Couldn't resist another play on the train this morning.
1: echoing RArcher's request to georeference images maps as well...
2: ... particularly because georeferenced terrains and spherically projected images have different orientations. N-S axis for spherical projection is the Y-axis and for georeferenced terrains is the Z-axis. Global textures have to be remapped with a pitch of -90° to correctly align with georeferenced terrain, and more painfully, changing longitude requires remapping using roll instead of yaw (which is easily done using image offset).
This had caused me some confusion earlier, so it is still only the "top" of the TG planet that gets terrain added. Given this, it is probably more useful at the moment to have polar axis on the z-axis so that we can at least see southern elevation data as well and can "rotate" the data to see other longitudes beyond +/-90°.
3: Georeferencing TERs isn't automatic. I got around this by adding a geotiff first and then swapping the file to a TER (TERs are 50% smaller)
4: There is a lean of the terrain towards 0,0,0 at large X,Z values.
I started on a terrain blending mask based on lat/long by combining parts of my rainbow network with my previous blending masks. I'll finish that off on another train trip. The network is a little more complex but it's pretty straight forward and will (should) work for any planet position. Performance remains to be seen, but it's mainly in preparation for when georeferencing is fine-tuned a bit more.
Images below (Grand Canyon) include global 0.05° resolution elevation data, 10m resolution USGS NED (both as TERs, not geotiff),no terrain blending mask and no fractal detail... longitude adjusted 100° to bring terrain towards the centre of visible data (90W, 90E), 8K blue marble texture remapped to pitch -90° and then rolled 100° in PTGui. Ocean in the last image is "real". Duplicate planet with water surface, all terrain on original planet <0m altitude displaced -5m to ensure minimum water depth.