News Release 

Machine learning greatly reduces uncertainty in understanding of paleozoic biodiversity

American Association for the Advancement of Science

Previous analyses of global paleobiodiversity have been coarsely resolved to roughly 10 million years, obscuring the effects of ecological processes and events that operate at shorter timescales. Now, by combining ancient marine fossils with modern machine learning and one of the world's most powerful supercomputers, researchers have composed a new record of Paleozoic biodiversity in which the age of average fossil layers can be resolved to within 26,000 years, the authors say. The computational approach allowed Jun-xuan Fan and colleagues to generate a new Cambrian-Triassic biodiversity curve with a refined temporal resolution of 26 ± 14.9 thousand years. "This new level of dating specificity is similar to moving from a system in which all people who lived in the same century are considered to be contemporaries to one in which only people who lived during the same 6-month period are deemed to be contemporaries," writes Peter Wagner in a related Perspective. To achieve this, Fan et al. developed a novel and custom-designed machine learning procedure and used the "Tianhe-2" supercomputer to synthesize data from nearly 11,000 Paleozoic marine invertebrate species recovered from more than 3,000 stratigraphic sections across China and Europe. The greatly improved resolution of the resulting paleobiodiversity curve clarified the timing of known diversification and extinction events while revealing many new, once-hidden aspects of Paleozoic biodiversity. "Supercomputer implementation that has resulted from this project will now become more or less standard in these types of biodiversity analyses throughout Earth sciences," said co-author Norman MacLeod in an accompanying video. The results also revealed a correlation between atmospheric carbon dioxide and paleobiodiversity change; however, due to the lack of a long-term and high-resolution paleoenvironmental data, more study is required to understand any causal links, the authors say.

###

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.