Effects of fish behaviour on abundance and length frequency estimates from in-trawl stereo cameras

Abstract

In-trawl stereo cameras can provide fine-scale spatial and temporal information on species along the trawl path and record small-sized and fragile organisms typically absent from catches. Reliable estimates of abundance and length frequency from in-trawl cameras will improve ecosystem understanding and lessen the need for physical catches on scientific surveys. However, determining these estimates from camera footage is challenging since the same individual can appear in multiple frames and swim repeatedly in and out of the camera’s field of view. The manual image analysis performed in this study provides important information on how the swimming behaviour of three abundant pelagic taxa in the Norwegian Sea, along with a camera’s field of view and frame rate, affect the number of repeated appearances. Moreover, these manual annotations serve as a valuable dataset for validating automatic image analyses. Our results show that, depending on the taxa, swimming orientation, length, density of individuals, and distance to the camera affect the extent of time an individual is observed. If the repeated appearance of individuals is not accounted for, taxa or length classes with fewer appearances are under-represented in relative abundance and lead to skewed length frequency distributions. Compared to herring and blue whiting, a large fraction of mesopelagic fishes remains undetected during automatic analysis (RetinaNet). Assessing the factors driving repeated appearances improves our understanding of in-trawl camera data and highlights the importance of integrating tracking with automatic image analysis.