Abstract
Urban drainage systems are critical to cities as they handle large volumes of runoff generated during rainfall events, thereby reducing the risks of floods. Current design practices for drainage systems relies on design storms – critical rainfall events of given intensity and duration corresponding to a given return period. However, rainfall conditions can be varied, in terms of intensity, duration, and profiles (i.e., the time distribution of rainfall). Systems optimized with respect to design storms are expected to perform well under similar and less intense rainfall conditions, but this may not always be guaranteed. Designing robust drainage systems – solutions that perform consistently under a broad range of rainfall events – is therefore a challenging task. In this thesis, we address this task by first identifying the flaws of existing design practices and then proposing a novel framework that tackles these flaws, thus supporting the design of optimal, and robust, drainage systems. In the first part of the thesis, we inspect existing design practices and identify when and why design storms may fail to produce robust solutions. To do this, we develop a computational framework that evaluates the robustness of drainage systems optimized for a design storm. The framework consists of four building blocks. First, we use sensitivity analysis to identify the most important decision variables [e.g., pipe expansions and low impact development (LID)], thereby reducing the complexity of the design problem. Second, we solve the problem using a multi-objective simulation optimization scheme, which yields a set of Pareto-efficient solutions optimizing various measures of performance. Following current practice in drainage system design, the first two steps rely on a design storm. Third, we simulate each solution under stochastic rainfall events characterized by different duration, intensity, and profile, and finally evaluate their robustness. The application of this framework to the Nhieu Loc-Thi Nghe basin (Ho Chi Minh City, Vietnam) reveals that that none of the Pareto-efficient solutions are robust across all rainfall events. In particular, we find that the optimized solutions underperform when rainfall intensities, duration, and profiles deviate from those of the design storms. The first part of this thesis thus elucidates the need to include stochastic rainfall events throughout the design process so as to obtain robust drainage solutions. In the next part of the thesis, we contribute a framework that builds on stochastic rainfall events –instead of design storms– for the entire design process. In particular, the proposed framework begins with the stochastic rainfall generation step and considers multiple rainfall events, representing a wide range of intensity, duration, and profiles, in both sensitivity analysis and optimization steps. To overcome the increase in computational requirements, we use emulation modelling techniques to replace theurban hydraulic simulator in the optimization step. We compare our proposed framework to the design storm based method and demonstrate that the proposed framework is more effective in finding drainage systems that are robust against a broad range of rainfall conditions. We also show that it is more efficient in terms of computational power required, solving the design problem 12 times (approximately 600 hours) faster than the conventional method.