SPI3S: SPectral Irradiance of the 3d Sun

EUV-observing instruments are limited in their numbers and have mainly been constrained to viewing the Sun from the ecliptic. For example, the Solar Dynamics Observatory (SDO; 2010-present) provides images of the Sun in EUV from the perspective of the Earth-Sun line. Two additional viewpoints are provided by the STEREO twin satellites pulling Ahead (STEREO-A; 2006-present) and falling Behind (STEREO-B; 2006-2014) of Earth’s orbit. No satellites observe the solar poles directly. However, a complete image of the 3D Sun is required to fully understand the dynamics of the Sun (from eruptive events to space weather in the solar system), to forecast EUV radiation to protect our assets in space, to relate the Sun to other stars in the universe, and to generalize our knowledge of the Sun-Earth system to other host stars.

To maximize the science return of multiple viewpoints, we propose a novel approach that (1) unifies and smoothly integrates data from multiple perspectives into a consistent 3D representation of the solar corona, and (2) allows us to make the first full-Sun EUV spectral irradiance measurements. More specifically, we leverage Neural Radiance Fields (NeRFs) which are neural networks that achieve state-of-the-art 3D scene representation and generate novel views from a limited number of input images. We adapted a Sun NeRF (SuNeRF) to generate a physically-consistent representation of the 3D Sun, with the inclusion of radiative transfer and geometric ray sampling that matches the physical reality of optically thin plasma in the solar atmosphere. SuNeRFs leverage existing multi-viewpoint observations and act as virtual instruments that can fly out of the ecliptic, that can view the poles, and that can be placed anywhere in the solar system at no additional cost. Data inferred from these novel viewpoints are then used as input into a deep learning irradiance estimator (MEGS-AI) trained on SDO/AIA and SDO/EVE observations to produce an EUV irradiance forecast for observations of the Sun that do not currently have irradiance spectra associated with them.

Ongoing work focuses on the development of an updated version of SuNeRFs that now leverages time sequences of multi-thermal EUV images captured from multiple viewpoints to generate a 4D data-driven reconstruction of the electron density and plasma temperature in the corona. The SuNeRFs model thus accounts for instrument temperature responses in order to incorporate EUV images from available sources. Additionally, we also explore the possibility of combining MEGS-AI with SuNeRFs reconstructions of the 3D corona to estimate irradiance at any point in the solar system, including L5 and Mars.

Our pipeline (SPI3S: SPectral Irradiance of a 3d Sun) is an example of how novel deep learning techniques can be used to significantly enhance observational capabilities by the creation of virtual instruments.