I have been thinking and maybe the surface level change problem can also be solved by using a differential method.
Let's say you use a second similar floating device with complete voyancy (floating at the surface of the liquid). If this second device is not affected by changes in the liquid voyancy then you can use it to measure just the surface level. Then by substracting the two signals the difference will be the difference in the voyancy of the liquid. To do that, both sensors outputs (after demodulation) can be fed to a differential amplifier ( easy to build, check the LM324 datasheet) and that will be your liquid voyancy signal to be send to Arduino analog input. That way you will not need to calibrate the system that much; but just at the beginning. I think you Project its completly doable that way.