Precision Farming in Aquaculture: Assessing gill health in Atlantic salmon (Salmo salar) using a non-invasive, AI-driven behavioural monitoring approach in commercial farms

Precision Farming in Aquaculture: Assessing gill health in Atlantic salmon (Salmo salar) using a non-invasive, AI-driven behavioural monitoring approach in commercial farms | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Precision Farming in Aquaculture: Assessing gill health in Atlantic salmon ( Salmo salar ) using a non-invasive, AI-driven behavioural monitoring approach in commercial farms Meredith Burke, Dragana Nikolic, Pieter Fabry, Hemang Rishi, Trevor Telfer, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6653910/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 11 You are reading this latest preprint version Abstract As the aquaculture industry grows, more sophisticated technology is required to monitor farms and ensure good fish welfare, in line with the precision livestock farming concept. Using behaviour as a non-invasive monitoring tool, combined with artificial intelligence, enables greater control over farm management. This study aimed to assess temporal changes in farmed Atlantic salmon ( Salmo salar ) group behavioural profiles related to fish health and welfare. A machine vision algorithm applied to feed cameras on commercial farms was used to determine whether changes in gill health would induce visible group behavioural changes. Video cameras were deployed in all cages at two Scottish Atlantic salmon marine farms. One cage at each farm was also equipped with additional cameras (5 and 4 at sites A and B, respectively) to provide higher spatial coverage of fish behaviour and distribution. The algorithm processed video footage from these cameras and produced behavioural data termed ‘activity’ (%), which encompasses fish abundance, speed, and shoal cohesion. Additionally, gill health, Operational Welfare Indicators (OWI), mortality, and Specific Feeding Rate (SFR) were scored weekly at both sites. During summer 2023, gill health issues arose at both farms, leading to fish stress reflected in the behavioural data. For two months prior to the onset of poor gill health, the average (± standard deviation) activity levels of the fish across all cages were 25.6 ± 10.5% and 24.9 ± 7.0% for Farm A and B, respectively. After gill health was compromised, the activity rose significantly for two months in all cages with a mean of 43.6 ± 15.1% and 32.6 ± 9.6%, respectively. A generalised linear mixed model revealed that Proliferative Gill Disease (PGD) was the main driver of this increase in activity. This increase in activity coincided with fish migration to the centre of the cage, meaning tighter shoaling, which is a normal stress response often seen in relation to predators and other environmental or health stressors. The use of behaviour as a non-invasive welfare indicator and the potential to use artificial intelligence to automate the process of behavioural identification allows farmers to improve welfare conditions and ensure industry sustainability. Atlantic salmon Aquaculture Fish behaviour Welfare Gill health Machine learning Precision farming Figures Figure 1 Figure 2 Figure 3 Figure 4 1. Introduction Measuring the behaviour of cultured fish is challenging but critical for aquaculture, as it provides valuable insights into fish health and welfare being one of the 5 domains used to understand the welfare status of the animals (Mellor and Beausoleil, 2015 ). However, observing behaviour is inherently difficult because fish are underwater, often with poor visibility, and it can be unclear which behaviours are most useful to monitor. Previous studies on the behaviour of farmed Atlantic salmon smolts ( Salmo salar ) in sea cages have focused on the environmental drivers of their swimming depth and speed (Oppedal et al., 2011 , Burke et al. 2025 ). Salmon appear to distribute themselves heterogeneously throughout the cage, largely basing their preferred depth on environmental gradients, such as temperature and photoperiod (Oppedal et al., 2011 ; O’Donncha et al., 2021 ). Fish shoaling, when multiple individuals group together socially, is a common anti-predator stress defence mechanism, however fish may shoal as a response to other stressors like changes in environmental or health status (Pitcher 1983 ; Jaervi and Uglem 1993 ; Hoare et al., 2004 ). A previous study found that an unpredictable stressor (confinement) resulted in a higher shoal cohesion response in sea bass (Cerqueira et al., 2020 ) and a temporal preference for higher temperatures in zebrafish as a stress coping mechanism (Rey et al. 2015 ). Other studies found higher shoaling in response to predator stimuli in banded killifish, minnows, and zebrafish, (Magurran and Pitcher, 1987; Krause and Godin, 1994 ; Hoare et al., 2004 ; Miller and Gerlai, 2012 ). However, there are few studies that explore the role that parasites and disease may play on the behaviour of fish in wild or sea cage systems, with most studies considering behaviours (e.g. locomotion, foraging, shelter use) in laboratory settings (e.g. Santos et al., 2011 ; Øverli et al., 2014 ; Ruehle and Poulin 2021 ). Studies investigating fish response to stressors have traditionally employed invasive methods such as using plasma cortisol analysis (e.g. Madaro et al., 2015 , Lai et al., 2021 , Samaras et al., 2021 ) requiring crowding, handling, anaesthetising, and blood sampling which are inherently stressful for the fish. However, using behaviour as a non-invasive metric for detecting poor fish health holds promise for enhancing fish welfare (Li et al., 2022 ). Behavioural changes often precede outward signs of illness, providing farmers with an early warning signal to prevent disease rather than merely treating the symptoms (Juell 1995 ). Additionally, changes in feeding behaviour serve as a critical indicator of potential disease or stress in fish. Research has demonstrated a decrease in feeding motivation in response to acute stressors across vertebrates (Folkedal et al., 2012 ). Studies have shown that all vertebrates exhibit stress-related suppression of appetite and food intake, as well as enhanced energy mobilisation (Carr 2002 ; Hvas et al., 2024 ). This phenomenon is evolutionarily advantageous, as appetite suppression reduces the risk of predation, given that foraging in the presence of predators heightens mortality rates (Carr 2002 ). Moreover, metabolic energy is directed toward vital activities such as respiration, locomotion, and tissue repair, all of which contribute to individual survival, prioritising these functions overgrowth and reproduction (Wendelaar Bonga, 1997 ; Deck et al., 2017 ). Specific feeding rate, a composite measure incorporating specific growth rate and feed conversion rate, is widely utilised by the aquaculture industry to gauge feeding efficacy. Higher specific feeding rates signify the implementation of efficient and effective feeding regimes (Kolsäter, 1995 ). These non-invasive tools represent examples of Operational Welfare Indicators (OWI), measures used to assess the overall welfare status of cultured fish and are pivotal for assessing fish welfare on-site within production systems and husbandry practices (Noble et al., 2018 ). OWIs encompass both indirect environment-based and direct animal-based measures, which can be further categorised at a group or individual level. Implementation of OWIs is indispensable for ensuring that the welfare needs of fish are adequately addressed. Standardised tools have been developed to monitor fish welfare through direct observations, focusing on a subset of individual animals within cages for external signs of injury or illness, as well as through assessments of group-level behaviours such as feeding patterns. Over the last decade, the increase in marine fish gill diseases is recognised as a major health and welfare problem worldwide (Boerlage et al., 2020 ). Compromised gill function disrupts fish respiration and osmoregulation, causing breathing difficulties, and loss of salt and blood gas control. There are many causes of gill disease including parasitic gill disease, viral/bacterial gill disease, plankton/algae associated gill disease, or chemical-related gill disease (Rodger 2007 ; Boerlage et al., 2020 ). The parasite Neoparamoeba perurans , through the development of Amoebic Gill Disease (AGD) is often the most reported due to the distinctive pathology in salmon and is recognisable on-site (Boerlage et al., 2020 ). AGD can cause typical losses to the Scottish salmon industry of 10–20%, with some sites as high as 70% (Marine Scotland, 2012 ; Shinn et al., 2015 ). It is also the cause of significant mortalities in commercial salmon fisheries in Australia, USA, Canada, France and Ireland. However, gill disease is often multifactorial, due to a combination of one or more of the above causes; this is called complex gill disease (CGD; Herrero et al., 2022 ). CGD encompasses Proliferative Gill Disease (PGD), which is a gill condition described in Scotland and Ireland as a generic term for the proliferative changes in the gill epithelium while causing gross lesions easily identified in the field (Boerlage et al., 2020 ). In Norway, a similar pathology-based diagnosis termed Proliferative Gill Inflammation (PGI) is distinguished by specific histopathological changes including lamellar vascular changes, inflammation, necrosis and epithelial cell hyperplasia (Herrero et al., 2018 ; Boerlage et al., 2020 ). Notably, PGD differs from PGI in that it exhibits proliferative histological features but with unknown aetiology (Herrero et al., 2018 ). It can result from the interaction of environmental and husbandry conditions as well as infection by pathogens and parasites (such as N. perurans ), particularly during the summer and fall months. While AGD can be diagnosed through gill sampling and histopathology to confirm the presence of the parasite, commercial producers typically assess gill health (both AGD and PGD) on-site using a categorical scoring system (Taylor et al., 2009 ). The progression of climate change and increased sea temperatures is directly linked to an increase in the organisms that cause damage to the gills and water quality changes, algal blooms, handling, and temperature changes can all have a cumulative stress effect increasing susceptibility to gill disease (Klinger et al., 2017 ; Foyle et al., 2020 ). Additionally, cumulative stressors can result in susceptibility to other diseases such as infection by copepod parasites, or sea lice ( Caligus elongatus and Lepeophtheirus salmonis ; Orr et al., 2020 ; Herrero et al., 2022 ). Sea lice have both lethal and non-lethal impacts on salmon, primarily resulting in chronic stress and reduced growth rate, with L. salmonis reported as the largest parasitic threat to the salmon aquaculture industry (Mustafa et al., 2000 ; Susdorf et al., 2017; Zhang et al., 2023 ). Advancements in artificial intelligence (AI) has resulted in improved monitoring techniques especially in the ocean sector where usual methods are not feasible. AI can enhance real-time or near-real time monitoring by automating the collection and analysis of data (Villegas-ch et al., 2024 ). AI-based algorithms have the capability to process and analyse large datasets, image or video data through machine vision techniques. AI can detect patterns and thereby any factors contributing to anomalies to those patterns. Thus, it can contribute to the development of early-warning systems, such as the detection of environmental or health-related parameters that may cause poor welfare and mortality (Mandal and Ghosh 2023 ). Moreover, they can collect high-quality data in a non-invasive manner, avoiding damage to the fish that often occurs with traditional manual methods (Liu et al., 2023). Machine vision techniques have been used in aquaculture in the past, primarily for biomass estimation, body length measurements, fish counting and feeding behaviours (Yang et al., 2021 ; Liu et al., 2023; Zhang et al., 2023 ). Fish behaviour in relation to stress has also been examined, however the focus is often on individual fish swimming speed, acceleration and turning angle (Li et al., 2022 ). While this type of analysis is relatively straightforward in tanks or recirculating aquaculture systems, it can be more difficult to assess in large commercial sea cages with low light, high noise, fish deformation, fish overlap and occlusion, and low field of view in larger cages (Yang et al., 2021 ). The present study utilised cameras deployed within cages in two commercial aquaculture farms in Scotland, focusing on group behaviour instead. This was quantified using an AI-based algorithm developed by Observe Technologies ( https://observe.tech/ ). The aim was to investigate fish group behaviours such as cohesive grouping or shoaling, feeding, and fish distribution in response to a common health-related stressor. This study hypothesised that changes in gill health would elicit observable group-level behavioural changes in Atlantic salmon. To test this, long-term behavioural data were collected and analysed a posteriori. The findings support the potential of precision fish farming (PFF), particularly when combined with system modelling and machine learning, as a significant advancement for aquaculture systems and a promising approach to addressing welfare-related stressors. 2. Methods 2.1 Study Sites The two commercial Atlantic salmon farms used in this study are located on the Isle of Harris and Lewis, in the northwest of Scotland (Fig. 1A,B,D). Farm A consists of 16 circular cages (13 stocked during the study) in two parallel rows, each 90 m in circumference and 15 m in depth with cylindrical nets (mesh size:18 mm) and Seal Pro anti-predator nets to a depth of 10 m (Table 1). Farm B contains 16 circular cages (10 stocked during the study) in two parallel rows, each 120 m in circumference and 10 m in depth (mesh size 18 mm) with Seal Pro anti-predator nets. Additionally, the fish were initially fed ad libitum and later transitioned to a twice-daily feeding schedule, which was adjusted based on their age class. Table 1 Details of the two fish farms used in this study. Farm Cages stocked during study Cage volume (m 3 ) Initial stocking week Mean smolt weight at stocking (g) Mean [range] current speed (m s − 1 ) * A 13 9 669 2023-02-13 109 0.06 [0-0.16] B 10 11 459 2023-05-08 87 0.09 [0-0.38] * The current speed was measured over 90 days prior to stocking the sites 2.2 Data Acquisition 2.2.1 Fish Behaviour There were 4–5 video camera modules (Sony model: IPCM-3516DS385-D29-AZ3611) installed by AKVA Group (www.akvagroup.com) within one, randomly allocated cage per site pointing towards the surface. These were deployed to obtain higher spatial resolution of fish behaviour and to identify optimal camera placement within the cage for capturing behavioural data from a high proportion of the population. In Farm A, they were positioned as 3 down the centre (approximately 4 m, 9 m, 14 m), and 2 at 9 m on the inner and outer areas of the cage, respectively (Fig. 1C left). As Farm B is a shallower site, only 2 cameras were placed down the centre (4 m and 9 m), with 2 at 5 m on the inner and outer areas of the cage, respectively (Fig. 1C right). The remaining cages were equipped with a single camera pointing upwards, positioned centrally at mid-depth (~ 9 m in Farm A and ~ 7 m in Farm B), to allow comparison with the higher resolution study cage. The cameras recorded at 25 frames per second, at a resolution of 1920 x 1080 pixels, and a video bit rate of 16 Mb per second. Videos were recorded continuously during daylight hours between May (June) and September (October) for Farm A (B). Daylight hours shift as summer progresses, with lowest duration in October at 12 hours and peaking in June at 18 hours. Each camera contained a pressure (TE connectivity US300; accuracy of ± 0.15% full scale output) and temperature sensor (RDO PRO-X; accuracy of ± 0.1ºC and resolution of 0.01 ºC), which recorded continuously for the duration of the study. Additionally, there were single cameras deployed in the centre of the remaining cages in each farm, allowing for activity data from these cameras to be compared to the study cage to investigate whether similar trends were observed using a single-camera system. Video recordings were analysed using an algorithm developed by by Observe Technologies (www.observe.tech), designed to extract behavioural data relevant to aquaculture operations. This algorithm employs a patented (EP3644717) convolutional neural network to heuristically estimate fish activity by identifying key behavioural features. These features include, but are not limited to, schooling cohesion, average distance from the camera, the number of fish detected per frame, and swimming speed. Rather than tracking individual fish, the model is trained globally, recognising behavioural patterns from annotated datasets. Since all features are processed within a shared network architecture until the final layer, their individual contributions to the overall activity score remain undefined. The model continuously refines its parameters through standard deep-learning optimisation techniques, generating a normalised activity score ranging from 0 to 100% (example in Fig. A.1). To ensure reliability, the model undergoes comprehensive validation using separate training, validation, and test datasets, following best practices in deep learning. Further, aquaculture specialists routinely assess the model’s outputs to verify their practical relevance, and for this study, they conducted an additional review to ensure accuracy before the data was provided to university researchers. Independent validation of the algorithm was also carried out by two observers from the University of Stirling, using video footage captured by the same AKVA feed cameras. The algorithm’s 'activity' output was correlated to fish abundance counts from 55 randomly chosen video stills using Pearson correlation (α = 0.05). Additionally, the Intraclass Correlation Coefficient (ICC) was calculated to assess interobserver reliability. 2.2.2 Fish Health and Welfare Each farm provided health and welfare-related parameters used in this study. Daily Specific Feeding Rate (SFR) was recorded for each cage as the % body weight fed: Where biomass was determined through sample and mortality weights and number of fish in each cage. At commercial aquaculture sites, farmers visually assess fish for OWIs, sea lice, and gill health. Given the demanding nature of farm operations and limited staffing, manual assessments remain a standard and practical protocol. On a weekly or bi-weekly basis (depending on water temperature), 10 fish per cage were examined for various Operational Welfare Indicators including fin and eye damage, skin lesions, scale loss and visual K-factor with severity scored from 0 (healthy) – 2 (severe; Table 2). Table 2 Operational Welfare Indicators (OWI) scoring used in this study. Eye damage Fin damage Skin lesions Scale Loss K-Factor 0 Both eyes perfect Dorsal, pectoral and caudal fins intact Skin intact All scales are intact, no patches of lost or damaged scales; Fish is ‘smooth’ Fish are well shaped and firm to touch; Has a ‘belly’. 1 Damage to one eye (unilateral) Such as: - Rupture - Cloudiness - Bleeding - Cataract - Swelling < half of fin remaining 1 + small wounds (< 10p coin); subcutaneous tissue intact Small areas of scale damage or loss present, but skin is intact (i.e. no bleeding or lesions); Affected area is no bigger than 10p coin Fish long, but not have as much depth as expected; Slightly soft to touch (esp. around the belly area) 2 Damage to both eyes Little of fin remaining Large (> 10p coin), severe wounds; muscle exposed Multiple areas of scale loss or damage; Skin may look dimpled due to water ingress Fish long and thin; head significantly oversized compared to body. Adapted from The Scottish Salmon Company (Bakkafrost Scotland) and FISHWELL handbook (Noble et al., 2018) These 10 sampled fish are then examined for sea lice with counts per fish for gravid females (females with egg sacs attached). Gravid females were chosen as they serve as a key indicator for assessors when determining whether treatment is necessary. Lastly, fish were assessed weekly for gill health using a standardised scoring protocol. Gill health was assessed by farmers using a standardised scorecard (Fig. A.2), evaluating lesions and lamellar thickening to assign AGD and PGD scores ranging from 0 (healthy) to 5 (necrotic; Table 3). The distinct visual differences between AGD and PGD allow for rapid on-farm evaluation and treatment decisions, as laboratory diagnostics can take time to yield results. While somewhat subjective, this method is supported by farmer training and extensive experience. Table 3 Gill health (AGD and PGD) scoring system used to estimate severity on-site. Score AGD PGD 0 Healthy No sign of infection Healthy, no thickening 1 Very light 1 white spot/light scarring Very slight thickening/few lamellae affected 2 Light 2–3 spots/small mucus patch Frequent thickening but tips only 3 Moderate Established thickened mucus patch/spot groupings up to 20% of gill Almost all lamellae have thickened tips/some have thickenings progressing to ~ 50% of the length of lamellae 4 Advanced Established lesions covering up to 50% of gill Most lamellae have thickening progressing to > 50% of the length of lamellae 5 Necrotic Extensive lesions covering most of gill surface (> 50%) Almost all lamellae are thickened along entire length AGD details adapted from Taylor et al., 2009; PGD details adapted from The Scottish Salmon Company (Bakkafrost Scotland) 2.3 Data collection This study was conducted for 4 months for each site, from May 10 to September 10, 2023 for Farm A and June 22 to October 13, 2023 for Farm B, these differences in time frames were related to the later stocking of Farm B while maximising data collection. At the beginning of July for Farm A and in mid-August for Farm B, there was an increase in the activity of the fish, prompting a divide in the dataset for 2 months pre/post rise in activity. Activity data was preprocessed through a rolling average of 15 minutes to identify trends and remove excess noise. Additionally, activity data collected during feeding periods were excluded to ensure that it did not influence the observed behavioural patterns. Utilising the depth sensors integrated into each camera, the depth of maximum activity was determined by identifying the camera with the highest activity at each time point and recording its corresponding depth. 2.4 Statistical Analysis All statistical analyses were conducted using the Python programming language (https://www.python.org/), with statistical significance set at α = 0.05. To independently validate the algorithm, 55 videos were randomly selected, and the number of fish visible on screen was counted by two evaluators. These abundance counts were then correlated with the corresponding activity (%) measured at each time point. The Intraclass Correlation Coefficient (ICC) was used to assess inter-rater reliability and to determine both absolute agreement and consistency between the evaluators. Mann-Whitney test was used to determine whether there were significant differences in activity, AGD, PGD, SFR, and mortality between the two time periods for each farm. The maximum activity was used for the multi-camera study cage, and this was aggregated at 15-minute intervals. Following this, model II linear regressions were performed to determine correlation between weekly-averaged maximum activity and gill health score with ANCOVA to determine significance for each farm independently. Coefficients of variation (COV) were used to determine the extent to which the level of dispersion of the depth of maximum activity was around the mean. Additionally, Aligned Ranked Transform ANOVA allowed for a comparison between each of the cages before and after the increase in activity. Lastly, a Generalised Linear Mixed Model was used to determine the effects of weekly-average temperature, AGD, PGD, and sea lice ( L. salmonis gravid females) on weekly-averaged activity with random effects included for cages and farms. All variables were z-transformed prior to analysis. Raw data is available through DataSTORRE (http://hdl.handle.net/11667/241). 3. Results 3.1 Model Validation To validate model efficacy, the algorithm was independently validated by using videos recorded using the same AKVA feed cameras. Evaluators assessed the algorithm’s 'activity' output and demonstrated a strong correlation with fish abundance counts from 55 clips (R 2 = 0.70, p < 0.001). Independent validation demonstrated a strong correlation (R² = 0.70, p < 0.001) between fish abundance observed in feeding camera footage by human assessors and the still images from the same videos used to calculate activity via the algorithm. The ICC determined high correlation both in the absolute agreement (ICC = 0.99) and the consistency (ICC = 0.98) between the two evaluators. Both reflect strong agreement between the observers that was statistically significant (p < 0.001). While fish speed and cohesion could not be separately validated, it is assumed that the 30% variation not captured is related to the speed and cohesion components. 3.2 Farm A: Activity and Health In Farm A, there were seasonal effects on temperature. At the start of the study, on May 10, the temperature was 10.0 ºC within the study cage, with an upward trend reaching a peak of 14.7 ºC by June 23 (Fig. 2 ). Throughout the remainder of the trial period the temperature maintained a consistent range, oscillating between 14.6 ºC and 15.3 ºC, ultimately settling at 15.4 ºC by the end of the study. While only temperature data was presented in the results for both farms, oxygen, turbidity, and salinity were also monitored throughout the study period, however these parameters showed only minor daily fluctuations and did not exhibit significant trends over the sampling period. As the behavioural responses were observed over several months, daily variations in these parameters were unlikely to significantly influence long-term behavioural patterns and therefore these results are included in the appendix (Fig. A.3). The depth of maximum fish activity, determined through the in-camera depth sensors, elucidated both the predominant depth distribution of the fish within the cage and the degree of their spatial cohesion (Fig. 2 ). There was a pronounced shift in the activity and depth of the fish on July 7. In the 2 months prior to July 7, the average depth at which fish congregated was 10.0 ± 2.8 m, with a COV of 34.5%. For 2 months after July 7, this depth of maximum activity then decreased to 9.0 ± 1.4 m (U = 1.32 \(\:\:\bullet\:\) 10 7 , p < 0.05) with a notable decrease in the COV to 18.7% as the fish coalesced towards the central region of the cage, indicative of heightened cohesion among individuals. This transition occurred rapidly, with the COV decreasing from 28.5% one week prior to July 7 to 13.1% the week after. Moreover, there was a significant increase in maximum activity from 37.5 ± 13.1% to 60.8 ± 11.5% (U = 2.5 \(\:\bullet\:\) 10 6 , p < 0.001) for 2 months before and after July 7, reflecting the intensified group cohesive behaviour as potential shoaling behaviour. This transition was rapid as well, from 32.3 ± 11.2% one week prior to 56.8 ± 17.0% on average one week post July 7. Concurrently, gill health assessments were conducted throughout this period, with the average AGD and PGD scores initially recorded as 0.4 ± 0.3 and 0.4 ± 0.2, respectively. Following the increase in activity, both AGD and PGD scores rose to 0.8 ± 0.7 and 1.2 ± 0.7 respectively. Subsequently, a correlation analysis revealed a non-significant association between AGD and activity (p > 0.05, R 2 = 0.07), while a significant correlation was observed between PGD and activity (p < 0.001, R 2 = 0.42; Fig. 3 ). There was a significant decline in the specific feeding rate (SFR) at Farm A, from an average of 2.3 ± 0.7 prior to the augmented activity levels of the fish, to an average of 1.1 ± 0.3 after (U = 1477, p < 0.001). Furthermore, mortality rates within the cage exhibited a slight, though non-significant, increase from 0.03 ± 0.02% to 0.08 ± 0.1% with the increased activity on July 7 (U = 23, p = 0.24). 3.3 Farm B: Activity and Health The study period for Farm B commenced later in the year, on June 22, with temperatures recorded at 11.8 ºC. Over the course of the study, temperatures steadily increased, reaching 14.7 ºC by September 7, before declining slightly to 13.3 ºC by October 13. On August 19, there was a significant increase in maximum fish activity from 32.5 ± 10.5% to 51.1 ± 19.0% (U = 5.9 \(\:\:\bullet\:\) 10 5 , p < 0.001). This occurred concurrently to a significant increase in depth from 5.8 ± 1.8 m to 6.0 ± 1.9 m (U = 1.16 \(\:\:\bullet\:\) 10 6 , p < 0.001), though it should be noted that this cage had a shallower depth overall. In a similar trend as Farm A, the maximum activity and COV transition quickly for the week prior and post-August 19, whereby the activity increased from 27.8 ± 10.2% to 48.7 ± 9.2%, and the COV declined sharply from 30.4–13.2%. In contrast to Farm A, Farm B exhibited overall lower gill health scores, with AGD scores recorded at 0.1 ± 0.2 prior to the heightened fish activity and increasing to 0.4 ± 0.3 after. Similarly, PGD scores were recorded at 0.01 ± 0.02 prior and 0.7 ± 0.4 post-increase in activity. A significant correlation was established between AGD and activity at this site (p < 0.05, R 2 = 0.35), however a stronger correlation was observed between PGD and activity (p 0.05). However, there was a significant increase in mortality rates, rising from 0.03 ± 0.01% to 0.2 ± 0.3% concurrent with the heightened activity observed on August 18 (U = 7, p < 0.01). 3.4 Farm-wide effects The investigation extended to the remaining cages at each site, each equipped with a single camera centrally positioned at mid-depth, around 9 m for Farm A and 7 m for Farm B. A similar pattern emerged, characterised by significant increases in activity concomitant with increased PGD scores, decreased SFR, and increased mortality rates (Fig. 4 ). In Farm A, 2 months prior to July 7, the activity levels were 25.6 ± 10.5%, increasing to 43.6 ± 15.1% 2 months thereafter, with all cages demonstrating a significant elevation ([3.8–22.9%; min – max difference within cages]; p < 0.01). Concurrently, AGD scores increased from 0.3 ± 0.06 to 0.51 ± 0.06 [0.04–0.4], with PGD scores exhibiting a more pronounced increase from 0.3 ± 0.09 to 1.1 ± 0.2 [0.4–1.2] across all cages, each showing a significant rise (p < 0.05). SFR exhibited a significant decline in all cages, from 1.3 ± 0.1 to 0.7 ± 0.1 ([-0.4– -1.2]; p < 0.05). Furthermore, there was a slight rise in mortality after July 7, on average from 0.2 ± 0.3% to 0.3 ± 0.3%, though this relatively small difference can be attributed to the low median mortality rate, with the upper quartiles in most cages indicating a trend towards higher mortality in the 2 months following July 7. There were 5 cages (6, 7, 8, 10 and 12) that did exhibit significantly elevated mortality (p < 0.05). Similarly, in Farm B, the average activity levels increased from 24.9 ± 7.0% to 32.6 ± 9.6% on August 18, with all cages, except cages 3 and 5, exhibiting significant increases ([5.2–16.8%; min – max difference within cages]; p < 0.05). It should be noted however that the latter two cages had trends towards an increase even though it was not statistically significant. AGD scores increased from an average of 0.1 ± 0.2 to 0.6 ± 0.4 [0.3–0.7], while PGD scores increased from 0.1 ± 0.2 to 0.9 ± 0.5 which was a significant increase in all cages ([0.6–1.2], p < 0.05). SFR decreased from average of 1.7 ± 0.3 to 1.4 ± 0.3 [-0.06 – -0.7]; all cages were significantly lower except for cages 2 and 8 (p < 0.05). The average mortality rate rose similarly to Farm A, from 0.2 ± 0.9% to 0.3 ± 0.5%, again with around half the cages (2, 4, 6, 8 and 9) showing significant increases in mortality (p < 0.05). Furthermore, a generalised linear mixed model was employed to examine the influence of temperature, AGD, PGD and sea lice (gravid L. salmonis ) on activity levels. The model revealed a significant effect of PGD on activity (0.32 [0.16–0.48], p 0.05). The model’s predictive capacity for activity was substantial with an R 2 marginal of 0.25 [0.12–0.39; 95% confidence interval], particularly when accounting for the variation at the cage and farm levels (R 2 conditional : 0.60 [0.54–0.64]). Regarding the Operational Welfare Indicators assessed at each farm, no significant increases were observed for any parameter at either farm (p > 0.05). Nonetheless, at Farm A, a trend toward leaner fish was discernible, evidenced by an increase in the occurrence of visual K-factor scores above 0 following the increase in activity, rising from 8 to 18 instances across all cages, though this was not mirrored at Farm B. 4. Discussion Gill disorders are a growing economic and welfare concern for the salmon aquaculture industry worldwide. Deterioration in gill health can precipitate significant mortality events, hinder salmon growth and production, and compromise overall welfare (Rozas-Serri, 2019 ). The escalating temperatures due to climate change exacerbate factors contributing to compromised gill health. These factors may encompass physical stressors, including abrasions of the gill surface, as well as pathogenic or parasitic agents such as N. perurans , causing AGD, hydrozoan jellyfish or other gelatinous zooplankton, as well as phytoplankton such as diatoms, which can instigate gill lesions (Kent et al., 1995 ; Baxter et al., 2011 ). The advancement of aquaculture monitoring has been significantly enhanced with the emergence of artificial intelligence (AI). The use of AI algorithms, through machine learning methodologies, facilitates the processing of extensive datasets, including the analysis of image and video data to discern patterns and trends, thereby enabling investigations such as the present study (Mandal and Ghosh, 2023 ). Leveraging AI for pattern recognition can result in the detection of any anomalous behaviours, indicative of stress and compromised health responses. In this study, an algorithm based on machine vision was employed to analyse videos sourced from commercial fish farms to quantify fish group behaviour and compare temporal behavioural profiles. The resulting ‘activity’ parameter is a combination of the relative abundance of fish within the video frame, perceived average inter-fish distances, and group schooling behaviour in one metric to describe the fish behaviour overall. Since the machine vision model is trained end to end, there is no overarching weighted model that dynamically adjusting feature importance. Instead, the model learns each feature independently while estimating relative activity. This innovative approach allows researchers, as well as farmers, to establish correlations between behaviours, specifically related to group social cohesion and fish abundance in certain cage locations, and external factors, a task that traditionally demands considerable time and effort. Moreover, while many algorithms focus on individual fish behaviour (Liu et al., 2023), this approach allows farmers to observe the behaviour of the fish as a group, giving a more holistic view. Deteriorated gill health was observed at two study sites situated within the Western Isles of Scotland. The initial occurrence was documented at Farm A in early June followed by a subsequent manifestation at Farm B in mid-August, 2023. Both sites exhibited increased levels of fish activity coinciding with higher gill health scores, particularly in terms of proliferative gill disease (PGD) scores, confirming the hypothesis that compromised gills would alter salmon behaviour. However, intriguingly, no significant correlations were discerned between the prevalence of Amoebic Gill Disease (AGD) specifically, ambient temperature or the presence of gravid female L. salmonis. The utilisation of gravid females in this case was to investigate whether activity could be used as an indicator for sea lice infestation, to inform managers and potentially allow for pre-emptive actions prior to visual inspections, as gravid females have been correlated to the external infection pressure (Kragesteen et al., 2021 ). However, the investigation failed to establish any significant correlation between the presence of gravid female lice and alterations in fish activity levels. It is important to note that there were few sea lice counted during the study period, as such, this may be of interest to explore further in future studies. Most studies focus on AGD as it can be detected relatively easily through histopathology and narrowed down to a singular amoeba rather than PGD which can be caused by a variety of infectious agents and has limited diagnostic test to detect it (Boerlage et al., 2022). Thus, the tests and treatments used for PGD are simply those used to test and treat AGD, specifically total histopathology score by Mitchell et al., ( 2012 ), or by gross gill observations described here. This study provides further information on how PGD can impact the behaviour and welfare of the fish, that cannot be observed through AGD tests alone. It should be noted that this study does not aim to diagnose the specific causes of PGD. Although turbidity remained similar throughout the study, few micro-jellyfish were observed and no changes in plankton counts, other potential factors influencing fish behaviour cannot be ruled out. Although a more detailed diagnostic approach could provide clarity on the etiology, such methods were beyond the scope of this farm-based study, where visual gill scoring by experienced farmers is the standard practice. This method has shown moderate to good agreement with histopathological diagnosis (Adams et al., 2004 ), though reliability decreases when symptoms are mild (Clark and Nowak, 1999 ). Refinements to reduce subjectivity and explore additional behavioural indicators, particularly in early or mild cases, could enhance diagnostic accuracy. Despite this limitation, the observable deterioration of gills was linked to behavioural changes occurring over several months, indicating a strong relationship between gill health and fish behaviour. The authors acknowledge this as a limitation and suggest that future research incorporating more precise diagnostic techniques could further elucidate the relationship between specific gill diseases and behavioural responses. Several OWIs were investigated in this study. These are important for fish welfare and can be easily implemented by farmers in their daily husbandry practices (Noble et al., 2018 ). Various types of welfare indicators exist, each with its own set of strengths and weaknesses. For instance, environment-based indicators such as temperature and oxygen levels are rendered ineffective in the absence of extremes, as observed in this study where such indicators remained relatively stable. Consequently, animal-based indicators, both at the group- (behaviour, feeding) and individual- (gill status, external inspections of the fish, k-factor) levels, were employed. As behaviour began to change and cohesive group behaviour or shoaling was more prominent, there was a rise in gill health scores. However, there was no significant change in the outward signs of illness (e.g. no change to the eyes, fins, skin). Despite the advantage of individual animal-based indicators (ABI) in detecting outward signs of poor health, the sampling of only 10 fish per cage out of a population exceeding 30,000 individuals highlights a weakness. Additionally, relying solely on individual indicators may be insufficient, as by the time outward signs of illness become apparent, the fish may already be compromised. In contrast, group-based indicators (GBI) offer a more holistic approach, allowing for the observation of most fish within the cages. Moreover, behaviour is the only welfare indicator that provides farmers direct insight into the subjective experience of fish, while OWIs (i.e. physiological or external parameters) often reflect indirect, consequential effects of environmental or health stressors (Noble et al., 2018 ) and are usually based on negative experiences. Though these indirect indicators remain valuable for comprehensive welfare assessments, behaviour uniquely captures the immediate responses of fish, making it a critical tool for early detection of welfare concerns in farm settings. This means that managers can have a better understanding of the state of the fish in a non-invasive way, simply through their position in the water column and proximity to each other (Martins et al., 2012 ; Noble et al., 2018 ). Changes to fish behaviour can be a warning sign that the fish are experiencing stress and reduced welfare. For example, social cohesion, such as schooling or shoaling, serves as an effective antipredator adaptation in fish (Morgan and Godin 1995; Krause et al., 2002 ; Huntingford et al., 2012 ). This response allows for the potential dilution effect, as it decreases the odds of being predated upon. This mechanism may also translate into a response to compromised health, as stressed fish are more likely to shoal (Kleinhappel et al., 2019 ). Moreover, a recent study on the effect of fish schooling on swimming noise production suggested that schooling fish emit lower sound levels, potentially shielding the group from predators while optimising hydrodynamic efficiency (Zhou et al., 2024 ). However, the consequences of this behaviour are not fully understood; while increased shoaling is typically associated with elevated risk of disease transmission, such as AGD, among nearby conspecifics, it may also confer protective benefits through mechanisms akin to herd immunity or the dilution effect (Mikheev, 2009 ). Further research is needed to clarify whether shoaling primarily facilitates the spread of pathogens or may serve as a defence strategy against them. Moreover, such aggregation behaviour can lead to overcrowding in localised areas of the cage, potentially reducing oxygen availability and further impacting welfare, although this was not observed in the present study. Acute stressors also disrupt feeding behaviour, leading to reduced food intake and growth, which can affect their energetic costs and ultimately impact their welfare (Morgenroth et al., 2024 ). This study observed a decline in specific feeding rates following the onset of gill damage. The impact of compromised gill health on feeding behaviour was demonstrated primarily at Farm A, where a significant decline in the SFR was observed following the onset of deteriorated gill health concomitant with increased activity levels. While a downward trend in SFR was also discernable at Farm B, it was not statistically significant. Reduced feeding often considered a primary symptom of stress in fish, alongside changes to swimming behaviour, underscores the importance of capturing these changes through metrics such as SFR (Conte, 2004 ). This study also documented rises in mortality within the majority of cages at Farm A, along with discernible trends towards increased mortality, albeit not statistically significant, at Farm B. While mortality is a commonly used group-based OWI in production systems, it is important to acknowledge the inherent limitation, as it signifies an irreversible stage in the deterioration of fish welfare, precluding the implementation of corrective measures (Ellis et al., 2012 ; Noble et al., 2018 ). The limitation is underscored by the findings in this study, as significant mortalities were observed at Farm A, correlating with the highest levels of AGD and PGD, alongside significant increases in activity. Conversely, Farm B, characterised by lesser but still notable increases in gill health issues and activity, did not exhibit significant rises in mortality in most of the cages. This discrepancy shows that relying on mortality alone as an indicator would overlook the manifestations of poor welfare discerned in this study through behavioural changes. If mortality is to remain an operational welfare indicator, it signals issues at too late a stage. To enhance its utility, individual-based assessments should be complemented by comprehensive behavioural monitoring. This combined approach enables earlier detection of welfare concerns, improving fish health and welfare outcomes. The effect sizes differed notably between the two sites, with Farm A exhibiting poorer gill health, heightened activity levels and mortality, and more pronounced reduction in SFR compared to Farm B. The discrepancy in gill health scores between the two farms likely underlies the variation in response, with Farm A recording higher scores, resulting in a lesser increase in activity, and absence of a significant decrease in SFR at Farm B. Several factors may contribute to the observed disparities between the two sites. Firstly, it is pertinent to acknowledge the highly contextual natures of PGD, with the specific pathogenic composition underlying PGD manifestation varying across sites (Król et al., 2020 ). Furthermore, differences in environmental exposure levels warrant consideration, particularly regarding the location of Farm B, which is characterised by robust tidal-driven currents, potentially facilitating the transport of harmful agents such as N. perurans , micro-jellyfish, plankton, and other debris capable of damaging the gills. Additionally, temporal discrepancies in fish stocking, with Farm A stocked 3 months prior to Farm B, result in differences in fish size between the two cohorts. Lastly, variations in broodstock between the sites may further contribute to disparities in response to gill health challenges. Previous studies have elucidated the influence of genetic background on behavioural traits and disease susceptibility in Atlantic salmon (Davidson et al., 1989 ; Bui et al., 2017). However, it is important to note that despite the comparatively smaller magnitude of change in activity observed at Farm B, both sites exhibited analogous behavioural shifts in response to compromised gill health. This suggests a conserved behavioural response related to survival among Atlantic salmon, irrespective of genetics, age, or environmental conditions, underscoring the impact of poor gill health-induced stress on fish behaviour. As climate change progresses and temperatures continue to rise, the agents causing PGD are expected to proliferate, leading to higher rates of illness. Over the past 4 decades, sea surface temperature around the UK have exhibited a warming trend of approximately 0.3 ℃ per decade (Cornes et al., 2023 ). This warming trend may broaden the reproductive periods and improve winter survival of jellyfish and plankton, causative agents of PGD, at mid-latitudes (Licandro et al., 2010 ; Boero et al., 2016 ; Kennerley et al., 2022 ). Moreover, research indicates that temperatures exceeding 12 ℃ can trigger outbreaks of AGD, potentially increasing the attachment and growth capacity of N. perurans under elevated temperature conditions (Benedicenti et al., 2019 ). Consequently, the establishment of efficient early detection mechanisms for illnesses is imperative to mitigate stress and mortality among farmed fish. (Barreto et al., 2021). This study provides insight into the role of behavior in detecting PGD, facilitating the implementation of preventative or treatment interventions. Conclusion In this study, a pronounced increase in fish activity occurred at the emergence of poor gill health in two Atlantic salmon farms in Scotland. During the summer and early autumn of 2023, both farms experienced a significant increase in fish activity concomitant with the onset of PGD. This was accompanied by a decline in SFR and increase in mortality rates, particularly evident in Farm A. Given the escalating temperatures associated with climate change, it is conceivable that the prevalence of agents precipitating PGD will escalate, underscoring the need for the aquaculture industry to devise effective mitigation strategies. Gill health is a leading cause of mortality in the global Atlantic salmon industry, thus is it is imperative that studies continue in this area to improve the welfare of farmed fish. The findings of this study hold significant implications for future studies, which can delve deeper into elucidating the activity thresholds indicative of impending gill health deterioration. This could lead to the development of a non-invasive early warning system for identifying compromised gill health. Such a tool could prove invaluable in proactively addressing welfare concerns in commercial aquaculture systems. The main challenges currently facing aquaculture are based on the improvement of best management practices and the advancement of monitoring systems, both for the aquatic environment and for assessing fish behaviour, health, and welfare. Implementing a systematic and integrated welfare assessment framework that spans the entire production cycle up to slaughter is essential to improve the understanding of basic needs of the fish, mitigating stressors and stress responses, and ultimately enhancing productivity and product quality. The development of Precision Fish Farming (PFF), incorporating system modelling and machine learning, represents a significant advancement in aquaculture technology. This approach offers promising solutions to welfare-related challenges, including those explored in the present study. Declarations Ethics approval: This study was approved by the Animal Welfare and Ethical Review Body (AWERB) at the University of Stirling (Approval No. 2022 8795 8155) and was conducted in accordance with institutional and national ethical guidelines for animal research. Consent for publication: Not applicable Availability of data and materials: The datasets supporting the conclusions of this article are available in the DataSTORRE repository, http://hdl.handle.net/11667/241. Competing Interests: The authors declare that they have no competing interests. Funding: This study was supported by Innovate UK Smart Grants (10028961), in partnership with Observe Technologies and AKVA Group. CRediT authorship contribution statement: MB: Writing – original draft, Formal analysis, Investigation, Visualisation . DN: Writing – review & editing, Methodology, Software, Formal analysis, Investigation, Resources, Data curation . PF: Writing – review & editing, Conceptualization, Methodology, Software, Validation, Resources, Data curation . HR: Conceptualization, Methodology, Software, Validation, Resources, Project administration, Funding acquisition. TT: Writing – review & editing, Conceptualization, Methodology, Supervision . SRP: Writing – review & editing, Conceptualization, Methodology, Supervision, Project administration, Funding acquisition Acknowledgments: The authors are grateful to the dedicated personnel at each farm for their diligent maintenance of the cameras. Special thanks to Michael Caldwell and AKVA Group for deploying the cameras and promptly addressing any technical issues that arose. References Adams MB, Ellard K, Nowak BF. Gross pathology and its relationship with histopathology of amoebic gill disease (AGD) in farmed Atlantic salmon, Salmo salar L. Journal of Fish Diseases. 2004;27(3):151–61. Barreto MO, Rey Planellas S, Yang Y, Phillips C, Descovich K. Emerging indicators of fish welfare in aquaculture. Reviews in Aquaculture. 2022;14(1):343–61. Baxter E, Rodger H, McAllen R, Doyle T. Gill disorders in marine-farmed salmon: investigating the role of hydrozoan jellyfish. Aquacult Environ Interact. 2011;1(3):245–57. Benedicenti O, Pottinger TG, Collins C, Secombes CJ. Effects of temperature on amoebic gill disease development: Does it play a role? Journal of Fish Diseases. 2019;42(9):1241–58. Boerlage AS, Ashby A, Herrero A, Reeves A, Gunn GJ, Rodger HD. Epidemiology of marine gill diseases in Atlantic salmon (Salmo salar) aquaculture: a review. Reviews in Aquaculture. 2020;12(4):2140–59. Boero F, Brotz L, Gibbons M, Piraino S, Zampardi S. Ocean warming: impacts and effects of ocean warming on jellyfish. In: Laffoley D, Baxter JM, editors. Explaining ocean warming: Causes, scale, effects and consequences . Gland, Switzerland: IUCN; 2016;213-237. Bui S, Dalvin S, Dempster T, Skulstad OF, Edvardsen RB, Wargelius A, Oppedal F. Susceptibility, behaviour, and retention of the parasitic salmon louse (Lepeophtheirus salmonis) differ with Atlantic salmon population origin. Journal of Fish Diseases. 2018;41(3):431–42. Burke M, Nikolic D, Fabry P, Rishi H, Telfer T, Rey Planellas S. Precision Farming in Aquaculture: Non-invasive monitoring of Atlantic salmon ( Salmo salar ) behaviour in response to environmental conditions in commercial sea cages for health and welfare assessment. Frontiers in Robotics and AI. 2025:1574161. Carr JA. Stress, Neuropeptides, and Feeding Behavior: A Comparative Perspective1. Integrative and Comparative Biology. 2002;42(3):582–90. Cerqueira M, Millot S, Felix A, Silva T, Oliveira GA, Oliveira CCV, Oliveira S, Rey S, MacKenzie S, Oliveira R. Cognitive appraisal in fish: stressor predictability modulates the physiological and neurobehavioural stress response in sea bass. Proceedings of the Royal Society B: Biological Sciences. 2020;287(1923):20192922. Clark A, Nowak BF. Field investigations of amoebic gill disease in Atlantic salmon, Salmo salar L., in Tasmania. Journal of Fish Diseases. 1999;22:433-443. Conte FS. Stress and the welfare of cultured fish. Applied Animal Behaviour Science. 2004;86(3):205–23. Cornes RC, Tinker J, Hermanson L, Oltmanns M, Hunter WR, Lloyd-Hartley H, Kent EC, Rabe B, Renshaw R. Climate change impacts on temperature around the UK and Ireland. MCCIP Science Review. 2023;1-18. Davidson WS, Birt TP, Green JM. A review of genetic variation in Atlantic salmon, Salmo salar L., and its importance for stock identification, enhancement programmes and aquaculture. Journal of Fish Biology. 1989;34(4):547–60. Deck CA, Honeycutt JL, Cheung E, Reynolds HM, Borski RJ. Assessing the Functional Role of Leptin in Energy Homeostasis and the Stress Response in Vertebrates. Frontiers in Endocrinology. 2017; 8. Ellis T, Berrill I, Lines J, Turnbull JF, Knowles TG. Mortality and fish welfare. Fish Physiol Biochem. 2012;38(1):189–99. Folkedal O, Stien LH, Torgersen T, Oppedal F, Olsen RE, Fosseidengen JE, Braithwaite V, Kristiansen, TS. Food anticipatory behaviour as an indicator of stress response and recovery in Atlantic salmon post-smolt after exposure to acute temperature fluctuation. Physiology & Behavior. 2012;105(2):350–6. Foyle KL, Hess S, Powell MD, Herbert NA. What Is Gill Health and What Is Its Role in Marine Finfish Aquaculture in the Face of a Changing Climate? Frontiers in Marine Science. 2020; 7(400):1-16. Herrero A, Thompson KD, Ashby A, Rodger HD, Dagleish MP. Complex Gill Disease: an Emerging Syndrome in Farmed Atlantic Salmon ( Salmo salar L.). Journal of Comparative Pathology. 2018;163:23–8. Herrero A, Rodger H, Hayward AD, Cousens C, Bron JE, Dagleish MP, Thompson K. Prospective Longitudinal Study of Putative Agents Involved in Complex Gill Disorder in Atlantic salmon (Salmo salar). Pathogens. 2022;11(8):878. Hoare DJ, Couzin ID, Godin JGJ, Krause J. Context-dependent group size choice in fish. Animal Behaviour. 2004;67(1):155–64. Huntingford, F., Coyle, S., Hunter, W., 2012. Avoiding Predators, in: Aquaculture and Behavior. John Wiley & Sons, Ltd, pp. 220–247. https://doi.org/10.1002/9781444354614.ch8 Hvas M, Kolarevic J, Noble C, Oppedal F, Stien LH. Fasting and its implications for fish welfare in Atlantic salmon aquaculture. Reviews in Aquaculture. 2024;16(3):1308-1332. Jaervi T, Uglem I. Predator training improves the anti-predator behaviour of hatchery reared Atlantic salmon(Salmo salar) smolt. Nordic journal of freshwater research Drottningholm. 1993;(68):63–71. Juell JE. The behaviour of Atlantic salmon in relation to efficient cage-rearing. Rev Fish Biol Fisheries. 1995;5(3):320–35. Kennerley A, Wood LE, Luisetti T, Ferrini S, Lorenzoni I. Economic impacts of jellyfish blooms on coastal recreation in a UK coastal town and potential management options. Ocean & Coastal Management. 2022;227:106284. Kent M, Whyte J, LaTrace C. Gill lesions and mortality in seawater pen-reared Atlantic salmon Salmo salar associated with a dense bloom of Skeletonema costatum and Thalassiosira species. Dis Aquat Org. 1995;22:77–81. Kleinhappel TK, Pike TW, Burman OHP. Stress-induced changes in group behaviour. Sci Rep. 2019;9(1):17200. Klinger DH, Levin SA, Watson JR. The growth of finfish in global open-ocean aquaculture under climate change. Proc R Soc B. 2017;284(1864):20170834. Kolsäter L. Feed management and reduction of aquaculture wastes. Water Science and Technology. 1995;31(10):213–8. Kragesteen T, Simonsen K, Visser A, Andersen K. Estimation of external infection pressure and salmon-louse population growth rate in Faroese salmon farms. Aquacult Environ Interact. 2021;13:21–32. Krause J, Godin JGJ. Shoal Choice in the Banded Killifish (Fundulus diaphanus, Teleostei, Cyprinodontidae): Effects of Predation Risk, Fish Size, Species Composition and Size of Shoals. Ethology. 1994;98(2):128–36. Krause J, Hensor EMA, Ruxton GD. Fish as prey. In: Hart PJB, Reynolds JD, editors. Handbook of Fish Biology and Fisheries . Hoboken, NJ: Wiley-Blackwell; 2002. Król E, Noguera P, Shaw S, Costelloe E, Gajardo K, Valdenegro V, Bickerdike R, Douglas A, Martin SAM. Integration of Transcriptome, Gross Morphology and Histopathology in the Gill of Sea Farmed Atlantic Salmon (Salmo salar): Lessons From Multi-Site Sampling. Front Genet 2020;11. Lai F, Royan MR, Gomes AS, Espe M, Aksnes A, Norberg B, Gelebart V, Rønnestad I. The stress response in Atlantic salmon (Salmo salar L.): identification and functional characterization of the corticotropin-releasing factor (crf) paralogs. General and Comparative Endocrinology. 2021;313:113894. Li D, Wang G, Du L, Zheng Y, Wang Z. Recent advances in intelligent recognition methods for fish stress behavior. Aquacultural Engineering. 2022;96:102222. Licandro P, Conway DVP, Daly Yahia MN, Fernandez de Puelles ML, Gasparini S, Hecq JH, Tranter P, Kirby RR. A blooming jellyfish in the northeast Atlantic and Mediterranean. Biology Letters. 2010;6(5):688–91. Liu J, Bienvenido F, Yang X, Zhao Z, Feng S, Zhou C. Nonintrusive and automatic quantitative analysis methods for fish behaviour in aquaculture. Aquaculture Research. 2022;53(8):2985–3000. Madaro A, Olsen RE, Kristiansen TS, Ebbesson LOE, Nilsen TO, Flik G, Gorissen M. Stress in Atlantic salmon: response to unpredictable chronic stress. Journal of Experimental Biology. 2015;218(16):2538–50. Magurran AE, Pitcher TJ, Dodd JM. Provenance, shoal size and the sociobiology of predator-evasion behaviour in minnow shoals. Proceedings of the Royal Society of London Series B Biological Sciences. 1997;229(1257):439–65. Mandal A, Ghosh AR. Role of artificial intelligence (AI) in fish growth and health status monitoring: a review on sustainable aquaculture. Aquaculture International. 2023;32:2791-2820. Marine Scotland, The Scottish Government. Amoebic gill disease. Topic Sheet No. 96 V1 . 2012. Available from: http://www.scotland.gov.uk/Resource/0039/00393037.pdf Martins CIM, Galhardo L, Noble C, Damsgård B, Spedicato MT, Zupa W, Beauchaud M, Kulczykowska E, Massabuau J-C, Carter T, Rey Planellas S, Kristiansen T. Behavioural indicators of welfare in farmed fish. Fish Physiology and Biochemistry. 2012;38(1):17–41.1. Mellor D, Beausoleil N. Extending the ‘Five Domains’ model for animal welfare assessment to incorporate positive welfare states. Animal welfare. 2015;24:241–253. https://doi.org/10.7120/09627286.24.3.241 Mikheev VN. Combined effects of predators and parasites on shoaling behavior of fishes. Journal of Ichthyology. 2009; 49(11):1032-1041. Miller N, Gerlai R. From schooling to shoaling: Patterns of collective motion in Zebrafish (Danio rerio). PLOS ONE. 2012;7(11):e48865. Mitchell SO, Baxter EJ, Holland C, Rodger HD. Development of a novel histopathological gill scoring protocol for assessment of gill health during a longitudinal study in marine-farmed Atlantic salmon (Salmo salar). Aquacult Int. 2012;20(5):813–25. Morgan MJ, Godin JGJ. Antipredator Benefits of Schooling Behaviour in a Cyprinodontid Fish, the Banded Killifish (Fundulus diaphanus). Zeitschrift für Tierpsychologie. 1985;70(3):236–46. Morgenroth D, Kvaestad B, Økland F, Finstad B, Olsen RE, Svendsen E, Rosten C, Axelsson M, Bloecher N, Føre M, Gräns A. Under the sea: How can we use heart rate and accelerometers to remotely assess fish welfare in salmon aquaculture? Aquaculture. 2024;579:740144. Mustafa A, MacWilliams C, Fernandez N, Matchett K, Conboy GA, Burka JF. Effects of sea lice (Lepeophtheirus salmonis Kröyer, 1837) infestation on macrophage functions in Atlantic salmon (Salmo salar L.). Fish & Shellfish Immunology. 2000;10(1):47–59. Noble C, Gismervik K, Iversen MH, Kolarevic J, Nilsson J, Stien LH, Turnbull JF. Welfare indicators for farmed Atlantic salmon: tools for assessing fish welfare. 2018. 351 p. O’Donncha F, Stockwell CL, Planellas SR, Micallef G, Palmes P, Webb C, Filgueira R, Grant J. Data Driven Insight Into Fish Behaviour and Their Use for Precision Aquaculture. Frontiers in Animal Science. 2021;2. Oppedal F, Dempster T, Stien LH. Environmental drivers of Atlantic salmon behaviour in sea-cages: A review. Aquaculture. 2011;311(1):1–18. Orr JA, Vinebrooke RD, Jackson MC, Kroeker KJ, Kordas RL, Mantyka-Pringle C, Van den Brink, De Laender F, Stoks R, Holmstrup M, Matthaei CD, Monk WA, Penk MR, Leuzinger S, Schäfer RB, Piggott JJ. Towards a unified study of multiple stressors: divisions and common goals across research disciplines. Proceedings of the Royal Society B: Biological Sciences. 2020;287(1926):20200421. Øverli Ø, Nordgreen J, Mejdell CM, Janczak AM, Kittilsen S, Johansen IB, Horsberg TE. Ectoparasitic sea lice (Lepeophtheirus salmonis) affect behavior and brain serotonergic activity in Atlantic salmon (Salmo salar L.): Perspectives on animal welfare. Physiology & Behavior. 2014;132:44–50. Pitcher TJ. Heuristic definitions of fish shoaling behaviour. Animal Behaviour. 1983;31(2):611–3. Rey S, Huntingford FA, Boltaña S, Vargas R, Knowles TG, Mackenzie S. Fish can show emotional fever: stress-induced hyperthermia in zebrafish. Proceedings of the Royal Society B. 2015;282(1819):20152266. Rodger HD. Gill disorders: an emerging problem for farmed Atlantic salmon ( Salmo salar ) in the marine environment? Fish Veterinary Journal. 2007;9:38–48. Rozas-Serri M. Gill diseases in marine salmon aquaculture with an emphasis on amoebic gill disease. CAB Reviews Perspectives in Agriculture Veterinary Science Nutrition and Natural Resources. 2019;14:1–15. Ruehle B, Poulin R. Hunger games: foraging behaviour and shelter use in fish under the context-dependent influence of parasitism. Parasitology Research. 2021;120(11):3681–92. Samaras A, Dimitroglou A, Kollias S, Skouradakis G, Papadakis IE, Pavlidis M. Cortisol concentration in scales is a valid indicator for the assessment of chronic stress in European sea bass, Dicentrarchus labrax L. Aquaculture. 2021;545:737257. Santos EGN, Cunha RA, Santos CP. Behavioral responses of Poecilia vivipara (Osteichthyies: Cyprinodontiformes) to experimental infections of Acanthocollaritrema umbilicatum (Digenea: Cryptogonimidae). Experimental Parasitology. 2011;127(2):522–6. Shinn AP, Pratoomyot J, Bron JE, Paladini G, Brooker EE, Brooker AJ. Economic costs of protistan and metazoan parasites to global mariculture. Parasitology. 2015;142(1):196–270. Susdorf R, Salama NKG, Todd CD, Hillman RJ, Elsmere P, Lusseau D. Context-dependent reduction in somatic condition of wild Atlantic salmon infested with sea lice. Marine Ecology Progress Series. 2018;606:91–104. Taylor RS, Muller WJ, Cook MT, Kube PD, Elliott NG. Gill observations in Atlantic salmon ( Salmo salar, L.) during repeated amoebic gill disease (AGD) field exposure and survival challenge. Aquaculture. 2009;290(1):1–8. Villegas-Ch W, García-Ortiz J, Sánchez-Viteri S. Toward Intelligent Monitoring in IoT: AI Applications for Real-Time Analysis and Prediction. IEEE Access. 2024;12:40368–86. Wendelaar Bonga SE. The stress response in fish. Physiological Reviews. 1997;77(3):591–625. Yang L, Liu Y, Yu H, Fang X, Song L, Li D, Chen Y. Computer Vision Models in Intelligent Aquaculture with Emphasis on Fish Detection and Behavior Analysis: A Review. Arch Computat Methods Eng. 2021;28(4):2785–816. Zhang D, Sogn-Grundvåg G, Tveterås R. The impact of parasitic sea lice on harvest quantities and sizes of farmed salmon. Aquaculture. 2023;576:739884. Zhang L, Li B, Sun X, Hong Q, Duan Q. Intelligent fish feeding based on machine vision: A review. Biosystems Engineering. 2023;231:133–64. Zhou J, Seo JH, Mittal R. Effect of schooling on flow generated sounds from carangiform swimmers. Bioinspired Biomimetic. 2024;19(3):036015. Additional Declarations No competing interests reported. Supplementary Files ASMAppendix.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 10 Jun, 2025 Reviews received at journal 03 Jun, 2025 Reviews received at journal 26 May, 2025 Reviewers agreed at journal 23 May, 2025 Reviewers agreed at journal 22 May, 2025 Reviewers agreed at journal 20 May, 2025 Reviewers agreed at journal 16 May, 2025 Reviewers invited by journal 16 May, 2025 Editor assigned by journal 14 May, 2025 Submission checks completed at journal 14 May, 2025 First submitted to journal 13 May, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6653910","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":457549475,"identity":"f762e61e-7425-417b-abaf-bf271074119a","order_by":0,"name":"Meredith Burke","email":"","orcid":"","institution":"University of Stirling","correspondingAuthor":false,"prefix":"","firstName":"Meredith","middleName":"","lastName":"Burke","suffix":""},{"id":457549476,"identity":"ff86b255-4122-45d9-b186-c91d27074ab0","order_by":1,"name":"Dragana Nikolic","email":"","orcid":"","institution":"Observe Technologies","correspondingAuthor":false,"prefix":"","firstName":"Dragana","middleName":"","lastName":"Nikolic","suffix":""},{"id":457549477,"identity":"0ef0b098-ef69-4211-a8bf-acd39e50f8b5","order_by":2,"name":"Pieter Fabry","email":"","orcid":"","institution":"Observe Technologies","correspondingAuthor":false,"prefix":"","firstName":"Pieter","middleName":"","lastName":"Fabry","suffix":""},{"id":457549478,"identity":"c4fe9c90-4268-47a3-9f9a-c371f8f319a7","order_by":3,"name":"Hemang Rishi","email":"","orcid":"","institution":"Observe Technologies","correspondingAuthor":false,"prefix":"","firstName":"Hemang","middleName":"","lastName":"Rishi","suffix":""},{"id":457549479,"identity":"2577fa2a-e82f-4339-9459-41a61edec0f0","order_by":4,"name":"Trevor Telfer","email":"","orcid":"","institution":"University of Stirling","correspondingAuthor":false,"prefix":"","firstName":"Trevor","middleName":"","lastName":"Telfer","suffix":""},{"id":457549480,"identity":"612ade5a-c09b-4232-8d5b-02e42f838f11","order_by":5,"name":"Sonia Rey Planellas","email":"data:image/png;base64,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","orcid":"","institution":"University of Stirling","correspondingAuthor":true,"prefix":"","firstName":"Sonia","middleName":"Rey","lastName":"Planellas","suffix":""}],"badges":[],"createdAt":"2025-05-13 09:38:03","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6653910/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6653910/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":83152043,"identity":"51d4d4ee-914f-4169-bf75-03462a09f0f1","added_by":"auto","created_at":"2025-05-20 14:01:10","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":440405,"visible":true,"origin":"","legend":"\u003cp\u003eOverview of the study site. (A) Map of Scotland showing the location of the of the Atlantic salmon aquaculture farms on the Western Isles. Satellite image providing an aerial view of the Farm A (B) and Farm B (D; from Google Earth) with study cages outlined in red. (C) Layout illustrating the placement of cameras within the two study cages.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6653910/v1/9e81f6f479c63be505fc6b18.png"},{"id":83152044,"identity":"07dcf722-a534-49b9-a53a-0fac394102f6","added_by":"auto","created_at":"2025-05-20 14:01:10","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":405534,"visible":true,"origin":"","legend":"\u003cp\u003eComparative figure for Farm A (left) an Farm B (Right) with: depth of maximum activity where colours indicate the hourly-averaged amount of activity present at that depth (A,E), weekly gill health scores (0 = healthy gills, 5 = necrotic tissue), where red indicates Proliferative Gill Disease (PGD) and blue indicates Amoebic Gill Disease (AGD; B,F), daily mortality as a percent loss within the cage (C,G), and hourly temperature at 8 m depth (D,H).\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6653910/v1/0cf6a8d758f04c73ba669298.png"},{"id":83152047,"identity":"ab235141-386d-4f17-af35-4bac47a1cc9b","added_by":"auto","created_at":"2025-05-20 14:01:10","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":176588,"visible":true,"origin":"","legend":"\u003cp\u003eThe model II linear regression for weekly activity and PGD scores for Farms A (orange) and B (blue) with confidence intervals (shaded region).\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6653910/v1/ca9cdd9bc91e8c93fc256b3d.png"},{"id":83153349,"identity":"2e9ba461-0cbe-448e-b0c2-4cb0c760324a","added_by":"auto","created_at":"2025-05-20 14:09:10","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":213625,"visible":true,"origin":"","legend":"\u003cp\u003eThe health data from each cage in Farm A (left) and Farm B (right) including: The weekly PGD scores (A. E), the daily activity of the fish (B, F), the daily specific feeding rate (SFR; C, G), and the daily mortality (D, H). The colours indicate average 2 months pre- (white) and post- (grey) increase in activity (July 7 [Farm A]; August 18 [Farm B]). Different letters denote significant differences within and among cages per farm.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6653910/v1/28a0c3b9f968d75eb54d717e.png"},{"id":83153830,"identity":"32946913-f125-4231-8ab1-d8d34d8cdf1a","added_by":"auto","created_at":"2025-05-20 14:17:12","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1970260,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6653910/v1/5459e1d5-014e-408c-8f20-2810aa6bca63.pdf"},{"id":83152050,"identity":"392041d2-fad1-4f75-9213-ba58cd0f8264","added_by":"auto","created_at":"2025-05-20 14:01:10","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":2070637,"visible":true,"origin":"","legend":"","description":"","filename":"ASMAppendix.docx","url":"https://assets-eu.researchsquare.com/files/rs-6653910/v1/e4eb41c4f85b0025f29bd841.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003ePrecision Farming in Aquaculture: Assessing gill health in Atlantic salmon (\u003cem\u003eSalmo salar\u003c/em\u003e) using a non-invasive, AI-driven behavioural monitoring approach in commercial farms\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eMeasuring the behaviour of cultured fish is challenging but critical for aquaculture, as it provides valuable insights into fish health and welfare being one of the 5 domains used to understand the welfare status of the animals (Mellor and Beausoleil, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). However, observing behaviour is inherently difficult because fish are underwater, often with poor visibility, and it can be unclear which behaviours are most useful to monitor. Previous studies on the behaviour of farmed Atlantic salmon smolts (\u003cem\u003eSalmo salar\u003c/em\u003e) in sea cages have focused on the environmental drivers of their swimming depth and speed (Oppedal et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2011\u003c/span\u003e, Burke et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Salmon appear to distribute themselves heterogeneously throughout the cage, largely basing their preferred depth on environmental gradients, such as temperature and photoperiod (Oppedal et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; O\u0026rsquo;Donncha et al., \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Fish shoaling, when multiple individuals group together socially, is a common anti-predator stress defence mechanism, however fish may shoal as a response to other stressors like changes in environmental or health status (Pitcher \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e1983\u003c/span\u003e; Jaervi and Uglem \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e1993\u003c/span\u003e; Hoare et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2004\u003c/span\u003e). A previous study found that an unpredictable stressor (confinement) resulted in a higher shoal cohesion response in sea bass (Cerqueira et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) and a temporal preference for higher temperatures in zebrafish as a stress coping mechanism (Rey et al. \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Other studies found higher shoaling in response to predator stimuli in banded killifish, minnows, and zebrafish, (Magurran and Pitcher, 1987; Krause and Godin, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e1994\u003c/span\u003e; Hoare et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Miller and Gerlai, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). However, there are few studies that explore the role that parasites and disease may play on the behaviour of fish in wild or sea cage systems, with most studies considering behaviours (e.g. locomotion, foraging, shelter use) in laboratory settings (e.g. Santos et al., \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; \u0026Oslash;verli et al., \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Ruehle and Poulin \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eStudies investigating fish response to stressors have traditionally employed invasive methods such as using plasma cortisol analysis (e.g. Madaro et al., \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2015\u003c/span\u003e, Lai et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2021\u003c/span\u003e, Samaras et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) requiring crowding, handling, anaesthetising, and blood sampling which are inherently stressful for the fish. However, using behaviour as a non-invasive metric for detecting poor fish health holds promise for enhancing fish welfare (Li et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Behavioural changes often precede outward signs of illness, providing farmers with an early warning signal to prevent disease rather than merely treating the symptoms (Juell \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e1995\u003c/span\u003e). Additionally, changes in feeding behaviour serve as a critical indicator of potential disease or stress in fish. Research has demonstrated a decrease in feeding motivation in response to acute stressors across vertebrates (Folkedal et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). Studies have shown that all vertebrates exhibit stress-related suppression of appetite and food intake, as well as enhanced energy mobilisation (Carr \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2002\u003c/span\u003e; Hvas et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This phenomenon is evolutionarily advantageous, as appetite suppression reduces the risk of predation, given that foraging in the presence of predators heightens mortality rates (Carr \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2002\u003c/span\u003e). Moreover, metabolic energy is directed toward vital activities such as respiration, locomotion, and tissue repair, all of which contribute to individual survival, prioritising these functions overgrowth and reproduction (Wendelaar Bonga, \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e1997\u003c/span\u003e; Deck et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Specific feeding rate, a composite measure incorporating specific growth rate and feed conversion rate, is widely utilised by the aquaculture industry to gauge feeding efficacy. Higher specific feeding rates signify the implementation of efficient and effective feeding regimes (Kols\u0026auml;ter, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e1995\u003c/span\u003e). These non-invasive tools represent examples of Operational Welfare Indicators (OWI), measures used to assess the overall welfare status of cultured fish and are pivotal for assessing fish welfare on-site within production systems and husbandry practices (Noble et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). OWIs encompass both indirect environment-based and direct animal-based measures, which can be further categorised at a group or individual level. Implementation of OWIs is indispensable for ensuring that the welfare needs of fish are adequately addressed. Standardised tools have been developed to monitor fish welfare through direct observations, focusing on a subset of individual animals within cages for external signs of injury or illness, as well as through assessments of group-level behaviours such as feeding patterns.\u003c/p\u003e \u003cp\u003eOver the last decade, the increase in marine fish gill diseases is recognised as a major health and welfare problem worldwide (Boerlage et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Compromised gill function disrupts fish respiration and osmoregulation, causing breathing difficulties, and loss of salt and blood gas control. There are many causes of gill disease including parasitic gill disease, viral/bacterial gill disease, plankton/algae associated gill disease, or chemical-related gill disease (Rodger \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Boerlage et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). The parasite \u003cem\u003eNeoparamoeba perurans\u003c/em\u003e, through the development of Amoebic Gill Disease (AGD) is often the most reported due to the distinctive pathology in salmon and is recognisable on-site (Boerlage et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). AGD can cause typical losses to the Scottish salmon industry of 10\u0026ndash;20%, with some sites as high as 70% (Marine Scotland, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Shinn et al., \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). It is also the cause of significant mortalities in commercial salmon fisheries in Australia, USA, Canada, France and Ireland. However, gill disease is often multifactorial, due to a combination of one or more of the above causes; this is called complex gill disease (CGD; Herrero et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). CGD encompasses Proliferative Gill Disease (PGD), which is a gill condition described in Scotland and Ireland as a generic term for the proliferative changes in the gill epithelium while causing gross lesions easily identified in the field (Boerlage et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). In Norway, a similar pathology-based diagnosis termed Proliferative Gill Inflammation (PGI) is distinguished by specific histopathological changes including lamellar vascular changes, inflammation, necrosis and epithelial cell hyperplasia (Herrero et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Boerlage et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Notably, PGD differs from PGI in that it exhibits proliferative histological features but with unknown aetiology (Herrero et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). It can result from the interaction of environmental and husbandry conditions as well as infection by pathogens and parasites (such as \u003cem\u003eN. perurans\u003c/em\u003e), particularly during the summer and fall months. While AGD can be diagnosed through gill sampling and histopathology to confirm the presence of the parasite, commercial producers typically assess gill health (both AGD and PGD) on-site using a categorical scoring system (Taylor et al., \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2009\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe progression of climate change and increased sea temperatures is directly linked to an increase in the organisms that cause damage to the gills and water quality changes, algal blooms, handling, and temperature changes can all have a cumulative stress effect increasing susceptibility to gill disease (Klinger et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Foyle et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Additionally, cumulative stressors can result in susceptibility to other diseases such as infection by copepod parasites, or sea lice (\u003cem\u003eCaligus elongatus\u003c/em\u003e and \u003cem\u003eLepeophtheirus salmonis\u003c/em\u003e; Orr et al., \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Herrero et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Sea lice have both lethal and non-lethal impacts on salmon, primarily resulting in chronic stress and reduced growth rate, with \u003cem\u003eL. salmonis\u003c/em\u003e reported as the largest parasitic threat to the salmon aquaculture industry (Mustafa et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2000\u003c/span\u003e; Susdorf et al., 2017; Zhang et al., \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAdvancements in artificial intelligence (AI) has resulted in improved monitoring techniques especially in the ocean sector where usual methods are not feasible. AI can enhance real-time or near-real time monitoring by automating the collection and analysis of data (Villegas-ch et al., \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). AI-based algorithms have the capability to process and analyse large datasets, image or video data through machine vision techniques. AI can detect patterns and thereby any factors contributing to anomalies to those patterns. Thus, it can contribute to the development of early-warning systems, such as the detection of environmental or health-related parameters that may cause poor welfare and mortality (Mandal and Ghosh \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Moreover, they can collect high-quality data in a non-invasive manner, avoiding damage to the fish that often occurs with traditional manual methods (Liu et al., 2023). Machine vision techniques have been used in aquaculture in the past, primarily for biomass estimation, body length measurements, fish counting and feeding behaviours (Yang et al., \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Liu et al., 2023; Zhang et al., \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Fish behaviour in relation to stress has also been examined, however the focus is often on individual fish swimming speed, acceleration and turning angle (Li et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). While this type of analysis is relatively straightforward in tanks or recirculating aquaculture systems, it can be more difficult to assess in large commercial sea cages with low light, high noise, fish deformation, fish overlap and occlusion, and low field of view in larger cages (Yang et al., \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The present study utilised cameras deployed within cages in two commercial aquaculture farms in Scotland, focusing on group behaviour instead. This was quantified using an AI-based algorithm developed by Observe Technologies (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://observe.tech/\u003c/span\u003e\u003cspan address=\"https://observe.tech/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). The aim was to investigate fish group behaviours such as cohesive grouping or shoaling, feeding, and fish distribution in response to a common health-related stressor. This study hypothesised that changes in gill health would elicit observable group-level behavioural changes in Atlantic salmon. To test this, long-term behavioural data were collected and analysed a posteriori. The findings support the potential of precision fish farming (PFF), particularly when combined with system modelling and machine learning, as a significant advancement for aquaculture systems and a promising approach to addressing welfare-related stressors.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\"\u003e\n \u003ch2\u003e2.1 Study Sites\u003c/h2\u003e\n \u003cp\u003eThe two commercial Atlantic salmon farms used in this study are located on the Isle of Harris and Lewis, in the northwest of Scotland (Fig.\u0026nbsp;1A,B,D).\u003c/p\u003e\n \u003cp\u003eFarm A consists of 16 circular cages (13 stocked during the study) in two parallel rows, each 90 m in circumference and 15 m in depth with cylindrical nets (mesh size:18 mm) and Seal Pro anti-predator nets to a depth of 10 m (Table 1). Farm B contains 16 circular cages (10 stocked during the study) in two parallel rows, each 120 m in circumference and 10 m in depth (mesh size 18 mm) with Seal Pro anti-predator nets. Additionally, the fish were initially fed \u003cem\u003ead libitum\u003c/em\u003e and later transitioned to a twice-daily feeding schedule, which was adjusted based on their age class.\u003c/p\u003e\n \u003cdiv\u003e\u0026nbsp;\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 1\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eDetails of the two fish farms used in this study.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFarm\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eCages stocked during study\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eCage\u003c/p\u003e\n \u003cp\u003evolume (m\u003csup\u003e3\u003c/sup\u003e)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eInitial stocking week\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMean smolt weight at stocking (g)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMean [range] current speed (m s\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e) *\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9 669\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2023-02-13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e109\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.06 [0-0.16]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11 459\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2023-05-08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.09 [0-0.38]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"6\"\u003e\n \u003cp\u003e* The current speed was measured over 90 days prior to stocking the sites\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\"\u003e\n \u003ch2\u003e2.2 Data Acquisition\u003c/h2\u003e\n \u003cdiv id=\"Sec5\"\u003e\n \u003ch2\u003e2.2.1 Fish Behaviour\u003c/h2\u003e\n \u003cp\u003eThere were 4\u0026ndash;5 video camera modules (Sony model: IPCM-3516DS385-D29-AZ3611) installed by AKVA Group (www.akvagroup.com) within one, randomly allocated cage per site pointing towards the surface. These were deployed to obtain higher spatial resolution of fish behaviour and to identify optimal camera placement within the cage for capturing behavioural data from a high proportion of the population. In Farm A, they were positioned as 3 down the centre (approximately 4 m, 9 m, 14 m), and 2 at 9 m on the inner and outer areas of the cage, respectively (Fig.\u0026nbsp;1C left). As Farm B is a shallower site, only 2 cameras were placed down the centre (4 m and 9 m), with 2 at 5 m on the inner and outer areas of the cage, respectively (Fig.\u0026nbsp;1C right). The remaining cages were equipped with a single camera pointing upwards, positioned centrally at mid-depth (~\u0026thinsp;9 m in Farm A and ~\u0026thinsp;7 m in Farm B), to allow comparison with the higher resolution study cage. The cameras recorded at 25 frames per second, at a resolution of 1920 x 1080 pixels, and a video bit rate of 16 Mb per second. Videos were recorded continuously during daylight hours between May (June) and September (October) for Farm A (B). Daylight hours shift as summer progresses, with lowest duration in October at 12 hours and peaking in June at 18 hours.\u003c/p\u003e\n \u003cp\u003eEach camera contained a pressure (TE connectivity US300; accuracy of \u0026plusmn;\u0026thinsp;0.15% full scale output) and temperature sensor (RDO PRO-X; accuracy of \u0026plusmn;\u0026thinsp;0.1\u0026ordm;C and resolution of 0.01 \u0026ordm;C), which recorded continuously for the duration of the study. Additionally, there were single cameras deployed in the centre of the remaining cages in each farm, allowing for activity data from these cameras to be compared to the study cage to investigate whether similar trends were observed using a single-camera system.\u003c/p\u003e\n \u003cp\u003eVideo recordings were analysed using an algorithm developed by by Observe Technologies (www.observe.tech), designed to extract behavioural data relevant to aquaculture operations. This algorithm employs a patented (EP3644717) convolutional neural network to heuristically estimate fish activity by identifying key behavioural features. These features include, but are not limited to, schooling cohesion, average distance from the camera, the number of fish detected per frame, and swimming speed. Rather than tracking individual fish, the model is trained globally, recognising behavioural patterns from annotated datasets. Since all features are processed within a shared network architecture until the final layer, their individual contributions to the overall activity score remain undefined. The model continuously refines its parameters through standard deep-learning optimisation techniques, generating a normalised activity score ranging from 0 to 100% (example in Fig. A.1).\u003c/p\u003e\n \u003cp\u003eTo ensure reliability, the model undergoes comprehensive validation using separate training, validation, and test datasets, following best practices in deep learning. Further, aquaculture specialists routinely assess the model\u0026rsquo;s outputs to verify their practical relevance, and for this study, they conducted an additional review to ensure accuracy before the data was provided to university researchers. Independent validation of the algorithm was also carried out by two observers from the University of Stirling, using video footage captured by the same AKVA feed cameras. The algorithm\u0026rsquo;s \u0026apos;activity\u0026apos; output was correlated to fish abundance counts from 55 randomly chosen video stills using Pearson correlation (\u0026alpha;\u0026thinsp;=\u0026thinsp;0.05). Additionally, the Intraclass Correlation Coefficient (ICC) was calculated to assess interobserver reliability.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec6\"\u003e\n \u003ch2\u003e2.2.2 Fish Health and Welfare\u003c/h2\u003e\n \u003cp\u003eEach farm provided health and welfare-related parameters used in this study. Daily Specific Feeding Rate (SFR) was recorded for each cage as the % body weight fed:\u003c/p\u003e\n \u003cdiv id=\"Equa\"\u003e\n \u003cdiv id=\"FileID_Equa\" name=\"EquationSource\"\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003eWhere biomass was determined through sample and mortality weights and number of fish in each cage.\u003c/p\u003e\n \u003cp\u003eAt commercial aquaculture sites, farmers visually assess fish for OWIs, sea lice, and gill health. Given the demanding nature of farm operations and limited staffing, manual assessments remain a standard and practical protocol. On a weekly or bi-weekly basis (depending on water temperature), 10 fish per cage were examined for various Operational Welfare Indicators including fin and eye damage, skin lesions, scale loss and visual K-factor with severity scored from 0 (healthy) \u0026ndash; 2 (severe; Table\u0026nbsp;2).\u003c/p\u003e\n \u003cdiv\u003e\u0026nbsp;\u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 2\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eOperational Welfare Indicators (OWI) scoring used in this study.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eEye damage\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFin damage\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSkin lesions\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eScale Loss\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eK-Factor\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBoth eyes perfect\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDorsal, pectoral and caudal fins intact\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSkin intact\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAll scales are intact, no patches of lost or damaged scales; Fish is \u0026lsquo;smooth\u0026rsquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFish are well shaped and firm to touch; Has a \u0026lsquo;belly\u0026rsquo;.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDamage to one eye (unilateral)\u003c/p\u003e\n \u003cp\u003eSuch as:\u003c/p\u003e\n \u003cp\u003e- Rupture\u003c/p\u003e\n \u003cp\u003e- Cloudiness\u003c/p\u003e\n \u003cp\u003e- Bleeding\u003c/p\u003e\n \u003cp\u003e- Cataract\u003c/p\u003e\n \u003cp\u003e- Swelling\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026lt; half of fin remaining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u0026thinsp;+\u0026thinsp;small wounds (\u0026lt;\u0026thinsp;10p coin); subcutaneous tissue intact\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSmall areas of scale damage or loss present, but skin is intact (i.e. no bleeding or lesions); Affected area is no bigger than 10p coin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFish long, but not have as much depth as expected; Slightly soft to touch (esp. around the belly area)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDamage to both eyes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLittle of fin remaining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLarge (\u0026gt;\u0026thinsp;10p coin), severe wounds; muscle exposed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMultiple areas of scale loss or damage; Skin may look dimpled due to water ingress\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFish long and thin; head significantly oversized compared to body.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"6\"\u003e\n \u003cp\u003eAdapted from The Scottish Salmon Company (Bakkafrost Scotland) and FISHWELL handbook (Noble et al., 2018)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eThese 10 sampled fish are then examined for sea lice with counts per fish for gravid females (females with egg sacs attached). Gravid females were chosen as they serve as a key indicator for assessors when determining whether treatment is necessary. Lastly, fish were assessed weekly for gill health using a standardised scoring protocol. Gill health was assessed by farmers using a standardised scorecard (Fig. A.2), evaluating lesions and lamellar thickening to assign AGD and PGD scores ranging from 0 (healthy) to 5 (necrotic; Table\u0026nbsp;3). The distinct visual differences between AGD and PGD allow for rapid on-farm evaluation and treatment decisions, as laboratory diagnostics can take time to yield results. While somewhat subjective, this method is supported by farmer training and extensive experience.\u003c/p\u003e\n \u003cdiv\u003e\u0026nbsp;\u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 3\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eGill health (AGD and PGD) scoring system used to estimate severity on-site.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eScore\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAGD\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePGD\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHealthy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo sign of infection\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHealthy, no thickening\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVery light\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1 white spot/light scarring\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVery slight thickening/few lamellae affected\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u0026ndash;3 spots/small mucus patch\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFrequent thickening but tips only\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eModerate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEstablished thickened mucus patch/spot groupings up to 20% of gill\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAlmost all lamellae have thickened tips/some have thickenings progressing to ~\u0026thinsp;50% of the length of lamellae\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAdvanced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEstablished lesions covering up to 50% of gill\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMost lamellae have thickening progressing to \u0026gt;\u0026thinsp;50% of the length of lamellae\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNecrotic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eExtensive lesions covering most of gill surface (\u0026gt;\u0026thinsp;50%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAlmost all lamellae are thickened along entire length\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\n \u003cp\u003eAGD details adapted from Taylor et al., 2009; PGD details adapted from The Scottish Salmon Company (Bakkafrost Scotland)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec7\"\u003e\n \u003ch2\u003e2.3 Data collection\u003c/h2\u003e\n \u003cp\u003eThis study was conducted for 4 months for each site, from May 10 to September 10, 2023 for Farm A and June 22 to October 13, 2023 for Farm B, these differences in time frames were related to the later stocking of Farm B while maximising data collection. At the beginning of July for Farm A and in mid-August for Farm B, there was an increase in the activity of the fish, prompting a divide in the dataset for 2 months pre/post rise in activity. Activity data was preprocessed through a rolling average of 15 minutes to identify trends and remove excess noise. Additionally, activity data collected during feeding periods were excluded to ensure that it did not influence the observed behavioural patterns. Utilising the depth sensors integrated into each camera, the depth of maximum activity was determined by identifying the camera with the highest activity at each time point and recording its corresponding depth.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec8\"\u003e\n \u003ch2\u003e2.4 Statistical Analysis\u003c/h2\u003e\n \u003cp\u003eAll statistical analyses were conducted using the Python programming language (https://www.python.org/), with statistical significance set at \u0026alpha;\u0026thinsp;=\u0026thinsp;0.05. To independently validate the algorithm, 55 videos were randomly selected, and the number of fish visible on screen was counted by two evaluators. These abundance counts were then correlated with the corresponding activity (%) measured at each time point. The Intraclass Correlation Coefficient (ICC) was used to assess inter-rater reliability and to determine both absolute agreement and consistency between the evaluators. Mann-Whitney test was used to determine whether there were significant differences in activity, AGD, PGD, SFR, and mortality between the two time periods for each farm. The maximum activity was used for the multi-camera study cage, and this was aggregated at 15-minute intervals. Following this, model II linear regressions were performed to determine correlation between weekly-averaged maximum activity and gill health score with ANCOVA to determine significance for each farm independently. Coefficients of variation (COV) were used to determine the extent to which the level of dispersion of the depth of maximum activity was around the mean. Additionally, Aligned Ranked Transform ANOVA allowed for a comparison between each of the cages before and after the increase in activity. Lastly, a Generalised Linear Mixed Model was used to determine the effects of weekly-average temperature, AGD, PGD, and sea lice (\u003cem\u003eL. salmonis\u003c/em\u003e gravid females) on weekly-averaged activity with random effects included for cages and farms. All variables were z-transformed prior to analysis. Raw data is available through DataSTORRE (http://hdl.handle.net/11667/241).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Model Validation\u003c/h2\u003e \u003cp\u003eTo validate model efficacy, the algorithm was independently validated by using videos recorded using the same AKVA feed cameras. Evaluators assessed the algorithm\u0026rsquo;s 'activity' output and demonstrated a strong correlation with fish abundance counts from 55 clips (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.70, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e \u003cp\u003eIndependent validation demonstrated a strong correlation (R\u0026sup2; = 0.70, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) between fish abundance observed in feeding camera footage by human assessors and the still images from the same videos used to calculate activity via the algorithm. The ICC determined high correlation both in the absolute agreement (ICC\u0026thinsp;=\u0026thinsp;0.99) and the consistency (ICC\u0026thinsp;=\u0026thinsp;0.98) between the two evaluators. Both reflect strong agreement between the observers that was statistically significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). While fish speed and cohesion could not be separately validated, it is assumed that the 30% variation not captured is related to the speed and cohesion components.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Farm A: Activity and Health\u003c/h2\u003e \u003cp\u003eIn Farm A, there were seasonal effects on temperature. At the start of the study, on May 10, the temperature was 10.0 \u0026ordm;C within the study cage, with an upward trend reaching a peak of 14.7 \u0026ordm;C by June 23 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Throughout the remainder of the trial period the temperature maintained a consistent range, oscillating between 14.6 \u0026ordm;C and 15.3 \u0026ordm;C, ultimately settling at 15.4 \u0026ordm;C by the end of the study. While only temperature data was presented in the results for both farms, oxygen, turbidity, and salinity were also monitored throughout the study period, however these parameters showed only minor daily fluctuations and did not exhibit significant trends over the sampling period. As the behavioural responses were observed over several months, daily variations in these parameters were unlikely to significantly influence long-term behavioural patterns and therefore these results are included in the appendix (Fig. A.3).\u003c/p\u003e \u003cp\u003eThe depth of maximum fish activity, determined through the in-camera depth sensors, elucidated both the predominant depth distribution of the fish within the cage and the degree of their spatial cohesion (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThere was a pronounced shift in the activity and depth of the fish on July 7. In the 2 months prior to July 7, the average depth at which fish congregated was 10.0\u0026thinsp;\u0026plusmn;\u0026thinsp;2.8 m, with a COV of 34.5%. For 2 months after July 7, this depth of maximum activity then decreased to 9.0\u0026thinsp;\u0026plusmn;\u0026thinsp;1.4 m (U\u0026thinsp;=\u0026thinsp;1.32\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\:\\bullet\\:\\)\u003c/span\u003e\u003c/span\u003e 10\u003csup\u003e7\u003c/sup\u003e, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) with a notable decrease in the COV to 18.7% as the fish coalesced towards the central region of the cage, indicative of heightened cohesion among individuals. This transition occurred rapidly, with the COV decreasing from 28.5% one week prior to July 7 to 13.1% the week after.\u003c/p\u003e \u003cp\u003eMoreover, there was a significant increase in maximum activity from 37.5\u0026thinsp;\u0026plusmn;\u0026thinsp;13.1% to 60.8\u0026thinsp;\u0026plusmn;\u0026thinsp;11.5% (U\u0026thinsp;=\u0026thinsp;2.5 \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\bullet\\:\\)\u003c/span\u003e\u003c/span\u003e 10\u003csup\u003e6\u003c/sup\u003e, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) for 2 months before and after July 7, reflecting the intensified group cohesive behaviour as potential shoaling behaviour. This transition was rapid as well, from 32.3\u0026thinsp;\u0026plusmn;\u0026thinsp;11.2% one week prior to 56.8\u0026thinsp;\u0026plusmn;\u0026thinsp;17.0% on average one week post July 7. Concurrently, gill health assessments were conducted throughout this period, with the average AGD and PGD scores initially recorded as 0.4\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3 and 0.4\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2, respectively. Following the increase in activity, both AGD and PGD scores rose to 0.8\u0026thinsp;\u0026plusmn;\u0026thinsp;0.7 and 1.2\u0026thinsp;\u0026plusmn;\u0026thinsp;0.7 respectively. Subsequently, a correlation analysis revealed a non-significant association between AGD and activity (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05, R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.07), while a significant correlation was observed between PGD and activity (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.42; Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThere was a significant decline in the specific feeding rate (SFR) at Farm A, from an average of 2.3\u0026thinsp;\u0026plusmn;\u0026thinsp;0.7 prior to the augmented activity levels of the fish, to an average of 1.1\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3 after (U\u0026thinsp;=\u0026thinsp;1477, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Furthermore, mortality rates within the cage exhibited a slight, though non-significant, increase from 0.03\u0026thinsp;\u0026plusmn;\u0026thinsp;0.02% to 0.08\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1% with the increased activity on July 7 (U\u0026thinsp;=\u0026thinsp;23, p\u0026thinsp;=\u0026thinsp;0.24).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Farm B: Activity and Health\u003c/h2\u003e \u003cp\u003eThe study period for Farm B commenced later in the year, on June 22, with temperatures recorded at 11.8 \u0026ordm;C. Over the course of the study, temperatures steadily increased, reaching 14.7 \u0026ordm;C by September 7, before declining slightly to 13.3 \u0026ordm;C by October 13.\u003c/p\u003e \u003cp\u003eOn August 19, there was a significant increase in maximum fish activity from 32.5\u0026thinsp;\u0026plusmn;\u0026thinsp;10.5% to 51.1\u0026thinsp;\u0026plusmn;\u0026thinsp;19.0% (U\u0026thinsp;=\u0026thinsp;5.9\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\:\\bullet\\:\\)\u003c/span\u003e\u003c/span\u003e 10\u003csup\u003e5\u003c/sup\u003e, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). This occurred concurrently to a significant increase in depth from 5.8\u0026thinsp;\u0026plusmn;\u0026thinsp;1.8 m to 6.0\u0026thinsp;\u0026plusmn;\u0026thinsp;1.9 m (U\u0026thinsp;=\u0026thinsp;1.16\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\:\\bullet\\:\\)\u003c/span\u003e\u003c/span\u003e 10\u003csup\u003e6\u003c/sup\u003e, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), though it should be noted that this cage had a shallower depth overall. In a similar trend as Farm A, the maximum activity and COV transition quickly for the week prior and post-August 19, whereby the activity increased from 27.8\u0026thinsp;\u0026plusmn;\u0026thinsp;10.2% to 48.7\u0026thinsp;\u0026plusmn;\u0026thinsp;9.2%, and the COV declined sharply from 30.4\u0026ndash;13.2%.\u003c/p\u003e \u003cp\u003eIn contrast to Farm A, Farm B exhibited overall lower gill health scores, with AGD scores recorded at 0.1\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2 prior to the heightened fish activity and increasing to 0.4\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3 after. Similarly, PGD scores were recorded at 0.01\u0026thinsp;\u0026plusmn;\u0026thinsp;0.02 prior and 0.7\u0026thinsp;\u0026plusmn;\u0026thinsp;0.4 post-increase in activity. A significant correlation was established between AGD and activity at this site (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.35), however a stronger correlation was observed between PGD and activity (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.62; Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eDespite a slight decrease in SFR from 1.4\u0026thinsp;\u0026plusmn;\u0026thinsp;0.6 to 1.3\u0026thinsp;\u0026plusmn;\u0026thinsp;0.4, this change was not statistically significant (U\u0026thinsp;=\u0026thinsp;1568, p\u0026thinsp;\u0026gt;\u0026thinsp;0.05). However, there was a significant increase in mortality rates, rising from 0.03\u0026thinsp;\u0026plusmn;\u0026thinsp;0.01% to 0.2\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3% concurrent with the heightened activity observed on August 18 (U\u0026thinsp;=\u0026thinsp;7, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Farm-wide effects\u003c/h2\u003e \u003cp\u003eThe investigation extended to the remaining cages at each site, each equipped with a single camera centrally positioned at mid-depth, around 9 m for Farm A and 7 m for Farm B. A similar pattern emerged, characterised by significant increases in activity concomitant with increased PGD scores, decreased SFR, and increased mortality rates (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn Farm A, 2 months prior to July 7, the activity levels were 25.6\u0026thinsp;\u0026plusmn;\u0026thinsp;10.5%, increasing to 43.6\u0026thinsp;\u0026plusmn;\u0026thinsp;15.1% 2 months thereafter, with all cages demonstrating a significant elevation ([3.8\u0026ndash;22.9%; min \u0026ndash; max difference within cages]; p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Concurrently, AGD scores increased from 0.3\u0026thinsp;\u0026plusmn;\u0026thinsp;0.06 to 0.51\u0026thinsp;\u0026plusmn;\u0026thinsp;0.06 [0.04\u0026ndash;0.4], with PGD scores exhibiting a more pronounced increase from 0.3\u0026thinsp;\u0026plusmn;\u0026thinsp;0.09 to 1.1\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2 [0.4\u0026ndash;1.2] across all cages, each showing a significant rise (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). SFR exhibited a significant decline in all cages, from 1.3\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1 to 0.7\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1 ([-0.4\u0026ndash; -1.2]; p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Furthermore, there was a slight rise in mortality after July 7, on average from 0.2\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3% to 0.3\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3%, though this relatively small difference can be attributed to the low median mortality rate, with the upper quartiles in most cages indicating a trend towards higher mortality in the 2 months following July 7. There were 5 cages (6, 7, 8, 10 and 12) that did exhibit significantly elevated mortality (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e \u003cp\u003eSimilarly, in Farm B, the average activity levels increased from 24.9\u0026thinsp;\u0026plusmn;\u0026thinsp;7.0% to 32.6\u0026thinsp;\u0026plusmn;\u0026thinsp;9.6% on August 18, with all cages, except cages 3 and 5, exhibiting significant increases ([5.2\u0026ndash;16.8%; min \u0026ndash; max difference within cages]; p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). It should be noted however that the latter two cages had trends towards an increase even though it was not statistically significant. AGD scores increased from an average of 0.1\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2 to 0.6\u0026thinsp;\u0026plusmn;\u0026thinsp;0.4 [0.3\u0026ndash;0.7], while PGD scores increased from 0.1\u0026thinsp;\u0026plusmn;\u0026thinsp;0.2 to 0.9\u0026thinsp;\u0026plusmn;\u0026thinsp;0.5 which was a significant increase in all cages ([0.6\u0026ndash;1.2], p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). SFR decreased from average of 1.7\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3 to 1.4\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3 [-0.06 \u0026ndash; -0.7]; all cages were significantly lower except for cages 2 and 8 (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). The average mortality rate rose similarly to Farm A, from 0.2\u0026thinsp;\u0026plusmn;\u0026thinsp;0.9% to 0.3\u0026thinsp;\u0026plusmn;\u0026thinsp;0.5%, again with around half the cages (2, 4, 6, 8 and 9) showing significant increases in mortality (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e \u003cp\u003eFurthermore, a generalised linear mixed model was employed to examine the influence of temperature, AGD, PGD and sea lice (gravid \u003cem\u003eL. salmonis\u003c/em\u003e) on activity levels. The model revealed a significant effect of PGD on activity (0.32 [0.16\u0026ndash;0.48], p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), while AGD, temperature and gravid female \u003cem\u003eL. salmonis\u003c/em\u003e were non-significant (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05). The model\u0026rsquo;s predictive capacity for activity was substantial with an R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003emarginal\u003c/sub\u003e of 0.25 [0.12\u0026ndash;0.39; 95% confidence interval], particularly when accounting for the variation at the cage and farm levels (R\u003csup\u003e2\u003c/sup\u003e\u003csub\u003econditional\u003c/sub\u003e: 0.60 [0.54\u0026ndash;0.64]).\u003c/p\u003e \u003cp\u003eRegarding the Operational Welfare Indicators assessed at each farm, no significant increases were observed for any parameter at either farm (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05). Nonetheless, at Farm A, a trend toward leaner fish was discernible, evidenced by an increase in the occurrence of visual K-factor scores above 0 following the increase in activity, rising from 8 to 18 instances across all cages, though this was not mirrored at Farm B.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eGill disorders are a growing economic and welfare concern for the salmon aquaculture industry worldwide. Deterioration in gill health can precipitate significant mortality events, hinder salmon growth and production, and compromise overall welfare (Rozas-Serri, \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). The escalating temperatures due to climate change exacerbate factors contributing to compromised gill health. These factors may encompass physical stressors, including abrasions of the gill surface, as well as pathogenic or parasitic agents such as \u003cem\u003eN. perurans\u003c/em\u003e, causing AGD, hydrozoan jellyfish or other gelatinous zooplankton, as well as phytoplankton such as diatoms, which can instigate gill lesions (Kent et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e1995\u003c/span\u003e; Baxter et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2011\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe advancement of aquaculture monitoring has been significantly enhanced with the emergence of artificial intelligence (AI). The use of AI algorithms, through machine learning methodologies, facilitates the processing of extensive datasets, including the analysis of image and video data to discern patterns and trends, thereby enabling investigations such as the present study (Mandal and Ghosh, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Leveraging AI for pattern recognition can result in the detection of any anomalous behaviours, indicative of stress and compromised health responses. In this study, an algorithm based on machine vision was employed to analyse videos sourced from commercial fish farms to quantify fish group behaviour and compare temporal behavioural profiles. The resulting ‘activity’ parameter is a combination of the relative abundance of fish within the video frame, perceived average inter-fish distances, and group schooling behaviour in one metric to describe the fish behaviour overall. Since the machine vision model is trained end to end, there is no overarching weighted model that dynamically adjusting feature importance. Instead, the model learns each feature independently while estimating relative activity. This innovative approach allows researchers, as well as farmers, to establish correlations between behaviours, specifically related to group social cohesion and fish abundance in certain cage locations, and external factors, a task that traditionally demands considerable time and effort. Moreover, while many algorithms focus on individual fish behaviour (Liu et al., 2023), this approach allows farmers to observe the behaviour of the fish as a group, giving a more holistic view.\u003c/p\u003e \u003cp\u003eDeteriorated gill health was observed at two study sites situated within the Western Isles of Scotland. The initial occurrence was documented at Farm A in early June followed by a subsequent manifestation at Farm B in mid-August, 2023. Both sites exhibited increased levels of fish activity coinciding with higher gill health scores, particularly in terms of proliferative gill disease (PGD) scores, confirming the hypothesis that compromised gills would alter salmon behaviour. However, intriguingly, no significant correlations were discerned between the prevalence of Amoebic Gill Disease (AGD) specifically, ambient temperature or the presence of gravid female \u003cem\u003eL. salmonis.\u003c/em\u003e The utilisation of gravid females in this case was to investigate whether activity could be used as an indicator for sea lice infestation, to inform managers and potentially allow for pre-emptive actions prior to visual inspections, as gravid females have been correlated to the external infection pressure (Kragesteen et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). However, the investigation failed to establish any significant correlation between the presence of gravid female lice and alterations in fish activity levels. It is important to note that there were few sea lice counted during the study period, as such, this may be of interest to explore further in future studies.\u003c/p\u003e \u003cp\u003eMost studies focus on AGD as it can be detected relatively easily through histopathology and narrowed down to a singular amoeba rather than PGD which can be caused by a variety of infectious agents and has limited diagnostic test to detect it (Boerlage et al., 2022). Thus, the tests and treatments used for PGD are simply those used to test and treat AGD, specifically total histopathology score by Mitchell et al., (\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2012\u003c/span\u003e), or by gross gill observations described here. This study provides further information on how PGD can impact the behaviour and welfare of the fish, that cannot be observed through AGD tests alone. It should be noted that this study does not aim to diagnose the specific causes of PGD. Although turbidity remained similar throughout the study, few micro-jellyfish were observed and no changes in plankton counts, other potential factors influencing fish behaviour cannot be ruled out. Although a more detailed diagnostic approach could provide clarity on the etiology, such methods were beyond the scope of this farm-based study, where visual gill scoring by experienced farmers is the standard practice. This method has shown moderate to good agreement with histopathological diagnosis (Adams et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2004\u003c/span\u003e), though reliability decreases when symptoms are mild (Clark and Nowak, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e1999\u003c/span\u003e). Refinements to reduce subjectivity and explore additional behavioural indicators, particularly in early or mild cases, could enhance diagnostic accuracy. Despite this limitation, the observable deterioration of gills was linked to behavioural changes occurring over several months, indicating a strong relationship between gill health and fish behaviour. The authors acknowledge this as a limitation and suggest that future research incorporating more precise diagnostic techniques could further elucidate the relationship between specific gill diseases and behavioural responses.\u003c/p\u003e \u003cp\u003eSeveral OWIs were investigated in this study. These are important for fish welfare and can be easily implemented by farmers in their daily husbandry practices (Noble et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Various types of welfare indicators exist, each with its own set of strengths and weaknesses. For instance, environment-based indicators such as temperature and oxygen levels are rendered ineffective in the absence of extremes, as observed in this study where such indicators remained relatively stable. Consequently, animal-based indicators, both at the group- (behaviour, feeding) and individual- (gill status, external inspections of the fish, k-factor) levels, were employed. As behaviour began to change and cohesive group behaviour or shoaling was more prominent, there was a rise in gill health scores. However, there was no significant change in the outward signs of illness (e.g. no change to the eyes, fins, skin). Despite the advantage of individual animal-based indicators (ABI) in detecting outward signs of poor health, the sampling of only 10 fish per cage out of a population exceeding 30,000 individuals highlights a weakness. Additionally, relying solely on individual indicators may be insufficient, as by the time outward signs of illness become apparent, the fish may already be compromised. In contrast, group-based indicators (GBI) offer a more holistic approach, allowing for the observation of most fish within the cages. Moreover, behaviour is the only welfare indicator that provides farmers direct insight into the subjective experience of fish, while OWIs (i.e. physiological or external parameters) often reflect indirect, consequential effects of environmental or health stressors (Noble et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) and are usually based on negative experiences. Though these indirect indicators remain valuable for comprehensive welfare assessments, behaviour uniquely captures the immediate responses of fish, making it a critical tool for early detection of welfare concerns in farm settings. This means that managers can have a better understanding of the state of the fish in a non-invasive way, simply through their position in the water column and proximity to each other (Martins et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Noble et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eChanges to fish behaviour can be a warning sign that the fish are experiencing stress and reduced welfare. For example, social cohesion, such as schooling or shoaling, serves as an effective antipredator adaptation in fish (Morgan and Godin 1995; Krause et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2002\u003c/span\u003e; Huntingford et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). This response allows for the potential dilution effect, as it decreases the odds of being predated upon. This mechanism may also translate into a response to compromised health, as stressed fish are more likely to shoal (Kleinhappel et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Moreover, a recent study on the effect of fish schooling on swimming noise production suggested that schooling fish emit lower sound levels, potentially shielding the group from predators while optimising hydrodynamic efficiency (Zhou et al., \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). However, the consequences of this behaviour are not fully understood; while increased shoaling is typically associated with elevated risk of disease transmission, such as AGD, among nearby conspecifics, it may also confer protective benefits through mechanisms akin to herd immunity or the dilution effect (Mikheev, \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). Further research is needed to clarify whether shoaling primarily facilitates the spread of pathogens or may serve as a defence strategy against them. Moreover, such aggregation behaviour can lead to overcrowding in localised areas of the cage, potentially reducing oxygen availability and further impacting welfare, although this was not observed in the present study.\u003c/p\u003e \u003cp\u003eAcute stressors also disrupt feeding behaviour, leading to reduced food intake and growth, which can affect their energetic costs and ultimately impact their welfare (Morgenroth et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This study observed a decline in specific feeding rates following the onset of gill damage. The impact of compromised gill health on feeding behaviour was demonstrated primarily at Farm A, where a significant decline in the SFR was observed following the onset of deteriorated gill health concomitant with increased activity levels. While a downward trend in SFR was also discernable at Farm B, it was not statistically significant. Reduced feeding often considered a primary symptom of stress in fish, alongside changes to swimming behaviour, underscores the importance of capturing these changes through metrics such as SFR (Conte, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2004\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThis study also documented rises in mortality within the majority of cages at Farm A, along with discernible trends towards increased mortality, albeit not statistically significant, at Farm B. While mortality is a commonly used group-based OWI in production systems, it is important to acknowledge the inherent limitation, as it signifies an irreversible stage in the deterioration of fish welfare, precluding the implementation of corrective measures (Ellis et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Noble et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). The limitation is underscored by the findings in this study, as significant mortalities were observed at Farm A, correlating with the highest levels of AGD and PGD, alongside significant increases in activity. Conversely, Farm B, characterised by lesser but still notable increases in gill health issues and activity, did not exhibit significant rises in mortality in most of the cages. This discrepancy shows that relying on mortality alone as an indicator would overlook the manifestations of poor welfare discerned in this study through behavioural changes. If mortality is to remain an operational welfare indicator, it signals issues at too late a stage. To enhance its utility, individual-based assessments should be complemented by comprehensive behavioural monitoring. This combined approach enables earlier detection of welfare concerns, improving fish health and welfare outcomes.\u003c/p\u003e \u003cp\u003eThe effect sizes differed notably between the two sites, with Farm A exhibiting poorer gill health, heightened activity levels and mortality, and more pronounced reduction in SFR compared to Farm B. The discrepancy in gill health scores between the two farms likely underlies the variation in response, with Farm A recording higher scores, resulting in a lesser increase in activity, and absence of a significant decrease in SFR at Farm B. Several factors may contribute to the observed disparities between the two sites. Firstly, it is pertinent to acknowledge the highly contextual natures of PGD, with the specific pathogenic composition underlying PGD manifestation varying across sites (Król et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Furthermore, differences in environmental exposure levels warrant consideration, particularly regarding the location of Farm B, which is characterised by robust tidal-driven currents, potentially facilitating the transport of harmful agents such as \u003cem\u003eN. perurans\u003c/em\u003e, micro-jellyfish, plankton, and other debris capable of damaging the gills. Additionally, temporal discrepancies in fish stocking, with Farm A stocked 3 months prior to Farm B, result in differences in fish size between the two cohorts. Lastly, variations in broodstock between the sites may further contribute to disparities in response to gill health challenges. Previous studies have elucidated the influence of genetic background on behavioural traits and disease susceptibility in Atlantic salmon (Davidson et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e1989\u003c/span\u003e; Bui et al., 2017). However, it is important to note that despite the comparatively smaller magnitude of change in activity observed at Farm B, both sites exhibited analogous behavioural shifts in response to compromised gill health. This suggests a conserved behavioural response related to survival among Atlantic salmon, irrespective of genetics, age, or environmental conditions, underscoring the impact of poor gill health-induced stress on fish behaviour.\u003c/p\u003e \u003cp\u003eAs climate change progresses and temperatures continue to rise, the agents causing PGD are expected to proliferate, leading to higher rates of illness. Over the past 4 decades, sea surface temperature around the UK have exhibited a warming trend of approximately 0.3 ℃ per decade (Cornes et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). This warming trend may broaden the reproductive periods and improve winter survival of jellyfish and plankton, causative agents of PGD, at mid-latitudes (Licandro et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Boero et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Kennerley et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Moreover, research indicates that temperatures exceeding 12 ℃ can trigger outbreaks of AGD, potentially increasing the attachment and growth capacity of \u003cem\u003eN. perurans\u003c/em\u003e under elevated temperature conditions (Benedicenti et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Consequently, the establishment of efficient early detection mechanisms for illnesses is imperative to mitigate stress and mortality among farmed fish. (Barreto et al., 2021). This study provides insight into the role of behavior in detecting PGD, facilitating the implementation of preventative or treatment interventions.\u003c/p\u003e "},{"header":"Conclusion","content":"\u003cp\u003eIn this study, a pronounced increase in fish activity occurred at the emergence of poor gill health in two Atlantic salmon farms in Scotland. During the summer and early autumn of 2023, both farms experienced a significant increase in fish activity concomitant with the onset of PGD. This was accompanied by a decline in SFR and increase in mortality rates, particularly evident in Farm A. Given the escalating temperatures associated with climate change, it is conceivable that the prevalence of agents precipitating PGD will escalate, underscoring the need for the aquaculture industry to devise effective mitigation strategies. Gill health is a leading cause of mortality in the global Atlantic salmon industry, thus is it is imperative that studies continue in this area to improve the welfare of farmed fish. The findings of this study hold significant implications for future studies, which can delve deeper into elucidating the activity thresholds indicative of impending gill health deterioration. This could lead to the development of a non-invasive early warning system for identifying compromised gill health. Such a tool could prove invaluable in proactively addressing welfare concerns in commercial aquaculture systems.\u003c/p\u003e\u003cp\u003eThe main challenges currently facing aquaculture are based on the improvement of best management practices and the advancement of monitoring systems, both for the aquatic environment and for assessing fish behaviour, health, and welfare. Implementing a systematic and integrated welfare assessment framework that spans the entire production cycle up to slaughter is essential to improve the understanding of basic needs of the fish, mitigating stressors and stress responses, and ultimately enhancing productivity and product quality. The development of Precision Fish Farming (PFF), incorporating system modelling and machine learning, represents a significant advancement in aquaculture technology. This approach offers promising solutions to welfare-related challenges, including those explored in the present study.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Animal Welfare and Ethical Review Body (AWERB) at the University of Stirling (Approval No. 2022 8795 8155) and was conducted in accordance with institutional and national ethical guidelines for animal research.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets supporting the conclusions of this article are available in the DataSTORRE repository, http://hdl.handle.net/11667/241.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported by Innovate UK Smart Grants (10028961), in partnership with Observe Technologies and AKVA Group.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCRediT authorship contribution statement:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMB:\u0026nbsp;\u003c/strong\u003eWriting \u0026ndash; original draft, Formal analysis, Investigation, Visualisation\u003cstrong\u003e. DN:\u0026nbsp;\u003c/strong\u003eWriting \u0026ndash; review \u0026amp; editing, Methodology, Software, Formal analysis, Investigation, Resources, Data curation\u003cstrong\u003e. PF:\u0026nbsp;\u003c/strong\u003eWriting \u0026ndash; review \u0026amp; editing,\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eConceptualization, Methodology, Software, Validation, Resources, Data curation\u003cstrong\u003e. HR:\u0026nbsp;\u003c/strong\u003eConceptualization, Methodology, Software, Validation, Resources, Project administration, Funding acquisition. \u003cstrong\u003eTT:\u0026nbsp;\u003c/strong\u003eWriting \u0026ndash; review \u0026amp; editing, Conceptualization, Methodology, Supervision\u003cstrong\u003e. SRP:\u0026nbsp;\u003c/strong\u003eWriting \u0026ndash; review \u0026amp; editing, Conceptualization, Methodology, Supervision, Project administration, Funding acquisition\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments:\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors are grateful to the dedicated personnel at each farm for their diligent maintenance of the cameras. Special thanks to Michael Caldwell and AKVA Group for deploying the cameras and promptly addressing any technical issues that arose.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eAdams MB, Ellard K, Nowak BF. Gross pathology and its relationship with histopathology of amoebic gill disease (AGD) in farmed Atlantic salmon, Salmo salar L. Journal of Fish Diseases. 2004;27(3):151–61.\u003c/li\u003e\n \u003cli\u003eBarreto MO, Rey Planellas S, Yang Y, Phillips C, Descovich K. Emerging indicators of fish welfare in aquaculture. Reviews in Aquaculture. 2022;14(1):343–61.\u003c/li\u003e\n \u003cli\u003eBaxter E, Rodger H, McAllen R, Doyle T. Gill disorders in marine-farmed salmon: investigating the role of hydrozoan jellyfish. Aquacult Environ Interact. 2011;1(3):245–57.\u003c/li\u003e\n \u003cli\u003eBenedicenti O, Pottinger TG, Collins C, Secombes CJ. Effects of temperature on amoebic gill disease development: Does it play a role? Journal of Fish Diseases. 2019;42(9):1241–58.\u003c/li\u003e\n \u003cli\u003eBoerlage AS, Ashby A, Herrero A, Reeves A, Gunn GJ, Rodger HD. Epidemiology of marine gill diseases in Atlantic salmon (Salmo salar) aquaculture: a review. Reviews in Aquaculture. 2020;12(4):2140–59.\u003c/li\u003e\n \u003cli\u003eBoero F, Brotz L, Gibbons M, Piraino S, Zampardi S. Ocean warming: impacts and effects of ocean warming on jellyfish. In: Laffoley D, Baxter JM, editors. \u003cem\u003eExplaining ocean warming: Causes, scale, effects and consequences\u003c/em\u003e. Gland, Switzerland: IUCN; 2016;213-237.\u003c/li\u003e\n \u003cli\u003eBui S, Dalvin S, Dempster T, Skulstad OF, Edvardsen RB, Wargelius A, Oppedal F. Susceptibility, behaviour, and retention of the parasitic salmon louse (Lepeophtheirus salmonis) differ with Atlantic salmon population origin. Journal of Fish Diseases. 2018;41(3):431–42.\u003c/li\u003e\n \u003cli\u003eBurke M, Nikolic D, Fabry P, Rishi H, Telfer T, Rey Planellas S. Precision Farming in Aquaculture: Non-invasive monitoring of Atlantic salmon (\u003cem\u003eSalmo salar\u003c/em\u003e) behaviour in response to environmental conditions in commercial sea cages for health and welfare assessment. Frontiers in Robotics and AI. 2025:1574161.\u003c/li\u003e\n \u003cli\u003eCarr JA. Stress, Neuropeptides, and Feeding Behavior: A Comparative Perspective1. Integrative and Comparative Biology. 2002;42(3):582–90.\u003c/li\u003e\n \u003cli\u003eCerqueira M, Millot S, Felix A, Silva T, Oliveira GA, Oliveira CCV, Oliveira S, Rey S, MacKenzie S, Oliveira R. Cognitive appraisal in fish: stressor predictability modulates the physiological and neurobehavioural stress response in sea bass. Proceedings of the Royal Society B: Biological Sciences. 2020;287(1923):20192922.\u003c/li\u003e\n \u003cli\u003eClark A, Nowak BF. Field investigations of amoebic gill disease in Atlantic salmon, Salmo salar L., in Tasmania. Journal of Fish Diseases. 1999;22:433-443.\u003c/li\u003e\n \u003cli\u003eConte FS. Stress and the welfare of cultured fish. Applied Animal Behaviour Science. 2004;86(3):205–23.\u003c/li\u003e\n \u003cli\u003eCornes RC, Tinker J, Hermanson L, Oltmanns M, Hunter WR, Lloyd-Hartley H, Kent EC, Rabe B, Renshaw R. Climate change impacts on temperature around the UK and Ireland. MCCIP Science Review. 2023;1-18.\u003c/li\u003e\n \u003cli\u003eDavidson WS, Birt TP, Green JM. A review of genetic variation in Atlantic salmon, Salmo salar L., and its importance for stock identification, enhancement programmes and aquaculture. Journal of Fish Biology. 1989;34(4):547–60.\u003c/li\u003e\n \u003cli\u003eDeck CA, Honeycutt JL, Cheung E, Reynolds HM, Borski RJ. Assessing the Functional Role of Leptin in Energy Homeostasis and the Stress Response in Vertebrates. Frontiers in Endocrinology. 2017; 8. \u003c/li\u003e\n \u003cli\u003eEllis T, Berrill I, Lines J, Turnbull JF, Knowles TG. Mortality and fish welfare. Fish Physiol Biochem. 2012;38(1):189–99.\u003c/li\u003e\n \u003cli\u003eFolkedal O, Stien LH, Torgersen T, Oppedal F, Olsen RE, Fosseidengen JE, Braithwaite V, Kristiansen, TS. Food anticipatory behaviour as an indicator of stress response and recovery in Atlantic salmon post-smolt after exposure to acute temperature fluctuation. Physiology \u0026amp; Behavior. 2012;105(2):350–6.\u003c/li\u003e\n \u003cli\u003eFoyle KL, Hess S, Powell MD, Herbert NA. What Is Gill Health and What Is Its Role in Marine Finfish Aquaculture in the Face of a Changing Climate? Frontiers in Marine Science. 2020; 7(400):1-16. \u003c/li\u003e\n \u003cli\u003eHerrero A, Thompson KD, Ashby A, Rodger HD, Dagleish MP. Complex Gill Disease: an Emerging Syndrome in Farmed Atlantic Salmon (\u003cem\u003eSalmo salar\u003c/em\u003e L.). Journal of Comparative Pathology. 2018;163:23–8.\u003c/li\u003e\n \u003cli\u003eHerrero A, Rodger H, Hayward AD, Cousens C, Bron JE, Dagleish MP, Thompson K. Prospective Longitudinal Study of Putative Agents Involved in Complex Gill Disorder in Atlantic salmon (Salmo salar). Pathogens. 2022;11(8):878.\u003c/li\u003e\n \u003cli\u003eHoare DJ, Couzin ID, Godin JGJ, Krause J. Context-dependent group size choice in fish. Animal Behaviour. 2004;67(1):155–64.\u003c/li\u003e\n \u003cli\u003eHuntingford, F., Coyle, S., Hunter, W., 2012. Avoiding Predators, in: Aquaculture and Behavior. John Wiley \u0026amp; Sons, Ltd, pp. 220–247. https://doi.org/10.1002/9781444354614.ch8\u003c/li\u003e\n \u003cli\u003eHvas M, Kolarevic J, Noble C, Oppedal F, Stien LH. Fasting and its implications for fish welfare in Atlantic salmon aquaculture. Reviews in Aquaculture. 2024;16(3):1308-1332.\u003c/li\u003e\n \u003cli\u003eJaervi T, Uglem I. Predator training improves the anti-predator behaviour of hatchery reared Atlantic salmon(Salmo salar) smolt. Nordic journal of freshwater research Drottningholm. 1993;(68):63–71.\u003c/li\u003e\n \u003cli\u003eJuell JE. The behaviour of Atlantic salmon in relation to efficient cage-rearing. Rev Fish Biol Fisheries. 1995;5(3):320–35.\u003c/li\u003e\n \u003cli\u003eKennerley A, Wood LE, Luisetti T, Ferrini S, Lorenzoni I. Economic impacts of jellyfish blooms on coastal recreation in a UK coastal town and potential management options. Ocean \u0026amp; Coastal Management. 2022;227:106284.\u003c/li\u003e\n \u003cli\u003eKent M, Whyte J, LaTrace C. Gill lesions and mortality in seawater pen-reared Atlantic salmon Salmo salar associated with a dense bloom of Skeletonema costatum and Thalassiosira species. Dis Aquat Org. 1995;22:77–81.\u003c/li\u003e\n \u003cli\u003eKleinhappel TK, Pike TW, Burman OHP. Stress-induced changes in group behaviour. Sci Rep. 2019;9(1):17200.\u003c/li\u003e\n \u003cli\u003eKlinger DH, Levin SA, Watson JR. The growth of finfish in global open-ocean aquaculture under climate change. Proc R Soc B. 2017;284(1864):20170834.\u003c/li\u003e\n \u003cli\u003eKolsäter L. Feed management and reduction of aquaculture wastes. Water Science and Technology. 1995;31(10):213–8.\u003c/li\u003e\n \u003cli\u003eKragesteen T, Simonsen K, Visser A, Andersen K. Estimation of external infection pressure and salmon-louse population growth rate in Faroese salmon farms. Aquacult Environ Interact. 2021;13:21–32.\u003c/li\u003e\n \u003cli\u003eKrause J, Godin JGJ. Shoal Choice in the Banded Killifish (Fundulus diaphanus, Teleostei, Cyprinodontidae): Effects of Predation Risk, Fish Size, Species Composition and Size of Shoals. Ethology. 1994;98(2):128–36.\u003c/li\u003e\n \u003cli\u003eKrause J, Hensor EMA, Ruxton GD. Fish as prey. In: Hart PJB, Reynolds JD, editors. \u003cem\u003eHandbook of Fish Biology and Fisheries\u003c/em\u003e. Hoboken, NJ: Wiley-Blackwell; 2002.\u003c/li\u003e\n \u003cli\u003eKról E, Noguera P, Shaw S, Costelloe E, Gajardo K, Valdenegro V, Bickerdike R, Douglas A, Martin SAM. Integration of Transcriptome, Gross Morphology and Histopathology in the Gill of Sea Farmed Atlantic Salmon (Salmo salar): Lessons From Multi-Site Sampling. Front Genet 2020;11. \u003c/li\u003e\n \u003cli\u003eLai F, Royan MR, Gomes AS, Espe M, Aksnes A, Norberg B, Gelebart V, Rønnestad I. The stress response in Atlantic salmon (Salmo salar L.): identification and functional characterization of the corticotropin-releasing factor (crf) paralogs. General and Comparative Endocrinology. 2021;313:113894.\u003c/li\u003e\n \u003cli\u003eLi D, Wang G, Du L, Zheng Y, Wang Z. Recent advances in intelligent recognition methods for fish stress behavior. Aquacultural Engineering. 2022;96:102222.\u003c/li\u003e\n \u003cli\u003eLicandro P, Conway DVP, Daly Yahia MN, Fernandez de Puelles ML, Gasparini S, Hecq JH, Tranter P, Kirby RR. A blooming jellyfish in the northeast Atlantic and Mediterranean. Biology Letters. 2010;6(5):688–91.\u003c/li\u003e\n \u003cli\u003eLiu J, Bienvenido F, Yang X, Zhao Z, Feng S, Zhou C. Nonintrusive and automatic quantitative analysis methods for fish behaviour in aquaculture. Aquaculture Research. 2022;53(8):2985–3000.\u003c/li\u003e\n \u003cli\u003eMadaro A, Olsen RE, Kristiansen TS, Ebbesson LOE, Nilsen TO, Flik G, Gorissen M. Stress in Atlantic salmon: response to unpredictable chronic stress. Journal of Experimental Biology. 2015;218(16):2538–50.\u003c/li\u003e\n \u003cli\u003eMagurran AE, Pitcher TJ, Dodd JM. Provenance, shoal size and the sociobiology of predator-evasion behaviour in minnow shoals. Proceedings of the Royal Society of London Series B Biological Sciences. 1997;229(1257):439–65.\u003c/li\u003e\n \u003cli\u003eMandal A, Ghosh AR. Role of artificial intelligence (AI) in fish growth and health status monitoring: a review on sustainable aquaculture. Aquaculture International. 2023;32:2791-2820. \u003c/li\u003e\n \u003cli\u003eMarine Scotland, The Scottish Government. Amoebic gill disease. \u003cem\u003eTopic Sheet No. 96 V1\u003c/em\u003e. 2012. Available from: http://www.scotland.gov.uk/Resource/0039/00393037.pdf\u003c/li\u003e\n \u003cli\u003eMartins CIM, Galhardo L, Noble C, Damsgård B, Spedicato MT, Zupa W, Beauchaud M, Kulczykowska E, Massabuau J-C, Carter T, Rey Planellas S, Kristiansen T. Behavioural indicators of welfare in farmed fish. Fish Physiology and Biochemistry. 2012;38(1):17–41.1.\u003c/li\u003e\n \u003cli\u003eMellor D, Beausoleil N. Extending the ‘Five Domains’ model for animal welfare assessment to incorporate positive welfare states. Animal welfare. 2015;24:241–253. https://doi.org/10.7120/09627286.24.3.241\u003c/li\u003e\n \u003cli\u003eMikheev VN. Combined effects of predators and parasites on shoaling behavior of fishes. Journal of Ichthyology. 2009; 49(11):1032-1041. \u003c/li\u003e\n \u003cli\u003eMiller N, Gerlai R. From schooling to shoaling: Patterns of collective motion in Zebrafish (Danio rerio). PLOS ONE. 2012;7(11):e48865.\u003c/li\u003e\n \u003cli\u003eMitchell SO, Baxter EJ, Holland C, Rodger HD. Development of a novel histopathological gill scoring protocol for assessment of gill health during a longitudinal study in marine-farmed Atlantic salmon (Salmo salar). Aquacult Int. 2012;20(5):813–25.\u003c/li\u003e\n \u003cli\u003eMorgan MJ, Godin JGJ. Antipredator Benefits of Schooling Behaviour in a Cyprinodontid Fish, the Banded Killifish (Fundulus diaphanus). Zeitschrift für Tierpsychologie. 1985;70(3):236–46.\u003c/li\u003e\n \u003cli\u003eMorgenroth D, Kvaestad B, Økland F, Finstad B, Olsen RE, Svendsen E, Rosten C, Axelsson M, Bloecher N, Føre M, Gräns A. Under the sea: How can we use heart rate and accelerometers to remotely assess fish welfare in salmon aquaculture? Aquaculture. 2024;579:740144.\u003c/li\u003e\n \u003cli\u003eMustafa A, MacWilliams C, Fernandez N, Matchett K, Conboy GA, Burka JF. Effects of sea lice (Lepeophtheirus salmonis Kröyer, 1837) infestation on macrophage functions in Atlantic salmon (Salmo salar L.). Fish \u0026amp; Shellfish Immunology. 2000;10(1):47–59.\u003c/li\u003e\n \u003cli\u003eNoble C, Gismervik K, Iversen MH, Kolarevic J, Nilsson J, Stien LH, Turnbull JF. Welfare indicators for farmed Atlantic salmon: tools for assessing fish welfare. 2018. 351 p.\u003c/li\u003e\n \u003cli\u003eO’Donncha F, Stockwell CL, Planellas SR, Micallef G, Palmes P, Webb C, Filgueira R, Grant J. Data Driven Insight Into Fish Behaviour and Their Use for Precision Aquaculture. Frontiers in Animal Science. 2021;2. \u003c/li\u003e\n \u003cli\u003eOppedal F, Dempster T, Stien LH. Environmental drivers of Atlantic salmon behaviour in sea-cages: A review. Aquaculture. 2011;311(1):1–18.\u003c/li\u003e\n \u003cli\u003eOrr JA, Vinebrooke RD, Jackson MC, Kroeker KJ, Kordas RL, Mantyka-Pringle C, Van den Brink, De Laender F, Stoks R, Holmstrup M, Matthaei CD, Monk WA, Penk MR, Leuzinger S, Schäfer RB, Piggott JJ. Towards a unified study of multiple stressors: divisions and common goals across research disciplines. Proceedings of the Royal Society B: Biological Sciences. 2020;287(1926):20200421.\u003c/li\u003e\n \u003cli\u003eØverli Ø, Nordgreen J, Mejdell CM, Janczak AM, Kittilsen S, Johansen IB, Horsberg TE. Ectoparasitic sea lice (Lepeophtheirus salmonis) affect behavior and brain serotonergic activity in Atlantic salmon (Salmo salar L.): Perspectives on animal welfare. Physiology \u0026amp; Behavior. 2014;132:44–50.\u003c/li\u003e\n \u003cli\u003ePitcher TJ. Heuristic definitions of fish shoaling behaviour. Animal Behaviour. 1983;31(2):611–3.\u003c/li\u003e\n \u003cli\u003eRey S, Huntingford FA, Boltaña S, Vargas R, Knowles TG, Mackenzie S. Fish can show emotional fever: stress-induced hyperthermia in zebrafish. Proceedings of the Royal Society B. 2015;282(1819):20152266.\u003c/li\u003e\n \u003cli\u003eRodger HD. Gill disorders: an emerging problem for farmed Atlantic salmon (\u003cem\u003eSalmo salar\u003c/em\u003e) in the marine environment? Fish Veterinary Journal. 2007;9:38–48.\u003c/li\u003e\n \u003cli\u003eRozas-Serri M. Gill diseases in marine salmon aquaculture with an emphasis on amoebic gill disease. CAB Reviews Perspectives in Agriculture Veterinary Science Nutrition and Natural Resources. 2019;14:1–15.\u003c/li\u003e\n \u003cli\u003eRuehle B, Poulin R. Hunger games: foraging behaviour and shelter use in fish under the context-dependent influence of parasitism. Parasitology Research. 2021;120(11):3681–92.\u003c/li\u003e\n \u003cli\u003eSamaras A, Dimitroglou A, Kollias S, Skouradakis G, Papadakis IE, Pavlidis M. Cortisol concentration in scales is a valid indicator for the assessment of chronic stress in European sea bass, Dicentrarchus labrax L. Aquaculture. 2021;545:737257.\u003c/li\u003e\n \u003cli\u003eSantos EGN, Cunha RA, Santos CP. Behavioral responses of Poecilia vivipara (Osteichthyies: Cyprinodontiformes) to experimental infections of Acanthocollaritrema umbilicatum (Digenea: Cryptogonimidae). Experimental Parasitology. 2011;127(2):522–6.\u003c/li\u003e\n \u003cli\u003eShinn AP, Pratoomyot J, Bron JE, Paladini G, Brooker EE, Brooker AJ. Economic costs of protistan and metazoan parasites to global mariculture. Parasitology. 2015;142(1):196–270.\u003c/li\u003e\n \u003cli\u003eSusdorf R, Salama NKG, Todd CD, Hillman RJ, Elsmere P, Lusseau D. Context-dependent reduction in somatic condition of wild Atlantic salmon infested with sea lice. Marine Ecology Progress Series. 2018;606:91–104.\u003c/li\u003e\n \u003cli\u003eTaylor RS, Muller WJ, Cook MT, Kube PD, Elliott NG. Gill observations in Atlantic salmon (\u003cem\u003eSalmo salar,\u003c/em\u003e L.) during repeated amoebic gill disease (AGD) field exposure and survival challenge. Aquaculture. 2009;290(1):1–8.\u003c/li\u003e\n \u003cli\u003eVillegas-Ch W, García-Ortiz J, Sánchez-Viteri S. Toward Intelligent Monitoring in IoT: AI Applications for Real-Time Analysis and Prediction. IEEE Access. 2024;12:40368–86.\u003c/li\u003e\n \u003cli\u003eWendelaar Bonga SE. The stress response in fish. Physiological Reviews. 1997;77(3):591–625.\u003c/li\u003e\n \u003cli\u003eYang L, Liu Y, Yu H, Fang X, Song L, Li D, Chen Y. Computer Vision Models in Intelligent Aquaculture with Emphasis on Fish Detection and Behavior Analysis: A Review. Arch Computat Methods Eng. 2021;28(4):2785–816.\u003c/li\u003e\n \u003cli\u003eZhang D, Sogn-Grundvåg G, Tveterås R. The impact of parasitic sea lice on harvest quantities and sizes of farmed salmon. Aquaculture. 2023;576:739884.\u003c/li\u003e\n \u003cli\u003eZhang L, Li B, Sun X, Hong Q, Duan Q. Intelligent fish feeding based on machine vision: A review. Biosystems Engineering. 2023;231:133–64.\u003c/li\u003e\n \u003cli\u003eZhou J, Seo JH, Mittal R. Effect of schooling on flow generated sounds from carangiform swimmers. Bioinspired Biomimetic. 2024;19(3):036015.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"aquaculture-science-and-management","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Aquaculture Science and Management](https://aquaculturesciencemanagement.biomedcentral.com/)","snPcode":"44365","submissionUrl":"https://submission.springernature.com/new-submission/44365/3","title":"Aquaculture Science and Management","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Atlantic salmon, Aquaculture, Fish behaviour, Welfare, Gill health, Machine learning, Precision farming","lastPublishedDoi":"10.21203/rs.3.rs-6653910/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6653910/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAs the aquaculture industry grows, more sophisticated technology is required to monitor farms and ensure good fish welfare, in line with the precision livestock farming concept. Using behaviour as a non-invasive monitoring tool, combined with artificial intelligence, enables greater control over farm management. This study aimed to assess temporal changes in farmed Atlantic salmon (\u003cem\u003eSalmo salar\u003c/em\u003e) group behavioural profiles related to fish health and welfare. A machine vision algorithm applied to feed cameras on commercial farms was used to determine whether changes in gill health would induce visible group behavioural changes. Video cameras were deployed in all cages at two Scottish Atlantic salmon marine farms. One cage at each farm was also equipped with additional cameras (5 and 4 at sites A and B, respectively) to provide higher spatial coverage of fish behaviour and distribution. The algorithm processed video footage from these cameras and produced behavioural data termed \u0026lsquo;activity\u0026rsquo; (%), which encompasses fish abundance, speed, and shoal cohesion. Additionally, gill health, Operational Welfare Indicators (OWI), mortality, and Specific Feeding Rate (SFR) were scored weekly at both sites. During summer 2023, gill health issues arose at both farms, leading to fish stress reflected in the behavioural data. For two months prior to the onset of poor gill health, the average (\u0026plusmn;\u0026thinsp;standard deviation) activity levels of the fish across all cages were 25.6\u0026thinsp;\u0026plusmn;\u0026thinsp;10.5% and 24.9\u0026thinsp;\u0026plusmn;\u0026thinsp;7.0% for Farm A and B, respectively. After gill health was compromised, the activity rose significantly for two months in all cages with a mean of 43.6\u0026thinsp;\u0026plusmn;\u0026thinsp;15.1% and 32.6\u0026thinsp;\u0026plusmn;\u0026thinsp;9.6%, respectively. A generalised linear mixed model revealed that Proliferative Gill Disease (PGD) was the main driver of this increase in activity. This increase in activity coincided with fish migration to the centre of the cage, meaning tighter shoaling, which is a normal stress response often seen in relation to predators and other environmental or health stressors. The use of behaviour as a non-invasive welfare indicator and the potential to use artificial intelligence to automate the process of behavioural identification allows farmers to improve welfare conditions and ensure industry sustainability.\u003c/p\u003e","manuscriptTitle":"Precision Farming in Aquaculture: Assessing gill health in Atlantic salmon (Salmo salar) using a non-invasive, AI-driven behavioural monitoring approach in commercial farms","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-20 14:01:06","doi":"10.21203/rs.3.rs-6653910/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-06-10T17:18:56+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-03T19:17:28+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-26T16:39:48+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"263557080830448783085520629661115296657","date":"2025-05-23T04:42:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"202115397493297244289230279526024594216","date":"2025-05-22T05:17:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"240911980657715300569673459568988847554","date":"2025-05-20T05:29:12+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"167535592186126172813968576836122777975","date":"2025-05-16T09:54:49+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-05-16T08:29:53+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-05-14T05:18:31+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-05-14T05:18:27+00:00","index":"","fulltext":""},{"type":"submitted","content":"Aquaculture Science and Management","date":"2025-05-13T09:22:05+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"aquaculture-science-and-management","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Aquaculture Science and Management](https://aquaculturesciencemanagement.biomedcentral.com/)","snPcode":"44365","submissionUrl":"https://submission.springernature.com/new-submission/44365/3","title":"Aquaculture Science and Management","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"19aa8a2c-0a7e-4446-b94d-85e776bf3396","owner":[],"postedDate":"May 20th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-07-26T08:08:14+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-20 14:01:06","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6653910","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6653910","identity":"rs-6653910","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}