Estimating relative abundance and species richness from video surveys of reef fishes

Underwater video sampling has become a common approach to index fish abundance and diversity, but little has been published on determining how much video to read. We used video data collected over a period of 6 years in the Gulf of Mexico to examine how the number of video frames read affects accuracy and precision of fish counts and estimates of species richness. To examine fish counts, we focused on case studies of red snapper (Lutjanus campechanus), vermilion snapper (Rhomboplites aurorubens), and scamp (Mycteroperca phenax). Using a bootstrap framework, we found that fish counts were unbiased when at least 5 of 1201 video frames within a 20-min video were read. The relative patterns of coefficients of variation (CVs) were nearly identical among species and declined as an inverse power function. Initial decreases in CVs were rapid as the number of frames read increased from 1 to 50. However, subsequent declines were modest, decreasing only by similar to 50% when the number of frames read increased by 300%. Estimated species richness increased asymptotically as the number of frames read increased from 25 to 200 frames, and reading 50 frames documented 86% of the species observed across all 1201 frames. Lastly, we used a generalized additive model to show that the most likely species to be missed were fast-swimming fishes that are solitary or form relatively small schools. Our results indicate that the most efficient use of resources (i.e., maximum information gained at the lowest cost) would be to read similar to 50 frames from each video.