flood forecasting based on machine learning pattern recognition and dynamic migration of parameters

Abstract

Study region: Typical basin in semi-arid and semi humid areas in the middle reaches of the Yellow River Study focus: Floods are among the most devastating natural disasters. Timely and accurate forecasting of runoff is crucial to safeguard human lives, minimize property damage, enhance the efficiency of reservoir power generation, and ensure the safety of water supply. In this study, the rainfall and flow data of 98 floods occurring between 1971 and 2014 in the Jingle sub-basin, a tributary of the Yellow River basin, China, were analyzed using dynamic clustering and random forest techniques to identify flood types and select appropriate model parameters. The Xin’anjiang model was then used for real-time flood forecasting. The results indicate that the rainfall characteristic indicators developed by the model can effectively identify potential flood types, and the model parameters determined by historical flood rates can be adapted and utilized for new forecasting tasks based on similarities. Ensemble forecasting results, which consider the probability of flood types, are superior to single fractal forecasting outcomes and diminish uncertainty. The proposed method can identify extreme flood events, facilitate flood classification and prediction, promote basin disaster mitigation, and enable the efficient use of water resources. New hydrological insights for the region: The proposed flood classification and identification method effectively analyzes flood events in the basin. The floods that may occur are characterized by relevant statistical characteristics of rainfall, allowing for selection of corresponding flood forecasting model parameters for improved accuracy in real-time flood forecasting and extended prediction period.

Keywords

• Rainfall characteristic
• Flooding
• Flood forecasting
• Parameter migration
• Xin’anjiang model
• Clustering