Forecasting the properties of the solar wind using simple pattern recognition

An accurate forecast of the solar wind plasma and magnetic field properties is a crucial capability for space weather prediction. However, thus far, it has been limited to the large-scale properties of the solar wind plasma or the arrival time of a coronal mass ejection from the Sun. As yet there are no reliable forecasts for the north-south interplanetary magnetic field component, B-n (or, equivalently, B-z). In this study, we develop a technique for predicting the magnetic and plasma state of the solar wind t hours into the future (where t can range from 6h to several weeks) based on a simple pattern recognition algorithm. At some time, t, the algorithm takes the previous t hours and compares it with a sliding window of t hours running back all the way through the data. For each window, a Euclidean distance is computed. These are ranked, and the top 50 are used as starting point realizations from which to make ensemble forecasts of the next t hours. We find that this approach works remarkably well for most solar wind parameters such as v, n(p), T-p, and even B-r and B-t, but only modestly better than our baseline model for B-n. We discuss why this is so and suggest how more sophisticated techniques might be applied to improve the prediction scheme.