Solar-Terrestrial Data Science: Prior Experience and Future Prospects

Abstract

Acquisition of relatively large data sets based on measurements in the interplanetary medium, throughout Earth's magnetosphere, and from ground-based platforms has been a hallmark of the heliophysics discipline for several decades. Early methods of time series analysis with such datasets revealed key causal physical relationships and led to successful forecast models of magnetospheric substorms and geomagnetic storms. Applying neural network methods and linear prediction filtering approaches provided tremendous insights into how solar wind-magnetosphere-ionosphere coupling worked under various forcing conditions. Some applications of neural net and related methods were viewed askance in earlier times because it was not obvious how to extract or infer the underlying physics of input-output relationships. Today, there are powerful new methods being developed in the data sciences that harken back to earlier successful specification and forecasting methods. This paper reviews briefly earlier work and looks at new prospects for heliophysics prediction methods.