Author(s): Lucie Pheulpin; Antonin Migaud; Nathalie Bertrand

Linked Author(s): Lucie Pheulpin, Nathalie Bertrand

Keywords: Uncertainty quantification; Global sensitivity analysis; Copula; Metamodel; Levee breaches

Abstract: To assess flooding risk, hydraulic models are used. However, these models have many uncertainties related to the lack of knowledge of input parameters (e.g. hydrograph parameters, levee breach parameters). To quantify the uncertainties and to evaluate the most influential parameters of the model, studies referred to as Uncertainty Quantification (UQ) and Global Sensitivity Analysis (GSA) can be used. In order to implement these methods, model inputs must be independent, which is not often the case. In addition, these kinds of methods require numerous simulations. Therefore, the purpose of this research is to propose a methodology to deal with dependent inputs in UQ and GSA studies for hydraulic models and to reduce the computational times. The proposed methodology is applied to the 2D hydraulic model of the Loire River (between Gien and Jargeau, in France) built with TELEMAC-2D. This river section is surrounded by levees where numerous historical breaches occurred. The study methodology is carried out in the following steps: first, the uncertain model inputs (hydrograph parameters, roughness coefficients and breach parameters) and the outputs of interest (water level at some points of the floodplain area) are set. Then, the inputs margins and the dependencies are defined by a statistical analysis using a real dataset (observed hydrograph and breach parameters) of the Loire River. In addition, the dependency structure between the inputs is represented by copulas. Since UQ and GSA require many simulations, kriging metamodels are built to increase the number of experiments in a short time period. Finally, UQ and GSA are carried out by considering the dependent inputs on one hand and the independent inputs on the other hand. With respect to the UQ, the outputs distribution slightly differs if the inputs are considered independent or not. Regarding to the GSA, the number of influencing parameters increases when inputs are considered dependent. This proves that some parameters, usually considered as not influencing, may actually be significantly impacting on the outputs. Although the differences are not considerable, these results suggest that dependencies should not be overlooked in 2D hydraulic models.

DOI: https://doi.org/10.3850/IAHR-39WC2521711920221433

Year: 2022