5

I have a sample of ~700 animal locations that I will be using for habitat selection analysis. I am trying to figure out a way to subsample my locations in a way that will maximize my sample size for analysis, but minimize spatial autocorrelation. I envision taking a subsample of the points such that no point is within X distance (the distance at which autocorrelation is no longer a problem) of another. I don't necessarily need to select the "maximum" number of points, it is more of a conceptual problem of balancing sample size with autocorrelation. Is there a way to do this in ArcGIS 10? The images below show the type of data and spatial clustering I am working with. The autocorrelation distance is roughly equal to that of the gray buffer (570 m), so subsampling based on that will severely limit my sample size. Is there a better approach that I am not thinking of?

Thanks for any input!

Locations with 90 m radius (white) and 570 m radius (gray) buffers

Inset from above

James
  • 155
  • 1
  • 8
  • A similar question to yours has been asked on this site, see link http://gis.stackexchange.com/questions/21616/select-maximum-number-of-points-more-than-x-meters-apart – artwork21 Feb 19 '13 at 23:14
  • A very similar question was recently asked at http://gis.stackexchange.com/questions/52261/method-for-finding-points-that-cover-an-extent. Both these questions seek to cover extents with disks that satisfy certain constraints (and similar techniques will solve them both). Abstractly, this question seems identical to http://gis.stackexchange.com/questions/27303/; the only difference between the two is the shape of the polygons involved. – whuber Feb 19 '13 at 23:48

1 Answers1

4

You can achieve the results you want using Create Random Points--just make sure to specify the constraining feature class as your animal locations point shapefile. This way, only your animal location points will be randomly sampled. You can also specify the minimum distance between points in the tool. The bottom image is the sample output, where 50 points were sampled from approximately 150 points. 200 m minimum distance between points was also specified.

With that said, I question why you would want to resample your dataset. The clustering you are observing is certainly an ecologically significant pattern. I do not know the type of analysis you are performing, though I would advise you to reassess whether or not spatial autocorrelation is inappropriate for your habitat assessment.

enter image description here

enter image description here

Aaron
  • 51,658
  • 28
  • 154
  • 317
  • Thank you- I didn't realize that the Create Random Points tool could be used in this manner. This is what I was trying to accomplish. I realize the clumping is an ecologically important pattern; I am now working through the best way to use my data with that, autocorrelation, and sample size in mind. – James Feb 21 '13 at 16:51