Thanks floobydust,
Yes the problem is clearly common-mode noise.
Test of the entire setup in the lab, "off ground", is giving great performance on the reading of the sensors. There is no noise on the analog front end, the readings are rock solid.
I have also measured critical points for noise with an oscilloscope (with proper grounding methods for the probe) and the 56V DC-DC converter has appropriate filtering (there's minimal noise on Vcc). There is actually a second stage regulation done by a 5V LDO, after the DC-DC. The module is a double sided PCB with proper grounding.
I am not too worried about ripple created by adding an isolated DC-DC on the NE555 frontend, since the reading is frequency of a square wave signal. If it is a problem it can be filtered with choke & PI filter as you mentioned.
When tested outside, the setup:
1) works if powered from a battery (UPS) and only one humidity sensor is in the ground
2) works if powered from mains and any sensors are isolated from ground (i.e. in a flower pot)
3) does not work if powered from a battery (UPS) and two or more humidity sensor is in the ground
4) does not work if powered from mains and one or more humidity sensor is in the ground
Basically, if there's more than one ground, there's a ground loop and things go bad.
Here's someone experiencing the same issue with a different setup:
https://electronics.stackexchange.com/questions/374749/soil-moisture-sensor-value-is-fluctuating-when-plugged-into-the-soilSome clues here as well:
http://www.irrometer.com/200ss.html"Devices reading WATERMARK sensors must isolate the sensor reading circuit from any earth ground. Mains powered devices will need transformer isolation to prevent a loop to ground through nearby equipment, and any communication line intended to be permanently connected to another grounded device will require optical isolation. If a ground loop exists, the readings will be wrong and current leakage may quickly destroy the sensor electrodes. Battery powered devices are more isolated by nature, but grounds for lightning protection etc. can cause the same problem.
In addition to general device isolation from ground, when reading multiple sensors the circuit must be designed to accommodate isolating sensors from each other. The wet soil in which sensors are installed creates a common conductive path between sensors. In effect, without isolation a device can be reading partially or fully between electrodes in different sensors rather than between electrodes inside each sensor.
Sensors must be powered individually, and the ground must be isolated from sensor to sensor. Even if powered individually, an open ground path to another un-powered sensor can't be allowed. This is best accomplished using multiplexers to open and close the appropriate channels."