Random encounter modelling as a viable method to estimate absolute abundance of reef fish

Abstract

Remote underwater video (RUV) surveys are increasingly replacing diver-based underwater visual censuses (UVCs) in fish ecology studies, especially on coral reefs. However, extracting reliable estimates of abundance or density from video footage is a major challenge, with most studies using a metric of relative abundance, MaxN (maximum number of concurrent conspecific individuals). Here we compare a video survey density estimation method used in terrestrial wildlife management, random encounter staying time modelling (REST) to MaxN from unbaited videos and density estimates from diver-based UVC. We assess and compare the three approaches according to labour intensiveness (survey time effort per m2), species detection, abundance estimates and potential biases against size or aggregation behaviour. We found that species detection in REST was highly sensitive to sampling effort due to the minimum number of detections required for model fitting. There was also considerable variation in REST-estimated absolute densities between species compared with other methods, however, each method was able to detect compositional differences among sites. While UVC was far less labour intensive than REST or MaxN in survey and processing effort, species richness was consistently lower in UVC observations compared to that from the two video-based approaches. Pairwise generalised linear mixed models between method abundances showed that UVC abundances tended to be higher than other methods for shoaling/schooling fish. Our study identified strengths and weaknesses in each method for specific research objectives that may prioritise species detections or precision in abundance estimates. Although labour intensive, we found REST a viable alternative to MaxN if absolute abundance estimates are preferred. If precise abundance estimates are not a priority and labour is limited, we found MaxN suitable for capturing species richness. Finally, UVC ranked best in labour effectiveness in terms of both person-hours and spatial coverage, and it provided good estimates of abundance, but generally lower species richness. Ultimately, method selection depends on study-specific priorities and trade-offs.