Copula-based analysis of flood peak, volume and duration dependence in Brazil

Código

I-EBHE0171

Autores

Thales Basílio da Circuncisão, Francisco Eustáquio Oliveira e Silva, Ana Clara de Sousa Matos, WILSON DOS SANTOS FERNANDES

Tema

WG 1.10: Hydrologic Design - Solutions & Communication

Resumo

The estimation of streamflow extreme events can be accomplished through a deterministic approach, which employs rainfall-runoff models, or through a probabilistic approach, which is based on frequency analyses. In the probabilistic approach, the literature predominantly focuses on a univariate frequency analysis of streamflow. However, in some hydrological applications, such as reservoir management and flood management, it is necessary to consider not only the peak flow but also the volume, duration, and shape associated with a flood event. Given the aforementioned limitations, it would be prudent to conduct a multivariate frequency analysis of these variables, based on the dependency relationships between them. This requires the evaluation of techniques for modelling the dependency between the random variables governing flood occurrence, structuring the joint probability distribution, and estimating the risk associated with extreme events. This work forms part of a study investigating the derivation of synthetic design hydrographs. It aims to evaluate the potential of asymmetric copula functions for modelling the dependence between random variables related to flood events. We tested copula functions to assess the dependence between flow, volume and duration for flood events observed in 38 Brazilian basins, in the Amazon, Tocantins-Araguaia, São Francisco and Paraná river basins. The basins were selected based on criteria such as drainage area, degree of regularization and temporal discretization of hydrological variable data. Our findings are consistent with those of other researchers who have postulated that copula functions are a valuable tool for the multivariate evaluation of these variables because of their flexibility and the extensive families of functions, which enable the representation of various types of dependencies, including different marginal distributions.