Multi-model prediction of ice jam occurrence probability at Toudaoguai section of the Yellow River based on ice thickness simulation with Unified Degree-Day Method

Le résumé fourni par la source

Study region The Toudaoguai reach of the Yellow River, China. Study focus By calibrating the parameters of the Unified Degree-Day Method, this study simulated and validated the formation and ablation process of ice cover at the Toudaoguai Cross-Section of the Yellow River from 2000 to 2024. For different ice jam occurrence scenarios, threshold analysis was conducted to determine the conditions for the discharge difference ( ΔQ ) between freeze-up and break-up dates, and the ice cover thickness on the break-up date. Furthermore, five machine learning models were employed to predict ice jam occurrences; the prediction accuracies of the threshold method and these models were evaluated to identify a method capable of accurate ice jam prediction. New hydrological insights for the region The results showed that the Unified Degree-Day Method with parameters α₀ = 0.85 and β = 0.032 yielded good agreement between simulated and measured ice cover thickness. The threshold conditions for potential ice jam occurrence were a discharge difference ( ΔQ ) exceeding 285 m³ /s and an ice cover thickness on the break-up date exceeding 0.25 m. Compared with the LSTM, SVM, KNN models, the CNN-BiLSTM, RF, and threshold models demonstrated higher accuracy in predicting ice jam occurrences, with accuracies reaching 94.00%, 93.97%, and 93.68%, respectively. These findings provide support for understanding the formation and ablation processes of ice cover, predicting ice jams, and preventing and mitigating ice floods in cold regions.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.