Comparison of PINN-SAMI3 and 4D-Var Data Assimilation Methods for Ionospheric Prediction

Purely data-driven ionospheric modeling inadequately adheres to fundamental physical laws. PINN-SAMI3 [1], as a novel data assimilation framework, integrates the control equations of the fully physical model SAMI3 with neural networks to construct a physics-informed ionospheric parameter inversion framework. This paper compares PINN-SAMI3 with traditional data assimilation methods EKF [2] and 4D-Var. The test results indicate that PINN-SAMI3 outperforms EKF but has lower accuracy than the 4D-Var method. This framework has important implications for the future construction of a hybrid model of space weather prediction that integrates physical models and neural networks.