Pushing High-Contrast Imaging with Hubble to the Limits

John Debes, debes@stsci.edu, and Bin Ren, bin.ren@jhu.edu

In 1980, a paper by D. W. Davies pointed out that a 2.4-meter telescope like Hubble could reasonably expect to detect an exoplanet in reflected light—provided that one could integrate sufficiently long to overcome the overwhelming background caused by a host star’s point spread function (PSF), the diffraction pattern of a telescope created by the shape of the primary mirror, the support structure/secondary mirror, and any pupil plane aberrations (Davies 1980). More careful analysis of Hubble’s expected optical performance showed that the roughness of the mirror and its associated aberrations made exoplanet imaging prohibitive for even the closest of stars (Brown & Burrows 1990).

These papers, among others dating back to before the mid-1970s, constitute a history of thought on how best to detect planetary systems through imaging, spectroscopy, and photometry. By 2016, many of these fledgling ideas have come to pass, including the direct imaging of exoplanets. Currently, Fomalhaut b stands as the best candidate for the first exoplanet detected in reflected light (Kalas et al. 2013), but its unusual orbit and properties have made its planetary status controversial (Lawler et al. 2015). A handful of other Jovian-mass planets have been imaged in near-infrared light, where extreme-adaptive-optics (AO)-fed coronagraphic instrument performance and the level of emission from young planet atmospheres combines favorably for detection (i.e., Marois et al. 2008; Macintosh et al. 2015). Coronagraphic operations have also evolved extensively over the years, with new techniques developed to push to deeper contrasts through image post-processing with large imaging reference libraries (i.e., Lafreniere et al. 2007; Soummer et al. 2012).

A formal start to NASA's Wide Field Infrared Survey Telescope (WFIRST) with its Coronagraphic Instrument (CGI) will mean that discoveries using the direct imaging of exoplanets in reflected light will accelerate. CGI will suppress the PSF of the telescope at 1 part per billion at just 100–200 mas from a target star, thus allowing the direct imaging of Jupiter analogs around nearby stars.

Behind this backdrop, it may be counterintuitive to expect that the 19-year-old STIS instrument on Hubble can contribute to this discussion. However, recent advances in coronagraphy with STIS place it at the frontiers of high-contrast imaging science, and uniquely positioned to inform the future operations of both the James Webb Space Telescope and WFIRST.

High contrast, small angular scales

As part of the innovative calibration program 14426 “Pushing the Limits of BAR5,” the STIS team asked the question: Using modern high-contrast imaging techniques developed on the ground and for Webb, what is the limit with which STIS can perform high-contrast imaging? To answer this question, the team selected the nearby bright star HD 33893, observed it over nine Hubble orbits with the STIS 50CORON aperture, placing the star behind the BAR5 aperture location (See Pushing Coronography Deeper with New Coronographic Modes; https://blogs.stsci.edu/newsletter/files/2014/05/STISrev.pdf).

The observing strategy was to: 1) collect as many photons as possible; 2) observe the star at multiple spacecraft orientations; and 3) and execute sub-pixel-sized dithers behind the BAR5 mask.

In a regime where the sensitivity to faint objects is determined by the PSF wings, any high-contrast observation needs enough light such that the photon noise present at a given angular separation is small when compared to the total flux of photons coming from the star. Hence, a bright star was selected to ensure plenty of counts in the PSF wings. For very bright stars, the total amount of time available to observe the star is limited by how often you read out the detector and the CCD’s read-out time. For faint stars, the contrast is eventually limited by how long the total exposure time is, or by the properties of the detector.

In an ideal situation, the observations would be limited by photon noise, but additional systematic noise occurs due to non-repeatabilities with which the star is placed behind the occulter from orbit to orbit, as well as to the focus and thermal evolution of the telescope. Previous high-contrast imaging has demonstrated that changing the on-sky orientation of the scene with respect to the PSF decreases this systematic noise (Debes et al. 2013; Schneider et al. 2014), also known as Azimuthal Differential Imaging (ADI).

We also pursued sub-pixel grid dithers to mitigate pointing uncertainties that significantly degrade contrast at the inner working angle of the occulter, which were originally designed by others to aid high-contrast imaging with Webb (Lajoie et al. 2015). Lajoie et al. (2015) found that the combination of dithering, changing spacecraft orientations, and post-processing with the Locally Optimized Combined Images (LOCI) algorithm (Lafrenière et al. 2007) or with the Karheunen-Loève Image Projection (KLIP) algorithm (Soummer et al. 2012) can lead to gains that approach the photon limit.

With the full dataset executed by February 26, 2016, we combined all of the images and determined achieved contrast levels using both classical ADI reductions and KLIP reductions of the data. In Figure 1 we present the results of our work on program 14426. We plot the contrast levels obtained for point sources recovered with a signal-to-noise ratio of 5. We determined these curves by inserting artificial PSFs from STIS and recovering them with small photometric apertures.

Figure 1: STIS coronagraphic contrast at small inner-working angles. This plot compares the raw contrast of the STIS corongraphic PSF wings (red curve) relative to the point source contrast achieved with Program 14426 with a classical Azimuthal Differential Imaging (ADI) reduction (black curve) and with post-processing using the Karheunen-Loève Image Projection (KLIP) algorithm (purple curve). Both results are slightly higher than the photon-limited prediction for contrast in an ideal case, but KLIP performs better than ADI interior to 0.8".

We find that we obtain 10-6 contrast at 0.5", and better than 10-6 contrast beyond 0.7", comparable or better than the best performance reported by GPI (NIR), SCExAO (NIR), and VLT/SPHERE-ZIMPOL (Visible) in the literature. We also obtain performance within a factor of a few of the photon limit estimated for the observations at these distances. The results of Program 14426 were publicized to the community through a Space Telescope Analysis Newsletter alert and through the STIS website (at http://www.stsci.edu/hst/stis/strategies/pushing/coronagraphy_bar5) on March 8, 2016. The BAR5 occulter location is now a fully supported coronagraphic aperture location within the Astronomer’s Proposal Tool as of Cycle 24.

Based on these results, STIS is unique as a high-contrast imager using total intensity visible light. Particularly in the northern hemisphere, it is the only visible light coronagraph available for small inner-working angles. It also complements images taken by SPHERE-ZIMPOL, which primarily works in polarized visible light. Since GPI and SPHERE have only recently started taking observations, it will be necessary to re-assess STIS’ performance relative to these instruments over the coming years.

STIS is also unique for high-contrast imaging beyond a few arcseconds, where its broad bandpass, low background, relatively large pixels, and low read-noise allow superior performance compared to the ground-based imagers. Figure 2 shows an ADI reduction of HR 8799, showing the depth one can obtain with six orbits at different spacecraft orientations with a total of exposure time of 13,800s. These images achieve point-source sensitivities of roughly V = 28.5, or 10-9 contrast far from the star. The field of view for STIS is significantly larger than any existing ground-based high-contrast imagers.

Figure 2: A deep signal-to-noise image of the surroundings to the planet hosting star HR 8799, taken as part of the GO program 12281 (PI: M. Clampin) using the WEDGEB2.5 coronagraphic position. The program used six spacecraft orientations for a total exposure time of 13800 seconds. Several background objects are easily detected, including a particularly faint object at r = 11.9" with a contrast of 20.4 magnitudes in V relative to the target star (small white box). Artificial point sources with a contrast of 19.5 magnitudes relative to the star (large rectangle) are detected down to 3.5" away from the star at a signal-to-noise greater than 5.

While the BAR5 occulter location is still relatively new, the first peer-reviewed high-contrast imaging results demonstrate its promise for the future. The nearby disk around HD 141569 was shown to have a bright inner disk at distances of 0.4" to 1.0" through STIS observations that used BAR5 (Konishi et al. 2016). The observations pushed in to an angular distance from the central star of 0.25", corresponding to a radial distance of 29 AU.


Brown, R. A., & Burrows, C. J. 1990, Icarus, 87, 484

Davies, D. W. 1980, Icarus, 42, 145

Debes, J., Perrin, M., & Schneider, G. 2013, retrieved from https://blogs.stsci.edu/newsletter/2013/04/10/the-unique-coronagraphic-capabilities-of-stis-direct-imaging/

Kalas, P., et al., 2013, ApJ, 775, 56

Konishi, M., et al. 2016, ApJL, 818, 23

Lajoie, C.-P., Soummer, R., Pueyo, L., et al. 2015, retrieved from https://blogs.stsci.edu/newsletter/files/2016/01/Lajoie.pdf

Lafrenière, D., et al., 2007, ApJ, 660, 770

Lawler, S. M., Greenstreet, S., & Gladman, B. 2015, ApJL, 802, 20

Schneider, G., et al. 2014, AJ, 148, 59

Soummer, R., Pueyo, L., & Larkin, J. 2012, ApJ, 755, 28