Lightning whistlers in DEMETER Satellite Data: Identification and Properties

Lightning whistlers play an important role in the loss of energetic electrons from the Van Allen radiation belts and their overall dynamics. Whistlers are generated by atmospheric lightning strokes and due to a few thousands of thunderstorms occurring simultaneously at any moment, they are very common in the satellite measurements. However, manual whistler identification is very time consuming and unfeasible on a large scale. In this work, we introduce an automatic whistler identification and analysis routine that identify individual whistlers and determine their dispersion from DEMETER satellite burst mode measurements. For this purpose, we use machine learning approach. Specifically, YOLOv11 and Faster R-CNN object detection techniques for whistler identification and genetic algorithm for analysis of their dispersion. We use a manually identified dataset of about 600 spectrogram images containing approximately 6,000 whistlers to train both models. Overall, we detect several millions of whistlers in DEMETER burst mode measurements. Comparing both models with whistler detection neural network onboard DEMETER we observe similar behavior between all three models. During the northern summer rich on thunderstorms, low-dispersion whistlers are observed more frequently in the Northern Hemisphere and high-dispersion whistlers are observed more frequently in the Southern Hemisphere. The results demonstrate that modern object detection techniques can be an eligible and robust approach for plasma wave identification and provide a valuable basis for future plasma wave studies.