5:15 PM - 6:30 PM
[HTT19-P01] Susceptibility assessment of post-seismic landslides with an index representative of seismic cracks
Keywords:post-seismic landslides, seismic cracks, landslide susceptibility assessment, WoE, Logistic regression
An intense earthquake can cause numerous landslides over a broad area, causing damage to human lives, property, and infrastructure. The strong tremors weaken the ground strength and induce cracks on the ground surface. Thus, following the earthquake, identification of the ground cracks is rapidly carried out via field survey and/or examining aerial photographs or topographic maps to find slopes in danger of sliding and prevent from further disaster. However, this operation requires more time and effort as the magnitude of the earthquake is more extensive. Hence, we proposed an index (DCI: Dense Crack Index) representative of the density of co-seismic cracks, which can be quickly and easily calculated using airborne LiDAR data before and after the earthquake. This study applied the index to an area (6 km2) located in the southwestern Aso Caldera, Kumamoto Prefecture, where the 2016 Kumamoto earthquake (Mw 7.0) struck, and the 198 co-seismic and 125 post-seismic slides were recognized in the following four months. They were categorized as shallow landslides. Then, we statistically estimated the susceptibility of post-seismic landslides and examined the improvement of the evaluation accuracy when DCI is incorporated in predictor variables.
First, we calculated the slope angle for each 1-m cell derived from LiDAR survey data acquired before and after the earthquake (Fig. 1-a), and then its standard deviation for 3 × 3 cells. We found that the standard deviation increased by more than 2 degrees after the earthquake in 75% of the locations where co-seismic cracks were identified (Fig. 1-b). The DCI was given as the spatial density of those cells, calculated by the kernel density with a nearby distance of 10 meters (Fig. 1-c).
Weight of Evidence (WoE) and Logistic Regression (LR) methods were applied to estimate the susceptibility for post-seismic slides. The response variable used in these two statistical models was the presence or absence of post-earthquake landslides on the cell, which was identified from aerial photographs and topographic maps obtained four months after the earthquake. Also, the slope angle, profile curvature, slope aspect, distance to edge of co-seismic landslides, distance to epicenter fault, and peak ground acceleration (PGA) were adopted as predictor variables in addition to DCI. This study considered the geology, vegetation, and rainfall conditions as the same due to the small size of the study area.
The analysis by the WoE method showed that steep slopes (slope: 40 - 55 degrees) near knick lines (profile curvature: -6 - -4 m-1) affected by strong tremors (PGA: 1050 - 1150 gal) with dense crack appearance (DCI: 0.2 - 1.0) were more likely to slide after the earthquake. This result is in agreement with the report of the field survey conducted in the study area (Kumamoto Prefecture, 2019). The accuracy of the susceptibility assessment by the WoE and LR methods was slightly higher when DCI was included as a predictor variable, with an Area Under the Curve (AUC) value of 0.80 - 0.90 for the ROC curves, compared to the models without DCI (0.70 - 0.85).
As such, the inclusion of DCI in statistical susceptibility assessment improved the accuracy of the models and allowed the identification of slopes in danger of sliding that were consistent with those reported from the field. These results suggest that DCI is an important index for estimating slopes destabilized by seismic motion.
Fig.1:Step of DCI calculation (a: Slope maps of pre- and post- earthquake, b: red indicates the locations where the standard deviation of the slope increased by more than 2 after the earthquake , c: DCI map)
First, we calculated the slope angle for each 1-m cell derived from LiDAR survey data acquired before and after the earthquake (Fig. 1-a), and then its standard deviation for 3 × 3 cells. We found that the standard deviation increased by more than 2 degrees after the earthquake in 75% of the locations where co-seismic cracks were identified (Fig. 1-b). The DCI was given as the spatial density of those cells, calculated by the kernel density with a nearby distance of 10 meters (Fig. 1-c).
Weight of Evidence (WoE) and Logistic Regression (LR) methods were applied to estimate the susceptibility for post-seismic slides. The response variable used in these two statistical models was the presence or absence of post-earthquake landslides on the cell, which was identified from aerial photographs and topographic maps obtained four months after the earthquake. Also, the slope angle, profile curvature, slope aspect, distance to edge of co-seismic landslides, distance to epicenter fault, and peak ground acceleration (PGA) were adopted as predictor variables in addition to DCI. This study considered the geology, vegetation, and rainfall conditions as the same due to the small size of the study area.
The analysis by the WoE method showed that steep slopes (slope: 40 - 55 degrees) near knick lines (profile curvature: -6 - -4 m-1) affected by strong tremors (PGA: 1050 - 1150 gal) with dense crack appearance (DCI: 0.2 - 1.0) were more likely to slide after the earthquake. This result is in agreement with the report of the field survey conducted in the study area (Kumamoto Prefecture, 2019). The accuracy of the susceptibility assessment by the WoE and LR methods was slightly higher when DCI was included as a predictor variable, with an Area Under the Curve (AUC) value of 0.80 - 0.90 for the ROC curves, compared to the models without DCI (0.70 - 0.85).
As such, the inclusion of DCI in statistical susceptibility assessment improved the accuracy of the models and allowed the identification of slopes in danger of sliding that were consistent with those reported from the field. These results suggest that DCI is an important index for estimating slopes destabilized by seismic motion.
Fig.1:Step of DCI calculation (a: Slope maps of pre- and post- earthquake, b: red indicates the locations where the standard deviation of the slope increased by more than 2 after the earthquake , c: DCI map)