I don’t know if there’s an easy/good way to achieve that. Cryptomattes and object ID’s don’t work with volumes. Rendering with all the buildings set to holdout doesn’t work well either; once the images are combined in Photoshop I end up with horrible white feathered transparent edges. Combining the images in the Blender compositor gives better results, but there’s still a tiny edge visible (which becomes even more noticeable after color correction). It’s especially bad with the cranes, making some parts of the crane seem nearly fully transparent.
I’m just going to render it without the cloud edges, already spent far too much time trying to figure this out 
EDIT: Nvm, decided to go the old school way and masking it out by hand. Finished result can be found here.