Automatic Real‐Time Ionogram Scaling Method for Global Ionosondes Using Artificial Intelligence

An ionosonde is an instrument used for local ionosphere remote sensing observations and has the longest history and widest global distribution. Ionograms observed by ionosondes, which depict the electron density profile as a function of virtual height, are critical for ionospheric modeling accuracy and space weather nowcasting. However, ionograms acquired from ionosondes worldwide differ substantially in noise characteristics, presentation style, and data format, which poses significant challenges for the development of automatic scaling methods. Meanwhile, the accuracy, processing speed and versatility of existing methods do not meet the requirements of space weather nowcasting. In this study, a new scaling method based on deep‐learning instance segmentation is proposed to accurately and simultaneously derive ionospheric parameters for ionosondes worldwide. The method consists of three steps: noise stripe removal, ionospheric echo detection, and pixel‐to‐ionogram coordinate system recovery. The performance of different embedded artificial intelligence models is benchmarked using the same manually labeled data set. Experimental results indicate that the proposed method achieves a maximum detection accuracy of 89.8% and scaling speed of 143.89 frame‐per‐second. Ionograms recorded by 19 ionosondes in East and Southeast Asia during 2000–2024 are analyzed using multiple methods. The proposed algorithm outperforms existing automatic ionogram scaling methods (e.g., ARTIST) by providing more comprehensive and precise ionospheric parameter outputs. Ionospheric scaling results indicate that the maximum errors between the proposed method and existing manual methods are low to f min = 0.941 MHz among critical frequencies and F = 24.659 km among virtual heights. These high‐accuracy scaling parameters enable potential future calibration of International Reference Ionosphere model.