Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userPhotometric Redshifts based on standard SED fitting procedures
81
0
0.0
(
0
)
Authors
Micol Bolzonella
Ask ChatGPT about the research
No Arabic abstract
In this paper we study the accuracy of photometric redshifts computed through a standard SED fitting procedure, where SEDs are obtained from broad-band photometry. We present our public code hyperz, which is presently available on the web. We introduce the method and we discuss the expected influence of the different observational conditions and theoretical assumptions. In particular, the set of templates used in the minimization procedure (age, metallicity, reddening, absorption in the Lyman forest, ...) is studied in detail, through both real and simulated data. The expected accuracy of photometric redshifts, as well as the fraction of catastrophic identifications and wrong detections, is given as a function of the redshift range, the set of filters considered, and the photometric accuracy. Special attention is paid to the results expected from real data.
rate research
Read More
Photometric redshifts (photo-zs) provide an alternative way to estimate the distances of large samples of galaxies and are therefore crucial to a large variety of cosmological problems. Among the various methods proposed over the years, supervised machine learning (ML) methods capable to interpolate the knowledge gained by means of spectroscopical data have proven to be very effective. METAPHOR (Machine-learning Estimation Tool for Accurate PHOtometric Redshifts) is a novel method designed to provide a reliable PDF (Probability density Function) of the error distribution of photometric redshifts predicted by ML methods. The method is implemented as a modular workflow, whose internal engine for photo-z estimation makes use of the MLPQNA neural network (Multi Layer Perceptron with Quasi Newton learning rule), with the possibility to easily replace the specific machine learning model chosen to predict photo-zs. After a short description of the software, we present a summary of results on public galaxy data (Sloan Digital Sky Survey - Data Release 9) and a comparison with a completely different method based on Spectral Energy Distribution (SED) template fitting.
We present robust statistical estimates of the accuracy of early-type galaxy stellar masses derived from spectral energy distribution (SED) fitting as functions of various empirical and theoretical assumptions. Using large samples consisting of 40,000 galaxies from the Sloan Digital Sky Survey, of which 5,000 are also in the UKIRT Infrared Deep Sky Survey, with spectroscopic redshifts in the range 0.05 leq z leq 0.095, we test the reliability of some commonly used stellar population models and extinction laws for computing stellar masses. Spectroscopic ages (t), metallicities (Z), and extinctions (A) are also computed from fits to SDSS spectra using various population models. These constraints are used in additional tests to estimate the systematic errors in the stellar masses derived from SED fitting, where t, Z, and A are typically left as free parameters. We find reasonable agreement in mass estimates among stellar population models, with variation of the IMF and extinction law yielding systematic biases on the mass of nearly a factor of 2, in agreement with other studies. Removing the near-infrared bands changes the statistical bias in mass by only 0.06 dex, adding uncertainties of 0.1 dex at the 95% CL. In contrast, we find that removing an ultraviolet band is more critical, introducing 2{sigma} uncertainties of 0.15 dex. Finally, we find that stellar masses are less affected by absence of metallicity and/or dust extinction knowledge. However, there is a definite systematic offset in the mass estimate when the stellar population age is unknown, up to a factor of 2.5 for very old (12 Gyr) stellar populations. We present the stellar masses for our sample, corrected for the measured systematic biases due to photometrically determined ages, finding that age errors produce lower stellar masses by 0.15 dex, with errors of 0.02 dex at the 95% CL for the median stellar age subsample.
Using a local reference sample of 21 galaxies, we compare observations of the $lambda$2.16 $mu$m Brackett-$gamma$ (Br$gamma$) hydrogen recombination line with predictions from the Prospector Bayesian inference framework, which was used to fit the broadband photometry of these systems. This is a clean test of the spectral-energy-distribution-derived star formation rates (SFRs), as dust is expected to be optically thin at this wavelength in nearly all galaxies; thus, the internal conversion of SFR to predicted line luminosity does not depend strongly on the adopted dust model and posterior dust parameters, as is the case for shorter-wavelength lines such as H$alpha$. We find that Prospector predicts Br$gamma$ luminosities and equivalent widths with small offsets ($sim$0.05 dex), and scatter ($sim$0.2 dex), consistent with measurement uncertainties, though we caution that the derived offset is dependent on the choice of stellar isochrones. We demonstrate that even when the Prospector-derived dust attenuation does not well describe, e.g., H$alpha$ line properties or observed reddening between H$alpha$ and Br$gamma$, the underlying SFRs are accurate, as verified by the dust-free Br$gamma$ comparison. Finally, we discuss in what ways Br$gamma$ might be able to help constrain model parameters when treated as an input to the model, and comment on its potential as an accurate monochromatic SFR indicator in the era of JWST multiobject near-IR spectroscopy.
Upcoming imaging surveys, such as LSST, will provide an unprecedented view of the Universe, but with limited resolution along the line-of-sight. Common ways to increase resolution in the third dimension, and reduce misclassifications, include observing a wider wavelength range and/or combining the broad-band imaging with higher spectral resolution data. The challenge with these approaches is matching the depth of these ancillary data with the original imaging survey. However, while a full 3D map is required for some science, there are many situations where only the statistical distribution of objects (dN/dz) in the line-of-sight direction is needed. In such situations, there is no need to measure the fluxes of individual objects in all of the surveys. Rather a stacking procedure can be used to perform an `ensemble photo-z. We show how a shallow, higher spectral resolution survey can be used to measure dN/dz for stacks of galaxies which coincide in a deeper, lower resolution survey. The galaxies in the deeper survey do not even need to appear individually in the shallow survey. We give a toy model example to illustrate tradeoffs and considerations for applying this method. This approach will allow deep imaging surveys to leverage the high resolution of spectroscopic and narrow/medium band surveys underway, even when the latter do not have the same reach to high redshift.
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.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
