Applying single- and multi-site weather generators for the estimation of extreme floods in the Austrian Alps

Abstract

The length and quality of observed discharge data is frequently insufficient to derive robust extreme discharge values, which are crucial for water management planning for the present and future climates. Using long timeseries of synthetic weather data generated by stochastic weather generators (SWGs) as input to rainfall-runoff models to derive extreme discharge peaks could be of value for basins with missing or short observation timeseries. While multi-site generators preserve spatial correlation between stations, their complexity also limits them in other ways, such as implementing only simple parametric distributions that don’t adequately represent tails of distributions and extreme events. Single-site generators offer more complex parametric distributions, but don’t preserve correlation between stations, making them unsuitable for distributed hydrological modeling. In the context of hydrological modeling with emphasis on extremes, the question arises if the lumped output of a multi-site generator outperforms a single-site generator in combination with a lumped hydrological model, or if the advantages of a heavy-tailed distribution can outweigh the averaging across a heterogenous catchment. This work examines the transfer of synthetic weather data into runoff extremes in the alpine watershed of the Austrian Ybbs River by driving the rainfall runoff model COSERO with meteorological timeseries from the single- and multi-site SWGs WeaGETS and MulGETS. A GEV distribution was fit to each timeseries of annual runoff maxima to derive events with return periods of 30, 100 and 300 years. The single-site generators and their superior parametric distribution functions did not outweigh the averaging effect, and hydrological simulations were greatly biased, significantly underestimating flood peaks. However, also the results of the flood frequency analysis using multi-site synthetic data underestimated the results using observed data for all return periods by at least 30 %. The findings show that the application of single- and multi-site SWGs to estimate runoff extremes may not be applicable and must be critically reviewed.