Authors: Griffin Goodwin (Georgia State University), Viacheslav Sadykov (Georgia State University), Petrus Martens (Georgia State University)

Active region (AR) vector magnetograms are frequently used to forecast solar flares as they provide an excellent source of preflare features that can be easily integrated into machine learning models. However, as ARs traverse farther from the solar disk center, the quality of magnetic field data degrades substantially. Near the limbs, magnetograms suffer from projection effects, which introduce observed (but not intrinsic) long-term trends within the data. These artificial variations complicate forecasting and may even introduce unnecessary false positive and negative predictions. To study the impacts of this issue on forecasting, we utilize a methodology proposed by Falconer et al. (2016) to correct for projection effects present in Georgia State University’s Space Weather Analytics for Solar Flares (SWAN-SF) benchmark data set. We then compare performance metrics across the corrected and uncorrected data. Our analysis shows that, though there are improvements to the false positive rates and Heidke skill scores after corrections are applied, the enhancements are minimal, averaging only a few percent. This suggests that projection effects do not have a large impact on forecasting, or a more complicated correction methodology may be needed to see improvements.