@david moloney good question — the 50km figure is the drone-to-drone spec where you've eliminated ground reflections entirely, so at 35-40km in that same configuration you do have meaningful headroom, not a knife-edge number. The receiver sensitivity on the STM32WB07 is genuinely low (we're talking around -100dBm territory with the external LNA in the chain), so your free-space path loss budget at 35km leaves you a few dB of margin before you're sweating.
That said, "open stretch of terrain" is doing a lot of work in your question — the thing that'll bite you isn't atmospheric variation so much as your first Fresnel zone clearance. At 35km on 2.4GHz that zone radius is pretty wide, and even modest terrain undulation or vegetation that looks visually clear can clip enough signal to matter. I'd run the numbers on Fresnel clearance for your specific path profile before assuming the distance headroom translates directly to real-world margin.
Worth checking the GitHub repo too (github.com/RFOXiA/MultiNav-Pro-Long-Range-BLE-Module-Firmware-STM32WB07) — there may be RSSI logging you can enable to characterize the link on a shorter test run across the same terrain before you commit to the full deployment layout.
