VAMPIRE: Using a Random Forest to Forecast Earth's Outer Van Allen Radiation Belt

The outer Van Allen radiation belt is highly dynamic in both strength and location, being driven by several distinct physical processes, making it difficult to predict for spacecraft operators. Forecasting models exist, in part, to minimise potential damage caused by this natural hazard. Both physics-based and machine learning models exist; generally, physics-based models allow for a deeper understanding of the system, while machine learning models offer a computationally cheap way to make a forecast, but do not always provide physical insight. We present VAMPIRE (Van Allen belt Multi-day Predictions by Implementing a Random forest for Electrons), a pair of simple machine learning models, along with an analysis of model feature importance, to both forecast and understand the physical drivers of the outer radiation belt. We use a random forest methodology to predict whether the daily maximum ∼2 MeV electron flux and daily fluence across the entirety of the outer belt crosses the alert levels, similar to the approach used by the UK Met Office. Both models show high levels of accuracy at both nowcasting and forecasting up to a week in advance. We use feature importance to determine the most important elements of each model, and demonstrate that these models also give an insight into the major drivers of the radiation belts, and the timescales on which they have an impact.