HSC simulations notes from 20/05/2021

[andrevitorelli]

• degraded images fdrom HST
• Problem: HSC is really good (seeing)
• Very deep, deeper than HST, you have to add noise, and gets to the limit of what you can do.
• using galsim
• using hsc pipeline PSFs - real data - coadded as HSC does (using same transformations as coadded imgs)
• steps: simple profiles, simple but using real gal images, testing how results vary with sim realism. If results dont change with some increased complexity then methods are robust.
• (1+m)true + c = obs
• every method will have m, c. tolerance on those depends on the accuracy you want in final results
  (...)
• exclusion factor: only galaxies that are isolated (original) - ie. no blending.
• third catalogue: some blends
• discrepancy sims/real data: blending factor, shape distributions, etc.

Thomas:

• what was the process for building COSMOS catalogue?

François:

• Mandelbaum 2012
• Leauthaud 2007
• GOOD_ZPHOT_SOURCE=1 caused problems in selection, because it will prefer deblended objects

Axel:

• When build the final sample, what happens in crowded fields with the detection?

François:

• we force the detection on the correct position

Axel, François:

• this was not a test of only shape measurements, but whole pipeline

Axel:

• advantage of simulating stamps instead of the full image?

François:

• you need to use cutouts

Axel:

• did galaxies have 90deg rotations

François:

• Yes

continuation:

François:

• fields have different properties (noise, seeing, etc)
• checks against different fields
• check of representativeness: different distributions of resolution (size), cmodel s/n
• for these checks to match degradation of images need to be correct

Andre:

• how do you control for ground truth of ellips

François:

• you dont
• lots of galaxies needed, you can measure average ellipticities. get 90 deg rotations to be sure avg cancel out

Andre:

• do you add anti-correlated noise when shearing the real image artificially?

François:

• not need to, because degradation adds much more noise

Axel:

• how do you apply shear on objects that are close.

François:

• the main result is multiplicative bias; it is a function of S/N
• additive bias depend on size and s/n
• vcompare differences of m, a beween diferent simulations
• greatest impact is the blends

Axel:

• what explains that having a model is good enough?

François:

• subdominant effect

Axel:

• can we do objects that have more different redshifts (blending biases w redshift)

Axel:

• when using real galaxies, we care only about shear not individual ellips.