AI technology maps new structures near Earth's core-mantle boundary

Researchers at the Chinese Academy of Sciences have utilized a new deep learning tool to map the Earth’s core-mantle boundary with unprecedented detail. By analyzing over 2 million seismic waveforms from approximately 5,000 earthquakes recorded between 1990 and 2024, the AI identified nearly 175,000 previously unrecorded seismic signals.

The study focused on PKP precursors, which are faint seismic waves that appear just before stronger waves as they interact with irregularities near the boundary where the solid mantle meets the liquid outer core. While previous maps depicted anomalies as scattered patches, the AI-driven analysis reveals that many of these features are actually linked, forming broad continuous bands.

These structures may consist of oceanic crust and sediment that sank millions of years ago, or debris from the ancient collision that formed the Moon. The findings highlight six previously under-sampled regions—including parts of high-latitude Eurasia, Central Asia, and the South Atlantic—as high-priority areas for future geophysical research.