I need Sentinel-1 GRD data unified with Landsat-8 data. Should I calibrate Landsat to sigma nought or Sentinel to spectral reflectance ? As a result I should receive layer stack for land cover classification.
Landsat reflectance and Sentinel 1A SAR backscattering are the result of different kinds of physical processes and characteristics of the surface, which are not easily “unified”. You could think up some indices, though, that you could somehow normalise measures for use in feature vectors, but you always need to understand the physics behind what you would want to derive from that (in classification, machine learning, etc.).
That’s actually an interesting field of work, to come up with generalised rules for land cover.