A basin of wet sand. You press in one corner, and water starts moving. But how exactly? Instead of an even circle, it might spread in tongues, linger in some spots, and rush through others. Underground, the same thing happens: oil, water, or gas migrate through rocks not by the textbook, but following the chaos of fractures and heterogeneities.

Geophysicists can 'listen' to such movements using repeated seismic surveys. They send sound waves and compare the echoes, like doing an ultrasound of the planet months or years apart. Differences in sound speed reveal zones where the rock composition has changed. But turning these echoes into a flow map is no easy task.

The authors of the new study took a different route. They created a 'reality generator'—a simulator based on fractional diffusion in space and time. This mathematical framework allows independent control of memory (the fluid 'remembers' the past and moves slower) and long-range interactions (the flow feels the influence of distant regions). By adding elliptical anisotropy, a preferred direction along fractures, the generator produced thousands of scenarios, from sluggish spreading to rapid breakthroughs.

On this variety, they trained a convolutional neural network. Its task: look at the difference between old and new seismic echoes and immediately output a map of velocity changes. The network learned not to memorize individual images, but to capture the physics itself — the link between signal shape and subsurface restructuring.

This approach is part of a larger trend toward 'physics-informed' neural networks. They aren't just fed data; they're taught the laws of nature from birth. As a result, subsurface flow diagnostics become faster and more reliable, and crucially, they no longer fear the complexity of the real world. The echo from the sand basin can now be deciphered, even if the water doesn't follow the rules.