Question on psf error map

[LyneVdV]

• psfr version: latest
• Python version: 3.9
• Operating System: Linux

Description

In the psf error map calculation, the expected noise level (i.e. sqrt(error_map_list[i])) is subtracted from the residuals (lines 358-360).
Here are the lines of code I am talking about:

        # subtract expected noise level
        if error_map_list[i] is not None:
            residuals -= np.sqrt(error_map_list[i])

If I understand correctly, the goal is to avoid double counting the errors, i.e. separate the "expected gaussian+poisson noise" to the one which comes from a true difference between the star and the estimated psf, the latter being the only one which should be taken into account in the psf_error_map. Is it correct ? If so, why are we subtracting simply instead of quadratically substracting ? I am not sure of the right answer, I just wanted to be sure we are consistent.

[sibirrer]

Hi @LyneVdV ,
thanks for the question. Indeed, quadratically subtracting would be the better/proper way. The question is just how that can be achieved when at the end we want the median (or mean) absolute difference of the model relative to the data for a set of multiple point sources. The way I've currently implemented it is more conservative in the sense that it suppresses the error map a bit.