Use of Machine Learning techniques to identify scenarios for implementing Nature-based Solutions that best prevent flooding

A paradigm shift is taking place in the way urban rainwater drainage is thought about, with the understanding that conventional drainage structures, known as gray infrastructures, end up collecting large volumes of water on impermeable surfaces, causing problems downstream. It is therefore necessary to consider systems that favor the interception and infiltration of water into the soil so that surface runoff is treated at the point where it is generated. Such systems are known as Nature-based Solutions (NbS), which comprise a set of structures that simulate natural drainage processes. These must be carefully designed and positioned to act at points in the urban landscape that, if impermeable, would favor water accumulation. This study consists of using Machine Learning (ML) techniques to identify the NbS layout that provides the best flood protection in an area of the city of Campinas, Brazil. To this end, a model in PCSWMM software is used, which will involve the implementation of eight NbS: bio-retention cell, infiltration trench, permeable pavement, rain barrel, vegetative swale, rain garden, green roof, and rooftop disconnection. Different rainfall scenarios are simulated to assess surface runoff generation in each subcatchment. The same volume of precipitation is considered, which is temporally and spatially distributed differently in each rainfall scenario, allowing the identification of differences in the floods generated. Using a database derived from the simulation results, Artificial Neural Networks (ANN) are applied to create a predictive model of surface runoff generated in a given rainfall event. An analysis of the variability of runoff in the different subcatchments is then performed, identifying how much the source of flow generation varies spatially when the rainfall configuration is modified. The NbS are then dimensioned with the help of the rainfall configuration most likely to cause flooding. The hydrological model is simulated several times, varying the positioning and quantity of NbS throughout the subcatchments, in order to generate data that, when applied to ANN, identifies the implementation scenario that best combats flooding in the studied area. The NbS are allocated so that each scenario generates the same implementation cost according to the Brazilian price benchmark, making each scenario have the same intensity of NbS allocation. The study presents a new methodology for sizing sustainable solutions, showing how much the use of ML techniques can assist in the design process of rainwater drainage for new developments. The study area considered is the Campinas International Hub for Sustainable Development, which hosts universities and research institutions. It will be expanded over the next few years, and the implementation of NbS on site will serve as a living laboratory for students who, on a daily basis, will be able to see in practice how sustainable solutions contribute to flood control.