Author(s): Isa Ebtehaj; Hossein Bonakdari

Linked Author(s): Hossein Bonakdari

Keywords: Evolutionary Polynomial Regression (EPR); Flash flood; Multi-objective genetic algorithm; Streamflow Forecasting; Water Resource Management

Abstract: Accurate flash flood forecasting enables proactive planning, preparedness, and response, which is essential for reducing the risks associated with it and minimizing their detrimental effects on society and the environment. The main aim of the current study is to develop a predictive machine-learning model based on the combination of the multi-objective genetic algorithm (MOGA) and evolutionary polynomial regression (EPR), with a particular emphasis on decision support. In this approach, the genetic algorithm was used to optimize the coefficient of all variables in all terms along with the bias term, and the two defined objectives are the accuracy and simplicity of the model. The developed EPR-MOGA model is employed for forecasting flash floods upstream of Quebec City. It utilizes 15-minute flow discharge data collected from the Saint-Charles station, enabling high-resolution forecasting of flash floods with lead times ranging from one to six hours. After excluding unavailable data due to freezing, a total of approximately 144,000 samples were collected from October 01,1997, to December 31,2022. The statistical indices (R2 = 0.989; NSE = 0.989; NRMSE = 12.996%; MARE = 10.924%; PBIAS = 0.072%) indicate that the developed model demonstrates satisfactory performance in flow discharge forecasting with a lead time of six hours. Moreover, the model accurately predicts flow discharge for extreme events (50- 100 years) with less than 4% relative error, providing reliable flash flood forecasting six hours in advance. The model enables timely decision-making and emergency preparedness by accurately predicting flash flood events. The ability to anticipate flash floods six hours in advance allows authorities to issue appropriate warnings, mobilize resources, and implement evacuation plans effectively. Furthermore, the model's high-resolution forecasting capability provides valuable insights for proactive planning and adaptive strategies, helping mitigate flash floods' adverse impacts. Its performance and reliability make it a valuable asset for enhancing resilience and reducing the impacts of flash floods on communities and the environment.

DOI: https://doi.org/10.3850/978-90-833476-1-5_iahr40wc-p1000-cd

Year: 2023