The flux method

For the stars we are observing ($4000 K \le T_\mathrm{eff} \le 6300 K$) and in the wavelength interval we are using (440 nm to 680 nm) the lines are only in absorption. So, every feature that overruns the continuum by more than the Poisson photon noise has to be removed. This assumes that the continuum is well defined.

In metal-rich stars the lines are so numerous that the mean and median values of the flux seriously underestimate the continuum. There are so few true continuum points that they cannot be successfully fit by a least-squares approach. Instead, the continuum is estimated by the value of the 50^th highest pixel of the order. A highest pixel was not chosen to avoid the strongest part of the Poisson photon noise. But in the most metal rich stars, even the value of the 50^th pixel slightly underestimates the continuum. Therefore an index has been defined, which is the number of pixels included between the continuum plus one sigma and the continuum minus five sigma, divided by the total number of pixels of the order. We have assumed Poisson noise statistics, and taken sigma equal to the square root of the estimated continuum expressed in number of photo-electrons. This index is well correlated with the difference between the true and the estimated continuum, because it gives an insight of the number and width of the lines which affect the shape of an order. An order with a few small lines will have a high value of its index and a continuum well estimated, whereas an order with many broad lines will have a small value of its index, few points belonging to the true continuum (often less than 50) which will be underestimated. From the estimated continuum and the index, the limit of the Poisson photon noise is determined and the pixels with a higher value are removed, giving them the flag value -100.0.

This method is especially useful for the cosmic rays that fall in the continuum (as opposed to those that fall into the lines).