Reconstructing Solar Wind Profiles Associated With Extreme Magnetic Storms: A Machine Learning Approach

Abstract

The lack of data on solar wind have prevented a detailed understanding of extreme magnetic storms. To address this issue, we apply a machine learning technique in the form of an Echo State Network (ESN) to reconstruct solar wind data for several extreme magnetic storms for which little or no solar wind data were previously available. Multiple geomagnetic activity indices are used as the input data for the ESN, which produces a continuous time series of solar wind parameters as output. As a result, the solar wind parameters for the largest storm event in March 1989 are obtained, and the minimum Bz is estimated to be −95 nT ±10 nT. Two different types of solar wind profiles are discussed for the extreme magnetic storms, a sheath-driven profile and a magnetic cloud-driven profile. The results reported here will be highly useful as input data for future simulation studies modeling extreme magnetic storms.