AI discovery of Earth core-mantle boundary structures

Researchers have utilized deep learning artificial intelligence to identify six previously unknown large-scale structures at the Earth’s core-mantle boundary, located approximately 2,900 kilometers below the surface.

By analyzing over two million seismic records from 1990 to 2024, the AI detected approximately 175,000 ‘PKP precursor waves’. These signals are extremely weak and arrive seconds before primary PKIKP waves, making them difficult to identify using traditional analytical methods. This represents a tenfold increase in data processed compared to previous global studies. The data revealed that these irregularities form continuous belts or bands beneath the North Atlantic, Northern Eurasia, South Atlantic, Southern Africa, the Pacific Ocean, and Antarctica.

Scientists suggest these structures may consist of material with varying densities or temperatures, potentially originating from ancient tectonic plates or chemically distinct material from Earth’s early formation. These findings are expected to improve understanding of mantle circulation, heat flow, and geological processes such as volcanic activity, earthquakes, and tsunamis.