The real-time storm prediction method, which is under active development at this time, has exhibited certain predictable characteristics, which are illustrated below. These examples are discussed in detail to help understand the results of our work. As the method is refined, we will also discuss how the refinements affect the method.
The solar wind structure shown in panels (a) and (b) is a classic magnetic cloud [Burlaga, 1988] that was observed by the IMP 8 satellite on 13 - 14 January 1988. The magnetic cloud consists of two long-duration Bz events. The duration and maximum Bz, tau' and Bzm', are accurately estimated after the inital 10-15% of each event. The calculated probability P1 = 1 is shown with a dashed line for the first event. This means that the structure is recognized as a long-duration Bz > 0 event to be followed by a Bz < 0 event of comparable duration and Bzm < 0. The solid line means that Bz < 0 and the SW event is geoeffective. The dashed curve in panel (b) shows the predicted Bz(t) profile, computed at the time marked by a diamond. The predicted Bz profile closely resembles the actual profile. This represents a remarkably accurate prediction of the magnetic field profile in the solar wind that has yet to arrive at the observer (nearly 30 hours upstream in this example).
The polarity of the Bz component appears to have a solar-cycle dependence: every other solar cycle, magnetic clouds tend to arrive with Bz > 0 first. We expect the above example to typify magnetic clouds during the next solar cycle. During the present cycle, the following example typifies magnetic clouds encountered at 1 AU.
This is a magnetic cloud, typical of the present solar cycle, with Bz < 0 arriving first. This situation allows shorter warning times, and often the onset (when Dst begins to decrease) of a storm has occurred when P1 becomes significant. However, the time before minimum Dst, the duration, and the severity can still the predicted in advance. This storm was marginal, having minimum Dst = -84 nT (provisional). This is correctly indicated by P1 hovering around 0.5.
This is an example of a miss. The storm on DOY 111 was caused by a long-duration Bz < 0 event (panel b), but this southward Bz period was interrupted by short northward periods. The algorithm resets each time Bz reverses sign, so that these southward Bz events were treated as uncorrelated events. In reality, the magnetosphere was continually stressed.
Another miss. The storm was caused by a high-speed solar wind stream (not shown here). Magnetic field data alone indicated no geoeffective structure. The algorithm does not look for such SW structures because they typically do not lead to very large storms.
