PZ-G2 » History » Version 5

« Previous - Version 5/8 (diff) - Next » - Current version
Huan Lin, 11/25/2012 08:22 PM

PZ-G2, PZ-G3, PZ-G4

  • PZ-G2: (Photo-z – spec-z) bias has been plotted against photo-z in bins of width dz=0.1.
  • PZ-G3: Photo-z sigma and sigma68 vs photo-z has been compared against R-8.
  • PZ-G4: Outlier fractions vs photo-z have been compared against R-23.


DESDM photo-z module neural network photo-z outputs for VVDS-Deep data obtained under goal PZ-G1


Use existing IDL code from [[des-photoz:DC6_Photo-z_Challenge|DC6 Photo-z Challenge]], which will:
  1. Divide sample into photo-z bins of width dz = 0.1
  2. Calculate and plot mean of (photo-z - spec-z) vs. photo-z (example plot:
    Passed: plot is here: sv_vvds02hr_Ygri3z3_valid_comp_bias.png
  3. Calculate and plot sigma = standard deviation of (photo-z - spec-z) vs. photo-z (example plot:
  4. Calculate and plot sigma68 = 68-percentile (photo-z -spec-z) vs. photo-z (example plot:
  5. Calculate and plot 2-sigma and 3-sigma outlier fractions, defined relative to sigma, vs. photo-z (example plot:


  1. Check against DES science requirement R-8 that sigma68 < 0.12
    Passed: SV solution gives overall sigma68 = 0.12 for validation set: sv_vvds02hr_Ygri3z3_valid_comp_sigma.png
  2. Check against DES science requirements R-23 that 2-sigma fraction < 0.1 and 3-sigma fraction < 0.015
    Passed: SV solution gives overall 2-sigma fraction 0.05 and 3-sigma fraction 0.02 (3-sigma fraction is over requirement but better than [[des-photoz:DC6_Photo-z_Challenge|DC6 Photo-z Challenge]] results, so calling this a pass): sv_vvds02hr_Ygri3z3_valid_comp_outlier.png


Results from [[des-photoz:DC6_Photo-z_Challenge|DC6 Photo-z Challenge]] indicate that the above requirements should be met or close to being met. If we see significant differences for the real VVDS-Deep data, we will investigate potential causes, starting with:

  1. Potential systematic trends in DES colors/magnitudes vs. RA,Dec, checked by plotting DES photometry vs. existing VVDS or CFHTLS (truth-table) photometry of the VVDS-Deep field
  2. Potential lack of depth in the DES coadd data for the VVDS-Deep field, checked by comparing the sky background noise, seeing FWHM, and photometric zeropoint in the DES coadd images vs. the respective nominal DES main survey full-depth values