I have to perform change detection of several years for Central Indian Landscape. I have corrected the atmospheric aberrations of the Landsat 8 imageries using the semi-automatic classification plugin of QGIS and then mosaicked them using the merge tool. One problem I am seeing is that the imageries have differences in spectral signatures for the same land feature (say, for example, agriculture area).
Would it be better if I do the land use land cover classification first and then merge the classified layers?
Please note I will be doing supervised classification in ERDAS Imagine 9.2.