Physics-guided AI reconstructs Earth's hidden mantle history from limited clues

Deep beneath your feet, rock is moving. Not quickly—only a few centimeters each year, the pace of a fingernail's growth. Yet that slow circulation of Earth's mantle, a rocky layer that makes up more than 80 percent of the planet's volume, drives the plate tectonics that shape continents, feeds volcanoes, and triggers earthquakes. The problem is that scientists cannot simply watch this process unfold. The mantle's history remains largely hidden, accessible only through indirect clues: surface geology, seismic observations, and the geological record left behind.

A team at the University of Tsukuba set out to test whether artificial intelligence could help recover that missing history. They developed what's called a physics-informed neural network—an AI model trained not just to fit data, but to obey the physical laws that govern how heat moves through rock and how material flows in the mantle. This distinction matters. A standard AI model might find a pattern in the available observations but produce a reconstruction that violates the actual physics of mantle convection. A physics-informed model faces a stricter test: its answer must behave the way the mantle actually behaves.

To validate the approach, the researchers created a controlled experiment. They built computer simulations of two-dimensional mantle convection—a simplified version of the real thing, but one where they knew the complete answer. They then hid parts of that history and gave the AI model only limited clues: observations of near-surface mantle motion and a snapshot of present-day temperature. The model received no direct information about past temperatures or deep-mantle flow through time. Despite these constraints, it successfully reconstructed the hidden temperature and flow history with high accuracy. The finding suggests that sparse observations can yield realistic results when combined with the right physical constraints.

The success hinged on using multiple types of evidence together. Near-surface motion provided clues about how material moved. The present-day temperature snapshot revealed something about the mantle's current structure. Separately, each piece of information left gaps. Combined, they gave the model enough anchors to recover a plausible history. The physical equations governing heat transport and fluid flow tied those pieces into one continuous reconstruction. This complementary approach proved essential—no single observation type was sufficient on its own.

Why is reconstructing mantle flow so difficult? The process unfolds slowly and far underground, leaving only indirect traces. It resembles trying to reconstruct a long movie from a few scattered scenes. Surface movements provide part of the plot. Today's deep structure provides the ending. The missing middle must be inferred carefully, and without physical constraints to guide the inference, any number of unrealistic histories might fit the available data.

The study is careful to mark its own limits. The researchers tested their method on a simplified two-dimensional system, a choice that made the problem manageable and allowed them to verify the reconstruction against a known reference. The real Earth is vastly more complicated. Its mantle is three-dimensional. Rock behaves differently under different temperatures and pressures. Plate boundaries, chemical variations, and other geological processes add layers of complexity. The study does not claim to have reconstructed Earth's actual mantle history. Instead, it demonstrates that the method can work in a controlled setting where the true answer is known.

That distinction is important. A successful simulation is a first step, not final proof. Applying the method to real geophysical data would require further development and careful handling of uncertainty. Yet the result is promising because it addresses a central problem in Earth science: scientists want to understand how the deep interior evolved, but the available observations are incomplete. A method that combines limited data with physical laws could help close part of that gap.

The implications reach beyond pure science. Understanding the history of mantle flow could help scientists interpret why Earth's surface changed in particular ways and connect present-day mantle structures to past geological events. With further development, physics-informed AI might help combine geological records, seismic images, and other geophysical observations into clearer reconstructions of mantle circulation. That could improve understanding of how deep-Earth processes link to plate tectonics, earthquakes, and volcanic activity. For now, the work remains at an early stage, tested only with synthetic data from simplified simulations. But it points toward a way to recover Earth's hidden history more robustly—not by relying on a single observation type, but by letting AI work with physics, not around it.