Whose bite? Evaluating the use of structure-from-motion for monitoring fish corallivory

Corallivory, or predation on corals, is a naturally occurring process that, at high levels, can impair coral growth, reproduction, and recovery. Traditionally, fish corallivory monitoring has been achieved through in situ visual surveys and analysis of 2D photoquadrats. However, 3D imaging techniques such as structure-from-motion (SfM) can provide a powerful tool to collect high-resolution colony-level data with limited field effort. Here, we conducted fish corallivory surveys using a recently developed SfM method at 10 sites around the island of O‘ahu, Hawai‘i, paired with in situ surveys to evaluate the accuracy of SfM as a fish corallivory monitoring tool. Our primary objectives were to test for differences in fish bite counts between methods and to understand how variables such as depth, hard coral cover, type of fish predator, and coral species further influenced that difference. Based on bite mark characteristics, we identified four types of fish predators: blennies, parrotfishes, scrapers, and excavators, with the latter two including triggerfish, pufferfish, and, to a lesser extent, parrotfish. Overall, fish bite counts varied significantly between methodologies, with higher counts recorded through SfM annotations and an average difference between methods of 17.34 bite marks m −2 ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" id="im1"><mml:mrow><mml:mover accent="true"><mml:mtext>x</mml:mtext><mml:mo>¯</mml:mo></mml:mover></mml:mrow></mml:math> ± SD: SfM = 39.64 ± 56.71; in situ = 22.30 ± 25.09 bite marks m − ²). The nature of these differences further varied depending on the type of predator and the coral species they consumed. Lastly, at deeper and higher coral cover sites, the difference in bite counts between methods was greater than at shallower and lower cover sites, with more bite counts recorded with SfM. These differences likely reflect inherent air and time limitations divers face that do not exist when annotating SfM models. Despite differences in absolute fish bite mark counts, both methods were consistent in the qualitative patterns of relative fish corallivory across sites, whereby methods aligned in the ranked order of sites from the least to most fish corallivory observed. Overall, our results indicate that SfM is a viable tool to quantify fish corallivory, with the added benefit of enhanced accuracy at sites where diver-based surveys are logistically limited.