Machine learning photometric redshifts

Apart from the template-fitting photometric redshifts included in the DR3 catalog, photometric redshifts based on two separate machine-learning techniques (MLPQNA and ANNz2) are also provided.

ANNz2 photometric redshifts

Photometric redshifts have been derived using the ANNz2 method (Sadeh et al. 2016, PASP, 128, 104502) for two KiDS DR3 subsets. One contains all sources with good GAaP photometry in all four filters (39,166,070 objects). The other includes bright KiDS galaxies (r<20.3 mag; 800,830 sources). Details of selections and photometric redshift derivation are provided in Bilicki et al. 2018, A&A, in press (arXiv:1709.04205). Please cite that paper together with the main KiDS DR3 one (de Jong et al. 2017, A&A, 604, A134, arXiv:1703.02991) if you use the data described here.

Full-depth catalog

Photometric redshifts were generated using a training sample from several spectroscopic fields. The released catalog contains all sources (both extended and point-like) with good GAaP photometry in all four filters (39,166,070 objects). Also included is a "fiducial" flag that identifies a more reliable subset, removing artefacts, stars and sources in parameter space not sampled by the training data; these filters result in a set of approximately 20.5 million objects. Depending on the application, additional filtering might still be needed to ensure reliable photo-zs. Please see Bilicki et al. (2018) for details.

Catalog format

Photometric redshifts derived using ANNz2 are provided in a single catalog, featuring the columns listed in the following table. Apart from the photo-z values, we provide source ID information and positions that allow straightforward association with the KiDS DR3 multi-band catalog, as well as basic photometric information, together with the fiducial flag mentioned above.

LabelFormatUnitDescription
ID23ASource identifier
RAJ2000DdegRight ascension (J2000)
DECJ2000DdegDeclination (J2000)
SG2DPHOTKSource classification
IMAFLAGS_ISO_bandJMask flag
MAG_GAAP_band_calibEmagCalibrated GAaP magnitude
MAGERR_GAAP_bandEmagError in GAaP magnitude
zphot_ANNz2DPhotometric redshift derived with ANNz2
fiducialIFlag defining fiducial selection

 

Catalog download

The full-depth catalog can be downloaded via the following link.

Bright-end catalog

Photometric redshifts were generated using a training sample from the Galaxy And Mass Assembly (GAMA) spectroscopic data and by applying an extended set of photometric features in photo-z derivation. The released catalog contains resolved sources (galaxies) extracted from DR3 by applying the magnitude cut of MAG_AUTO_R_calib ≤ 20.3, star-galaxy separation parameter SG2DPHOT = 0, and by requiring that all the four ugri GAaP magnitudes, and their errors, are measured; this gives 800,830 objects in total. Additional filtering, as described in Bilicki et al. (2018), is needed to remove residual artefacts.

Catalog format

Photometric redshifts derived using ANNz2 are provided in a single catalog, featuring the columns listed in the following table. Apart from the photo-z values, we provide source ID information and positions that allow straightforward association with the KiDS DR3 multi-band catalog, as well as basic photometric information.

LabelFormatUnitDescription
ID23ASource identifier
RAJ2000DdegRight ascension (J2000)
DECJ2000DdegDeclination (J2000)
IMAFLAGS_ISO_bandJMask flag
MAG_AUTO_band_calibEmagCalibrated Kron-like elliptical aperture magnitude
MAGERR_AUTO_bandEmagRMS error for MAG_AUTO
MAG_ISO_band_calibEmagCalibrated isophotal magnitude
MAGERR_ISO_bandEmagRMS error for MAG_ISO
MAG_GAAP_band_calibEmagCalibrated GAaP magnitude
MAGERR_GAAP_bandEmagError in GAaP magnitude
zphot_ANNz2DPhotometric redshift derived with ANNz2

 

Catalog download

The bright-end catalog can be downloaded via the following link.

MLPQNA photometric redshifts and PDFs

For the DR3 data set photometric redshifts have been derived using the Multi Layer Perceptron with Quasi Newton Algorithm (MLPQNA) technique, following the a similar strategy as was done for DR2 (Cavuoti et al. 2015, MNRAS, 452, 3100). Driven by the magnitude ranges covered by the spectroscopic knowledge base (SDSS DR9 and GAMA DR2), the total number of sources with MLPQNA photo-z's is 8,582,152. Photo-z probability distribution functions (PDFs) were also derived.

Catalog format

The best-guess photometric redshifts derived using the MLPQNA technique are provided in a single catalogue, featuring the columns listed in the following table. Apart from the photo-z values the only content consists of source ID information and positions that allow straightforward association with the KiDS DR3 multi-band catalogue.

LabelFormatUnitDescription
ID25ASource identifier
SLIDJAstro-WISE SourceList identifier
SIDJAstro-WISE Source identifier
RAJ2000DdegRight ascension
DECJ2000DdegDeclination
Z_MLPQNADBest MLPQNA predicted photometric redshift

 

Also available are photo-z Probability Distribution Functions (PDFs) based on the MLPQNA technique. These are provided in separate catalog files per DR3 survey tile, and their format is specified in the following table . Again, included data is limited to the PDF bins and source ID and position information that can be used to associate to the DR3 multi-band catalogue.

LabelFormatUnitDescription
ID25ASource identifier
SLIDJAstro-WISE SourceList identifier
SIDJAstro-WISE Source identifier
RAJ2000DdegRight ascension
DECJ2000DdegDeclination
PDF001DMLPQNA photo-z PDF bin 1 (z=0.01)
.........
PDF350DMLPQNA photo-z PDF bin 350 (z=3.5)

 

Catalog download

The catalog with the best photo-z estimates can be downloaded via the following link.

The PDFs are distributed in a separate FITS catalog for each of the DR3 survey tiles; the total size of these catalogs is 23 GB. Downloading of these files is possible with wget using the input files linked below. The also provided md5sums file can be used to verify your downloaded files with the command "md5sum -c kids_dr3.2_mlpqnapdf_md5sums.txt".