Biomass and size estimates of fish are most valuable information to assure optimal management in Norwegian Atlantic salmon production. Such estimates are largely based on models and small samples of few individuals, and improvements are required for better representativity. Advances in machine learning triggered fast development of automated size estimation using stereovision cameras. These measurement systems can result in precise measurements of individual fish. Nevertheless, fish behavior needs to be considered for a proper description of the true mean and variation of fish sizes of cage populations because salmon distribution is not homogeneous in sea cages. Most studies on salmon distribution were performed in relatively small sea cages and additional experiments at commercial scale are required to verify that previous findings are applicable under industrial conditions. In this case study, we monitored vertical and horizontal distributions of salmon in a commercial sea cage. Vertical size stratification was investigated using stereovision cameras, and salmon and growth-stunted fish abundances and swimming speed were measured using cameras recording simultaneously in 16 positions: four horizontal positions and four depths (1 m, 5–6 m, 9–11 m and 14–16 m), four times during one day. Our results suggest a heterogeneous vertical distribution with longer individuals found deeper in the sea cage at high abundance and swimming more slowly than smaller individuals, including growth-stunted fish, found in low abundance close to the surface. Cumulative mean fork-lengths per depth were compared to the estimated true mean fork-length for the whole population of the cage and results show that automatic biomass and size estimation based on a single depth likely results in biased size estimation for the cage population. On the contrary, horizontal distribution appeared homogeneous, and our data supports the assumption that fish were exhibiting a circular swimming pattern, as previously described in small sea cages and under slow current conditions. Our study adds data to currently still scarce knowledge about fish behavior and distribution at commercial scale and provides indications on the use of new technologies for biomass and size estimation. Further studies on longer timelines and different locations are necessary to further generalize behavioral models.
Read full abstract