Bayesian assessment of modeling uncertainties application to fire risk assessment

Uncertainties in the prediction of models can be important contributors to the uncertainties in probabilistic risk assessment results, yet are usually not treated. This paper discusses the assessment of uncertainties associated with a simple model when large amounts of benchmarking data are not available, and applies a Bayesian approach for quantification. The application is to a problem of interest in fire risk assessment: estimation of the velocity of flame spread over a horizontal cable tray. The uncertainties in the available predictive model are formally assessed and possible reasons for the deviation between predictions and observations are discussed