A data-driven approach to compare behavioral and physiological metrics of gilthead sea bream exposed to playback ship noise to a control group

A data analysis approach is proposed to compare motion and physiological metrics of a test group of gilthead sea bream ( Sparus aurata ) exposed to a playback of shipping noise to a control group. Two groups, Test and Control, were observed for one month, with playbacks of ship noise emitted in the Test tank only. A combination of MobileNetV2 and DeepSORT is used to process video footage of the fish allowed collection of millions of fish trajectories for motion analysis. When tested once, the method showed a potential to discover non-linear differences between the behavior of the fish in the Test and Control groups by observing changes in the distribution of the fish speed, acceleration, movement curvature and dispersion in the tank. The results include a trend of reduced weight gain in the noise-exposed group, although growth rates were not significantly different, together with decreases in three beneficial metabolites and increases in two potentially harmful metabolites. These results were obtained for a playback of shipping noise at a sound pressure level that is only 10 dB above the ambient noise in the water tank, where both short- and long-term effects on the fish were observed. The results therefore represent preliminary evidence of potential stress responses rather than conclusive proof of physiological stress. Accordingly, this study should be viewed as a proof-of-concept demonstration of a multimodal approach that integrates behavioral and physiological indicators. • Statistical comparison of motion and physiological metrics between groups of gilthead sea bream fish. • Compared to the control group, non-linear differences were observed in movement, weight trends, and blood metabolite profiles. • The fish exposed to noise showed a consistent trend of lower weight gain, although growth rates were not significantly different. • Fish exposed to noise showed a decrease in 3 beneficial and an increase in 2 harmful metabolites. • Movement was analyzed from video footage using a combination of MobileNetV2 and DeepSORT for trajectory analysis.