It seems that a range of at least 440 km is possible with the LoRa protocol (i.e. there is no time-of-flight assumption as in GSM).
The correct way to answer this question is by looking at the link budget for your transmit/receive arrangement. Although the basic calculations are simple, knowing the right way to do the calculation is not so simple.
To receive a usable signal, the receiver needs a certain signal-to-noise ratio (determined by the tolerable error rate and the modulation characteristics). You may find some online examples of how to calculate this (for LoRa or something similar).
Signal comes from transmit power, plus antenna gain, minus free-space loss (the range calculation) minus any shading from non-line-of-sight, plus antenna gain.
Noise comes from the receive environment or thermal noise (whichever is greatest) and amplifier noise figure, plus any multi-path interference which is not delay compensated in the receiver.
Assuming a 16km range is possible with a simple antenna (spherical uniform radiation), you're asking for a 3.125 times range increase or a 9.77 times increase in power. This is conveniently about 10 dB, so as a rough approximation you need a 5 dB antenna gain above the 'trivial' antenna at each end. If you aim for 7 dB at each end, this gives you a small margin for other factors you've not accounted for, imperfections in your assembly, etc.
A further complication is that the quoted 16km range is plausibly within the horizon of an antenna close to the earth, but to achieve 50km line of sight, you would need to raise one or both ends by many metres.