Speakers
Description
Cumulative exposure to ionizing radiation at aviation altitudes poses significant health risks for aircrews and, at higher altitudes, astronauts. Physics-based models are commonly used to estimate radiation levels during flight; however, they often do not fully capture the rapidly varying and complex nature of atmospheric radiation, limiting real-time prediction accuracy.To address this limitation, we explore machine learning (ML) approaches to improve the analysis and nowcasting of aviation radiation.
Using newly compiled, ML-ready aviation radiation datasets, we train supervised ML models to identify nonlinear relationships between geospace environmental parameters and measured radiation effective dose rates. Our results show that a gradient boosting (XGBoost) model trained on the concurrent properties of the geospace environment improves radiation prediction accuracy by ~9% compared to the considered physics-based NAIRAS-v3 model. Feature importance analysis and Shapley Additive Explanations (SHAP) indicate key geospace parameters, including solar wind and solar polar fields, play a dominant role in controlling radiation variability at flight altitudes.
In a complementary observational study, we examine the role of secondary cosmic-ray muons in aviation radiation environments at altitudes below 15 km. Atmospheric muon flux measurements obtained from a CubeSat prototype developed by the Nuclear Physics Group at Georgia State University are analyzed alongside radiation doses modeled by NAIRAS-v3. Correlation analysis demonstrates a strong, statistically significant positive relationship between measured muon counts per minute and modeled radiation dose rates (µSv/h), with a Pearson correlation coefficient of r = 0.93.