Global Sensitivity Analysis in a 2D high resolution hydraulic modeling application - Sobol index maps to rank uncertain parameters

Modern technologies, like aerial photogrammetry, are getting used by hydraulic modeler community to include the topographic information in hydraulic models. Such data allow production of 3D datasets of High Resolution (HR) which include classes of tin features in Digital Elevation Models (DEM). Even though this category has a high level of inframetric accuracy, nevertheless errors remain in measurements and hypothesis under the DEM elaboration. Moreover, modeller optimization of spatial discretization in order to balance flood computation time and hydraulic models, impact accuracy. Presented work performed a Global Sensitivity Analysis (GSA), investigating specifically on uncertainties related to the own error of high resolution topographic dataset, and the modeler choices when including topographic data in 2D hydraulic codes. Objective of the approach is to spatially rank influence of identified uncertain parameter related on results variability (Sobol index). A coupling between a 2D hydraulic code (FullSWOF_2D) with a parametric environment (Promethee) has been developed over a High performance Computing (HPC) structure. This settlement allows performing a GSA, going through a Monte Carlo uncertainty propagation step followed by a post treatment step with R environment, to produce Sobol index maps. The study has been performed over a 17.5 km2 area of the Var river using the estimated hydrograph of 1994 flood and HR classified topographic data -average accuracy of 0.3m-. Three uncertain parameters were studied: the measurement error (var. E), the level of details of the above-ground elements in DEM (buildings, sidewalks, etc.) (var. S), and the spatial discretization resolution (grid cell size for regular mesh) (var. R). A stochastic sampling of the results has been performed with a Monte-Carlo approach. Sensitivity index maps have been produced at areas of interest, enhancing the relative weight of each uncertain parameter on variability of calculated overland flow. Results quantify the importance of uncertainty introduced by modeler choices.