A Hybrid Model for Fast and Probabilistic Urban Pluvial Flood Prediction

Abstract

Urban flooding is highly uncertain, so the use of probabilistic approaches in early flood warning is encouraged. While well‐established 1‐D/2‐D hydrodynamic sewer models do exist, their deterministic nature and long computational time undermine their applicability for real‐time urban flood nowcasting. Aiming at meeting the needs of fast and probabilistic flood modeling, a new hybrid modeling method integrating a suite of lumped hydrological models and logistic regression is proposed. The lumped models are configured using graph theory techniques, based on the sewer system's topology and characteristics, to account for the spatial heterogeneity and physical processes and properties. The logistic regression models are calibrated to lumped models' results and Yes/No flooding information. Due to its conceptual and data‐driven nature, the hybrid model makes fast probabilistic flood predictions at manhole locations. In two case studies, the results showed that when incorporating most dominant physical processes and properties in the model setup, the hybrid model can achieve up to 86% accuracy for flood warnings issued at 50% probability, with 96% computation saving, compared with a traditional 1‐D hydrodynamic model. When such a detailed model is in place, the hybrid model is set up easily, but the accuracy could be further increased when historical flooding observations are considered.

Flood4castRTF: A Real-Time Urban Flood Forecasting Model

“…To overcome the difficulties of the detailed physically based models for real-time flood forecasting, alternative modelling approaches have been proposed. Different approaches can be used to reduce the computational demand of the flood models: simplifying the 2D shallow water equations by for example omitting the inertia terms (e.g., [18]), using cellular automata approaches [19], using simplified, non-physicalbased methods [1], or by applying empirical/data driven surrogate models [20] or a hybrid approach using a series of lumped models in combination with logistic regression [21]. Despite these recent advantages in simplifying and reducing the computational demand of urban flood models, several difficulties remain.…”

A Rapid Prediction Model of Urban Flood Inundation in a High-Risk Area Coupling Machine Learning and Numerical Simulation Approaches

“…In recent years, some researchers have begun to try to use a combination of numerical simulation models and machine learning methods. In response to the demand for rapid and probabilistic flood modeling, Li and Willems (2020) proposed a hybrid modeling method that combines a lumped hydrological model with logistic regression. Hou et al (2021) used a hydrodynamic model with a 2D shallow water equation as the governing equation, and proposed a rapid prediction model combining a hydrodynamic model and machine learning algorithm.…”

A database system for querying of river networks: facilitating monitoring and prediction applications

“…The implementation of complementary data-driven approaches, which focus on the relationships between input and desired outputs, have become increasingly popular and have proven to successfully represent hydrological processes (Solomatine & Ostfeld 2008;Ahani et al 2018), although most emphasis regarding data-driven approaches in hydrology has been focused on water quality rather than quantity predictions (Maier et al 2010). An exception is the work of Li and Willems (Li & Willems 2020), who present a hybrid approach to predict flooding in sewages and surrounding areas. The presented approach uses a graph based model and graph algorithms to determine the flow path in the sewage network.…”

“…Adequate computational tools and frameworks are necessary for establishing data-driven hydrological models so that they are easy to use for hydrologists and water managers. Hydrological applications depend on geospatial representations and interconnections to trace the flow path of water through the river system, similar to the tracing used by Li and Willems (Li & Willems 2020). Databases are well suited to store and represent this information by translating the geospatial representation into a topology.…”