Moisture sensing

Lowest cost/fast response moisture sensing seems to be capacitive. The Honeywell HCH-1000-001 moisture sensing capacitor based on polyimide is $2.47 in quantities of 500. Getting a quantity of 1 from Newark or Digikey is on the order $6, and with shipping that's $10-30. This sensor isn't that special or high tech, it's basically a plate capacitor with moisture absorbing polyimide dielectric that changes values from roughly 300 pF at 0% RH to 355 pF at 50% RH to 380 pF at 100% RH. One might as well build one from just a thin film polyimide plastic itself, by electroless coating of metal on a small area (the total area decides the sensitivity/response time - small device time having fast response of water absorption, but lower overall change in capacitance magnitude.)
One would need a very sensitive PIC based capacitance meter targeted for this zone. This is a better way to measure capacitance, compared to off the shelf digital output ones, because one can throw two or more different frequencies generated by the PIC, say 100 Hz, 10 kHz, (100 kHz and 1 MHz needing bnc coax), and measure the RC time constant. Water absorbption into the polymer may have a characteristic ratio at different frequencies, compared to stray capacitance from nearby soil, or even electrical conductivity from absorbed salts such as nitrates. Different polymers might actually work better, such as PVAc, or PC. I see some polyimide tape at http://www.grainger.com/Grainger/items/3ZRA4?cm_mmc=Google%20Base-_-Adhesives,%20Sealants%20and%20Tape-_-Tapes-_-3ZRA4,
which could be tested initially on a large size with may layers stacked, and taped to alufoil, as a cheap capacitive moisture sensor. You have to calibrate out the stray capacitances anyway, and moisture sensitivity is what ultimately matters. Polyimide is very stable over time, but the degree that the adhesive glue contributes to the overall moisture sensitivity, and its variation over time might be an issue. One might need three different sensors for 3 different ranges, unless it's determined that soil moisture should be controlled between say 30-50 wt%, and then the relative humidity in air around it be say 60 to 90%, and then one sensor fine tuned for that exact region. Calibration of different RC time constants vs. humidity would have to be done with predetermined oven dried gypsum/water wt% mix ratios as humidity standards, and the constants stored in the PIC's eeprom. CaSO4.xH2O is a fairly pure chemical standard, as opposed to different soils, which vary widely.
I've had the opportunity in the past to work with Agilent HP4285A/HP16453B fixture dielectric constant measurement instruments for PVC that top out at 30MHz. There are some variations vs. frequency that might be water sensitive, and yield better calibrations compared to single frequency designs. Ideally water itself has a maximum dielectric constant change / tan delta near 800 MHz - 3 GHz, but that is out of the direct range of PIC's, plus special low signal loss waveguides would be needed at that frequency, unless the sensor and measurement conversion board are next to each other. Eventually we'll see.
I was originally going to use LCD's for a sensor, but simply adding a serial adapter, with an earphone plug to female DB9 convertor, to each microcontroller, and hooking up a laptop hyperterminal to it to see what it says, and what it's supposed to transmit wirelessly, is so much simpler and cheaper. You'd have to walk around with a laptop or serial pocket pc to debug why something is not talking on the wireless and to isolate fault. I still simmering in thoughts about the wireless part, and that is almost the most crucial thing expense wise here.

PS 9/15/09 - there is nothing high tech about the plate capacitor, except the way one of the metal surfaces is either porous/foamy, or has many tiny holes embedded in it, for fast moisture sorbtion throughout the dielectric. I guess one can duplicate it with alufoil and a needle pin, though not very reproducibly, and each individual item would have a very unique, but hopefully timestable and calibratable behavior.