Tracing Uncertainty Contributors in the Multi‐Hazard Risk Analysis for Compound Extremes

Abstract In this study, an iterative factorial multimodel Bayesian copula (IFMBC) framework was developed for revealing uncertainties in risk inferences of compound extremes under the consideration of diverse model structures and parameter sets. Particularly, an iterative factorial analysis (IFA) method would be advanced in IFMBC to track the dominant contributors to the imprecise predictions of multi‐hazard risks. The developed IFMBC framework was applied for the risk assessment of compound floods at two estuarine systems (i.e., Washington and Philadelphia) in US. The results indicate that the most likely compound events, under predefined return periods, exhibit noticeable uncertainties. Those uncertainties also present multiple hotspots which may be attributed to different impacts from different factors. By applying the IFA method, the results suggest the copula structure would likely be ranked as one of the top 2 impact factors for predictions of failure probabilities (FPs) in the scenarios of AND, and Kendall, with its contributions higher than 30% for FP in Kendall (more than 40% at Washington) and more than 25% for FP in Kendall (larger than 40% at Philadelphia). In comparison, the copula structure may not pose a visible effect on the predictive uncertainty for FP in OR, with its contribution possibly less than 5% under long‐term service time periods. However, the marginal distributions would have higher effects on FP in OR than the effects on the other two FPs. Particularly, the marginal distribution for the extreme variable with high skewness and kurtosis values tends to be ranked as one of the most significant impact factors for FP in OR. Also, the overall impacts from parametric uncertainties in both marginal and dependence models cannot be neglected for the predictions of all three FPs with their contributions probably larger than 20% under a short service time period. Compared with the traditional multilevel factorial analysis, the IFA method can provide more reliable characterization for uncertainty contributors in multi‐hazard risk analyses, since the traditional method seems to significantly overestimate the contributions from parameter uncertainties.