Modelling Suspended Sediment Transport in the Kelkit River, Turkey: A Machine Learning Perspective

Abstract

Suspended sediment (SS) transport is a critical process in river systems, influencing water quality, ecosystem dynamics, and engineering infrastructure. Traditional modeling approaches often encounter challenges in accurately capturing the complex dynamics of sediment transport. In this study, we have employed machine learning methodologies to enhance the predictive capabilities of SS transport models. We used data from the Kelkit River, Turkey, collected during the period of 1967–2012. The study focuses on developing and evaluating two models: Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM) networks. Both models demonstrated high accuracy, with the LSTM model showing superior performance, achieving a coefficient of determination (R2) of 0.991 during training and 0.961 during testing. These results highlight the potential of machine learning techniques, particularly deep learning models, in effectively predicting suspended sediment transport. Future research should consider incorporating additional environmental variables and testing these models in different river basins to validate their generalizability. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.