Automatic Detection of Atmospherics and Tweek Atmospherics in Radio Spectrograms Based on a Deep Learning Approach
Abstract
Lightning strikes can be routinely observed by the radio receivers that measure electromagnetic signals in the very low frequency range. The acquired pulses called atmospherics provide valuable details about source lightning discharges and also about the Earth-ionosphere waveguide where they can propagate thousands of kilometers.
The automatic acquisition of the events requires also automatic methods for extraction of atmospherics' details to confirm the observed trends with statistical significance. For this purpose, we have developed a method based on a deep learning approach to automatically obtain the type of atmospherics, their exact time, and the frequency range from the frequency-time spectrograms.
The method that employs convolutional neural networks is designed to be adaptable to scientific needs and provide reliable results according to the requirements on the sensitivity to events, computation performance, and precision of extracted details. The efficiency and specific steps of our method are demonstrated for data recorded by the ELMAVAN-G instrument.
The comprehensive description of the method allows its usage for regular characterization of the ionospheric D-layer or for other similar applications in the future.