Improving MAG4's Forecasting of Drivers of Severe Space Weather

The Sun produces large flares and coronal mass ejections (CMEs) which often produce solar proton events that endanger astronauts. MAG4 (magnetogram forecasting) is a near-real-time large-database forecasting technique for forecasting an active region's (AR's) next-day production rate of major flares and CMEs from an AR's free-energy proxy and the AR's short-term previous flare productivity. The free-energy proxy is measured from an HMI vector magnetogram of the AR. MAG4 presently uses a deprojected HMI AR vector magnetogram to estimate what the MDI AR line-of-sight magnetogram would look like, and then applies the AR's free-energy proxy measured from that to forecasting curves derived from MAG4's large database of MDI AR line-of-sight magnetograms and histories of major-flare production of 3,000 AR's. In the present work, we quantify the improvements in MAG4's major-flare forecasting performance that result from using HMI forecasting curves obtained from MAG4's comparably large database of HMI AR vector magnetograms instead of using MDI forecasting curves. We use a Monte Carlo division to divide the AR magnetograms into two halves: a control sample and an experimental sample. For each of 3,000 random divisions, we make a forecasting curve from the control sample and then use that curve to make a forecast for each AR magnetogram in the experimental sample. For each random division, these forecasts are then compared to the observed major-flare productivity of the experimental-sample ARs and the Heidke skill score is calculated (The Heidke score can range from negative infinity to 1: negative values are worse than chance, 0 is no better than chance, 1 is prefect prediction). From histograms of the differences in the Heidke scores from the two alternative MAG4 magnetogram databases, we find an increase in Heidke score of over 0.1 from using HMI forecasting curves instead of MDI forecasting curves, improving the accuracy by about 10%. MAG4 will soon be upgraded to use HMI forecasting curves for more accurate forecasts.

This work was funded by the NSF's Solar-Terrestrial Division and the UAH/ NASA MSFC NSF Research Experience for Undergraduates (REU) Program.