11:55 AM - 12:15 PM
[U12-04] AI Technologies Applied to Space Weather Forecasting
★Invited Papers
Keywords:Space Weather Forecasting, Machine-learning, Solar Flares, Deep Neural Networks
AI technologies, including machine learning, have been applied to space weather forecasting, solar physics, and space science. Machine learning models have already been implemented for forecasting solar flares, and used for estimating solar wind speed, predicting ionospheric disturbances and geomagnetic storms, and increasing the resolution of solar observation data. In addition, recently, machine learning methods that incorporate physical laws and methods for deriving physical laws from data using machine learning have been developed. Therefore, the data science approaches have great potential in space weather forecasting, solar physics, and space science.
Since the launch of the U.S. solar observation satellite in 2010, the large amount of data has been accumulated and made available to the public. The application of machine learning to this data has been active since 2015, when the solar maximum period passed, and the first solar flare prediction models using deep neural networks were published in 2018. Improving the accuracy of solar flare predictions has been a long-standing challenge in space weather forecasting, and we challenged to exceed the accuracy of manual predictions by the application of machine learning techniques to solar observation data. Furthermore, generative AI techniques have been applied to the estimation of the far-side solar magnetograms and to noise and seeing removal techniques for ground-based observations.
We have developed an operational solar flare prediction model using deep neural networks, named Deep Flare Net, which is used in daily space weather forecast meetings. In solar physics, research has been conducted to elucidate the elementary processes of solar flare occurrence mechanisms. On the basis of this knowledge and experience, the model is trained to extract the characteristics of sunspot magnetic fields and X-ray observation data that appear as precursor phenomena before a solar flare from a large amount of data, enabling more accurate prediction. The prediction model consists of three parts. (1) Automatic detection of active regions from solar magnetograms, (2) extraction of 79 physical features from each region, and (3) input of these features into deep neural networks to predict the largest flares that will occur within 24 hours.
Furthermore, the application of AI technology is currently expanding to space weather forecasting around the Earth, such as in the magnetosphere and ionosphere. The challenge is to improve forecast accuracy by successfully combining the advantages of AI forecasting and numerical simulation forecasting. Another challenge is to promote private-sector use of space weather forecasting by combining AI forecasting technology with countermeasures for impact on social infrastructure. In this presentation, examples of applications of machine learning and AI technologies to space weather forecasting will be presented and future issues will be discussed.
Since the launch of the U.S. solar observation satellite in 2010, the large amount of data has been accumulated and made available to the public. The application of machine learning to this data has been active since 2015, when the solar maximum period passed, and the first solar flare prediction models using deep neural networks were published in 2018. Improving the accuracy of solar flare predictions has been a long-standing challenge in space weather forecasting, and we challenged to exceed the accuracy of manual predictions by the application of machine learning techniques to solar observation data. Furthermore, generative AI techniques have been applied to the estimation of the far-side solar magnetograms and to noise and seeing removal techniques for ground-based observations.
We have developed an operational solar flare prediction model using deep neural networks, named Deep Flare Net, which is used in daily space weather forecast meetings. In solar physics, research has been conducted to elucidate the elementary processes of solar flare occurrence mechanisms. On the basis of this knowledge and experience, the model is trained to extract the characteristics of sunspot magnetic fields and X-ray observation data that appear as precursor phenomena before a solar flare from a large amount of data, enabling more accurate prediction. The prediction model consists of three parts. (1) Automatic detection of active regions from solar magnetograms, (2) extraction of 79 physical features from each region, and (3) input of these features into deep neural networks to predict the largest flares that will occur within 24 hours.
Furthermore, the application of AI technology is currently expanding to space weather forecasting around the Earth, such as in the magnetosphere and ionosphere. The challenge is to improve forecast accuracy by successfully combining the advantages of AI forecasting and numerical simulation forecasting. Another challenge is to promote private-sector use of space weather forecasting by combining AI forecasting technology with countermeasures for impact on social infrastructure. In this presentation, examples of applications of machine learning and AI technologies to space weather forecasting will be presented and future issues will be discussed.