Estimating the Probability of Earthquake-Induced Landslides
Abstract
The development of a regionally applicable, predictive model for earthquake-triggered landslides is needed to improve mitigation decisions at the community level. The distribution of landslides triggered by the 1994 Northridge earthquake in the Oat Mountain and Simi Valley quadrangles of southern California provided an inventory of failures against which to evaluate the significance of a variety of physical variables in probabilistic models of static slope stability. Through a cooperative project, the California Division of Mines and Geology provided 10-meter resolution data on elevation, slope angle, coincidence of bedding plane and topographic slope, distribution of pre-Northridge landslides, internal friction angle and cohesive strength of individual geologic units. Hydrologic factors were not evaluated since failures in the study area were dominated by shallow, disrupted landslides in dry materials. Previous studies indicate that 10-meter digital elevation data is required to properly characterize the short, steep slopes on which many earthquake-induced landslides occur. However, to explore the robustness of the model at different spatial resolutions, models were developed at the 10, 50, and 100-meter resolution using classification and regression tree (CART) analysis and logistic regression techniques. Multiple resampling algorithms were tested for each variable in order to observe how resampling affects the statistical properties of each grid, and how relationships between variables within the model change with increasing resolution. Various transformations of the independent variables were used to see which had the strongest relationship with the probability of failure. These transformations were based on deterministic relationships in the factor of safety equation. Preliminary results were similar for all spatial scales. Topographic variables dominate the predictive capability of the models. The distribution of prior landslides and the coincidence of slope with bedding plane had poor predictive capability. The predictive value of cohesive strength and internal friction is still unresolved. Although several strength variables were statistically significant, the explanatory power of the model was little improved over models based solely on topographic variables. Finally, 10-meter resolution digital elevation data was necessary to constrain locally steep slopes. However, regional slopes were also statistically significant, suggesting that coarser resolution models may refine finer resolution models by exposing larger scale controls on the distribution of landslides.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2001
- Bibcode:
- 2001AGUFMNG51A0455M
- Keywords:
-
- 1869 Stochastic processes;
- 6309 Decision making under uncertainty;
- 7223 Seismic hazard assessment and prediction