artifacts in OSIRIS data

[jeansom]

Hi, I am reducing OSIRIS H-band data of a globular cluster in the Fornax dwarf galaxy. This data was taken in Oct. 2019 with the 0.1” plate scale. It consisted of four 900 sec science exposures and three 900 sec sky exposures.

I've attached some plots of the reduced data for one of the science exposures. I am using the latest version of the DRP and have used apply_mask.py script with the 15 sigma mask. The top spectrum shows the spectrum before sky subtraction, and I've marked a handful of downward spikes with red lines. I see these spikes in all of my exposures, and they are always at the same wavelengths. Do you know what could cause such spikes? Below that spectrum, I have also shown a sky subtracted spectrum. The spikes are somewhat removed, but not very well.

On the left of the plot, I've shown a slice of the cube at the wavelength of one of the spikes. The color shows the flux, as shown in the colorbar. They don't seem to be caused by one bad patch, but show up as almost a flux gradient.

Do you know what could be causing the negative spikes? This is the first time I have used OSIRIS data, so is this kind of thing expected? I have also compiled the xml files I'm using for the reduction into one file and have attached it.
fornax.txt

[followthesheep]

A couple of questions: is this just this night or do you see this on other nights? What is the fraction of the flux in the dips compared to the peaks?

This may be related to the flux assignment issue that is a known issue with the way the pipeline works. Because the spectra are very closely packed together, the deconvolution process can result in positive and negative flux 32 channels away (total flux is still conserved). This becomes more prominent for bright lines. Typically, these residuals are about 6-7% of the peak flux. In effect, this adds extra correlated noise to the spectrum. We are investigating ways for better extraction, but at this time, this is the best the pipeline can do.

For more information about this, see there are two OSIRIS hackathon reports that are useful to read:

https://drive.google.com/file/d/0B_YkzZoUSrX-RENMLUpvX2FWTlE/view?usp=sharing

https://drive.google.com/file/d/0B_YkzZoUSrX-YnpCRjVZRkRPWnM/view?usp=sharing

[jeansom]

I only have observations from this one night, unfortunately. The peaks and dips do seem to line up with +- ~32 pixels, and the flux ratio in the residuals vs the peaks tends to be roughly 6-7% like you said. Okay, looks like this is most likely the issue you mentioned. Thank you so much for the help!

One more quick (separate) question: I noticed that line 321 of mosaic_000.pro (link) calculates the noise in each pixel as sqrt(sum(noise^2 in each pixel)/N). My naive understanding of how this code is working makes me think it should be sqrt(sum(noise^2 in each pixel))/N (ie, take the N outside of the sqrt). Am I misunderstanding? Thanks.