Hi all! I will stay with english because I believe this issue to be of larger importance. I agree with you Lothar, the text is often difficult to understand if you have no basic knowledge about the subject. However, that doesn't reduce its quality. I also did not understand parts of the stuff of major importance to me - the flat fielding. For that reason I contacted Phil Massey and here is the discussion. As an important result, the response function as often used in some groups is out of business.
_____________________________________________
On Nov 17, 2010, at 4:29 AM, <Thomas.Eversberg> wrote:
Hi Phil,
as a former student of Tony Moffat I am still active in spectroscopy, although as an amateur (see
http://www.stsci.de/index_e.htm andhttp://
www.stsci.de/wr140/index_e.htm) and supporting the german amateur spectroscopists (
http://spektroskopie.fg-vds.de/index_e.htm). I now discovered your and Margaret Hanson’s text Astronomical Spectroscopy in astro-ph. I have some questions about the flat fielding procedure and especially about the response function and its necessity.
Until now I understand the flat as being necessary for getting rid of the local bumps and wiggles in the CCD image, due to the CCD window, grating or other optics inside the instrumental chain. The flat counts should be high enough not to degenerate the original spectra, as you said in your paper.
However, I had a number of discussions about flat normalization and the instrumental response function, and I still do not understand the necessity of both.
From my point of view, wavelength dependent effects as described in page 35 (bottom) (e.g. different color temperature) would only change the slope of the flat reduced spectra. This slope can then be flattened by standard rectification to 1. The bumps are gone and fine. The only problem I can imagine is the reduced S/N at low counts in the flat so that the spectral S/N is degraded at the low count pixel regions as described on pages 55-57. I thought normalizing the flat is a kind of a try to make spectral rectification easy and did not consider it as an always necessary standard procedure. Am I right here or do I miss something?
Second, the response function, obtained from a standard star spectrum would help finding the continuum in broad emission line stars like WR. However, I do not understand the advantage to introduce an additional normalization procedure with the S/N effects described in your paper for other stars than WR. Neglecting the color temperature and the spectral slope to be rectified, a dome flat should do a good job. What do I miss here? Is this procedure only for extremely difficult cases?
I would highly appreciate if you could give me some ideas.
Cheers, Thomas
Von: Philip Massey [mailto:
massey@lowell.edu]
Gesendet: Mittwoch, 17. November 2010 16:15
An: Eversberg, Thomas
Betreff: Re: AW: flat fielding
Zitat:
Hi, Thomas---
I'm not exactly sure I understand your two concerns/questions, so let me just try to restate it in different words. Flat-fielding spectra accomplishes three things: (a) it takes out the pixel-to-pixel sensitivity response of the chip, (b) it makes the spatial profile flat, or at least uniform (non-lumpy) in order to make it easier to perform decent sky subtraction, and (c) it also can be used to take out the low-order sensitivity variations with wavelength to make it easier to either normalize the final stellar spectrum or easier to calibrate to flux using standard stars.
I understand and agree with all issues. I am only confused by the “spatial profile”. The direction perpendicular to the dispersion is somewhat negligible from my point of view. A gradient over some pixel perpendicular to the dispersion should not be problem. But the 2D consideration includes the 1D, of course. So, basically, I agree.
Zitat:
This step is actually pretty critical if you are in the regime where good sky subtraction matters. Most spectrographs have vignetting so that the peak counts will be in the middle. So, you want to make sure that when you go to subtract sky by interpolating from either side of your object that you're not making a mistake. In other words, the spatial direction should be flat after flat fielding.
Ah, now I understand that the sky is the target of this procedure and I fully agree. In fact, during the recent years I did only work on bright stars of at least 8 mag and sky subtraction was not critical.
_____________________________________________
Zitat:
We need to normalize the flat field in the wavelength and spatial direction so as to not mess with the Poisson statistics in order to apply optimal extraction or to co-add the spectra properly---in other words, we want to preserve the original counts (more or less).
I try to understand this but I am not sure if I do. S/N should be larger at higher flat values than at lower ones for a non-normalized flat. And sigma should be the same, even if you do not normalize. So, you want to take the RELATIVE counts properly into account having an almost constant S/N in the flat?
Zitat:
No. The S/N in the flat is not changed by normalizing. What you want to do is preserve the absolute counts in the stellar spectrum.
Oh yes, of course. I was mixed up here. That means I do not want to affect the stellar counts by a non perfect flat, right?
_____________________________________________
Zitat:
If the spectrum of the flat field has the same bumps and wiggles as the spectrum of the star, then simply normalizing the flat field by a constant satisfies this.
What exactly do you mean with a constant? Just a number? Why not dividing the spectral data by the original flat. The spectral gradient will normally change but the bumps are also gone. [/color]
Zitat:
Yes, by a constant I mean just a number. Why you need to do that is as per above: you want the counts in your stellar spectrum to be reserved. In other words, if you have 5000 ADUs in a pixel in your stellar spectrum, and the gain is 1.5, then you know what the variance is going to be sqrt(5000*1.5)/1.5 in ADUs. You need to know this if you are to do anything like optimal extraction.
However, it could be that the flat field exposure has bumps in it that aren't
intrinsic to the spectrograph and detector. In that case dividing by the flat field exposure normalized by a constant will ADD bumps, not remove them.
Ok, I definitely need to distinguish between intrinsic features in the optical chain and spectroscopic feature introduced by external sources. Right?
_____________________________________________
Zitat:
However, often dome flat exposures have bumps and wiggles that the stellar exposure doesn't have because the reflectivity of the paint used on the spot is not uniform in wavelength: i.e., how much the spot reflects at 4000A is not going to be the same that it reflects at 8000A---the spot
is not truly "white". Worse, there may be appreciable bumps and wiggles (when plotted against wavelength) due to the paint having dips or peaks in its reflectivity. In that case one wants to fit a higher order function to the flat-field and remove those bumps and wiggles.
I understand and agree. This is something I never realized in detail. But why should I then apply a high order fit to get rid of the bumps in the lat? I could do the same after flat division in the stellar spectrum. Is it because flat fit is easier to perform? In addition, if I apply a high order fit to get rid of those spectral dips how can I then get rid of the bumps introduced by optical features (e.g., dust on the CCD window or on the grating, vignetting). Can I do both? If yes, how?
Zitat:
No. You don't know a priori how flat and smooth your stellar spectrum should be. It's not just Wolf-Rayet stars that have features.
OK, that is what I expected.
_____________________________________________
Zitat:
If standard stars had wavelength calibration at every 1 A rather than at every 50A (typically) then indeed you could take out bumps and wiggles by using the response curve. However, just like spectral normalization, you usually want to use a low order curve (5th or 6th cubic spline, say) to fit the response function. You can't go to 30th or 50th or something, which is what you would need to take out the bumps and wiggles you might introduce by the flat field.
Yes, I agree. But what is the response function for, in detail? I can imagine that the color temperature difference between target star and incandescent lamp and screen is a problem. But that should only introduce a final gradient after division of the spectrum by the flat. And that gradient will be rectified by a low order spline anyway. Neglecting different color temperatures I still do not understand the necessity of a response function from a standard star (which has not necessarily the same color temperature as the target).
Zitat:
Many of us flux our data; i.e., we are not interested in just normalizing the spectrum at the end of the process, but actually need ergs/sec/A. You need the standard star spectrum for that. I agree if you are just normalizing your spectrum at the end then using a response function will just make things a little easier for you.
Ok, that is what I expected. This is a hot discussion issue in the amateur domain. Many permanently use a response function and apply it to the data. And then they rectify the spectra. I believe that are unnecessary remainings from absolute measurements of the stellar flux.
_____________________________________________
Zitat:
Using low order to normalize stellar spectra is important not just with Wolf-Rayet stars with broad features, but particularly with early-type stars where the features can be very weak, if you want to then do anything quantitative with the data.
Yes, I understand
Zitat:
I hope this has helped.
cheers,
phil
Thanks a lot, Phil, I really learned something. As the webmaster of our spectroscopy section (
http://spektroskopie.fg-vds.de/index_e.htm) I placed your paper to the literature and marked it as a recommended one (also in the news ticker). This paper is very helpful. Very good work, indeed.
Cheers, Thomas
Zitat:
Hi, Thomas---
It sounds like we are now both on the same wavelength. I'm glad my responses were useful, and I'm glad you liked our paper! Good luck in your work!!
cheers,
phil
Yes, that’s right. I will also announce our email in our respective discussion forum. This paper is of high use for everybody!
Cheers from Bonn and thanks again, Thomas
Registrieren
Anmelden
FAQ
Suche
