Numerical simulations of the August 2002 flooding event in the Elbe catchment area

On 12-13 August 2002, sustained heavy rainfall with peak amounts of more than 300 mm / 36h in the Erzgebirge mountains led to one of the most catastrophic flooding events Central Europe has ever experienced. The region most heavily struck by the flooding was the catchment area of the Elbe river in eastern Germany, but there were also substantial damages in parts of Austria and the Czech Republic. Although the responsible weather situation was captured fairly well by the operational forecast models, there were large deficits in the quantitative precipitation forecasts. Even the regional-scale 48-hour forecasts greatly underpredicted the rainfall amounts and misplaced the primary rainband by more than 100 km. To get a better insight into the predictability problems of this event, high-resolution numerical simulations have been performed with the Penn State/NCAR mesoscale model MM5. The model is run a four-domain configuration with a finest horizontal resolution of 1 km. Sensitivity experiments are performed with coarser resolutions (3 km and 9 km), with different cloud microphysical parameterizations and with a different date of initialization. Moreover, tests with 1 km resolution but the smoothed topography of the 9-km runs are conducted in order to isolate the contribution of the model topography to the differences between the 1-km runs and the 9-km runs.

The results show that the high-resolution runs reproduce the observed structure of the precipitation field very well. In particular, the location of the rainfall maximum is correct to within 15 km. The quantitative agreement between model results and observations is fairly good in regions with 36h-rainfall accumulations below 100 mm, but the orographic enhancement of the precipitation along the Erzgebirge ridge tends to be underestimated. For observed 36h-rainfall accumulations exceeding 200 mm, the negative bias typically ranges between 15% and 30% with fairly little dependence on the microphysical parameterization. In the coarser-resolved runs, the bias at high precipitation amounts becomes even more pronounced, and it is found that this behavior is not only due to the fact that a coarser resolution is associated with a smoother topography. A possible reason for this behavior could be a systematic deficiency in the microphysical parameterizations, limiting their ability to simulate heavy orographic precipitation. Another possible contribution to the resolution-dependence of the simulated precipitation amounts might arise from the fact that the convection parameterization used in the 9-km runs does not capture the effect of embedded convective cells.