A comparison of methodologies used to predict earthquake-induced landslides

The rigid sliding-block analysis introduced by Newmark in 1965 has become a popular method for assessing the stability of slopes during earthquakes. Estimates of sliding displacement calculated using this methodology serve as an index of seismic performance and are used for mapping seismic landslide hazard potential. The original approach of rigorously integrating ground acceleration time-histories to compute estimates of sliding displacement has been replaced by the use of simple, empirical models that predict displacement as a function of a slope's yield acceleration and one or more measures of ground shaking. To be useful the results of these models must be compared with observations of landslides from previous earthquakes. Seven different empirical models were evaluated by comparing predicted displacements with an inventory of observed landslides from the 1994 Northridge, California earthquake. Using a comprehensive set of ground motion data and shear strength properties from the Northridge earthquake, sliding displacements were calculated within a geographic information system (GIS) and the accuracy of each model was computed. The influence of factors such as landslide size, geologic unit, slope angle, and material strength on the prediction of landslides was also evaluated. The results were used to show that the accuracy of the predictive models depends less on the model used and more on the uncertainty in the model parameters, specifically the assigned shear strength values. Because current approaches do not take into account the spatial variability of strength within individual geologic units, the accuracy of the predictive models is controlled by the distribution of slope angles within observed and predicted landslide cells. Assigning overly conservative (low) shear strength values results in a higher percentage of landslides accurately identified, but also results in a large over-estimation of the seismic landslide hazard.