Bayesian updating in natural hazard risk assessment

M. Graf, K. Nishijima, Michael Havbro Faber

Research output: Contribution to journalJournal articleResearchpeer-review

Abstract

Probabilistic models are typically implemented into risk management systems using whatever relevant information is available prior to the implementation. However, in the course of time more information becomes available and it is of significant practical importance to be able to update the probabilistic models based on the new information. The present paper investigates a Bayesian approach for the updating of probabilistic models in the context of risk management of natural hazards. Bayesian probabilistic networks are proposed to form the basic tool for the probabilistic representation of knowledge and uncertainties. Updating of models is performed by instantiating the variables of the Bayesian probabilistic networks corresponding to observations from events of natural hazards. This approach, however, necessitates that large Bayesian probabilistic networks can be efficiently handled and for that purpose a compact object based representation of Bayesian probabilistic networks is suggested. The proposed methodology is applied to three illustrative examples considering updating of fragility model parameters. It is illustrated how commonly applied techniques for model updating in natural hazards risk assessments may lead to somewhat biased model parameter estimates. Furthermore, it is shown how available information on hazard intensities as well as on damages of structures can be utilised at the same time for the updating of the fragility model parameters in a consistent and efficient way. © Institution of Engineers Australia, 2009.
Original languageEnglish
JournalAustralian Journal of Structural Engineering
Volume9
Issue number1
Pages (from-to)35-44
ISSN1328-7982
Publication statusPublished - 2009
Externally publishedYes

Fingerprint Dive into the research topics of 'Bayesian updating in natural hazard risk assessment'. Together they form a unique fingerprint.

Cite this