Past (more) perfect
New technique may sharpen accuracy for measuring past temperatures.
Ateam of astrophysicists, palaeontologists and mathematicians used machine-learning algorithms – originally developed by gravitational wave astrophysicists – to improve the accuracy of a “paleothermometer”, which looks at fossil evidence of past climate change to predict Earth’s future.
Ice cores and tree rings are both examples of paleothermometers. By studying the trapped air bubbles within ice or the oxygen isotope ratio of tree ring cellulose, researchers can reconstruct the composition of Earth’s atmosphere over millions of years.
Led by palaeontologist
Tom Dunkley Jones, from the University of Birmingham,
UK, the team instead studied biomarkers left over from singlecelled organisms called archaea, dating as far back as the Cretaceous (145–66 million years ago).
Archaea produce compounds called Glycerol Dialkyl Glycerol Tetraethers (GDGTS). In modern oceans, the abundance of
GDGT varies with the local sea temperature, “most likely driven by the need for increased cellmembrane stability and rigidity at higher temperatures,” the researchers explain in their paper in the journal Climate of the Past.
Archaea preserved in ancient marine sediments therefore have the potential to provide a long-term geologic record of the planet’s surface temperatures.
“After several decades of study, the best available models are only able to measure temperature from GDGT concentrations with an accuracy of around 6°C,” says co-researcher Ilya Mandel, gravitational wave astrophysicist at Australia’s
ARC Centre of Excellence in Gravitational Wave Discovery (Ozgrav).
The team turned to machine learning tools that are used in gravitational wave astronomy to create predictive models of merging objects like black holes and neutron stars.
The accuracy of the model nearly doubled, from 6°C to 3.6°C.