SOIL MOISTURE SENSOR EVALUATION FOR CASES97 Introduction For the CASES97 field experiment soil moisture was an important variable to aid in an understanding of the water budget for the Walnut River watershed. Each PAM/ASTER site was equiped with a Campbell CS615 soil moisture probe. These probes have variable calibrations which can only be determined in-situ. In order to obtain useful data from the probes, more reliable comparison data was required. To this effect, two soil moisture inter-comparison (IC) measurement methods were used. The first method used a soil coring tool to obtain the combination of gravimetric moisture and bulk density. The second method was using a TRIME, handheld TDR system (TRIME-FM). The TRIME-FM system was fitted with a P3 (three prong probe) 15cm long as opposed to a P2 (two prong probe). The P3 probe is more immune to soil conductivity errors than the P2 probe. The TRIME data was in volumetric percent moisture (the same as the CS615). The soil core data was converted from gravimetric to volumetric by multiplying by the bulk density. The soil core samples provided a fundamental reference measurement of soil moisture. The TRIME system was an untested instrument at the beginning of the CASES97 project, but the anticipation was that it could provide a much easier method of obtaining near reference quality data. Therefore, the purpose of this report is twofold. First, an evaluation of the TRIME-FM sensor is given in comparison to the soil core samples. Second, an attempt is made at using the IC data for generating calibration equations for the CS615 probes. Sampling Discrepencies Fundamentally, the three methods of obtaining soil moisture sampled the soil medium in three different ways. Although some effort was made to measure the same sample depth for each of the techniques, it was not possible to do this well. The following paragraphs elaborate on each measurement. CS615 Sampling Location The CS615 probes are 30cm long. Since only the top section of the soil was of interest, the probes were buried horizontally at a 5cm depth. Standard burial technique involved digging a trench with a vertical wall, inserting the CS615 into the vertical wall, and then backfilling the hole. The probe measures moisture approximately 2cm on either side of itself, so the measurement depth was 3cm to 7cm. Soil Core Samples The soil coring tool consists of an 8cm long sample tube, which can be split using two 3cm rings, and two 1cm rings. The 1cm rings are installed on the top and bottom, so the the center contains the 3cm rings. The top and bottom cm of a sample is discarded, so the measured soil core consists of the sample from about 1cm to 7cm. TRIME Measurements The TRIME-FM probe has 15cm rods. In order to match the soil depth range of the CS615 and soil cores more closely, the probe was installed at a 45 degree angle. This gave a nominal range from 0cm to 10cm. For some measurements a 45 degree triangle was used to facilitate proper insertion. Soil Core Versus TRIME A number of ICs taken during the project consisited of a TRIME measurment at a site in conjunction with a core sample. The first graph shows time on the X axis, and soil moisture by volume on the Y axis. Soil cores and TRIME measurements are plotted as points. The TRIME data is plotted by station number, and the soil cores are plotted by dots. The combination data points are circled. The first few show some wide scatter, but subsequently there is excellent agreement between the TRIME and soil core measurements. The first few points with the wide discrepencies may have errors in the measurement technique. The main cluster of measurements were taken from May 5 through May 10, by a single technician. All these data points show excellent agreement. The second plot in this report shows the soil moisture from a core sample plotted against the absolute error (TRIME - core). Most of the data points indicate that the TRIME reads slightly lower than the core samples. This could either be due to the instrument, or due to the different depth range which the TRIME is averaging over. CS615 Calibrations The remainder of the plots can be divided into two groups. The first group shows attempted second order calibration fits for all eight stations. The remaining pages show improved fits and the resulting soil moisture data for the project. Also shown on the time plots are the TRIME (denoted by T) and soil core (denoted by c) data points. On some plots nominal "factory" curves are also shown in dashed line types and colors. There are two fits shown for each station, the first is without, and the second is with soil temperature compensation. To complicate things a bit, two different versions of the CS615 probe were used. Table 1 shows the probe types and site information. Table 1 CS615 Probes and Sites Site CS615 1 new grassland 2 old grassland 3 new corn, newly planted 4 new grassland, poor soil 5 old wheat 6 old wheat 7 new soybeans??? 8 new grassland Since the TRIME data demonstrated itself to be of high quality, and there was far more of it than the soil core data, the TRIME data was used for obtaining calibrations. However, stations 7 and 8 only had one TRIME sample, so the soil cores were used instead at these sites. The documentation on the CS615 probes and previous experience has shown that a second order polynomial would be sufficient for a fit function. Thus: soil_mois (%vol) = a0 + a1 * raw + a2 * raw^2 An Splus function ("fun.cases.qsoil.fit") was written to expedite the calibration procedure. An initial fit was done using the least squares method. It is apparent from the graphs, that most of the fits were quite poor outside the range of data points. What seems obvious from intuition and visual inspection of the intial plots is that two things are required for a good fit. First, a fairly wide range of moisture values for the set of ICs is needed, and second, low IC sample errors are needed. In order to obtain a better understanding of what was required for a good set of IC data points, several statistical calculations were done. Table 2 shows some of these statistics. The parameter "Fit Fraction", is essentially a measure of the range of raw IC data divided by the range of raw data for the whole project. Of all the first cut statisitics, the covariance and correllation of the fit data show the most promise as a fit indicators. However, none of the indicators does an ideal job. Table 2 First Round Fits Site Fit Fit IC rh Site Points Fract Range covar corrrel Initial Fit 1 5 .606 13.2 .18 .91 bad low end--too flat 2 5 .926 11 .32 .84 flattend high end 3 6 .361 5.4 .16 .64 flattend bottom end 4 7 1.258 9 .18 .82 dampened high end 5 3 .621 13.95 1.33 .997 excellent 6 6 1.148 26 1.16 .65 dampened high end 7 5 .542 11.14 .19 .93 terrible--U curve 8 6 1.18 15.25 .45 .98 good In order to obtain better fits, some data manipulation was required. The Splus routine was changed to allow modification of the initial fit. After the initial fit was generated, it was plotted on the screen. Then additional fit points could be added on an iterative basis to try and improve the shape of the curve. Initially this was done by adding points in the range of the IC data. This did a fairly reasonable job of improving the curves, but it was an iterative process. A better method which proved to be extremely effective was to add anchoring points outside the range of the data. For the stations with the new CS615 probe this was easily done by adding a point at the X axis crossing. This crossing occurs in dry air where the probe is expected to output a value of 0.7 msec. The old CS615 probes do not have this axis crossing. For these probes, an anchor was chosen based on the expected fit of the data. Table 3 shows the final third order fits generated. Table 3 Good Fit Site A0 A1 A2 Added Fit points (x,y) 1 38.1 -115.1 87.17 (.7,.2) 2 12.87 -17.34 9.47 (1.5,8.6) 3 -42.5 73.23 -17.28 (.7,.1) %3a -2.48 -8.99 17.65 (.7,0), (1.82,43.35), (1.2,12.3), % (1.77, 44.5), (1.36,14.7), (1.64, 26.2) 4 -20.142 26.53 3.69 (.7,.2) 5 52.47 -67.5 26.5 none 6 -15.5 -5.35 11.2 (2.46, 49), (1.5, 1.06) 7 31.27 -89.5 64.1 (.7,0) 8 38.8 -103.6 68.98 (.7,.1) These fits and the respective soil moisture time plots for the project are shown as the last 16 pages of graphs. Stations 3 and 4 show relatively flat soil moisture curves throughout the project. This is most likely due to the minimal vegatation on these two sites. Conclusions The CASES97 project showed encouraging results for measuring soil moisture with PAM stations. There are four conclusions which can be drawn. First, the TRIME is suitable for taking reference points. Combination measurements with the soil coring tool will remain useful to verify the operation of the TRIME, but even by itself, the TRIME unit seems trustworthy enough to use for IC measurements. Second, due to logistics, it will not be possible to obtain the perfect wet-dry soil calibration points. For smaller PAM networks, more IC points can be gathered, but for a larger network, the logistics become a significant obstacle. One possible method to compensate for this might be to generate a wet-soil data point on station tear-down. This would involve using a watering can to uniformly wet the soil around the CS615 probe to saturation. Some unknown waiting period would be required for the moisture to stabilize. Then a corresponding TRIME data point could be obtained in the same area. An alternative to this would be making a concerted effort after a heavy rainstorm to obtain soil moisture IC samples. This was done during the CASES97 project, and was beneficial in fitting the data. Third, adding data points at the X axis crossing is the easiest way to improve the curve fit. This is easier to do with the with the new CS615 probes, and it could in fact be added as an automatic calculation rather than a user input. Finally, the older CS615 probes should be replaced with new units. This will help to simplify the calibration procedure. The older units nominally use a 3rd order polynomial rather than a second order one which could be a problem with longer term projects under a wider range of soil moistures. An additional problem is simply the confusion which results from having two variations of the instrument in the field. By using a single type of probe, it would be possible to use the factory calibration as a default for real-time project use.