Résumé
Annualised average damage (AAD) is a widely used indicator, both in research and in operational use, for two purposes: the evaluation of the flood exposure of a territory and the estimation of the effectiveness of flood prevention policies. The AAD synthesizes rich information resulting from the combination of hydrological (relationship between the rarity and intensity of events), hydraulic (spatial extent and intensity of floods), geographical (location and characteristic of stakes), and vulnerability (potential damage) modelling. By construction, the ADD allows to follow the evolution of hydrology or land use, whether they are due to the evolution of the climate, of the society or to flood prevention policies. As hydraulic modelling is costly to calibrate, in practice, the AAD is usually estimated on the basis of a set of specific flood scenarios.The objective of our presentation is to discuss the influence of the choice of these scenarios (flood sampling) according to the expected use of the AAD (exposure diagnosis vs. project effectiveness). To do so, we build a digital experiment that mime the sampling of floods encountered in practice while keeping full control of the key parameters in the estimation of the AAD. This digital experiment is made up of a stochastic and parametric generator of flood scenarios (hydrograms at floodplain inlets), a Saint-Venant 1.5D-network hydraulic model, whose spatial representation directly arises from a digital terrain model, and a collection of spatially arranged stakes, whose vulnerability is represented in the form of multivariate damage functions (height and duration of submersion). This controlled digital experiment allows the evaluation of various types of policies, alone or combined: diking system, upstream reservoir, adaptation or relocation of the stakes, in a stationary or non-stationary climatic context.Based on the simulation of a 10,000-year chronicle of flood events, we calculate a reference AAD (empirical average as an unbiased estimator of AAD mathematical expectation). This reference allows us to discuss the accuracy of the estimation of the AAD from a set of flood scenarios (sampling effect), and ultimately, the strategies adopted for the choice of flood scenarios (sampling design).