Reconstructing Earth’s atmospheric oxygenation history using machine learning

Reconstructing historical atmospheric oxygen (O(2)) levels at finer temporal resolution is a top priority for exploring the evolution of life on Earth. This goal, however, is challenged by gaps in traditionally employed sediment-hosted geochemical proxy data. Here, we propose an independent strategy—machine learning with global mafic igneous geochemistry big data to explore atmospheric oxygenation over the last 4.0 billion years. We observe an overall two-step rise of atmospheric O(2) similar to the published curves derived from independent sediment-hosted paleo-oxybarometers but with a more detailed fabric of O(2) fluctuations superimposed. These additional, shorter-term fluctuations are also consistent with previous but less well-established suggestions of O(2) variability. We conclude from this agreement that Earth’s oxygenated atmosphere may therefore be at least partly a natural consequence of mantle cooling and specifically that evolving mantle melts collectively have helped modulate the balance of early O(2) sources and sinks.