Deep within the Earth, machine learning has uncovered nearly 175 tisíc faint seismic signals that reveal a surprisingly chaotic boundary between the solid mantle and the liquid outer core, according to a study published in the Journal of Geophysical Research: Solid Earth and Science Advances.
Machine Learning Uncovers Massive Seismic Dataset
Researchers analyzed over two million seismic waveforms from earthquakes with a magnitude of at least 6.0 recorded between 1990 and 2024, according to the studies. Traditional methods relied on manual or semi-automated searching, which made global mapping painfully slow. By combining deep learning with repeated manual checks, the team identified 174 929 PKP precursors—roughly ten times more than all previous global datasets combined.
Did you know? PKP waves are pressure seismic waves that travel through the mantle, pass into the outer core, and head back up through the mantle toward a measuring station. The faint precursors emerge when energy scatters off tiny structures near the core-mantle boundary.
Mapping Ultra-Low Velocity Zones Near the Core
Schoolbook diagrams neatly divide the Earth into the crust, mantle, outer core, and inner core, but the actual boundaries are far more complex, according to geophysical findings. Near the base of the mantle, scientists study ultra-low velocity zones (ULVZ) where seismic waves slow down dramatically. These regions may relate to extreme temperatures, partial melting, distinct chemistry, or remnants of ancient oceanic crust pushed downward over hundreds of millions of years.
Instead of isolated spots, the expanded dataset shows that many scattering areas form continuous bands across the deep interior, according to the research. The team mapped six previously undocumented areas with a high probability of significant heterogeneities.
Interpreting Deep Earth Structures
Researchers caution that the algorithm did not discover a hidden continent or definitively identify the physical composition of these structures, according to the study text. The breakthrough lies in data volume and spatial coverage, which highlights where seismic waves scatter strongly. Determining their exact nature requires further investigation.
Several processes likely contribute to these anomalies. Subducted oceanic plates can sink deep into the mantle, creating what models describe as a geological graveyard of ancient surfaces. At the same time, extreme pressure and rising core temperatures can lower the melting point of rocks, forming partially molten pockets that slow down seismic waves.
Frequently Asked Questions
What are PKP precursors?
They are faint seismic signals that arrive before the main PKP wave, created when energy scatters off small structures deep inside the Earth near the core-mantle boundary.
Did AI discover a new continent under the Earth?
No. The machine learning system processed decades of seismic data to find weak signals much faster than humans could, revealing new statistical patterns rather than physical landmasses.
How do scientists study the Earth’s core without drilling?
Because direct drilling is impossible, researchers rely on natural earthquakes, using the planet itself as a giant tomography machine to track how seismic waves travel through different layers.
Pro Tip for Researchers
Combining automated deep learning classifiers with rigorous manual verification remains the most effective way to extract faint signals from decades of global seismic archives.
Stay Informed on Geophysical Discoveries
Want to track more breakthroughs in Earth sciences and machine learning? Subscribe to our newsletter for regular updates on how researchers peer into the deepest layers of our planet.