Statistical seismic landslide hazard analysis: An example from Taiwan

Following the work of Lee et al. (2008a), a statistical approach is applied to seismic landslide hazard analysis for the whole of Taiwan. All the work is done using new data sets, which include a new and carefully mapped Chi-Chi earthquake-induced landslide inventory, a 5-m DEM, and a new version of the 1:50,000-scale geologic map of Taiwan. Landslide causative factors used in the susceptibility analysis include the slope gradient, slope aspect, terrain roughness, slope roughness, total curvature, total slope height, and lithology. The corrected Arias intensity taking topographic amplification into consideration is used as a triggering factor.Firstly, a susceptibility model is built using the 1999 Chi-Chi shallow landslides as a training data set and multivariate logistic regression as the analytical tool. This model is validated by using the 1998 Jueili earthquake-induced landslide data. Then, a probability-of-failure curve is established by comparing the Chi-Chi landslide data and the susceptibility values, after which the spatial probability of landslide occurrence is drawn. The temporal probability may be accounted for with the triggering factor (the hazard level of the Arias intensity), which was obtained through regular probabilistic seismic hazard analysis. Finally, the susceptibility model and the probability-of-failure curve are applied to the whole of Taiwan using the topographically corrected 475-year Arias intensity as a triggering factor to complete a seismic shallow-landslide probability map for ground-motions having a 475-year return period.