Determination of solar cycle variations of midlatitude ELF/VLF chorus and hiss via automated signal detection

Abstract

An automated algorithm for detecting chorus and hiss emissions in ground-based extremely low frequency/very low frequency (ELF/VLF) wave receiver data is developed and applied to 10 years of data collected at Palmer Station, Antarctica (L = 2.4, 50 degrees S invariant latitude). The algorithm consists of three major processing steps. First, sferics and power line hum are removed from the broadband data. Second, individual events are detected and a set of 19 scalar event parameters are determined. Finally, on the basis of the parameters, detected events are categorized by a sequential pair of neural networks as either chorus, hiss, or noise. The detector runs on a modern 8-core computer at a speed of 350x real time. Results of training indicate that the neural networks are capable of differentiating between noise and emissions with a 92% success rate and between chorus and hiss with an 84% success rate. Data collected at Palmer from May 2000 to May 2010 were processed, and yearly and seasonal trends of chorus and hiss are analyzed. Yearly occurrence rates of chorus and hiss are strongly dependent on the geomagnetic disturbance level, as measured by K-p and AE, whereas seasonal occurrence rates are more strongly dependent on variations of the day/night terminator and associated variations in ionospheric absorption.