Your tenacity is remarkable but flies in the face of any of the advice being given.
The formula will have to be run each time you sleep and as you are sleeping for long periods before the drone’s next visit, it will have to be run a reasonably regular intervals as the temperature changes, which also has an impact on the battery voltage as well.
And then using the main clock which will vary based on temperature & battery voltage (which varies on temperature) to measure the WDT looks like it’s about to enter a circular loop.
All on a device that has next to no RAM left to run after LMIC is implemented and can result in issues if you use too much of what’s left over for it’s actual activities.
Using a more recent ATmega that can support a 32KHz crystal for a RTC (in this case, a Real Time Counter, there isn’t an internal Real Time Clock in the time & calendar sense on ATmega) would give you more RAM, more flash, chug along nicely with its internal main clock and be reasonably accurate with it’s sleep timer.
And if you don’t use LoRaWAN, you can get the drone to broadcast a time signal so each day the devices can re-calibrate their sense of time.
On the plus side, this has given me some ideas for some other ways to tackle some technical problems I’ve seen of late, so 