Core Concepts
The proposed 4D DeepONet-STEC model can accurately predict the 4D temporal-spatial integrated ionospheric Slant Total Electron Content (STEC) parameter for specified ground station-satellite ray paths globally.
Abstract
The paper presents a novel 4D ionospheric parameter prediction framework based on the deep operator regression network, called DeepONet-STEC. The key highlights are:
The DeepONet-STEC model is capable of simultaneously providing high accuracy spatial estimation and temporal prediction of the STEC values for global GNSS data.
The model utilizes kernel methods to construct the input function from the real observation data, and adopts a deep operator regression architecture to learn the nonlinear mapping between the input function and the STEC output.
The model is validated using both simulated and real observation datasets under global and regional GNSS station geometry scenarios, covering both quiet and active solar magnetic storm periods.
In the quiet period, the three-day 72 hour prediction using the DeepONet-STEC model achieves high accuracy with RMSE less than 0.7 TECU and R-squared over 0.96 for the simulated data. For the real observation data, the model still maintains good performance with RMSE around 1.3-1.7 TECU and R-squared over 0.75.
Under active solar magnetic storm periods, the DeepONet-STEC model also demonstrated its robustness and superiority compared to traditional deep learning methods.
The proposed framework presents a neural operator regression architecture for accurately predicting the 4D temporal-spatial ionospheric parameter, which can be further extended for various space applications.
Stats
The maximum STEC value in quiet periods is 42 TECU.
The maximum STEC value in storm periods is 243 TECU.
Quotes
"The proposed 4D DeepONet-STEC model can accurately predict the 4D temporal-spatial integrated ionospheric Slant Total Electron Content (STEC) parameter for specified ground station-satellite ray paths globally."
"In the quiet period, the three-day 72 hour prediction using the DeepONet-STEC model achieves high accuracy with RMSE less than 0.7 TECU and R-squared over 0.96 for the simulated data."
"Under active solar magnetic storm periods, the DeepONet-STEC model also demonstrated its robustness and superiority compared to traditional deep learning methods."