Processing Pipeline

The pipeline can be broken down into the following steps:

Image selection

While the majority of the Suprime-Cam archive was processed by the SCLA, some images could not be processed. Images with obvious defects including bad tracking, electronic issues with the readout of one or more of the chips, odd illumination patterns caused by a nearby bright object, or bad seeing were rejected.

For some filters, the filter transmission function were not available at the time of processing; these data could not be photometrically calibrated. Most of the narrow band filters (except N-A-L656, N-B-L711, N-B-816 and N-B-L921) are in this category.

For several images, the astrometric calibration failed. In some cases, this was due to a lack of usuable point sources, either because of a nebula or bright, extended galaxy in the image, or due to a short exposure time. Occasionally, the star-matching algorithm became confused, particularly in crowded fields. This last problem was exacerbated by the fact that many of the Suprime-Cam images are much deeper than the astrometric reference catalog from GAIA.

Detrending with SDFRED

The individual images are retrieved from the STARS archive and stored in the CADC VOspace. The basic detrending is done using the sdfred tools provided by Subaru. The basic recipe prescribed by the sdfred user manual was followed, although wrapper scripts are used to automate the process. The data was grouped by observing run, and the different calibration products (biases, darks, flat-fields) built per run. In some cases this was not possible, because the relevant calibration images hadn't been taken, in which case calibration products from the previous or subsquent runs were used.

Bias and dark frames were built using the standard recipe. For the flat-fields, three kinds were generated, in increasing order of preference: dome, twilight and night-sky. The last type, object flats, are built using the science images themselves. While testing has shown that the object flats are superiour in terms of photometric flatness and depth, there are not always a sufficient number of images taken during a run, in which case twilight or dome flats were used.

The masking of the auto-guider shade was applied.

The Suprime-Cam detectors were changed in 2008, with a corresponding change in SDFRED version.

Astrometric calibration with GAIA

The astrometric calibration uses GAIA DR2 as the reference. Each image is calibrated in two stages. The first stage operates chip-by-chip and models the distortion as a 2nd order polynomial in the chip positions, x and y. In the second stage, the distortion is modeled as a 5th order polynomial in r, the distance from the center of the mosaic, plus a 1st order fit in x and y, equivalent to the CD matrix. This distortion fit is done over the entire mosaic simultaneously. Doing a global fit in this manner greatly reduces the number of parameters necessary. Only (6 parameters / chip) × 10 chips + 5 = 65 parameters are needed for the mosaic. Conversely (20 / chip) × 10 chips = 200 parameters are needed for a 3rd order polynomial fit in x and y; given that there are typically ~1000 GAIA stars in a Suprime-Cam field-of-view for a typical high-galactic latitude field, which raises the possibility of over-fitting. The median astrometric residuals are 55 mas with respect to GAIA. Although the distortion is measured using this radial model, it is converted into a polynomial in x and y and stored using the FITS PV keywords for compatibility with other software. This software that performed this task was adapted from the MegaPipe pipeline.

Photometric calibration with Pan-STARRS

The photometric calibration is based on the Pan-STARRS DR1 photometric catalog. The Pan-STARRS grizy photometry is converted into the Suprime Cam passbands. The full transmission functions of each Suprime-Cam filter is computed, including the filter itself, the quantum efficiency of the detectors the transmission of the camera, the reflectivity of the primary mirror and the transmission of the atmosphere at a nominal airmass of 1.25. The transmission functions of the PS1 filters are taken from Tonry et al. (2012). Standard stellar spectra from Pickles (1998) and CALSPEC are multiplied by the filter transmission functions to produce synthetic photometry. This synthetic photometry is used to compute a polynomial conversion between the Pan-STARRS and Suprime-Cam photometric systems.

This is relatively straightforward for the Suprime-Cam grizy passbands; the bands are fairly similar and consequently the color terms are fairly small. For the BVRI bands, the color terms are larger and typically only valid over a narrow range in color.

The narrow- and intermediate-band filters can also be calibrated in this manner, athough again the transformation is valid only over a small color range. For a number of filters, the transmission functions are no longer available. These data were not processed.

The details of the filter curves and transformation are given on the Suprime-Cam filter page.

For each image, the Pan-STARRS photometry for that patch of sky is transformmed into that image's bandpass and used as in-field standards. This allows images taken under all conditions to be photometrically calibrated. The photometric calibration done in this manner is typically good to about 10-20 mmag with respect to Pan-STARRS, which itself is calibrated to slightly better than 10 mmag.

The photometric calibration proceeds in two stages. In the first, a relative photometric calibration is applied to correct the mosaic to single zero-point. This correction is computed for each observing run and applied to each image from that run. The second step computes. the global zero-point for the whole mosaic taking into acount the varying atmospheric transmission.

Stacking with SWarp

For each observing run, a mask is generated to eliminate dead/hot pixels and bad columns. The mask changes only slightly between runs, but obviously changes significantly at the point where the Suprime-Cam detectors were replaced in 2008. WeightWatcher is run on the images to mask cosmic rays.

The Suprime-Cam images are stacked onto a series of tiles covering the sky. The tiles are the same set used by MegaPipe 2.0. The tiles are measure 10000×10000 pixels and are spaced by 0.5 degrees in RA (with the appropriate cos(Dec) factor) and 0.5 degrees in Dec. An example of part of the tiling scheme superimposed on the footprint of some Suprime-Cam images is shown in the figure below. The advantage of such a scheme is obvious in the upper half of this figure where multiple Suprime-Cam images partially overlap one another, but there is no clear footprint on which to stack.

Example of stacking

Where possible, the Suprime-Cam images are also stacked in groups corresponding to the original observing pattern. A friends-of-friends algorithm is run on the image catalog, grouping together images whose centers are within 0.1 degrees of each. If the group contains 4 or more images in band, a stack is built on this footprint. If multiple bands are available in a group, all the bands will be stacked onto the same footprint. On the bottom half of Figure 2, one can see two obvious concentrations of Suprime- Cam images, where the observing pattern was constrained to relatively small dithers. Separate stacks are made for each of these two groups.

The stacking is done using SWarp. The images are resampled according to the astrometric calibration, scaled according to the photometric calibration. The images are combined using a clipped mean. The resulting stacks and corresponding catalogs are made available for download. The individual fully detrended and calibrated Suprime-Cam images are also made available for astronomers interested in time-domain astronomy, such as variable sources and solar system objects.

Date modified: