Neural Network Prediction of Maximum Scour in Bends of Sand-Bed Rivers

Abstract

Scoured streambeds can undermine revetments built on concave banks in bends of sand-bed rivers formed in alluvium. For this reason, protection work along the banks needs to extend to at least the lowest level reached along the toe of the underwater slope. In this investigation, 202 onsite measurements of bend scour assembled from several previous studies of streams with beds composed primarily of sand-sized sediment are evaluated. The data are used to fit coefficients of an artificial neural network model that offers a robust and easy to apply relation for rapidly predicting the maximum depth of bend scour in sand-bed rivers. A calculation of model variance by a similar neural network provides a means of generating the upper prediction limit of bend-scour depth. Consequently, one can determine the safety margin needed for a design scour depth that is statistically well-reasoned and not disproportionately large.