More Diagnosis of Solar Flare Probability from Chromosphere Image Sequences

Abstract : This report describes a continuation of the work reported in AFRL-RV-PS-TP-2012-0005. First, we nearly doubled the original set of 46 ISOON H image sequences to 90. Second, the flare probability diagnosis algorithm was modified from the last version in the previous report to use the 1-minute image time change of the leading eigenvectors as predictors, and flaring level indicators prescribed from the whole sequence eigenvector patterns and the x-ray flux rise as predictors. Third, a new approach was investigated that attempted a single flare probability diagnosis for the whole image sequence. Frequency of occurrence of 1-minute eigenvector changes for an entire sequence over a designated number of size bins served as the predictor vector elements for each sequence, and the sequence-prescribed flaring level indicator was the predictand. Fourth, the latter method was limited to using occurrence frequency over just the pre-flare image times. The three new methods and the legacy algorithm were evaluated using three development sets of 60 sequences to train the candidate algorithm, and the applying resulting discriminant vectors to the three application sets of 30 independent sequences. The first two methods, which diagnosed flare category probability at individual image times, performed the best. Lastly, we investigated augmentation of H eigenvectors with ISOON Doppler velocity eigenvectors to see if flare category discrimination could be improved.