STScI Logo

Hubble Space Telescope
How well do integral pixel dithers in the WFC work?

Integral pixel dithers in the WFC:
We are considering small dithers to get rid of detector blemishes,
remove hot pixels, and average over flat fielding errors.  If we use
integral pixel shifts, images can be registered and combined without
interpolation.  How likely are we to get integral pixel shifts if we
ask for them?  It seems that if the error in the integer shifts
is much larger than the amount of pointing drift during a typical
exposure, then interpolation will be required for image registration.
Interpolation of undersampled images introduces photometric errors, which
we would like to avoid.


See the related FAQ on "dithering for hot pixels", with specific
2x2 pixel dither patterns that you can use.

You can certainly ask for integer pixel shifts, see the ACS Instrument
Science Report ACS/ISR 01-07 available from our web site

One of the predefined patterns offers exactly this, with a shift kept
as small as reasonable to achieve the goals (which you describe) yet
minimizes effects such as distortion, more below, see pattern number 14
in that ISR.
Obviously there will be several factors that come into play to limit
the achievable accuracy in practice. Within the instrument, the
geometric distortions introduce a change of plate scale across the
field of view of order 10%. Hence from center to edge the image scale
change is (about) 5%, so for pattern#14, a 5 pixel shift at the center
is (about) 5.25 pixels or 4.75 pixels at the edges.
HST itself and the instrument are likely to introduce additional
systematics. Obviously we don't yet know how large they are for ACS,
but for WFPC2 and STIS Ron Gilliland sees something of order 10 mas
pointing drift within an orbit, and more over longer timescales. These
are probably due to a mix of thermal effects within the telescope and
instruments primarily, although other things such as differential
velocity aberration may play a small role.  The pointing error on the
small angle maneuver commanded to the telescope is not likely to be the
dominant error, to the best of my knowledge, although from orbit to
orbit as guide stars are reaquired there may again be a small shift.
The undersampling of ACS images, particularly in the I band where you
are working, is not bad so I would expect relative image offsets to be
measurable to a high degree of accuracy.
On balance, I think it would be sensible to make a small, approximately
integer shift between exposures or exposure sequences, but of course it
is your call.
Incidentally, you will at some point have to decide what to do about
the geometric distortions. If you carry out photometry on the raw,
distorted image then you need to worry about what the distortion is
doing to the flat field (essentially, sky flats), or else you can
resample the image onto a uniform grid and carry out the photometry
there, which in principle should be consistent across the image. Also,
if weak lensing requires precise shape measurement, again you may need
to be concerned about the geometry.