11:00 AM - 1:00 PM
[PEM12-P01] Long-term analysis of electron density observed by the MU radar and auto-scaled ionosonde data
Keywords:MU radar, electron density, ionosonde, auto-scaling
The MU radar, located in Shigaraki-Cho, Koka City, Shiga Prefecture, is a large atmospheric radar designed to observe the middle and upper atmosphere, and has been observing the ionospheric F region as an Incoherent Scatter (IS) radar regularly since 1986. The IS radar emits radio waves into the upper atmosphere and is capable of estimating various physical quantities in the ionosphere that contribute to the intensity and spectrum of the scattered waves. Electron and ion temperatures, plasma drift velocities and echo power are regularly observed by the MU radar (http://www.rish.kyoto-u.ac.jp/mu/isdata/). First, we focus on the statistics of the echo power, which is proportional to the electron density. The peak echo power measured by the MU radar is calibrated as the electron density that corresponds to the foF2 measured by Kokubunji ionosonde. We compare the electron density observed by the MU radar with the International Reference Ionosphere (IRI) model.
In order to utilze the ionosonde at Shigaraki MU Observatory, we have developed an auto-scaling system of ionograms. A representative model of instance segmentation, which is called Mask Region-Convolutional Neural Network (R-CNN), is modifid in this research to detect foF2 values from ionograms obtained by Shigaraki ionosonde. The system was trained by 426 ionogram images in which ionospheric traces are labelled manually. The obtained foF2 values are generally consistent with those from Kokubunji ionosonde. The errors tend to be large especially in the period from April to August, which may be affected by strong sporadic-E layers. This system can be applied to other ionosonde systems where the auto-scaling system has not been established.
In order to utilze the ionosonde at Shigaraki MU Observatory, we have developed an auto-scaling system of ionograms. A representative model of instance segmentation, which is called Mask Region-Convolutional Neural Network (R-CNN), is modifid in this research to detect foF2 values from ionograms obtained by Shigaraki ionosonde. The system was trained by 426 ionogram images in which ionospheric traces are labelled manually. The obtained foF2 values are generally consistent with those from Kokubunji ionosonde. The errors tend to be large especially in the period from April to August, which may be affected by strong sporadic-E layers. This system can be applied to other ionosonde systems where the auto-scaling system has not been established.