Analysis of Remotely Sensed Channel Width Observations Using High-Accuracy Shoreline Tracking on Alaskan Rivers

Remote sensing has been vital for analyzing river flow in resource-limited regions; however, uncertainty associated with satellite observations can lead to inaccurate results. Discharge calculated using remotely sensed river geometry is likely biased due to uncertainty originating in the original imagery. Thus, it is critical to assess uncertainty within individual variables, so error can be mitigated. Here I sought to evaluate the potential error in river geometry as measured by remote sensing against those measured on the ground. I used a Bad Elf Global Navigation Satellite System (GNSS) Surveyor (20-50cm accuracy after post-processing) to determine the shoreline of three Alaskan rivers (i.e., the Knik, Tanana, and Nenana) during the summer of 2021. Remotely sensed river shorelines derived from a surface water mask generated using DeepWaterMap on WorldView imagery (1.2m spatial resolution). The distance to shoreline was assessed every 10m by measuring the length of a perpendicular line to the river's centerline. For satellite imagery, the shore-river boundary was set as the landward extent of the classified water mask. Ground-based measurements of the shoreline extent were used as a control. Shoreline error is the difference between satellite and ground-based estimates of distance to shoreline, with positive and negative values being over- and underestimates, respectively. Shoreline error here documents the error on one side of the river; and is not identical to width error, which accounts for both sides of the channel. Shoreline bias is the average shoreline error, and the mean absolute error (MAE) is the absolute value of mean shoreline error. Overall, satellite-derived shoreline distance was overestimated by 4.6m (± 9.7m) with a MAE of 8.2m (± 6.9m) across all rivers. Georegistration error may play a role in the overestimation of shoreline. Shoreline error driven by georegistration error can lead to width error may be less than shoreline error, while less georegistration error can lead to width error up to twice as large as shoreline error. These results provide implications for the accuracy of modern high-resolution imagery while giving insight into the source of width errors in today's datasets.