Using Machine Learning Techniques to Forecast Solar Energetic Particles

Abstract:

Solar energetic particles (SEPs) endanger satellites and astronauts in orbit, and can disrupt air traffic and spaceflight communication. The ability to forecast these events in advance is vital, both, economically, and for the safety of air and space faring passengers. The method of particle acceleration and transport is still an area of active research and physics-based models of such processes are currently unable to provide timely information which can be leveraged for issuing meaningful warnings. Thus, forecasters at NOAA’s Space Weather Prediction Center (SWPC) make use of a statistical model (PROTONS) to inform real-time decisions. We have applied modern machine learning techniques to the observational dataset upon which PROTONS was built, in an effort to improve the model performance. We will present results from the application of logistic regression, boosted decision trees and support vector machines to make a binary classification, i.e., whether or not there will be an SEP event, based on the physical parameters associated with solar flares. Results will be compared to those from the Space Weather Prediction Center’s existing model.