Reconstructing bed topography of a shallow river from close-range aerial imagery: Multi-UAV experimental campaign in the Izera river (SW Poland/N Czechia)
Matylda Witekcor, Grzegorz Walusiak, Michał Halicki, Joanna Remisz, Dorota Borowicz, Krzysztof Parzóch, Łukasz Kasprzak, Jakub Langhammer, Michal Gallay, Jakub Miřijovský, Ján Šašak, Ján Kaňuk, Theodora Lendzioch, Robert Minařík, Stanislav Popelka, Tomasz Niedzielski
Abstract
The objective of the study is to reconstruct the bathymetry of a shallow river using the Structure-from-Motion algorithm and to evaluate the robustness of the approach by exploring its repeatability and reproducibility. To address this issue, the multi-UAV experimental campaign was organized in SW Poland/N Czechia, and focused on the smoothly flowing Izera river. Unmanned aerial vehicles used to conduct the experiment were: one fixed-wing aircraft (eBee by SenseFly) and three multi-rotor aircrafts (DJI Matrice 210-RTK V2, DJI Mavic 2 Pro, DJI Phantom 4 Pro). Riverbed topography was reconstructed 48 times (2 sites × 4 platforms × 3 resolutions × 2 repetitions) using Structure-from-Motion algorithm, corrected for refraction. The reference elevations of riverbed were obtained using leveling, while water surface elevations were determined by a terrestrial laser scanner. Root mean square errors (RMSE) of the reconstructions of river bottom topography (mostly underestimated) varied between 2.6 cm and 109.3 cm, with the median of 17.5 cm. The error of reconstruction of river bottom topography falls within a typical error range known for through-water photogrammetry. RMSE of the repeatability test for all platforms is smaller than 10 cm in 57 % (up to 76 % for only multi-rotor aircrafts) of the cases and for the reproducibility test in 47 % (up to 60 % for only multi-rotor aircrafts) of the cases. Hence, it was found that the studied approach is robust in terms of its repeatability and reproducibility, predominantly when multi-rotor aircrafts are utilized. The effectiveness of the bottom reconstruction, however, depends on the channel morphology, and specifically decreases at the vicinity of the river banks. • 48 river bathymetry reconstructions: 2 sites, 4 UAVs, 3 resolutions, 2 repetitions. • Structure-from-Motion algorithm and refraction correction were employed. • Median RMSE of riverbed reconstructions was 17.5 cm, varying from 2.6 to 109.3 cm. • Better accuracies were obtained for multi-rotors; resolution revealed no influence. • Reproducibility and repeatability tests confirmed the robustness of this approach.