11:45 AM - 12:00 PM
[SCG53-05] Verification of the predictive performance of the early aftershock-shaking forecast method based on extreme value statistics
Keywords:Extreme value statistics, Continuous seismogram, Forecast of aftershock shaking, verification of the predictive performance
Due to the numerous aftershocks occurring immediately after a major earthquake, the overlap of waveforms on seismograms leads to failure in detection of P- and S-waves. As a result, both the quality and quantity of earthquake catalogs deteriorate, making it challenging to understand and forecast the actual aftershock activity. As one solution to this difficulty, a method has been proposed to predict the shaking at target observation stations through extreme value analysis of seismograms recorded shortly after a major earthquake (Sawazaki, 2021). In this method, the maximum amplitudes within fixed time intervals are utilized as they follow a type of extreme value distribution known as non-stationary Frechet distribution (NFD). By employing Bayesian estimation of the control parameters of NFD, this method predicts the maximum amplitude and the frequency of shaking exceeding certain thresholds for approximately one week ahead within a few hours of the mainshock. In this study, we applied this method to three aftershock sequences occurred in Japan, which are characterized by different magnitudes and occurrence patterns, and verified the predictive performance of maximum amplitude and the number of felt earthquakes.
The target aftershock sequences in this study are from the 2008 Iwate-Miyagi Earthquake (MJ7.2), the mainshock of the 2016 Kumamoto Earthquake (MJ7.3), and the 2018 earthquake in northern Osaka Prefecture (MJ6.1). We analyzed Hi-net seismograms for three days, which continuously recorded the shakings following these earthquakes. For the Iwate-Miyagi Earthquake and the Kumamoto Earthquake, Hi-net stations within 100 km of the epicenter were used, while for the earthquake in northern Osaka Prefecture, Hi-net stations within 50 km were utilized. Hi-net records were corrected for instrument response up to 0.2 Hz, and the waveforms recorded mechanical saturation were replaced with concurrent KiK-net records.
As a result of the analysis, the prediction of maximum amplitude tended to be systematically overestimated except for the earthquake in northern Osaka Prefecture. On the other hand, regarding the frequency of felt earthquakes (nearly equivalent to Hi-net records exceeding 0.02 cm/s), the observed numbers were generally evenly distributed within the predicted range. This difference could be attributed to saturation of maximum amplitude with respect to magnitude, the upper limit of the range where the Gutenberg-Richter relation holds, or a combination of both possibilities.
In the case of the Kumamoto Earthquake, strong shaking and many felt earthquakes were predicted not only near the epicenter of the mainshock but also near the areas of induced seismic activity such as Beppu and Aso regions, and these shakings were actually observed. On the other hand, for the Iwate-Miyagi Earthquake, although the magnitude of the mainshock was similar to the Kumamoto mainshock, the shaking due to aftershocks was predicted to be smaller compared to the Kumamoto aftershocks. These results indicate that it is possible to make predictions reflecting the areal distribution of seismic activity, which cannot be handled by methods such as the Generic model, which predicts aftershock activity based solely on the magnitude of the mainshock, or the method proposed by Omi et al. (2016, 2019) that does not use information of the aftershock location.
Acknowledgements: This study is funded by Grants-in-Aid for Scientific Research C (Grant Number 21K03686) and Seismology TowArd Research innovation with data of Earthquake (STAR-E).
The target aftershock sequences in this study are from the 2008 Iwate-Miyagi Earthquake (MJ7.2), the mainshock of the 2016 Kumamoto Earthquake (MJ7.3), and the 2018 earthquake in northern Osaka Prefecture (MJ6.1). We analyzed Hi-net seismograms for three days, which continuously recorded the shakings following these earthquakes. For the Iwate-Miyagi Earthquake and the Kumamoto Earthquake, Hi-net stations within 100 km of the epicenter were used, while for the earthquake in northern Osaka Prefecture, Hi-net stations within 50 km were utilized. Hi-net records were corrected for instrument response up to 0.2 Hz, and the waveforms recorded mechanical saturation were replaced with concurrent KiK-net records.
As a result of the analysis, the prediction of maximum amplitude tended to be systematically overestimated except for the earthquake in northern Osaka Prefecture. On the other hand, regarding the frequency of felt earthquakes (nearly equivalent to Hi-net records exceeding 0.02 cm/s), the observed numbers were generally evenly distributed within the predicted range. This difference could be attributed to saturation of maximum amplitude with respect to magnitude, the upper limit of the range where the Gutenberg-Richter relation holds, or a combination of both possibilities.
In the case of the Kumamoto Earthquake, strong shaking and many felt earthquakes were predicted not only near the epicenter of the mainshock but also near the areas of induced seismic activity such as Beppu and Aso regions, and these shakings were actually observed. On the other hand, for the Iwate-Miyagi Earthquake, although the magnitude of the mainshock was similar to the Kumamoto mainshock, the shaking due to aftershocks was predicted to be smaller compared to the Kumamoto aftershocks. These results indicate that it is possible to make predictions reflecting the areal distribution of seismic activity, which cannot be handled by methods such as the Generic model, which predicts aftershock activity based solely on the magnitude of the mainshock, or the method proposed by Omi et al. (2016, 2019) that does not use information of the aftershock location.
Acknowledgements: This study is funded by Grants-in-Aid for Scientific Research C (Grant Number 21K03686) and Seismology TowArd Research innovation with data of Earthquake (STAR-E).