I have a land cover layer from which I've extracted the tree canopy class and set all values that are not tree to NULL. I've also converted it to a vector and exploded it to make single part features.

I want to classify each patch into "small" (being very small trees along streets or in fields), "medium" (groups of backyard trees clustered together/ in small parks) and "large" (legitimate forested areas such as town forest and the like).

However, there are some medium and large patches that are connected by a very small "bridge patch" resulting in humongous patches that should be a large and medium patch, respectively. It's a mostly residential area bordering public parkland so there is much more dense tree canopy than you see here (also a 6 sq mile area).

I have access to data like roads, town boundaries, etc. etc. No special data or anything not available publicly (from USGS or other sources). I do have the Spatial Analyst and Patch Analyst extension (among others) as well as an ArcGIS for Desktop Advanced License.

For a little visualization of the data I'm working with:

enter image description here


I do have an update on this problem. For whatever reason, the issue with the buffer I mentioned above did not occur the second time I tried it and produced a layer that I then used to create the forest patches. I did try the thin tool to see what kind of layer it produced and ended up with something looking like it came straight out of Tron. I did try "expanding" this layer but I ended up losing much of the integrity of the original TC layer.

  • 1
    You could try resampling the tree canopy raster to a lower resolution - this would eliminate the "bridge" patches, while preserving the larger contiguous areas, to give you the "medium" and "large" clusters. Maintain the current "small" clusters from the high-resolution version. – Stephen Lead Dec 11 '14 at 3:54
  • As a raster you can use Thin resources.arcgis.com/en/help/main/10.1/index.html#//… to remove single pixels and thin bridges then expand back resources.arcgis.com/en/help/main/10.1/index.html#//…. As a vector you can buffer by a negative amount then buffer back by the same value as a positive number. Let me know if any of these sound right and I'll expand into an answer. – Michael Stimson Dec 11 '14 at 4:04
  • I have not tried the Thin or resampling methods yet and I will definitely give them a try when I get a chance and will post any updates. As for the buffer, the Patch Analyst extension has a synonymous tool that creates "core areas" link. But when buffering these core polygons I either get a (seemingly nonsensical) error or buffers that do not conform to the original vector's shape (I assume because of the irregularity of the tree canopy). Thank you so much for your responses @StephenLead and @MichaelMiles-Stimson! – Jeremy Dec 11 '14 at 5:16
  • While not exactly the same question, you might want to employ the techniques used in this question: gis.stackexchange.com/q/41064/7424 which is pretty much what Michael Miles-Stimson mentioned. – Fezter Dec 11 '14 at 6:07

You should take a look at the Neighborhood tools in the Spatial Analyst toolset. I have solved similar problems with the Focal Statistics tool, using the circle neighborhood and a majority statistic type. Then, convert to polygons and if necessary use Michael's idea of buffering twice to separate patches.

| improve this answer | |

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.