Line Searches

Lines in the spectra are identified by fitting a polynomial to the spectrum, and then searching for deviations from the fit. First, a polynomial of order fit_order1 (default 1) is fit to the whole spectrum. Subsequently, all points from the original spectrum which deviate by more than DET_THRESH$\times\sigma$ from the fit are removed from the fit region, where sigma is the rms of the deviations of the spectrum from the fitted polynomial. For each deviant point removed from the fit region, the 2 neighboring points are also removed. A new polynomial is then fitted to the reduced fitting region, and the whole process is iterated until the fitting region is constant or the maximum number of n_iteration iterations has been reached. After n_iterlow iterations, the order of the polynomial is increased to fit_order2.

After this iteration, contiguous regions are identified among the rejected points, i.e. neighboring points which all fit the rejection criterion. Contiguous regions which contain at least MIN_NPOINTS points are considered to be spectral.

After subtracting the polynomial from the spectrum, a Gaussian is fitted to each of the line regions separately. The fit is weighted by the error estimate of the spectrum. The parameters $\sigma_i$, peak flux P_i and wavelengths $\lambda_i$ are recorded, where i labels the lines.

Subsequently, a search for overlapping lines is carried out. If any two lines are overlapping in the sense

$$(\lambda_1-\lambda_2) < {\tt blend\_factor} \cdot ( \sigma_1-\sigma_2),$$ (28)

then those 2 lines are merged, i.e. a single Gaussian is fit to the 2 contiguous regions identified before.

The five lines with the highest signal/noise ratio are finally recorded both in the catalogue and in the spectrum header.

In figure 3.17, an example of an output spectrum is shown. Lines are numbered in order of significance (Line 0 being the most significant and Line 4 being the least). Note that the broken horizontal line along the bottom of the graph shows the continuum through which the polynomial was fit. The solid darker lines are the Gaussian fits through the line.

 

Figure 3.17: Example of the paper product delivered by Calnic C. The extracted spectrum is shown with error bars. Superimposed is the fit of a polynomial to the continuum. The bold lines are the Gaussian fits to the identified lines. The regions of the polynomial fit as well as the lines are indicated at the bottom of the plot. The lines are labeled starting with the most significant line.