A Deep Learning Approach for Automatic Ionogram Parameters Recognition With Convolutional Neural Networks

Abstract Typical ionosondes operate with >5 min time intervals, which is enough to obtain regular parameters of the ionosphere, but insufficient to observe short‐term processes in the Earth's ionosphere. The key point for this study is to increase the ionosondes data time resolution by automatization of ionogram scaling routine. In this study we show the results of implementation of deep learning approach for ionogram parameters scaling. We trained the model on 13 years ionogram data set of Sodankyla ionosonde at high latitude region (67°N). We tested our autoscaling program tool on 2021 years data set and evaluate errors between operator and automatic parameters scaling. The root mean square errors for critical frequencies foF2, foF1, foE, foEs, fmin, fbEs and virtual heights h′F, h′E, h′Es are estimated as 0.12 MHz (2 pixels), 0.07 MHz (1.16 pixels), 0.15 MHz (2.5 pixels), 0.33 MHz (5.5 pixels), 0.15 MHz (2.5 pixels), 0.17 MHz (2.83 pixels), 7.7 km (1.34 pixels), 7.0 km (1.22 pixels), 7.1 km (1.24 pixels), respectively.