Changing data frame boundaries for pattern analysis

Discussion created by leclairaj on Feb 6, 2014
Latest reply on Feb 9, 2014 by obllhcb
Hi folks,

I've been working with Arc for a couple of years but only recently got in to using the spatial statistics tool set and I'm running in to a couple of problems I feel like I should be able to fix easily but can't figure out how.

I've got a data set mapping small scale disturbance features generated through feature extraction in Envi. The image I analyzed is perfectly square, but in this image there are a few large active sand dunes that were identical to the features I was looking for from a spectral point of view. At the segmentation scales I had to use, they were torpedoing my accuracy so I created a grid that I imposed over the image to mask any grid cells with active sand dune in them. This has left a decent number of large, irregularly shaped holes in my classification output.

I'm now trying to do an average nearest neighbor analysis, but the default frame size is the minimum rectangle that fits the input feature class (my classification-result shapefile), which in this case is the entire frame of the original image. It appears to be considering the areas that I masked and never classified as not having any features, rather than having no values, which seems to be skewing the results to the extremely clustered side.

Is there a way to get tools in the spatial statistics menu to recognize certain areas of your data frame as having no data and exclude them from analysis, or do these tools require a square/rectangular processing frame? I have also classified each grid cell according to major landcover type (forest, grassland, etc.), and I'd like to compare how the clustering of these features varies between them, but none of these things are polite enough to have square or rectangular boundaries, or even be continuous on the landscape, so I'm not sure how to proceed. The tools let you specify an 'area', but a single area number could never describe the shape of my classified/not classified footprint.

I should also add that this was a binary classification, so things were either this feature, or 'unclassified'. When I moved in to Arc I deleted all 'unclassified' polygons from the shape file, so what I'm working with now is just a bunch of single class polygons that are spatially discontinuous (imagine spattering a wall with paint, then deleting the wall while keeping the paint drops). Would doing a union on a background shapefile that gave a different class ID for unclassified vs. not classified surface help in any way?

Any help would be greatly appreciated.