Correlation between bandwidth and frequency of plasmaspheric hiss uncovered with unsupervised machine learning

TL;DR

This study employs unsupervised machine learning to categorize plasmaspheric hiss spectral shapes, revealing a strong negative correlation between hiss frequency and bandwidth, supporting wave growth theories.

Contribution

It introduces an unsupervised machine learning approach to classify spectral shapes of plasmaspheric hiss, uncovering spatial patterns and correlations not seen in averaged data.

Findings

Strong negative correlation between hiss frequency and bandwidth

Spectral shape categorization reveals diverse patterns

Supports in situ wave growth hypothesis

Abstract

Previous statistical studies of plasmaspheric hiss investigated the averaged shape of the magnetic field power spectra at various points in the magnetosphere. However, this approach does not consider the fact that very diverse spectral shapes exist at a given L-shell and magnetic local time. Averaging the data together means that important features of the spectral shapes are lost. In this paper, we use an unsupervised machine learning technique to categorize plasmaspheric hiss. In contrast to the previous studies, this technique allows us to identify power spectra that have "similar" shapes and study their spatial distribution without averaging together vastly different spectral shapes. We show that strong negative correlations exist between the hiss frequency and bandwidth, which suggests that the observed patterns are consistent with in situ wave growth.