Application to a Storm

(Work in Progess)

The storm

To find a storm, we search through the list provided by S. Watari (2017) (see here for the list itself). We pick the St. Patrics day storm, since everyone is a big fan of this storm.

The chosen storm is a sudden commencement related to a partial halo CME. The storm occured between 2015/03/17 and 2015/03/22.

Unfortunately, during this start of this storm (first ~1500 minutes) there is no good quality GOES data. To fill the gap in data, we simply forward-fill the last good measurement to fill in the gaps. This means that the dynamics in the first 1500 minutes are driving by the other features input to the model.

In both models (though more significantly for the index model), we observe a flux dropout at at the beginning of the storm correpsonding to the CME hitting the magnetosphere (see the spike in SYM-H). The index model then has an enhancement of the higher energy electron fluxes to ~100x the initial value. These fluxes remain high even during the recover phase of the storm. In contrast, the solar wind model shows a much more gradual increase in the fluxes, though reaching the same eventual level.

During the storm there is an increase in the uncertainty of the predictions - this is likely due to storms being a sparsely represented phenomena in the training dataset. We also notice that the solar wind model shows significantly more uncertainty over the storm.

For such reconstructions, it may be that the results are improved by combining the data available between GOES 13, 14 and 15 (these plots use only GOES 15 data).