Recent Developments in the Optimal Extraction of UVES Spectra
- PDF / 290,983 Bytes
- 5 Pages / 441 x 666 pts Page_size
- 18 Downloads / 156 Views
2 3
4
Rovsing A/S, Dyregaardsvej 2, DK-2740 Skovlunde, Denmark; [email protected] ESO, Karl-Schwarzschild-Strasse 2, 85748 Garching, Germany INAF – Osservatorio Astronomico di Trieste, via G.B. Tiepolo 11, Trieste I-34131, Italy ESO, Alonso de Cord´ ova 3107, Vitacura, Santiago, Chile
1 Introduction The UVES data reduction pipeline [1] uses optimal extraction to achieve higher signal-to-noise (S/N) of faint objects, corresponding to an increase in effective exposure time up to 70% compared with a simple aperture extraction (see Horne [2] and the further developments by Marsh [3] and Mukai [4] for an introduction). Initial releases of the UVES pipeline had limitations in the extraction quality at certain S/N ranges. We describe the implementation in version 3 of the pipeline, working in the common pipeline library CPL [5] context, which shows significant improvements with respect to earlier versions. The new implementation is used operationally since the beginning of P79.
2 Algorithm In order to measure the spatial profile a preliminary estimate and subtraction of the sky is carried out by taking the median of all pixels (after masking out the object using a rough object localization). The algorithm then follows Horne’s scheme but with the following differences: – The spatial profile is measured either using an analytical (Gaussian or Moffat) profile (as described in [4]), or by resampling the empirical profile to a grid with a resolution of 0.2 pixels in the spatial direction, and fitting a low order polynomial to the spatial profile at each resampled position.1 See also Fig. 3. 1 While resampling the data is often avoided because it introduces resampling noise [3], this resampling noise is smoothed when fitting a low-degree polynomial to the spatial profile. However, when the model profile is later used to extract the data, it is important that the model is rebinned to the sampling of the data rather than the other way around. Mukai dubbed this “virtual resampling”.
404
J. M. Larsen et al.
– In order to fully exploit the peculiarities of the echelle format the free parameters of the respective models (analytical profile centroid and width, or virtually resampled profile at each spatial position) are modelled as 2D polynomials in wavelength and order number. In this way, regions (which may span entire orders) where the profile cannot be accurately determined due to very low signal are interpolated from neighbouring regions having presumably higher S/N. – Horne’s formula for the optimally extracted flux (which is equivalent to profile fitting at every wavelength [2]) assumes that the sky background has been already subtracted, and furthermore that the interpolated sky level is effectively noise-free. Because of the short slits typically used in echelle spectrography (to ensure order separation), the assumption of a noise-free sky determination may not be valid; we therefore generalized the method to give combined optimal sky and object flux estimates by minimization of (fi − (Si + F pi ))2 (1) χ2 = σi2 i where fi and σi2 are the flu
Data Loading...
