I have agricultural data in form of polygons which I would like to test for spatial clusterings/spatial agglomerations.

All in all I have about 40 variables which I can aggregate and standardize in different ways. One way of standardization could be for example to calculate production values per capita within each polygon. Another way could be to calculate production values per ha within each polygon.

All ways of standardization and aggregation produce different maps with different spatial patterns: clusterings and no-clusterings. So as a base for my later analysis I wan't to identify such aggregation/standardization combinations that produce strong spatial clusterings. Therefore I would need to compare the different results from aggregation and standardization.

Of course I could do this looking manually at the maps (see example below). But this is quite subjective and only in some case you can make clear distinctions. Imagine doing this for 40 variables and let’s say 8 possible ways of data preparation… So I would prefer to use some objective measurement i.e. spatial statistics.

I use R and Arc GIS. Has anyone an idea how to implement such an analysis?

The examples below show Banana Production once without standardization and once standardized per capita. they look very similar, but which one is more spatially clustered? Without Standardization Standardized

  • 1
    Have a look at the tools in this topic gis.stackexchange.com/questions/3189/… – radek Feb 8 '13 at 12:55
  • I know some of these tools. I think they are basically designed for multivariate cluster-analysis. But in my case i would first like to see a univariate measurement of clusters. for point data it might be something similar to density analysis or hot spot analysis. But i don't know if there is anything similar for polygon data. – Dspanes Feb 8 '13 at 13:20
  • 1
    ok I think maybe HotSpot Analysis in ArcGIS might do the job... – Dspanes Feb 8 '13 at 13:35
  • Give LISA a shot - available in both ArcGIS and R. – radek Feb 8 '13 at 13:40
  • 5
    Unless there has been a major change in ESRI's Python code I would strongly recommend NOT using the ArcGIS LISA model on polygon data. The code converts to polygon centroids and does not use neighbor adjacency, which is quite incorrect. It is straightforward to run a LISA using a 1st or 2nd order neighbor contingency matrix in the R spdep package. A nice alternative is GeoDA (geodacenter.asu.edu). – Jeffrey Evans Feb 8 '13 at 15:57

Morans I would give you one measure of spatial clustering.

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.