DeepEM: A Deep Neural Network for DEM Inversion
Abstract
The differential emission measure (DEM) is a critical component for understanding the physics of the solar corona and thus estimating the Extreme UV (EUV) solar spectral irradiance. However, inferring DEM maps from observations is an ill-posed inverse problem that requires the expensive optimisation of basis functions on a pixel-by-pixel basis, using multiple image channels (e.g. Cheung et al 2015). In the case of data obtained from the Atmospheric Imaging Assembly (AIA) onboard the Solar Dynamics Observatory (SDO), this involves six EUV channels at a resolution of 4096 x 4096 pixels. As a means to reduce the cost of calculating these DEMs, we propose the use of neural networks. Neural networks can be seen as a technique that enables complex transformations of the input data that takes advantage of the informational content contained in those data. This makes it possible for them to 'learn' physically meaningful relationships. Here we present a deep neural network implementation for DEM inversion that is able to perform millions of DEM calculations per second, with similar performance to basis pursuit (Cheung et al 2015). This work was performed at NASA's Frontier Development Lab, a public-private initiative to apply AI techniques to accelerate space science discovery and exploration.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMSM31D3536C
- Keywords:
-
- 1942 Machine learning;
- INFORMATICSDE: 7924 Forecasting;
- SPACE WEATHERDE: 7959 Models;
- SPACE WEATHERDE: 7999 General or miscellaneous;
- SPACE WEATHER