AUTHOR=Bacheler Nathan M. , Shertzer Kyle W. , Schobernd Zebulon H. , Coggins Lewis G. TITLE=Calibration of fish counts in video surveys: a case study from the Southeast Reef Fish Survey JOURNAL=Frontiers in Marine Science VOLUME=10 YEAR=2023 URL=https://www.frontiersin.org/journals/marine-science/articles/10.3389/fmars.2023.1183955 DOI=10.3389/fmars.2023.1183955 ISSN=2296-7745 ABSTRACT=

Changes to sampling gears or vessels can influence the catchability or detectability of fish, leading to biased trends in abundance. Despite the widespread use of underwater video cameras to index fish abundance and the rapid advances in video technology, few studies have focused on calibrating data from different cameras used in underwater video surveys. We describe a side-by-side calibration study (N = 143 paired videos) undertaken in 2014 to account for a camera change in the Southeast Reef Fish Survey, a regional-scale, multi-species reef fish survey along the southeast United States Atlantic coast. Slope estimates from linear regression for the 16 species included in the analyses ranged from 0.21 to 0.98, with an overall mean of 0.57, suggesting that original cameras (Canon Vixia HF-S200) observed an average of 43% fewer fish than newer cameras (GoPro Hero 3+). Some reef fish species had limited calibration sample sizes, such that borrowing calibration information from related or unrelated species was justified in some cases. We also applied calibrations to 11-year video time series of relative abundance of scamp Mycteroperca phenax and red snapper Lutjanus campechanus (N = 13,072 videos), showing that calibrations were critical to separating changes in camera sightability from true changes in abundance. We recommend calibrating data from video cameras anytime changes occur, and pairing video cameras to the extent possible to control for the spatial and temporal variability inherent in fish populations and environmental conditions. Following these guidelines, researchers will be able to maintain the integrity of valuable long-term video datasets despite intentional or unavoidable changes to video cameras over time.