We investigated the physical conditions present in solar active regions that lead to solar eruptions. Machine Learning methods require large databases. Quantitative data about active region development were already available from 2010 onward based on observations from the Helioseismic and Magnetic Imager (HMI) in the Space-weather HMI Active Region Patches (SHARPs) data series. We extended that record back in time to 1996 using measurements from its predecessor instrument, the Michelson Doppler Imager (MDI). We published the Space-weather MDI Active Region Patch (SMARP) catalog, making it available to the community, and we used it to investigate sympathetic flares (flares that occur in association with others).

The earlier MDI instrument did not provide measurements of the vector magnetic field, which is available for the SHARPs after 2010. Quantities based on the vector magnetic field are highly useful for predicting solar eruptions. To mitigate this limitation, we then trained a ML model to first synthesize coronal extreme ultraviolet (EUV) images and photospheric magnetograms by rendering results of a non-linear force-free field (NLFFF) model computed using the vector magnetic field measurements. That allowed us to train another ML model to solve the difficult inverse problem of deducing the vector coronal magnetic field from the synthesized extreme ultraviolet (EUV) images. We then began the more challenging step of using actual EUV images from the Atmospheric Imaging Assembly (AIA) along with HMI magnetograms to derive information about the vector magnetic field in the corona. The goal was to learn how to recover additional information during the MDI era from the available line-of-sight magnetograms and EUV images. While significant progress was made, COVID-related delays and personnel changes prevented us from completing this task.

The third outcome was the development of a greatly enhanced dataset (SHARP-E) including all significant active regions observed since May 2010. This comprehensive database complements the SHARPs by adding many additional physical parameters computed using the electric field determined from time series of HMI magnetic field and velocity observations. The electric field was derived as part of the Coronal Global Evolutionary Model (CGEM) project by considering temporal and spatial variations of both the vector magnetic and velocity fields measured in the photosphere. More than 232 quantities are computed at each 12-minute time step for each region. We then completed a preliminary analysis of the usefulness of the quantities for flare forecasting, comparing them with results based on the magnetic-only parameters typically used for prediction. Ranking their effectiveness based on the True Skill Score (considering only one quantity at a time), we found that eight of the top ten ranked parameters were from the newly derived parameters (see figures for a list of the top ten and their relative TSSs). The computed parameters have recently been made publicly available at the Stanford Data Repository (https://purl.stanford.edu/jz065wt1927) and will be presented at meetings and in publications in the near future.

The SMARP and SHARP-E data sets greatly increase the time range and capability of active region catalogs useful for machine learning applications for prediction of solar eruptions. Interpretation of these data opens the door to improved understanding of the conditions leading up to important space weather events.

Last Modified: 10/26/2024
Modified by: Jon T Hoeksema