Added new page on blending bias

[thuiop]

Issue #27

I added a page corresponding to the blending bias in the shear biases section, including a quick introduction to what is blending, some examples of how blending can affect shear measurement, and a few numbers and references on how much the bias amounts to (from DESY3 and Metadetection paper, as well as a 2011 paper for selection blending bias).

[thuiop]

One thing I would like some feedback on : the examples are made are really simple but also really exaggerated (the shear is huge in particular, which is not very realistic for weak lensing) ; do you think it would be better to have some more realistic blends, even if it is less visual ? Do you have any idea of additionnal examples I could provide, or do you think those are enough ?

[aguinot]

First I try to give you some answers:

• do you think it would be better to have some more realistic blends, even if it is less visual ?: This is fine, I prefer something visual (even if not realistic) in this case.
• Do you have any idea of additionnal examples I could provide, or do you think those are enough ?: I think you have covered most situations. The only missing one would be the impact of different redshift vs same redshift but you would thousands of galaxies for that so I think it's fine!

Now I have some comments:

• In 4.4.1: we need to refer to a general definition of shear bias which explain what it is and why this is important. I don't think we have such section at the moment (right?)
• In 4.4.2: You should make subsection here. like "No blend", "Blending effect", "Blending sensitive to shear", "Deblending"
• In 4.4.2: For the first example (galaxy alone) you should make it clear that this is the "truth" you will refer to for the other example.
• In 4.4.2: For the other I would add the bias in addition to the measured shear like: g1 | (g1-g1_true)/g1_true to emphasize the effect of the blend.
• In 4.4.2: It could be nice to show the light profile in 1D to clearly see the overlap. Without noise it is easy to do with profile=np.sum(gal_img, axis=0) and then you can do the histogram plt.hist(profile, histtype='step'). And what would look very incredible is to have two color one for each galaxies and a third line for the combination of the two profile. You could add that next to image. Just a suggestion, fill free to ignore it.
• In 4.4.2: I think the picture 5 is not necessary.
• In 4.4.2: Maybe you should make clear that the only difference between the two blended case is only the shear.
• In 4.4.2: when you mentioned SExtractor and SEP a ref (SExtractor and SEP) could be nice if the reader want to have a look. You can also point to the section 4.1 of the SExtractor paper that explain how the deblending is done.
• In 4.4.2: Maybe an example where a simple deblending fail to justify the need for more powerful tools could help..
• In 4.4.3: You can remove the "-" sign in front of the bias since it is expressed as a percentage.
• In 4.4.4: You should say that part of the bias here is also coming from the fact that you don't identify all the blends.

We can go through this during the telecon if we have time. I think most of them minor things. I think you have covered the problem very well!

[sfarrens]

Follow up from today's discussion, a few things to do before the next session (12/03/21):

• Pull the latest commits from the develop branch
• Move the blending bias page to the "BIAS AND CALIBRATION" section
• Create a new section called "PRE-PROCESSING" in-between the introduction and the PSF section
• Extract the deblending aspects of your notebook and put them in this new section