Evaluating the Performance of Solar Flare Forecasting Methods

Abstract

The number of published approaches to solar flare forecasting using photospheric magnetic field observations has proliferated recently, with widely varying claims about how well each works. As different analysis techniques and data sets were used, it is essentially impossible to directly compare the results. A systematic comparison is presented here using three parameters based on the published literature that characterize the photospheric magnetic field itself, plus one that characterizes the coronal magnetic topology. Forecasts based on the statistical method of discriminant analysis are made for each of these parameters, and their ability to predict major flares is quantified using skill scores. Despite widely varying statements regarding their forecasting utility in the original studies describing these four parameters, there is no clear distinction in their performance here, thus demonstrating the importance of using standard verification statistics.