Deep-learning for ionogram automatic scaling

Scientists can study the global ionospheric weather by manually or automatically scaling ionograms recorded by global ionosondes to obtain characteristic values of D, E, F regions in the ionosphere. Therefore, fast and accurate ionogram scaling is crucial to real-time space weather monitoring, which is closely related to the performance of space-borne and ground-based technological systems as well as life on earth. The significant increase in data collections during recent years makes an impossible task for human experts to manually scale large amounts of ionograms in time. While the scaling accuracy of traditional automatic methods is less than that by human experts, making them insufficient for scientific tasks. Deep-learning is currently attracting immense research interest in many scaling tasks due to its powerful ability to deal with huge data collections. In this study, we present a deep-learning method for ionogram automatic scaling (DIAS) that can rapidly scale ionograms precisely from the ionosonde data. We trained and tested on data recorded by Wuhan ionosonde located at 114.4 degrees E and 30.5 degrees N. Our results show that the proposed deep-learning method improved the precision and recall rate by 8%, 17%, respectively, compared to using Automatic Real-Time Ionogram Scaling with True-height (ARTIST), which is the most-widely-used automatically scaling routine, in scaling E, F1 and F2 layers. The scaling accuracy of the ionograms provided by our deep-learning model is close to that by human experts, which suggests that the ionograms provided by our deep-learning method can be applied directly to global ionospheric weather nowcasting. Therefore, this study may contribute greatly to improve our knowledge of the ionospheric space. (C) 2020 COSPAR. Published by Elsevier Ltd. All rights reserved.