Item

Abstract

Accurate modeling of the plasmasphere is essential for understanding inner magnetospheric dynamics. In this study, we present an updated version of the PINE (Plasma density in the Inner magnetosphere Neural network-based Empirical) model, improved by integrating data from both the Van Allen Probes (VAP) and Arase missions. Intercalibration at L-MLT conjunctions indicates that electron density measurements between the two satellites are highly consistent, which allows us to integrate these observations in model development. We evaluate three PINE models trained separately on VAP, Arase, and combined data sets, assessing performance using root mean square errors and mean errors across independent test sets. The combined model trained on VAP and Arase data sets consistently outperforms the models trained on the individual data sets, achieving higher accuracy, lower prediction bias, and lower error variance across all magnetic local times and L-shells. By incorporating Arase measurements, the updated PINE model extends coverage to lower altitudes and L-shells, improving performance in the ionosphere–plasmasphere transition region, where previous models were limited. Furthermore, we demonstrate its ability to reconstruct global plasmaspheric structures by comparing model outputs with independent extreme ultraviolet observations of the plasmapause from the IMAGE mission. These results highlight the benefits of multi-mission data integration in improving model robustness and generalizability in space weather applications.