Probabilistic seismic landslide hazard maps including epistemic uncertainty

Abstract

Seismic landslide hazard maps are used in regional planning to identify zones that require detailed, site-specific studies. Most seismic landslide hazard maps are based on predicted sliding block displacements for a given level of shaking, but this deterministic approach does not fully consider the aleatory variability in the sliding displacement prediction or the epistemic uncertainty in the soil and slope properties. Probabilistic approaches incorporate aleatory variability through a sliding displacement hazard curve and they incorporate epistemic uncertainties (e.g., soil shear strength, empirical displacement model) through a logic-tree approach. This fully probabilistic approach can be implemented efficiently on a regional scale for seismic landslide hazard maps using ground motion hazard data and the Mean λD Threshold approach. This paper applies the probabilistic approach to develop a seismic landslide hazard map for Anchorage, Alaska. The results show that incorporating epistemic uncertainty can increase the area of high seismic landslide hazard by a factor of 2 to 3 as compared to analyses without any epistemic uncertainty. Additionally, incorporating a logic-tree avoids using overly conservative input parameters in a deterministic approach to capture these uncertainties, which can lead to an unrealistic inflation of the seismic landslide hazard.