Artificial Intelligence and Objective‐Function Methods Can Identify Bankfull River Channel Extents

Abstract

AbstractBankfull channel extents are of fundamental importance in fluvial geomorphology, to describe the geomorphic character of a river, and to provide a boundary for further processing of morphologic and hydraulic attributes. With ever‐increasing availability of high‐resolution spatial data (e.g., lidar, aerial photography), manual delineation of channel extents is a bottleneck which limits the geomorphic insights that can be gained from that data. To address this limitation, we developed and tested two automated channel delineation methods that define bankfull according to different conceptualisations of bankfull extent: (a) a cross‐sectional method called HydXS that identifies the elevation which maximizes hydraulic depth (cross‐section area/wetted width); and (b) a neural network image segmentation model based on a pretrained model (ResNet‐18), retrained with images derived from a digital elevation model. The cross‐sectional method outperformed the neural network method overall. Its prediction accuracy varied according to channel size and type, with overall precision of 0.87 and recall of 0.80. The neural network method was strongest in larger streams, and outperformed the cross‐sectional method in channel sections with inset benches. A tool to delineate morphological bankfull conditions can allow us to more efficiently implement high‐resolution and large‐scale analyses of channel morphology (e.g., regional hydraulic geometry, channel evolution, physical complexity/habitat surveys), and improve management of fluvial geomorphology and stressors.