CCD exposures skipped during coaddition

[sfu (Shenming Fu)]

Hi, I’m testing v26 on DECam data. I would like to coadd three g-band exposures (937419,937931,938234). The seeing is 0.9-1" with teff~0.7-1, so the observing conditions are great.

The single frame processing was successful, but after coaddition, I found at tract=9355, patch=26 the deepCoadd image from assembleCoadd looked like this.

I double checked the calexp images and they looked good. I tried using the jointcal results or not, but the results kept the same.

I also check the images from makeWarp. Here are the deepCoadd_directWarp images of those three exposures.

The psfMatchedWarp images are very similar to the makeWarp images.

I’m using the default configuration, and I suspect some config parameters may let the program skip some ccd images during makeWarp (or finalVisitSummary), and then the program just interpolated the pixel values from nearby ones. There are many config parameters so I’m not sure which one(s) may need to be adjusted…

Could you give me some help?

[sfu (Shenming Fu)]

After discussion with @antenglert I realized there are some parameters in the config of makeWarp. (Are they based on some science requirements of LSST/HSC/DES?)

# Maximum median ellipticity residual
config.select.maxEllipResidual=0.007

# Maximum scatter in the size residuals
config.select.maxSizeScatter=None

# Maximum scatter in the size residuals, scaled by the median size
config.select.maxScaledSizeScatter=0.009

# Maximum delta (max - min) of model PSF trace radius values evaluated on a grid on the unmasked detector pixels (pixel).
config.select.maxPsfTraceRadiusDelta=0.7

In the config of assembleCoadd, there are the same parameters, and also in assembleStaticSkyModel.

I set them to be None except the assembleStaticSkyModel ones. Then I ran the coaddition again using these new parameters and it worked.

[arunkannawadi]

Yes, they are based on HSC largely for filtering out bad visits from going into the coadds and ruining them altogether.