Tiny, Illegitimate Watersheds created by Watershed Tool

[EvanCollins]

I have recently been trying to delineate watersheds for several thousand points in my study area. To accomplish this, I have written a simple script tool that that takes a point shapefile, flow direction raster, and flow accumulation raster as inputs.  Essentially, the tool follows these steps: select point, snap pour point, delineate watershed, convert watershed raster to polygon, append polygon to a shapefile.  I'm overall pleased with the results, however, the tool seems to have generated about 200 more watersheds than it should have.  These "watersheds" are all about the same size (typically just one pixel) and fall along the edges of the larger legitimate watersheds (I have attached a map to help clarify this).  My data are all in the same state plane projection and the processing extent was set to match the flow direction raster I was using. 
Has anyone ever seen this problem when running the Watershed tool?  Are there any suggestions to prevent them from being generated?  The biggest problem they create for me is that they desynchronize the point IDs and the watershed IDs.  I could simply delete the illegitimate watersheds, but that doesn't really help me associate the points to their respective watershed.  Any help would be greatly appreciated.  Let me know if anything needs more explaining.

Thanks,
E
  [ATTACH=CONFIG]25233[/ATTACH]

[MarkBoucher]

Are you filling the sinks before you perform flow direction?

[curtvprice · Accepted solution]

Do these watersheds tiny have the same grid code as a larger watershed polygon?

If so these are simply diagonally-connected pieces of that watershed and can be combined to the main watershed using the Dissolve tool.

[EvanCollins]

Thanks for the responses.  I should have updated my post.  I think the tiny watersheds where created during the raster to polygon conversion.  I removed them with the eliminate tool, which appears to have been designed to address this problem.  It seems curtvprice's suggestion would work also.  Next time I encounter the problem, I will see if a dissolve could be a better way to handle the small, spurious watersheds.
Thanks again,
Evan

[curtvprice]

Next time I encounter the problem, I will see if a dissolve could be a better way to handle the small, spurious watersheds.

If you do want to zap away these small watersheds pieces, the raster tool equivalent of Eliminate is to use these tools to locate these small isolated regions and nibble them away:

RegionGroup (FOUR option)
ZonalGeometry (calculate AREA)
Con (to select small areas)
Nibble

[ShaunWalbridge]

Concur that these techniques both work well, I used both when creating a global 1km hydrologic model. One issue to keep in mind is determining the correct size of erroneous polygons to remove depends on the scale of the input data and the nature of the terrain you're working with.