RIVERBED GRAIN SIZE CHANGES FOLLOWING A SEDIMENT PULSE MEASURED USING UAV IMAGERY AND COMPUTER VISION

In mountain streams, the size of the sediment grains that comprise the bed surface is an important parameter that has implications for fish habitat, sediment transport, and channel stability. Despite its importance, measuring grain size in the field is time-consuming and often challenging. This is especially true following large events like landslides or dam releases, when the grain size of the bed may respond quickly. In this study, we use GrainID, a machine-learning-aided computer vision model, on repeat aerial photographs to track surface grain size changes in a small mountain stream in response to a sediment pulse released by the removal of an upstream culvert. We found that although the bed fined overall, we observed local coarsening as the bed adjusted to the pulse. We also highlight limitations of this method and recommend best practices for using computer vision models on aerial imagery, which will be useful as the application of UAVs (drones) to geomorphic data collection becomes more common.