Multivariate flood frequency analysis, involving flood peak flow, volume and duration, has been traditionally accomplished by employing available functional bivariate and multivariate frequency distributions that have a restriction on the marginals to be from the same family of distributions. The copula concept overcomes this restriction by allowing a combination of arbitrarily chosen marginal types. It also provides a wider choice of admissible dependence structure as compared to the conventional approach. The availability of a vast variety of copula types makes the selection of an appropriate copula family for different hydrological applications a non-trivial task. Graphical and analytic goodness-of-fit tests for testing the suitability of copulas are beginning to evolve and are being developed; there is limited experience of their usage at present, especially in the hydrological field. This paper provides a step-wise procedure for copula selection and illustrates its application to bivariate flood frequency analysis, involving flood peak flow and volume data. Several graphical procedures, tail dependence characteristics, and formal goodness-of-fit tests involving a parametric bootstrap-based technique are considered while investigating the relative applicability of six copula families. The Clayton copula has been identified as a valid model for the particular flood peak flow and volume data set considered in the study.
Keywords: bivariate flood frequency, Chi-plot, Clayton copula, copula, generating functions, K-plot