Characteristics and influencing factors of flash floods in Bavaria (Germany) – an analysis using a hydrological-hydraulic model ensemble

In recent years, heavy precipitation and flash flood events frequently occurred in Germany. The project HiOS (reference map for surface runoff and flash floods) focusses on the analysis of these events using conceptual lumped precipitation runoff models, distributed raster-based water balance models (LARSIM and WaSiM), as well as a hydrodynamic model internally coupled with infiltration routines (TELEMAC-2D). The objective of our research is to analyze which factors and processes foster flash floods, and how they may be represented in models. We show a comprehensive methodological comparison using simulation results of some events in Bavaria. These do not include erosion and log jam scenarios.

The catchments distributed across whole Bavaria considering a variety of catchment characteristics and varying in size between 1.2 and 164km². All models are driven by 5 minute pseudo-calibrated radar precipitation data of the German Weather Service (YW product), which are available for entire Germany in a 1km² raster. The distributed water balance models are available using high-resolution cell grids. WaSiM uses a regular grid size of 50m, whereas LARSIM is run using 100m cells and an embedded hydrological response unit scheme. All TELEMAC-2D meshes are built with a standard mesh size of 5m in the catchment and 2m in the settled area of interest, while important hydrodynamic structures are resolved more in detail.

We want to highlight the variety of applied hydrological and hydrodynamic model approaches of runoff generation and concentration, whereby both, simple conceptual and complex physical methods are included. Runoff generation processes are represented using the SCS-CN method, a modified Lutz-Südbayern approach, a Xinjiang-bucket model combined with a Green&Ampt infiltration routine, as well as a layer-resolving Richards model. Beyond that, some of these consider silting up and soil crack formation. Runoff concentration processes are assessed by constant translation, Strickler flow time index method, a combination of Williams and Kalinin-Miljukov method, as well as finally with two-dimensionally resolved shallow water equations.

As expected, runoff generation is influenced by land use and soil parametrization. However, the amount of created runoff differs a lot changing the method of simulation. Furthermore, the runoff volume reacts quite sensitive to small changes in the preceding saturation conditions. Runoff concentration is influenced by slope, retention capacity of the flood plain, the network of drainages, as well as the formation of polders by water-crossing structures such as traffic infrastructure. Our results therefore clearly show the individual characteristics of extreme events depending on the catchment properties, which are reflected by the demands concerning the modelling techniques. The findings of this study illustrate the importance of improved radar-derived precipitation observations as well as the need for a spatially distributed and layered soil moisture product to enhance flash flood modelling using hydrological models.