A Computer Vision Framework for Estimating Body Size and Condition in Multiple Cetacean Species

ABSTRACT Quantifying body size and condition accurately is an important goal for studies in ecophysiology and bioenergetics, as well as conservation biology. Photogrammetric measurement of body morphometrics from aerial photographs taken by unmanned aerial vehicles (UAVs) is now widely used in cetacean research, generating large datasets that are increasingly time consuming to process. While many studies have developed algorithms and procedures for accurately estimating body size and condition, recent automated artificial intelligence software systems allow expedited body morphometric measurements from UAV images, offering the potential to enhance rapid assessment across species, populations, and time frames. We used the WHALESHAPE software, a machine learning tool, to quantify body morphometrics in four cetacean species (~300 bowhead whales Balaena mysticetus , ~300 gray whales Eschrichtius robustus , ~300 humpback whales Megaptera novaeangliae , and ~100 blue whales Balaenoptera musculus ), which we compared to prior data (~470 measurements) on southern right whales ( Eubalaena australis ), collected using the same software. The software was trained separately for each species and successfully quantified body size and condition metrics (body volume and length) across all species. After species‐specific correction, body length estimates were within 5% of manual measurements for 94.6% of bowhead whale images and 96.4% of blue whale images, and within 10% for the majority of gray whale and humpback whale images. Residual body volume mean absolute errors after correction and filtering were lowest for bowhead whales (7.3%), moderate for gray and humpback whales (10.1% and 9.4%, respectively), and highest for blue whales (12.0%), likely reflecting greater sensitivity of volume estimation to errors in width prediction and challenges related to body posture and image quality. Overall, our results demonstrate that with relatively modest investment, WHALESHAPE can be successfully applied across cetacean species of varying size and shape.