Enhancing Solar Driver Forecasting with Multivariate Transformers

When Predicting future geomagnetic and solar storms charged particles from flares or CMEs reach Earth, and evaluating their potential impacts requires accurate atmospheric heating and transient solar wind activity solar driver forecasts. To assess forecast performance increase, sometimes resulting in geomagnetic and for a given prediction framework, Space Environment solar storms. With a history of such storms disrupting Technologies (SET) provides a benchmarking communications and power systems and significantly dataset using an archived data set spanning 6 increasing atmospheric drag for Low Earth Orbit years and 15,000 forecasts across Solar Cycle 24 [6]. In (LEO) satellites, accurate space weather activity this work, we employ a multivariate approach using forecasting presents a critical enabling technology for a transformer deep neural network to learn the mapping mitigating space weather-induced outages and satellite from historical solar drivers to future drivers, conjunction risk [1].