In the vast expanse of space, the solar wind—a stream of charged particles released from the upper atmosphere of the Sun—plays a pivotal role in shaping the heliosphere and interacting with planetary atmospheres. On Earth, these high-speed solar wind streams can wreak havoc on satellite operations and power grids, making accurate forecasting of solar wind speed a critical endeavor. A groundbreaking study led by Dattaraj B. Dhuri from the Center for Astrophysics and Space Science at New York University Abu Dhabi has introduced a novel neural network framework that promises to revolutionize solar wind forecasting, with significant implications for the energy sector.
The research, published in the Astrophysical Journal Supplement Series, presents an encoder–decoder neural network framework designed to forecast the daily averaged solar wind speed for the subsequent four days. This model leverages two different modes of solar observations: historical solar wind data from prior solar rotations and extreme ultraviolet (EUV) coronal observations up to four days prior to the current time. The decoder then outputs the daily averaged solar wind speed from four days prior to the current time to four days into the future.
“The encoder–decoder framework is trained with two different modes of solar observations, which allows us to capture a more comprehensive picture of the solar wind dynamics,” explains Dhuri. “This multimodal approach significantly improves the accuracy of our forecasts, which is crucial for mitigating the impacts of space weather on Earth’s infrastructure.”
The model’s performance is impressive, with root mean square errors (RMSEs) of 55, 58, 58, and 58 km/s and Pearson correlations of 0.78, 0.66, 0.64, and 0.63 for 1 to 4 days in advance, respectively. When tested on unseen data from 2019 to 2023, the model demonstrated robustness with an RMSE of 53 km/s and a Pearson correlation of 0.55 for a four-day advance prediction. These results represent a significant improvement over previous works, paving the way for developing comprehensive multimodal deep learning models for operational solar wind forecasting.
The implications of this research are far-reaching, particularly for the energy sector. Accurate forecasting of solar wind speed can help energy companies anticipate and mitigate the impacts of space weather on power grids and satellite operations. “By providing more reliable forecasts, we can help energy companies make informed decisions to protect their infrastructure and ensure the stability of their operations,” says Dhuri.
This innovative approach not only enhances our understanding of solar-terrestrial interactions but also underscores the potential of neural networks in space weather forecasting. As we continue to advance in this field, the integration of multimodal data and sophisticated machine learning techniques will be key to developing more accurate and reliable forecasting models.
In the words of Dhuri, “This research is a stepping stone towards a future where we can better predict and prepare for the impacts of space weather on our technological infrastructure.” With the encoder–decoder neural network framework, we are one step closer to achieving that goal.