Comparing methods for modeling and detecting flood nonstationarity: an Austrian case study

Abstract:

Changes in flood regimes have a huge impact on mid- and long-term flood management strategies. These strategies are often based on design floods defined by return periods, which could be ill-defined if nonstationarity is present. Therefore it is crucial when analyzing existent flood time series to detect and, where possible, attribute flood nonstationarities to changing hydroclimatic and land-use processes. Moreover, the choice of model used to detect and model nonstationary flood regimes can have a profound impact on design flood estimates.

This work presents the preliminary results of applying different non-stationarity detection methods on annual peak discharge time series at more than 400 gauging stations in Austria. The type of nonstationarities analyzed include trends (linear and non-linear), change points, clustering beyond stochastic randomness, and detection of flood-rich and flood-poor periods. Austria has a wide variety of physiographic regions with distinct landscapes, elevations and climates which allows us to interpret the spatial patterns obtained with the nonstationarity detection methods in terms of the dominant flood generation mechanisms.