Estimation of riverbank erosion by combining channel morphological models with AI techniques

Abstract

Simulating riverbank erosion involves modeling the complex interactions between water flow, sediment transport, and bank stability. The present study employs a novel approach, combining the Self Organizing Map (SOM) algorithm and the Long Short-Term Memory (LSTM) network, to capture the behavior of riverbank erosion. A two-stage training procedure is promoted to enhance predictive accuracy and emphasize the need for preprocessing input data related to hydrological conditions, geomorphological features, and soil properties. Pivotal variables that can characterize changes in channel geometry are designed as the output targets, such as the vertical displacement of the riverbed, the horizontal displacement of the riverbank, and the channel width. The goal of this study is to create an alternative to addressing challenges associated with predicting riverbank erosion by utilizing new training methods of artificial intelligence (AI) models. The proposed method performs well in assessing three output variables, showing low mean relative errors, less than 0.081, and high correlation coefficients above 0.981, along with R-squared values over 0.963. These results highlight the effectiveness of this method in accurately describing cross-sectional changes. The proposed method is designed with practical applications in mind, satisfying the need for methods that are not only accurate but also operationally efficient.