1:45 PM - 3:15 PM
[SSS09-P07] Comparison methods of the spatial distributions of the seismic intensity
Keywords:ground motion, similarity
When a large earthquake occurs in Japan, the Headquarters for Earthquake Research Promotion releases the evaluation of the earthquake, which includes information on similar earthquakes that have occurred in the past. In such evaluations, source information such as the location and magnitude of the earthquake, mechanism solution, and interpretation of the earthquake location such as an interplate or intraplate earthquake are mainly used. For example, the Kagan angle has been proposed as an indicator of the difference between the mechanistic solutions of two earthquakes. Two Fukushima-oki earthquakes in February 2021 (Mj 7.3) and in March 2022 (Mj 7.4) occurred close to each other and were both in the subducting Pacific Plate. The peak ground acceleration due to the 2022 earthquake was reported to be larger than that of the 2021 earthquake in a wide area that is north of the epicenter (Suzuki et al., 2022, SSJ Fall Meeting). Since strong ground motions due to earthquakes are a direct cause of damage, it is important to evaluate not only the similarity of the earthquake source but also that of the ground motions with past earthquakes to predict and understand the damage. In this study, we investigate methods to evaluate the similarity of the spatial distribution of the seismic intensity between different earthquakes.
We used K-NET and KiK-net data from 255 earthquakes with a maximum seismic intensity of 5 lower or higher from 1997 to 2019. 29 of the 255 earthquakes were excluded from the analysis because the target ground motion data was contaminated by that of other earthquakes, and our final data set contains 226 earthquakes. For each earthquake, the real-time seismic intensity at each station was calculated, and spatial interpolation of the real-time seismic intensity was made every second to produce the every-second seismic intensity distribution. The seismic intensity distribution data for each earthquake was obtained by taking the maximum value at each mesh. The seismic intensity distribution data were generated on a 250-m mesh and consisted of 381,234 meshes. The data of the every-second seismic intensity distribution was created by: 1) converting the real-time seismic intensity into the peak ground velocity, 2) estimating the peak basement velocity using the site amplification factor by J-SHIS, 3) spatially interpolating using the inverse distance weighting method, 4) estimating the peak ground velocity using the site amplification factor at each mesh, and 5) converting the peak ground velocity into the equivalent value of the measured seismic intensity.
The seismic intensity distribution data of the 226 earthquakes were examined to evaluate the similarity using the following five methods: 1) Cosine distance, 2) Manhattan distance (L1-distance), 3) Euclidean distance (L2-distance), 4) Variance Reduction, and 5) Jaccard coefficient. For the comparison of the 2016 Kumamoto foreshock (Mj 6.5, maximum intensity 7), mainshock (Mj 7.3, maximum intensity 7), and aftershock (Mj 5.8, maximum intensity 6 upper) on April 16, 2016, all methods show the similar results in which the similarity of the aftershock to the foreshock is highest and that of the aftershock to the mainshock is lowest, which could be caused by the difference of the magnitude and the extent of the seismic intensity distribution. For the 2005 Miyagi-oki earthquake (Mj 7.2, maximum intensity of 6 lower), the 2011 Sanriku-oki earthquake (Mj 7.3, maximum intensity of 5 lower), and the 2011 Tohoku earthquake (Mj 9.0, maximum intensity of 7), the Cosine distance and Jaccard coefficient showed the highest similarity for the Miyagi-oki and Tohoku earthquakes. On the other hand, the other methods gave the highest similarity between the Miyagi-oki and Sanriku-oki earthquakes. We will examine the relationship between the comparison methods and the dimensional compression of the seismic intensity distribution data as a preprocessing method for comparison.
We used K-NET and KiK-net data from 255 earthquakes with a maximum seismic intensity of 5 lower or higher from 1997 to 2019. 29 of the 255 earthquakes were excluded from the analysis because the target ground motion data was contaminated by that of other earthquakes, and our final data set contains 226 earthquakes. For each earthquake, the real-time seismic intensity at each station was calculated, and spatial interpolation of the real-time seismic intensity was made every second to produce the every-second seismic intensity distribution. The seismic intensity distribution data for each earthquake was obtained by taking the maximum value at each mesh. The seismic intensity distribution data were generated on a 250-m mesh and consisted of 381,234 meshes. The data of the every-second seismic intensity distribution was created by: 1) converting the real-time seismic intensity into the peak ground velocity, 2) estimating the peak basement velocity using the site amplification factor by J-SHIS, 3) spatially interpolating using the inverse distance weighting method, 4) estimating the peak ground velocity using the site amplification factor at each mesh, and 5) converting the peak ground velocity into the equivalent value of the measured seismic intensity.
The seismic intensity distribution data of the 226 earthquakes were examined to evaluate the similarity using the following five methods: 1) Cosine distance, 2) Manhattan distance (L1-distance), 3) Euclidean distance (L2-distance), 4) Variance Reduction, and 5) Jaccard coefficient. For the comparison of the 2016 Kumamoto foreshock (Mj 6.5, maximum intensity 7), mainshock (Mj 7.3, maximum intensity 7), and aftershock (Mj 5.8, maximum intensity 6 upper) on April 16, 2016, all methods show the similar results in which the similarity of the aftershock to the foreshock is highest and that of the aftershock to the mainshock is lowest, which could be caused by the difference of the magnitude and the extent of the seismic intensity distribution. For the 2005 Miyagi-oki earthquake (Mj 7.2, maximum intensity of 6 lower), the 2011 Sanriku-oki earthquake (Mj 7.3, maximum intensity of 5 lower), and the 2011 Tohoku earthquake (Mj 9.0, maximum intensity of 7), the Cosine distance and Jaccard coefficient showed the highest similarity for the Miyagi-oki and Tohoku earthquakes. On the other hand, the other methods gave the highest similarity between the Miyagi-oki and Sanriku-oki earthquakes. We will examine the relationship between the comparison methods and the dimensional compression of the seismic intensity distribution data as a preprocessing method for comparison.