Astronomy and Astrophysics, volume 535A, 106-106 (2011/11-1)
Automatic stellar spectra parameterisation in the IR CaII triplet region.
KORDOPATIS G., RECIO-BLANCO A., DE LAVERNY P., BIJAOUI A., HILL V., GILMORE G., WYSE R.F.G. and ORDENOVIC C.
Abstract (from CDS):
Galactic archaeology aims to determine the evolution of the Galaxy from the chemical and kinematical properties of its individual stars. This requires the analysis of data from large spectroscopic surveys, with sample sizes in tens of thousands at present, with millions of stars being reached in the near future. Such large samples require automated analysis techniques and classification algorithms to obtain robust estimates of the stellar parameter values. Several on-going and planned spectroscopic surveys have selected their wavelength region to contain the IR CaII triplet (~λλ 8500Å) and the work presented in this paper focuses on the automatic analysis of such spectra. We aim to develop and test an automatic method by which one can obtain estimates of values of the stellar atmospheric parameters (effective temperature, surface gravity, overall metallicity) from a stellar spectrum. We also explore the degeneracies in parameter space, estimate the uncertainties in the derived parameter values and investigate the consequences of these limitations for achieving the goals of galactic archaeology. We investigated two algorithms, both of which compare the observed spectrum to a grid of synthetic spectra, but each uses a different mathematical approach for finding the optimum match and hence the best values of the stellar parameters. Our investigation of these algorithms' robustness can be widely applied because it amplifies the main problems that the other methods can encounter. The first algorithm, MATISSE, derives the values of each stellar parameter through a local fit to the spectrum such that each pixel in wavelength space is treated separately. The sensitivity of the flux at each wavelength to the value of a given stellar parameter is determined from the synthetic spectra. The observed spectrum is then projected using these sensitivity vectors to give an estimated value of the stellar parameters. This value depends on finding the true minimum in the fit and the algorithm must avoid being trapped in false local minima. The second algorithm, DEGAS, uses a pattern-recognition approach and consequently has a more global vision of the parameter space. The best-fit synthetic spectrum is derived through a series of comparisons between the observed and synthetic spectra, summed over wavelength pixels, with additional refinements in the set of synthetic spectra after each stage, i.e. a decision tree. We identified physical degeneracies in different regions of the H-R diagram: hot dwarf and giant stars share the same spectral signatures. Furthermore, it is very difficult to determine an accurate value for the surface gravity of cooler dwarfs. These effects are intensified when the lack of information increases, which happens for low-metallicity stars or spectra with low signal-to-noise ratios (SNRs). Our results demonstrate that the local projection method is preferred for spectra with high SNR, whereas the decision-tree method is preferred for spectra of lower SNR. We therefore propose a hybrid approach, combining these methods, and demonstrate that sufficiently accurate results for the purposes of galactic archaeology studies are retrieved down to SNR∼20 for typical parameter values of stars belonging to the local thin or thick disc, and for SNR down to ∼50 for the more metal-poor giant stars of the halo. If unappreciated, degeneracies in stellar parameters can introduce biases and systematic errors in derived quantities for target stars such as distances and full space motions. These can be minimised using the knowledge gained by thorough testing of the proposed stellar classification algorithm, which in turn lead to robust automated methods for the coming extensive spectroscopic surveys of stars in the Local Group.
stars: fundamental parameters - stars: abundances - techniques: spectroscopic - methods: data analysis