Regional earthquake-induced landslide assessments using a data-informed probabilistic approach

The seismically active state of California is characterized by the potential for permanent ground displacements of slopes, such as those in the form of landslides, during earthquake events. The adverse impacts of these landslides on the built environment, infrastructure, nature, and human lives are significant. It is of critical importance in geotechnical engineering to adequately assess, mitigate and/or minimize these impacts for effective seismic landslide risk management. In California, several thousands of kilometers of spatially distributed lifeline infrastructure, such as natural gas pipelines, traverse regions that are susceptible to earthquake-induced landslides, creating a need for a state-wide seismic risk assessment of natural gas pipelines. These assessments require the spatial distributions of earthquake-induced landslides within these regions as well as the characteristics of the landslides (such as displacement amount, size, and direction of movement) as inputs into pipeline fragility models. To accurately estimate displacements on a regional scale, there is a need for adequate characterization of shear strength properties of geologic units, slope properties and groundwater table. The first goal of this dissertation is to develop a data-informed, statistical approach to characterizing shear strength parameters for geologic units across California. The approach involves jointly analyzing geotechnical data from the California Geological Survey (CGS) borehole database and a California statewide geologic map, while incorporating professional judgment derived in consultation with geologists/geotechnical engineers in California. The second goal is to compute seismic landslide displacements on a regional scale and use this information to define seismic landslide zones along with the attributes of size, movement, and direction. The geologically assigned strength parameters are used in conjunction with slope properties and the groundwater table depth to estimate yield acceleration and the yield acceleration and ground shaking are used to compute sliding displacements at a 10-m grid resolution across the state. Due to the state-wide scale of the analysis, epistemic uncertainties associated with the input parameters, as well as those associated with the empirical displacement models, are incorporated using a logic tree approach. The grid-level values of sliding displacement are used to identify landslide zones and to compute probability distributions of the amount of movement, the landslide size, and the direction of landslide movement that can be used for regional-scale pipeline risk assessments. A new framework for defining landslide zones that involves integrating geomorphic landforms with sliding displacements to delineate boundaries for individual landslide zones is developed. An illustration of the application of the framework is also presented for a small study area in Southern California. The final objective of the work is to evaluate the predictive performance of the framework using landslide inventories from four historical earthquakes events. These analyses demonstrated that the seismic landslide framework captures a large percentage of observed landslides (63% to 95% across the four earthquake events), but that there is the potential for significant overprediction. This overprediction is most significant when the event-specific conditions (e.g., groundwater table) are not consistent with the values used in the logic tree. The seismic landslide predictions can be improved by incorporating more localized shear strength, slope properties, and groundwater table data into the analyses where available to minimize the degree of uncertainties that accompanies regional-scale assessments.