No Arabic abstract
Machine learning techniques, specifically the k-nearest neighbour algorithm applied to optical band colours, have had some success in predicting photometric redshifts of quasi-stellar objects (QSOs): Although the mean of differences between the spectroscopic and photometric redshifts is close to zero, the distribution of these differences remains wide and distinctly non-Gaussian. As per our previous empirical estimate of photometric redshifts, we find that the predictions can be significantly improved by adding colours from other wavebands, namely the near-infrared and ultraviolet. Self-testing this, by using half of the 33 643 strong QSO sample to train the algorithm, results in a significantly narrower spread for the remaining half of the sample. Using the whole QSO sample to train the algorithm, the same set of magnitudes return a similar spread for a sample of radio sources (quasars). Although the matching coincidence is relatively low (739 of the 3663 sources having photometry in the relevant bands), this is still significantly larger than from the empirical method (2%) and thus may provide a method with which to obtain redshifts for the vast number of continuum radio sources expected to be detected with the next generation of large radio telescopes.
From SDSS commissioning photometric and spectroscopic data, we investigate the utility of photometric redshift techniques to the task of estimating QSO redshifts. We consider empirical methods (e.g. nearest-neighbor searches and polynomial fitting), standard spectral template fitting and hybrid approaches (i.e. training spectral templates from spectroscopic and photometric observations of QSOs). We find that in all cases, due to the presence of strong emission-lines within the QSO spectra, the nearest-neighbor and template fitting methods are superior to the polynomial fitting approach. Applying a novel reconstruction technique, we can, from the SDSS multicolor photometry, reconstruct a statistical representation of the underlying SEDs of the SDSS QSOs. Although, the reconstructed templates are based on only broadband photometry the common emission lines present within the QSO spectra can be recovered in the resulting spectral energy distributions. The technique should be useful in searching for spectral differences among QSOs at a given redshift, in searching for spectral evolution of QSOs, in comparing photometric redshifts for objects beyond the SDSS spectroscopic sample with those in the well calibrated photometric redshifts for objects brighter than 20th magnitude and in searching for systematic and time variable effects in the SDSS broad band photometric and spectral photometric calibrations.
Machine learning (ML) is a standard approach for estimating the redshifts of galaxies when only photometric information is available. ML photo-z solutions have traditionally ignored the morphological information available in galaxy images or partly included it in the form of hand-crafted features, with mixed results. We train a morphology-aware photometric redshift machine using modern deep learning tools. It uses a custom architecture that jointly trains on galaxy fluxes, colors and images. Galaxy-integrated quantities are fed to a Multi-Layer Perceptron (MLP) branch while images are fed to a convolutional (convnet) branch that can learn relevant morphological features. This split MLP-convnet architecture, which aims to disentangle strong photometric features from comparatively weak morphological ones, proves important for strong performance: a regular convnet-only architecture, while exposed to all available photometric information in images, delivers comparatively poor performance. We present a cross-validated MLP-convnet model trained on 130,000 SDSS-DR12 galaxies that outperforms a hyperoptimized Gradient Boosting solution (hyperopt+XGBoost), as well as the equivalent MLP-only architecture, on the redshift bias metric. The 4-fold cross-validated MLP-convnet model achieves a bias $delta z / (1+z) =-0.70 pm 1 times 10^{-3} $, approaching the performance of a reference ANNZ2 ensemble of 100 distinct models trained on a comparable dataset. The relative performance of the morphology-aware and morphology-blind models indicates that galaxy morphology does improve ML-based photometric redshift estimation.
At present, the precision of deep ultraviolet photometry is somewhat limited by the dearth of faint ultraviolet standard stars. In an effort to improve this situation, we present a uniform catalog of eleven new faint (u sim17) ultraviolet standard stars. High-precision photometry of these stars has been taken from the Sloan Digital Sky Survey and Galaxy Evolution Explorer and combined with new data from the Swift Ultraviolet Optical Telescope to provide precise photometric measures extending from the Near Infrared to the Far Ultraviolet. These stars were chosen because they are known to be hot (20,000 < T_eff < 50,000 K) DA white dwarfs with published Sloan spectra that should be photometrically stable. This careful selection allows us to compare the combined photometry and Sloan spectroscopy to models of pure hydrogen atmospheres to both constrain the underlying properties of the white dwarfs and test the ability of white dwarf models to predict the photometric measures. We find that the photometry provides good constraint on white dwarf temperatures, which demonstrates the ability of Swift/UVOT to investigate the properties of hot luminous stars. We further find that the models reproduce the photometric measures in all eleven passbands to within their systematic uncertainties. Within the limits of our photometry, we find the standard stars to be photometrically stable. This success indicates that the models can be used to calibrate additional filters to our standard system, permitting easier comparison of photometry from heterogeneous sources. The largest source of uncertainty in the model fitting is the uncertainty in the foreground reddening curve, a problem that is especially acute in the UV.
We extend the SDSS Stripe 82 Standard Stars Catalog with post-2007 SDSS imaging data. This improved version lists averaged SDSS ugriz photometry for nearly a million stars brighter than r~22 mag. With 2-3x more measurements per star, random errors are 1.4-1.7x smaller than in the original catalog, and about 3x smaller than for individual SDSS runs. Random errors in the new catalog are ~< 0.01 mag for stars brighter than 20.0, 21.0, 21.0, 20.5, and 19.0 mag in u, g, r, i, and z-bands, respectively. We achieve this error threshold by using the Gaia Early Data Release 3 (EDR3) Gmag photometry to derive gray photometric zeropoint corrections, as functions of R.A. and Declination, for the SDSS catalog, and use the Gaia BP-RP colour to derive corrections in the ugiz bands, relative to the r-band. The quality of the recalibrated photometry, tested against Pan-STARRS1, DES, CFIS and GALEX surveys, indicates spatial variations of photometric zeropoints <=0.01 mag (RMS), with typical values of 3-7 millimag in the R.A., and 1-2 millimag in the Declination directions, except for <~6 millimag scatter in the u-band. We also report a few minor photometric problems with other surveys considered here, including a magnitude-dependent ~0.01 mag bias between 16 < G_Gaia < 20 in the Gaia EDR3. Our new, publicly available catalog offers robust calibration of ugriz photometry below 1% level, and will be helpful during the commissioning of the Vera C. Rubin Observatory Legacy Survey of Space and Time.
The scientific value of the next generation of large continuum surveys would be greatly increased if the redshifts of the newly detected sources could be rapidly and reliably estimated. Given the observational expense of obtaining spectroscopic redshifts for the large number of new detections expected, there has been substantial recent work on using machine learning techniques to obtain photometric redshifts. Here we compare the accuracy of the predicted photometric redshifts obtained from Deep Learning(DL) with the k-Nearest Neighbour (kNN) and the Decision Tree Regression (DTR) algorithms. We find using a combination of near-infrared, visible and ultraviolet magnitudes, trained upon a sample of SDSS QSOs, that the kNN and DL algorithms produce the best self-validation result with a standard deviation of {sigma} = 0.24. Testing on various sub-samples, we find that the DL algorithm generally has lower values of {sigma}, in addition to exhibiting a better performance in other measures. Our DL method, which uses an easy to implement off-the-shelf algorithm with no filtering nor removal of outliers, performs similarly to other, more complex, algorithms, resulting in an accuracy of {Delta}z < 0.1$ up to z ~ 2.5. Applying the DL algorithm trained on our 70,000 strong sample to other independent (radio-selected) datasets, we find {sigma} < 0.36 over a wide range of radio flux densities. This indicates much potential in using this method to determine photometric redshifts of quasars detected with the Square Kilometre Array.