Here I'm dumping my work over the past month, to remind myself at the end of the project what I've done.

After consulting with Dave Gochis, I came up with about a 35-gallon plastic keg (borrowed from Steve Semmer) that I have half-filled with 150 lbs of play sand.  I can weigh this entire keg and add water to obtain a bulk gravimetric soil moisture content.  I roll/shake the entire keg to ensure that the water is thoroughly mixed with the sand and vibrate it (hit the sides with a pipe) to ensure that the bulk density remains similar (as determined by level marks inside the keg).  With the vigorous mixing, it is necessary to remove and reinstall the probes under test each reading.

Multiple Qsoil measurements show that the mixing process works reasonably well, though differences of up to 2% soil moisture still remain.  However, the Csoil measurements range by as much as a factor of 2, showing little consistency between readings with different probes and little (though some) consistency with Qsoil.  I have to conclude that this is a result of the installation process and thus have tried repeatability tests at constant Qsoil.  There still is about 25% variability between the same probe installed several different times in the same media.  (It is true that most of these tests were done near field capacity and should be redone at a lower Qsoil value.)  I have carefully dug up the probe each time and found that the probes are not being deformed significantly by my latest installation procedure.  There is little variation with the same probe installation as a function of time.  (We set it to make a reading every 3 hours and I run it for a day or more.)  I also get about 25% variability between different probes in the same media.  With all of this variability, I obviously do not see the expected linear variation of Csoil with Qsoil.  Clearly, this lack of repeatability seriously compromises our confidence in the data from these probes.

To understand this problem, I have also looked at asoil and lambdasoil values individually.  asoil (derived from Tau63) varies by as much as 30% from the mean of 5 probes tested, where the Hukseflux spec is 20%.  lambdasoil (derived mostly from Vpile_on) varies by 20% with a spec of only 5%.  Thus, either there is a problem with the Vpile measurement (that could be installation related, or Semmer wants to check the A/D gain programming), or the sensor gain factor E is incorrect.  We probably can check the E values in Agar gel, but won't have time for this prior to PCAPS.  This is a TODO item.  We might as well check the resistance R as well.

I attach the plot that I made of these lab data.  Note that the symbols 1--5 refer to TP01 sensors 1,2,4,10,11 respectively.

P.S. (Apr 2011):  These files are on my PC/projects/pcaps/soilkeg.qq.  Adjusting the values using the readings in Agar taken in Sept 2010 reduces the scatter (to generally within the 20%/5% specification), especially in Lambda, though there still is more variation in Csoil than I'd like.  Still haven't measured Rs.

P.P.S. See new logbook entry Apr 2011 about post-project testing.