Evaluating the sensitivity of multi-dimensional model predictions of salmon habitat to the source of remotely sensed river bathymetry

Multi-dimensional numerical models are fundamental tools for investigating biophysical processes in aquatic ecosystems. Remote sensing techniques increase the feasibility of applying such models at riverscape scales, but tests of model performance on large rivers have been limited. We evaluated the potential to develop two-dimensional (2D) and three-dimensional (3D) hydrodynamic models for a 1.6-km reach of a large gravel-bed river using three sources of remotely sensed river bathymetry. We estimated depth from hyperspectral image data acquired from conventional and uncrewed aircraft and multispectral satellite imagery. Our results indicated that modeled water depth errors were similar between 2D and 3D models, with depth residuals that were comparable to the uncertainty associated with the bathymetry used as input. We found good agreement between measured and modeled depth-averaged velocities generated by 2D and 3D models, while 3D models provided superior predictions of near-bed velocities. We found that optimal model performance occurred for lower flow resistance values than previously reported in the literature, possibly as a consequence of the high-resolution bathymetry used as model input. Model predictions of winter-run Chinook salmon (Oncorhynchus tshawytscha) spawning and rearing habitat were not sensitive to the source of bathymetric information, but bioenergetic predictions related to adult holding costs were influenced by the input bathymetry. Our results suggest that hyperspectral imagery acquired from piloted and/or uncrewed aircraft can be used to map the bathymetry of clear-flowing, relatively shallow large rivers with sufficient accuracy to support multi-dimensional flow model development; models developed from multispectral satellite imagery had more limited predictive capability.