A team of researchers from IRTA and the Universitat Rovira e Virgili, in Spain, has developed an Artificial Intelligence-based system that could transform the production of Sole (Solea senegalensis). The model is capable of predicting spawning nights with an accuracy of 90% to 100% through an automated and continuous analysis of reproductive behaviour throughout the night — a task that until now relied entirely on manual observation.
The greaest challenge in farming this species remains the inability of captive-born broodstock to successfully complete courtship and spawning. Addressing this limitation is essential to achieving sector self-sufficiency and reducing dependence on wild-caught specimens. The issue affects males in particular, although the underlying causes are still unknown. As a result, producers are forced each season to rely on wild broodstock – a limited, costly and ultimately unsustainable resource that also adds regulatory and logistical uncertainty.
While a definitive solution to this biological problem remains elusive, the system developed at IRTA can reliably anticipate the nights on which wild broodstock are most likely to spawn.
To achieve this, the researchers combined computer vision (YOLOv8), individual tracking algorithms (DeepSORT) and a predictive model capable of integrating key behaviours such as Rest the Head, Guardian, Follow and locomotor activity. Among these, locomotor activity emerged as the strongest predictor, helping streamline future implementations and reduce operational costs.
A system with the potential to revolutionise broodstock management
This automated approach has the potential to completely reshape reproductive dynamics in captivity. By analysis full nights ( 17:00 – 23:00), detecting reproductive behaviours even when subtle, and issuing alerts when the likelihood of spawning increase, farms can better capitalize on periods when captive-bred fish show reproductive activity, optimise egg collection without relying on constant manual surveillance, schedule staff only when necessary and gather large volumes of data to select better-performing captive-born broodstock.
Altogether, this significantly enhances the reproductive efficiency of farmed stocks and reduces the need to introduce wild specimens to maintain production. In the long term, improvements in data-driven genetic selection and husbandry practices could finally close the gap between wild and captive-born fish.
The authors highlight that this technology is compatible with monitoring system already installed in most facilities and can be integrated into automation processes associated with Aquaculture 4.0. They also note that it could be adapted for other species facing similar reproductive challenges, extending its potential impact across the sector.
For an industry that has spent years striving for reproductive self-sufficiency in sole farming, this development marks a turning point: artificial intelligence is no longer a distant promise but a concrete, reliable tool ready to be implemented in production systems.
Reference:
Qadir, A., Duncan, N., González-López, W. Á., Fatsini, E., & Serratosa, F. (2025). Automated prediction of spawning nights using machine learning analysis of flatfish behaviour. Smart Agricultural Technology, 101668.
https://doi.org/10.1016/j.atech.2025.101668
Related
Latest In-depth
-
Bacteriohages in aquaculture: What should we know?
Phage therapy – a natural alternative to antibiotics – is gaining ground in aquaculture