ITI for the Sun: Improved intercalibration of multi-instrument heliophysics data series with Instrument-To-Instrument translation
Abstract
In solar physics, the study of long-term evolution typically exceeds the lifetime of single instruments. Data-driven approaches are limited in terms of homogeneous historical data samples. We demonstrate how machine learning enabled Instrument-To-Instrument translation (ITI; Jarolim et al. 2022) can leverage recent instrumental improvements and provide a so far unused resource to foster novel research and accelerate data-driven solar research.In this study, we aim at providing a uniform data series of EUV observations from SDO/AIA, STEREO/EUVI and SOHO/EIT. The application of unpaired image-to-image translation methods to standard reduced SDO/AIA data shows a sensitivity for insufficiently corrected device degradation, leading to differences between the calibrated series. Here, we apply the auto-calibration from Dos Santos et al. (2021), to obtain a more consistent calibration for the SDO/AIA series and use the ITI framework to translate observations from STEREO/EUVI and SOHO/EIT to the same domain. We demonstrate that with this adjustment we can achieve a better calibration between the three instruments. Comparisons of aligned observations demonstrate high perceptual quality and a strong similarity to reference observations.The resulting data series covers uniform observations dating back to 1996, including simultaneous observations from multiple vantage points. This method paves the way towards a new generation of solar cycle studies of the solar EUV corona, contributes additional samples for datadriven methods and enables the application of automated methods that were developed specifically for SDO/AIA data to the full EUV data series without further adjustments.
- Publication:
-
Proceedings of the 2nd Machine Learning in Heliophysics
- Pub Date:
- March 2022
- Bibcode:
- 2022mlph.conf...39J