i'm not really experienced in spatial stats yet, but i'm growing into it.

I basically want to ascertain if certain values in a raster are a) autocorrelated and b) are more likely to exist in a certain spatial area (k-means? i'm unsure)

the data is a single raster showing land use change from 1 year to another (sorry I can't post the data) and there are about 50 possible changes (some far more prevalent than others). Quickly viewing the raster, it is clear that some changes are more prevalent in northern extremes, some in areas of upland farming etc etc, patterns do exist. But I want to prove this with stats.

For a) Local Moran's I (using a simple binary queen's spatial-weights matrix) gives us indication of spatial autocorrelation - this is useful for finding 'clumps' of similar data, correct?

For b) I'd like to explore whether each change combination is more likely to exist in a certain part of the UK (in Scotland, in western extremes etc). Would this be some sort of k-means clustering?

I'm doing all this in R,

thanks for any help (this question had some good info re Moran's I), (this question seems to start out with a similar goal, finding regional patterns in surface sea temps but fizzled out).

  • $\begingroup$ ArcMap's Geostatistical wizard could show you a spatial trend like that, and perhaps you'd find some sort of hotspot analysis useful (would require conversion to points). That's not R, but it might give you some more terms to search on. $\endgroup$
    – J Kelly
    Jan 12 '16 at 12:59
  • $\begingroup$ @J Kelly, thanks. I'm a bit reluctant to use anything that requires vector data due to the size of my rasters. Also i think hot spot analyses such as Getis-Ord Gi* wont work well with categorical data. I may have a look at k-modes. $\endgroup$
    – Sam
    Jan 12 '16 at 15:25

Your Answer

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

Browse other questions tagged or ask your own question.