Machine Learning-Based Detection and Characterisation of Magnetopause Crossings in the Lunar-Distance Magnetotail

Abstract

The Earth’s lunar-distance magnetopause is a highly dynamic boundary. Its position plays a key role in shaping weather events within the magnetotail and other space-weather phenom- ena. Statistical studies of the magnetotail require the identification of magnetopause crossings from spacecraft observations to study its dynamics. However, current detections have some shortcomings. Detecting magnetopause crossings can be time-consuming with rule-based methods and manual inspection, especially at the lunar distance due to its variability. Recent automated classifiers do cover larger datasets, but often miss the dynamic nature of the mag- netopause, due to their coarse timing and limited number of detected events. This limits the usability of these datasets for detailed studies of magnetotail dynamics. This thesis addresses these challenges by developing and evaluating machine learning ap- proaches for detecting magnetopause crossings in the lunar-distance magnetotail using AR- TEMIS mission data. First, a gradient-boosted decision tree is used as a baseline, trained to classify magnetosheath and magnetotail samples. Afterwards, a double masked autoen- coder (MAE) Transformer is introduced. This model uses a reconstruction-based method to detect changes in plasma regimes and shifts from the magnetotail to the magnetosheath and vice versa. Both models are trained and validated using a labelled dataset, combining ion spectrograms, plasma moments, and a list of known magnetopause crossings. The results show that the MAE Transformer achieves higher precision with similar recall, and improves timing accuracy compared to the baseline. The MAE transformer is applied to twelve years of ARTEMIS P1 and P2 data, detecting around 3000 magnetopause crossings. The spatial distribution of these crossings matches well with empirical magnetopause models, and a clear correlation to the solar cycle is observed. Outliers in the detected crossings are linked to solar and geomagnetic activity.