Upgrades of the ESPERTA forecast tool for solar proton events

Abstract

The Empirical model for Solar Proton Events Real Time Alert (ESPERTA) exploits three solar parameters (flare longitude, soft X-ray fluence, and radio fluence) to provide a timely prediction for the occurrence of solar proton events (SPEs, i.e., when the $>$10MeV proton flux is $\geq$10 pfu) after the emission of a $\geq$ M2 flare. In addition, it makes a prediction for the more geoeffective SPEs for which the $>$10 MeV proton flux is $\geq$ 100 pfu. In this paper, we study two different ways to upgrade the ESPERTA model and implement it in real time: 1) by using ground based observations from the LOFAR stations; 2) by applying a novel machine learning algorithm to flare-based parameters to provide early warnings of SPE occurrence together with a fine-tuned radiation storm level. As a last step, we perform a preliminary study using a neural network to forecast the proton flux profile to complement the ESPERTA tool. We evaluate the models over flare and SPE data the last two solar cycles and discuss the performance and the limits and advantages of the three methods.