Telescopes operated at the diffraction limit suffer of straylight issues. This is because the geometry of the telescope itself and the limited optical quality of optical elements (e.g. micro-roughness of mirrors) leads to scattering of photons within the beam path. In ground-based observations, dust on optical surfaces and turbulence-induced changes of the refractive index of Earth's atmosphere ('seeing') further increase the amount of straylight.

The contamination by straylight can be quantified via the point spread function (PSF) of the telescope. Deconvolution of the observed data with a measured telescope PSF allows to remove straylight in an a posteriori process. Since seeing changes the PSF on timescales of seconds, it is not possible to remove the atmospheric straylight using this method. In that case, the PSF can only be approximated by a combination of analytical functions of Gaussian or Lorentzian shape. In solar physics, the approximated PSF is sometimes justified by a comparison of the granular contrast in observation with the one in numerical simulation.

In this contribution, we will demonstrate how a deconvolution with an approximated PSF can yield unwanted side effects when applied to spectroscopic data. The deconvolution does not only change the intensity in spatial dimension (at a given position of the image), but also in spectral dimension (it alters the position of spectral lines). Using data obtained with the Japanese solar space telescope HINODE, we show that the flow-field of sunspot penumbrae is strongly influenced by the amounts of straylight correction. An over-correction may even change the sign of a flow.