Accurate subtraction of the bright night sky emission lines in the near-infrared is crucial, given that the object being
observed is often several magnitudes fainter than the sky background. Most integral field spectrographs (IFS) have a
modest field of view (FoV), and it is often not possible to achieve good sky subtraction by nodding the object within the
FoV, as is common practice for long slit spectrographs. In principle, it should be possible to use sky background
information from one part of the FoV (typically the periphery) to subtract the sky from all other parts of the IFS FoV.
However, this has never been achieved in practice.
We show that the reason on-IFU sky subtraction does not work is that the spectrograph spectral response function (line
spread function, or LSF) varies strongly with wavelength, position within the field of view, and telescope pointing
(flexure). By micro-stepping the grating of the SINFONI IFS at the ESO-VLT, we have been able to hyper-sample the
spectral PSF and reconstruct detailed LSF profiles for all wavelengths and all field points for SINFONI H band data.
Using this information, we can conclusively demonstrate improvements in observing efficiency by over a factor of two.
Our technique not only removes the need for separate sky exposures, but can also improve the noise of the sky
background measurement itself, providing further potential gain over pairwise frame subtraction. We explain our
algorithms, including non-parametric descriptions of the LSF, and present the results from applying our method to
archival SINFONI data.