There is a neat Surface Reflectance product of Landsat imagery. I know the difference between SR and raw products: SR presents calculated values, that are supposed to represent state of the Earth surface, while "normal" products are disturbed by all kinds of scattering on the way from the Sun to the surface and back to the sensor.
Now, I'm performing a land classification of terrain to extract boundaries between two types of land use. I have many images and this is unsupervised classification. Do I really need SR product here? Would I need it, if it was supervised classification? I suppose it's no in both cases, as I don't directly compare any values between images. I just need to know exact position of this boundary.