Global AI model may protect freshwater fish from extinction

Nearly one in three freshwater fish species now faces possible extinction, a statistic that reflects mounting pressure on rivers and lakes worldwide. Habitat loss, dams, pollution, invasive species, and climate shifts often interact in complicated ways. That complexity has made it difficult for conservation managers to know where to act before populations collapse.

A research team led by University of Maine assistant professor Christina Murphy set out to change that timing problem. Instead of focusing only on species already in trouble, they built a system designed to flag risks earlier, while prevention is still realistic.

After five years of work, the group developed a machine learning model that evaluates extinction risk for more than 10,000 freshwater fish species worldwide. Their findings were published in Nature Communications.

The results point toward a hopeful conclusion: many species could still be protected before they reach endangered status.

Traditional conservation assessments often rely on population trends, geographic ranges, and known threats. Those factors remain important, but they do not always capture the broader context shaping species survival.

Murphy’s team incorporated 52 variables drawn from 12 global data sources, most from the International Union for Conservation of Nature Red List. The model included environmental conditions, human economic factors, and species characteristics, while deliberately excluding variables already used directly in conservation listings.

Researchers trained artificial intelligence to analyze millions of nonlinear relationships across species and habitats. The system then classified species as either imperiled or non-imperiled.

The model achieved an overall accuracy of about 88 percent. It correctly identified non-imperiled species roughly 90 percent of the time, compared with about 81 percent accuracy for imperiled ones.

That difference turned out to be revealing.

“Our results suggest conservation works like human health: the signals of ‘well-being’ are often more consistent than the many pathways to illness,” said co-author J. Andres Olivos, a postdoctoral researcher at Oregon State University.

In other words, stable ecological conditions appear more predictable than the many combinations of pressures that drive species toward extinction.

Environmental and socioeconomic factors played the largest role in predicting conservation status. Intrinsic biological traits contributed less than 10 percent of the model’s predictive power.

Species classified as non-imperiled tended to live in regions with reliable water availability, moderate levels of river impoundment, minimal habitat disruption, and relatively low human footprint. Stable economic conditions also correlated with healthier fish populations.

Extreme values, whether environmental or economic, often aligned with higher extinction risk.

One of the strongest predictors was hydro-geomorphic diversity, which reflects how many different habitat types exist within a species’ range. High diversity sometimes indicated fragmented environments and poor connectivity between habitats, conditions known to contribute to population decline.

Taxonomic grouping also mattered. Species within the same evolutionary orders often responded similarly to environmental stress, which helped the model detect patterns across related fish.

The analysis also revealed an unexpected factor: how much scientists know about a species.

Fish with either very limited data or extensive research were more likely to be classified as imperiled. That pattern suggests that uncertainty and risk aversion may influence conservation decisions.

Nearly half of the assessed species lacked at least 30 attributes in available datasets, highlighting significant information gaps. Geographic and taxonomic biases in conservation assessments may also affect outcomes, since species from wealthier regions and those described earlier in scientific history tend to receive more attention.

The authors caution that their findings represent a first step toward understanding global patterns rather than definitive explanations for every species. Local threats within a species’ range may still escape detection in broad analyses.

Murphy began the project in 2020 as a postdoctoral researcher at Oregon State University, working with associate professor Ivan Arismendi, Olivos, and collaborators from the U.S. Geological Survey, the U.S. Forest Service, and the University of Girona in Spain.

One goal was to move conservation planning away from crisis response.

“People sometimes go in to protect species when it’s already too late,” Arismendi said. “With our model, decision makers can deploy resources in advance before a species becomes imperiled.”

Murphy emphasized that the model highlights conditions that already support species success.

“This uses new metrics to identify what is working to keep species from being listed,” she said. “Managers may be able to protect a lot of fish.”

Because many species share similar environmental needs, the approach could allow conservation programs to protect multiple species simultaneously rather than tackling each individually.

Researchers also suggest that similar modeling frameworks could be adapted for birds, plants, and other wildlife groups.

The study offers conservation planners a way to prioritize prevention instead of emergency recovery.

By identifying regions where environmental and socioeconomic conditions support healthy fish populations, managers could target protections before declines occur.

That shift could reduce costs, regulatory barriers, and ecological damage associated with late-stage interventions, potentially improving outcomes for freshwater ecosystems worldwide.

Research findings are available online in the journal Nature Communications.

The original story "Global AI model may protect freshwater fish from extinction" is published in The Brighter Side of News.