I'm using GRASS to perform an unsupervised classification on a time series of LANDSAT images (monthly temporal res, over 30 years), in order to compare land cover change in a particular area.

I'm concerned that if I run i.cluster for each map in the time series the signatures used for the classes may be slightly different for each map, affecting the validity of the comparison. Would this be the case? If so, is there a way to ensure that the class signatures used on each map in the time series are identical throughout?

My initial idea was to run i.cluster on a particular map where all classes are well represented, then use the sigfile generated from this as the input sigfile to i.maxlik for all maps.

  • 1
    One way to get a common signature for a collection of related grids is to displace them all to different locations and mosaic those (with some bands of null values between them all): create a signature for that. This works around the potential problem that any one image might not display the full variety of classes.
    – whuber
    Commented Nov 22, 2012 at 0:19
  • 1
    Thanks @whuber. This sounds like a good idea, however, the area I'm looking at consists of 11 LANDSAT tiles mosaicked together. Disolacing these and mosaicking them together will produce a massive file and take a long time. Instead, could I make an image group of all the points in the time series and generate a signature for this?
    – si_2012
    Commented Nov 22, 2012 at 9:34
  • 1
    Your data set seems to be very large and probably it'll be a hard task to atmospherically correct plus radiometricaly normalise your images. The latter, if successful, might drive in to a significant improvement towards radiometric cross-comparisons. Commented Jan 5, 2013 at 1:55
  • Good advice @Nikos - I'm going to write a script using the R package 'landsat' to carry out PIF identification for relative normalisation of the tiles. I'll update this post when I have a result.
    – si_2012
    Commented Feb 26, 2013 at 17:57
  • 1
    I am not suggesting that histogram matching (i.histo.match in grass 7) suits your needs, but you might want to have a look at it also. Commented Feb 26, 2013 at 22:21

1 Answer 1


Did you know this wiki? Maybe it helps you:


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.