stscijgbot-jp
commented
3 months ago Charles Proffitt Following up on this, a potential easier option than modifying the association generator might be to tweak the parameters of the bkg_subtract step instead. The four other nod positions all get combined together using a sigma-clipped mean, and it looks like the default 3-sigma clipping isn't doing a great job in the case of the short S1600A1 slit at the longest wavelengths (i.e., largest PSF). If I change it to use 1.5-sigma clipping instead though (steps.bkg_subtract.sigma) it seems to do dramatically better at eliminating the overlapping source flux from the composite background.
Might this be an option for you instead? This way the only changes necessary would be for the NIRSpec team to deliver new pars-backgroundstep parameter reference files for whichever slits were problematic, and could adjust the sigma-clipping as necessary.
Issue JP-3551 was created on JIRA by Charles Proffitt:
For the five point dither pattern in the NIRSpec S1600A1 in FS mode, the primary dither spacing is only 0.2571", (with some additional variation if spatial sub-dithers are also included). This leads to significant overlap of the PSF wings of point sources in each image with the wings of the PSF in the background images and can reduce the net extracted flux by about 25% relative to the baseline calibration when the usual nod subtraction is done. But since we have 5 dither positions, this could be significantly reduced by not including the one or two immediately adjacent dither positions as background members in the level 2 files. One way to do this would be to require a minimum separation in the cross-dispersion direction before allowing another association member to be used as a background member for a given science exposure. Taking into account the additional sub-pixel dither offsets, and the dither patterns used for other slits, I believe that a requirement that the cross dispersion position of background members be offset by no less than 0.375" would have the desired effect. For my tests, I used the YOFFSET keyword to determine the offset between the dithers, but this might not be a general solution as one doesn't want to impact the use of a dedicated background observation at a different pointing. This should perhaps be a tunable parameter in a reference file in case other instruments or setups would benefit from a similar restriction. Even if we eventually come up with an improved scheme for correcting for the flux changes resulting from nod overlap in general, not subtracting such closely spaced members from each other should still improve data quality as subtracting the overlap and then adding back in an approximate correction would add noise, and so this fix would remain in place.
A workaround is available, as the user can always edit the level 2 association files to exclude the nearest background members and then re-run spec2 and spec3, but a large fraction of the NIRSpec fixed slit calibration observations taken of standard stars use the S1600A1 with a 5 position dither and it is a bit awkward for the pipeline to give default reductions that are so far off in flux for standard stars. If nothing else it has generated a number of help desk questions.
The rate files from Obs 160 and 107 of PID 1538 should provide good examples, as this star is bright enough that there is no significant real sky background.