What's this purple-lavender smudge?

Hmm. Are things working OK? Thumbs aren’t appearing on my last two posts and I’m getting a “service unavailable error” on some refreshes…
Purple smudge below:


Web server is having issues, it seems. 503 errors

Yes, there was an outage. I think it’s back now.

The purple line is continuing horizontally so it’s probably a boundary of different exposure bricks.

This should be a normal galaxy. However, there appears to be an image glitch so a band does not appear. Similar to if the G was missing in the RGB image, for example.

am i right?

edit: @spcgoat was a bit faster ;-D

1 Like

Yep, if you look at the different single image cutouts there are some partial exposures and that causes the purple effect.

Does every exposure count equally or are they weighted somehow?

Good question, unfortunately I don’t currently have an answer. I think @dstn has the best expertise in this area. :slight_smile:


i think i found it: on page 24 is something with weighend images.

We then create five separate stacks for the purpose ofsource detection: a weighted sum of all the (PSF-convolved) CCDs in a given band (resulting in threesuch stacks); a weighted sum of all three bands to optimize for a “flat” SED (i.e., zero AB mag color);and a weighted sum of all three bands to optimize for a “red” SED (i.e., with colorsg−r= 1 magandr−z= 1 mag). While these image stacks are weighted sums of the convolved images, the inputimages are not all convolved to a common PSF

but… well… yes :wink: i have no idea

edit: This is getting crazier and crazier… you can get the source code and dive into it. It seems the tool is named “tractor”


and it is from @dstn ?! wtf… that is genius!

After a quick look at the source code, I guess it should be somewhere in the area …
But that’s really tough stuff.
/tractor/projects/desi/cfht.py (updated)

For this particular spot, there’s something weird going on. At first I thought it was going to be a small gap between images where we have no r-band = green coverage, but there is coverage, but one of the images has an unmasked flaw that pulls the signal low. I raised this with the team.

For the Legacy Surveys data processing, the code is https://github.com/legacysurvey/legacypipe – and these images are the “coadds” (not the “detection map”, which are used for initially finding stars and galaxies). The coadds are weighted.


That explains so much for e.g. why the cosmic rays that have not been filtered out are so present. and why missing image data show through as artifacts.

> print('Creating coadds...')
>         # Produce per-band coadds, for plots

have i found it? (hehe, it’s like searching for “interesting things” :slight_smile: )

Please let me/us know what the team thinks.
And thank you for sharing your time and knowledge.

Nope, I am still getting 503 errors.

Okay, it really should be back now :slight_smile: There was an outage at NERSC that caused this downtime.