnitude is to some degree deterministic by the combination of deep learning and physics-based features. Meanwhile, MEANet might have potential for earthquake early warning.

Summary To rapidly and accurately provide alerts at target sites near the epicenter, we develop an on-site alert-level earthquake early warning (EEW) strategy involving P-wave signals and machine-learning-based prediction equations. These prediction equations are established for magnitude estimation and peak ground velocity (PGV) prediction accounting for multiple feature inputs and the support vector machine (SVM). These prediction equations are called SVM-M model for estimating magnitude and SVM-PGV model for predicting PGV, respectively. According to comparison between the predicted magnitude and PGV values with the predicted threshold values (M = 5.7 and PGV = 9.12 cm/s, respectively), different alert level (0, 1, 2, 3) is issued at the different recording site when the predicted magnitude or PGV values exceed the given threshold values. Alert level 3 means that both the predicted magnitude and the predicted PGV exceed a given threshold, and there may be serious damage in this recording site. We apply the method to three destructive earthquake events (M ≥ 6.5) occurred in Japan, and our results indicate that with regard to the performance of SVM-PGV model for predicting PGV, at 3 s after P-wave arrival, the percentage of successful alarms (SAs) for these three events is higher than 95 per cent, 73 per cent and 94 per cent, respectively, and the percentage of false alarms approaches 0. Additionally, with regard to the performance of SVM-M model for estimating magnitude, at 3 s after P-wave arrival, the percentage of SAs for these three events exceeds 95 per cent, and the percentage of missed alarms approaches 0. Moreover, almost all stations in the areas PGV ≥ 16 cm/s (IMM ≥ Ⅶ) near the epicenter issue alert level 3. The proposed method provides potential applications in EEW system.

No abstract

Introduction: China is one of the most seismically active countries in the world. It is an important task for a Chinese earthquake early warning system to quickly obtain robust magnitude estimation. However, within the first few seconds after P-wave arrival, there is considerable scatter in magnitude estimation for traditional methods based on a single early warning parameter.Methods: To explore the feasibility of using a convolutional neural network for magnitude estimation in China, establish a magnitude estimation model suitable for China and provide more robust magnitude estimation based on strong-motion data from China, we propose a new approach combining a convolutional neural network and transfer learning (TL) to construct a magnitude estimation model (TLDCNN-M) in this study.Results and Discussion: Our results show that for the same test dataset, in terms of the mean absolute error and standard deviation of magnitude estimation errors, the TLDCNN-M model has better performance than traditional methods and convolutional neural network models without using TL. Meanwhile, we apply the method to the 2022 Lushan M6.1 earthquake occurred in Sichuan province, China. At 3 s after the earliest P phase, the magnitude estimation error is less than 0.5. With the increase in time after the earliest P phase, the magnitude estimation is close to the catalog magnitude; at 10 s after the earliest P phase, the magnitude estimation error is less than 0.2.

No abstract