Classifying Signatures of Sudden Ionospheric Disturbances

TL;DR

This paper presents an automated method using machine learning to identify noisy data in ground-based measurements of Sudden Ionospheric Disturbances, improving data quality for scientific analysis.

Contribution

The study develops a feature-based support vector machine classifier to automatically distinguish clean from contaminated SID data sets.

Findings

Classifier achieves True Skill Score of ~0.75

Daytime and nighttime signal differences are key features

Method enhances data reliability for ionospheric research

Abstract

Solar activity, such as flares, produce bursts of high-energy radiation that temporarily enhance the D-region of the ionosphere and attenuate low-frequency radio waves. To track these Sudden Ionospheric Disturbances (SIDs), which disrupt communication signals and perturb satellite orbits, Scherrer et al. (2008) developed an international, ground-based network of around 500 SID monitors that measure the signal strength of low-frequency radio waves. However, these monitors suffer from a host of noise contamination issues that preclude their use for rigorous scientific analysis. As such, we attempt to create an algorithm to automatically identify noisy, contaminated SID data sets from clean ones. To do so, we develop a set of features to characterize times series measurements from SID monitors and use these features, along with a binary classifer called a support vector machine, to automatically assess the quality of the SID data. We compute the True Skill Score, a metric that measures the performance of our classifier, and find that it is ~0.75+/-0.06. We find features characterizing the difference between the daytime and nighttime signal strength of low-frequency radio waves most effectively discern noisy data sets from clean ones.