Interpretable Machine Learning Procedure Unravels Hidden Interplanetary Drivers of the Low Latitude Dayside Magnetopause

Interplanetary parameters such as solar wind and interplanetary magnetic fields (IMF) drive the shape and size of the magnetopause jointly, which has complex relationships. In this study, we proposed an interpretable machine learning procedure to disentangle the influences of interplanetary parameters on the magnetopause standoff distance (MSD) and sort their importance in the MSD simulation. A magnetopause crossings database from the THEMIS satellites and interplanetary parameters from OMNI during the period of 2007-2016 are utilized to construct machine learning magnetopause models. SHapley Additive exPlanations (SHAP) is the basis of the interpretable procedure, which introduces interpretability and makes the machine learning magnetopause model to be a “white box”. The solar wind dynamic pressure and IMF B_Z are widely considered the top two important parameters that drive the MSD. However, the interpretable procedure suggests that the IMF magnitude (i.e. strength of the IMF) leads B_Z as the second most important interplanetary driver. This ranking result is unexpected, and it implies that the role of IMF magnitude is underestimated although magnetic pressure, which is a function of the IMF magnitude was considered in previous studies. The examination of disentangled effects of interplanetary parameters reveals that the combined influence of the IMF magnitude and B_Z can cause an MSD sag near B_Z = 5 nT. This is for the first time we conduct the interpretable concept into the machine learning model in the study of the magnetosphere.