Development of a new decision-based framework for risk assessment and management of landslides

Landslide is a complicated multi-hazard event that can turn into a great disaster. Landslide risk assessment and management process involves decision making to find an optimal loss-mitigation approach. A good, representative decision analysis process relies on reasonably capturing all possible scenarios that would happen in a landslide event and assessing the probabilities. The nature of a wide range engineering problems, including landslide risk assessment and management, requires that decisions to be made based on limited information and in the face of extreme uncertainty. Bayesian updating method is a more rational approach to account for extreme events and data irrelevancy. In Bayesian, a prior sample space is assumed, the probabilities are then updated by the likelihood function based on historical data. This procedure makes the Bayesian approach capable of accounting for extreme events and data irrelevancy. The prior sample space has a significant effect on updated probability distribution, especially in the case of rare or limited data. The main motivation of this research is to defensibly account for the uncertain in extreme events without unrealistic and unnecessary assumptions. Therefore, in this research, new decision-based framework is introduced for establishing non-informative prior sample space as a starting point in assessing the probabilities. In addition, this research introduces and applies new methods in the formulation of likelihood function in order to minimize the amount of unnecessary prescribed assumptions in the model. Two major risk management case studies are included in this dissertation: a rockslide in Western Norway, and the landslide in Oso, Washington. The objective of the case studies are to demonstrate the application of the new framework, introduced in this research to establish prior and formulate likelihood, in the real-world landslide risk assessment and management problems. These case studies showed that the framework suggested by this dissertation is a rational and defensible approach to account for the extreme uncertainties. The expected contribution of this research is in the field of risk assessment and management of landslides (or other natural hazards). The Decision Entropy is a theory underdevelopment toward becoming a rational method to establish non-informative prior sample space. The fact that the axioms of Decision Entropy Theory requires equally probable states for preference, degrees of preference, and information about preference of alternatives, makes this theory an impartial and objective approach to establish prior sample space in which no unnecessary assumption is included. Furthermore, axioms of this theory provide the tool to account for extreme events, unknown uncertainties, and irrelevancy of data in the risk analysis. However, the theory needs more development to make it easier to be used in the real-world risk assessment problem. Besides starting with a non-informative prior, the likelihood function should be formulated so that it does not include information more than what really is available. Using probability models that account for the renewal process of event, or accounts for the facts that events are non-stationary and correlated, has a greater advantage toward proper assessment of probabilities. Accounting for in-completeness and irrelevancy of data is also necessary to be considered in the formulation of likelihood function.