A new method for measuring group behaviours of fish shoals from recorded videos taken in near aquaculture conditions

• Two indexes quantifying dispersion and activity of a fish group are presented.
• The reliability of these indexes was tested on fish shoal simulations.
• The indexes were also tested in real conditions using videos of trout in aquaria.
• We found them to faithfully translate simulated and real behaviour modifications.

Monitoring fish shoal behaviour is a growing concern for scientists studying fish stress and welfare. This study presents an algorithm developed to calculate, from videos taken from above aquaria, two indexes characterizing fish shoal behaviour. These two indexes quantify the dispersion and the swimming activity of the fish shoal in the aquaria. The reliability of these indexes was tested on fish shoal simulations following the rules of Reynolds' model on flocks, herds and schools. Since coordinates of each simulated fish in the shoal was known, these simulations provided true values of dispersion and swimming speed of each fish in the shoal, which were compared to values calculated using the presented algorithm. Further, the two indexes were tested on videos of rainbow trout in aquaria. Behavioural variations of the shoal were estimated before and after food distribution in one test, and before and after a four hours confinement stress in a second test. Data resulting from simulations indicate that the two indexes are sensitive to the simulated changes in the cohesion or the swimming speed of the group. Thus, indexes faithfully translated true values in simulations, with a minimum of 94% of the total variation in true values explained by indexes. Furthermore, the two indexes were sensitive to shoal behaviour modifications observed in the two case studies. Indeed, as expected, a strong group dispersion decrease associated with an important swimming activity increase could be detected just after food distribution using our method. Similarly, our indexes were sensitive to a group behaviour change observed after the four hours confinement stress. Finally, our method was compared to Israeli's, and was found to be more sensitive and more accurate in our conditions. This method provides therefore a sensitive, non-invasive, simple and widely applicable tool to quantify behavioural changes associated with various challenges in aquacultural conditions.