Robust optimal design of urban drainage systems: A data-driven approach

Abstract

The design of urban drainage systems often relies on the coupling of optimization algorithms and process-based models, with the latter informed by design storms. However, such practice may not yield systems that are robust with respect to a broad array of rainfall conditions. To tackle this problem, we contribute a novel computational framework characterized by two distinguishing features. First, we use multiple rainfall events throughout the design process; and, second, we rely on faster emulators (instead of process-based models) to manage the inevitable increase in the computational requirements. In particular, our framework moves through five steps accounting for the generation of stochastic rainfall events, the selection of decision variables (via sensitivity analysis), emulator identification, emulation-based optimization, and ex-post robustness analysis. To test our framework, we implement it on a 33-km2 basin in Ho Chi Minh City (Vietnam) and benchmark it against a more traditional approach based on design storms. Results show that there are untapped opportunities for improving drainage performance without altering investment costs. For example, we found that a budget of $30 million can yield an overflow reduction of about 85% or 50% depending on whether we rely on stochastic rainfall events or design storms. Importantly, such improvements do not affect the computational requirements, owing to the use of emulators. The framework is applicable to other basins and could be readily extended to other sources of uncertainty, such as land use change, or climate change scenarios, which are expected to pose additional strains on urban drainage systems.