01-23-2019 08:06 AM - edited 01-23-2019 08:54 AM
Proximity Spread should do it I think, but currently has a problem (refer to IM-46021 if you have access).
Ooops - had't noticed that Proximity Spread creates an "OrignalSourceValue" attribute. So it's workling correctly and is exactly what you need to do a "Nibble" model,
01-23-2019 09:19 AM
03-06-2019 10:21 AM - edited 03-06-2019 12:52 PM
I sat down and started to flesh this model out as an article for the Spatial Models Tutorial section and realised that this model does pretty much the exact same thing:
The results of applying sequential Focal Majority filters will definitely be different from that produced by Proximity Spread, but for relatively small gaps/holes they are almost identical. In some instances I might argue that the Focal approach result is better.
The Focal Majority approach also appeared to be much faster than Proximity Spread, with the possible downside of using temporary disk space. However, interestingly, Proximity Spread does not seem to be slowed down as you increase the MaximumCost (i.e. distance to fill), whereas the Focal Majority is definitely going to take linearly longer as you increase the number of iterations (i.e. distance). So the Proximity Spread may be a better option if you have large holes to fill and want to specify an arbitrarily large distance to force all holes to be filled completely.
If you were attempting to build Thiessen Polygons from very sparse samples I would use Proximity Spread. But for filling relatively small areas of unknown values, the Focal Majority approach could be a solid option to consider.
03-07-2019 12:18 PM
Here's the more elegant, fully featured version of the Spatial Model: