How to handle Hubble images

While we’re squirreling away processing the Hubble data on IC 2497 and Hanny’s Voorwerp, and starting to get some science out of them, here’s a guide to the kinds of things needed to get science from Hubble images and make them presentable. To demonstrate, I’ll use a galaxy that shows up in the opposite corner of the field in exposures with the Wide-Field Camera 3 (WFC3). This piece of the SDSS roughly matches what we covered in the deep-red and UV filters:

Now we’ll zoom in on the galaxy at lower left, SDSS ID 588016891171635444 with a redshift z=0.165, over three times as far away as Hanny’s Voorwerp and IC 2497 (so we should see less detail in it).

Now we’ll pull out a region 29 arcseconds long around this galaxy from the recent WFC observation in an I filter (around 8100 A). We take two exposures in each filter, moving the telescope slightly in between. The reasons for this show up when we look at one of the individual images:

Pretty terrifying. All those blotches and streaks mark pixels where cosmic rays – energetic charged subatomic particles – have struck the detector, registering as strongly as a bright source of actual light. (In the context of our Voorwerp project, probing the history of active galaxies and their accretion onto massive black holes, it is ironic that some of the most energetic cosmic rays have been attributed to such active galaxies, accelerating particles so close to the speed of light that they cross tens of millions of light-years without their paths being appreciably curved by galactic magnetic field). This happens on the ground as well, but the situation is worse with Hubble data – first, for the obvious reason that it’s above the protection given by the atmosphere, and second because its images are so sharp that we can’t always distinguish a cosmic-ray strike from a star image. So the best approach to remove them is to take multiple exposures and reject anything that’s only there in one image. Ideally one would take many images so that little signal is lost to cosmic rays, but this comes at a price. The amplifier on each CCD chip induces a certain level of noise at the end of each exposure when the data values are read into an external memory, so there is serious loss of data quality unless this noise is less than the statistical fluctuations in the background sky brightness. A common compromise, which we did, is to break a half-orbit period of visibility (about 50 minutes in this part of the sky) into two exposures, so the images can be combined while rejecting the cosmic rays and regaining almost all the image pixels. In this step, it is now usual practice to also offset the telescope by a few pixels between exposures (“dithering”), so that any bad pixels which are not already mapped can be rejected in the same way. The reconstruction process interleaving the pixel values is known as drizzlng. Here is the default “pipeline” processing drizzle version of our galaxy as delivered from the Hubble archive:

Now it’s obvious that it’s a nice barred spiral – did everyone click “clockwise” in Zoo 1? We should be able to do a bit better by rerunning the software on my own machine, and changing the parameters used to recognize cosmic rays and the amount of space to delete around each one. (For IT aficionados, a Python script calls a series of lower-level routines from the STSDAS and IRAF packages). After some help from colleagues at STScI, I have that running on my laptop and get this result:

Almost there. For some science purposes, this will do it. For example, for an inventory of potential star clusters, we’d work from this version and use the statistics of image size versus peak intensity to weed out cosmic rays that sneaked past the software. For visualization, I can interactively remove things that I’m pretty sure are cosmic rays, and get this nice-looking result:

That was actually half-size – to see every pixel, look here. It’s impressive to recall that the primary mirrors of Hubble and the SDSS survey telescope are almost the same size. Getting past the atmosphere and lengthening the exposure by 60 times make a big difference!

And now I have another galaxy I really want to finish this process for, preferably together with some narrowband data in [O III] and H-alpha expected in about 9 hours…