There’s already enough trouble on this planet that we don’t need new troubles coming from the sun. Unfortunately, we haven’t been able to destroy this merciless star yet, so we leave it to it. But NASA at least maybe soon can let us know when one of its deadly flames will plunge our terrestrial systems into disarray.
Understanding and predicting space weather is an important part of NASA’s work. There’s no air up there, so no one can hear you scream, “Wow, how’s this radiation!” We therefore rely on a set of satellites to detect and relay this important data to us.
One such measurement is that of the solar wind, “a constant stream of matter from the sun”. Even NASA couldn’t find anything nice to say about it! Normally, this current is absorbed or dissipated by our magnetosphere, but if there is a solar storm, it can be powerful enough to overwhelm local defenses.
When this happens, it can put electronic devices in jeopardy, as these charged particles can flip bits or disrupt volatile memory like RAM and solid-state memory. NASA concerned that even telegraph stations were not safe, exploding in the largest solar storm on record, The Carrington Incident of 1859.
While we cannot prevent these amazing events from happening, we can be better prepared for them if we know they will come. But usually by the time we know, they’re basically here. But how can we predict such infrequent and chaotic events?
View of NASA’s SOHO satellite overwhelmed during a 2003 solar storm. Image credits: NASA
A joint project between NASA, the US Geological Survey, and the Department of Energy at the Frontier Development Laboratory looked at this problem, and the answer is exactly what you’d expect: machine learning.
The team collected data on solar flares from multiple sun-tracking satellites, as well as from ground stations that monitor geomagnetic disruptions (called perturbations), like those affecting to technology. The deep learning model they designed identified patterns in such a way that the former led to the latter, and they called the resulting system DAGGER: D.eep leONErninggmagnetic field perturseesaw.
Yes, it’s a stretch. But it seems to work.
Using the geomagnetic storms that hit Earth in 2011 and 2015 as test data, the team found that DAGGER could quickly and accurately predict their impact globally. This combines the strengths of the previous methods while avoiding their disadvantages. As NASA said:
Previous prediction models have used AI to make local geomagnetic forecasts for specific locations on Earth. Other models that do not use AI have made untimely global predictions. DAGGER is the first to combine AI’s rapid analysis with real-world measurements from space and across the Earth to produce regularly updated predictions that are both fast and accurate for website worldwide.
It may take a while before you get a solar alert on your phone asking you to pull over or your car could stop working (which won’t really happen…it might happen) , but it can make a big difference when we know there’s a vulnerable infrastructure that could have a sudden shutdown. Better a few minutes warning than nothing!
You can read the paper describing the DAGGER model, which is open source, by the way, in this issue of the journal. Space weather.
