PINNBARDS: A Physics-Informed Framework for Decoding Deep Solar Magnetic Structure and Forecasting Flare-Producing Active Region Emergence a Week in Advance
Dr. Mausumi Dikpati
NSF NCAR
1 July, 2026, 2pm
Abstract:
Most space-weather forecasts rely on diagnostics of active regions (ARs) after they emerge at the solar surface, limiting warning times to hours. Extending forecasts to days or weeks requires predicting where and when new, flare-producing ARs will emerge. Observations show that large and complex ARs are organized within slowly evolving, longitudinally warped toroidal activity bands rather than appearing randomly across the Sun. These global patterns likely originate from the dynamics of deep-seated tachocline toroidal magnetic fields. We present PINNBARDS (Physics-Informed Neural Network–Based Active Region Distribution Simulator), a framework that combines observed AR distributions with the governing equations of a global MHD shallow-water tachocline model to decode dynamically self-consistent subsurface magnetic and flow fields. These decoded state vectors are used to initialize a global MHD model, which is then integrated forward in time to forecast future AR emergence.
We forecast AR emergences and their eruption into multiple big X-flares for a few case studies in Solar Cycle 25, including (i) AR 13590 emergence and eruption into X-6.37 flare on 22 February 2024, (ii) ARs 13664/8 emergence and eruption into X-8.79 flare on 2024 Mother’s Day, (iii) AR 14274 emergence and eruption into X-5.1 on 11 November 2025, and (iv) AR 14366 emergence and eruption into X-8.11 flare on 2 February, 2026. All of these were associated with multiple flares and/or CMEs. For example, in the first case study, AR 13590 emerged during 18-20 February 2024 and subsequently produced multiple flares with the biggest one, an X-6.3 flare, on February 22. Simulations initialized by the PINNBARDS-decoded state-vectors one week earlier indicate the development of a strong localized bulging within a tipped-away segment of the northern toroidal band at the correct longitude and hemisphere. The evolving pressure-gradient structure promotes plasma drainage and enhanced magnetic buoyancy, creating favorable conditions for flux emergence. The persistence of the bulging and the pressure-gradient are consistent with the sustained magnetic-flux injection observed in AR 13590 prior to its eruption. Similarly, other case studies from Solar Cycle 25 suggest that major flare-producing ARs emerge preferentially within globally organized toroidal-band structures, which can be identified days to weeks in advance. These results demonstrate a promising pathway toward week-ahead forecasting of space-weather-producing active regions' emergence through model integration initialized by PINNBARDS-decoded data.