Enlarge / The cycle of building and breaking up of supercontinents seems to drive long-term climate trends. (credit: Walter Myers/Stocktrek Images)

Global temperature records go back less than two centuries. But that doesn't mean we have no idea what the world was doing before we started building thermometers. There are various things—tree rings, isotope ratios, and more—that register temperatures in the past. Using these temperature proxies, we've managed to reconstruct thousands of years of our planet's climate.

But going back further is difficult. Fewer proxies get preserved over longer times, and samples get rarer. By the time we go back past a million years, it's difficult to find enough proxies from around the globe and the same time period to reconstruct a global temperature. There are a few exceptions, like the Paleocene-Eocene Thermal Maximum (PETM), a burst of sudden warming about 55 million years ago, but few events that old are nearly as well understood.

Now, researchers have used a combination of proxy records and climate models to reconstruct the Earth's climate for the last half-billion years, providing a global record of temperatures stretching all the way back to near the Cambrian explosion of complex life. The record shows that, with one apparent exception, carbon dioxide and global temperatures have been tightly linked. Which is somewhat surprising, given the other changes the Earth has experienced over this time.

Enlarge / Artist's conception of the state of the Earth during its global glaciations. (credit: NASA)

Earth has gone through many geologic phases, but it did have one striking period of stasis: Our planet experienced a tropical environment where algae and single-celled organisms flourished for almost 2 billion years. Then things changed drastically as the planet was plunged into a deep freeze.

It was previously unclear when Earth became a gargantuan freezer. Now, University College London researchers have found evidence in an outcrop of rocks in Scotland, known as the Port Askaig Formation, that show evidence of the transition from a tropical Earth to a frozen one 717 million years ago. This marks the onset of the Sturtian glaciation and would be the first of two "snowball Earth" events during which much of the planet’s surface was covered in ice. It is thought that multicellular life began to emerge after Earth thawed.

Found in the Scottish islands known as the Garvellachs, this outcrop within the Port Askaig Formation is unique because it offers the first conclusive evidence of when a tropical Earth froze over—underlying layers that are a timeline from a warmer era to a frigid one. Other rocks that formed during the same time period in other parts of the world lack this transitional evidence because ancient glaciers most likely scraped it off.

Enlarge / Image of some of the atmospheric circulation seen during NeuralGCM runs. (credit: Google)

Right now, the world's best weather forecast model is a General Circulation Model, or GCM, put together by the European Center for Medium-Range Weather Forecasts. A GCM is in part based on code that calculates the physics of various atmospheric processes that we understand well. For a lot of the rest, GCMs rely on what's termed "parameterization," which attempts to use empirically determined relationships to approximate what's going on with processes where we don't fully understand the physics.

Lately, GCMs have faced some competition from machine-learning techniques, which train AI systems to recognize patterns in meteorological data and use those to predict the conditions that will result over the next few days. Their forecasts, however, tend to get a bit vague after more than a few days and can't deal with the sort of long-term factors that need to be considered when GCMs are used to study climate change.

On Monday, a team from Google's AI group and the European Centre for Medium-Range Weather Forecasts are announcing NeuralGCM, a system that mixes physics-based atmospheric circulation with AI parameterization of other meteorological influences. Neural GCM is computationally efficient and performs very well in weather forecast benchmarks. Strikingly, it can also produce reasonable-looking output for runs that cover decades, potentially allowing it to address some climate-relevant questions. While it can't handle a lot of what we use climate models for, there are some obvious routes for potential improvements.