UMASEP

UMASEP is a set of event prediction tools. Each tool predicts SEP events of a specific energy range by using two models: a well-connected and a poorly-connected prediction model. These tools are also able to predict "all-clear" situations. Each well-connected prediction model makes use of the lag-correlation of solar electromagnetic flux with the particle flux at near-earth. If the correlation is high, the model infers that there is a magnetic connection through which particles are arriving. If, additionally, the intensity of the flux of the associated solar event is also high, then the well-connected prediction model issues an SEP event prediction. Each poorly-connected prediction model checks whether the differential proton flux behavior is similar to that in the beginning phases of previous historically events and thus deduce similar consequences. When a UMASEP tool predicts an SEP event, it also launches machine-learning-based predictions of the SEP integral proton intensity and fluence for a specific energy bin.

The UMASEP prediction tools are:

• UMASEP-10 (Núñez, 2011; Núñez, 2022), for predicting >10 MeV SEP proton events and the SEP fluence for the energy bin 15.1 - 21.9 MeV;
• UMASEP-30 for predicting >30 MeV SEP proton events and the SEP fluence for the energy bin 31.6 - 45.7 MeV;
• UMASEP-50 for predicting >50 MeV SEP proton events and the SEP fluence for the energy bin 45.7 - 66.1 MeV;
• UMASEP-100 (Núñez, 2015) for predicting >100 MeV SEP proton events and the SEP fluence for the energy bin 95.6 - 138.3 MeV;
• HESPERIA UMASEP-500 (Núñez et al, 2017) for predicting GLE and >500 MeV proton events.

The UMASEP-10 tool has a new model that uses machine learning techniques to predict the occurrence of >10 MeV SEP events from solar data only (without analyzing proton data). Currently, this component processes flare and radio burst data from the NOAA/SWPC Event Report to make predictions of the occurrence and the peak intensity of >10 MeV SEP events. This new model works together with the aforementioned well- and poorly-connected models.

GOES soft X-ray flux, GOES differential and integral proton flux. In the case of UMASEP-10, the model also processes Solar flare and radio burst data from the NOAA/SWPC Event Report.

The temporal cadence of the analyzed input data is 5 minutes in the case of UMASEP-10 and UMASEP-100, and 1 minute in the case of HESPERIA UMASEP-500.

The five tools predict SEP/GLE event occurrences. These tools also predict the peak SEP flux. The forecast window of these predictions depends on the type of predicted situation:

In the case the tool recognizes precursors of a well-connected SEP event, the prediction window is: 7 hours for >10 MeV protons (UMASEP-10), 6 hours for >30 MeV protons (UMASEP-30), 5 hours for >50 MeV protons (UMASEP-50), 3 hours for >100 MeV protons (UMASEP-100), 1 hour for GLE/>500 MeV events (HESPERIA UMASEP-500), and 7 hours for the 15-138 MeV SEP fluence spectrum

In the case the expected SEP event is not well-connected, the prediction window is: Up to 80 hours for >10 MeV protons (UMASEP-10), 24 hours for >30 MeV protons (UMASEP-30), 20 hours for >50 MeV protons (UMASEP-50), 16 hours for >100 MeV protons (UMASEP-100)

Version 3.0 was deployed on February 22, 2022 at the CCMC.

Version 3.0.4 was deployed on April 14, 2022 at the CCMC.

Version 3.1 was deployed on June 29, 2022 at the CCMC. List of changes include the new Machine-Learning-based peak intensity prediction capability of the UMASEP tools (including UMASEP-500).

Version 3.2.1 was deployed on September 19, 2022 at the CCMC. UMASEP 3.2.1 includes the following updates:

• Adds the fluences chunk to the JSON output files of UMASEP-10, UMASEP-30, UMASEP-50 and UMASEP-100.
• Includes knowledge files with better forecasting performance in terms of the 'Mean Absolute Error' of the integral proton intensity and fluence spectrum predictions.
• Includes an improved version of the fluence spectrum prediction module.

Version 3.2.2 was deployed on December 2, 2022 at the CCMC. UMASEP 3.2.2 includes an update to the poorly-connected prediction model of the UMASEP-100 tool, which yields a better forecasting performance in terms of the "False Alarm Ratio" using data of SC23-SC25. This version also includes minor updates to UMASEP-10, UMASEP-30 and UMASEP-50.