Reconstructing He <scp>i</scp> 10830 Å Images Using H<i>α</i> Images through Deep Learning

Abstract He i 10830 Å, as an important spectrum line to diagnose the solar chromosphere and corona, has had consistent observations within the past two decades. This study aims to reconstruct synthetic He i 10830 Å images, addressing the limited availability of historical data compared to the extensive record of H α images spanning over a century. To achieve this, we generate He i 10830 Å images from H α using a deep learning method, pix2pixHD. For model development, we use He i 10830 Å images from the National Solar Observatory (NSO)/Synoptic Optical Long-term Investigations of the Sun and H α images from NSO/Global Oscillation Network Group. Our model achieves a high correlation coefficient (CC) of 0.867 to reconstruct full-disk He i 10830 Å images. For solar structures like active regions, nonpolar, and polar crown filaments, we can achieve CCs of 0.903, 0.844, and 0.871, respectively. The model also shows reasonable performance on coronal holes with a CC of 0.536. Moreover, the model effectively generalized to data from multiple observatories, producing reliable results. In the early 2000s, when He i 10830 Å data was scarce, our model successfully reconstructed a scenario of an X-class flare eruption in the He i 10830 Å line covering the full observing period. This reconstruction included the formation of dark flare ribbons during the flare and postflare phases, showing a strong match with the postflare scenario observed by the Mauna Loa Solar Observatory/Chromospheric Helium Imaging Photometer.