5:15 PM - 6:45 PM
[PEM11-P08] Improvement of solar wind prediction model using segmentation images of coronal holes
Keywords:Solar wind, Coronal hole, Deep learning, Space Weather
The solar wind is a stream of charged particles released from the solar corona into the interplanetary space. Coronal holes, observed as dark regions in the extreme ultraviolet (EUV) images, are thought to be the source of the solar wind. Currently, the solar wind prediction is studied to reduce space weather hazards caused by geomagnetic storms and so on. Hemapriya et al. (2021) has proposed a solar wind prediction model using deep learning techniques and EUV images, and it achieved the prediction of solar wind speed with RMSE of 76.3±1.87[km/s] and the correlation coefficient of 0.57±0.02. However, the solar wind prediction is difficult because observation data that clearly captured high speed solar wind streams from a coronal hole are not enough. Hence, the prediction model should learn the relationships between input parameters and solar wind speed efficiently. Furthermore, structures other than coronal holes are also captured in EUV images, which can make learning process more difficult.
Here, we aim to improve the solar wind prediction model with a limited number of observation data by using segmentation images of coronal holes in addition to EUV images. We use two types of segmentation images of coronal holes: the first one obtained by an image processing method, named CHIMERA, and the second one obtained by a deep learning model, which detected brighter coronal holes than those by CHIMERA. As in the previous study, we use the daily averaged solar wind speed measured at the L1 point. The data set is in the period of 2015 - 2018.
First, we extract some coronal hole features from segmentation images and input them into a multilayer perceptron (MLP) model to predict solar wind speeds. The model achieves RMSE of 80.9±1.11 [km/s] with segmentation images by CHIMERA, while 78.7±0.94 [km/s] with segmentation images by a deep learning technique.
Second, we use convolutional neural networks (CNN) to predict the solar wind speed using coronal hole segmentation images. As a result, the model achieves RMSE of 68.8±1.71 [km/s] with coronal hole segmentation images by a deep learning technique, which is better than the previous studies.
In summary, our prediction model can learn the relationships between solar EUV images and the solar wind speed efficiently with the help of segmentation images obtained by a deep learning model. We also find that brighter coronal holes can contribute to the solar wind speed prediction as well as dark coronal holes.
Here, we aim to improve the solar wind prediction model with a limited number of observation data by using segmentation images of coronal holes in addition to EUV images. We use two types of segmentation images of coronal holes: the first one obtained by an image processing method, named CHIMERA, and the second one obtained by a deep learning model, which detected brighter coronal holes than those by CHIMERA. As in the previous study, we use the daily averaged solar wind speed measured at the L1 point. The data set is in the period of 2015 - 2018.
First, we extract some coronal hole features from segmentation images and input them into a multilayer perceptron (MLP) model to predict solar wind speeds. The model achieves RMSE of 80.9±1.11 [km/s] with segmentation images by CHIMERA, while 78.7±0.94 [km/s] with segmentation images by a deep learning technique.
Second, we use convolutional neural networks (CNN) to predict the solar wind speed using coronal hole segmentation images. As a result, the model achieves RMSE of 68.8±1.71 [km/s] with coronal hole segmentation images by a deep learning technique, which is better than the previous studies.
In summary, our prediction model can learn the relationships between solar EUV images and the solar wind speed efficiently with the help of segmentation images obtained by a deep learning model. We also find that brighter coronal holes can contribute to the solar wind speed prediction as well as dark coronal holes.