Evaluation of the Relative Importance of Storm Characteristics on the TEC Enhancements in the US Sector

The evolution of the total electron content (TEC) in the ionosphere in the aftermath of a large geomagnetic storm is complex and depends on a number of factors. In this study, we aim to use data science methods to ascertain the relative importance of a number of characteristics of large equinoctial geomagnetic storms in the solar cycles 23 and 24 on the TEC variations in the US sector. A random forest-based machine-learning method is often used to quantify the importance of a driver on an observed variable. However, the dataset consists of relatively small number of large storms and hence, this is an atypical application of a learning-driven method. Therefore, in addition to random forest computed importance, we also use statistical tools (correlation coefficients and multivariate linear regression) for the same data-set. A thorough comparison of the three different methodologies is used to understand the degree of agreement in determining the key variables that are most correlated with the TEC fluctuations.

The geographical domain under study comprises six geomagnetically different regions in the contiguous US sector. We find that different storm characteristics are important in understanding the ionospheric variability in each of these regions. The study describes a way to leverage the benefits of different analysis tools to reach a statistically reliable conclusion that addresses an important science question.