Estimating detection probabilities for reef fishes on baited underwater video

Using video to survey fish populations has become widespread due to technological advances of cameras, and it is often assumed that detection of species is constant (or 100%) in video surveys. We tested these assumptions by analyzing paired video and chevron trap data (n = 11091) from a region-wide survey along the southeast US Atlantic coast (2015-2023). Using nonlinear logistic regression, we related the presence or absence of 8 reef-associated fish species on video (when they were confirmed to be present based on trap catch) to predictor variables including water clarity and current direction relative to the camera view after standardizing for variable site abundance. Empirical video detection estimates ranged from 0.49 for bank sea bass Centropristis ocyurus , a relatively small, plainly colored, demersal species, to 0.91 for red porgy Pagrus pagrus , a larger, distinct species that tends to circle the baited trap-video array. Regression models only explained 9-28% of the variation in video detection for the 8 species, suggesting video detection was variable and difficult to predict. After accounting for variable site abundance, the effects of water clarity and current direction on video detection were inconsistent among species, as were the effects of maximum substrate relief, bottom depth, water temperature, and percent of the bottom that was rock. Video detection of predatory and scavenging fish species was not strongly affected by characteristics of the water or seafloor, but video also clearly did not detect all fish species at a site. Therefore, imperfect detection must be acknowledged and accounted for in video surveys, perhaps by using a paired-gear sampling approach.