Continuing with our exploration of the weird effects we are seeing in the different u-band images, I wanted to explore how things might differ between the different bands. The first thing I tried was looking at masking effects. I took u and u* band patches in the same region of the sky, and got an even distribution of cutouts. I then estimated the background for each cutout, and masked the cutout as well. This plot shows the difference in the number of masked pixels.
I also check the difference in the background estimates for these cutouts. I have plotted here the difference for the estimate, as well as the RMS between u and u*. This seems to recover what we see for the estimates for multiple patches, showing that the background estimate is lower for u, but the background RMS is higher for u*.
As we discussed, we also wanted to see some profiles extracted from u or u*. I generated a simulated galaxy and convolved it with a u-band PSF to give us a simple model to test. I didn’t change the galaxy since we are only looking at changes in the profiles at different places in the image.
I added this galaxy to each cutout and masked out companions. I then extracted the profile, saving the ones that succeeded for both u and u*. The below figure shows the differences between the profiles of 63 cutouts, as well as the median profile plotted in black. It appears that u-band is slightly fainter on average than u*, but definitely not by much.
I also did some work figuring out how to properly use the Scoville Densities data. It is a single datacube but it also has a 3D WCS. I had to do some work fixing the WCS because for whatever reason it was not loading in Python. Doing some checks leads me to believe it is working fine now.
From the README I can construct 3 arrays of redshift information, containing the central, low, and high values for each slice. When we “collapse” this information, we can simply collapse the arrays in the same fashion and we will be able to seamlessly use the data at any level of coarseness we need.
Here is a plot showing the slice width as a function of redshift, based on the arrays I have available.
Lastly, the test of the model matrix code on i-band images went smoothly. I do not see any downward trends in the i-band images, giving me confidence that is the input images causing the different effect. However, some thing I DID notice was that the cutoff values are different between the i-band test and the original results. Not by much, and the trend is still there as expected, but I think the code is quite sensitive to randomness. After all, we have random selections all throughout the workflow (KDE selection, image selection, and bootstrapping). It makes sense to me that we won’t get the same values every time.
To explore this, I have 10 jobs currently running on Cedar. All of these run the model matrix code on i-band images, which will give us multiple trials on a single band. This should allow us to have some better knowledge on the uncertainty of the cutoff values themselves, which I think is very important for the paper.
HSC Research 