9:45 AM - 10:00 AM
[HDS09-04] Prototyping of real-time liquefaction damage estimation system and verification of estimation accuracy
Keywords:liquefaction, damage estimation, real-time, system development, multi-hazard
A real-time damage estimation system (Fujiwara et.al, 2019) that estimates damage immediately after an earthquake is effective for prompt disaster response, but currently only targets damage caused by shaking. On the other hand, the damage caused by the earthquake includes liquefaction and landslides, and we have started to develop a damage estimation system for these. In the previous report (Naito et al., 2021), we developed a system to predict the probability of liquefaction based on Senna et al. (2018), but there are problems such as the estimation result may be underestimated. Therefore, in this report, we have developed the liquefaction estimation system based on Senna et al. (2021) to improve the estimation accuracy and improved real-time performance.
This system automatically acquires the 250m mesh maximum velocity distribution from the real-time damage estimation system, and uses the topography classification group and the liquefaction probability formula using the maximum velocity proposed by Senna et al. (2021). The liquefaction probability is estimated, and the liquefaction risk rate is calculated by multiplying this by the liquefaction area ratio, which is the ratio of the liquefaction area in the mesh. In addition, using population and building data, the liquefaction-exposed population distribution and liquefaction-exposed building distribution are calculated in units of 250 m mesh and municipalities. Furthermore, based on the method of Shimizu et al. (2016), the total number of buildings destroyed by liquefaction is calculated in units of 250 m mesh and municipalities. These estimation results are visualized as a Web system.
For the purpose of verifying the estimation accuracy of this system, we compared the estimated liquefaction probability with the liquefaction mesh published by J-SHIS Labs for the 2016 Kumamoto earthquake, 2016 Tottori-ken Chubu earthquake, and 2018 Hokkaido Eastern Iburi earthquake. The former indicates that traces of liquefaction such as sand boil are confirmed in even one place in the mesh based on the interpretation results of the aerial photograph, and the latter indicates the probability that liquefaction will occur even in one place in the mesh. Therefore, both are by definition equal. The target area is divided by a larger 10km mesh, the ratio of the liquefaction occurrence mesh in each area is regarded as the liquefaction occurrence probability (measured value), and the average value of the liquefaction occurrence probability for each 10km mesh in this system (predicted value) are compared.
Taking the Kumamoto earthquake as an example, the estimation results by this system and the liquefaction meshes are aggregated every 10 km, and the measured values (Fig.1), the predicted values in this presentation (Fig.2), and the predicted values in the previous report (Fig.3) are compared. It can be seen that the predicted value, which was underestimated in the previous report, is approaching the measured value in this presentation. For a quantitative comparison, the correct answer rate for each mesh was calculated with the liquefaction occurrence probability of 1% as the threshold value. The overall accuracy was 74% in the previous report, but it improved to 82% in this presentation. In addition, the accuracy was 87% for the Iburi eastern earthquake and 94% for the Tottori prefecture central earthquake, both of which were able to confirm relatively high prediction accuracy.
The prototype of this system has started operation for internal use. For example, in the Hyuga-nada earthquake that occurred around 1:08 on January 22, 2022, a high liquefaction occurrence probability is predicted in the Nishi-Oita area of Oita prefecture (Fig.4), which is consistent with the situation of local damage in the media.
In the future, we would like to proceed with verification of actual damage for more earthquakes and to develop a system that integrates multi-hazard such as landslides.
This system automatically acquires the 250m mesh maximum velocity distribution from the real-time damage estimation system, and uses the topography classification group and the liquefaction probability formula using the maximum velocity proposed by Senna et al. (2021). The liquefaction probability is estimated, and the liquefaction risk rate is calculated by multiplying this by the liquefaction area ratio, which is the ratio of the liquefaction area in the mesh. In addition, using population and building data, the liquefaction-exposed population distribution and liquefaction-exposed building distribution are calculated in units of 250 m mesh and municipalities. Furthermore, based on the method of Shimizu et al. (2016), the total number of buildings destroyed by liquefaction is calculated in units of 250 m mesh and municipalities. These estimation results are visualized as a Web system.
For the purpose of verifying the estimation accuracy of this system, we compared the estimated liquefaction probability with the liquefaction mesh published by J-SHIS Labs for the 2016 Kumamoto earthquake, 2016 Tottori-ken Chubu earthquake, and 2018 Hokkaido Eastern Iburi earthquake. The former indicates that traces of liquefaction such as sand boil are confirmed in even one place in the mesh based on the interpretation results of the aerial photograph, and the latter indicates the probability that liquefaction will occur even in one place in the mesh. Therefore, both are by definition equal. The target area is divided by a larger 10km mesh, the ratio of the liquefaction occurrence mesh in each area is regarded as the liquefaction occurrence probability (measured value), and the average value of the liquefaction occurrence probability for each 10km mesh in this system (predicted value) are compared.
Taking the Kumamoto earthquake as an example, the estimation results by this system and the liquefaction meshes are aggregated every 10 km, and the measured values (Fig.1), the predicted values in this presentation (Fig.2), and the predicted values in the previous report (Fig.3) are compared. It can be seen that the predicted value, which was underestimated in the previous report, is approaching the measured value in this presentation. For a quantitative comparison, the correct answer rate for each mesh was calculated with the liquefaction occurrence probability of 1% as the threshold value. The overall accuracy was 74% in the previous report, but it improved to 82% in this presentation. In addition, the accuracy was 87% for the Iburi eastern earthquake and 94% for the Tottori prefecture central earthquake, both of which were able to confirm relatively high prediction accuracy.
The prototype of this system has started operation for internal use. For example, in the Hyuga-nada earthquake that occurred around 1:08 on January 22, 2022, a high liquefaction occurrence probability is predicted in the Nishi-Oita area of Oita prefecture (Fig.4), which is consistent with the situation of local damage in the media.
In the future, we would like to proceed with verification of actual damage for more earthquakes and to develop a system that integrates multi-hazard such as landslides.