Pyrite - that brassy mineral your geology teacher called "fool's gold" - has been keeping receipts on Earth's atmosphere for over three billion years. And a team of researchers just taught an algorithm to read them.

The Planet That Learned to Breathe

Here's the thing about oxygen: we take it for granted, but Earth spent most of its existence as a suffocating wasteland where you'd keel over faster than you could say "primordial soup." The air we're breathing right now? It's basically a cosmic accident that took billions of years to happen.

Scientists have long known about two major oxygen booms - the Great Oxidation Event around 2.4 billion years ago (when cyanobacteria figured out photosynthesis and promptly caused a mass extinction), and the Neoproterozoic Oxygenation Event roughly 700-540 million years ago (which set the stage for complex critters like, eventually, us). But the finer details? Those have been murky at best.

Enter Zhang, Tang, and Cheng with their shiny new dataset: trace element concentrations from pyrite grains spanning 3.5 billion years of geological history. They fed this mountain of geochemical data into machine learning algorithms and asked a simple question: what's actually driving oxygen levels on this planet?

Fool's Gold, Actual Treasure

Why pyrite? Turns out, when sedimentary pyrite forms in ancient oceans, it captures trace elements like a chemical time capsule. Elements like molybdenum, uranium, and vanadium get incorporated into the mineral structure, and their concentrations tell you about ocean chemistry and, by extension, how much oxygen was floating around.

The researchers identified two distinct groups of trace elements in their analysis. One group - the redox-sensitive metals - responds directly to oxygen availability. The other reflects hydrothermal influences, those underwater volcanic vents that belch chemicals into the deep ocean. Separating these signals is like distinguishing between background noise and the actual music, and that's exactly where machine learning earns its keep.

Life Pumps, Continents Shuffle

The big reveal: Earth's long-term oxygen trend tracks remarkably well with biosphere expansion. More life means more photosynthesis means more oxygen. The correlation between those two major oxygenation events (GOE and NOE) and biological expansion isn't subtle - it's practically a duet.

But biology isn't working alone. Superimposed on that steady biological drumbeat are shorter-term fluctuations that dance to a tectonic rhythm. Continental assembly - when landmasses smoosh together into supercontinents - tends to correlate with oxygenation episodes. The mechanism makes sense: mountain building accelerates weathering, washing nutrients into the ocean, fertilizing photosynthetic organisms, and burying organic carbon before it can rot and consume oxygen.

Supercontinent breakup, on the other hand, associates with more reducing (low-oxygen) conditions. Large igneous provinces pump volcanic gases into the atmosphere, overwhelming the oxygen supply. It's like the planet alternates between exhaling and holding its breath every few hundred million years.

A Feedback Loop the Size of a Planet

What emerges is a picture of Earth as a coupled system - biology, geology, and chemistry locked in a feedback loop spanning billions of years. Cyanobacteria didn't just oxygenate the atmosphere; they transformed the planet's chemistry, which in turn shaped which organisms could evolve, which further changed the chemistry, and so on.

This framework does more than satisfy academic curiosity. Understanding how Earth maintained habitable conditions through billions of years of upheaval - snowball glaciations, supercontinent cycles, mass extinctions - informs our search for life on other worlds. If oxygen accumulation requires this particular symphony of biology and tectonics, that constrains where else we might find breathable atmospheres.

The study also highlights how machine learning combined with geochemical big data can extract signals from the geological record at resolutions previously impossible. When you're trying to reconstruct conditions from 2 billion years ago using microscopic mineral grains, every analytical edge matters.

The Bottom Line

Life made Earth's oxygen. Tectonics modulated its variability. And machine learning, trained on billions of years of chemical evidence locked in humble pyrite, finally sorted out who did what when. Not bad for fool's gold.

References

• Zhang, Z-J., Tang, D-J., & Cheng, Q-M. (2026). Biosphere expansion drives Earth's secular oxygenation while tectonics modulate oxygen variability revealed by machine learning. Proceedings of the National Academy of Sciences. DOI: 10.1073/pnas.2536681123

• Chen, C., et al. (2022). Reconstructing Earth's atmospheric oxygenation history using machine learning. Nature Communications, 13, 5862. DOI: 10.1038/s41467-022-33388-5

• Large, R.R., et al. (2014). Trace element content of sedimentary pyrite as a new proxy for deep-time ocean - atmosphere evolution. Earth and Planetary Science Letters. ScienceDirect

• Shields-Zhou, G. & Och, L. (2011). The case for a Neoproterozoic Oxygenation Event: Geochemical evidence and biological consequences. GSA Today, 21(3), 4-11. Geological Society of America

• Schirrmeister, B.E., et al. (2015). Cyanobacteria and the Great Oxidation Event: evidence from genes and fossils. Palaeontology. PMC4755140

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.