SMARTS 1.5m reduced spectral format

The reduced data are saved in a FITS file. The data are saved in an
N x M vector. N is the number of points along the spectrum.
M is 6 if background is subtracted, and 4 otherwise.
If the data are flux calibrated, then M=8 or 6.

If background has been subtracted, the individual vectors are:
   M=0: background-subtracted spectrum extracted using boxcar extraction
   M=1: background vector extracted using boxcar extraction
   M=2: error on the spectrum
   M=3: error on the background
   M=4: spectrum extracted using a Gaussian extraction
   M=5: background vector extracted using Gaussian extraction
   M=6: flux/count. Multiply data(*,0) x data(*,6) for the flux.
   M=7: flux/count. Multiply data(*,1) x data(*,7) for the flux.

If no background has been subtracted, the individual vectors are:
   M=0: spectrum extracted using boxcar extraction
   M=1: error on the spectrum
   M=2: spectrum extracted using a Gaussian extraction
   M=3: background vector extracted using Gaussian extraction
   M=4: flux/count. Multiply data(*,0) x data(*,4) for the flux.
   M=5: flux/count. Multiply data(*,1) x data(*,5) for the flux.

The wavelength vector is linear, with the wavelength of the first
point and the reciprocal dispersion (A/bin) given in the header
keywords WAVE_0 and DELTA_W, respectively

Extraction methods:
   Boxcar extraction: Extract counts within +/- X pixels of the peak.
   Extract background in regions to either side of peak. Subtract median
   background level.
   The default width is determined by fitting a Gaussian to the spatial cut.
   This method can give more accurate absolute fluxes, because the extraction
   region widths can be set by hand.

   Gaussian extraction: Fit the spectrum in the cross-dispersion direction
   with a Gaussian. The extracted flux is the integral of the fitted 
   Gaussian, so it is less susceptible to cosmic rays in the spectrum.
   A gaussian often fits the wings poorly when the seeing is poor (or even
   average), and so generally gets fewer counts than the boxcar extraction.
   The Gaussian extraction does not do a good job if there is another bright
   source nearby.