Initial research by the Tokai National Higher Education and Research System is making potentially harmful solar events easier to predict accurately

With all the effects of climate change and global warming on the weather, people hardly ever wonder about weather conditions in outer space.

However, at the 2024 Annual Spring Meeting of the Astronomical Society of Japan held on 11 March 11, 2024, the Tokai National Higher Education and Research System (THERS) announced how AI-based research on space weather is vital to our understanding of the hazards affecting Earth and space-travel projects.

Routinely, space weather effects cause failures or glitches in communications equipment; disturbances in satellite positioning, and can complicate air route planning. Research on space weather observation represents an increasingly important challenge both in Japan and globally.

Factors affecting space weather range from solar flares, solar eruptive events such as coronal mass ejections, and the resultant solar energetic particles (SEP) that radiate throughout space. Direct exposure to SEP can lead to lethal effects on humans stationed in orbit and other scientific stations.

How AI can help

Working with a database of past SEP events and soft X-ray intensity, longitude, and flare duration time provided by the National Oceanic and Atmospheric Administration’s (NOAA) Space Weather Prediction Center, as well as sunspot-linked magnetic field data provided by the National Aeronautics and Space Administration (NASA) in the US, THERS had set out to find actionable trends that could be used to forecast future space weather patterns.

Dealing with the voluminous amounts of complex data required the use of a supercomputer, as well as an AI platform to analyze and profile the conditions leading to SEP events. The AI platform employs a machine learning technology named “wide learning explainable AI”. Also called “wide-and-deep learning”, this machine learning model combines the strengths of both linear models and deep learning models in AI to harness the strengths of each model.

As a result of their approach, THERS has made some important initial discoveries:

  • The soft X-ray intensity and duration time of flares represent an important factor in the generation of SEP events.
  • The first flare generated in sunspot regions with a low number of flares is likely to cause an increase in SEP. As flares tend to occur continuously at certain sunspots, common methods to predict flare occurrence have been based on historical data of the last flare. However, with the latest wide learning approach, THERS suggest that the prediction of the first flare, which occurs in the sunspot region where the flare activity of the previous day was relatively weak, represents an important factor in the prediction of SEP events, providing a new guideline for the future research and development of space weather forecasts.
  • The numerical model constructed in the research can enable prediction of the increase of SEP with the same accuracy as that of conventional prediction methods, with room for further accuracy in future, three-dimensional magnetic field model data around sunspots.

THER’s technology partner for the research is Fujitsu, which supplied the AI platform as well as the supercomputer. The plan is to continue to analyze the prediction of first solar flares as an important factor in SEP events, accelerate research towards new findings related to SEP events, and enhance space-weather forecasting for the benefit of all.