Can anyone help me in suggesting methods for change detection for landsat-8 and landsat-5 images?

The software that I am using is ArcMap 10.2 and i need to find the increase in builtup area from the two images.

I tried it in ArcMap 10.2 using image analysis tools's difference tool but unfortunately it is not enabled.

My work is to get the increase in builtup area from landsat 2005 image and landsat8 2015 image.

  • Welcome to GIS SE! As a new user be sure to take the Tour. As it stands I think your question is too broad because you have not included any details about what you have researched (and tried) prior to posting. So we know where you are up to and where you are stuck please use the edit button beneath your question to revise it with additional details.
    – PolyGeo
    Commented Jul 11, 2015 at 4:53

1 Answer 1


Since your're working in a cross-sensor scenario (in which the 2 images you are trying to compare possess slightly different ranges for the spectral channels and probably have also different radiometric resolution / number of channels / etc) direct image differencing cannot be done, so don't worry about the difference tool. It's likely that you will obtain bad results, unless you're lucky (images are in reflectance and changes easy).

The simplest solution is to first classify images (you'll need to create single ground truths for each image), and then compare maps. This is called post-classification comparison. After this process you may need to filter the change detection, since many transitions obtained this way are either wrong or even impossible to obtain.

A second solution, is to resample pixels to the same size (images as well, possibly), stack both images in a single hypercube, and classify this data with a ground truth indicating all the transitions occurring (both no changes and changes). This is called direct multidate classification.

A third solution, which (to my knowledge) will require some coding, is to apply a projection/rotation/transformation to each single time image so that the distribution of the unchanged pixels in the transformed space are matched. Them, you apply standard image differencing-based methods. This is called feature-representation-transfer, a specific strategy from domain adaptation literature.

I guess that in your situation, the first one is the simplest to implement. If you still have some questions, don't hesitate to ask, as I personally worked a bit in cross-sensor change detection stuff!

Good luck!

(PS: I don't understand why your question has been flagged as too broad, honestly. It seems a legit one! )

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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