Bayesian network for characterizing model uncertainty of liquefaction potential evaluation models

Knowledge about model error or model uncertainty is essential for liquefaction analysis. Model uncertainty characterization is generally not easy due to the presence of a large number of uncertain model input parameters. The Bayesian network is a versatile tool for analyzing problems involving a large number of uncertain variables. In this paper, a Bayesian network is developed to determine the model uncertainty of liquefaction evaluation models considering the parameter uncertainties. An approximate variable elimination algorithm is suggested to reduce the computational work in model uncertainty characterization. A weighted likelihood function is used to consider the sampling bias in the calibration database. The model uncertainty of a liquefaction model is studied to illustrate the proposed method. It is found that the model is on average biased towards the conservative side. Ignoring the model uncertainty is a convenient assumption, but it may result in either overestimation or underestimation of the reliability index.