Forecasting extreme geomagnetic storms using different statistical models

Extreme geomagnetic storm events could cause hazardous damage to the technological infrastructure that increasingly underpins modern society. Being able to forecast the next extreme geomagnetic storm is thus crucial to navigating risks in the 21^st century. There have been quite a number of studies on using extreme value theory to forecast future extreme geomagnetic storms. However, to the best of our knowledge, each study selects one model and one estimation method. In this study, we demonstrate that different estimation methods for the same extreme value model and different extreme value models can produce very different estimates of return levels when applied to the same dataset. We propose to use the average of the estimated return levels from different models and different estimation methods to produce more robust and reliable forecasts.

We apply this method to the geomagnetic field data measured in every minute in the period between 1994 and 2019 at Eyrewell, Canterbury, New Zealand. We focus on the horizontal components, and estimate the return levels of the ramp change in the horizontal component. The resulting return levels for the ramp change in the horizontal component using different types of models and different estimation methods show the importance of using model-averaged forecasts.

In this presentation, we also demonstrate forecasts of future geomagnetic storms using counting processes applied to the wavelet spectrum.