5:15 PM - 7:15 PM
[SCG63-P04] Toward a more robust P-wave-based PLUM algorithm: continuous monitoring of the UD-component realtime seismic intensity
Keywords:Ground motion prediction, Earthquake early warning, P wave, PLUM algorithm, Wavefield-based ground motion prediction
Earthquake early warning (EEW) algorithms that predict ground motions directly from observation of the ongoing wavefield have been proposed to improve the ground motion prediction performance for complex scenarios such as large earthquakes (M>8) with non-negligible finite faults and multiple earthquakes that occur simultaneously (e.g., Kodera et al., 2018). The Japan Meteorological Agency (JMA) has introduced and been operating the PLUM algorithm (Kodera et al., 2018) in its EEW system since March 2018, and the PLUM implementation has led to several improvements such as the reduction of missed warnings. PLUM, which predicts future ground motion assuming that ongoing strong motions will propagate without attenuation within a short distance (30 km), has an advantage in that the algorithm can be implemented easily for an actual operational system; however, PLUM also has a technical issue in that the algorithm would not provide long lead times because high seismic intensities could not be predicted until strong motions caused by S-waves are observed within 30 km from prediction points. To improve the PLUM timeliness, Kodera (2018) proposed a P-wave-based PLUM algorithm that uses S-wave intensities predicted from the corresponding P-waves. The proposed algorithm provides S-wave intensity predictions by (1) detecting P-waves continuously with the ground motion polarization and (2) predicting S-wave intensities if the detection algorithm declares P-wave, adding a constant value (1.0) to the UD-component of the real-time seismic intensities (assuming that the UD-component real-time intensity corresponds to the P-wave intensity). The prediction performance was influenced strongly by the P-wave detection capability. On the other hand, there were events where overprediction did not occur when the strong motion prediction with the UD real-time intensity was applied to entire time windows including S-waves. This implies that the strong motion prediction based on the UD real-time intensity could skip the P-wave detection process. In this study, we investigated an algorithm that predicts future strong motions using the UD real-time intensity without the P-wave detection process.
We tested the proposed algorithm with KiK-net (surface) and JMA intensity meter stations for earthquakes from 2012 to 2019 with M>=6 and the maximum observed JMA intensity of >= 5-lower (~30 events in total). First, constant values added to the UD real-time intensity were obtained on a station-by-station basis, calculating the differences between the maximum UD and three-component intensities. The obtained values ranged from 0.5 to 1.5, and the majority were ~1.0. The standard deviation ranged from 0.1~0.3 for many stations, which means that the maximum intensities could be predicted within an error of 0.1~0.3 from the UD intensity. We also calculated actual intensity prediction errors for every station and found that the prediction error depended on the observed intensity. With the observed intensity of ~4 or more, the prediction error was within 0.5 intensity units; on the other hand, with a lower observed intensity, the predicted intensity tended to be overestimated. This implies that we may need to introduce an observed intensity threshold above which the proposed prediction algorithm performs, for the actual implementation of the proposed algorithm. Finally, we evaluated available lead times by actually adding the obtained constant values to the UD intensity. If the intensity threshold was set to 4.5, the available lead times were ~1 to 3 s; for the intensity threshold of 3.5, the lead times were ~1 to 10 s. Those results indicate that future ground motions could be predicted effectively from the UD intensity by the proposed algorithm.
Acknowledgements: In this study, we used waveform records of KiK-net, maintained by NIED.
We tested the proposed algorithm with KiK-net (surface) and JMA intensity meter stations for earthquakes from 2012 to 2019 with M>=6 and the maximum observed JMA intensity of >= 5-lower (~30 events in total). First, constant values added to the UD real-time intensity were obtained on a station-by-station basis, calculating the differences between the maximum UD and three-component intensities. The obtained values ranged from 0.5 to 1.5, and the majority were ~1.0. The standard deviation ranged from 0.1~0.3 for many stations, which means that the maximum intensities could be predicted within an error of 0.1~0.3 from the UD intensity. We also calculated actual intensity prediction errors for every station and found that the prediction error depended on the observed intensity. With the observed intensity of ~4 or more, the prediction error was within 0.5 intensity units; on the other hand, with a lower observed intensity, the predicted intensity tended to be overestimated. This implies that we may need to introduce an observed intensity threshold above which the proposed prediction algorithm performs, for the actual implementation of the proposed algorithm. Finally, we evaluated available lead times by actually adding the obtained constant values to the UD intensity. If the intensity threshold was set to 4.5, the available lead times were ~1 to 3 s; for the intensity threshold of 3.5, the lead times were ~1 to 10 s. Those results indicate that future ground motions could be predicted effectively from the UD intensity by the proposed algorithm.
Acknowledgements: In this study, we used waveform records of KiK-net, maintained by NIED.