VZJ sign up for etocs
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via ISI Web of Science (8)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Fares, A.
Right arrow Articles by Parsons, L. R.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Fares, A.
Right arrow Articles by Parsons, L. R.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Fares, A.
Right arrow Articles by Parsons, L. R.
Related Collections
Right arrow Vadose Zone Processes and Chemical Transport
Right arrow Other Geophysical Methods
Right arrow Soil Hydrology
Published in Vadose Zone Journal 3:1390-1399 (2004)
© 2004 Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA

ORIGINAL RESEARCH

Dual Field Calibration of Capacitance and Neutron Soil Water Sensors in a Shrinking–Swelling Clay Soil

A. Faresa,*, P. Bussb, M. Daltonb, A. I. El-Kadic and L. R. Parsonsd

a Natural Resources and Environmental Management Dep., Univ. of Hawaii at Manoa, Honolulu, HI 96822
b Irrigation and Hydrology Dep., Sentek, Pty. Ltd., Stepney, Adelaide, South Australia
c Dep. of Geology and Geophysics and Water Resources Research Center, 1680 East-West Road, Univ. of Hawaii at Manoa, Honolulu, HI 96822
d Univ. of Florida, Institute of Food and Agricultural Sciences, Citrus Research and Education Center, 700 Experiment Station Road, Lake Alfred, FL 33850
* Corresponding author (AFares{at}Hawaii.edu)

Contribution of the College of Tropical Agriculture and Human Resources, Univ. of Hawaii, Honolulu, HI.


Received 23 October 2003.



   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 SUMMARY AND CONCLUSIONS
 REFERENCES
 
Multisensor capacitance sensors (MCS) are now popular alternatives to neutron scattering (NS) soil water–monitoring devices. The objectives of this study were to (i) quantify the effect of clay shrinkage–swelling on the soil bulk density; (ii) determine field calibration equations for an MCS and an NS device; and (iii) compare the performance of the MCS with a NS meter under field conditions. The calibration was conducted in a duplex soil with sandy clay loam overlying clay in South Australia. Six access tubes were installed in a 6 by 8 m grid. Three moisture treatments were replicated twice for every moisture level. The bulk density of the top 20 cm increased with increasing water content; this increase was more pronounced in the upper 10-cm horizon, which could be attributed to soil compaction. However, a negative correlation was obtained between bulk density and water content in the 30- to 100-cm depth layers reflecting the shrinking and swelling properties of the fine-textured subsoil. Results also show highly significant effects of sampling depth and moisture level on NS and MCS readings. Compared with linear calibration, a three-parameter model improved NS calibration and/or minimized the root mean square errors for 6 out of the 10 sampling depths. Except for the 10-cm sampling depth, individual calibration for each 10-cm soil layer improved the accuracy of the MCS as compared with the use of single calibration equation for the entire profile. Site-specific calibration improved the accuracy of both the NS and MCS soil water–monitoring devices.

Abbreviations: EM, electromagnetic • MCS, multisensor capacitance sensor • NS, neutron scattering • SF, scaled frequency • TDR, time domain reflectometry


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 SUMMARY AND CONCLUSIONS
 REFERENCES
 
COMPETITION IS INCREASING between agricultural, domestic, and industrial water use. Optimal management of water resources is especially critical when water supplies are limited and/or potential environmental problems exist. One key requirement to optimal soil and water management is accurate knowledge of the soil water content. Most early in situ measurements of the water content were made with neutron moisture meters, often referred to also as neutron scattering (NS). First introduced in the 1950s (e.g., Gardner and Kirkham, 1952), NS proved to be very popular as a research and teaching tool within the scientific community, but also for application to a wide range of practical agricultural, environmental, and engineering problems. Its widespread use resulted partly from the ease and speed of measurement and the nondestructive nature of its water content measurement as compared with conventional gravimetric methods. Over the years, many NS calibration equations have been proposed (e.g., Allen and Segura, 1990; Corbeels et al., 1999). One of the first guides on how to use the neutron probe was written by Greacen et al. (1981); recent guides were also published by Hignett and Evett (2002) and IAEA (2003).

Since the release of the first prototype, NS has seen several improvements such as weight and size reductions and the introduction of more efficient detectors that also used safer radioactive sources. However, despite these improvements, safety regulations requiring costly licensing and training of users and considerable regulation have caused the NS method to remain expensive to maintain and difficult or impossible to use in some situations, particularly for unattended monitoring (Evett, 2000).

Neutron scattering soil water measurements are known to be affected by soil access tube material type, its thickness and size, and air gaps between access tubes and the soil profile (e.g., Allen and Segura, 1990). The inside diameter of the access tube should be large enough for the probe to move freely without friction (IAEA, 2003). An air gap of 1.5 to 4 mm is often considered satisfactory provided the hole is straight (e.g., Prebble et al., 1981). Neutron meter count ratios were found to be influenced by material type, access tube type (Abeele, 1979; Allen and Segura, 1990), and thickness (Allen and Segura, 1990). Large air gaps between the probe and the access tube may result in errors due to eccentric positioning (Schrale, 1976). The effect of an air gap on neutron counts is similar to having a larger tube diameter (Abeele, 1979; Allen and Segura, 1990). An inverse linear correlation exists between NS count ratio and access tube diameter of PVC and steel types (Abeele, 1979). Allen and Segura (1990) found that on average, count ratios for PVC access tubes with no air gaps were 6% higher than those for tubes with 10-mm air gaps (i.e., for auger holes having a diameter approximately 20 mm larger than the tube) and 14% higher than for tubes with 20-mm air gaps (auger holes having a diameter approximately 40 mm larger than the tube).

Many vadose zone hydrologic studies now involve automated measurements of the soil water content. Recent advances in the use of electromagnetic (EM) methods have focused on measurements that respond to changes in soil water content (Topp and Ferré, 2002; Robinson, 2003). These EM methods include remote sensing instruments or in situ devices such as time domain reflectometry (TDR) and capacitance sensors (Hoekstra and Delaney, 1974; Topp and Ferré, 2002). Time domain reflectometry is now also increasingly used for volumetric water content measurements (Davis and Chudobiak, 1975; Topp et al., 1980). More details about this method can be found in the literature (e.g., Topp et al., 1980; Evett, 1998; Topp and Ferré 2002; Robinson, 2003).

The capacitance method for water content estimation was introduced to the scientific community as early as the 1930s (Smith-Rose, 1933); however, it was only in the late 1980s that commercial capacitance probe prototypes were developed and tested under laboratory (Dean et al., 1987) and field (Bell et al., 1987) conditions. During the 1990s several capacitance sensors were commercialized. The MCS is now increasingly being used for soil water content measurements for a variety of applications including irrigation scheduling (Starr and Paltineanu, 1998; Fares and Alva, 1999, 2000). In addition to the manufacture's calibration, MCS has been calibrated in the laboratory (Mead et al., 1995; Paltineanu and Starr, 1997; Baumhardt et al., 2000) and partially under field conditions (Morgan et al., 1999). Description of the MCS system, and its principles of operation are given elsewhere (Buss, 1993; Paltineanu and Starr, 1997; Alva and Fares, 1998; Fares and Alva, 2000).

Measuring devices are usually calibrated by obtaining readings of the instrument for a range of independently determined values of the parameter to be measured. The relationship between the readings and the values then provide a calibration curve (Greacen et al., 1981). In practice, response of the MSC to the independent parameter (i.e., water content) depends to a certain extent also on other properties of the medium (e.g., bulk density and soil salinity), thus complicating the calibration process. Given the important implications of calibration to soil water monitoring, we first give several definitions related to calibration: accuracy, precision, and robustness of a soil water–monitoring device. These three terms are sometimes incorrectly thought to have the same meaning (Williams and Sinclair, 1981).

Precision may be defined as the degree of refinement in the performance of an operation, or the degree of perfection in the instruments and methods used to obtain a result. Precision has also been defined as a measure of variability of an observation around the statistical true value (Allmaras and Kempthorne, 2002). Precision is an indication of the uniformity or reproducibility of a result (Kendall and Buckland, 1957) and as such relates to the quality of an operation by which a result is obtained. A precise instrument is an instrument that always gives the same result for some set of conditions even though this result may not necessarily be correct. A measure of the precision of an estimate is given by the reciprocal of its variance, the measure of the random error (Williams and Sinclair, 1981). Lesser precision is reflected by a greater variance. Accuracy, on the other hand, may be defined as the degree of conformity with a standard. It relates to the quality of the result and as such is different from precision, which relates to the quality of the operation by which the result is obtained. Accordingly, an ideal system is both accurate and precise. It should have a nonchanging degree of precision and conformity with the standard. A reasonable measure of accuracy in some cases is the RMSE (Allmaras and Kempthorne, 2002).

The objectives of this study were to: (i) quantify the effect of clay shrinkage and swelling on the soil bulk density; (ii) determine field calibration equations for the MCS and NS devices for a shrinking–swelling clay soil; and (iii) evaluate the performance of MCS as compared with NS under field conditions.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS AND DISCUSSION
 SUMMARY AND CONCLUSIONS
 REFERENCES
 
Experimental Site
Field calibration was conducted at the South Australian Research and Development Institute, Nuriootpa Research Station, Nuriootpa, Barossa Valley, South Australia. The Barossa Valley has a Mediterranean climate with approximately 500 mm yr–1 of rainfall that occurs predominately during fall, winter, and spring (Stace et al., 1968). The soil at the field site is classified as a Red-Brown Earths. Selected physical and chemical properties of a typical Red-Brown Earths soil are presented in Table 1. Soils of this type have three important features, including (i) brown to gray-brown coarse- to medium-textured surface soils (upper 35 cm); (ii) red-brown clay subsoils (35–130 cm below the soil surface); and (iii) calcareous material in the deeper subsoil (>130 cm below the soil surface). Based on the soil-textural difference, the top 100 cm of the soil profile was divided into an upper horizon (0–35 cm) and a lower horizon (35–100 cm). This terminology is used in the rest of this paper. Soil water retention parameters for the two major surface and subsurface horizons as reported by Stace et al. (1968) are presented in Table 2.


View this table:
[in this window]
[in a new window]
  		Table 1. Selected physical and chemical properties of a Red-Brown Earths soil from Victoria, Australia.

 

View this table:
[in this window]
[in a new window]
  		Table 2. Factorial analysis of variance tests for the loamy and clay horizons.

 
Access Tube Installation
Six access tubes were installed in two rows, 6 m apart, using a 6 by 8 m grid system. The tubes were installed based on the installation procedures recommended by the manufacturer to ensure good contact between the soil and the access tube. Detailed instructions about the installation procedure are described by Paltineanu and Starr (1997), the manufacturer's calibration manual (Calibration of Sentek Pty Ltd Soil Moisture Sensors, 2001), and Starr and Paltineanu (2002). The tubes were 1.5 m long and had an inside diameter of 5.1 cm and an outside diameter of 5.7 cm. Each tube had an extended section of 10-cm length above the ground surface. A plastic cap was firmly fitted to the upper end of each access tube.

Soil Water Monitoring Devices
Three moisture treatments (wet, moist, and dry) were used to cover much of the soil water content range at the site with two access tubes per moisture level. Water was applied to the wet treatments for several days using four low-volume drippers per access tube. Dry treatments were covered with plastic for more than a month before sampling. Moist treatment tubes were left uncovered.

Two types of soil water–monitoring devices were used: an EnviroSCAN capacitance system (Sentek, Pty Ltd, Adelaide, South Australia) and a Campbell Pacific Nuclear neutron probe meter CPN 503 Hydroprobe (CPN Company, Martinez, CA) with a 50 mCi 241Am–9Be source and a 10B–F3 detector. The neutron probe had a 3.8-cm diameter. To eliminate the effect of spatial variability on measurements of the two devices, the same access tube was used for the two soil water–monitoring devices. These access tubes were manufactured for use with MSC. This dual use of the same access tube left an air gap of 6 mm between the NS and the walls of access tubes which may have reduced optimal performance of the NS meter. If we extrapolate Allen and Segura's (1990) data assuming inverse linear correlation between the NS count ratio and access tube diameter (Abeele, 1979), a 6-mm air gap in this study would result in a 3.6% reduction in NS count ratio as compared to a no–air gap situation.

Water Content Measurements
The purpose of calibration is to establish a mathematical relationship between the volumetric water content and the corresponding device readings (e.g., the count ratio for NS). The following procedures were followed during sampling, beginning with the dry treatment access tubes. Locked in its shield, the NS was placed on a wooden stand, 75 cm above the ground, and four standard counts were taken before sampling. The NS meter was placed on the extended section of each access tube, 10 cm above the ground surface, before readings were made. Precaution was taken to position the meter case on the access tube so that the NS probe was centered at each depth of measurement. Thus, the first measurement was taken with the probe centered at 10 cm below the soil surface. Sampling with the NS probe continued at 10-cm increments to a depth of 1 m. The counts in the access tubes were for periods of 32 s. Sampling per depth was replicated three times.

Estimation of water contents near the soil surface using NS technology is complicated by escape of neutrons into the atmosphere (Jensen, 1993). The loss of a substantial fraction of neutrons from the surface soil during near-surface readings (<30-cm depth) necessitates separate depth-specific calibrations for this zone (Hignett and Evett, 2002). Following NS probe measurements at all depths, a ten-sensor MCS probe was inserted into each access tube and data were logged automatically at 1-min intervals for at least 10 min. MCS sensors were placed at 10-cm intervals so that the first and tenth sensors were centered at 10- and 100-cm depths, respectively. The MCS data logger reads the frequency response at each sensor (field count). During downloading of the data from the data logger, a normalization equation was used to transform the stored data (field counts) into a scaled frequency defined as follows:

[1]
where air, water, and field counts refer to the frequency counts of each sensor placed in air (inside the PVC tube) at room temperature (22°C), in a water bath at room temperature (22°C), and in the field, respectively.

Immediately after the measurements, a mechanical backhoe was used to excavate a 1.5-m deep trench 0.6 m away from each access tube. The trench allowed taking soil samples around the access tubes with minimum compaction. Three undisturbed soil samples (7.2-cm diameter by 6.0-cm height) adjacent to and evenly spaced around the access tubes were taken every 10 cm to a depth of 1 m as shown by Hignett and Evett (2002). The soil samples were taken vertically such that the center of the sampling ring was in the middle of each 10-cm layer. We used for this purpose a standard volumetric soil sampler, the Tanner sampler kit (Opal Engineering, Adelaide, South Australia), with a cross-sectional area of the cutting edge less than 5% of the area of the soil sample (Hignett and Evett, 2002) to lessen compaction of the samples. The invoked sampling method caused the first soil sample to be centered at 10-cm depth. A high precision Sartorius basic laboratory balance (0.01 g) (Data Weighing Systems, Inc., Elk Grove, IL), powered with a car battery, was used to immediately weigh the samples. Samples were firmly covered with a plastic foil and placed in a transporting tray. At the end of the day, soil samples were placed in a drying oven at 105°C for 72 h and weighed again to determine the water content. Bulk densities (Mg m–3), and water contents on a mass (kg kg–1) and volume (m3 m–3) basis were determined for each soil sample.

Statistical Analysis
A factorial general analysis of variance was conducted using measured bulk densities, corresponding gravimetric water contents, sampling depths (10–100 cm), and moisture levels (dry, moist, and wet). Bulk density was the dependent variable for this analysis, while gravimetric water content, moisture level, and sampling depth served as the covariance factors. The Statistix software package (Analytical Software, 2003) was used to perform the general analyses of variance. Two additional factorial general analyses of variance were conducted using NS count ratios and MSC scaled frequencies and their corresponding volumetric water content, sampling depth (10–100 cm), and moisture level (dry, moist, and wet) data. Tukey's mean separation test was used to separate the effect of moisture level and sampling depth when statistically significant interactions were found between them.


   RESULTS AND DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
SUMMARY AND CONCLUSIONS
 REFERENCES
 
Effect of Shrinking–Swelling on Bulk Density and Water Content
Soil bulk density significantly responded to the individual effects of moisture level (wet, moist, or dry), sampling depth, and their interaction (Table 2 and Fig. 1) , thus indicating changes in the influence of moisture level on bulk density with sampling depth. Tukey's mean separation test was used to obtain detailed information about the differences among the bulk density averages at three different moisture levels. Tukey's mean separation test, or simply Tukey's test, is considered one of the most conservative tests (Analytical Software, 2003). The analysis shows that the three moisture treatments form two statistically different groups, A and B (dry 1.68 A, moist 1.66 A, and wet 1.58 B). The means of the dry and the moist treatments were statistically similar since their means are followed by the same letter (A) but are statistically different from that of the wet treatment (B). The same procedure, Tukey's test, was used to study differences among the bulk density means at different sampling depths. Results are shown in Table 3. The bulk densities at 10 and 40 cm were the smallest and statistically homogeneous; however, they were different from the bulk densities at all other depths. The bulk density at 50 cm was statistically not different from that at 30 cm but differed from the bulk densities at all other depths. Significant differences were found between the 90- and 50-cm sampling depths. There was no statistical difference between the 70-, 80-, 60-, 20-, and 100-cm sampling depths on the one hand and the 90-cm sampling depth in the other hand. Except for the 20-cm sampling depth, the top 50 cm of the soil profile had the smallest bulk density as compared with the lower 50-cm portion of the profile.



View larger version (18K):
[in this window]
[in a new window]
  		Fig. 1. Effect of water content on soil bulk density as a function of depth for three moisture levels (dry, moist, wet). Results are averages of six data points. Standard deviations around the mean are also indicated.

 

View this table:
[in this window]
[in a new window]
  		Table 3. Mean separation across moisture level treatments (dry, moist, and wet). Means of treatments followed with the same letter are statistically not different.

 
The statistical significance of the moisture level–sampling depth interaction indicates a change in the effect of moisture level on bulk density with sampling depth. Tukey's mean separation procedure results show that there are eleven groups in which the means are statistically not different from one another. The largest and smallest bulk density values were for the low moisture–90-cm and the low moisture–10-cm sampling depth combinations, respectively. Regardless of the depth of sampling, large bulk density values are more likely at low or medium moisture levels than at wet moisture levels.

Regression analyses quantifying the effect of water content on bulk density were conducted for each soil depth (Table 4). All analyses were highly significant (P < 0.001). Based on the regression analyses, approximately, 55 to 90% of the bulk density variations can be explained by variations in the gravimetric water content, with a range of RMSE between 0.0005 and 0.004 g g–1. A linear correlation between bulk density and water content existed for all depths; however, this correlation was positive for the top two soil depths (10 and 20 cm) and negative for the other eight depths (30–100 cm). In other words, the soil bulk density increased with an increase in water content in the top 20 cm but decreased with water content at deeper depths (30–100 cm).


View this table:
[in this window]
[in a new window]
  		Table 4. Linear regression equations quantifying linear relationship between bulk density and water content at individual sampling depth (10–100 cm).

 
The observed increase of bulk density of the top 20 cm may be attributed to soil compaction, with the slope of the regression model serving as an indicator of the degree of compaction. The amount of compaction of the top 10 cm was 40% higher than that of the 20-cm depth (Table 4). Soil compaction generally increases bulk density and soil strength and decreases air-filled porosity, infiltration rate, and hydraulic conductivity (Froehlich and McNabb, 1984). The degree of soil compaction decreases with depth as described by the Boussinesq equation (Chancellor, 1994). Our data support this explanation since the degree of compaction decreases with depth.

The negative correlation between bulk density and water content for the 30- to 100-cm depths reflects the shrinkage and swelling properties of the fine-textured subsurface soil. As the clay soil loses water, its volume shrinks and the bulk density increases. The slopes of the regression equation at each depth below 20 cm (Table 4) represented average rates of decrease of the bulk density as a result of soil water losses. This rate varied between –2.56 g cm–3 at 80 cm and –1.40 g cm–3 at 60 cm.

Gravimetric water contents at each sampling depth were multiplied by their corresponding bulk density as reported in Table 1 for a typical Red-Brown Earths soil. The resulting volumetric water contents were regressed against actual volumetric water contents calculated using field- determined bulk densities and gravimetric water contents for each soil sample. These data were plotted for the top two depths (10 and 20 cm; Fig. 2a) and the eight depths below (30–100 cm; Fig. 2b). For the top two depths, the use of a single bulk density based on literature values resulted in underestimates of the volumetric water content of the soil samples (the majority of the data were above the 1:1 line). However, data for subsequent soil depths (30–100 cm) showed mixed results. The use of published average bulk density values underestimated actual water contents for relatively dry to moist conditions and overestimated water contents for relatively wet conditions. Failure to use correct bulk densities at the corresponding water content hence may result in as much as 20% error in estimating volumetric water content (Fig. 2a, 2b). Gardner et al. (1991) reported that relative errors of 30% in the volumetric water content are possible during drying if the correct bulk density is not used. This shows that bulk density at every location where gravimetric water contents are measured.



View larger version (23K):
[in this window]
[in a new window]
  		Fig. 2. Relationship between actual volumetric water contents determined from individual gravimetric water content measurements, and bulk density measurements and water contents determined based on individual gravimetric water content, and literature values of the bulk densities for (a) the top two depths and (b) lower eight depths of the soil profile.

 
Substantial experimental evidence exists of soil volume changes in the field over a wide range of soil water contents (Yule and Ritchie, 1980; Corbeels et al., 1999). Several algebraic expressions (Fox, 1964; Bronswijk, 1990) are available to describe the changes in bulk density as a function of gravimetric water content. A number of empirical relationships have similarly been proposed (Greacen and Hignett, 1979) and tested in the field (Jayawardane et al., 1983) to correct neutron ratios for bulk density effects.

Calibration of the Neutron Probe
Results in Table 5 show high statistical significance of sampling depth and moisture level and their combined effect on NS readings as a function of water content (P < 0.0001). The significance of sampling depth–moisture level interaction indicates that there is a change in the influence of moisture level at different sampling depths. Tukey's mean separation test showed that the mean NS count ratios as a function of the moisture level formed three statistically different groups (wet 0.3602 A, moist 0.2940 B, and dry 0.2493 C). Depending on sampling depth, the NS counts formed three statistically different groups (depths 30–100 cm, group A; depths 20 and 30 cm, group B; and the 10-cm depth, group C). The mean separation of the sampling depth–moisture level interaction produced 10 different groups for which the means were not statistically different from one another. The general trend was that the larger mean NS counts correspond to wet moisture level treatments and to depths of 40 cm and more. The smallest three means were for the top three sampling depths (10, 20, and 30 cm) at the dry moisture level. In the other hand, the seven largest NS means were for depths of 40 cm and below at the wet moisture level. These statistical analyses confirm positive linear relationships between NS counts and the moisture content; higher water content leads to high NS counts. Since depth was statistically significant, regression analyses to establish relationships between water content and NS counts were conducted at every sampling depth.


View this table:
[in this window]
[in a new window]
  		Table 5. Results of the factorial general analysis of variance for neutron scattering count ratio data as a function of water content, sampling depth, and moisture level.

 
Regression analyses were performed using a two-parameter linear model (Table 6) and a three-parameter power model (Table 7) for the 10-cm depth measurements, the 20- and 30-cm depths combined, and the seven deeper 10-cm depths combined. These analyses correlated readings at each depth with the geometric means of the water content of the three soil cores collected at each sampling depth. Linear regression is the standard approach used by manufacturers as well as in previous studies (e.g., Greacen et al., 1981; Jayawardane et al., 1983; IAEA, 2003). The trenches that were dug 0.6 m away from each access tube before soil sampling were very helpful in allowing soil sampling at all 10 depths with almost no soil compaction or sample compression around the access tubes. A few data points were disregarded as a result of experimental errors. Except for those of the 40- and 80-cm depths, all regression analyses showed regression coefficients equal to or larger than 0.87. The correlation coefficient was as high as 0.97 for the 10-cm depth, thus indicating that 97% of the variations in the water content were accounted for by the NS meter's readings using a linear regression model. As a practical indicator of the accuracy of the NS meter, RMSE of these different regression analyses varied substantially (Table 6) between 0.01 and 0.07 cm3 cm–3. Low accuracy as indicated by high RMSEs coincides with poor model correlation as reflected by a low regression coefficient, as was the case for the 40-cm depth (Table 6). Developing a single calibration equation for the entire soil profile has little practical use in our case because of the layered nature of the soil profile. Thus, three separate calibration equations were used to cover the entire soil profile; namely, for the top 10-cm surface layer (to account for neutron losses from the soil profile), for the 20- and 30-cm depths, and for the lower part of the soil profile. These three calibration equations had correlation coefficients of 0.91, 0.97, and 0.87 and RMSE values of 0.029, 0.019, and 0.022 cm3 cm–3, respectively. It should be noted that the range of water contents of the sandy clay loam surface soil was much broader (0.038–0.35 cm3 cm–3) than that of the clay subsoil (0.183–0.406 cm3 cm–3).


View this table:
[in this window]
[in a new window]
  		Table 6. Neutron scattering calibration equations for different soil profiles using a linear model.

 

View this table:
[in this window]
[in a new window]
  		Table 7. Results of the factorial general analysis of variance for multisensor capacitance system scaled-frequency data as a function of water content, sampling depth, and moisture level.

 
Compared with the remainder of the soil profile, calibration equations for the two soil depths 40 and 50 cm show relatively poor correlation coefficients (0.61 and 0.62, respectively) and high RMSE values (0.050 and 0.068 cm3cm–3). The correlation coefficient for the 50-cm depth substantially improved to 0.91 and the RMSE became much smaller (0.018) when one of the six data points was disregarded. The poor calibration at the 40 cm may be attributed to the abrupt textural change from the sandy clay loam surface soil and the clay subsoil. Because NS has a relatively large zone of influence, its readings at this depth are affected by the water contents of both horizons. This shows that caution should be exercised when calibrating or monitoring water contents of heterogeneous soil profiles. Greacen et al. (1981) pointed out that calibration becomes much more uncertain in soils with strong clay and water content gradients in the profile. Such situations arise when the sphere of influence of the probe covers layers of soil of different density, composition and water content.

Although linear regression is the standard approach for calibration of NS, we also tested a three-parameter power model to evaluate its performance in fitting the same calibration data (Table 7). Compared with linear calibration, the power calibration model substantially improved the correlation coefficients and/or minimized the RMSEs for 6 out of the 10 sampling-depth calibration models.

Calibration of the Multisensor Capacitance System
The response of the scaled frequency to water content variations as a function of sampling depth and moisture level was statistically highly significant (Table 8). Scaled frequencies increased significantly with the increasing of moisture levels leading to three statistically different groups (wet 0.8890 A, moist 0.8254 B, and dry 0.6901 C). The smallest mean scaled frequencies across sampling depths were found for the 10-, 20-, and 30-cm depths. They were also statistically different from the scaled frequency at all other sampling depths. The mean of the scaled frequency for the transitional 40-cm depth was statistically different from the top three sampling depths but statistically not different from the two depths (50 and 60 cm) below this layer. The 80-cm depth had the highest mean scaled frequency.


View this table:
[in this window]
[in a new window]
  		Table 8. Neutron scattering calibration equations for different soil layers using a three-parameter power model.

 
The interaction between moisture level and sampling depths was also statistically significant; Tukey's mean separation procedure was again used to separate these effects. Large and small scaled-frequency values were reported for the lower seven sampling depths (40–100 cm) and the top three sampling depths (0–30 cm). Of the possible 30 group combinations, 16 groups did not show significantly different means from one another. The clear soil physical property differences between the top 35- and the lower 65-cm parts of the soil profile (Table 1) explain their different responses to the same treatments. These results suggest that separate calibration equations are needed for these two different parts of the soil profile.

Following previous studies (Buss, 1993; Mead et al., 1995; Paltineanu & Starr; 1997; Morgan et al., 1999; Baumhardt et al., 2000), we used three-parameter power relationships for calibrating the MCS. Figure 3 shows the fitted power equations representing the relationship between the measured volumetric water contents and the recorded scaled frequencies. Separate calibration equations were generated for the upper sandy clay loam surface soil and the lower clay subsurface (Table 9 and Fig. 3). Calibration of sandy clay loam surface soil covered a wider water content range (0.038–0.350 cm3cm–3) and had a larger correlation coefficient (0.98) than the clay subsoil which had a water content range of 0.183 to 0.406 cm3cm–3. However, the regression model for the lower horizon had a RMSE that was 37% smaller than that of the upper horizon. Although the range of water contents of the clay subsoil was relatively small, it covered the practical range of plant-available water defined as the difference between water content at field capacity and permanent wilting point. Using the calibration equation outside this range may produce inaccurate water content values.



View larger version (23K):
[in this window]
[in a new window]
  		Fig. 3. Plot of field-measured water contents for the entire soil profile (0–100 cm) versus MSC scaled-frequency (SF) readings.

 

View this table:
[in this window]
[in a new window]
  		Table 9. Calibration equations of the multisensor capacitance system for different soil layers.

 
The data of the entire soil profile were also pooled and fitted to determine one calibration equation for the entire soil profile (Table 9 and Fig. 3). As expected, the resulting model had a smaller correlation coefficient and a larger RMSE as compared with separate models for the two soil textures. The pooled model was closer to the calibration equation of the lower clay subsoil than to the upper sandy clay loam surface soil (Fig. 3). The use of one equation to represent the entire soil profile explained 88% of relationship between the scaled frequency and the water content, with a RMSE of 0.032 cm3 cm–3 (Table 9). With the exception of the upper 10-cm layer, using individual calibration equations for each 10-cm soil depth improved the accuracy of the MCS as compared to the use of a single calibration equation for the entire soil profile.

To compare our results with previous MCS calibration research (Paltineanu and Starr, 1997; Baumhardt et al., 2000), calibration equations provided by the manufacturer are plotted in Fig. 4 along with the field data and the fitted calibration equation for the entire profile. Our calibration model had a much smaller correlation coefficient and a larger RMSE as compared with the laboratory calibrations conducted by Paltineanu and Starr (1997) and Baumhardt et al. (2000). Paltineanu and Starr (1997) pointed out that field calibration will rarely produces RMSE values that will be similar to laboratory-derived values. Our study suggests that site-specific calibration will improve the accuracy of soil water monitoring as compared with the use of manufacturer or laboratory calibration results. Also, high contents of shrinking–swelling clay minerals will affect the performance of the MCS consistent with previous studies by Baumhardt et al. (2000) and Paltineanu and Starr (1997).



View larger version (23K):
[in this window]
[in a new window]
  		Fig. 4. Field-measured water contents (circle with black dots) for the top 3 depths of the soil profile (0–35 cm) regressed against capacitance probe scaled-frequency (SF) readings. Plots of field-measured water contents vs. MSC scaled frequencies for the upper (0–35 cm) and lower (35–100) soil horizons, calibration equations for the upper (solid line) and lower (dotted line) horizons, and the profile as a whole (dashed line).

 
Accuracy of the calibration exercise necessitates determining the bulk densities for each soil sample included in the analysis. Figure 5 shows the calibration equations based on literature values and carefully measured values of bulk densities for the top 3 and the lower 7 sampling depths (Fig. 5a) and for the entire soil profile (Fig. 5b). Calibration equations using bulk density reported in the literature underestimate water contents as compared with calibration equations based on volumetric water content data calculated from sample specific bulk densities.



View larger version (25K):
[in this window]
[in a new window]
  		Fig. 5. MCS calibration models developed for the top 3 depths of the soil profile (0–35 cm), the lower 7 depths (35–105 cm) A, and the entire soil profile B using site-determined and literature values of the bulk densities.

 
Finally, we also compared the NS manufacturer calibration equation to the calibration equation determined using field gravimetric water contents multiplied with either literature-based soil bulk densities or sampled specific bulk densities (Fig. 6a, 6b, 6c) . Results showed that field-based calibration equations using site-specific or average bulk densities were very different from the default calibration equation. In particular, the default calibration equation for the upper 10-cm soil layer underestimated the water content by as much as 50%.



View larger version (28K):
[in this window]
[in a new window]
  		Fig. 6. Effect of using manufacturer calibration equation, site-specific bulk density, and bulk densities from the literature on NS calibration models developed for (a) the surface 10-cm depth, (b) following two depths (20 and 30 cm), and (c) the lower seven depths (35–105 cm).

 

   SUMMARY AND CONCLUSIONS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 SUMMARY AND CONCLUSIONS
REFERENCES
 
A field calibration was conducted for the MCS and NS devices in a shrinking–swelling soil profile. The soil used in this calibration had two distinct soil textures: a sand clay loam surface (0–35 cm) and a clayey subsoil (35–100 cm). Three soil water content treatments (dry, moist, and wet) were used. To eliminate the effect of spatial variability on the measurements, both devices used the same access tube. Results showed up to 20% larger soil bulk densities at the smaller water contents. NS and MCS regression equations were generated for the two horizons. The NS linear calibration equations for the combined 20- and 30-cm layers and the lower profile depths had a similar correlation coefficient of 0.87 and RMSEs of 0.019 and 0.022 cm3 cm–3, respectively. Calibration of the sandy clay loam surface soil involved a much wider water content range (0.038–0.350 cm3cm–3) than that of the clay subsoil (0.183–0.406 cm3cm–3). The NS data points also were fitted using three-parameter power regression models. This approach slightly improved the correlation coefficient and lowered the RMSEs for some of sampling depths. We also obtained three-parameter power calibrations between the measured volumetric water contents and the scaled frequencies of the MCS for the top sandy clay loam surface soil and the clay subsurface soil. The regression model for the lower horizon had an RMSE that was 37% less than that of the upper sandy clay loam surface soil. Our study shows that site-specific calibration will improve the accuracy of soil water–monitoring devices as compared to using manufacturer calibration equations.


   ACKNOWLEDGMENTS
 
Appreciation is expressed to Shannon Pudney of the South Australian Research and Development Institute, Nuriootpa Research Station, Nuriootpa, South Australia, for her help during the field experiments. Thanks also to Nicole Wilson and Tanya Caceres for their help in the field. Special thanks to Louise Clark at the School of Earth and Environmental Science, University of Adelaide, South Australia, for sharing her manuscript and the organic carbon data. A special thanks goes to Susie Olden at Sentek, Pty, Ltd. for her valuable comments.


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 SUMMARY AND CONCLUSIONS
 REFERENCES
 




This article has been cited by other articles:


Home page
 Soil Sci.Home page
A. Fares, H. Hamdhani, and D. M. Jenkins
Temperature-Dependent Scaled Frequency: Improved Accuracy of Multisensor Capacitance Probes
Soil Sci. Soc. Am. J., May 16, 2007; 71(3): 894 - 900.
[Abstract] [Full Text] [PDF]

Home page
 Vadose Zone JHome page
V. Polyakov, A. Fares, and M. H. Ryder
Calibration of a Capacitance System for Measuring Water Content of Tropical Soil
Vadose Zone J., November 11, 2005; 4(4): 1004 - 1010.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via ISI Web of Science (8)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Fares, A.
Right arrow Articles by Parsons, L. R.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Fares, A.
Right arrow Articles by Parsons, L. R.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Fares, A.
Right arrow Articles by Parsons, L. R.
Related Collections
Right arrow Vadose Zone Processes and Chemical Transport
Right arrow Other Geophysical Methods
Right arrow Soil Hydrology

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Crop Science
Journal of Natural Resources
and Life Sciences Education		 Soil Science Society of America Journal
Journal of Plant Registrations Journal of
Environmental Quality		 The Plant Genome