Exploring the potential of utilizing unsupervised machine learning for urban drainage sensor placement under future rainfall uncertainty

Abstract

Recently, advanced informatics and sensing techniques show promise of enabling a new generation of smart stormwater systems, where real-time sensors are deployed to detect flooding hotspots. Existing stormwater design criteria assume that historical rainfall frequency and intensity are reliable predictors to place real-time sensing devices. However, nonstationarity in rainfall due to climate change violates this assumption by disturbing hydrologic regimes and relocating flooding spots. This paper proposes a novel methodology of combining unsupervised machine learning (Agglomerative Clustering) and analysis of variance (ANOVA) to optimize the sensor placement under uncertain rainfalls. An urban drainage network located in Salt Lake City, Utah, USA, is chosen as the case study to demonstrate the application of the proposed method. Results show that: i) the proposed Agglomerative Clustering and ANOVA integrated approach can efficiently and accurately pinpoint sensor locations for drainage flooding detection; ii) rainfall uncertainty has limited impacts on the number of sensors, but it induces significant effects on sensor locations from the historical period (2000–2009) to the future period (2040–2049). By exploring the effects of climate nonstationarity on sensor placement, this work aims to help engineers and decision-makers better respond to the changing climates and rainfall extremes in urban drainage catchments.