Knowledge, Innovation and Technology in Scientific Research Vol. 1

Precision aquaculture increasingly converts continuous environmental, behavioural and physiological data into management signals, creating an opportunity to recognise compromised fish welfare before mortality, injury or production loss becomes evident. This critical narrative review examines whether current biosensors, behavioural phenotyping systems and real-time stress-detection approaches can support biologically valid and operationally useful welfare assessment in farmed finfish. The literature was selected from open scholarly sources through 4 July 2026, with emphasis on studies that linked sensor outputs to established welfare indicators, stress physiology or production-relevant events. Evidence was synthesised across implantable and externally attached biosensors, heart-rate and acceleration bio-logging, biochemical sensing, computer vision, acoustic telemetry and machine-learning approaches. The strongest evidence concerns repeated or continuous measurement of heart rate, acceleration, depth, swimming activity and ventilation, particularly in Atlantic salmon and other intensively farmed species. Yet most sensor outputs are proxies rather than direct welfare states. Cortisol, heart rate, ventilation and altered swimming can each reflect stress, activity, circadian state, temperature, oxygen availability or coping style, and their interpretation is therefore context dependent. Animal-borne devices provide rich physiological information but can themselves alter growth or behaviour and usually sample only sentinel fish. Camera-based systems are less invasive and increasingly field capable, but performance is vulnerable to turbidity, occlusion, lighting variation, density and domain shift. Emerging water-based cortisol sensors reduce handling, although group-level dilution, hydrodynamics and biological lag complicate inference. The evidence supports a transition from single-threshold alarms towards multimodal, context-aware baselines that combine environmental exposure with behavioural and physiological response. The principal translational challenge is no longer signal acquisition alone, but prospective validation of welfare inference, decision thresholds and intervention benefit under commercial conditions.