You can possibly do a divide by 3 or 4 of the cellphone mast map since there's a range of competing networks.
LoRaWan receivers are more sensitive, and have a lower datarate, so we needn't necessarily end up with more than the cells.
LoRaWAN doesn't really have deadzones around transceivers -- they seem to be able to turn the gain right down, and the antennas tend to be simple dipoles. AutoData rates will also help with access. I've talked to mine from within the same office with no problem.
Trilateration for determining position of the sender works better with more receivers -- we'd need at least 3 to cover each point.
I'll run some more simulations with basestations along the main roads you suggest.