Markhvida, M; Cremen, G; Grujic, O; Baker, J; (2021) Methods for evaluation and treatment of epistemic uncertainty in portfolio losses due to earthquakes. In: Proceedings of the 17th World Conference on Earthquake Engineering, 17WCEE. IAEE: Sendai, Japan.

Preview

Text Markhvida_et_al_(2020),_Epistemic,_WCEE (1).pdf - Other Download (544kB) | Preview

Abstract

Assessment of seismic losses in a portfolio of buildings can be a challenging task, since there can be large epistemic uncertainties associated with the different steps of the probabilistic seismic risk analysis: hazard estimation, exposure modeling, fragility functions, and damage-to-loss estimation. Refining models and gathering more data to reduce the epistemic uncertainties can require substantial time investment and incur significant costs; therefore, to make this process more efficient, variables that drive the epistemic uncertainty must be identified. This paper explores the use of two sensitivity analysis methods to evaluate the effect of uncertain variables on the epistemic uncertainty of portfolio losses from earthquakes: 1) a well established variance-based sensitivity analysis technique and 2) a novel method that leverages regression tree ensemble methods with functional outputs. The two methods are examined using a fictional portfolio of 20 buildings in the San Francisco Bay Area. The results from the methods are compared, and advantages and disadvantages of the regression tree ensemble method are highlighted. Also discussed are recommendations for treatment of uncertain input variables based on insights about epistemic uncertainty in the losses.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item