LAMOST Stellar Atmospheric Parameter System

With the launch of LAMOST Survey, more than 10000 stellar spectra will be getted on each observing night. Spectra contain important informations about celestial bodies. Extraction of the atmospheric physical parameters of stars through the stellar spectra is a basic work in astronomy. The study of stellar spectra of the celestial bodies plays an important role. Numerous methods have been developed in order to extract atmospheric parameter estimates from stellar spectra in a fast, efficient, and automated way. In this study, a stellar atmospheric parameter extraction system is designed, implemented to meet the requiement of LAMOST. The main research work are as follows:1. Preprocess for the observed spectra of LAMOST. By comparing the observation wavelength to laboratory wavelengths of 11 strong absorption lines, we can calculate the radial velocity. In order to get a good continuum fitting of 3850-9000 A, we first divide the spectrum to two parts:blue (3850-6000A) and red (6000-9000A). The blue and red part are fitted by polynomial separately, and then connected to be fitted by polynomial gain. In this system, the line indices of 83 characteristic lines are calculated for feature extraction.2. Grid template matching is used to extract stellar atmospheric parameters. Two sets of theoretical spectra template grid are generated using the Kurucz model. One set contains the g-r color index and normalized spectrum of 4400-5500A, the other set contains only the normalized spectrum of 4400-5500A. The distance between observed spectra and the theoretical template spectrum is defined, and Nelder-Mead algorithm is used to search the minimum value. The parameters of the closest theoretical spectrum is believed to be the atmospheric parameters of the observed spectra. Monte Carlo Method is used to simulate the noise distribution, in order to obtain the error of stellar atmospheric parameters.3. PC A is used to reduce dimensionality of stars spectral data, the Neural Network is used to extract stellar atmospheric parameters. Red and blue part of the spectrum are reduced to 25 dimensions separately.50 dimensions are regarded as the neural network's input, the three atmospheric parameters as output. A three-layer Neural Network is built with10 hidden intermediate nodes. Theoretical spectra and SLOAN spectra (measured by SSPP) are used as training data and two sets of neural network system are obtained.4. Stellar atmospheric parameters are extracted through the chi-square minimization technique. First of all, two different sets of theoretical spectra templates are generated, and chi-square distance between the observed spectrum and theoretical spectra are defined. Half power point (HPP) is used to estimate the initial temperature to reduce the computation, and then polynomial fitting technology with pruning is used to get the minimum, so the effective temperature is obtained. Surface gravity and metal abundances values are obtained in the same way. The second set of template will use the same temperature as getted by the first set, but the second set of templates will calculate alpha element abundances in the future. In this system, effective temperature predicted from observation g-r color index and Ballmer (Blamer) lines strength are introduced. Two theoretical and three empirical temperatures estimates are obtained finally.5. Galactic Open and Globular Clusters are used for Validation of metal abundances. The parameters extracted from high-resolution spectra of other telescope are assessed to be true values. These true values of metal abundances are used to correct the result of our system. The offset and dispersion of every algorithm are obtained. The weights are given by their offset and dispersion, and new results are re-weighted to obtain a good accuracy.