AI‑Driven Seismic Scan Reveals Six New Structures at Earth’s Core‑Mantle Boundary

By employing artificial intelligence on seismic recordings, scientists have uncovered six separate, previously unknown formations at the interface that separates Earth’s liquid outer core from its solid mantle, a zone situated about 2,900 km beneath the surface.

The finding, reported in the Journal of Geophysical Research: Solid Earth, results from an innovative method that feeds deep‑learning algorithms with thousands of earthquake records. Training the model to detect minute changes in seismic wave speed and direction enabled the researchers to chart heterogeneities that traditional techniques overlooked.

Since no instrument can directly access the core‑mantle boundary, researchers depend on earthquake‑generated seismic waves to deduce its characteristics. These waves are modified as they pass through variations in composition, temperature and phase within the Earth. The AI‑based analysis pinpointed six areas where the waveforms consistently diverged, indicating atypical features like localized compositional anomalies or thermal plumes.

These results contribute to an expanding literature that depicts the core‑mantle interface as a dynamic, heterogeneous region instead of a smooth, uniform layer. Earlier work has documented extensive low‑shear‑velocity provinces and ultra‑low velocity zones, yet the six structures described in this study are smaller and more isolated, suggesting a level of complexity that was previously unrecognized.

The consequences touch on a number of geophysical enigmas, such as the forces behind mantle convection, the origin of Earth’s magnetic field, and the planet’s long‑term thermal evolution. Boundary anomalies may alter the transfer of heat from the scorching core to the mantle above, possibly modulating the intensity of mantle plumes that emerge as volcanic hotspots.

The lead authors warn that, although the deep‑learning model yields a strong statistical signal, additional verification using independent datasets and different analytical methods is required. Upcoming efforts could incorporate data from dense seismic arrays—like those operating in the United States and Europe—to sharpen the spatial resolution of the detected structures.

The research highlights the expanding influence of machine learning in Earth sciences, providing a potent supplement to conventional seismology. As computational techniques grow more advanced, scientists expect to reveal further concealed features deep within the planet, enhancing our grasp of the processes that sculpt Earth’s interior.