Machine Learning Utilization for Bed Load Transport in Gravel-Bed Rivers

Abstract

Three data-driven techniques, namely artificial neural networks, adaptive-network-based fuzzy inference system, and symbolic regression based on genetic programming, are employed for the prediction of bed load transport rates in gravel-bed steep mountainous streams and rivers in Idaho (U.S.A.), and the potential of several input variables is investigated. The input combinations that were tested are based, mainly, on unit stream power, stream power, and shear stress, and exhibited similarly good performance, with respect to the machine learning technique used, accentuating the importance of the regression model. The derived models are robust, generalize very well in unseen data, and generated results superior to those of some of the widely used bed load formulae, without the need to set a threshold for the initiation of motion, and consequently avoid predicting erroneous zero transport rates.