Downscaling of Precipitation in the Upper Danube Catchment Area

This work has been carried out in the framework of the project
GLOWA-Danube (GLObal WAter cycle) where a joint effort is made by
several groups to model the interaction of the water cycle and
society in the Upper Danube catchment area. In particular regional
climate models are used to simulate and eventually predict
precipitation in this research area, while other groups convert
this information into river runoff estimates and groundwater
fluxes. It has been agreed in the project that precipitation data
and other meteorological data must be handed over to the
hydrological groups with a spatial resolution of 1 km. Long term
runs with regional climate models are, however, not feasible at 1
km resolution, because they would exceed available computer
resources by far. Therefore, a pragmatic downscaling method for
precipitation must be implemented which provides data of 1 km
resolution on the basis of model results of fairly coarse
resolution. This downscaling uses extensively climatological
precipitation observations where such downscaling relations can be
derived. These observed rates are then adapted to the model
results. The data are provided by the German Weather Service (DWD)
and the Austrian Weather Service (ZAMG). The years 1991-2000 have
been chosen as a reference period for the analysis. The climate
simulation is carried out by the mesoscale model MM5 at a
resolution of 45 km. The model MM5 offers a wide range of
parameterizations with respect to convective processes, the
boundary layer, cloud microphysics, and the radiation balance, all
directly or indirectly responsible for generating precipitation.
Sensitivity studies are performed to find the best configuration
for the research area and reference period. A variety of methods is
tested to generate observed and simulated climatological time
series of precipitation. In particular, a linear average, a running
average, a Fourier analysis, and spline interpolation are
intercompared. In the end, spline interpolation between monthly
values showed the best results for both time series and is used as
a basis for the downscaling method. The downscaling method has to
correct two major discrepancies between the observed precipitation
distribution at the 1 km resolution and the simulated distribution
at the 45 km resolution. First, these are small scale details
related to topography in the rainfall distribution at the 1 km
resolution, which lack in the 45 km resolution. Second, there is an
unrealistic southward shift of the rainfall maximum at the northern
rim of the Alps in the simulations, which needs to be corrected. A
specific correction factor is introduced for each problem. The
correlation between the spatial distribution of observed and
simulated distributions increases after using the correction
factors. Due to the climatological relationships, the results time
periods of 10 days and longer are superior to those for periods
shorter than 10 days. The precipitation distribution depends, of
course, on the wind direction in particular so near the Alps. Wind
direction and wind speed are simulated by the MM5 model and
combined with the correction factors described above. The
correlation between the spatial distribution of observed and
simulated precipitation increases more if the wind direction
dependent correction factors are introduced. These improved
correction factors depend less on climatological relationships and
perform therefore better for shorter time periods. Additionally,
they will be able to respond better on changes in the weather
regime in future climates. Altogether, this investigation provides
a new pragmatic method to downscale model simulations on the basis
of observations. This method will be used in the project
GLOWA-Danube.