AI pinpoints when fish reach their thermal limit

Researchers at Nagoya University and the University of Fukui have created an artificial intelligence (AI)-based system that identifies the moment when fish lose their balance due to increased water temperatures - a key indicator of thermal stress. The system uses DeepLabCut, a deep-learning technology that tracks the fish's body posture from video footage, and ResNet34, another deep-learning method used for image classification.

The team, led by Tomoya Nakayama and Yoshiya Matsuo, collaborated with Associate Professor Tatsuto Hasegawa and Takuya Kato. Their work was published on September 10, 2026, in the journal Scientific Reports.

In aquatic ecosystems, rising water temperatures due to climate change can be detrimental for fish and other species adapted to specific temperature conditions. Since fish are ectothermic animals unable to regulate their body temperature, changes in water temperature directly impact their physiology and behavior. Therefore, understanding the thermal limits of various fish species is crucial for predicting their response to climate change.

Traditionally, researchers have assessed fish thermal tolerance by observing video recordings of fish as they gradually adapt to different water temperatures and identifying the point at which the fish can no longer maintain its balance. However, this method is prone to subjectivity, leading to variations in results and requiring significant time and effort to analyze large populations of fish.

To address these limitations, the research team developed an automated system that photographs individual fish in separate compartments and employs DeepLabCut for precise tracking of seven body landmarks, including the fish's nose, fins, body segments, and tail. By combining the information on these movements with video data, an AI model determines the fish's state and automatically flags when the fish reaches the thermal loss of equilibrium (LOE) point.

The accuracy of the AI-based determinations was found to be comparable to those made by experienced researchers, and the system produced consistent results upon repeated analyses of the same videos, making it well-suited for large-scale experiments.

The researchers conducted trials using different strains of medaka and closely related fish species to investigate their thermal tolerance. They discovered that temperature tolerance varied among the strains, with higher-latitude populations generally demonstrating greater cold tolerance. Notably, the Japanese medaka (Oryzias latipes) exhibited the highest cold tolerance, while the recently described Taiwanese medaka relative (Oryzias cabaranensis) showed greater cold tolerance than expected based on its habitat's latitude.

This finding suggests that factors beyond latitude may play a role in species adaptation to temperature changes.

Overall, the researchers have demonstrated a reliable AI method for objectively and reproducibly evaluating fish thermal tolerance, which has the potential to facilitate large-scale comparisons among strains and species. This technology could significantly contribute to predicting the impacts of climate change on fish populations and inform conservation efforts.