Research Projects

Develop high-contrast image processing techniques

The direct observation of exoplanets is challenged by the extreme flux ratios between these faint objects and their host star (typically 10-6 for a young and warm Jupiter-mass planet) and by their short apparent separation to the star (0.5’’ for a Jupiter-like planet at 10pc). Exoplanet signals are thus buried under the bright glare of the stellar hallo, within a few resolution elements separation from the star. While complex corographic instruments are necessary to reject the bulk of the starlight upstream from the science camera (e.g. N'Diaye et al. 2016), image processing allows to further remove the residual starlight in the images. This is done by subtracting from the raw image a model of the instrument response image.

The main challenge in high-contrast image processing is to create accurate instrument response models matching the science observations, in spite of the complex spatial pattern and variable behavior of the instrument response. We rely on empirical models computed from image libraries representative of the diversity of patterns produced by the instrument to efficiently subtract the starlight from our raw images and reveal faint exoplanets in the processed images.

State-of-the-art

On ground-based telescopes, the standard is to use the science target itself to create representative image libraries. The most common observing method uses the natural celestial rotation overnight to introduce an Angular Diversity between each image in the library, allowing us to disentangle the signal of putative planets, which thus rotate in the field from one image to another, from starlight patterns, which are quasi-static in the library (Marois et al. 2006, Fig. 1 left). Combined with least-square modeling algorithms (e.g. LOCI: Lafrenière et al. 2007, PCA: Soummer et al. 2012), this method is quite efficient at detecting giant exoplanets at large separations (see Beuzit et al. 2019). However, it is blind to planets at short separations, where the angular diversity is too low to distinguish them from speckles (see Ruane et al. 2019). Similarly, it strongly affects azimuthally extended objects like circumstellar disks (see Milli et al. 2012).

On space telescopes like Hubble, which do not benefit from Earth's natural rotation, this observing method cannot be used to create large image libraries, or at the cost of multiple complex spacecraft maneuvers. Space telescope observations instead rely on simple one-to-one image subtraction to remove the contaminating starlight, using either the image of a reference star as instrument response model (Weinberger et al. 1999) or the image of the science target obtained with a different spacecraft orientation (“roll”, Lowrance et al. 2005), mimicking the ground-based Angular Diversity method while minimizing spacecraft maneuvers and overheads. With no library to sample the diversity and dynamics in starlight paterns, these methods are moderately efficient and leave strong residuals in the processed images, preventing the detection of exoplanets.

Archival Diversity

Building on the pioneering work of Lafrenière et al. (2009) and Soummer et al. (2011), I developped and optimized with P.I. R. Soummer's team a new image processing concept, Archival Diversity, to improve on the modest performance obtained with the Hubble Space Telescope (HST). Instead of using images from a single reference star, this method consists in using all the reference stars images available in the archives of an instrument to build a master image library that finely samples the diversity of response patterns produced by the instrument over its lifetime (Choquet et al. 2014, Fig. 1 right). After selecting the images from this library most correlated to a given science image and using the least-square modeling algorithms used with ground-based data (LOCI, PCA), Archival Diversity computes highly representative instrument response and significantly improves the starlight subtraction. On data from the HST NICMOS infrared instrument, this method routinely improves planet detection limits by 20 to 35 compared to basic reference or roll image subtraction.

The ALICE program

The ALICE program (Archival Legacy Investigations of Circumstellar environment) is an HST archival program led by R. Soummer (STScI) to consistently reprocess the entire coronagraphic archive of the NICMOS instrument with Archival Diversity. With the ALICE team, we reprocessed about 400 datasets for which we delivered high-level science products to the community on the MAST archive under the ALICE program (Hagan et al. 2018). We used this opportunity to propose a standard data format, the HCI-fits format, to present and exchange high-contrast imaging processed datasets withing the communicty and make multi-epoch and multi-instrument studies easier.

While re-analysing the NICMOS coronagraphic archive, we re-detected and discovered numerous astrophysical sources, including faint point sources, protoplanetary disks, and debris disks. All are available on the ALICE archive and several have been published in specialized astrophysics journals (e.g. Soummer et al. 2014, Milli et al. 2015, Mazoyer et al. 2016, Choquet et al. 2015, 2016, 2017, 2018). An extension of the ALICE program to the STIS coronagraphic archive has also been performed by collaborator B. Ren, which also allowed the re-detection of several debris disk systems (see Ren et al. 2017)

If you need support on existing ALICE products or custom processing on any dataset (e.g. larger field-of-view), feel free to contact me at elodie.choquet@lam.fr.

The ESCAPE program

Performing the reverse engineering and adapting the Archival Diversity method to ground-based high-contrast observations is yet another full research project. The challenge with ground-based data lies in 1/ the much stronger temporal diversity in starlight patterns caused by the atmospheric turbulence, even if partly corrected by extreme adaptive optics systems, and 2/ the very high number of frames stacked in their archives (105 images, compared to the few hundreds of images in the HST archives).

I am currently co-supervizing the PhD thesis of C. Xie with A. Vigan, started in Oct 2020, aiming at developing image selection methods to allow the application of the Archival Diversity to VLT SPHERE observations (Beuzit et al. 2019). This is the onset of the ESCAPE project (Exoplanetary Systems with our Coronagraphic Archives Processing Engine) I am initiating and which aims at developing innovative post-processing methods for high-contrast imaging instruments.

Collaborators

The ALICE team: R. Soummer (PI), E. Choquet, L. Pueyo, M. Perrin, B. Hagan, C. Chen, G. Schneider, A. Rajan, D. Hines, D. Golimowski, A. Sivaramakrishnan, T. Barman, J. Gagné, D. Lafreniere, B. Macintosh, C. Marois, D. Mawet.

The ESCAPE team: E. Choquet (PI), L. Altinier, R. Mayer, N. Godoy, A. Lau, C. Xie, A. Vigan, D. Mary.

High-contrast image processing techniques
Figure 1: High contrast image processing concepts.
Left: Angular Diversity method, which assembles image libraries from the science target itself with a rotation of the field of view between each image to disantabgle circumstellar objects (e.g. planets) from starligth contaminents. This method tends to subtract any signal at short separation from the star (e.g planet 2) and azimuthaly extended objects (e.g. circumstellar disks) along with the starlight.
Right: Archival Diversity method, which assembles libraries with images of reference stars gathered in its archives, thus sampling many realization of the random instrument responses across its lifetime. Using reference stars only, this method does not suffer from self-subtraction effects.