Author(s): Mohammad Uzair Anwar Qureshi; Mohammad Shaheen; Ousmane Seidou; Juraj Cunderlik; Hossein Bonakdari
Linked Author(s):
Keywords: Ottawa River; Artificial Intelligence; HEC-RAS; Machine Learning; Hydrodynamic Modeling
Abstract: The frequency and intensity of extreme flood events in the Ottawa River Basin have escalated due to climate change, underscoring the limitations of traditional hydrodynamic models like HEC-RAS in providing real-time hydrological predictions. Although Artificial Intelligence (AI) models rely on high-quality historical data for training and testing, recent hydrological anomalies, such as the 2019 flood, which surpassed the scale of a 100-year event, have exposed the inadequacy of historical data alone for reliable flood forecasting. This study explores the integration of the HEC-RAS hydrodynamic model with an Extreme Learning Machine (ELM) to enhance flood prediction accuracy in a changing climate. The research focuses on a 14 km section of the Ottawa River, covering 22 cross-sections in flood-prone areas such as the town of Quyon & Constance Bay. The HEC-RAS model is calibrated using a wide range of discharge data & river bathymetric surveys to simulate the river’s hydraulic behavior under various flow conditions. The calibrated results are then used to develop a robust AI-driven model that incorporates real-time data and adapts to previously unforeseen flood events. The integration of traditional hydrodynamic modeling with AI techniques aims to produce more reliable and timely flood predictions, improving flood risk management and decision-making in the context of climate variability. The ELM's reliability analysis demonstrates its effectiveness in peak flow forecasting, with Mean Absolute Relative Error (MARE) values remaining below 10% across various discharge scenarios and lead times, underscoring the model’s precision in predicting high-impact flood events.
Year: 2024