Representing external hazard initiating events using a Bayesian approach and a generalized extreme value model

Abstract

•This paper details how Bayesian inference can be used to evaluate the Generalized Extreme Value (GEV) model, including how to quantify parameter, and modeling uncertainty. The GEV is potentially useful to represent the occurrence of natural hazards as an initiating event in computational risk analysis.•This paper demonstrates how to incorporate “inferred information†such as paleo-data (data collected before the time of continuous records or direct measurements) with an observed set of data.•This paper demonstrates an approach to probabilistic weighting of data in order to emphasize different parts of an observed data set.•This paper incorporates physical limits into the statistical-based GEV modeling approach.•This paper includes trending in order to account for medium- to long-term variations such as climate change.The objective of this paper is to demonstrate a Bayesian approach to using Generalized Extreme Value (GEV) models for representing a hazard-magnitude-frequency relationship within a computational risk assessment (CRA) framework. This paper will provide discussion and demonstration of:1How Bayesian inference can be used to evaluate the GEV model, including a detailed treatment of data, parameter, and modeling uncertainty.2Three synthetic examples where the underlying data-producing mechanisms are fully known in order to understand how GEVs can be used to represent the occurrence of extreme events. The three examples used in this paper are (i) a linear-type model, (ii) a logarithmic-type model, and (iii) an exponential-type model.3Incorporation of inferred information such as “paleo data†with an observed data set.4Probabilistic weighting of data in order to emphasize different parts of an observed data set over other less-important data.5Incorporation of physics or physical limits into the statistical-based GEV modeling approach.6Inclusion of trending within the GEV approach in order to account for medium- to long-term variations such as climate change.