Introduction
Aquaculture has grown significantly in recent years, necessitating the development of advanced monitoring techniques to ensure fish welfare and optimize management practices. Understanding how fish respond to environmental and anthropogenic factors is key to improving welfare standards. Surgically implanted b iologgers, capable of measuring heart rate (HR) and external acceleration (ACC), are increasingly used to monitor fish behaviour and physiological states, providing valuable insights into activity patterns.
As the aquaculture industry shifts toward more sustainable and ethical practices, welfare monitoring is becoming increasingly important. Within the context of Precision Fish Farming , these technologies facilitate informed decision-making and enhance farm management (Morgenroth et al., 2024; Brijs et al., 2021). Among them , biologgers stand out for their ability to measure HR and ACC, offering insights into fish physiology and behaviou r under farming conditions. However, despite extensive research on northern species such as Atlantic salmon ( e.g. Brijs et al., 2018) , few studies have focused on Mediterranean species such as the European seabass (Dicentrarchus labrax).
This study assesses the behavioural and physiological responses of seabass reared in open-sea cages using data from implanted biologgers , applying machine learning techniques to classify welfare-related behavioural states and highlighting the potential of integrating sensor technology and data analysis to support welfare monitoring in aquaculture.
Materials and Methods
The study was conducted at the LIMIA-IRFAP aquaculture facilities (Balearic Islands, Spain ), using 1000 adult seabass reared in four open-sea cages. Sixteen fish were implanted with biologgers (DST milli HRT-ACT, Star-Oddi®) to monitor HR and ACC over two 15-day periods in March and July. Surgical implantation followed the protocols described by Cabrera-Álvarez et al. (2024), ensuring minimal impact on fish recovery and behaviour. Data were recorded every five minutes and retrieved post-mortem. To classify behavioural states, a Random Forest (RF) model was developed with data obtained from swimming and crowding stress challenges, enabling the definition of four key states: Resting, Regular Activity, Reactive Stress Response, and Proactive Stress Response . The RF model was trained using 540 labelled observations, with independent testing to ensure its robustness. During the swim challenge, fish were exposed to incremental swimming speeds (0.2 to 1.0 m/s), while the stress challenge included progressive stressors such as reduced water levels and net chasing. Behavioural states were labelled as follows: nighttime observations were categorized as Resting, swimming speeds between 0.2 -0.6 m/s as Regular Activity, stress-induced peaks as Proactive Stress Response, and post-stress periods as Reactive Stress Response. Following validation, the RF model was used to classify the data in the present study.
All procedures involving fish were approved by the Ethical Committee of Animal Experimentation (CEEA-UIB, Spain; Ref. 206/12/22), the Dutch Central Committee for Animal Experiments (CCD ; project AVD40100202115078, 15 November 2021), and the Animal Experiments Committee (DEC) and Authority for Animal Welfare (IvD) of Wageningen University and Research (experiment 2021.D-0005.004) . They were performed by trained personnel in accordance with the European Directive 2010/63/UE and Spanish Royal Decree RD53/2013, ensuring optimal animal care, health, and welfare.
Results and Discussion
The biologger data revealed distinct behavioural patterns in seabass under farm conditions. The RF model successfully classified behavioural states with an overall accuracy of 67.38% on the test dataset and over 73% during cross-validation. The model performed particularly well in detecting stress-related behaviours, correctly identifying Proactive and Reactive Stress Responses in over 80% of cases. These stress states were characterized primarily by elevated ACC values, with HR also contributing in reactive contexts.
When applying the RF model to the biologgers data obtained from the seabass reared in open-sea cages, circadian behavioural rhythms that aligned with operational routines were detected. Resting states predominated at night and on weekends when human activity around the cages was minimal. Conversely, stress-related behaviours were more frequent during daylight hours, particularly in July, when water temperatures and maritime traffic were higher. These findings indicate that both environmental and anthropogenic factors significantly influence seabass behaviour and welfare.
Overall, by combining sensor-based monitoring with behavioural classification, this study presents a practical approach to real-time welfare assessment in open-sea aquaculture. The method not only helps identifying stressing events but also provides insights into baseline activity and seasonal activity patterns. These findings underscore the value of integrating behavioural data into farm management strategies and support the development of PFF technologies tailored to open-sea aquaculture systems.
References
Brijs , J., Sandblom, E., Axelsson, M., Sundell, K., Sundh, H., Huyben , D., ... & Gräns, A. (2018). The final countdown: Continuous physiological welfare evaluation of farmed fish during common aquaculture practices before and during harvest. Aquaculture, 495, 903-911.
Brijs , J., Føre , M., Gräns, A., Clark, T. D., Axelsson, M., & Johansen, J. L. (2021). Bio-sensing technologies in aquaculture: how remote monitoring can bring us closer to our farm animals. Philosophical Transactions of the Royal Society B, 376(1830), 20200218.
Cabrera-Álvarez, M. J., Arechavala-Lopez , P., Mignucci , A., Oliveira, A. R., Soares, F., & Saraiva, J. L. (2024). Environmental enrichment reduces the effects of husbandry stressors in gilthead seabream broodstock. Aquaculture Reports, 37, 102256.
Morgenroth, D., B. Kvaestad , F. Økland, B. Finstad, R. Olsen, E. Svendsen, C. Rosten, M. Axelsson, N. Bloecher, M. Føre y A. Gräns . 2024. Under the sea: How can we use heart rate and accelerometers to remotely assess fish welfare in salmon aquaculture?. Aquaculture. 579, 740144.
Acknowledgements
The authors acknowledge the CARUS and LIMIA team for their help. PA-L was supported by a Ramón y Cajal (Ref. RYC2020-029629-I) postdoctoral grant from the State Research Agency of the Spanish Government.