From disrupted GPS signals to risks for satellites and astronauts, severe space weather can have far-reaching consequences. But researchers from the US National Science Foundation National Center for Atmospheric Research (NSF NCAR) and Southwest Research Institute (SwRI), have developed a tool with huge potential to mitigate these impacts. Mausumi Dikpati, the senior scientist who led the team, discusses how PINNBARDS (PINN-Based Active Regions Distribution Simulator), a newly developed physics-informed, AI-enabled modeling framework, could transform forecasting by predicting flare-producing solar active regions weeks before they emerge
Your new tool, PINNBARDS, is described as enabling the first steps toward space-weather forecasting weeks in advance rather than hours. What key scientific or technological breakthrough made this longer forecasting horizon possible?
Major space-weather events are driven by large, complex solar active regions. Once these regions appear at the sun’s surface, eruptions can follow within hours, leaving little time to prepare. The breakthrough behind PINNBARDS is that it allows us to look before these regions appear. Large active regions take weeks to rise from beneath the sun’s surface, and PINNBARDS reconstructs the hidden magnetic conditions that can be fed to a physical model, which controls where and when they will emerge. By translating the observed surface magnetic patterns into physically consistent internal magnetic states, the physical model can be run forward in time, providing weeks of advance warning instead of hours. That shift, from reacting after emergence to anticipating it before it happens, is what enables the longer forecasting horizon.
How does integrating AI with physics-based models improve predictive reliability compared with either approach used on its own?
AI alone is powerful but ultimately statistical – it can recognize patterns but does not obey the laws of physics. Physics-based models, on the other hand, are reliable only if they are started from accurate initial conditions, which have long been missing. PINNBARDS combines the two: AI extracts hidden physical information from observations, and physics ensures that the reconstruction of magnetic state-vectors by PINNBARDS is physically consistent, and hence can forecast realistically. This hybrid approach provides accurate initialization and accurate predictions, making forecasts more reliable than either AI or physics models used on their own.
The system reconstructs subsurface magnetic conditions beneath the sun’s surface. What observational data feeds into the model, and how confident are you that these reconstructions accurately represent the sun’s internal magnetic dynamics?
PINNBARDS uses solar surface magnetograms – synoptic maps of the sun’s magnetic field. Large-scale, organized magnetic patterns seen at the surface must originate less-turbulent, deeper inside the sun, where magnetic structures are more stable. Physics-based models have already shown that such subsurface magnetic structures can explain major historical storms, like the ‘Halloween storms’ in 2003, as well as recent ‘Mother’s Day storms.’ What PINNBARDS brings here as a breakthrough is the ability to run these models in a data-driven way, anchored directly to observations. My recent case studies (in preparation for the next publication) show that the reconstructed subsurface magnetic states successfully predict the emergence of flare-prone active regions, giving us strong confidence that the method captures the sun’s internal magnetic dynamics realistically.
If forecasting active regions weeks ahead becomes operational, what would be the most immediate benefits for industries and agencies that rely on space-weather alerts, such as satellite operators, power grid managers or human spaceflight programs?
Severe space-weather events can disrupt satellites, GPS, communications, power grids and human spaceflight. With only hours of warning, operators have limited options. Weeks-ahead forecasts would allow agencies and industries to plan protective actions well in advance, such as adjusting satellite operations, delaying launches or preparing power-grid safeguards. For example, the May 2024 Mother’s Day solar storms disrupted GPS-dependent precision agriculture. With weeks of advance notice, many of these impacts could be reduced or avoided altogether.
What are the next research or technical steps required to move PINNBARDS from a research demonstration to a tool that could eventually support operational space-weather forecasting?
The next step is to transition from case studies to routine forecasting. This includes producing daily, updated forecasts that identify when and where large, flare-prone active regions are likely to emerge weeks in advance, as well as identifying quieter periods when space-weather risk is low. Both types of information are valuable: early warnings help mitigate hazards, while forecasts of quiet intervals help operators safely plan launches and infrastructure activities. Continued validation, automation and collaboration with operational forecasting centers will be key to making PINNBARDS an operational tool.
