Principal component analysis as a tool to extract Sq variation from the geomagnetic field observations: conditions of applicability

TL;DR

This study evaluates the effectiveness of principal component analysis (PCA) in extracting Sq geomagnetic variations from observational data, highlighting its automatic extraction capability for Y and Z components and the need for reference comparisons for the X component.

Contribution

It demonstrates the conditions under which PCA can reliably extract Sq variations from geomagnetic data, comparing different reference series and emphasizing the importance of similarity metrics like DTW.

Findings

PCA automatically extracts Sq variation for Y and Z components.

For X component, reference series comparison is necessary.

DIFI3 model-based reference series outperforms other models.

Abstract

In this paper, we analyze the applicability of the principal component analysis (PCA) as a tool to extract the Sq variation of the geomagnetic field. We tested different geomagnetic field components and used data measured at different levels of the solar and geomagnetic activity and during different months. Geomagnetic field variations obtained with PCA were classified as SqPCA using two types of reference series: SqIQD series calculated using geomagnetically quiet days and simulations of the ionospheric field with models. The results for the X and Y and Z components are essentially different. The Sq variation is always filtered to the first PCA mode for the Y and Z components. Thus, PCA can automatically extract the Sq variation from the observations of the Y and Z components of the geomagnetic field. For the X component, the automatic extraction of the Sq variation is not possible, and a complimentary analysis, like a comparison to a reference series, is always needed. We tested two types of reference series: the mean SqIQD and the outputs of the CM5 and DIFI3 models. Our results show that both the data-based and model-based reference series can be used but the DIFI3 model performs better. We also recommend estimating the similarity of the series not with the correlation analysis but using metrics that account for possible local stretching/compressing of the compared series, for example, the dynamic time warping (DTW) distance.