Probabilistic forecasting of solar flares from vector magnetogram data

Discriminant analysis is a statistical approach for assigning a measurement to one of several mutually exclusive groups. Presented here is an application of the approach to solar flare forecasting, adapted to provide the probability that a measurement belongs to either group, the groups in this case being solar active regions which produced a flare within 24 hours and those that remained flare quiet. The technique is demonstrated for a large database of vector magnetic field measurements obtained by the University of Hawai'i Imaging Vector Magnetograph. For a large combination of variables characterizing the photospheric magnetic field, the results are compared to a Bayesian approach for solar flare prediction, and to the method employed by the U.S. Space Environment Center (SEC). Although quantitative comparison is difficult as the present application provides active region (rather than whole‐Sun) forecasts, and the present database covers only part of one solar cycle, the performance of the method appears comparable to the other approaches.