Light and Shadow in Mapping Alpine Snowpack With Unmanned Aerial Vehicles in the Absence of Ground Control Points

Abstract

Recent developments in unmanned aerial vehicles (UAV) and structure from motion (SfM) algorithms have shown reliable results for retrieving snow depth distribution. However, their ability to obtain accurate results usually relies on deploying and measuring the exact position of ground control points (GCP) for georeferencing the information. Commercial UAVs can now provide real time kinematic (RTK) positioning of the images with centimetric accuracy. Nonetheless, their operational applicability for observing snow distribution in highly heterogeneous mountain areas has not been evaluated. This study presents a complete assessment of the reliability of snow depth observations from a fixed-wing UAV working in RTK mode with an RGB camera. During the 2018–2019 season, seven field campaigns (13 UAV flights) were undertaken covering 0.48 km2 in complex alpine terrain in the Pyrenees. The UAV observations obtained under different light conditions and flight block configurations (altitude and image overlaps) in the same day were evaluated with terrestrial laser scanner acquisitions. Two SfM processing options of the UAV images were also compared. When the study area received direct solar light, the results were comparable to previous studies that had used GCPs, with an average root mean squared error of 0.19 m and an average absolute snow volume discrepancy lower than 4%. However, when large areas were under shadow from the terrain, or solar light was affected by clouds, the estimated error tripled. The quality of the snow depth maps was little affected by the snow covered area and the flight mission configurations.