Intercomparison of UAV platforms for mapping snow depth distribution in complex alpine terrain

Abstract

[EN]Unmanned Aerial Vehicles (UAVs) offer great flexibility in acquiring images in inaccessible study areas, which are then processed with stereo-matching techniques through Structure-from-Motion (SfM) algorithms. This procedure allows generating high spatial resolution 3D point clouds. The high accuracy of these 3D models allows the production of detailed snow depth distribution maps through the comparison of point clouds from different dates. In this way, UAVs allow monitoring of remote areas that were not achievable previously. The large number of works evaluating this novel technique has not, to date, conducted a systematic evaluation of concurrent snowpack observations with different UAV devices. Taking into account this, and also bearing in mind that potential users of this technique may be interested in exploiting ready-to-use commercial devices, we conducted an evaluation of the snow depth distribution maps with different commercial UAVs. During the 2018-19 snow season, two multi-rotors (Parrot Anafi and DJI Mavic Pro2) and one fixed-wing device (SenseFly eBee plus) were used on three different dates over a small test area (5 ha) within Izas Experimental Catchment in the Central Pyrenees. Simultaneously, snowpack distribution was retrieved with a Terrestrial Laser Scanner (TLS, RIEGL LPM-321) and was considered as ground truth. Three different georeferencing methods (Ground Control Points, ICP algorithm over snow-free areas and RTK-GPS positioning) were tested, showing equivalent performances under optimum illumination conditions. Additionally, for the three acquisition dates, both multi-rotors were flown at two distinct altitudes (50 and 75 m) to evaluate impact on the obtained snow depth maps. The evaluation with the TLS showed an equivalent performance of the two multi-rotors, with mean RMSE below 0.23 m and maximum volume deviations of less than 5%. Flying altitudes did not show significant differences in the obtained maps. These results were obtained under contrasted snow surface characteristics. This study reveals that under good illumination conditions and in relatively small areas, affordable commercial UAVs provide reliable estimations of snow distribution compared to more sophisticated and expensive close-range remote sensing techniques. Results obtained under overcast skies were poor, demonstrating that UAV observations require clear-sky conditions and acquisitions around noon to guarantee a homogenous illumination of the study area.