First Ground-Based Adaptive Optics Observations of Neptune and Proteus

High angular resolution (0.15”) K-band images of Neptune were obtained in August 1995, with the University-of-Hawaii adaptive optics system mounted on the 3.6-m Canada-France-Hawaii telescope. The images show bright high contrast features that are believed to be high altitude clouds. They confirm that low latitude (< 30 ̊) cloud activity has shifted since Voyager from the south hemisphere to the north hemisphere, whereas higher latitude activity seems more permanent. Proteus can be seen at the locations predicted from Voyager data. Its K-magnitude is 19.0 ± 0.03. The corresponding geometrical albedo is identical to that measured in the visible by Voyager. Introduction Voyager 2 images of Neptune revealed an unexpectedly high atmospheric activity, raising questions about the driving source of energy (Smith et al., 1989). Attempts have been made to follow from Earth the evolution of atmospheric features seen by Voyager, both with ground-based telescopes and with the Hubble Space Telescope (HST). HST has provided the highest angular resolution images since Voyager, first in 1991 (before repair), (Sromovsky et al. 1995), then in 1994 (Hammel et al., 1995), but only at wavelengths below 1 μm. Ground-based observations made in the K-band (2.2 μm) benefit from a higher contrast, because clouds are seen against a planet background darkened by strong methane absorption, but the angular resolution is severely limited by seeing. We present here what we believe to be the first ground-based adaptive optics observations of Neptune. Adaptive optics (AO) is a means for real-time compensation of wave-front distortions produced by turbulence in the Earth’s atmosphere. This technique has allowed us to produce the sharpest K-band images of Neptune ever obtained with an angular resolution of 0”.15, providing 15 resolution elements across Neptune’s equatorial diameter. Observations and data reduction Observations were made with an experimental AO system specifically developed at the University of Hawaii (UH) for astronomical observations (Roddier et al., 1991). A small (30-mm diameter) image of the telescope entrance aperture is formed on a 13-actuator deformable mirror which compensates wave-front aberrations. The mirror –called bimorph– consists of two piezoelectric wafers glued together. The correction rate (1.2 kHz) is sufficiently high to compensate wave-front distortions introduced by the atmosphere. Near infrared (1-2.5 μm) images are recorded with a 1024 x 1024 pixels HgCdTe infrared camera developed by Hodapp et al. (1995). Immediately before the camera, light shortward from 1 μm is reflected toward a wave-front sensor. An array of 13 photoncounting avalanche photodiodes detect any unbalanced illumination between oppositely defocused images of a guide source. The output signals relate to local errors in the wave-front curvature. They are used to drive the deformable mirror through a computer. For the observations presented here, Triton was used as a guide source. At the time of the observations, the angular distance between Triton and Neptune was 14”. This was found to be close enough for the atmospheric distortions to be almost the same for Neptune and Triton. The instrument was mounted at the Cassegrain f/35 focus of the 3.6-m Canada-France-Hawaii Telescope (CFHT) with the main purpose of observing the rings of Saturn as the Earth was crossing the ring plane (August 1995). We took opportunity of this observing run to record a few observations of Neptune. The data consist of four 5-minute exposures taken through a standard K-band filter. The exposures were taken on August 12, 1995, starting about half an hour after midnight (local time), that is on August 12 around 10:30 am (UT). For each of the 4 exposures, the exact midexposure time in decimal hours is: 10.4683, 10.5928, 10.6994, 10.8167. The plate scale and North direction were calibrated using the Saturn data. The pixel size is 0”.035, or 740 km at Neptune’s distance (for comparison, the pixel size of HST images is 990 km). The top of each frame is slightly eastward from north, at a position angle of +0.65 degrees. The effect of planet rotation is clearly seen when comparing successive frames. Adopting a 16 hour rotation period (Sromovsky et al. 1995), the blur produced at the sub-Earth point by a five minute exposure is one pixel, and was deemed acceptable. The blur due to the motion of the guide source (Triton) relative to Neptune was 0.4 pixel. A remotely controlled offset mirror in the AO system allows the telescope to be pointed anywhere up to 20” away from the guide source. This mirror was used to record Neptune on four different locations on the 1024 x 1024 infrared array, thus avoiding the need for recording a separate background exposure. Sky flats were recorded at sunrise for flat-field corrections. Each image was flatfielded with these sky flats and background subtracted. Adaptive optics produces images with a sharp diffraction-limited core. However, the core is surrounded with a halo of light scattered by uncompensated small scale wave-front errors. The halo produces a loss of contrast in the recorded images. The situation is similar to (but not as severe as) that of the impaired HST. The image contrast can be restored by deconvolution. To do that, we have estimated the point spread function (PSF) in two different ways. Firstly we used images of Triton recorded on the same frame. This has the advantage that they were taken at the same time under exactly the same conditions. However, the angular diameter of Triton was 0.13” which is comparable to the PSF full width at half maximum, that is Triton was almost resolved. Secondly we used stellar images taken some time later. In this case the source was unresolved, but the seeing conditions may have changed. Both PSFs yielded similar results. Best results were obtained by taking the median of all the available references (Triton and stellar). This approach has the advantage of improving the signal-to-noise ratio (SNR) on the halo which surrounds the PSF, a factor we found important for proper removal of the image blur caused by this halo. Because Neptune is quite dark in the K-band and the image is highly magnified, intensity levels in a 5-minute exposure are quite low and each of the four images is noisy. As a result, deconvolution of individual images was found to be ineffective. To improve the signal-to-noise ratio, we considered adding all four images. However, when doing this the rotation of the planet produces a significant blur. Over a 20 minute interval, a 16h rotation period produces a blur of the order of one resolution element. To avoid this blur, we decided to remap each image slightly as if it were taken precisely at the same time, arbitrarily chosen at 10h 38m (UT). The Jacobian of the geometric transformation was calculated assuming that irradiance was independent of direction, an approximation deemed acceptable for small corrections. We attempted to determine the rotation period that would best match the cloud positions in the first and the last image. We clearly found that a longer period was needed for the low latitude clouds in the northern hemisphere than for the high latitude clouds in the southern hemisphere. The best match gave 21h for the north and 18h for the south. Given the uncertainty on these values (± 1 hour at least), they are not inconsistent with Voyager results (Hammel et al., 1989). Because remapping is a non-linear transformation, one should in principle deconvolve each image first, then remap and co-add them to improve the SNR. However, the deconvolution of individual images being ineffective and the correction due to remapping being small, we remapped and co-added the images first. The resulting image was then deconvolved, with the benefit of a better SNR. We used the Lucy-Richardson algorithm. This algorithm is best suited to restore optical data which are dominated by photon shot noise with Poisson statistics, and gives a maximum likelihood estimate. It was widely and successfully used to restore early images from the impaired HST, the PSF of which is similar to that of our images. It is also widely used to further improve images produced by adaptive optics. A discussion on the application of this algorithm to such images can be found in Roddier et al. (1996). We were able to use up to 20 iterations of the algorithm with good results. The resulting deconvolved image is shown here in Fig. 1. Results and discussion