SPRINTS: A Framework for Solar‐Driven Event Forecasting and Research. Issue 10 (19th October 2017)

• Record Type: Journal Article
• Authors: Engell, A. J.; Falconer, D. A.; Schuh, M.; Loomis, J.; Bissett, D.
• Is Part Of: Space weather. Volume 15:Issue 10(2017:Oct.)
• Page Start: 1321
• Page End: 1346
• Publication Date: 2017-10-19
• Subjects: forecasting -- automatic detections -- machine learning -- solar energetic particles; Space environment -- Periodicals; 551.509992
• DOI: 10.1002/2017SW001660
• Languages: English
• ISSNs: 1542-7390

Abstract

Abstract: Capabilities to predict onset and time profiles of solar‐driven events, including solar X‐ray flares; solar energetic particles (SEP); coronal mass ejections; and high‐speed streams, are critical in mitigating their potential impacts. We introduce the Space Radiation Intelligence System (SPRINTS). This NASA‐invested technology integrates preeruptive metadata and forecasts from the MAG4 system with posteruptive metadata in order to produce high fidelity and preeruptive to posteruptive transitional forecasts for solar‐driven events. To catalog start and end times of the four solar‐driven events, SPRINTS is capable of generating posteruptive forecasts based on automatic detections employed on 30+ years of GOES X‐ray and particle data as well as 20+ years of ACE and DSCOVR solar wind data. The prediction results for 1986–2016 presented here are from the SPRINTS posteruptive capability for forecasting SEPs leveraging GOES X‐ray metadata. We present onset, peak flux, and time profile SEP forecast metrics and results based on expert‐guided, statistical, and machine‐learned decision tree models. Operating on data from a 20 year period, a machine‐learned decision tree model provided the best results for predicting an S1 event (on the NOAA Space Weather Prediction Center Solar Radiation Scale): 86% probability of detection and 37% false alarm rate. Five flare‐related metadata sets were leveraged in the decision tree modeling. Consistently, flare‐integrated flux, flare heliolongitude, and flare decay‐phase duration were found to be the top three forecasting parameters, while flare magnitude and flare latitude had little to no impact on the SEP forecast model. For the solar‐driven events of March 2012, we demonstrate SPRINTS abilities to forecast solar flares, SEP onset, and SEP evolution. Key Points: New operational framework for forecasting solar‐driven events; designed for research to operations and operations to research Combines preeruptive and posteruptive forecasts; exploits automatic detections, data science applications, and machine learning in models 86% probability of detection and 37% false alarm rate for NOAA SWPC S1 events and demonstrates ability to predict SEP event evolution