Thread by @mikewulder

@K_VanTricht Relative diffs in digital numbers is key for classifier. But, some level of correction typically needed over large areas (multiple scenes), e.g., TOA to calibrate and accommodate for differences in view angle, etc.

@K_VanTricht If using ARD, suggest surface reflectance as physical values and more robust over time and space. Enable use of same image database for change detection and empirical modeling.

@K_VanTricht Lots to consider! DN -> TOA -> SR, when and how? Topic has developed, but great (classic!) paper on topic:

Song et al. 2001. Classification and Change Detection Using #Landsat TM Data: When and How to Correct Atmospheric Effects? #landcover

sciencedirect.com/science/articl…