2005AJ....130.2804R

We develop and test a method for the estimation of metallicities ([Fe/H]) and carbon abundance ratios ([C/Fe]) for carbon-enhanced metal-poor (CEMP) stars based on the application of artificial neural networks, regressions, and synthesis models to medium-resolution (1-2 Å) spectra and J-K colors. We calibrate this method by comparison with metallicities and carbon abundance determinations for 118 stars with available high-resolution analyses reported in the recent literature. The neural network and regression approaches make use of a previously defined set of line-strength indices quantifying the strength of the Ca II K line and the CH G band, in conjunction with J-K colors from the Two Micron All Sky Survey Point Source Catalog. The use of near-IR colors, as opposed to broadband B-V colors, is required because of the potentially large affect of strong molecular carbon bands on bluer color indices. We also explore the practicality of obtaining estimates of carbon abundances for metal-poor stars from the spectral information alone, i.e., without the additional information provided by photometry, as many future samples of CEMP stars may lack such data. We find that although photometric information is required for the estimation of [Fe/H], it provides little improvement in our derived estimates of [C/Fe], and hence, estimates of carbon-to-iron ratios based solely on line indices appear sufficiently accurate for most purposes. Although we find that the spectral synthesis approach yields the most accurate estimates of [C/Fe], in particular for the stars with the strongest molecular bands, it is only marginally better than is obtained from the line index approaches. Using these methods we are able to reproduce the previously measured [Fe/H] and [C/Fe] determinations with an accuracy of ~0.25 dex for stars in the metallicity interval -5.5≤[Fe/H]≤-1.0 and with 0.2≤(J-K)₀≤0.8. At higher metallicity, the Ca II K line begins to saturate, especially for the cool stars in our program, and hence, this approach is not useful in some cases. As a first application, we estimate the abundances of [Fe/H] and [C/Fe] for the 56 stars identified as possibly carbon-rich, relative to stars of similar metal abundance, in the sample of ``strong G-band'' stars discussed by Beers, Preston, and Shectman.