Systematic Analysis of Machine Learning and Feature Selection Techniques for Prediction of the Kp Index

Abstract

The Kp index is a measure of the midlatitude global geomagnetic activity and represents short‐term magnetic variations driven by solar wind plasma and interplanetary magnetic field. The Kp index is one of the most widely used indicators for space weather alerts and serves as input to various models, such as for the thermosphere and the radiation belts. It is therefore crucial to predict the Kp index accurately. Previous work in this area has mostly employed artificial neural networks to nowcast Kp, based their inferences on the recent history of Kp and on solar wind measurements at L1. In this study, we systematically test how different machine learning techniques perform on the task of nowcasting and forecasting Kp for prediction horizons of up to 12 hr. Additionally, we investigate different methods of machine learning and information theory for selecting the optimal inputs to a predictive model. We illustrate how these methods can be applied to select the most important inputs to a predictive model of Kp and to significantly reduce input dimensionality. We compare our best performing models based on a reduced set of optimal inputs with the existing models of Kp, using different test intervals, and show how this selection can affect model performance.

Probabilistic prediction of geomagnetic storms and theKpindex

“…We have further demonstrated that including a simple, operationally-available proxy for the likelihood of solar activity improves the prediction of geomagnetic storms. The inability of K p prediction models to predict larger storms (K p ≥ 5) well from L1 solar wind data has previously been discussed in the literature (see, e.g., Zhelavskaya et al, 2019), and this work shows that including solar X-ray flux can directly improve the prediction of high levels of geomagnetic activity. In this work we found that including solar X-ray flux in our model features reduces the overall RMSE by 0.01, from 0.78 to 0.77.…”

“…Capturing uncertainty, providing probabilistic predictions and improving our ability to capture transient behavior are all within reach with modern tools and do not require sacrificing model predictive performance. We hope that future work continues to bring together recent advances in feature selection (e.g., Zhelavskaya et al, 2019), model design to accommodate probabilistic prediction, and more complex solar data sources such as solar magnetograms, to provide accurate forecasting of strong geomagnetic activity with longer lead times. Fig.…”

“…Further work is planned to investigate this work as well as incorporating other recent advances that will help improve the fidelity of our predictive model. We also note that Zhelavskaya et al (2019) explored methods for selecting model features and reducing a large number of candidate features to a smaller selection of those that are most important. Relative to their work we use a small number of features, all of which were selected based on a physically-motivated interpretation and subject to the constraint of being products generally available operationally.…”

The Space Weather Atmosphere Models and Indices (SWAMI) project: Overview and first results

“…While tests of Shprits et al (2019) provided a systematic error estimation for the models based on neural networks, it remained unclear if the accuracy could further be improved by utilizing different methods for regression. To address this question, Zhelavskaya et al (2019) compared the performance of a number of machine learning (ML) methods for predicting the Kp index. In particular, they used Linear Regression (LR), Gradient Boosting (GB) (e.g.…”

“…MI is a quantity that represents the "amount of information" that can be obtained from one random variable through observation of another random variable. See Figure 1 in Zhelavskaya et al (2019) for further illustration of these methods. The obtained results (Fig.…”