Researchers at the Chinese Academy of Sciences used a deep-learning neural network to analyze over two million seismograms from 1990 to 2024, uncovering 174,929 high-quality PKP wave weak signals. Published in the Journal of Geophysical Research: Solid Earth, this AI-driven geological map reveals interconnected thermo-chemical structures and six previously undocumented seismic anomalies.
Decoding Earth’s Deep Mantle with Deep Learning
For decades, mapping the deep interior of our planet has remained a grueling manual endeavor. Geophysicists sift through historical seismic archives searching for faint signatures known as PKP wave precursors. These signals scatter off minute, heterogeneous structures near the boundary where the lower mantle transitions into a liquid core, roughly 2,900 kilometers beneath our feet. These deep dynamics drive tectonic plate motion, earthquakes, and volcanic activity.
Manual searches through millions of seismic recordings demand immense time and labor. To bypass this bottleneck, researchers deployed a specialized neural network. The algorithm automatically sorted records by quality and crunched more than two million seismograms generated by approximately 5,000 earthquakes spanning a 35-year archive from 1990 to 2024, according to findings published in the Journal of Geophysical Research: Solid Earth.
Human oversight kept the machine-learning pipeline grounded. Geophysicists manually verified and corrected model errors, feeding those polished examples back into the training loop to refine algorithmic accuracy. Ultimately, the artificial intelligence pipeline flagged 174,929 high-quality weak signals—roughly ten times the volume captured by all historical manual studies combined.
From Isolated Anomalies to Interconnected Belts
This massive dataset completely reframes how geologists visualize the Earth’s lower mantle. Past mapping efforts typically exposed isolated, random structural anomalies scattered across the core-mantle boundary. The new AI-processed data reveals that these distinct regions actually link together, forming vast, continuous belts.
Researchers hypothesize that these extensive “thermo-chemical accumulations” possess ancient origins. One leading geological theory suggests they represent remnants of oceanic crust dragged deep into the Earth by ancient tectonic subduction. An alternative hypothesis points toward fragments of Theia, the Mars-sized protoplanet theorized to have collided with the early Earth billions of years ago in an impact that helped form the Moon. Under extreme pressures and scorching deep-Earth temperatures, these foreign materials likely underwent partial melting and severe phase changes.
Unmapped Regions and Global Seismic Distribution
The resulting map highlights structural complexities previously obscured by the limits of manual data processing. Specifically, the updated data identifies six previously undocumented anomalous regions. These newly charted zones sit beneath high-latitude locations, including parts of Eurasia, Central Asia, and the South Atlantic Ocean.
By leveraging neural networks to automate the extraction of weak seismic wave precursors, this research bridges a decades-old observational gap in solid-earth geophysics. It transforms archival data into a high-resolution window into the planet’s deepest mechanics.
>