From: Andrew Fruchter (fruchter@stsci.edu)
Date: Tue Dec 09 2003 - 07:48:00 PST
Removing bad pixels, cosmic rays etc, is not a problem with sub-pixel
dithering Vallery. See the paper I sent around.
Also, the new software is only really needed in the regime where the
S/N is greater than or comparable to the changes in the wavelength
dependent flat-field changes. We will be lucky to be in that regime,
so one can do what the ACS Team does -- if you want to be quick and
dirty -- and just drizzle together the images using a wavelength
independent flat-field. But we here are hoping to make the full proper
reduction fairly automatic.
Better resolution is indeed helpful. You want to resolve the features
not convolve them with something their own size (if possible). You
only loose S/N by doing so.
Cheers,
Andy
On Tuesday, December 9, 2003, at 05:43 AM, Vallery Stanishev wrote:
> Dear All,
>
> The document I wrote was not meant to be a complete description
> of the reduction or complete analysis. And apparently there
> are some points that are not clear.
>
> About Rachel's comment #2. Yes, I first flat-field the images,
> and then subtract the background and extract the spectra.
>
> About #3. Before combining Riess' images I shifted them
> according to what is given in the fits headers. Here by
> 'shift' I do mean sub-pixel shift. I did this with IRAF's
> IMSHIFT, which allows sub-pixel shifts.
> Perhaps this is not the best way to handle this, but since
> the software to deal ACS/grism dithered images is not
> out yet I used IRAF.
>
> About the dithering, sub-pixels, etc.. Riess' images were
> dithered by a few pixels only. On all 16 images the SN was
> within a square of 20x20 pixels. I also don't know if this
> is large enough to start seeing differences in the wavelength
> calibration (but this may be tracked down form the field dependence
> of the wavelength solution given in the aXe's conf. file).
> My guess would be that this 20-pixels shift is
> fairly small to affect the wavelength calibration and hence
> the flux calibration of Riess' spectrum.
>
> I agree that the point of doing sub-pixel dithering is
> to improve the resolution. If the software to handle such
> data is about to be released we can use sub-pixel dithering.
> However, I'm not quite sure that the spectral resolution is
> of big concern for us. Given the broad SN features, especially
> at z>1, a spectral resolution of ~90A seems to be enough to type the
> object and to get the redshift with reasonable accuracy.
> Moreover, for these very faint high-redshift SNe the S/N
> will not be high and we shouldn't hope to get something
> more than that. I would be more concerned about bad pixels
> and in this aspect I think that integer pixels dithering
> should be enough.
>
> Regards,
> Vallery
>
>
> Rachel G. wrote:
>>> As Vallery points out, there is a discrepancy with Adam's spectrum,
>>> especially on the red end. Please send comments to Vallery and cc to
>>> this mail list - hstsearch@lbl.gov.
>> Two immediate comments on Vallery's conclusions regarding Riess' data:
>> 1. We should be able to handle sub-pixel dithered data. It isn't an
>> easy problem, but there is existing software that can properly handle
>> such data. I can look into this.
>> 2. The background subtraction is an important step. Remember
>> flat-fielding is wavelength dependent, so the background you subtract
>> has to be flattened properly as well. This could very well be the
>> source of some of the differences with Riess' reduction (along with
>> #1).
>> 3. The correlated noise you see in your shift and add of Riess' data
>> cannot be directly compared to a straight sum. If you simply sum
>> Riess' data ignoring image shifts, you should find the data have
>> better SNR than ours (although they could have been taken when
>> background levels were higher). But even in the present sum, one can
>> see Riess' spectrum has more interesting structure (features) than do
>> ours.
>> Definitely summing sub-pixel dithered data has to be handled more
>> carefully.
>> In addition, Given there looks to be a bit of a lull in the search
>> planning (for a few days perhaps) I will finish my write-up of the
>> reductions and analysis I did. The details of how this stuff is done
>> is important and I suggest we compare results before moving on.
>> Rachel
>
>
This archive was generated by hypermail 2.1.4 : Tue Dec 09 2003 - 07:47:33 PST