Predictions of Equatorial Vertical Plasma Drift Using TEC Data and a Neural Network Model

Abstract

Vertical plasma drift, v_z, plays a key role in the dynamics, morphology, and space weather effects of the equatorial and low latitude ionosphere. Modeling the drift has been an on-going effort for climatology-based prediction. To address daily prediction, the Vertical drIfts: Predicting Equatorial ionospheRic dynamics (VIPER) model has been developed. VIPER is a machine learning model that is trained on total electron content (TEC) data to predict low-latitude vertical plasma drift observed by the C/NOFS mission across the period 2009–2015. The uniqueness of VIPER is that it uses TEC data for the prediction, and the data is globally and readily available. A Gaussian fitting routine is developed to strengthen the link between TEC and v_z. VIPER is a multi-layer perceptron framework with Monte Carlo (MC) uncertainty estimation capabilities. It has a mean absolute error of 8.3 m/s, an R of 0.89/1, and a skill of 0.78/1, all of which are strong scores. The model is capped at quiet and unsettled activity levels (Kp < 3). MC analysis reveals that predictions should be interpreted as distributions and the uncertainty can vary with distributions of TEC data and regions of prediction even if the predicted value is the same. VIPER offers longitudinally global coverage and uncertainty estimation capabilities. It could also be expanded to handle storm-time conditions with additional work.