You can't calibrate away the RF noise introduced by transmitters that change frequently in time and space.
That's not to suggest that you cannot find a way to discriminate between the water vapor and the 5G transmissions, but you can't just take a sample on a low-humidity day and subtract that from new samples. If the metrics are below the new noise floor, merely throwing machine learning at the problem will not solve it.
I don't mean taking RF data in account at all. I mean just collecting temperature, air pressure, humidity and wind speed data from many points in time-space.
That's not to suggest that you cannot find a way to discriminate between the water vapor and the 5G transmissions, but you can't just take a sample on a low-humidity day and subtract that from new samples. If the metrics are below the new noise floor, merely throwing machine learning at the problem will not solve it.