Scientists have identified six previously unknown structures deep within the Earth, situated at the boundary between the planet's viscous mantle and its liquid outer core. These formations, located approximately 1,800 miles (2,900 kilometers) beneath the surface, were detected using seismic waves from earthquakes. These waves subtly alter their path when interacting with the "deep-seated scatterers," making them nearly invisible except through their influence on seismic wave propagation.
The findings, published in the journal JGR Solid Earth, suggest these structures may be remnants of material from the Earth's outer layers that were drawn deep into the planet. Potential compositions include fragments of continental crust or even the remains of Theia, a protoplanet believed to have impacted Earth about 4.5 billion years ago, contributing to the moon's formation. Under the immense pressure and high temperatures at the mantle boundary, these materials could transform, partially melt, or undergo mineralogical changes, resulting in "thermochemical piles" distinct from the surrounding mantle.
The boundary between the mantle and outer core is a region of significant geological activity, characterized by a temperature difference of around 1,000°C (1,800°F) and dramatic changes in density. These conditions drive convection currents in the mantle, known as mantle plumes, which influence surface volcanic activity.
To detect these deep structures, researchers focused on a specific type of seismic wave called a PKP precursor. These weaker waves arrive shortly before the primary seismic waves from an earthquake and are scattered by heterogeneities at the mantle boundary. Due to their faint nature, manually identifying these precursors within vast seismic data sets is inefficient and subjective.
To overcome this challenge, the research team developed and trained an artificial intelligence (AI) model to automatically identify PKP precursors. After training the AI on human-identified waves, it was applied to over two million recordings from 5,000 earthquakes. The model successfully identified 174,929 high-quality PKP precursor signals, a number more than ten times that of all previous studies combined. This extensive dataset provided an unprecedented view of the mantle boundary, revealing large areas of previously undocumented structures. The AI identified six specific regions that likely host significant heterogeneities, marking them as priority targets for future exploration of Earth's deep interior.
Previous research had noted fragmented structures in isolated locations. However, this new analysis indicates these fragments are connected, forming larger, continuous belts. While the exact composition and formation mechanism of these structures remain subjects of ongoing research, the increasing analytical power of AI models applied to decades of collected seismic data promises to refine our understanding of Earth's deep interior geodynamics.