Machine Learning Study Suggests Spinosaurus Lived Mostly Underwater While Early Marine Reptiles Kept Returning to Land

A new study from Yale University suggests that Spinosaurus, the giant predatory dinosaur of ancient northern Africa, spent the vast majority of its time submerged in water. The same analysis found that some early marine reptiles long assumed to be fully aquatic actually spent considerable time on land. The findings were published in the journal Current Biology.

Many major groups of mammals and reptiles made an evolutionary U-turn, returning to water after their ancestors had conquered land. According to the researchers, these groups evolved in strikingly similar ways, developing flippers and other swimming features, which makes them textbook examples of convergent evolution. Yet scientists have long disagreed about when particular extinct groups became fully adapted to life in water, because fossils often offer incomplete or conflicting clues. Some skeletal features point to swimming, while others suggest an animal could still move and feed on land.

To tackle the problem, the team examined hundreds of fossil and modern specimens held at the Yale Peabody Museum and dozens of other institutions worldwide. They collected new data, including more than 11,000 measurements, along with photographs and CT scans. Lead author Caleb Gordon, who completed his doctorate at Yale earlier this year and is now a postdoctoral researcher at the Florida Museum of Natural History, developed relatively simple machine learning models trained on living animals. The models link skeletal proportions to known lifestyles and account for evolutionary relationships among species. They can then estimate how extinct animals used aquatic environments and predict soft tissue features, such as flippers, that rarely survive in fossils.

The researchers also revived a statistical technique from the 1940s. Co-author Jacques Gauthier said it was developed during World War II to estimate the probability that an approaching blip on a radar screen was an enemy plane. Gordon used it to assess whether an extinct animal had limbs modified for swimming. The researchers found that forelimb proportions were especially useful for distinguishing lifestyles, and their models predicted with more than 90 percent accuracy whether an animal had soft tissue flippers and whether it was highly or fully aquatic.

For Spinosaurus, which lived roughly 113 million to 94 million years ago, the debate has been lively. Some evidence portrays an animal that dived and chased prey underwater like a seal or penguin, while other findings suggest a heron-like wader hunting along shorelines and in shallows. Gordon said the team confidently recovered highly aquatic habits for Spinosaurus, indicating it spent the vast majority of its time submerged. The results do not settle every question about how it hunted, but they strengthen the case for the more aquatic view.

Mesosaurs, small marine reptiles that lived between 290 million and 274 million years ago in what is now southern Africa and eastern South America, told a different story. They gave birth to live young, which some researchers had taken as a sign of a fully aquatic life. But the new analysis indicates their limbs were not adapted for living entirely in water. Gordon said they spent a lot of time on land, like a modern alligator or platypus. More broadly, the researchers report that all Paleozoic marine reptiles, those that lived before the dinosaurs appeared, kept returning to land regularly, and even the most water-adapted among them remained amphibious.

Senior author Bhart-Anjan Bhullar, an associate professor of Earth and planetary sciences at Yale, said that the method removes subjectivity from the process of reconstructing life habits and lets the data speak for themselves. Traditionally, scientists have argued for particular lifestyles on the basis of one preferred set of skeletal features or another.

The source material does not describe specific limitations of the study, and its conclusions rest on models that infer behavior from skeletal proportions, so they are estimates rather than direct observations. As next steps, the researchers suggest the approach could eventually be applied to other major transitions, such as when human ancestors began walking on two legs and how the dinosaur ancestors of birds evolved flight. The study's additional co-authors are Lisa Freisem of Yale and Christopher Griffin of Princeton University.