Intelligent prediction of thalweg migration during flood seasons in the braided reach of the lower Yellow River

High rates of thalweg migration frequently occurred in the braided reach, posing a significant threat to the stability of river regime and safety of flood control. However, there is a deficiency in the prediction technique of thalweg migration in the braided reach. Based on the long-term measurement of flow-sediment, cross-sectional profiles and sensing images, the influential factors controlling thalweg migration were quantified and classified. The intelligent models were then constructed to predict the thalweg migration during a flood season in the braided reach, including prediction models for migration probability and migration rate. Two methods were proposed to enhance model accuracy, including filtering the input factors for a section-by-section model and integrating sections with the same main factors into a combination model. Results indicate that several factors were dominant across most sections, including the distance between the second deepest point and the thalweg point, channel geometry and thalweg migration during previous periods. The accuracy of migration probability and rate was improved based on the combination model. By filtering the input factors, high accuracy was achieved in the prediction of thalweg migration in the validation set. The average accuracy for predicting migration direction reached 0.81, and the relative errors of the migration rate were below 1.0% at 59% of the sections. The models also demonstrated good performance for the recent years 2021 and 2022. The predicted thalweg direction was completely accurate at 80% of the sections, and the relative errors of thalweg migration rate were lower than 1.0 at 74%–78% of all sections, with the NSE of 0.58 and 0.72.