Nowcasting of Auroral Electron Precipitation Using an Artificial Neural Network

The impact of space weather phenomena on human society continues to grow as more technology is developed that is exposed to the space environment. As a result of modern data-driven sciences, we are now able to complement our physics understanding of the effect of the Sun’s variability on Earth with advanced machine learning techniques. A model applying the artificial neural network (ANN) methodology for predicting auroral electron energy flux in the atmosphere is presented. The ANN is developed using high-resolution solar wind, magnetic field, and plasma data from spacecraft orbiting the 1st Lagrange point, in addition to several ground-based magnetometers, and observations from the Defense Meteorological Satellite Program (DMSP). Different combinations of inputs were tested to find the optimal set of parameters. The model is then trained and validated using the Adam optimization method over a complete solar cycle. The resulting validation root-mean-square error after training is 0.12, and the Pearson correlation coefficient is equal to 0.93. Trained ANN output was then compared with OVATION Prime, which is the operational model running at the NOAA Space Weather Prediction Center, for multiple geomagnetic storm events. The resulting network in combination with other statistical models may provide further accuracy in short-term auroral forecasting.