SPATIAL CORRELATION OF MODELED GROUND SHAKING AND TRIGGERED LANDSLIDES: A CASE STUDY FROM THE 2006 KIHOLO BAY, HAWAII EARTHQUAKE

We compared modeled strong ground motion from the 2006 Kiholo Bay, Hawaii earthquake (M 6.7) to the distribution of triggered landslides to determine if spatial variation in ground shaking correlated with observed landslide locations. The earthquake triggered a high concentration of falls and slides in rock and soil in the Kohala Mountains on the northern coast of the island of Hawaii; the landslides were mapped at 1:24,00 scale using 1:12,000-scale color aerial photography taken after the earthquake. The landslides occurred in steep-walled canyons in an area of uniform lithology; therefore we hypothesized that landslide concentrations should be strongly controlled by spatial variation in strong shaking. We performed a 3D finite-element analysis based on a finite-fault slip model to simulate earthquake strong motion to compare with the landslide distribution. Throughout the area affected by landslides, we simulated velocity and dynamic shear strain with both 3 Hz and 5 Hz upper frequencies. The highest velocities were associated with ridge tops and convexities within the canyons, but the highest shear strains occurred within the canyons where most of the landslides occurred. Maximum velocities at both 3 Hz and 5 Hz did a good job of predicting topographic amplifications on ridge tops and other convexities but were rather poor predictors of landslide locations. One possible reason for this is that the shallow, disrupted landslides that were triggered could be sensitive to higher frequencies (5-10 Hz) of ground shaking that were impractical to model. High levels of dynamic shear strain correlated much more closely with the landslide distribution, which suggests that dynamic shear strain is a good measure of slope deformation during earthquakes and therefore might correlate well with the process of co-seismic landslide initiation on steep slopes. Modeling from other well-recorded earthquakes is needed to confirm this result.