It is always better to use the reflectance data to achieve your work. Actually, there are two kinds of reflectance data: one is at-satellite reflectance (without atmospheric correction), the other is land surface reflectance (atmospheric effects are taken into account). For the first one, you can just use any calibration procedure provided by different software, e.g., ENVI5.1/5.2/5.3 to convert at-satellite radiance (DN) into at-satellite reflectance. Regarding the second one, you have to conduct atmospheric correction, either using image-based approaches such as COST model by Chavez (1996) with which the haze value of each band was determined by DOS (dark-object subtraction) (Chavez 1988) [Note: the "Coastal" band may not be corrected but it will not influence your forest/land cover mapping work even you do not use this band], or using FLAASH model to conduct such correction.
Thanks for your answer. I have already converted the data to at sensor reflectance, but I was wondering if this can be used for forest type classification as well. Or is it better to get land surface relectance (the second one you mentioned) and then do the classification to get more better results.
As long as you are deriving your training data (or spectral signatures) for classification from the same image, then may not need to use Surface Reluctance. But if you applying some automated feature recognition or you want repeatable training data, then you might need to the conversion,
Actually I have 4 different Landsat 8 scenes and my study area falls in all these scenes and I want to do unsupervised classification. So I was thinking to convert to surface reflectance. But manual conversion through formula in Arc Map seem quite confusing. Do you have any document/manual for converting to surface reflectance