Efficient identification of pre-flare features in SDO/AIA images through use of spatial Fourier transforms

In this "Methods " paper, we investigate how to compress SDO/AIA data by transforming the AIA source maps into the Fourier domain at a limited set of spatial frequency points. Specifically, we show that compression factors of one order of magnitude or more can be achieved without significant loss of information. The exploration of data compression techniques is motivated by our plan to train Neural Networks on AIA data to identify features that lead to a solar flare. Because the data is spatially resolved and polychromatic (as opposed to spatially-integrated, such as GOES, or monochromatic, such as magnetograms), the network can be trained to recognize features representing changes in plasma properties (e.g., temperature, density), in addition to temporal changes revealed by Sun-integrated data or physical restructuring revealed by monochromatic spatially-resolved data. However, given the immense size of a suitable training set of SDO/AIA data (more than 10(11) pixels, requiring more than one TB of memory), some form of data compression scheme is highly desirable and, in this paper, we propose a Fourier based one. Numerical experiments show that, not only Fourier maps retain more information on the original AIA images compared to straightforward binning of spatial pixels, but also that certain types of changes in source structure (e.g., thinning or thickening of an elongated filamentary structure) may be equally, if not more, recognizable in the spatial frequency domain. We conclude by describing a program of work designed to exploit the use of spatial Fourier transform maps to identify features in four-dimensional data hypercubes containing spatial, spectral, and temporal information of the state of the solar plasma prior to possible flaring activity.