EDP Sciences logo
Subscriber Authentication Point
Search
Menu
Home All issues Volume 695 (March 2025) A&A, 695 (2025) A125 Full HTML
Issue
A&A
Volume 695, March 2025
Solar Orbiter First Results (Nominal Mission Phase)
Article Number A125
Number of page(s) 9
Section The Sun and the Heliosphere
DOI https://doi.org/10.1051/0004-6361/202452304
Published online 12 March 2025

© The Authors 2025

Licence Creative CommonsOpen Access article, published by EDP Sciences, under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This article is published in open access under the Subscribe to Open model. Subscribe to A&A to support open access publication.

1. Introduction

Solar Orbiter (Müller et al. 2020) is a spacecraft orbiting the Sun that can capture images of the Sun in closer proximity than any previous spacecraft. Additionally, it enables the exploration of the Sun’s uncharted polar regions for the first time. The Solar Orbiter/Extreme Ultraviolet Imagers (EUI; Rochus et al. 2020) include the Full Sun Imager (FSI), which observes the entire solar disk. Since the FSI only captures two channels (174 and 304 Å), this poses challenges for performing differential emission measure (DEM) analysis. Another instrument for observing extreme UV (EUV), the Solar Dynamics Observatory (SDO; Pesnell et al. 2012)/Atmospheric Imaging Assembly (AIA; Lemen et al. 2012), which has been in operation since May 2010, captures seven EUV (94, 131, 171, 193, 211, 304, and 335 Å) images from four telescopes. This instrument has two channels (171 and 304 Å) similar to those of the Solar Orbiter/EUI/FSI.

Differential emission measure is a technique used to quantify the density and temperature of emitting multithermal plasma in the solar atmosphere. The main peaks of the DEM profile represent the dominant plasma temperatures, and the width of the peak indicates whether the plasma is close to isothermal or multithermal at those temperatures (Schmelz et al. 2011). DEM analysis is a complex and ill-posed mathematical problem that requires data from multiple channels to accurately determine the plasma characteristics. The DEM equation is defined as

D N i = ∫ 0 ∞ K i ( T ij ) DEM ( T j ) d T $$ \begin{aligned} DN_i = \int _{0}^{\infty } K_i(T_{ij}) \text{ DEM}(T_j) \ dT \end{aligned} $$(1)

where DNi (i = 1,..., N) is the observed digital number of the i-th channel. In this study, we used six channels to calculate the DEMs (N = 6); Kij is the response function of a specific instrument, which indicates its sensitivity to plasma that is emitting at different temperature bins, j. More information about the response function can be found in Boerner et al. (2012). To effectively calculate the DEMs of the solar atmosphere, a combination of various EUV observations is essential. Therefore, numerous studies have utilized data from instruments like SDO/AIA for DEM analyses (e.g. Lee et al. 2017; Mandal et al. 2024). There are various methods for calculating DEMs (Hannah & Kontar 2012; Cheung et al. 2015; Morgan & Pickering 2019). Our analysis employs the method developed by Hannah & Kontar (2012), which uses a regularization technique to address uncertainties inherent to DEM analysis and thus ensures more reliable results.

Recently, significant progress has been made in applying deep learning to image-to-image translation for solar observations. For example, solar (E)UV images have been successfully translated into magnetograms (e.g., Kim et al. 2019; Jeong et al. 2020, 2022). In contrast, Park et al. (2019) generated (E)UV images from magnetograms. Lee et al. (2021) transformed Galileo’s historical sunspot sketches into modern EUV images, and Son et al. (2021) generated He I images from EUV images, effectively mitigating the limitations of ground-based observations through the use of satellite observations. Additionally, Salvatelli et al. (2019) demonstrated the capability to translate between different EUV image channels, and Lim et al. (2021) suggested how to select specific EUV channels to improve the translation performance among EUV channels. On the other hand, Salvatelli et al. (2022) explored the limitations of synthetic EUV images generated via deep learning, that is, the translation performance declines during extreme events such as solar flares. More recently, Park et al. (2023) made a pixel-to-pixel translation to generate EUV images and demonstrated that AI-generated data can be successfully used for DEM calculations.

In this study, we aim to address whether we can properly determine DEMs from Solar Orbiter/EUI/FSI using AI-generated data. To do this we generated five EUV channel datasets from FSI 174 and 304 Å using deep learning models. The paper is organized as follows. In Sect. 2, we describe the structure of our deep learning model and explain how we tested, trained, and applied it. We also explain how we applied the DEM method to the data. In Sect. 3, we describe the data and calibration of the Solar Orbiter/EUI/FSI and the SDO/AIA datasets. In Sect. 4, we present the results of the deep learning models and DEMs. A brief summary and conclusion are given in Sect. 5.

2. Methods

Here we explain the overall process shown in Figure 1. To perform this study, we first trained our deep learning models for SDO/AIA 171 and 304 Å to generate each channel (94, 131, 193, 211, and 335 Å). Next, we applied the trained models to EUI/FSI 174 and 304 Å data. Lastly, we determined the DEMs from FSI 174 Å data and AI-generated five-channel data.

thumbnail Fig. 1.

Overview of our study. (a) Training process. This image is reproduced from Jeong et al. (2022). The generator (G) takes input images (AIA 171 and AIA 304 Å) and translates them into new images. The red box at the boundary of the images indicates the generator images. The discriminator (D) distinguishes between real pairs and fake pairs. The inspector (I) guides the generator by computing concordance CC values, ensuring the generated images are not only realistic but also scientifically accurate according to the data relationships. To utilize this model, we first trained five models for each channel (94, 131, 193, 211, and 335 Å) using only the AIA observed dataset. (b) Applying process. After we successfully trained the models, we applied the generative models to the FSI observed dataset. The dashed blue line shows the results when a deep learning model was applied to the FSI dataset. Finally, the observed FSI 174 Å with AI-generated sets could determine the DEMs.

2.1. Deep learning model

We used the Pix2PixCC model created by Jeong et al. (2022), the structure of which is shown in Figure 1a. This model is an improved version of Pix2PixHD (Wang et al. 2018). It combines the original loss functions with a concordance correlation coefficient (CC) (Lin 1989) loss to enhance performance. We first trained the models to generate data for five channels (94, 131, 193, 211, and 335 Å) from AIA 171 and AIA 304 Å. Then, we applied the trained models to the FSI 174 and 304 Å channels to generate the five-channel EUV data, which correspond to synthetic EUI/FSI data to be used as the input for DEM analysis.

2.2. Differential emission measure

For the present work, we used the DEM code that was initially developed by Hannah & Kontar (2012) with Python codes1. This method provides error bars for both the temperature and the DEM value at that temperature. We computed the DEMs within a temperature range of 0.1 MK to 10 MK. First, we created AI-generated five-channel FSI data from FSI 174 and 304 Å pairs using deep learning models. Then we calculated the DEMs from FSI 174 Å and the five-channel datasets (Figure 1b). Here we did not use FSI 304 Å to calculate the DEMs because its He II 304 Å line is optically thick (Dolliou et al. 2023).

We used the temperature response functions calculated for both FSI and AIA using the CHIANTI database version 10.1 (Dere et al. 2023), and these were calculated for a density of n = 109 cm−3, a value typical for coronal loops. The calculated temperature response functions of AIA 171 Å and FSI 174 Å have similar profiles but different peak values; the response function of AIA 171 Å tends to be slightly lower than that of FSI 174 Å. Based on Boerner et al. (2012), the response function is given by a product of the effective area and gain of the Charge-Coupled Device (CCD) camera system. The effective area, Aeff, provides information about the efficiency of the telescope optics, and the effective area is given by combinations of the geometrical collecting area, reflectance, transmission efficiency, quantum efficiency, and an additional correction. Due to different optical structures, the temperature response functions of SDO/AIA and Solar Orbiter/EUI/FSI are different. To compensate for this difference, we multiplied the FSI 174 Å temperature response function by 0.7, which corresponds to the ratio of the maximum values of both functions across the two instruments. Wright et al. (2017) employed a similar method across two different instruments (Hinode/XRT and SDO/AIA).

3. Data

3.1. SDO/AIA datasets

We used Level-1 SDO/AIA data from JSOC Data Export2. The data we used were taken at 00:00 UT from each day from January 2011 to December 2021, covering approximately one solar cycle. We discarded datasets with a QUALITY header value that was not zero to ensure the quality of the data. Then, we rotated the images to align the solar axis to the north, located the solar disk at the center of the image, and divided it by the exposure time. We also calibrated the degradation of the AIA imagers to set all the median values to 2011 January 01, 00:00 UT (Ugarte-Urra et al. 2015; Jeong et al. 2022). We took the logarithm base 2 and normalized it to a range of [−1, 1] to provide input for the deep learning models. The original image resolution was 4096 × 4096 pixels, which we resized to 1024 × 1024 pixels. We set the solar disk radius to 400 pixels, which is discussed in the next Sect. 3.2. We divided the datasets into training and testing sets, and considered the solar inclination and the elliptic orbit while maintaining a two-week gap between the two sets to prevent overlap. We selected 11 months for the training datasets and 1 month for the testing datasets, shifting the test month forward each time. For example, in the 2011 datasets, January was used for the testing dataset and the remaining months for the training datasets. In 2012, February was used for the testing and the rest for the training.

3.2. Solar Orbiter/EUI/FSI datasets

We used Level-1 Solar Orbiter/EUI/FSI images after January 2022 from EUI data Release 6.03. The FSI was tasked with capturing EUV images of the entire solar disk at 174 Å and 304 Å. The original image resolution of FSI was 3072 × 3072. We removed the last 32 pixels since they were bad pixels. Due to the Solar Orbiter’s elliptical orbit around the Sun, the apparent size of the solar disk in FSI images varied significantly throughout the mission. For example, at a distance of 1.01 AU, the solar disk had a radius of 213 pixels; at 0.29 AU, it had a radius of 739 pixels. To standardize input data for our deep learning models, which require consistency in radius pixels, we preprocessed all the images to have a reference radius of 400 pixels. For example, images with radii pixels larger than 400 pixels were resized, while those with smaller radii were cropped and then upscaled to maintain a consistent size. In addition, to calibrate the FSI degradation and the effect of its orbit, we set all the median values of the data to those at 00:00 UT on 2011 January 01, as used for the AIA data.

3.3. Intercalibration between AIA and FSI for deep learning

Since we trained our model using the AIA dataset, the application dataset (Solar Orbiter observations) should have been similar to the AIA dataset. However, there was a relative difference in brightness due to the optical differences between the two instruments: AIA has two mirrors while EUI/FSI has one. This led to slightly different temperature response functions and resulted in different contrasts in the images (Müller et al. 2020). Jarolim et al. (2024) also mentioned the same problem, noting that the two instruments need an intercalibration process. To correct the small discrepancies between AIA and FSI, we first calibrated the data to have the same median and then used third-order fitting to match the FSI 174 Å data with the AIA 171 Å data and the FSI 304 Å data with the AIA 304 Å data. To intercalibrate between the images from the two instruments, we drew a scatter plot of digital number (DN) within the range of ±45° longitude and latitude on the heliographic map, when the two instruments were at the inferior conjunction (at 08:30 UT on 2022 March 07 and at 21:30 UT on 2023 March 28, considering the time delay of the light). Further details can be found in Appendix A.

4. Result and discussion

4.1. Generation of EUV images

We generated the SDO/AIA 94, 131, 193, 211, and 335 Å images from the SDO/AIA 171 Å and AIA 304 Å test datasets. Figure 2 shows a comparison of AIA observations and AI-generated ones on 2012 February 26. The comparison indicates that deep learning models successfully generate various coronal features on the solar disk. For instance, the shapes of coronal holes are similar. Additionally, the structures of active regions in the solar atmosphere are quite similar in both the observed and AI-generated images, suggesting that the AI models effectively generate the characteristics of active solar regions.

thumbnail Fig. 2.

Comparison of observed and AI-generated AIA images at 00:00 UT on 2012 February 26. The top row shows observed AIA images in the 94, 131, 193, 211, and 335 Å channels. The bottom row shows AI-generated images from AIA 171 and 304 Å for the same channels. We note that the same conditions for each channel are used when plotting the images.

Table 1 shows the average metrics for the test datasets. We calculated the Pearson CC, the root mean square error (RMSE), and the normalized RMSE (NRMSE) between the observed data and the AI-generated data within one solar radius. In calculating these metrics, we restored the normalization and logarithm processes performed during the preprocessing of the dataset and calculated the comparison without any binning. The Pearson CC values between the AIA data and our deep learning models for the 94, 131, 193, 211, and 335 Å channels are 0.87, 0.96, 0.95, 0.95, and 0.93, respectively. The corresponding RMSE values are 0.50, 1.26, 38.96, 13.55, and 0.84. The NRMSE values are 0.012, 0.008, 0.012, 0.010, and 0.011. According to the metrics, the model that generates 131 Å yields the best performance, while 94 Å performs the worst. The 94 Å channel was intended to observe primarily the Fe XVIII at ∼7 MK, while the input channels, 171 and 304 Å, observe relatively lower temperatures of ∼7 × 105 K and ∼5 × 104 K, respectively (O’Dwyer et al. 2010; Viall et al. 2020). In addition, the 94 Å images are relatively noisy (Park et al. 2023).

Table 1.

Averaged metric values of SDO/AIA observed data and AI-generated data for the test dataset.

Figure 3 shows the comparison between observed AIA images and AI-generated ones from FSI 174 and 304 Å when the two instruments, SDO and Solar Orbiter, were in inferior conjunction (at 08:38 UT on 2022 March 07). At this moment, the Solar Orbiter was closer to the Sun than the SDO, at a distance of 0.5 AU. The overall structures are quite similar to each other. However, we found small discrepancies in the contrast, particularly in extremely dark or bright regions (e.g., coronal holes and active regions). To our knowledge, the response functions between AIA channels and EUI channels are slightly different (F. Auchère, private communication). These differences arise because the structures of the instruments are quite different, as mentioned in Sect. 3.3.

thumbnail Fig. 3.

Comparison of observed images at 08:38 UT on 2022 March 07 and AI-Generated AIA images at 08:33 UT. The top row shows observed AIA images in the 94, 131, 193, 211, and 335 Å channels. The bottom row shows AI-generated images for the same channels, using input from third-order fitted FSI 174 and 304 Å data. We note that the same conditions are used in Figure 2.

4.2. Comparison of differential emission measures

Now we can compare the DEMs from SDO and those from Solar Orbiter/EUI/FSI 174 Å together with AI-generated ones, when the Solar Orbiter and the SDO were in inferior conjunction (at 08:38 UT on 2022 March 07). This comparison could validate the hypothesis that our models successfully generate five channels from FSI 174 and 304 Å. To compare the DEMs between these two datasets, we selected the same coronal loop shown in Figure 4a. Before comparing the two datasets, we restored the dynamic scale, which ranges from [0, 214 − 1]. We calculated the DEMs for each pixel within the contour of the region of interest (ROI) and then averaged them. Figure 4b illustrates the positions of the two spacecraft, with the plot showing the Solar Orbiter at 0.5 AU and in conjunction with SDO. Figure 4c shows the comparison between the DEMs determined from FSI 174 Å with the AI-generated dataset and the DEM determined from the AIA dataset. In determining the DEM for the Solar Orbiter, we used FSI 174 Å data that had been calibrated for degradation by only the median value (not third-order-fitted) and used the temperature response function of FSI 174 Å (see Sect. 3.3). The results of the two DEMs in Figure 4c are mostly consistent with each other. The temperature of the main peak in both DEMs is approximately log T = 6.0, and the DEM values at the main peak are within the error bar range. This similarity shows that the AI-generated images using our deep learning model can reproduce the DEMs well.

thumbnail Fig. 4.

Results of the DEM calculation at 08:33 UT on 2022 March 07. (a) ROI for the DEM calculations. The red box in the first column of images marks the region shown in the remaining column of images. The white contour in the remaining column highlights the specific ROI. (b) Positions of the Solar Orbiter (purple dot) and SDO (green dot). (c) Comparison of DEM results from FSI 174 Å + AI-generated (purple line) images and AIA images. The vertical error bars represent the uncertainties of the DEM values, while the horizontal error bars indicate the temperature resolution.

In a similar way, we determined the DEMs from two datasets when the Solar Orbiter was at different vantage points (at 00:00 UT on 2023 April 04), and the results are shown in Figure 5. Figure 5b illustrates the positions of the Solar Orbiter and the SDO when the Solar Orbiter is positioned at 0.33 AU and located 25 degrees ahead of the SDO. We selected the ROI from AIA and calculated it based on the solar coordinate system; then we used the calculated ROI in the FSI dataset. These different viewing angles might cause uncertainty in higher coronal structures, so we directed the ROI toward a low-height coronal structure. Figure 5a shows one of these coronal structures, specifically an active region that was growing. Figure 5c presents the results of the DEMs. The main peak temperature in both DEMs is almost consistent at around log T = 6.2, and the widths of the DEMs are quite similar. These facts suggest that our method can successfully determine DEMs even if the two observations are from different vantage points as well as from different datasets.

thumbnail Fig. 5.

Results of the DEM calculation at 00:00 UT on 2023 October 16. The layout of this figure is the same as in Figure 4.

5. Summary and conclusion

In this study, we have demonstrated that we can properly determine the DEM from Solar Orbiter/EUI/FSI using deep learning, which can be a great extension of the coronal structures of the FSI. Using the Pix2PixCC model, we successfully generated five-EUV-channel data (94, 131, 193, 211, and 335 Å) from 174 and 304 Å observations of the Solar Orbiter/EUI/FSI datasets. We then determined the DEMs from FSI 174 Å together with AI-generated data for two cases. These DEMs are very consistent with those from the observed AIA dataset. This suggests that deep learning models successfully decompose the information of FSI 174 and 304 Å into each of the five channels.

In order to achieve more accurate results, further research on intercalibration techniques between the two instruments (SDO/AIA and Solar Orbiter/EUI/FSI) is essential. As the Solar Orbiter mission progresses, these intercalibration methods can be refined and improved, thereby improving the determination of DEMs.

While selecting the ROIs, calculations based on solar coordinates can introduce errors into the analysis because the morphology of a coronal loop is three-dimensional. Therefore, a more accurate method is needed to compare the DEMs using the data from the two instruments. For instance, Aschwanden et al. (2008b) performed triangulation and aligned the same loop using data from two spacecraft (STEREO A and B).

We plan to examine the DEM results observed from the same coronal loop when viewed stereoscopically. Analyzing the variations in the line-of-sight density and temperature distribution when observed from different angles will contribute to a more comprehensive understanding of the solar corona’s three-dimensional structure (Aschwanden et al. 2008a). Additionally, this research could produce a global temperature map of the Sun.

We think our study can provide benefits for future missions at Lagrangian points L4 (Cho et al. 2023; Lee et al. 2024; Moon et al. 2024) and L5 (Vourlidas 2015). Since these missions cannot carry various channel imagers due to weight constraints, the ability to determine DEMs using fewer channels would be advantageous. Additionally, such missions would allow for long-term stereoscopic studies of the coronal structure from wide angles and enable researchers to determine coronal temperature changes from L5, via Earth, to L4.

Acknowledgments

We deeply appreciate the anonymous reviewers and editors for their thoughtful comments and detailed feedback, which have substantially contributed to improving the overall quality of this manuscript. Their perceptive suggestions have greatly strengthened both the clarity and the scientific depth of our work. This work was supported by the BK21 FOUR program through National Research Foundation of Korea (NRF) under Ministry of Education (MoE) (Kyung Hee University, Human Education Team for the Next Generation of Space Exploration), the Korea Astronomy and Space Science Institute under the R&D program (Project No. 2024-1-850-12) supervised by the Ministry of Science and ICT (MSIT), Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korean government MSIT (No. RS-2023-00234488, Development of solar synoptic magnetograms using deep learning, 15%), and Basic Science Research Program through the NRF funded by the MoE (NRF-2021R1I1A1A01049615, RS-2023-00248916, and RS-2024-00337363). Also, this research is partially funded by the BK21 FOUR program of Graduate School, Kyung Hee University (GS-1-JO-NON-20242364). CHIANTI is a collaborative project involving George Mason University, the University of Michigan (USA), University of Cambridge (UK) and NASA Goddard Space Flight Center (USA). This study uses SDO and Solar Orbiter data. We acknowledge the Joint Science Operations Center (JSOC) for providing SDO/AIA data and Solar Influences Data Analysis Center (SIDC) for providing Solar Orbiter/EUI data. This work was supported by the use of PyTorch (Paszke et al. 2019), Astropy (Astropy Collaboration 2013), aiapy (Barnes et al. 2020), SunPy (Barnes et al. 2023), and NumPy (Harris et al. 2020), which provided essential tools for deep learning development and solar data preprocessing.

References

  1. Aschwanden, M. J., Nitta, N. V., Wuelser, J.-P., & Lemen, J. R. 2008a, ApJ, 680, 1477 [NASA ADS] [CrossRef] [Google Scholar]
  2. Aschwanden, M. J., Wülser, J.-P., Nitta, N. V., & Lemen, J. R. 2008b, ApJ, 679, 827 [NASA ADS] [CrossRef] [Google Scholar]
  3. Astropy Collaboration (Robitaille, T. P., et al.) 2013, A&A, 558, A33 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  4. Barnes, W., Cheung, M., Bobra, M., et al. 2020, J. Open Source Software, 5, 2801 [CrossRef] [Google Scholar]
  5. Barnes, W. T., Christe, S., Freij, N., et al. 2023, Front. Astron. Space Sci., 10, 1076726 [NASA ADS] [CrossRef] [Google Scholar]
  6. Boerner, P., Edwards, C., Lemen, J., et al. 2012, Sol. Phys., 275, 41 [Google Scholar]
  7. Cheung, M. C. M., Boerner, P., Schrijver, C. J., et al. 2015, ApJ, 807, 143 [Google Scholar]
  8. Cho, K.-S., Hwang, J., Han, J.-Y., et al. 2023, J. Korean Astron. Soc., 56, 263 [NASA ADS] [Google Scholar]
  9. Dere, K. P., Del Zanna, G., Young, P. R., & Landi, E. 2023, ApJS, 268, 52 [NASA ADS] [CrossRef] [Google Scholar]
  10. Dolliou, A., Parenti, S., Auchère, F., et al. 2023, A&A, 671, A64 (SO Nominal Mission Phase SI) [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  11. Hannah, I. G., & Kontar, E. P. 2012, A&A, 539, A146 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  12. Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357 [Google Scholar]
  13. Jarolim, R., Tremblay, B., Muñoz-Jaramillo, A., et al. 2024, ApJ, 961, L31 [NASA ADS] [Google Scholar]
  14. Jeong, H.-J., Moon, Y.-J., Park, E., & Lee, H. 2020, ApJ, 903, L25 [Google Scholar]
  15. Jeong, H.-J., Moon, Y.-J., Park, E., Lee, H., & Baek, J.-H. 2022, ApJS, 262, 50 [CrossRef] [Google Scholar]
  16. Kim, T., Park, E., Lee, H., et al. 2019, Nat. Astron., 3, 397 [Google Scholar]
  17. Lee, J.-Y., Raymond, J. C., Reeves, K. K., Moon, Y.-J., & Kim, K.-S. 2017, ApJ, 844, 3 [NASA ADS] [Google Scholar]
  18. Lee, H., Park, E., & Moon, Y.-J. 2021, ApJ, 907, 118 [NASA ADS] [Google Scholar]
  19. Lee, D.-Y., Kim, R.-S., Choi, K.-E., et al. 2024, J. Astron. Space Sci., 41, 1 [NASA ADS] [CrossRef] [Google Scholar]
  20. Lemen, J. R., Title, A. M., Akin, D. J., et al. 2012, Sol. Phys., 275, 17 [Google Scholar]
  21. Lin, L. I-K. 1989, Biometrics, 45, 255 [CrossRef] [Google Scholar]
  22. Lim, D., Moon, Y.-J., Park, E., & Lee, J.-Y. 2021, ApJ, 915, L31 [NASA ADS] [CrossRef] [Google Scholar]
  23. Mandal, S., Peter, H., Klimchuk, J. A., et al. 2024, A&A, 682, L9 (SO Nominal Mission Phase SI) [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  24. Moon, Y.-J., Cho, K.-S., Park, S.-H., et al. 2024, J. Korean Astron. Soc., 57, 35 [NASA ADS] [Google Scholar]
  25. Morgan, H., & Pickering, J. 2019, Sol. Phys., 294, 135 [NASA ADS] [CrossRef] [Google Scholar]
  26. Müller, D., St. Cyr, O. C., Zouganelis, I., et al. 2020, A&A, 642, A1 [Google Scholar]
  27. O’Dwyer, B., Del Zanna, G., Mason, H. E., Weber, M. A., & Tripathi, D. 2010, A&A, 521, A21 [Google Scholar]
  28. Park, E., Moon, Y.-J., Lee, J.-Y., et al. 2019, ApJ, 884, L23 [NASA ADS] [CrossRef] [Google Scholar]
  29. Park, E., Lee, H., Moon, Y.-J., et al. 2023, ApJS, 264, 33 [CrossRef] [Google Scholar]
  30. Paszke, A., Gross, S., Massa, F., et al. 2019, arXiv e-prints [arXiv:1912.01703] [Google Scholar]
  31. Pesnell, W. D., Thompson, B. J., & Chamberlin, P. C. 2012, Sol. Phys., 275, 3 [Google Scholar]
  32. Rochus, P., Auchère, F., Berghmans, D., et al. 2020, A&A, 642, A8 [NASA ADS] [CrossRef] [EDP Sciences] [Google Scholar]
  33. Salvatelli, V., Bose, S., Neuberg, B., et al. 2019, arXiv e-prints [arXiv:1911.04006] [Google Scholar]
  34. Salvatelli, V., dos Santos, L. F. G., Bose, S., et al. 2022, ApJ, 937, 100 [NASA ADS] [CrossRef] [Google Scholar]
  35. Schmelz, J. T., Worley, B. T., Anderson, D. J., et al. 2011, ApJ, 739, 33 [NASA ADS] [CrossRef] [Google Scholar]
  36. Son, J., Cha, J., Moon, Y.-J., et al. 2021, ApJ, 920, 101 [NASA ADS] [CrossRef] [Google Scholar]
  37. Ugarte-Urra, I., Upton, L., Warren, H. P., & Hathaway, D. H. 2015, ApJ, 815, 90 [NASA ADS] [CrossRef] [Google Scholar]
  38. Viall, N. M., Kucera, T. A., & Karpen, J. T. 2020, ApJ, 905, 15 [NASA ADS] [CrossRef] [Google Scholar]
  39. Vourlidas, A. 2015, Space Weather, 13, 197 [NASA ADS] [CrossRef] [Google Scholar]
  40. Wang, T. C., Liu, M. Y., Zhu, J. Y., et al. 2018, in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 8798 [CrossRef] [Google Scholar]
  41. Wright, P. J., Hannah, I. G., Grefenstette, B. W., et al. 2017, ApJ, 844, 132 [NASA ADS] [CrossRef] [Google Scholar]

Appendix A: Intercalibration between AIA and FSI

Here, we reveal more information about results of our intercalibrations mentioned in Section 3.3. For the intercalibration between two different datasets, at the top row in Figure A.1 we present scatter plots of DN/s within the range of ±45° longitude and latitude of the heliographic map, when the two instruments are at the inferior conjunction (2022-03-07 08:30 UT and 2023-03-28 21:30 UT). As a result of the intercalibration, the standard deviation differences between the original data of AIA 304 Å and EUI 304 Å decreases from 19.32 to 16.04 when we apply the third-order fitting calibration. Similarly, the standard deviation differences between AIA 171 Å and the FSI 174 Å decreases from 105.21 to 85.23. Moreover, these intercalibrations could gradually improve as the mission progresses. We present these results in the middle and bottom row in Figure A.1.

Appendix B: Additional Figures

Here, we present additional cases of DEM results. Figure B1 highlights strands of coronal features, while Figure B2 highlights the outer coronal features. Both figures reveal the usefulness of our method in accurately determining DEMs for various solar structures, including off-disk features. The temperature of the main peak in Figure B1 is consistent at approximately log T = 6.2. Similarly, the temperature of the main peak in Figure B2 is also consistent at approximately log T = 6.2.

thumbnail Fig. B.1.

Scatter plots and histograms between two datasets. Each column shows the intercalibration between different channel of AIA and EUI. The solid lines in the scatter plots correspond to the third-order fitting, passing through zero, representing the relationship between FSI and AIA. Here, the x-axis corresponds to the DN/s of FSI observations and the y-axis corresponds to the DN/s of AIA observations. The histograms represent the distribution of differences between the original data and differences after third-order fitting. For both wavelengths, the standard deviations improve after intercalibration.

thumbnail Fig. B.2.

The DEM determination of the strands of coronal features at 08:40 UT on 2024-03-20, when the two instruments were observing the same side of the disk. (a), (b), (c) layout of this figure is the same as in Figure 4.

thumbnail Fig. B.3.

The DEM determination of outer coronal features at 21:30 UT on 2023-03-28, when the two instruments were observing the same side of the disk. (a), (b), (c) layout of this figure is the same as in Figure 4.

All Tables

Table 1.

Averaged metric values of SDO/AIA observed data and AI-generated data for the test dataset.

In the text

All Figures

thumbnail Fig. 1.

Overview of our study. (a) Training process. This image is reproduced from Jeong et al. (2022). The generator (G) takes input images (AIA 171 and AIA 304 Å) and translates them into new images. The red box at the boundary of the images indicates the generator images. The discriminator (D) distinguishes between real pairs and fake pairs. The inspector (I) guides the generator by computing concordance CC values, ensuring the generated images are not only realistic but also scientifically accurate according to the data relationships. To utilize this model, we first trained five models for each channel (94, 131, 193, 211, and 335 Å) using only the AIA observed dataset. (b) Applying process. After we successfully trained the models, we applied the generative models to the FSI observed dataset. The dashed blue line shows the results when a deep learning model was applied to the FSI dataset. Finally, the observed FSI 174 Å with AI-generated sets could determine the DEMs.
In the text
thumbnail Fig. 2.

Comparison of observed and AI-generated AIA images at 00:00 UT on 2012 February 26. The top row shows observed AIA images in the 94, 131, 193, 211, and 335 Å channels. The bottom row shows AI-generated images from AIA 171 and 304 Å for the same channels. We note that the same conditions for each channel are used when plotting the images.
In the text
thumbnail Fig. 3.

Comparison of observed images at 08:38 UT on 2022 March 07 and AI-Generated AIA images at 08:33 UT. The top row shows observed AIA images in the 94, 131, 193, 211, and 335 Å channels. The bottom row shows AI-generated images for the same channels, using input from third-order fitted FSI 174 and 304 Å data. We note that the same conditions are used in Figure 2.
In the text
thumbnail Fig. 4.

Results of the DEM calculation at 08:33 UT on 2022 March 07. (a) ROI for the DEM calculations. The red box in the first column of images marks the region shown in the remaining column of images. The white contour in the remaining column highlights the specific ROI. (b) Positions of the Solar Orbiter (purple dot) and SDO (green dot). (c) Comparison of DEM results from FSI 174 Å + AI-generated (purple line) images and AIA images. The vertical error bars represent the uncertainties of the DEM values, while the horizontal error bars indicate the temperature resolution.
In the text
thumbnail Fig. 5.

Results of the DEM calculation at 00:00 UT on 2023 October 16. The layout of this figure is the same as in Figure 4.
In the text
thumbnail Fig. B.1.

Scatter plots and histograms between two datasets. Each column shows the intercalibration between different channel of AIA and EUI. The solid lines in the scatter plots correspond to the third-order fitting, passing through zero, representing the relationship between FSI and AIA. Here, the x-axis corresponds to the DN/s of FSI observations and the y-axis corresponds to the DN/s of AIA observations. The histograms represent the distribution of differences between the original data and differences after third-order fitting. For both wavelengths, the standard deviations improve after intercalibration.
In the text
thumbnail Fig. B.2.

The DEM determination of the strands of coronal features at 08:40 UT on 2024-03-20, when the two instruments were observing the same side of the disk. (a), (b), (c) layout of this figure is the same as in Figure 4.
In the text
thumbnail Fig. B.3.

The DEM determination of outer coronal features at 21:30 UT on 2023-03-28, when the two instruments were observing the same side of the disk. (a), (b), (c) layout of this figure is the same as in Figure 4.
In the text

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.