Currently working on a 3rd generation cloud-terrain interaction model...
Aerial view of Haleakala, Hawaii. A low convective layer, winds from north (left), first 0 kt (all winds are aloft winds - winds on the ground are perhaps half of that due to the boundary layer effect):
The layer pretty much rims the mountain, the lava flows to the south heat up better and give more pronounced clouds than the forests to the north. Now we turn on 5 kt of winds:
Immediately, activity in the north is increased, clouds start being pushed to higher elevations, whereas the activity in the south (in the lee of the mountain) decreases. Same scene at 10 kt:
And at 20 kt - cloud now reach all the way up to the summit (as they have to in order to water the vegetation belt) although the layer latitude itself is very small:
Finally at 30 kt:
Activity in the south is basically gone, strong cloud cover in the north, being pushed even over the summit.
Another test case which I wanted to get right: Geneva and the Jura mountains. Usually, the convective weather is such that clouds develop over the Jura whereas the valley remains pretty clear and then push into the valley only in the late afternoon.
It's not perfect, but at least the cloud density over Geneva is significantly decreased and over the Jura strong Cb towers develop.
Right now the whole setup is startup only, i.e. it is not used or preserved when clouds are allowed to drift - I'll work on a dynamical variant which has the same cloud population in the steady-state long-term limit next. And, sadly, it asks for a factor 2 more terrain elevation than the 2nd generation model that is implemented right now - which is not pleasant (there's no way to get around this since I need the terrain gradient upwind of the cloud, which means sampling at least 2 points instead of one). So, presumably the 3rd generation placement model will be optional, since it doesn't do anything new if the terrain is reasonably level.