WL data processing requirements for DC2 validation

[EiffL]

I'm opening this thread to document the WL working group needs in terms of data products for DC2 validation, and ultimately for DC2 science-ready data products. This will serve as an input to the DM-DC2 and Data Access task force.

The discussion in particular has been on which bands to run through metacal, how to pipe the metacal photometry to the photo-z pipeline, and what area do we need for robust validation.

Thanks to Heather we now have a confluence page where these requirements can be written down:
https://confluence.slac.stanford.edu/display/LSSTDESC/WL+requirements

Pinging some relevant people @joezuntz @esheldon @johannct @heather999

[esheldon]

The inclusion of bands with poorly determined PSFs is an ongoing issue we are dealing with in DES. It is probable that ultimately we will treat g separately.

However, the shear-dependent detection effects will dominate for these DC2 runs. I propose that for DC2 we just keep it simple and do griz all together, assuming that any biases will be sub-dominant.

[johannct]

For the record, current setup, straight from the package example files, is riz

[EiffL]

Just to link the two together, the decision we make here should answer LSSTDESC/ImageProcessingPipelines#108

[EiffL]

@joezuntz gave a very nice overview today at the 3x2pt telecon of the impact of using 3 (r,i,z) or 4 (g,r,i,z) bands in metacal for photometric binning.

To summarize the discussion:

• We need metacal photometry simply for the purpose of putting galaxies into photo-z bins, this can be done with any number of approaches (like BTDs, or SOMs) and does not necessarily require the main Photoz pipeline to be ran on this metacal catalog.

• When including the photometry from the g band, Joe gets much cleaner bins
  
  (Fraction of objects placed in the correct photo-z bins)
  which in turn translates into much better FoM:
  
  So, it makes it very likely we want to have the 4 bands for that purpose

• Now, the issue is that adding the g band may degrade the quality of the shape measurement, in particular we don't know how well the PSF is modeled in the g band.

At the end of the discussion, @esheldon recommended using the g-band for now in DC2.

[katrinheitmann]

@fjaviersanchez @nsevilla Can you take a look at this please and see if you want to track it as part of the SRV group activities? Thanks!