Physical units on Coadds pixel values

Dear all,

I noticed the min-max for for the pixel values on coadds are significantly different than the single visits. How are they rescaled? I saw in Jim’s paper the coadds are computed as a direct weighted average of the resampled CCD images, but didn’t catch anything on how the pixel values are rescaled.

I’m also trying to gain some intuition for the values. For example, for single visit exposures I can convert ADU to electrons per pixel by using the gain. Is there a way to convert the coadd pixel values to electrons or is that not well defined for a coadd?


You should treat the scaling of the pixel values as arbitrary, and use the attached PhotoCalib object to do any calibration.

Why do you want the values in electrons? You can’t obtain errors using Poisson statistics, because the values have been rescaled and correlated. Our best guess of the pixel errors (which is often not a very accurate guess due to variance lost to covariance during resampling) is in the variance array.

Hi Paul,

Thanks for your reply. We are looking to see if low surface brightness features of satellite streaks that extend beyond the edges of the masks that are placed over them during safeClipCoadd crop up in coadds. It is looking like they do not, but we wanted to plot the flux in coadds in units that are on the “same footing” as the single image visits. Is there a way to scale back the coadd so its on the same footing as single image visits so we can compare coadd ADUs directly to single image visit ADUs?


One option is to compare the calibrated values of both. An alternative is to compare the coadd to the warps, which should have the same photometric calibration as the coadd, so you can directly compare the pixel values (you might even get decent results with a simple subtraction, depending on the distribution of seeing values that went into the coadd).

I recommend using the photoCalib.calibrateImage() from both the calexp (or as @price says, the warp, so that the pixel positions line up directly with the coadd) and the coadd respectively to calibrate both of them to nJy, then you can compare them both with the same physical units.

sounds good, thanks all!

If you calibrate the calexp, bear in mind that there’s also a Jacobian term from the coordinate transformation that needs to be accounted for.