I'm doing some forest change detection across several years using L2_Landsat8 Surface Reflectance data. And I am implementing the method available in R (landsat::relnorm) for relative radiometric normalization of the images, so they can be compared between dates. But someone mentioned that the SR_Landsat8 images are already normalized, although I haven't managed to find any information confirming this. Not sure if he was referring to something else.

Could someone clarify whether I need to normalize all the images of the time series using a master image or whether Level 2 Surface Reflectance data are ready to be compared between dates?

1 Answer 1


The surface reflectance product should be all you need. This is a level 2 product, so it is radiometrically calibrated and atmospherically corrected. Hence it should be ready for comparison in a time series. See also Landsat 8 Collection Level 1 and 2

You need a radiometric calibration to get your image from digital number (the raw output of the sensor) to a physical quantity (reflectance). https://gis.stackexchange.com/a/173631/134898 gave a great explanation on this. Usually, for satellite images there are processing algorithms in place, like the Landsat Surface Reflectance Product, so you don't have to do all that yourself (unless you really want to have more control over your data).

  • Thanks danscr. Sorry, I did't explained myself correctly. I understand that there is the Radiometric calibration that converts DN to Reflectance, but the Relative Radiometric Normalization it is used to make images (already converted to Top of the atmosphere Reflectance or Surface Reflectance) but obtained at different dates comparable, so is possible to perform change detection analysis. The method uses a master image (the best image of the time series, and all the other images get "corrected" in relation with the master. And is about this later correction I am asking about.
    – Ana
    Jan 10, 2019 at 15:55
  • I don't have first hand experience with time series analysis, but as far as I know it's common to use just the surface reflectance. I believe the relative normalization is more useful in scenarios where sensor calibration, atmospheric correction etc. are not available. From a physics point of view surface reflectance should be fine, you're analysing the (potential) change of the reflected signal of a surface against (potential) changes of the surface.
    – danscr
    Jan 11, 2019 at 9:03
  • The documentation of the landsat R package raises some more concerns. It says there the correction is one of the most aggresive ones. Without being too familiar with the method, I'd say it could mean you could lose some valuable information in your time series, i.e. some breakpoints might disappear. This method might be intended for level 1 products to make an own calibration. On the other hand, if you do see a lot of noise in your time series, you could consider a moving average over space or time. I can't give a definite answer though what's best as this isn't my area of expertise.
    – danscr
    Jan 11, 2019 at 9:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.