Global Ionospheric VTEC Data Completion Method Based on Aggregated Contextual-Transformation Generative Adversarial Nets
Rong Wang, Yibin Yao, Peng Chen, Leran Fu, Xin Gao, Liang Zhang · IEEE Transactions on Geoscience and Remote Sensing · 2024
The determination of ionospheric total electron content (TEC) is crucial for various applications in space weather. However, due to the uneven distribution of stations, complex data processing and reconstruction algorithms are usually required to obtain a seamless global TEC map. To reduce the difficulty of this process, we show the possibility of reconstructing a high-resolution global TEC map based on the image inpainting method aggregated contextual-transformation generative adversarial nets (AOT-GANs). First, the AOT-GAN model is used to learn the data fitting process and TEC spatial distribution characteristics in the UQRG, and then the data completion performance and generalization ability of the model are verified under different data-missing conditions and geomagnetic activity levels. The verification results show that the model has relatively reliable completion results under different conditions. The average root mean squared error (RMSE) between the filled results and UQRG is mainly concentrated in$2\sim 3$TECU, and the average structural similarity index measure (SSIM) index is mainly concentrated in$0.95\sim 0.97$. Even during the geomagnetic storm periods and the land data missing rate exceeds 30%, more than 92% of the biases are still within ±5 TECU. In addition, the model can also achieve considerable results when completing CODG, with more than 63% of the biases within ±1 TECU and more than 92% of the biases within ±5 TECU. Finally, the Massachusetts Institute of Technology (MIT)-TEC map is filled using the trained model. When part of MIT-TEC is removed, the biases between the completed result and the original MIT-TEC are relatively small. For the ocean area, the completed MIT-TEC map has the lowest RMSE and the STD level is similar to the ESAG product. The completed MIT-TEC maps not only maintain the global structure of TEC but also have rich details.