BEARALERTS: A successful flare prediction system

We describe our BEARALERT program of predicting solar flares or rapid development of activity in certain sunspot groups. The purpose of the program is to test our understanding of the flare process by making public predictions via electronic mail. Neither the exact timing of the flare nor the possibility of emergence of new active regions can be predicted. But high-resolution observations of the magnetic configuration, Ha brightness and structure and other properties of a region enabled us to announce the onset of 15 of 23 major active regions over a two-year period, and 15 of 32 BEARALERTS were followed by this activity. We used high-resolution real-time data available at the Big Bear Solar Observatory (BBSO). The criteria for prediction are given and discussed, along with those for filament eruption. The success fo the BEARALERT is evaluated by counting the M- and X-class flares in six days following the alert and comparing these results with those of a number of other predictive schemes. We find the single regions chosen had about 30% more flares than the whole disk on random days, or several times more than individual regions chosen at random. There was a gain of 1.5 to 2.0 times in flare frequency compared to regions selected by spot size or complexity. We also find an improvement of 20–40% over large or complex regions that have had some flares already. The ratio of improvement has increased with time as we gained experience. In the 24-hr period following each alert, one or more M-class or greater flares occurred 72% of the time. We also checked the possibility of prediction by the 152-day interval which some workers have claimed, but found those results slightly worse than random and considerably inferior to the BEARALERTS. All of the particularly active regions that were missed either occurred during bad weather at BBSO or were missed because we only issued alerts for one region at a time