Statistical Analysis of Electromagnetic Ion Cyclotron Rising‐Tone Emissions Based on Deep Learning

Abstract

AbstractSeveral studies have shown the importance of electromagnetic ion cyclotron (EMIC) rising‐tone emissions to the rapid precipitation of energetic radiation belt electrons. Based on a large number of Van Allen Probes observations from October 2012 to July 2019, we identify EMIC rising‐tone emissions using a convolutional neural network (CNN), a modern deep learning technique. Results of training indicate that the CNN is capable of identifying EMIC rising‐tone emissions with a recall of 99.3%. The statistical analysis of the wave events identified reveals that the average occurrence rate of the events is about 0.016%, with a high occurrence rate from the forenoon to the dusk sector at L > 5. There are also events observed at L < 5, which are scattered at almost all magnetic local times. The events in the hydrogen and helium bands have comparable wave amplitudes on average, but the larger amplitude events tend to occur around noon and in the afternoon sector in the hydrogen and helium bands, respectively. In addition, the frequency sweep rate tends to increase with the wave frequency. The frequency sweep rates of the hydrogen band EMIC rising‐tone emissions are about 6 times larger than those of the helium band events. There is also a positive correlation between the wave amplitudes and the sweep rates of the hydrogen band emissions.