Blind source separation

Figure 2: from top to bottom: average solar spectral irradiance, the altitude at which wavelengths are predominantly absorbed, and the variability over one solar cycle. Data are from TIMED and SORCE.

Figure 3 : Comparison between various solar proxies (small letters ) and different spectral bands in the UV range (capital letters X, E, F, L and M). The shorter the distance between two letters, the easier it is to reconstruct one from the other. This comparison only applies to variations occurring on time scales of 17-34 days.

Blind source separation and hyperspectral unmixing have recently become very active fields of research. The objective is to recover individual source signals from their mixture with the least possible information about these sources or their mixing.

We were the first to apply this concept to solar UV spectra, with the aim to identify elementary spectra. Surprisingly, most of the solar spectral variability in the UV can be reproduced by the linear combination of three elementary sources that correspond to emissions coming from specific layers of the solar atmosphere (read more). See Figure 4.

We are now using this approach to decompose multispectral solar images into their sources and extract from these information about the temperature of the solar corona. Our objective is to deliver in real-time three-temperature maps of the Sun for space weather applications, see for example Figure 5.

Figure 4 : Excerpt of the three elementary spectra (in red) and average solar spectrum in the UV (blue). The elementary spectra are normalised to the average spectrum.

Original images from SDO/AIA in 6 wavelengths, in the EUV

(9.4, 13.1, 17.1, 19.3, 21.1, 33.5 nm)

Their 3 source images (upper row) and an excerpt (bottom row)

Temperature map obtained by assigning the blue channel to the cold source, the green one to the intermediate source and the red channel to the hottest one. This single picture condenses all the salient features contained in the original images.

Left picture : temperature map from sources. Right picture : classical composite map from spectal bands.

Figure 5 : Example of the 3-temperature map construction from solar images taken in the EUV.

T. Dudok de Wit

24/3/2011

To come next

Solar spectral variability

Higher order spectra

Ultraviolet emissions represent only about 9% of the total power that is radiated by the Sun and yet they receive much more attention because of their high variability, see Figure 2. These emissions may also provide one of the links by which the Sun affects climate (read more.).

Most UV emissions cannot be measured from ground because of their absorption in the upper atmosphere. In addition to that, space borne instruments suffer from degradation. For these reasons, there have been many attempts to reconstruct the solar UV spectrum from ground-based solar proxies such as the sunspot number.

We have developed a statistical technique that allows for the first  time the relevance of all these solar proxies to be evaluated in a single glance, see Figure 3. (read more)

Real-time reconstruction of the solar spectrum

Since January 2011 we deliver real-time spectra of the Sun from the extreme-UV up to the near-infrared using solar magnetograms from the SDO satellite

link:  http://lpc2e.cnrs-orleans.fr/~soteria/

These spectra are updated every 3hrs. They are made by segmenting solar magnetograms from HMI in regions of different intensities and then using an artificial neural network to reconstruct the contribution of each of these regions (takin into account their location on the Sun) to the spectral variability. This work was funded by the FP7 SOTERIA project.

Real-time spectra of the Sun

Figure 1 : Comparison between the observed (blue) and modelled (red) irradiance of the Lyman-alpha line at 121.57 nm. The observations are from SORCE/SOLSTICE.