The Development of Spatial Attention U‐Net for the Recovery of Ionospheric Measurements and Extraction of Ionospheric Parameters

• Published in: Radio science 2022-08, Vol.57 (8), p.n/a
• Main Authors: Huang, Guan‐Han, Dmitriev, Alexei V., Lin, Chia‐Hsien, Chang, Yu‐Chi, Hsieh, Mon‐Chai, Tsogtbaatar, Enkhtuya, Mendoza, Merlin M., Hsu, Hao‐Wei, Lin, Yu‐Chiang, Tsai, Lung‐Chih, Li, Yung‐Hui
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Critical frequencies; Datasets; Deep learning; E region; F 2 region; image segmentation; ionosonde; Ionosphere; Labels; Training
• ISSN: 0048-6604; 1944-799X
• DOI: 10.1029/2022RS007471

We train a deep learning artificial neural network model, Spatial Attention U‐Net, to recover useful ionospheric signals from noisy ionogram data measured by Hualien's Vertical Incidence Pulsed Ionospheric Radar. Our results show that the model can well identify F2 layer ordinary and extraordinary modes (F2o and F2x) and the combined signals of the E layer (ordinary and extraordinary modes and sporadic Es). The model is also capable of identifying some signals that were not labeled. The performance of the model can be significantly degraded by insufficient number of samples in the data set. From the recovered signals, we determine the critical frequencies of F2o and F2x and the intersection frequency between the two signals. The difference between the two critical frequencies is peaking at 0.63 MHz, with the uncertainty being 0.18 MHz. Plain Language Summary A large amount of images are retrieved by a specialized instrument designed to make observations in the ionosphere. These images are contaminated by instrumental noises. In order to recover useful signals from these noises, we train a deep learning model. A data set containing the labeled signals is used for both training and validating the model performance. The desired signals are manually labeled using a labeling software. By comparing the model predictions with the labels, the results show that the model can well identify the elongated, overlapping, or compact signals. The model is also capable of correcting some missing and incorrect labels. The performance of the model is sensitive to the data number of the corresponding labels fed to the model during training. The recovered useful signals are then used to estimate physical quantities that are important for the study of ionospheric physics. Key Points A deep learning model is applied to ionogram recovery The model can well identify the combined signals of the sporadic E layer and the ordinary and extraordinary modes of the F2 layer signal Critical frequencies of the modes and the intersection frequency between them are derived