Evaluation of UAV Flight Planning Strategies on the Accuracy of Glacier Mapping in Rugged Antarctic Terrains: A Case Study of Horseshoe Island

Changes in the Earth’s radiation balance have increased global average temperatures, accelerating glacier melt in Antarctica. Monitoring this process has become of great importance for future modeling studies. Therefore, obtaining accurate, reliable, and most importantly rapid, logistically feasible, and cost-effective measurements of glacier melt is critically important for regions with harsh conditions such as Antarctica. The Ground Sampling Distance (GSD) and positional accuracy of photogrammetric products generated by unmanned aerial vehicle (UAV) photogrammetry are critical to accurately assessing snow and glacier dynamics. In this context, the study presents accuracy assessments and planning recommendations for various flight plan scenarios in a rugged area with slopes up to 40% and partial snow and glacial cover. The study was conducted on a test site on Horseshoe Island, and the UAV flight missions were performed under two operational modes: (i) terrain-follow and (ii) fixed-height. The flight missions were real-time kinematic (RTK)-supported; however, ground control points (GCPs) were also placed and measured prior to the flights. The images were evaluated to assess the positional accuracy achieved through RTK-supported processing and GCP-based georectification, with each flight mode analyzed separately. Fixed-height flight planning resulted in lower spatial accuracy than terrain-follow mode, indicating that maintaining a constant flight height over irregular topography reduces spatial resolution due to variable image scales. It was also observed that it performs better in terms of key points per image and bundle block adjustment (BBA)-Average reprojection error in terrain-follow scenarios. In terms of 3D positional accuracy, using a limited number of GCPs resulted in root mean square error (RMSE) values of approximately 1–2 cm in the horizontal and 2–3 cm in the vertical components; however, the RTK-supported flights exhibited higher errors in the Z component, with values of 12.18 cm for the fixed-height and 6.85 cm for the terrain-follow flights.