Weather IA
climat

How the Five Megaclimatic States of Earth Explain Mass Extinctions

An international study reveals that over 539 million years, Earth has experienced only five megaclimatic states. Each transition between these regimes coincided with a peak in mass extinctions, including the Big Five. The work combines fossils, isotopes, and artificial intelligence to detect these climate shifts.

WE

Rédaction Weather IA

mercredi 23 septembre 2026 à 19:158 min
Partager :Twitter/XFacebookWhatsApp
How the Five Megaclimatic States of Earth Explain Mass Extinctions

Over 539 million years, Earth has occupied only five major climatic states, and each transition coincided with a rise in mass extinctions. This discovery, presented by an international team led by paleobiologist Corinne "Cori" Myers of the University of New Mexico, challenges the traditional view of a "random" climate. By linking regime changes to extinction episodes, researchers offer a new perspective on the mechanisms that shaped terrestrial biodiversity and could illuminate future risks related to rapid climate change.

What the study reveals: precise results and key figures

Analyzing over 12,000 data points from sediment cores, isotopic archives, and fossils, scientists identified five distinct megaclimatic states, summarized as periods of "intense glacial climate", "stable temperate climate", "warm and humid climate", "moderate arid climate", and "hyper-tropical climate". Each state lasted between 30 and 120 million years, with abrupt transitions that usually last less than one million years.

Transitions between these states occurred near the five recognized mass extinctions – Ordovician-Silurian, Devonian, Permian-Triassic, Triassic-Jurassic, and Cretaceous-Paleogene. For example, the shift from intense glacial climate to warm and humid climate around 252 million years ago preceded the Permian extinction, which eliminated nearly 96% of marine species. Authors report that extinction incidence increases by 3.5 times during the 200,000 years surrounding each transition, a statistically robust signal beyond natural noise.

Applying a predictive model based on machine learning, researchers could reproduce extinction patterns with 87% accuracy. This ability to anticipate extinction peaks from past climate changes suggests that future rapid transitions – such as those induced by current warming – could trigger similar biodiversity losses if critical thresholds are crossed.

How: research method and data used

The core of the study relies on a convolutional neural network trained to recognize climatic signatures in paleoclimate time series data. Network inputs include measurements of isotopic ratios (oxygen-18/16, carbon-13/12), microfossil records, as well as derived variables from modern climate models like ECMWF reanalyses and Copernicus observations. This machine learning approach allowed classifying millions of years of data into five distinct states while quantifying the prediction uncertainty associated with each classification.

To calibrate results, scientists

Was this article helpful?

Commentaires

Connectez-vous pour laisser un commentaire