I would tend to agree about tracing the edge of the canopy. This would be consistent with the way we typically map buildings as roof extents and don’t map a true footprint. In some cases I think we also concede the “true” coastline to one side of mangrove areas for similar visibility issues.
My tendency would be to show natural=wood
crossing the road if the road can’t be clearly seen from the available aerial imagery (or vice versa seeing sky from the road) and to separate them if it can be.
In the orchard case I would also leave joined if it was the same functional unit. I would also be against splitting landuse=forest
for forestry roads, but I don’t think we’ll get any sort of conclusion if we end up re-arguing natural
vs landuse
for “areas with trees”.