A new machine-learning tool gives space weather forecasters an advantage
Computers can learn to find solar flares and other events in vast streams of solar images and help NOAA forecasters issue timely alerts, according to a new study. The machine-learning technique, developed by scientists at NCEI and the Cooperative Institute for Research in Environmental Sciences (CIRES) at the University of Colorado (CU) Boulder, searches massive amounts of satellite data to pick out features that are significant for space weather. Changing conditions on the sun and in space can affect various technologies on Earth, blocking radio communications, damaging power grids, and diminishing navigation system accuracy.
“Being able to process solar data in real time is important because flares erupting on the sun impact Earth over the course of minutes. These techniques provide a rapid, continuously updated overview of solar features and can point us to areas requiring more scrutiny,” says Rob Steenburgh, a forecaster in the NOAA Space Weather Prediction Center in Boulder.
The research was published in the Journal of Space Weather and Space Climate.
Capturing Information From Steady Streams of Solar Data
To predict incoming space weather, forecasters summarize current conditions on the sun twice daily. Today, they use hand-drawn maps labeled with various solar features—including active regions, filaments, and coronal hole boundaries. But solar imagers produce a new set of observations every few minutes. For example, the Solar Ultraviolet Imager (SUVI) on NOAA’s GOES-R Series satellite runs on a 4-minute cycle, collecting data in six different wavelengths every cycle.
Just keeping up with all of that data could take up a lot of a forecaster’s time. “We need tools to process solar data into digestible chunks,” says Dan Seaton, a scientist with CIRES and NCEI, one of the paper’s co-authors.
Training Computers to Identify Solar Features
J. Marcus Hughes, an NCEI and CIRES scientist who is also a graduate student in computer science at CU Boulder and lead author of the study, created a computer algorithm that can look at all the SUVI images simultaneously and spot patterns in the data. With his colleagues, Hughes created a database of expert-labeled maps of the sun and used those images to teach a computer to identify solar features important for forecasting. “We didn’t tell it how to identify those features, but what to look for—things like flares, coronal holes, bright regions, filaments, and prominences. The computer learns the how through the algorithm,” Hughes says.
The algorithm identifies solar features using a decision-tree approach that follows a set of simple rules to distinguish between different traits. It examines an image one pixel at a time and decides, for example, whether that pixel is brighter or dimmer than a certain threshold before sending it down a branch of the tree. This repeats until, at the very bottom of the tree, each pixel fits only one category or feature—a flare, for example.
The algorithm learns hundreds of decision trees—and makes hundreds of decisions along each tree—to distinguish between different solar features and determine the “majority vote” for each pixel. Once the system is trained, it can classify millions of pixels in seconds, supporting forecasts that could be routine or require an alert or warning.
“This technique is really good at using all the data simultaneously,” Hughes says. “Because the algorithm learns so rapidly it can help forecasters understand what’s happening on the sun far more quickly than they currently do.”
Improving Solar Models and Space Weather Forecasts
The technique also sees patterns humans can’t. “It can sometimes find features we had difficulty identifying correctly ourselves. So machine learning can direct our scientific inquiry and identify important characteristics of features we didn’t know to look for,” Seaton says.
The algorithm’s skill at finding patterns is not only useful for short-term forecasting, but also for helping scientists evaluate long-term solar data and improve models of the sun. “Because the algorithm can look at 20 years’ worth of images and find patterns in the data, we’ll be able to answer questions and solve long-term problems that have been intractable,” Seaton says.
NCEI and SWPC are still testing the tool for tracking changing solar conditions so forecasters can issue more accurate watches, warnings, and alerts. The tool could be made officially operational as early as the end of 2019.