Multivariate Design Flood Estimation in Large River Basins under Consideration of Tributary Impacts and Flood Types

Abstract:

Flood events are not subject to a common set of forces. Instead, floods may differ strongly in their genesis and different runoff-generating processes are associated with different flood types. The complexity of design flood estimation increases with spatial dimension of the basin by the superposition of flood waves from tributaries. This requires spatial event analysis. If the peak is regarded as the only relevant feature of a flood, the detailed consideration of the contributions of the subareas and the wave superposition appears to be negligible. If, instead, additional characteristics like volume and duration are considered, flood generation and event typing are to be included in the formation of quasi-homogeneous subsets of events in order to consider a greater differentiation of flood generation conditions in the statistics. In this way, not only the runoff-generating process, considered by flood types, determines shape and peaks of flood waves but also the superposition of floods at confluences of tributaries, especially in large river basins. This is considered here by a volume-based runoff analysis of the contributions of sub-basins and its application in a statistical simulation model of samples of synthetic flood events, characterized by their peaks and volumes and differentiated by the spatial distributions of runoff from tributaries. The statistical distribution of simulated event series shows a high accordance with the observed distribution. Long synthetic samples are derived to apply a multivariate statistic model for flood frequency analysis, which considers the flood type of the confluence gauge as well as of the tributaries and delivers a conditional probability of flood peak and volume given the tributaries. The approach is applied to three large river basins in Europe. It became obvious that the highest floods are driven by other mechanisms than smaller and more frequent flood events. In this way the developed model is able to generate extraordinary flood events which are observed infrequent in observed data. It forms the basis for a new consideration of multivariate design floods.