ALICE Data Release: A Revaluation of HST-NICMOS Coronagraphic Images

The input data for the ALICE program come from the Legacy Archive PSF Library and Circumstellar Environments (LAPLACE) program⁸ (HST program AR-11279, PI: G. Schneider; Schneider et al. 2010). This program delivered a homogeneous re-calibration of a large fraction of the raw NICMOS coronagraphic archive, optimized for imaging at separations close to the coronagraphic-mask inner working angle (radius of 03) using PSF subtraction techniques. This re-calibration was performed using contemporary flat-field frames optimally matched to the location of the coronagraphic mask (as opposed to epochal flats), and using an improved bad-pixel correction. During the second era of NICMOS operations (after replacement of its cooling system), dark calibration observations were obtained with less frequency than during the first era, and the LAPLACE program delivered two versions of re-calibrated images: one using observation-optimized dark frames when available, and one using synthetic model dark frames.

For consistency, the ALICE program re-analyzed all the non-polarimetric data from the LAPLACE program that were calibrated with contemporary flats for NICMOS Era 1,⁹ and with contemporary flats and observed dark frames for NICMOS Era 2,¹⁰ which represent 72% of the non-polarimetric NICMOS archival data sets. Furthermore, we also re-analyzed a few selected data sets that were not re-calibrated by the LAPLACE program (from HST programs 7248, 10897, and 11155).

The outputs of the ALICE program—and delivered data described through this document—all come from the input data detailed above. Data that were not reprocessed may have: (1) not been part of the LAPLACE inputs described above, (2) suffered from bad acquisition images not centered on the coronagraphic mask, or (3) failed sub-pixel alignment within a PSF library. The complete list of NICMOS coronagraphic programs (non-polarimetric data) is detailed in Table 8 in Appendix A. The table also reports the number of targets per HST program that have been partly or entirely reprocessed as part of the ALICE program. The complete list of reprocessed images for each HST program, target, and filter set can be accessed on the MAST archive. Details on how these data were re-analyzed (field of view, alignment procedure, PSF subtraction, and analysis methods) are described in Choquet et al. (2014a).

NICMOS data from 401 targets were used as input for the ALICE program, some observed in multiple HST programs (451 different target observations) and/or with multiple filters. In total, there were 590 different data sets, i.e., 590 unique observations with respect to program, target, and filter. The number of input images that were re-analyzed by the ALICE program amounts to a total of 4879 images (83% of them were acquired with the F110W or F160W filters). Reprocessed data are delivered for 494 data sets out of the 590 input ones. Of the 96 data sets not delivered, 56 of them were entirely composed of bad acquisition images and 40 others could not be properly aligned within a PSF library (often because the target was not a bright star or the target was a binary or contained a bright companion causing the alignment to fail). The majority of these data sets were acquired with the F110W or F160W filters (408 data sets). This amounts to a total of 3955 reprocessed images (86% in F110W or F160W), delivered back to the community.

The PSF libraries assembled by the ALICE pipeline are much larger in the F110W and F160W filters than in the other filters, as the F110W and F160W were commonly used for surveys and specific object characterizations. The mediumband and narrowband filters were exclusively used for characterization of known circumstellar objects. Furthermore, as ALICE's reprocessing is based on MRDI and excludes images with detected circumstellar material from the reference PSF libraries (except in "planet mode," which includes images of the science target acquired in a different orientation of the spacecraft, when available; Choquet et al. 2014a), the mediumband and narrowband filter data were reprocessed with PSF libraries of very small sizes (see Table 1). ALICE's reprocessing for these data sets is not expected to improve much upon classical PSF subtraction techniques. The real added value from the ALICE program mostly concerns NICMOS data acquired with the F110W and F160W filters.

We refer to a data set as the set of images acquired with the same filter element, as part of the same HST program, and under the same target identifier (HST TARGNAME keyword). Data sets are thus composed of a homogeneous group of images as designed by the PI of the program. A data set may combine images acquired with different spacecraft orientations, with different exposure times, and different epochs (within a year). We assume that any potential astrophysical source remains unchanged in all the images of a data set when combining them in the main ALICE science FITS file.

A few targets have several data sets with the same filter element, in different HST programs (despite HST policies not to re-observe a target in the same mode), or within the same program (target re-observed after a failed acquisition).

The main science output of the ALICE program is a multi-extension FITS file (Pence et al. 2010) that gathers all the high-level information about the reprocessed data set.

In an effort to facilitate high-level data exchange and to make exoplanet population statistical analyses easy, we developed a specific format for these data, and we propose that it becomes a standard for reduced products of high-contrast imaging data (Choquet et al. 2014b). We describe below the main choices we made for the Version 2 of this format and provide a detailed description of its content. We hereafter call this format the HCI-FITS file, for High-Contrast Imaging FITS file. It is inspired from the OI-FITS format, which is the standard for calibrated data exchange from optical interferometers (Pauls et al. 2004, 2005). In addition to the products delivered by ALICE, the HCI-FITS format presented here has also been implemented in the Direct Imaging Virtual Archive (DIVA), another large-scale database dedicated to high-contrast imaging data (Vigan et al. 2017).¹¹

A single FITS file—To enable high-level science analyses of high-contrast imaging data, several products are mandatory (e.g., the reduced image, detection limits, source detections). In order to prevent information loss when exchanging data, all the information must be gathered in a single file. The FITS file format offers the structure needed to achieve that, through the use of extensions. Extensions may contain different types of data, including images (IMAGE extension type) which is appropriate for the reduced images, sensitivity and SNR maps, and multi-dimension tables (BINTABLE extension type), which is appropriate for radial detection limits, characteristics of potentially detected sources, or general characteristics of the reduced products. Moreover, having a single file as reduced product makes databases more convenient to both implement and use.

A flexible standard—We developed HCI-FITS format to be compatible with any type of data set, regardless of the instrument, or observing mode, or processing method used to obtain the final reduced products. It can be used for both ground-based and space observation, for coronagraphic and saturated imaging, for broadband imaging, integral-field spectroscopy, and polarimetric imaging. Depending on the observer/analyst's choice to present the reduced data, a HCI-FITS file may contain products for either a single image or an image cube. For example, this format supports all options between combining all reduced images from an Integral Field Spectrograph (IFS) in one broadband image and keeping each reduced image separated in a spectral cube, while tracking specific image characteristics in all cases.

Structure of the HCI-FITS format—We identified five main products that are necessary for a high-level use of reduced HCI data: the reduced images, the SNR maps, the sensitivity maps (or "noise" maps), the radial detection limits (or "contrast curves"), and the characteristics of any detected point sources. The HCI-FITS format is thus composed of 6 extensions, one for each of these products, plus one which tracks the main characteristics of each image provided in the file. The extensions may appear in any order in the FITS file, but must have mandatory EXTNAME values to enable compatibility between files. The structure of the HCI-FITS format is provided in Table 2. The SOURCE_DETECTION extension is optional but must respect the specified format if present. This structure is not exclusive and may include additional data in other extensions (e.g., intermediate products such as the instrument PSF image). Reading software or codes should not presume the presence of such additional extensions.

The mandatory products enable analyses such as detection limit comparisons and astrophysical signal comparisons, but does not enable precise characterization of an unreported signal. Such characterization requires a forward modeling process (Lagrange et al. 2010; Milli et al. 2012; Soummer et al. 2012; Pueyo 2016) for which intermediate products are needed (instrument PSF image, raw data, eigenimages of the PSF library). Such detailed characterization is out of the scope of the HCI-FITS format use.

The DATA_INFORMATION extension is critical to identify the characteristics of each reduced image in the file. It is the extension that makes this format compatible with any collection of high-contrast images. It is a BINTABLE extension that must be composed of 12 fields that track the field orientation, polarization state, epoch, and spectral information of the images. It must have as many rows as images provided in the file, and if several images are provided, the order must be the same in the DATA_INFORMATION table as in the image cubes. We present in Table 3 the structure of this table.

The DETECTION_LIMIT extension is also BINTABLE and reports the radial point-source detection limits for each image present in the file. It must be composed of two mandatory fields reporting the separation from the star and the corresponding detection limit. The header of this extension must indicate the confidence level of the detection limit using the mandatory NSIGMA keyword. As the high-contrast imaging community is currently in the process of improving the definition of detection limits (Mawet et al. 2014; Jensen-Clem et al. 2018), we note that the header of this extension may be further developed in subsequent versions of the HCI-FITS format. The structure of this extension is presented in Table 4.

The third optional BINTABLE extension reports the characteristics of the point sources detected in the data. For each source it must indicate its astrometry, photometry, and SNR in each image provided in the file.

Specifics of the ALICE HCI-FITS files—For the specific case of ALICE, the HCI-FITS files always contain the reduced data for each combined-roll and for the combination of all images, so the ALICE products present cubes of N_im = N_roll + 1, where N_roll is the number of spacecraft orientations used to observe the target.

As the ALICE data were reprocessed using the PCA-based KLIP algorithm using large PSF libraries, we provide in the REDUCED_DATA header some specific keywords describing the reduction parameters we used for the data set (see Table 5).

The images in the SENSITIVITY_MAP extension are computed from the temporal variance of the residual speckle field through the PSF library. To do so, we reprocessed the reference images from the PSF library with the same parameters as the science images, and rotated-combined groups of them with the same numbers, weights, and angles as for the science combined images. We then compute the covariance matrix of these combined reference images, convolve it with a 1λ/D aperture, and compute its square root to estimate the temporal speckle noise map per resolution element. The images in the SNR_MAP extension are computed by convolving the reduced images from the REDUCED_DATA extension with the same aperture, and dividing it with the images from the SENSITIVITY_MAP. The tables provided in the DETECTION_LIMIT extension are the radial averages of the images in the SENSITIVITY_MAP extension, computed in 2-pixel wide annuli. They are normalized by the stellar flux converted to count s⁻¹ (keyword STARFLUX in the primary header) to give a measure of the point-source detection limit in terms of contrast to the star. We provide in the header of these three extensions keywords describing the parameters used to compute these metrics (see Table 5). It is important to note that the sensitivity map and the detection limit delivered here do not include the processing throughput.

The characteristics of the detected sources in extension SOURCE_DETECTION are computed from a matched-filter process with a synthetic, unocculted NICMOS PSF, computed for the corresponding filter element with the TinyTIM software package (Krist et al. 2011). The source astrometry is determined with the position that maximizes the cross-correlation between the reduced images and the normalized synthetic PSF. The photometry of the source is retrieved with the maximum value of the cross-correlation, subtracted from the local background level, corrected from post-processing oversubtraction with analytical forward modeling (Soummer et al. 2012), and corrected from correlation losses between the synthetic and the real PSF by using photometric calibration data acquired on calibration white dwarfs. The contrast is computed by normalizing the photometry by the stellar flux converted in count s⁻¹ (keyword STARFLUX in the primary header). Table 6 provides a description of the SOURCE_DETECTION extension.

The primary header of the ALICE HCI-FITS files is described in Table 9 in Appendix B. It gathers a selection of keywords useful at a science level, and comes from the raw HST FITS header, LAPLACE program added keywords, and from our work.

In addition to the main HCI-FITS file, which provides science metrics for the combined products of a data set, we also provide a "Products" folder gathering intermediate products for each image in the data set. These files are complementary to the combined HCI-FITS products. While the purpose of main HCI-FITS file is to provide high-level science metrics to quantify the detection limits and detected sources in the data set, the data image products are useful for diagnostic and astrophysical signal forward modeling.

For each exposure, we provide a multi-extension FITS file gathering the products described in Table 7. Unlike the main HCI-FITS file, these products are specific to ALICE and their format is not compatible with all types of high-contrast science products.

The REDUCED_IMAGE extension provides the reduced image computed by the ALICE pipeline using the KLIP algorithm. The image is not derotated and is presented with the same field orientation as the raw NICMOS image. The star is centered on pixel (41, 41) with pixel (1, 1) at the bottom left of the image. The header of the extension provides detailed information on the reduction parameters used to compute the image (see Table 10 in Appendix C).

The RAW_IMAGE extension provides the raw NICMOS image calibrated by the LAPLACE program. The LAPLACE image has been cropped to a field of view smaller than the NICMOS full frame (80 × 80 pixels or ∼6.1 × 6.1 arcsec), and the star is centered on pixel (41, 41). The header of the extension lists the keywords provided in the raw HST file, as well as the position of the star center in the full NICMOS field of view (see Table 10 in Appendix C).

The REDUCTION_ZONE extension provides a binary image of the reduction zone (pixels with 0 value were excluded from the reduction). In most cases, the reduction zone correspond to the full image except for a central mask of a few pixels radius. The parameters used to define the reduction zone are provided in the extension header (see Table 10 in Appendix C).

The EIGEN_IMAGES extension provides the cube of the first principal components of the PSF library used for the PSF subtraction, truncated at the number of components actually used to reduce the data. This cube can be used to analytically compute the impact of the PSF subtraction process on an astrophysical source using forward modeling.

The REF_FILE_NAMES extension provides a table with the file name of the NICMOS images composing the PSF library used to reduce the image. The table also provides the position of the star center in these reference images in the full NICMOS field of view, around which they have been aligned and cropped to 80 × 80 pixels.

Finally, we also deliver for each data set a "Preview" folder that contains PDF and CSV files of the content of each extension of the main HCI-FITS file. The PSF files show images of each frame of the REDUCED_DATA extension (one version with the point-source detections circled and one version without) of the first frame (North-combined frame) of the SNR_MAP extension and of the DETECTION_LIMIT extension. The CSV files show all the data contained in each BINTABLE extension (DATA_INFORMATION, DETECTION_LIMIT, SOURCE_DETECTION).