Author(s): Agnieszka Indiana Olbert; Sogol Moradian; Michael Puchley; Thomas K. J. Mcdermott
Linked Author(s):
Keywords: Compound Flood Flood Forecasts Machine Learning Data-driven Systems Early Warning
Abstract: The aim of this research is to develop a state-of-the-art forecasting system of compound floods by using process- and data-driven approaches. Unlike traditional systems, this novel system integrates multiple data sources such as meteorological predictions, tidal records, river flow measurements, and historical flood events to identify complex patterns and interdependencies that contribute to compound fluvial-pluvial-coastal floods. Here, in total, seven ML models including: Support Vector Regression (SVR), Support Vector Machine (SVM), Gaussian Process Regression (GPR), Radial Basis Function (RBF), Linear Regression (LR), Decision Tree (DT), and Artificial Neural Network (ANN) were used. ML algorithms process the input data, offering more precise and timely predictions of when and where floods are likely to occur. By improving the accuracy and integration of flood forecasts, the project enhances early-warning systems and decision-making processes, thus increasing community resilience and better preparedness for the growing threat of compound coastal-fluvial floods. This system addresses the limitations of traditional hydrodynamic model-based approaches, where the trade-offs between high numerical accuracy and computational efficiency often restrict their use in short-term flood forecasting. Reliable flood forecasting brings numerous societal benefits, including enhanced flood preparedness, greater confidence in decision-making, optimized resource allocation, minimized flood-related damages, and, in some cases, potential flood prevention.
Year: 2025