Predicting well-connected SEP events from observations of solar EUVs and energetic protons

Abstract

This study shows a quantitative assessment of the use of Extreme Ultraviolet (EUV) observations in the prediction of Solar Energetic Proton (SEP) events. The UMASEP scheme (Space Weather, 9, S07003, 2011; 13, 2015, 807–819) forecasts the occurrence and the intensity of the first hours of SEP events. In order to predict well-connected events, this scheme correlates Solar Soft X-rays (SXR) with differential proton fluxes of the GOES satellites. In this study, we explore the use of the EUV time history from GOES-EUVS and SDO-AIA instruments in the UMASEP scheme. This study presents the results of the prediction of the occurrence of well-connected >10 MeV SEP events, for the period from May 2010 to December 2017, in terms of Probability of Detection (POD), False Alarm Ratio (FAR), Critical Success Index (CSI), and the average and median of the warning times. The UMASEP/EUV-based models were calibrated using GOES and SDO data from May 2010 to October 2014, and validated using out-of-sample SDO data from November 2014 to December 2017. The best results were obtained by those models that used EUV data in the range 50–340 Å. We conclude that the UMASEP/EUV-based models yield similar or better POD results, and similar or worse FAR results, than those of the current real-time UMASEP/SXR-based model. The reason for the higher POD of the UMASEP/EUV-based models in the range 50–340 Å, was due to the high percentage of successful predictions of well-connected SEP events associated with <C4 flares and behind-the-limb flares, which amounted to 25% of all the well-connected events during the period May 2010 to December 2017. By using all the available data (2010–2017), this study also concluded that the simultaneous use of SXRs and EUVs in 94 Å in the UMASEP-10 tool for predicting all >10 MeV SEP events, improves the overall performance, obtaining a POD of 92.9% (39/42) compared with 81% (34/42) of the current tool, and a slightly worse FAR of 31.6% (18/57) compared with 29.2% (14/58) of the current tool.