Solar Flares and Satellites: Testing the Sensitivity of the Iterative Ensemble Kalman Smoother

Abstract

Space weather (SWx), the complex set of conditions between the Sun and the Earth, is difficult to predict. However, accurate forecasting of the conditions in the interplanetary medium is essential due to the dangers that solar storms pose to technology on the Earth's surface and in the atmosphere. One way to improve forecasting is with data assimilation (DA), a technique that integrates downstream observations into estimates of the solar wind near the Sun. In this thesis, the efficacy of an iterative Ensemble Kalman Smoother (iEnKS) coupled with a reduced-dimension propagation model (HUXt) is investigated. Prior work has been done to assimilate observations from two satellites— STEREO-A and STEREO-B—into this DA algorithm. However, STEREO-B no longer provides operational data. The iEnKS algorithm has thus been tweaked to assimilate observations from a tertiary source — the Advanced Composition Explorer (ACE). An experiment was designed to test the performance of the iEnKS with and without STEREO-B observations over the year of 2012, chosen due to the interception of a large coronal mass ejection (CME) by STEREO-A near the midpoint of the temporal window. iEnKS performance was compared with the performance of a Variational DA technique developed a few years prior.

It was found that when STEREO-B observations were removed from the iEnKS, root-mean squared error (RMSE) at each of the three satellites over the forecast period increased between 3-4\%. This consistency was not observed with the Variational DA method in the same circumstances. Additionally, even after the removal of STEREO-B observations, the iEnKS performed more accurately than the Variational method in either case. This points to the effectiveness of the iEnKS as well as its response to changes in observation sources. Further research is needed to determine if error could further decrease with full-dimensional model coupling or the further optimization of the cost-function solving algorithm present in the iEnKS. However, these results are promising in terms of the operational implementation of DA for SWx applications.