OK so the light bulb finally came on and it's like 1000 watts or so.
Thanks again to Ulco for the sample file (and his previous posts on this subject).
This thread in particular was very helpful. A post in it by Martin is worth quoting at length:
QuoteThe shader uses the terrain's state from the last compute terrain. This compute terrain does NOT need to have smoothing enabled.
For best results the surface layer shader asks to enable smoothing, so do so.
The surface layer shader then generates a smoothed version of your last compute terrain before that surface layer shader.
The smoothed result's smoothness depends on the gradient patch size of that compute terrain.
Indeed, the smaller the gradient patch the more details the smoothed version of the terrain will have.
Yet, it will always be smoother than the unsmoothed, so with the right settings you will always see some kind of effect of IU with DI.
Now the intersection zone is a parameter which determines a kind of step size to read the smoothed version of the terrain as provided by the last compute terrain.
It will compare both states and determines where the two "intersect", which is mostly in depressions.
You can imagine this by having a sinus shaped terrain from -0.5 to 0.5m, which upon smoothing is flat @ 0m (for example).
With DI you can now "fill" the negative/depressed parts of the sinus shaped parts.
With correct settings you can even "fill" them completely.
Small values for intersection zone allow for detailled "filling" of smaller depressions.
High values for intersection zone uses large chunks of smoothed terrain and allows for filling large depressions, but as a consequence can also add a thick layer over non-depressed surfaces.
Here is what I take from this:
The patch size of the preceding compute terrain drives the scale of the intersection. For large-scale alpine scenes, use a very large patch size. For nuanced, small-scale scenes, use a very small patch size.
Smoothing in the i/u layer must be enabled and set to 1, regardless of the type of intersection (favor rises, depressions, or displacement intersection).
Intersection zone is the vertical depth of the, well, intersection zone. In meters.
Intersection shift vertically adjusts the offset of this zone. In meters.
Min intersection shift can be used to choke the distribution of the intersection.
The closer it is to intersection shift, the more contrast in the distribution.
For my scene, the compute terrain must follow the canal displacement because the canal is going to be drained. (There will be a bit of water in the bottom in the final version.) This is a relatively small-scale detail (~12m wide), so a large patch size (as in my original scene), won't work. The answer is to use a small patch size. It's now set to 3.
The snow surface layer has i/u set to favor depressions with an intersection zone of 1 and an intersection shift of -0.5 to expose the very slight vertical displacements. A breakup shader adds more texture.
The small patch size means the hills in the background are now completely covered with snow, but that looks natural here and they will be covered with trees anyway.
Finally, a question for
@Matt:
The thread referred to above includes several references to detailed, in-progress documentation for i/u. If it isn't too much trouble, can this to be made available on the wiki or forum? I wonder if it might be worthwhile, even if it's not complete.