Neural-network photo-z solutions have been derived from grizY DESDM coadd catalog of full-depth observations of the VVDS-Deep 2-hr field.


  1. grizY observations to full DES depth (800 sec each in g,r, 1000 sec i,z, 500 sec Y) or greater, under main survey observing conditions, of VVDS-Deep 02hr field (RA = 02:26, Dec = -04:30)
    • VVDS-Deep CDFS field (RA = 03:32, Dec = -27:48) can serve as an alternative, but isn't as ideal as it has fewer redshifts. Can however be used in addition to the 02 hr field data.
    • If time available, u-band data totaling 400 sec or more in exposure time will be useful to test the factor of 2 improvement in photo-z sigma that is forecast for z < 0.3 galaxies when u band is included
  2. Standard DESDM processing of above data to generate coadd images and catalogs, with PSF homogenization preferred, though not required
  3. SExtractor coadd catalogs including grizY model fitting photometry and detmodel colors


  1. SExtractor catalog objects will be positionally matched to VVDS-Deep high quality (flags = 3,4) spectroscopic redshifts
  2. Matched objects will be divided randomly into training and validation samples
  3. Photo-z solution will then be derived from training and validation sets using DESDM photo-z module code: neural network photo-z's plus nearest neighbor error estimates
    • Input quantities to neural network are grizY MAG_AUTO and MAG_DETMODEL from SExtractor catalogs
    • Output quantities are photo-z's and photo-z errors


PZ-G1 PASSED: Solution was derived from SV observations of the supernova X3 field, containing the VVDS-Deep 02 hr field. Here is the photo-z scatter plot for the validation set: photoz_sv_vvds02hr_Ygri3z3_valid.png, and full details are posted here,

Quality of photo-z solution will be assessed under goals PZ-G2, PZ-G3, and PZ-G4


Quality of photo-z solution will be assessed under goals PZ-G2, PZ-G3, and PZ-G4