Neural Network Prediction of Alluvial Stream Bedforms

Abstract

Substantial laboratory, field, and theoretical studies have been carried out by many to understand alluvial stream bedform origin, their shapes, equilibrium with the flow, and their depositional structure. The findings of these analyses are often presented as phase or stability diagrams in which the dependence of the various bed states on two or three variable quantities is depicted graphically. However, the several hydrodynamic and sediment-related parameters that control bedform development in alluvial channels makes the construction of stability diagrams that display the complex interactions clearly and consistently problematic. In this study, alluvial stream bedforms are studied using a theory-guided data science approach that assures logical reasoning when analyzing physical phenomena with large amounts of data. First, a theoretical evaluation of parameters that influence bedform development is carried out, followed by a classification of the bedform type with an artificial neural network (ANN) trained using a sizeable collection of 2,412 samples (2,144 from laboratory flumes and 268 from natural streams). The neural network provides reliable predictions of bedform states and distinguishes between laboratory flumes and natural stream channels.