Predicting the strength of geomagnetic storms that can harm satellites and power grids is vital in a technological dependant society. Right now, these predictions rely on data from monitoring satellites located between the Sun and the Earth. However, there's a built-in problem: we can only measure the dangerous solar material after it passes these satellites. This means we cannot reliably predict the start of a storm before that material arrives at the satellites, severely limiting how far ahead we can warn about the most damaging phase.

We tested how far ahead we can actually predict storm strength using data from these satellites. We studied two major storms from 2024 using a neural network. We found we can only reliably predict the start and peak intensity of a sudden storm about 3 hours before it hits Earth. Trying to predict this damaging peak further ahead (4-6 hours) is inherently limited because the storm hasn't even reached the monitoring satellites yet. However, predictions work better (up to 6 hours ahead) for the storm's recovery phase (when things are calming down) and for slower types of space weather, as the needed information is already available.