Interpolating Electron Diffusion Coefficients in Earth's Radiation Belts: A Neural Network Model Inspired by Video Frame Interpolation

Abstract Many numerical models have been built to uncover the physical mechanisms governing radiation belt dynamics and enable real‐time forecasting. The diffusion coefficient, as a critical input for these models, is usually obtained through linear interpolation from precomputed diffusion coefficient libraries. However, linear interpolation unavoidably introduces overlap issues, compromising the accuracy of the simulation results. In this study, we propose a diffusion Coefficient Interpolation Neural nEt (CINE) model, inspired by video frame interpolation, to address this overlap issue. Without preexisting diffusion coefficients for training, the CINE model successfully interpolates diffusion coefficients induced by distinct physical mechanisms (e.g., wave‐particle interactions and atmospheric collisions). We further analyze optimal interpolation intervals for diffusion coefficients (ΔL ≤ 0.8 for hiss waves and ΔL ≤ 0.2 for atmospheric collisions) based on a threshold of Structural Similarity Index Measure (SSIM) = 0.98. CINE can also be used to interpolate other parameters (e.g., phase space density) to reduce overlap issues. These enable the CINE model to efficiently accelerate radiation belt simulation on high‐resolution grids, serve for real‐time space weather forecasting and protect the safety of satellites and spaceflight.