Estimating Three‐Dimensional Posture to Correct Drone‐Based Measurements of Free‐Swimming Marine Mammals: An Application to Dugongs in Shallow Waters
Chiaki Yamato, Helen L. Sneath, Janet M. LanyonABSTRACT
Drone‐based measurements of free‐ranging aquatic wildlife are inevitably affected by posture variations, yet this issue has remained largely unaddressed. A widely used approach is to visually inspect body curvature based on overhead shots and exclude animals exhibiting strong body curvature, which may result in a substantially reduced dataset and inconsistent quality control. We developed a drone‐based method of estimating body measurements of dugongs using dual drones (positioned vertically and obliquely relative to the animal) whilst considering body posture (Method dual ). We reconstructed 3D coordinates of the body's dorsal line and scaled these based on altitude obtained with Real‐Time Kinematic measurements. Estimation error was evaluated using 25 free‐swimming dugongs for which direct body measures were obtained out‐of‐water and compared with a conventional method based only on overhead shots (Method single ). General surfacing posture of animals was further examined using 188 wild dugongs. Method dual achieved significantly smaller absolute errors in body length (BL) than Method single both when the body was relatively curved (0.039 ± 0.033 m or 2.0% ± 1.6% of BL versus 0.114 ± 0.084 m or 4.8% ± 3.8%) and straight (0.033 ± 0.028 m or 1.3% ± 1.1% versus 0.05 ± 0.042 m or 1.5% ± 1.2%). Method dual provided reliable estimation even when the body curve ratio (ratio of apparent BL from the overhead shot to the posture‐corrected 3D BL) was lower than 0.919 (i.e., body curve is stronger), whereas Method single did not. The mean body curve ratio of free‐swimming animals was 0.926 ± 0.049 (0.751–0.992; n = 188) and 31.9% had a body curve ratio below 0.919. The proposed method enables data collection from surfacing events that would otherwise have been discarded as poor quality, which may be particularly useful when encounter rate is low. This method also helps establish quantitative thresholds for data inclusion.