Bias correction and resampling of RACMO output for the hydrological modelling of the Rhine

Extreme discharges of the Rhine are likely to change as a result of the changing climate. A common way to assess impacts of climate change is to use Regional Climate Model (RCM) output to drive impact models. For the assessment of very rare discharge events in a large river basin (e.g. with return periods of 1250 years) there are two major problems. First, available RCM simulations are usually way too short for the robust estimation of such rare events. Second, RCM output is generally too biased for direct use in impact models.

Nearest Neighbour Resampling (NRR) stochastically extends meteorological time series to any length. The generated synthetic time series are subject to the same characteristics as the reference time series and are generally thought to contain rare multi-day extremes in accordance with the time series length. The discharge of the Rhine at Lobith is closely related to the upstream precipitation of multiple preceding days. So, hydrological modelling on the basis of very long synthetic time series may result in more robust estimation of very rare discharge events at Lobith.

The RACMO simulation used for this study was nested in a simulation with the Global Circulation Model ECHAM5 forced by the SRES emission scenario A1B. The simulation is biased with respect to a reference observational data set with precipitation and temperature. Too high spatial and temporal coherency of the rainfall events result in too many wet days. A wet-day adjustment is developed that leaves the Probability Density Function (PDF) of wet-day amounts largely intact and slightly reduces the spatial and temporal coherency. These daily, local scale adjustments also improve the large-scale and multi-day variability. Yet, an additional power-law correction is necessary to efficiently reduce the remaining biases in average precipitation and the coefficient of variability (CV). For the daily temperature a shifting and scaling are sufficient to correct for biases in average temperature and standard deviation.

After the successive application of the Nearest Neighbour Resampling and the bias correction (BC) still some small biases remain. Yet, rare large-scale multi-day rainfall events in the bias-corrected RCM output are very well reproduced compared to the results of the synthetic time series based on historical data. This justifies the application of the generated and corrected time series for hydrological modelling and assessment of
extreme discharges at Lobith.

Rare 10-day large-scale precipitation events upstream of Lobith with return periods between 10 and 1250 years increase 7% to 10% according the RACMO simulation and the proposed methods. Rare events with return periods of 1250 years in the current climate will have a 3 to 4 times higher occurrence probability around 2050.