# Relationships Between Characteristics of the Line-of-Sight Magnetic Field and Solar Flare Forecasts

## TL;DR

This study demonstrates that line-of-sight magnetic field characteristics, especially those of the Polarity Inversion Line, can effectively forecast solar flares with high accuracy, outperforming current expert predictions and highlighting the importance of PIL features.

## Contribution

The paper introduces a PIL-based feature extraction method and shows its effectiveness in flare prediction using only LOS magnetograms, achieving high TSS scores.

## Key findings

- PIL characteristics outperform global AR features in flare forecasting.
- Support Vector Machine with Sigmoid kernel yields highest TSS scores.
- LOS magnetogram-based predictions are nearly as effective as vector magnetogram methods.

## Abstract

We analyze the relationship between the flare X-ray peak flux, and characteristics of the Polarity Inversion Line (PIL) and Active Regions (AR), derived from line-of-sight (LOS) magnetograms. The PIL detection algorithm based on a magnetogram segmentation procedure is applied for each AR with 1 hour cadence. The PIL and AR characteristics are associated with the AR flare history and divided into flaring and non-flaring cases. Effectiveness of the derived characteristics for flare forecasting is determined by the number of non-flaring cases separated from flaring cases by a certain threshold, and by their Fisher ranking score. The Support Vector Machine (SVM) classifier trained only on the PIL characteristics is used for the flare prediction. We have obtained the following results: (1) the PIL characteristics are more effective than global characteristics of ARs, (2) the highest True Skill Statistics (TSS) values of 0.76$\pm$0.03 for $\geq$M1.0 flares and 0.84$\pm$0.07 for $\geq$X1.0 flares are obtained using the "Sigmoid" SVM kernel, (3) the TSS scores obtained using only the LOS magnetograms are slightly lower than the scores obtained using vector magnetograms, but significantly better than current expert-based predictions, (4) for prediction of $\geq$M1.0 class flares 74.4 % of all cases, and 91.2 % for $\geq$X1.0 class, can be pre-classified as negative with no significant effect on the results, (5) the inclusion of global AR characteristics does not improve the forecast. The study confirms the unique role of the PIL region characteristics in the flare initiation process, and demonstrate possibilities of flare forecasting using only the line-of-sight magnetograms.