AI Model Reconstructs Hidden History of Earth's Mantle Flow From Surface Clues

A researcher at the University of Tsukuba has developed an AI model that can reconstruct the hidden history of Earth's mantle circulation using only near-surface observations and a present-day snapshot of mantle temperature. The findings were published in the Journal of Geophysical Research: Machine Learning and Computation.

The mantle, a rocky layer that makes up more than 80% of Earth's volume, circulates at just a few centimeters a year, driving plate tectonics and influencing earthquakes and volcanic activity — yet its history remains poorly understood because direct observation of the deep Earth is extremely limited. Scientists currently rely on geological records of past surface movement and geophysical imaging, such as seismic observations, to infer the mantle's structure and behavior, but reconstructing its historical flow has remained a major challenge.

The new model is a physics-informed neural network, trained not only to fit observational data but also to satisfy the physical equations that govern heat transport and fluid flow in the mantle. To test it, the researcher generated computer simulations of two-dimensional mantle thermal convection as a reference solution, then gave the model only synthetic near-surface motion data and a present-day mantle temperature snapshot — withholding information about past temperatures and deep-mantle flow.

The model successfully reconstructed the withheld historical features with high accuracy, indicating that combining different types of geophysical information is key to recovering realistic histories of mantle convection. With further development and testing on real geophysical data, the approach could become a tool for revealing how Earth's deep interior has evolved over time, according to the study.