Offbeat

Science

Machine learning helps geoboffins spot huge beds of hot rocks 1,000km across deep below Earth's surface

Large structures were detected as anomalies in seismic waves processed by an algorithm


Geophysicists have uncovered large swathes of hot, dense rock lying nearly 3,000 kilometers beneath Earth's surface, hidden below the Pacific Ocean, thanks to an unsupervised learning algorithm.

These structures, described as "ultralow-velocity zones" (ULVZs) in a study published in Science on Friday, were detected by measuring the echoes of seismic waves emitted from earthquakes underground. "Ultralow-velocity zones are areas of unusually dense, hot material at the core-mantle boundary," Doyeon Kim, first author of the paper and a scientist at Cornell University, explained to The Register: "Seismic waves travel up to 30 per cent slower in ULVZs than through surrounding mantle materials."

The reflections of these seismic signals reveal that there are obstacles in the way as the waves travel in between the source of the earthquake and reach the seismometer. By studying the recordings and measuring the amount of diffraction in the echoes, the researchers can estimate how dense or large these objects are.

"We found echoes on about 40 per cent of all seismic wave paths," said Vedran Lekić, co-author of the study and an associate professor of geology at the University of Maryland. "That was surprising because we were expecting them to be more rare, and what that means is the anomalous structures at the core-mantle boundary are much more widespread than previously thought."

Some of the large areas of rock had previously been discovered, but there were new regions that haven't been seen before, like the one lurking below the Marquesas Islands, a tiny group of volcanic islands in the middle of the Southern Pacific Ocean. The team also found that the ultralow-velocity zone underneath the Hawaiian Islands is much larger than previously thought.

"The scale of typical ULVZs found elsewhere are on the order of about 100km. But what we discuss in our paper is an order of magnitude larger than those typical ULVZs, a structure about 1,000km across. ULVZs of this extraordinary size are called 'mega-ULVZs'. We detected two mega-ULVZs, one beneath Hawaii and the other beneath Marquesas Islands, both are located near the edge of the Pacific LLSVPs," Kim told us.

The team collected 7,000 seismogram recordings of the echoes and fed them to a machine-learning algorithm known as Sequencer. Developed by researchers at Tel Aviv University and John Hopkins University, the algorithm uses unsupervised learning to "reveal the main trend or sequence in a dataset." In other words, Sequencer is good at highlighting any anomalies – or the echoes – in data.

"Machine learning in earth science is growing rapidly and a method like Sequencer allows us to be able to systematically detect seismic echoes and get new insights into the structures at the base of the mantle, which have remained largely enigmatic," said Kim.

The researchers don't know what these structures are made out of and how they might affect Earth's plate tectonics over time. "Although still enigmatic we think all of these lower mantle structures are somewhat dynamically related as our mantle convects and the Earth keeps cooling down," he concluded.

"This mantle convection is the driving mechanism for hotspot volcanism and plate tectonics. And the Earth is the only terrestrial planet in our solar system where we uniquely observe plate tectonics and rarely observe such manifestation of volcanisms seen from the surface. Therefore, understanding mantle structures and their characteristics is important to learn how the Earth has evolved or evolving as opposed to other planets." ®

Send us news
18 Comments

Nvidia rival Cerebras says it's revived Moore's Law with third-gen waferscale chips

Startup is also working with Qualcomm on optimized models for its Cloud AI 100 Ultra inference chips

Can AI shorten PC replacement cycles? Dell seems to think so

Might be wishful thinking from finance exec following two-year computer sector recession

Meta seeks ASIC designers for ML accelerators and datacenter SoCs

Appears to be struggling to find them, even in India, as it's re-posted job ads

Quilter's AI design service nabs $10M to make circuit board design easier

Claims what took weeks to do by hand, machine learning can do in hours

NASA finally launches PACE Earth science satellite

'New era of ocean science' hoped to follow debut of billion-dollar plankton-spotter

IBM Japan and NTT think they can make datacenter aircon adjust to different workloads

They're measuring server exhaust temperatures to detect power consumption

What is Model Collapse and how to avoid it

We chat to AI expert Ilia Shumailov about the pitfalls of using machines to train machines

Robots with a 'Berliner Schnauze' may appear more trustworthy to locals

Dialect study a mixed bag when it comes to droids speaking highbrow German

DPD chatbot blasts courier company, swears, and dabbles in awful poetry

So this is the AI revolution

Big Cloud deploys thousands of GPUs for AI – yet most appear under-utilized

If AWS, Microsoft, Google were anywhere close to capacity, their revenues would be way higher

Daughter of George Carlin horrified someone cloned her dad with AI for hour special

Seven words you can't say – This won't backfire and ruin art forever

EU lassos tech giants in bid to rein in the AI Wild West

Interpretation, debate, and judges set to decide how the rubber hits the road