Hi, thanks for your very detailed answer.
After one evening of tests, I tend to agree with you: I also think that the variation in the readings is not related to the converter but instead some mechanical change in the capacitor circuit due to the heat.
My configuration is well described by your first example: a series of parallel tracks on one layer, covered by solder mask (but when the circuit will be mounted there will be a thin plastic cover to avoid direct contact of rain), and the heating resistors at the bottom. You can see the top in this (quite bad) photo:
I'm actually using only the first six tracks for the capacitor, because I was always out of scale with the entire surface connected, so I cut the rest of the tracks away from the circuit.
Below the tracks (just there) there are the four resistors. The IC is located on the upper side of the board, where you see the other small tracks and vias.
Right now I'm doing all the tests indoor, so there is no water involved. Using your notation, I have only CA and CB involved in the measurement. I do have a plane on the bottom side of the PCB, but it is floating, just to distribute the heat -I didn't thought it could be harmful.
In my circuit, I'm using not using a differential reading, just an absolute measurement of the capacity:
because I didn't expected this behavior.
Anyway, in the previous graphs the temperature is on the right axis, in ºC, while the readings of the converter are just the converter output, not scaled, truncated at 16 bits.
Yesterday I run a series of heating-cooling cycles to see what was happening and I noticed that the difference between the starting value and the end value was reducing itself, until I arrived to the ninth run where there was no difference. You can see the graphs here:
The series 1 was the first that the 9 the last. The heating duration was manually set, so the curves are a bit different, but the overall behavior is clear.
Now, what could I do to fix this problem?
If I use another series of tracks on the same PCB and configure the converter for differential reading, I will not be able to read the presence of rain anymore, because it will be present on both capacitors.
Do you think that after a series of cycles the substrate of the PCB will reach some stability?
Thanks for your help!