I am using IRAF/daophot to build a PSF for an image that is a median-combined version of a number of exposures. Currently I have the datapars parameter "noise" set to "poisson" and have set "gain" and "readnoi" to the values appropriate to the camera used. However, I know it is incorrect for daophot to use Poisson statistics applicable to a single exposure on a median-combined image. For example, when I do external analysis I account for image combination in the expected variance of a pixel value by writing:

sigma^2 = N/(gain*n_images) + R^2/(gain^2*n_images)

Where n_images is the number of median-combined exposures in the image being analyzed and N is counts in ADU. On the other hand when daophot computes the PSF it must be assuming n_images = 1, which I would like to change. Is there any way or parameter to communicate information about the number of frames used to make a stacked image to daophot so that its routines use correct statistical models? Thanks for any help.

I'll leave it to a photometry expert to comment in detail, but you might find the following reference useful:

http://www.noao.edu/noao/noaonews/sep99/node41.html

This refers to a paper by Massey but the link appears to be dead at the moment. You might also search the forums for additional information on photometry of stacked images. See e.g.

http://iraf.net/phpBB2/viewtopic.php?t=79415&highlight=photometry+stack

Otherwise, AFAIK there's no parameter or other method to pass in the number of images used in a combined image, I'm hoping somebody can correct me.

Cheers,
-Mike

My interpretation of what you want to do is that for a stack the concept of gain and rdnoise have to be interpreted as a model for the noise. This standard Poisson model for noise is widely used and is fairly applicable to single CCD images but is less useful for processed images including stacks. But given a task has a fixed noise model you then have to find some way to set values for gain and rdnoise that approximate the behavior of the noise in the image of interest; a stack in your case. Since stacks are often made from interpolated single images (to get them on the same pixel scale) the noise is correlated. Again, the best you can do is try and approximate the noise by a Poisson model.

The rdnoise is simple, the standard deviation of the sky. The gain is harder to define empirically.

I hope this helps.

Yours,
Frank Valdes