Estimation of future discharges of the River Rhine in the SWURVE project

The change in the flow regime of the river Rhine by the end of the 21st century was one of the eight case studies in the European project SWURVE (Sustainable Water: Uncertainty, Risk and Vulnerability in Europe). Both the potential changes in high river flows (flood protection) and low river flows (inland navigation) were of interest. The river flow simulations were done with the RhineFlow model, a distributed water balance model of the Rhine basin with a temporal resolution of 10 days and a spatial resolution of 3 km × 3 km. For SWURVE the model was recalibrated with a larger meteorological dataset than used in earlier studies of climate change impacts on the discharge of the river Rhine.

The output of the Hadley Centre regional climate models HadRM2 and HadRM3H was extensively used in the SWURVE project. These regional models cover nearly the whole of Europe and part of the Atlantic Ocean with a grid resolution of 50 km × 50 km, and are driven at their boundaries by a global model. For the older HadRM2 model two simulation runs were available: a 30-year control run, representing the climate of the second half of the 20th century, and a 20-year anomaly run, representing the climate for the period 2080 – 2099. A 1% increase in equivalent CO2 after 1989 was assumed. Sulphate aerosol forcing was not included. The HadRM3H simulations were performed with SRES emission scenarios. For the A2 emission scenario three simulation runs were available for the period 2070-2099. There were also three simulation runs for the period 1961-1990 (control climate).

The precipitation and temperature biases in the control simulation are of similar magnitude for the HadRM2 and HadRM3H simulations (about 1 °C for the seasonal mean temperature and up to 40% for the seasonal mean precipitation over the basin). Both models show an increase of 4.5 °C in the basin-averaged annual mean temperature at the end of the 21st century. The mean precipitation increases in winter and decreases in summer. The decrease in summer precipitation in the HadRM3H simulations is as large as 40%. In both the HadRM2 and HadRM3H simulations, the decrease in mean summer precipitation is accompanied by a significant increase in the coefficient of variation (CV: standard deviation divided by the mean) of the 10-day precipitation totals. The HadRM2 simulations also show a significant increase in the CV in the winter season, but the CV decreases in winter in the HadRM3H simulations. The latter is accompanied by a relatively small increase in the largest quantiles of the 10-day precipitation distribution in winter.

Different scenario time series for application in the RhineFlow model were produced from the regional climate model output. The observed data were perturbed in two different ways with the seasonal mean changes in the HadRM2 experiment. For the HadRM3H simulations a simple perturbation of the data for present-day conditions was compared with the direct use of the climate model output in RhineFlow. The HadRM2 scenario 1 was obtained by perturbing the observed precipitation and temperature data with the seasonal mean changes in the HadRM2 experiment. The change in potential evaporation was based on an empirical relation between the changes in open water evaporation and temperature. The CV of the 10-day precipitation totals remains unchanged in this simple scenario. A more advanced scenario was produced by adjusting the standard deviations of the observed precipitation and temperature data as well (HadRM2 scenario 2). Unlike the HadRM2 scenario 1, the seasonal changes in the mean precipitation and temperature in the HadRM3H simulations were applied to the bias-corrected control runs rather than to the observed data (HadRM3H scenario 1). The bias-corrected model output was also directly used as input into the RhineFlow model (HadRM3H scenario 2). The estimated potential evaporation from the HadRM3H output could not be used in these scenarios. The dependence between this potential evaporation and the simulated soil moisture by RhineFlow appeared to be too strong, resulting in a bias in the simulated summer flows. This bias could be suppressed by deriving potential evaporation from temperature only.

All four scenarios result in an increase in the mean winter discharge at Lobith (German – Netherlands border) of 20-30%. The mean summer discharge decreases by 30% in the HadRM2 scenarios and by 40% in the HadRM3H scenarios. The realism of the latter could be questioned because of indications of a too strong hydrological feedback in the HadRM3H control climate, i.e. dry conditions (and also wet conditions) tend to be too persistent. The changes in the seasonal mean flows are accompanied by more extreme flood peaks and an increased frequency of low flows.

In contrast to the change in the seasonal mean flows, the changes in the annual maximum flows are very sensitive to the method of scenario construction. The relative increase of the 1000-year event is for the HadRM2 scenario 2 more than twice as large as that for the HadRM2 scenario 1, mainly because the latter did not account for the increase in the CV of the 10-day precipitation totals in winter. The direct use of the HadRM3H output (HadRM3H scenario 2) resulted in a much smaller increase in the 1000-year event than the perturbation of the control run (HadRM3H scenario 1). The HadRM3H scenario 2 should be considered as the more appropriate one because it includes the changes in the variability and the shape of the distributions of the meteorological inputs as well as the changes in their correlations. The relative increase of about 10% of the 1000-year event in the HadRM3H scenario 2 is small compared to the relative increase of 37% in the most appropriate scenario (HadRM2 scenario 2) from the HadRM2 simulations. An important factor causing these differences is that the CV of the 10-day precipitation totals in winter increases in the HadRM2 simulations and decreases in the HadRM3H simulations. Apart from the large climate-model dependence of the change in the 1000-year event, there are additional uncertainties due to the limited length of the scenario series (35 years for the HadRM2 scenarios and 90 years for the HadRM3H scenarios) and the extrapolation beyond present-day conditions.

The change in the frequency of low flows is much less sensitive to the method of scenario construction than the change in the annual maximum distribution. The results depend, however, on the regional climate model used. The proportion of the 10-day periods that the Rhine discharge is below the present-day 5th percentile roughly doubles in the HadRM2 scenarios and quadruples in the HadRM3H scenarios. For the latter scenarios the additional transport costs due to drought increase from about 80 million euro a year for present-day conditions to about 500 million euro a year, assuming no changes in the number of vessels and in the vessel types. This cost estimate should be used with care because of the low confidence in the summer drying as simulated by the HadRM3H model and because of systematic errors in the RhineFlow simulations of low flows.