Due to recent flood events, risk assessment has become a topic of highest public interest. The definition of endangered or vulnerable areas is based on numerical models of the water flow. The most influential input for such models is the topography provided as a Digital Terrain Model of the Watercourse (DTM-W). For capturing terrain data of inundation areas Airborne Laser Scanning (ALS) has become the prime data source. It combines cost efficiency, high degree of automation, high point density of typically 1-10 points per m(2) and good height accuracy of less than 15cm. For all these reasons ALS is particularly suitable for deriving precise DTMs as basis for Computational Fluid Dynamic (CFD) models. The quality of such models depends crucially on how well vegetation or other off-terrain objects have been removed in the DTM generation process. The task of removing off-terrain points from the ALS measurements is commonly referred to as filtering. Traditional laser scanners only supply range measurements to the reflecting objects and, thus, the filtering process has to rely on geometric criteria. The latest generation of ALS systems record the full backscattered waveform, from which physical quantities like echo width and backscatter cross section can be derived. An advanced filtering technique based on the well established method of robust interpolation is presented exploiting the echo width for a more robust and reliable classification of the point cloud into ground and off-terrain points resulting in a more precise DTM-W. Besides filtering, exact sensor calibration, fine adjustment of ALS-strip data, proper fusion of ALS and additional river bed data as well as elimination of random measurement errors are important issues for generating a precise DTM-W based on the ALS point cloud. The higher DTM resolution provided by modern sensors comes along with an increased amount of data. Thus, a direct use of the high resolution DTM-W as the geometric basis for CFD models is impossible. Currently available mesh generators for CFD models basically focus on physical parameters of the calculation grid like angle criterion, aspect ratio and expansion ratio. The detailed shape of the terrain as provided by modern ALS systems is often neglected. A DTM data reduction approach is presented, considering both the physical aspects mentioned above as well as the preservation of relevant terrain details. The method starts with an initial TIN-approximation of the DTM comprising structure lines and a coarse grid. The TIN is subsequently refined by adding additional grid points until a certain height tolerance is met. A spatially adaptive data density, where terrain parts being sensitive for the CFD model are mapped with more details than parts of minor importance, can be achieved by introducing individual height tolerances in the iterative refinement process. In order to obtain a high quality computation grid the resulting surface approximation is professionally conditioned to meet specific hydraulic requirements. Finally, practical results of CFD models based on different geometry variants are presented and discussed. It will be shown that a very detailed description of the topography can indeed be established in CFD models, resulting in more realistic flow simulations and more precise boundaries of potential flooding areas. An example is shown in Figure 1. [GRAPHICS] .
