Speaker
Description
Solar Energetic Particle (SEP) events pose significant threats to space-based assets and human spaceflight. Accurate and timely prediction of SEP event profiles is crucial for space weather forecasting. This study presents a physics-based SEP event modeling approach, enhanced by Bayesian optimization techniques, and applies it to the well-observed October 2021 SEP event. This event, originating from an M1.6 flare at 06:19 UT on October 9, 2021, and an associated halo CME with an initial velocity of 712 km/s at 07:12 UT, resulted in a >0.1 pfu @ 10 MeV proton flux observed by GOES-17. Its uniqueness lies in the availability of multi-spacecraft in-situ proton flux observations from PSP, SolO, BepiColombo, SOHO, and STEREO-A, enabling robust validation of our coupled simulation.
Our numerical approach integrates CME propagation with an SEP solver. First, CME propagation is reconstructed using the SUSANOO-CME model (Shiota et al. 2016), driven by ISEE IPS data (Iwai et al. 2019), to derive CME shock parameters. High-energy particles are then injected at the shock position based on Diffusive Shock Acceleration (DSA) theory (Hu et al. 2017), following Li, Shalchi et al. (2012) for the injection threshold. An injection efficiency ε and a parameter ξ inversely proportional to the particle acceleration rate at the shock front are treated as hyperparameters. Particles are propagated upstream using the Focused Transport Equation (Ruffolo 1995), with the parallel mean free path λ optimized as another hyperparameter. This framework is detailed in Minoshima et al. (submitted).
Model performance is evaluated using the Mean Absolute Error (MAE) of the logarithm of observed and simulated fluxes, after interpolating both to 5-minute intervals. We optimize the hyperparameters (λ,ξ,ε) using Bayesian optimization techniques implemented in Optuna (Akiba et al. 2019). We conducted trials for global optimization, minimizing the MAE across all spacecraft and all energy bands, as well as trials for individual spacecraft optimization, minimizing the MAE for each spacecraft across all its energy bands.
Our results show that the optimal global parameter for the parallel mean free path λ is very high, which is consistent with Palmerio et al. (2024) assuming scatter-free transportation. Individual spacecraft optimization yielded a lower mean MAE, indicating a better fit. Furthermore, by calculating Permutation Importance using the pairs of numerical calculation results and parameter values obtained from Optuna, we found that the particle injection efficiency ε at the shock position is a significantly important parameter, accounting for the majority of the importance in our numerical modeling. Future work includes optimizing parameters per magnetic field line connecting to each sector of the shock and applying these methods to a larger number of SEP events to establish robust initial parameter sets for predictive numerical models, incorporating further data-driven enhancements.
