Earthquake magnitude estimation using a two-step convolutional neural network.
Saved in:
| Title: | Earthquake magnitude estimation using a two-step convolutional neural network. |
|---|---|
| Authors: | Liu, Xinliang1 (AUTHOR), Ren, Tao1 (AUTHOR) chinarentao@163.com, Chen, Hongfeng2 (AUTHOR), Dimirovski, Georgi M.3 (AUTHOR), Meng, Fanchun1 (AUTHOR), Wang, Pengyu1 (AUTHOR) |
| Source: | Journal of Seismology. Feb2025, Vol. 29 Issue 1, p241-256. 16p. |
| Subject Terms: | *Convolutional neural networks, *Magnitude estimation, *Earthquakes, *Artificial intelligence, *Image processing, *Earthquake magnitude |
| Abstract: | In this paper, an efficient two-step convolutional neural network (CNN) procedure is proposed to estimate earthquake magnitude using raw waveform data up to only 4 s after the P wave onset. In the proposed procedure, magnitude estimation is split into classification task and regression task. The classification task trains a CNN model to estimate the magnitude range by employing unsure responses that represent the classification decision boundary. In addition, the regression task trains two CNN models to estimate the specific magnitudes of large and small earthquakes, respectively. After training, the classification model achieves an accuracy of 98.63%. The mean absolute error (MAE) of the large earthquake regression and the small earthquake regression models are 0.26 and 0.46, respectively. The ideology behind the two-step procedure effectively address two main issues in earthquake early warning (EEW) systems: reducing missed alert caused by seismometer saturation and improving the accuracy of estimating specific magnitudes. Currently, this procedure has been connected to China Earthquake Networks Center (CENC) for real-time monitoring. [ABSTRACT FROM AUTHOR] |
| Database: | Energy & Power Source |
|
Full text is not displayed to guests.
Login for full access.
|
|
| Abstract: | In this paper, an efficient two-step convolutional neural network (CNN) procedure is proposed to estimate earthquake magnitude using raw waveform data up to only 4 s after the P wave onset. In the proposed procedure, magnitude estimation is split into classification task and regression task. The classification task trains a CNN model to estimate the magnitude range by employing unsure responses that represent the classification decision boundary. In addition, the regression task trains two CNN models to estimate the specific magnitudes of large and small earthquakes, respectively. After training, the classification model achieves an accuracy of 98.63%. The mean absolute error (MAE) of the large earthquake regression and the small earthquake regression models are 0.26 and 0.46, respectively. The ideology behind the two-step procedure effectively address two main issues in earthquake early warning (EEW) systems: reducing missed alert caused by seismometer saturation and improving the accuracy of estimating specific magnitudes. Currently, this procedure has been connected to China Earthquake Networks Center (CENC) for real-time monitoring. [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 13834649 |
| DOI: | 10.1007/s10950-024-10258-9 |
