Urban drainage infrastructure has to adapt to challenges such as climate change, but also to trends such as population and settlement development. At the same time, the opportunities offered by digitization can be used to meet these challenges. In this thesis, therefore, two geoinformation-based models are developed, which use the advancing digitization to contribute to the adaptation of urban drainage to demographic and climate change. An important aspect of municipal flood protection are nature-oriented measures for the rainwater management as part of water-sensitive urban development. According to the recognized rules of technology, the overriding goal of rainwater management is to approximate the natural water balance. The estimate of this target value is difficult to determine. The potential offered by decentralized rainwater management measures to achieve or approach this target as well. One geoinformation-based model pursues these two goals. The first goal is the estimation of the water balance deficit. The second goal is a limitation of the potential measures for the best compensation of this deficit. For this purpose, structural, geological and hydrogeological restrictions are taken into account. The model works exclusively with geo-referenced data and thus offers direct localization of the results. Furthermore, the model combines the approach of the empirically determined water balance distribution functions of the DWA-A 102 and a GIS-based calculation. The advantages of this combination are the scaling of a selectable study area and the on the fly consideration of restrictions for the measures of decentralized rainwater management. The second geoinformation-based model is an artificial neural network (ANN) for calculating flood areas caused by heavy rain. The trained ANN is suitable for known areas, works in real time and substitute a conventional hydrodynamic model. Therefore, a supervised learning multi-layer feed-forward ANN was set up and trained. After that, it was then examined and assessed for its suitability. For the substitution of a hydrodynamic numerical model for calculation of flooding risks (flood areas or surface runoff) by an ANN, the ANN must 'learn' the hydrological processes of the conventional model. For this learning process, the ANN requires input values that essentially determine the hydrological processes (e.g. precipitation, digital elevation model, type of sealing, barriers and culverts, etc.). Moreover, the AAN requires target values. For the ANN, the target values are areas with a certain height water level caused by a precipitation event (flood areas with flood level). These target values were first calculated using a hydrodynamic numerical model.The developed model shows that ANNs are more suitable than conventional hydrodynamic numerical models for certain applications due to their computing speed. Even if the generalizability of the trained ANN to other model areas could not be successfully demonstrated, the ANN however was able to predict good results for a sufficiently known range of values.The biggest advantage of the ANN is the low computing time for predicting flood areas. In contrast to hydrodynamic numerical models, the ANN predicts flood areas with sufficient accuracy in a few seconds. The application possibilities of the developed model thus specialize in cases in which these strengths are indispensable. One indispensability is the prediction of flood area caused by spontaneously and spatially limited heavy rain. This indispensability is also a demand of the municipal flood protection.

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2021; Aachen : RWTH Aachen University 1 Online-Ressource : Illustrationen, Diagramme (2021). = Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2021

Published by RWTH Aachen University, Aachen