Evaluating the Interpretation Uncertainty From the Manual Streambank Delineation

ABSTRACT Streambank delineation enables the development of two important parameters: streambank position and channel width. For older aerial images that cannot be easily processed by automatic methods, manual interpretation of streambanks is a necessary task. In this study, the streambanks of a 70 km reach of the Canadian River in Oklahoma, USA were manually delineated by three individuals, using eight, red‐green‐blue band composite imagery. We quantified uncertainty across the three interpretations and evaluated how, sinuosity, flow rate on the image, and recent flood activity influence uncertainty. We developed two streambank calculation methodologies, namely the centroid method and direct width method. The centroid method produced three uncertainty estimates for the parameters streambank position, absolute channel width, and signed channel width, while the direct width method produced an uncertainty estimate for absolute channel width. Interpretation uncertainty was defined as the median error value for skewed distributions or as the mean value for normal distributions. Interpretation uncertainty estimates ranged from −0.1 to 2.8 m. These uncertainty values are 0.1 to 3 times the 1 m pixel image resolution and are an order of magnitude lower than the median width of the channel, which changed from 106 m in 2013 to 71 m in 2021. Increased channel sinuosity was shown to increase the median and interquartile range of the uncertainty metrics, while more outliers were observed for images taken soon after two‐year return flows. Overall, we conclude that manual delineation is a reliable method to obtain streambank position and channel width in large, geomorphically complex rivers.