Automatic Recognition And Research Of Space Hurricane Based On Deep Learning

During geomagnetically active periods,the solar wind easily invades the Earth’s magnetosphere and can cause strong auroral activity in the auroral ellipse,while during geomagnetically quiet periods,the efficiency of the solar wind invasion of the Earth is significantly reduced,but sometimes there may still be localized energy injections in the polar cap region that are comparable to magnetic storm activity.Space hurricane is a newly discovered typhoon-like phenomenon occurring in the ionosphere and magnetosphere of the Earth’s polar cap region,optically manifested as large-scale cyclone-like auroral structures in the polar cap region,which can cause solar wind energy injections comparable to super magnetic storms and greatly enhance the efficiency of solar wind-magnetosphere-ionosphere energy coupling.In addition,the extreme space environment caused by Space hurricane can directly affect radio communication,navigation and positioning,and over-the-horizon radar detection in the relevant regions.The automatic identification of Space hurricane based on deep learning methods and the in-depth study of their observation characteristics and occurrence laws can help deepen the understanding of the solar wind-magnetosphereionosphere coupling process during the geomagnetic calm period,provide a physical basis for space weather modeling and forecasting in polar regions,and have important application values for ensuring the reliable operation of radio communication and navigation systems in trans-polar regions.In this paper,we study the observation characteristics,formation and evolution mechanism of Space hurricane in detail by using the observation and particle detection of DMSP,combined with the synergistic observation of ground-based radar,and propose an automatic Space hurricane identification method based on deep learning.The main results are summarized as follows:1.The observed characteristics of Space hurricane occurring in the northern and southern hemispheres are obtained.Space hurricane in both the northern and southern hemispheres occur during the northward IMF,and those in the northern hemisphere occur in summer and the positive By condition,while those in the southern hemisphere occur in summer and the negative By condition;using DMSP satellite auroral imaging and particle detection data,it is found that Space hurricane in both the northern and southern hemispheres occur within the polar caps and have cyclonic auroral bright spot structures with strong convective shear at their edges,accompanied by upward field currents and inverted "V" electron energy spectra,as well as significant electron temperature increase.Combined with the ground-based incoherent scattering radar observations in the northern hemisphere,it is found that when the radar beam sweeps through the Space hurricane,it is accompanied by significant electron settling,electron temperature increase,ion temperature increase,and convective velocity shear.The occurrence conditions and observational characteristics of Space hurricane in the northern and southern hemispheres suggest that the formation of Space hurricane is associated with tail-valve reconnection occurring in the high-latitude tail-valve region at the top of the magnetosphere.2.An automatic identification method for Space hurricane is proposed.Based on the 2011 and 2012 Space hurricane datasets,six typical deep learning models are trained using a transfer learning approach,and the effect of automatic model learning of Space hurricane features is observed by means of feature visualization.The results of the test set(2005-2021 except 2011 and 2012)show that the EfficientNetB2 model works best in the northern hemisphere with an accuracy of 97.7%,while the EfficientNetB0 model with the best effect in the southern hemisphere has an accuracy of only 90%due to problems such as smaller data sets and satellite orbits in the southern hemisphere.Finally,based on the automatically identified Space hurricane events,combined with manual discrimination,a total of 835 northern hemisphere Space hurricane events and 229 southern hemisphere Space hurricane events are found.3.The occurrence conditions and spatial and temporal distribution patterns of Space hurricane in the northern and southern hemispheres are given statistically.The distribution of these events with the solar activity cycle,seasons and UT,as well as the dependence on IMF and solar wind parameters are analyzed.It is found that:the occurrence of Space hurricane is significantly positively correlated with the solar activity cycle,mainly occurring in summer in the respective hemispheres,with a higher incidence in the late afternoon;Space hurricane mainly occur at 15 MLT and near the geomagnetic latitude of 85°;in the northern hemisphere,Space hurricane mainly occur under+By dominated northward IMF conditions,while in the southern hemisphere,Space hurricane mainly occur under-By dominated northward IMF conditions;the conditions of low solar wind number density and low solar wind dynamic pressure seem to be more favorable for Space hurricane events.These results suggest that the larger solar altitude angle and dipole tilt(in summer)under IMF northward and By positive(negative)conditions help the solar wind invade the magnetospheric tail flap region of the earth’s northern hemisphere(southern hemisphere)and open the solar wind-magnetosphereionosphere coupling channel,allowing Space hurricane formation.The results of this paper will deepen the knowledge of the observational characteristics and physical laws of Space hurricane and help to better understand the solar windmagnetosphere-ionosphere coupling process,which can provide a physical basis for space weather modeling and applications in polar regions.