Getis and Ord G* in GeoDa

This video illustrates how to run the Getis-Ord G* statistic in GeoDa using a distance based weights matrix.

[@sergiomorellmonzo1780]

I am trying to apply the Gi index to a remote sensing problem. My goal is to extract features from a high resolution image (1m) for a semantic image segmentation process. I do not have an ENVI license and I want to try to process my data with GeoDa. However my data is very heavy (7 Shapefiles of 10milions of points). Which method is faster? ? Queens / Rook neighborhood rule or Euclidean distance based? Thank you

[@SEERLABUNIVOFFLORIDA]

Hello. Your decision should based on the pattern you are searching for. If you use Rook, you will have fewer neighbors than queen. Or you will be more conservative in defining neighbors. Distance based versus neighbor-based should be determined by your research question and data. Shapefiles with lots of points should still run. It may take a while, but it will work. GeoDa will be faster (much!) than ArcGIS for raster data (here I assume your points are centroids of cells with values as shapefiles). I hope that helps.

[@sergiomorellmonzo1780]

@@SEERLABUNIVOFFLORIDA Thanks for the tips!

[@SEERLABUNIVOFFLORIDA]

@@sergiomorellmonzo1780 sure. You may also look at cran.r-project.org/web/packages/AMOEBA/index.html The AMOEBA package in R. It will also run quickly on your data. Instead of using a fixed matrix, it will use a random seed and search for clusters - there are several papers on it - and define cluster shape. This may be useful with your RS data.