Photometric redshifts

[ran zhang]

I have reviewed the paper titled "The clustering of DESI-like luminous red galaxies using photometric redshifts" and took note of the mention of utilizing photometric parameters with machine learning methods to predict a σNMAD of 0.02106.

However, when I created my dataset using DESI DR9 data and calculated the σNMAD based on the official predicted redshift and spectral redshift, I astonishingly achieved a value as low as 0.009 or even lower (ensuring to use data with z_training=False).

I am curious to understand whether this discrepancy is due to the randomness in the data I selected or if the DESI DR9 data utilized a model with better performance for its predictions. 

[Rongpu Zhou]

Hi Ran,

The ~0.02 photo-z error that you quoted is for the DESI LRGs. The z_training=False objects are mostly SDSS and BOSS galaxies (which are downsampled from the training data) and have much smaller photo-z errors, and are presumably what you are selecting.

Best,

Rongpu