Probabilistic Forecasts of Storm Sudden Commencements from Interplanetary Shocks Using Machine Learning

Abstract:

Rapid changes in the surface geomagnetic field can induce potentially damaging currents in artificial conductors on the ground. One phenomenon that can produce such changes in the Earth’s magnetic field are Sudden Commencements (SCs), which are associated with sharp increases in solar wind dynamic pressure usually caused by interplanetary shocks. SCs may also be followed by other longer lasting magnetospheric phenomena that cause large fluctuations in the geomagnetic field, geomagnetic storms for example. An SC that is followed by a geomagnetic storm may be termed a Storm Sudden Commencement (SSC). Therefore, the capability of predicting SC (or SSC) occurrence bears a high importance for the forecasting of space weather-related phenomena.

We investigate the ability of several different machine learning models to provide probabilistic predictions as to whether interplanetary shocks observed upstream of the Earth at L1 will lead to immediate or longer lasting magnetospheric activity (i.e. an SC or SSC, respectively). Four models are tested including linear (Logistic Regression), non-linear (Naive Bayes and Gaussian Process) and ensemble (Random Forest) models. They are shown to provide skilful and reliable forecasts of SCs, strongly outperforming climatological forecasts. The most powerful predictive parameters are also evaluated and discussed. The models are also shown to produce skilful forecasts of SSCs, though with less reliability than was found for SCs. The most important parameters for these predictions are compared and contrasted with the previous forecasts of SCs, which are suggestive of differing driving processes and interactions. Finally, the response of the different models is explored with hypothetical extreme data beyond current observations, showing dramatically different extrapolations, highlighting the significance of model selection.