[P-EM13] Study of coupling processes in solar-terrestrial system

[PEM13-06] Automatic detection and statistical analysis of MSTIDs based on deep learning

*Peng Liu¹, Tatsuhiro Yokoyama¹ (1.Kyoto University)

Keywords:TEC, MSTIDs, ionosphere, deep learning, FCN, instance segmentation

Medium-scale traveling ionospheric disturbances (MSTIDs) are the most typical irregularities of nighttime mid-latitude ionosphere, which are usually associated with the periodical variation of total electron content (TEC) at F region and cause the degradation of satellite positioning accuracy. Using detrended TEC map provided by Japan GEONET GPS of Geospatial Information Authority, MSTIDs are observed as wavy structures in the plasma density at F-region heights with horizontal wavelengths of 100-1000 km. To detect MSTIDs and analyze the parameter of them automatically, we propose a fully convolutional network (FCN) based on instance segmentation, trained by 1500 nighttime detrended TEC images from January to July of 2019. This network could detect the position of nighttime MSTIDs from detrended TEC maps with up to 85% precision, then derive out the parameters such as wavelength, central coordinates, period, direction, duration and occurrence rate. we have processed the data from 1997 to 2020 (2 solar cycles) with this network, and statistical results are consistent with the previous research, especially effects of solar activity and seasonal dependence. So far, our research is the first one to apply the fully convolutional network to ionosphere irregularity automatic detection and analyzation.