Introduction
Farmed aquatic animals are subjected to numerous handling operations that can be challenging for their health and welfare . One such operation is crowding , where fish are subjected to reduced rearing volumes to facilitate handling , which is key in the rearing cycle. If fish are repeatedly crowd ed, there may be potential cumulative effects upon fish welfare and these effects can be detrimental (Espmark et al., 2015; Grøntvedt et al., 2015; Roth, 2016; Gismervik et al., 2017). Methodologies based on operational welfare indicators (OWIs), Laboratory-based welfare indicators (LABWIs), health indicators and new emerging technologies have been developed to monitor and audit the welfare of farmed animals, including fish (Noble et al., 2018). Its application can help audit, refine, and optimize crowding procedures. The main goal of this study was to quantify and audit fish behaviour during crowding operations in order to create automatic frameworks and protocols that can assist in documenting/optimising fish welfare during crowding operations.
Materials and Methods
The effect of four different crowding intensities on Atlantic salmon welfare was explored by the project Crowd Monitor (Norwegian Seafood Research Fund, pr. num. 901595) by means of 12 tanks of 3300 l, holding ca. 120 fish (>500g) within a clockwise water flow. Each crowding level was triplicated (4x3 design) and a GoPro hero4 black was installed over the tank to capture the entire water surface in the field of view . Each tank was recorded for ca. three hours: around half an hour before crowding, two hours during crowding and half an hour after crowding. One tank with intermediate crowding intensity was used for the present study. After observing the original video, b oth swimming structure impacted by crowding onset and its gradual increase during crowding were hypothesized . Hereafter , fish swimming behaviour was quantified through computer vision (Python 3.0, OpenCV library). Firstly , median frames were calculated for the 30-minute period before and after crowding. The 2h crowding period was split in four 30-minute periods and the median frames were also generated (Fig. 1A) . Theoretically, a still and more structured fish swimming would be mirrored in median frames showing fish silhouettes oriented towards the water flow , whereas chaotic swimming and low swimming structure would render blurred median frames . Further metrics were quantified for median frames’ validation. Five frames of every 30-minute period were subsampled and the coordinates from heads, anterior dorsal fins and tail peduncles were extracted (ca. 50 fish per frame) to calculate a) individual swimming orientations and ii) swimming angles as proxies of swimming structure . Heart rate as a proxy of activity/stress to cross-validate mean frames, swimming orientation and swimming angle was obtained from intraperitoneal tags (n = 4 fish) . Eventually, fish swimming behaviour was continuously quantified over the whole sample video by mean histogram values at three distinct levels , i.e., within the crowding area, rear- and front-zone of the crowding area. The fish abundance in the front and rear zone of the crowder were also calculated (Fig. 1B).
Results and discussion
The largest proportion of individuals swimming against the water flow was detected before crowding, followed by the post-crowding period. T he crowding period showed the largest variability in swimming orientation and swimming angle . Overall, and consistent with median frames observation, the mean histogram colour unveiled a gradual increase in fish abundance in the front of the crowder, in contrast to the rear zone (Fig. 1B) . This is consistent with the visual analysis of median frames, revealing clearer fish silhouettes swimming against the water flow in the crowd front , and avoidance of the rear zone . B oth the largest shift in the number of fish swimming against the water flow and a decreas ing heart rate coincided one hour after the crowding onset. This could be linked to a decrease in activity/stress during crowding . Overall, the current behavioural approach detected fish gradually coping with crowding conditions by structuring swimming and avoiding the rear zone of the crowding area . Uncontrolled factors such as chaotic swimming and/or interactions with the crowder might jeopardize fish welfare status. Therefore, f urther research could focus on the optimal thresholds of duration , repetition and/or the speed of crowding activities regarding welfare status.
References
Espmark, Å., Kolarevic, J., Aas-Hansen, Ø., Nilsson, J. 2015. Pumping og håndtering av smolt. NOFIMA Rapportnr.:6/2015. ISBN 978-82-8296-262-9.
Gismervik K, Nielsen K.V., Lind M.B., Viljugrein H. Mekanisk avlusing med FLS-avlusersystem- dokumentasjon av fiskevelferd og effekt mot lus. Veterinærinstituttets rapportserie 6-2017. Oslo: Veterinærinstituttet; 2017. ISSN 1890-3290
Grantvedt RN, Nerbevik IKG, Viljugrein H, Lillehaug A, Nilsen H, Gjevre AG. Termisk avlusning av laksefisk - dokumentasjon av fiskevelferd og effekt. Veterinaerinstituttets rapportserie 13-2015
Noble, C., Gismervik , K., Iversen, M. H., Kolarevic, J., Nilsson, J., Stien, L. H., ... & AS, N. (2018). Welfare Indicators for farmed Atlantic salmon: tools for assessing fish welfare. ISBN: 978-82-8296-556-9
Bjørn Roth 2016. Avlusing av laksefisk med Optilice : Effekt på avlusing og fiskevelferd. NOFIMA Rapportnr.: 59/2016. ISBN: 978-82-8296-458-6