Consistency of Extreme Flood Estimation Approaches

Abstract

Estimations of low-probability flood events are frequently used to plan infrastructure and to determine the dimensions of flood protection measures. Several well-established methods exist for estimating low-probability floods. However, a global assessment of the consistency of these methods is difficult to achieve because the “true value” of an extreme flood is not observable. A detailed comparison performed on a given case study brings useful information about the statistical and hydrological processes involved in different methods. In the present study, the following three methods of estimating low-probability floods are compared: a purely statistical method (ordinary extreme value statistics), a statistical method based on stochastic rainfall-runoff simulation (SCHADEX method), and a deterministic method (physically based estimation of the probable maximum flood, PMF). These methods are tested for two different Swiss catchments; the results show that the 10,000-year return level flood estimations exceed the PMF estimations by 3% and 18%. The analysis shows that the plausibility of an extreme flood estimation does not only depend on the applied method but also on its ability to represent flood-triggering processes, including precipitation input, spatio-temporal precipitation distribution, and runoff.