MERR values reported in DAOPHOT tasks
ameisner wrote on Jan 19, 2009
Recently I noticed that the MERR values reported for a given star by the tasks DAOPHOT/phot and DAOPHOT/peak (or DAOPHOT/nstar) are substantially different. I found the explicit formula for the phot MERR in the phot task help, but there is no description of the algorithm used to determine MERR by tasks like nstar and peak that fit psf models to stars. In between running phot and nstar or peak, I'm not changing any of the parameters (like epadu etc.) that enter into the determination of the phot MERR. The reason that this really bothers me is that MERR for a given star is consistently *larger* in tasks like nstar and peak that fit stars with a custom psf. It seems to me that using a psf which incorporates an analytic part and a table of residuals should improve one's estimate of the magnitude of a star being fit and therefore lead to lower MERR values. I don't think this can be attributed to a poor psf model, since I consistently get MERR_peak ~ (a few)*MERR_phot even when I use really good data from WIYN with sub-arcsecond seeing and use many tens of stars to build the psf. My question mainly is whether the relative sizes of MERR that I'm seeing are to be expected, and also on what basis is MERR computed in peak/nstar? MERR_phot is clearly statistical in nature...are MERR_peak/MERR_nstar supposed to be some sort of systematic error associated with a given psf model? Thanks.
Philip Massey wrote on Jan 19, 2009
The error can never be any smaller than that determined from photon statistics, and that is what phot reports. Consider the simple case that you use a measuring aperture of radius 5 pixels in phot. The error is then just determined by the read-noise, the gain, and the number of counts in the aperture. Now, consider fitting a PSF with a fitting radius of 3 pixels. Statistically the error HAS to be larger, even if the PSF model is perfect, as fewer photons are involved in making the fit to the PSF.
In addition to this, if the "chi" of the fit is greater than 1, then that would tell you that the fit isn't as good as you expect from photon statistics, and the associated error will be greater.
PSF-fitting is great, particularly in crowded regions, but it isn't magical.
See: Stetson (1987), PASP, 99, 191
---phil massey
In addition to this, if the "chi" of the fit is greater than 1, then that would tell you that the fit isn't as good as you expect from photon statistics, and the associated error will be greater.
PSF-fitting is great, particularly in crowded regions, but it isn't magical.
See: Stetson (1987), PASP, 99, 191
---phil massey
Jason Quinn wrote on Jan 19, 2009
massey
The error can never be any smaller than that determined from photon statistics, and that is what phot reports. Consider the simple case that you use a measuring aperture of radius 5 pixels in phot. The error is then just determined by the read-noise, the gain, and the number of counts in the aperture. Now, consider fitting a PSF with a fitting radius of 3 pixels. Statistically the error HAS to be larger, even if the PSF model is perfect, as fewer photons are involved in making the fit to the PSF.
In addition to this, if the "chi" of the fit is greater than 1, then that would tell you that the fit isn't as good as you expect from photon statistics, and the associated error will be greater.
PSF-fitting is great, particularly in crowded regions, but it isn't magical.
See: Stetson (1987), PASP, 99, 191
---phil massey
And if you are doing non-relative photometry, I've always found the zero-point offset a big source of systematic error (comparable to the photon noise) using the IRAF version of DAOPHOT.
Jason
Last post on Jan 19, 2009