Monitoring Space Weather: Using Automated, Accurate Neural Network Based Whistler Segmentation for Whistler Inversion

It is challenging, yet important, to measure the—ever-changing—cold electron density in the plasmasphere. The cold inside and outside of plasmapause a key parameter for radiation belt dynamics. One indirect measurement through finding velocity dispersion relation exhibited by lightning induced whistlers. main difficulty method comes from low signal-to-noise ratios most ground-based whistler components. To provide accurate -shell measurements, components need be detectable noisy background, their characteristics reliably determined. For this reason, precise segmentation needed on spectrogram image. Here, we present fully automated way perform such an image leveraging power convolutional neural networks, state-of-the-art computer vision tasks. Testing proposed against manually, semi-manually segmented data set achieved 10% relative prediction error 80% traces, while shell, 5% 90% cases. By segmenting more than 1 million additional real traces Rothera station Antarctica, logged over 9 years, seasonal changes average were found. variations match previously published findings, confirm capabilities technique.