Traditional air quality models struggle with changing emissions, while pure AI systems operate as opaque "black boxes" that can break the laws of physics. To solve this, researchers developed ProNet, a framework that embeds AI directly into atmospheric equations. Tested across 338 cities, ProNet outperformed experienced human forecasters and accurately predicted extreme events—like severe dust storms—without producing "AI hallucinations." Crucially, it also acts as a scientific diagnostic tool, using routine hourly data to uncover and correct hidden flaws in existing chemical models.

Using satellite observations, researchers tracked water storage recovery after 468 severe droughts in large river basins worldwide. Most events were followed by recovery, but persistent deficits clustered in drylands and some inland basins. Recovery trajectories varied across basins and events, reflecting water availability after drought. Recurrent droughts and intensive human water use were associated with prolonged deficits.

Carbonate cerium and neodymium isotope records from South China show that low-oxygen waters invaded shallow habitats twice during the Late Ordovician Mass Extinction, with a brief recovery between the two pulses. Earth-system models suggest that rapid climate swings reorganized continental weathering and nutrient delivery, amplifying marine oxygen loss and ecological stress.

When two tectonic plates collide, one plate sinks below the other – a process known as “subduction”. An international research team led by Göttingen University has discovered a rock in Papua New Guinea that, four million years ago, travelled over 90 km down into a subduction zone. The rock records an unusual journey, with higher-than-expected temperature conditions at a depth of around 45 km. This change from very hot subduction at this depth to the more common colder subduction at its final depth challenges scientists’ view of the thermal structure of subduction zones. The results were published in Nature Geoscience.

Global climate and weather forecasting use predictive models which divide the Earth’s atmosphere into a grid, with each section typically covering tens to hundreds of kilometers. Now a new model, run by a team at the University of Tokyo, has narrowed this grid spacing down to just 220 meters. This higher resolution removed instances of “popcornlike” rain, bursts of heavy rain which appear on simulations but not in real life, providing more accurate predictions of extreme local weather from a global model.