Using Recurrent Neural Networks to Provide Improved Predictions of Ionospheric Electron Density and Model Storm Activity

The ionosphere has a large impact on various systems, especially wireless communications, so knowing its state is of interest to many parties. By monitoring the ionosphere and the surrounding Earth-Sun system and using such observations, predictions of ionospheric activity can be improved. With this data, we aim to provide an accurate representation of the electron density of the ionosphere, an important indicator of the ionosphere's state. By building upon already existing techniques for predicting electron density, specifically recent machine learning ones, we have created a recurrent model that links solar wind and interplanetary magnetic field (IMF) values to observed measurements of electron density. As we are using machine learning techniques to do this, we are able to avoid complex calculations and definitions of physical processes to create an accurate and fast model of the ionosphere. Our current machine learning model uses historical IMF observations, and we have demonstrated a noticeable improvement over models without such inputs. This improved model has been used to create predictions of electromagnetic storms to allow for further analysis of the storm-time ionosphere. We also aim to further improve our approach by introducing location-based smoothing through convolution as well as integration of physics-based model predictions to provide a better electron density predictor.