11:00 AM - 11:15 AM
[HDS11-07] Enhancement and verification of the real-time damage estimation system for landslide and liquefaction
Keywords:Landslide, Slope failure, Liquefaction, Damage estimation, Realtime, Earthquake disaster
The “Real-time damage estimation system (J-RISQ)” (Fujiwara et al., 2019) has been in stable operation since 2013 by the NIED. In this presentation, we will explain the enhancement of J-RISQ for immediate estimation of landslide and liquefaction damage, associated with earthquakes. We also report the results of validation of the J-RISQ for several earthquakes.
The landslide damage estimation system converts the maximum velocity distribution obtained from J-RISQ into acceleration using the formula by Ooi et al. (2002), and then calculates the collapse probability using the Rokko formula (Method 1) by Uchida et al. In addition to this method, we have added a function using the modified Rokko formula (Method 2) proposed by Kamiya et al. (2012).
The liquefaction damage estimation system uses the maximum velocity distribution obtained from J-RISQ to calculate the probability of liquefaction using the model proposed by Senna et al. (2021). In addition to the microtopography classification in which the parameters are grouped in consideration of elevation, specific height, and water distance (Method A), the method allow the selection of microtopography classification without grouping (Method B).
The estimation results from each system were verified using damage data from earthquakes that have occurred in recent years.
The landslide damage estimation data for the Chuetsu, Kumamoto, Iburi, and Noto earthquakes was developed using J-RISQ. As correct data for landslides, we used the inventory published by the Geospatial Information Authority of Japan (GSI) as the “Data Release Site for SGDAS” (Iwahashi et al., 2022) and the slope failure data from the Noto earthquake (GSI, 2024). In the former case, a collapse was considered to exist when the points were located within a 250 m mesh, without distinguishing between landslide and slope failure. In the latter case, the slope failure data overlapping with 250 m mesh was assumed to be collapsed and used as the ground truth for each. The estimation results were considered to be collapsed when the probability was estimated to be 0.1% or higher by the landslide damage estimation system. The Recall ranged from about 54 to 87% for each earthquake, and the Precision ranged from about 14 to 49% for each earthquake. Comparing Method 1 and Method 2, it was confirmed that Method 1 tended to have a higher recall and Method 2 had a higher precision.
The liquefaction damage estimation data was for the Tottori, Kumamoto, Iburi and Noto earthquakes were generated by J-RISQ. The liquefaction mesh data developed by Senna et al. (2021,2024) were used as the ground truth. For the estimation results, liquefaction was considered to be present when the probability of liquefaction was greater than 1%. As a result, the results show that the mesh of the Iburi earthquake has a Recall of about 48%, but the Recall of the other earthquakes is about 81-92%, all of which indicate that the liquefaction mesh is well predicted. On the other hand, the Precision is about 4-13% but the Precision can be improved by increasing the liquefaction threshold value. Although the comparison of methods A and B showed different results for each earthquake, the Recall exceeded 80% for both methods, suggesting that both methods are effective in identifying areas with a high possibility of liquefaction.
The developed landslide and liquefaction damage estimation system is in operation at the NIED, and the Web API has been developed to enable external data acquisition, and preparations are underway to make the system available to disaster response agencies and others on a limited basis.
(Acknowledgement) Part of this research was conducted under the “Strategic Innovation Program (SIP)” project entitled “Subproject A: Instantaneous and wide-area understanding and sharing of disaster information”.
The landslide damage estimation system converts the maximum velocity distribution obtained from J-RISQ into acceleration using the formula by Ooi et al. (2002), and then calculates the collapse probability using the Rokko formula (Method 1) by Uchida et al. In addition to this method, we have added a function using the modified Rokko formula (Method 2) proposed by Kamiya et al. (2012).
The liquefaction damage estimation system uses the maximum velocity distribution obtained from J-RISQ to calculate the probability of liquefaction using the model proposed by Senna et al. (2021). In addition to the microtopography classification in which the parameters are grouped in consideration of elevation, specific height, and water distance (Method A), the method allow the selection of microtopography classification without grouping (Method B).
The estimation results from each system were verified using damage data from earthquakes that have occurred in recent years.
The landslide damage estimation data for the Chuetsu, Kumamoto, Iburi, and Noto earthquakes was developed using J-RISQ. As correct data for landslides, we used the inventory published by the Geospatial Information Authority of Japan (GSI) as the “Data Release Site for SGDAS” (Iwahashi et al., 2022) and the slope failure data from the Noto earthquake (GSI, 2024). In the former case, a collapse was considered to exist when the points were located within a 250 m mesh, without distinguishing between landslide and slope failure. In the latter case, the slope failure data overlapping with 250 m mesh was assumed to be collapsed and used as the ground truth for each. The estimation results were considered to be collapsed when the probability was estimated to be 0.1% or higher by the landslide damage estimation system. The Recall ranged from about 54 to 87% for each earthquake, and the Precision ranged from about 14 to 49% for each earthquake. Comparing Method 1 and Method 2, it was confirmed that Method 1 tended to have a higher recall and Method 2 had a higher precision.
The liquefaction damage estimation data was for the Tottori, Kumamoto, Iburi and Noto earthquakes were generated by J-RISQ. The liquefaction mesh data developed by Senna et al. (2021,2024) were used as the ground truth. For the estimation results, liquefaction was considered to be present when the probability of liquefaction was greater than 1%. As a result, the results show that the mesh of the Iburi earthquake has a Recall of about 48%, but the Recall of the other earthquakes is about 81-92%, all of which indicate that the liquefaction mesh is well predicted. On the other hand, the Precision is about 4-13% but the Precision can be improved by increasing the liquefaction threshold value. Although the comparison of methods A and B showed different results for each earthquake, the Recall exceeded 80% for both methods, suggesting that both methods are effective in identifying areas with a high possibility of liquefaction.
The developed landslide and liquefaction damage estimation system is in operation at the NIED, and the Web API has been developed to enable external data acquisition, and preparations are underway to make the system available to disaster response agencies and others on a limited basis.
(Acknowledgement) Part of this research was conducted under the “Strategic Innovation Program (SIP)” project entitled “Subproject A: Instantaneous and wide-area understanding and sharing of disaster information”.