Always work from a copy. Both Spatial Adjustment and Rubbersheet tools modify data in place.
Clean identity links text file
The interactive Spatial Adjustments tool saves links as fixed width text, using multiple delimiters (tabs) where needed to line things up. The XYToLine tool doesn't understand these, so pre-process input link files to remove extra delimiters:
Convert text to Displacement and Identity links
Both displacement links and identity links are saved in the same file. There is no marker to set the types apart; the identity links merely have the same coordinates for source and destination. (This likely accounts for the lengthy pause when loading the links table interactively. ArcMap needs to scan for and separate the identity links.)
- Create displacement vectors from text file with XYToLine
- Make a Feature Layer or selected set from 0 length lines
- Copy the selected identity links to a new feature class, then create XY Event point layer from it, (Extra step necessary because XY Event ignores the selected set from the feature layer.), then save to "Identity Links" feature class
- Invert the selection, then save to "Displacement Links" feature class
- Delete the intermediate feature class
The model is tested and works, but the exported script fails on assigning the spatial reference. It should still be good for study material.
Not much to say here, provided no mistakes in above the tool will just work. Make sure interpolation method is set to match Spatial Adjustment tool. We used Natural Neighbour.
The tool is under "Editing Tools >> Conflation", Advanced license required.
Compare the Rubbersheet geoprocessing tool results with the Spatial Adjustment tool results and verify it makes sense with your data. In our tests the two do not produce identical results. They're very close, 94%, but not identical.
In our data the differences have only been observed on outer edges. This is also where our identity links lie. Figuring out if the variance is about different identity calculation approaches or edge of data is left as an exercise for a future endeavour, though I lean to edge of data. There is less than 0.15m variance along borders with identity links, and up to 3m elsewhere.
2017-07-05: Bug Confirmation
Esri has confirmed there is a bug up to and including ArcGIS Desktop 10.5:
BUG-000105702 - Rubbersheet Features tool does not produce the same
output as the Rubbersheet Method from the Spatial Adjustment toolbar
when using the same displacement links.
The bug is specific to the Rubbersheet method ... but they are also
seeing differences with the projective method. ... found that when
using the rubbersheet method the vertices appear to vary when they
are outside of an imaginary bounding box around the displacement
links. Therefore, one potential workaround would be to add more
displacement points so that all of the vertices are inside that
imaginary bounding polygon.
When asked which method is most correct they responded "the Spatial Adjustment Toolbar". I'm cautious about accepting the statement wholesale because previous experience with vertex coordinate drift was because the graphical interface (ArcMap) took math shortcuts for performance improvements during interactive edit sessions. (It used the single precision of the data frame rather than that of the dataset.) Admittedly that was a long time ago, circa 2004. Either way, the proof will in the testing and results. ;-)