Resumen
The issue of regional design flood composition should be considered when it comes to the analysis of multiple sections. However, the uncertainty accompanied in the process of regional design flood composition point identification is often overlooked in the literature. The purpose of this paper, therefore, is to uncover the sensibility of marginal distribution selection and the impact of sampling uncertainty caused by the limited records on two copula-based conditional regional design flood composition methods, i.e., the conditional expectation regional design flood composition (CEC) method and the conditional most likely regional design flood composition (CMLC) method, which are developed to derive the combinations of maximum 30-day flood volumes at the two sub-basins above Bengbu hydrological station for given univariate return periods. An experiment combing different marginal distributions was conducted to explore the former uncertainty source, while a conditional copula-based parametric bootstrapping (CC-PB) procedure together with five metrics (i.e., horizontal standard deviation, vertical standard deviation, area of 25%, 50%, 75% BCIs (bivariate confidence intervals)) were designed and employed subsequently to evaluate the latter uncertainty source. The results indicated that the CEC and CMLC point identification was closely bound up with the different combinations of univariate distributions in spite of the comparatively tiny difference of the fitting performances of seven candidate univariate distributions, and was greatly affected by the sampling uncertainty due to the limited observations, which should arouse critical attention. Both of the analyzed sources of uncertainty increased with the growing T (univariate return period). As for the comparison of the two proposed methods, it seemed that the uncertainty due to the marginal selection had a slight larger impact on the CEC scheme than the CMLC scheme; but in terms of sampling uncertainty, the CMLC method performed slightly stable for large floods, while when considering moderate and small floods, the CEC method performed better.