QE variations with position on NICMOS detectors are strong. Because the grism images cannot be flatfielded, this leads to significant structure in the images. Since the QE variations also depend significantly on the wavelength, the expected background in both the grism image and the direct image will depend on the spectrum of the background in space. The best way to properly account for the background at the time the data were taken is to use a chopping mode so that the target occupies a different part of the detector each time it is observed. Background estimates can then be constructed for any part of the detector and subtracted from the grism/spectrum image of interest. In cases where only one single pointing is available (e.g. for parallel observations), two methods methods of background subtraction are available to the user.
The first one is to use only one grism image at the time, and to interpolate from the regions of blank sky surrounding a spectrum of interest. The second option is to use a pre-defined background and to scale it appropriately to the local background level. The former method, which is described in section 3.4.2, has the advantage that it is more robust and is free of assumptions. The later method has the advantage that it might provide a better estimate of the variations of the background at the exact position of the spectrum (see section 3.4.3).