Friday, June 29, 2012

Building the Hubble Image Mosaics for CANDELS

When we use Hubble to observe the CANDELS fields, there are quite a few steps needed to turn the original ("raw") data into the final nice combined multi-wavelength mosaics that we use for science and for color images - it's not exactly an instant process!

First, the observations for each CANDELS field are spread over many months, and each Hubble observation consists of many exposures - an "exposure" is the basic building block for how we construct the CANDELS observations. Since the galaxies we're studying are so faint, each exposure typically needs to last for about 15-20 minutes just to be able to collect enough light, and even then many of the galaxies still look very faint on a single exposure, so we need to combine many exposures of each portion of the sky in order to see these galaxies and even fainter ones. In addition, the Hubble cameras cover a fairly small portion of the sky (only about 2 - 3 arcminutes, or about 1/10th the diameter of the full moon), so we need to obtain many adjacent exposures to cover the CANDELS fields. Finally, since each Hubble camera can only observe in one filter (or one color) at a time, we need to take multiple exposures in different filters to build up the multi-color mosaics. Over the 3-year lifetime of the CANDELS project, we end up obtaining over 7,000 exposures in total!

Once we have all the different exposures for a CANDELS field, including multiple exposures in different filters and at different pointing centers across the field, we are ready to start working with them. The first steps involve "calibration", where we remove electronic artifacts that are produced by the cameras, and ensure that we are left only with real objects in each exposure. Related to this is the step of removing bad pixels that can sometimes be present on the detectors, along with "cosmic ray" hits, which are small bright random spots on each exposure caused by high-energy particles, known as "cosmic rays", hitting the detector. These can be identified and removed by comparing a sequence of exposures of the same part of the sky - the real astronomical sources (stars, galaxies) will show up at the same locations in all the exposures, while the cosmic rays will be different in each exposure and can then be identified and excluded later on.

Hubble mosaic of GOODS-S, one of the CANDELS fields. Each point of light is a galaxy, with over 20,000 galaxies detected in this image. Red represents the infrared wavelength images (which is invisible to our eyes, but can be detected by the cameras on Hubble), while blue represents shorter wavelengths. More than 2,000 separate exposures were used to create these mosaics, requiring a total of more than 1 million seconds of Hubble observing time.
Credit: Anton Koekemoer
Many of the CANDELS team members help to visually inspect the exposures to make sure that there are no other issues - for example, other satellites in space sometimes move across the field while Hubble is taking the exposure, which will leave a bright streak across the image! These are then masked out and excluded later on. Bright stars can also sometimes cause reflection 'ghosts', which we need to be aware of. The images also need to be aligned properly - Hubble can point with fairly good accuracy (about 1 arcsecond), but we actually need about 100 times better accuracy to eventually be able to combine the images, since the pixels on the cameras have a typical size of 0.04 - 0.12 arcseconds. We have written computer programs that do this by first identifying all the objects in each exposure, then matching them up between different exposures to find and remove the shifts. This process also aligns the images to other catalogs of objects from other telescopes on the ground, which helps us later when comparing the new Hubble results to data from other telescopes.

After the images have been aligned, they are ready to be stitched together, or combined, into a set of "mosaics", using other computer programs that we have written. This step takes all the exposures of a CANDELS field in a given filter, and combines them in an optimally weighted way that resembles dropping small spots onto a single output image - this process is called "drizzling"! Also, since a multitude of images are combined together in this way in a single step, the software for doing this is called "Multidrizzle"! The end result from this is a large combined mosaic, for each different filter, of each CANDELS field, together with a few related mosaics that indicate how much exposure time went into each portion of the field, and how much background noise is present across each CANDELS field.

The mosaics are large! The current ones can span up to 30,000x70,000 pixels on a side, meaning that this single mosaic image would correspond to a 2,000-Megapixel camera! The resulting file size is 8 Gigabytes, which is actually too large for some of our software programs to cope with so we generally break them up into smaller more digestible chunks, or make "binned" versions at a coarser pixel scale, for comparison with data from other telescopes that have lower spatial resolution than Hubble. We use these mosaics to make the final color pictures, which are "approximate true-color", in the sense that we use different images from filters covering a range of different wavelengths. However, since the longest wavelengths are in the infrared (which would be invisible to our eyes), we generally use red for those images, and use blue and green for the images obtained at shorter wavelengths.

This is then where the image processing ends and the science begins - we use these mosaics to build catalogs of galaxies and measure their shapes and other properties, all of which are the subject of the various scientific results that we are pursuing in CANDELS. So, that's where the real fun begins!

No comments:

Post a Comment