Prediction of radiation belt relativistic electron phase space density using artificial neural networks.

This article focuses on the development and evaluation of an artificial neural network (ANN) model to predict the phase space density (PSD) of relativistic electrons (>1 MeV) in Earth's outer radiation belt, using Van Allen Probe-A data from 2012 to 2019 alongside 12 solar wind and geomagnetic indices. The model predicts electron PSD at μ = 1000 MeV/G with two values of the second adiabatic invariant K (0.08 and 0.17 G^1/2R_E) across L* values from 2.0 to 5.5, achieving high accuracy within the core region (L* = 3.0–5.5) with root mean square errors around 0.13, prediction efficiencies near 0.99, and Pearson correlation coefficients above 0.995. Statistical results show that over 99.9% of predictions differ from observations by less than one order of magnitude, and the model effectively reproduces temporal PSD variations during both quiet and active geomagnetic conditions. The study highlights the model's potential to improve understanding of radiation belt electron dynamics and contribute to future space weather warning systems, with plans to incorporate additional data sources and interpretive machine learning techniques in future work.