Automated Morphometric Modelling of an Endangered Freshwater Cetacean from Unscaled UAV Imagery Using an Internal Scaling Framework

With the rapid advancement of drone technology, aerial photogrammetry has been increasingly applied to morphometric research, particularly for endangered species, owing to its fully non-invasive nature. Photogrammetry-derived morphometric models (e.g., three-dimensional reconstructions) provide critical insights into individual health condition and population demographic characteristics. Conventional aerial photogrammetry function through conversion of the pixel dimension to real-life unites, and its accuracy is susceptible to deviation on intrinsic and extrinsic parameters (e.g., image distortion and flight attitude). Using the endangered Yangtze finless porpoise (Neophocaena asiaeorientalis asiaeorientalis) as a model species, we demonstrated that the allometric development of blowhole length (distance from the tip of the rostrum to the posterior margin of the blowhole) results in significant age-specific differences in the blowhole length-to-body length ratio, providing an internal scaling metric that is independent of image scale. Although the small body size and agile swimming behavior of this species challenges aerial photogrammetric surveys, a recent rescue operation provided a unique opportunity allowing direct and accurate morphometric measurements for validating the internal scaling framework and novel segmented weighted integration model. We further developed a computer program that automates pixel extraction, metric conversion, model implementation, parameter calculation, and data tabulation. Significant differences in the blowhole-to-body length ratio were observed among calves (n = 11), juveniles (n = 36), and adults (n = 35) (ANOVA, P < 0.01), with all pairwise comparisons remaining significant (calves vs. juveniles, P < 0.001; calves vs. adults, P < 0.001; juveniles vs. adults, P < 0.01). The model demonstrated high accuracy in estimating total body length from unscaled images without prior knowledge of flight altitude or camera focal length, achieving a mean absolute error of 6.69 cm and a mean absolute percentage error of 5.02% across 62 test images. Our study presented an automated and accurate approach for estimating body length from a single unscaled drone image of dorsal view using internal scaling framework and segmented weighted integration model, providing a practical tool for rapid morphometric assessment of marine mammals.