GRID-SITES: Gridded Solar Iterative Temperature Emission Solver for Fast DEM Inversion

TL;DR

Grid-SITES significantly accelerates solar DEM inversion by grouping similar pixels, enabling faster analysis of large datasets with minimal added uncertainty, suitable for high-resolution solar imaging.

Contribution

Introduces Grid-SITES, a novel gridding approach that enhances DEM inversion efficiency by grouping pixels, applicable to large datasets and compatible with existing inversion methods.

Findings

Achieves up to 30x speed increase for 10 MP images.

Maintains low uncertainty despite data quantization.

Efficiency improves with sequential processing of image series.

Abstract

The increasing size of solar datasets demands highly efficient and robust analysis methods. This paper presents an approach that can increase the computational efficiency of differential emission measure (DEM) inversions by an order of magnitude or higher, with the efficiency factor increasing with the size of the input dataset. The method, named the Gridded Solar Iterative Temperature Emission Solver (Grid-SITES) is based on grouping pixels according to the similarity of their intensities in multiple channels, and solving for one DEM per group. This is shown to be a valid approach, given a sufficiently high number of grid bins for each channel. The increase in uncertainty arising from the quantisation of the input data is small compared to the general measurement and calibration uncertainties. In this paper, we use the Solar Iterative Temperature Emission Solver (SITES) as the core method for the DEM inversion, although Grid-SITES provides a general framework which may be used with any DEM inversion method, or indeed any large multi-dimensional data inversion problem. The method is particularly efficient for processing larger images, offering a factor of 30 increase in speed for a 10 megapixel image. For a time series of observations, the gridded results can be passed sequentially to each new image, with new populated bins added as required. This process leads to increasing efficiency with each new image, with potential for a $\approx$100 increase in efficiency dependent on the size of the images.