Earth's magnetic field is like an invisible river that meanders beneath our feet. Navigators and geologists rely on maps of this river, but instruments only take readings at scattered points, leaving blank spots in between. Filling those gaps is like trying to sketch the river's course from a few pebbles found on the bank.
Traditional interpolation methods often get it wrong: they mistake random sensor crackle for a real bend in the field. The new approach works differently. It first learns from examples what the river looks like without interference, then, looking at sparse points, reconstructs a smooth flow. It relies on a principle familiar to geologists: neighboring parts of the field are connected, and this connection helps distinguish noise from true signal.
This technique isn't just for magnetic field maps. Any measurements where data is collected point-by-point with interference, from soil sensors to body-worn health monitors, can be reconstructed the same way: first learn the physics of the process, then fill in the missing pieces.
Earth's magnetic field is weaker than the field of a fridge magnet, yet even bacteria can sense it to navigate through water.