Hi Laszlo,
Apologies for the delay of response. I agree that the 100 Megaohms resistor can give you high input impedance and dc return path at the same time. But I don’t think that the diodes necessarily solve the problem. Their leakage will add to the bias current of the amplifier. If the two currents can’t cancel at some voltage, then you will still have the same problem. AD8626, for example, uses n-JFETs at the input, so the leakage current flows out of the gate. The leakage of the two BAV199 diodes should match pretty well when the voltage across them is equal, so they will cancel. Then the current flowing out of the op-amp input will start to raise the voltage of the input node. The question will be whether the leakage current from the bottom diode will increase enough to cancel the bias current before the amplifier saturates. If it has to move very far, you’ve already lost a lot of dc information.
There are other tricks as well. Bench-top instruments such as DMM’s use galvanic isolation to achieve high input impedance. Something like CN-0067 might have a similar effect. I’m not sure if it would work in soil, but another thing that is done in some medical systems is deriving the common-mode voltage from the inputs, integrating it, and driving a third electrode with it, which in effect sinks the input bias current (and noise current) in the other two electrodes through the measurement medium and controls the common-mode voltage of the measurement electrodes. These are just a couple of ideas without knowing much about your system.
You may also want to check our instrumentation amplifier product like AD8220, AD8235 and AD8236. All of them has a very low input offset current (~1pA). This can simplify your circuit instead of using 2 buffers and difference amplifier.
Best regards,
Emman