RWT can generate underwater areas when the elevation provider has no data.
This uses a rather primitive algorithm based on perlin noise.
If you have any suggestions for improvement, I am always open to discussion.

Detecting water by color will not work on satellite imagery.
I tried this a few years ago.
To detect water, you need to have a specially prepared image.
The next version of RWT will have such feature.
Here a post about it with video:
https://forum.infinity-code.com/viewtop … 6916#p6916

If the user has programming experience, he can use RWT API to modify the generated underwater values and achieve the result he wants.

The tools you described go beyond the concept of an asset to generate real-world objects.
I think that it would be more logical to create such tools in the asset like River Auto Material.

As someone focused on naval simulation, it would be invaluable to me if there were water depths in this, even if the data were of low-resolution or was synthetically computed according to some heuristic (or a combination of those).

You could add the option to have your add-on detect water areas (alt=0, color=bluish to greenish with slow gradients) and replace them with some algorithmically-created depths.

The user could assist in this.  Once the tool had located a contiguous body of water, the user could click on a few points within it and specify the depth s/he wanted those points to be.  The tool would then create a noisy but fairly smooth fake bed for the lake or ocean that satisfied the constraints.  Paint tools could also be used, perhaps while enforcing bounds for max/min depths.

Other tools could be to let the user trace a closed area of water and specify the average depth of that area.  Any overlap of such areas would have an additive effect whereby the most recently drawn area has pre-eminence.  A color coding mode in the editor would help the user see where s/he has been in terms of authoring, as well as to see the depth map in 2D.

tone