SSSAJ Journal of Natural Resources and Life Sciences Education
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 (2)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Zhang, N.
Right arrow Articles by Loughin, T. M.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Zhang, N.
Right arrow Articles by Loughin, T. M.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Zhang, N.
Right arrow Articles by Loughin, T. M.
Related Collections
Right arrow Water Content
Right arrow Other Geophysical Methods
Right arrow Soil Physics
Published in Soil Sci. Soc. Am. J. 68:1515-1525 (2004).
© 2004 Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA

DIVISION S-1—SOIL PHYSICS

Simultaneous Measurement of Soil Water Content and Salinity Using a Frequency-Response Method

N. Zhanga,*, G. Fana, K. H. Leea, G. J. Kluitenbergb and T. M. Loughinc

a Dep. of Biological and Agricultural Engineering, Kansas State Univ., Manhattan, KS 66502
b Dep. of Agronomy, Kansas State Univ., Manhattan, KS 66502
c Dep. of Statistics, Kansas State Univ., Manhattan, KS 66502

* Corresponding author (zhangn{at}ksu.edu)


   ABSTRACT
TOP
 NOTES
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Laboratory tests were conducted to simultaneously measure soil water content and salinity using a four-electrode Wenner array sensor. The sensor was modified to enhance the capacitive effect. Soil bulk density and the depth to which the electrodes penetrate into the soil were strictly controlled during the experiments. Sinusoidal current signals with a constant amplitude and frequencies ranging from 1 Hz to 15 MHz were sent to the outer electrodes of the Wenner array, whereas voltage outputs were measured from the inner electrodes. Frequency-response data were analyzed using the partial least-squares method to establish calibration models for simultaneously predicting water content and salinity from the frequency-response patterns. The R2 values for predicting water content and salinity at the 30-mm penetration depth reached 0.89 and 0.91, respectively, whereas the root-mean-square errors for the volumetric water content and salinity measurements were 0.019 m3 m–3 and 0.173 cmol kg–1, respectively. Test results showed that, in general, the calibration models predicted the water content more accurately than salinity. The depth to which the sensor penetrates into the soil has a strong effect on the measurement accuracy. This study has demonstrated that the modified Wenner array sensor and the frequency-response method have a potential for simultaneously measuring soil capacitive and conductive properties. However, numerous difficulties, including contact resistance, depth control, and the effect of soil type, will need to be addressed to improve the measurement accuracy.

Abbreviations: EC, electrical conductivity • PLS, partial least squares • RMS, root mean square • TDR, time domain reflectometry


   INTRODUCTION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
SOIL PLAYS A KEY ROLE in crop production as a physical support and a reservoir of water and nutrients. Site-specific crop management decisions for optimized input rates of water, fertilizer, pesticides, and seeds are largely based on physical, chemical, and biological properties of soils. Traditional soil surveys and accompanying soil databases are too general for use in site-specific farming systems, and the current method of intensive grid sampling requires a sizeable investment of money and time. Therefore, inexpensive sensors that are capable of measuring multiple soil properties in real-time are needed.

Soil permittivity is a good indicator of several important soil properties closely related to crop productivity. Soil relative permittivity is composed of a real component, the dielectric constant, and an imaginary component, the equivalent dielectric loss (Kraus, 1984) (Eq. [1] and [2]).


[1]

[2]
where {epsilon} equals permittivity, {epsilon}r equals relative permittivity, {epsilon}0 equals permittivity of free space (8.85 x 10–12 F m–1), {epsilon}'r equals dielectric constant (the real part of relative permittivity), {epsilon}''r equals equivalent dielectric loss taking into account the conductive loss (the imaginary part of relative permittivity), and j = .

The complex form of soil permittivity is analogous to impedance, which is composed of a resistive and a reactive (capacitive and inductive) component with a phase difference of 90°. Soil permittivity is strongly affected by frequency. In fact, both the real and imaginary parts of relative permittivity are functions of frequency (Eq. [3] and [4]). Thus, the frequency, at which the dielectric permittivity is measured, plays a strong role on the effectiveness of the measurement.


[3]

[4]
where {epsilon}''d equals dielectric loss, {sigma}/({omega}{epsilon}0) equals conductive loss, {sigma} equals electrical conductivity (EC), {omega} equals angular frequency, and {delta} = loss angle.

The capacitive (dielectric) behavior of soils has been used to measure soil volumetric water content. A capacitance probe uses a pair of electrodes that insert into the soil to form a capacitor. This capacitor and an inductor in an oscillator circuit determine the oscillation frequency. Changes in volumetric water content vary the capacitance and, in turn, change the oscillation frequency. Test results indicated that solution ionic conductivity within soil water always causes increases in the dielectric permittivity measurement, indicating the effect of the imaginary part of relative permittivity on soil dielectric characteristics (Robinson et al., 1998).

The conductive behavior of soils has been used to measure bulk soil EC with sensors of various designs, including a contact-type design, which places four electrodes in a Wenner array configuration (Fig. 1) on the immediate surface of soil, and a noncontact, nondestructive design that uses the electromagnetic induction principle (McNeill, 1980). These sensors operate in the low frequency range. Wenner array sensors have been tested extensively and were found highly accurate in measuring soil salinity, because soil salinity is the major factor determining bulk soil EC. However, since soil water content also has a strong effect on bulk soil EC, measurement of soil salinity using this type of sensor has to be made under known soil water conditions, such as "2 to 3 d after irrigations" (Rhoades and Ingvalson, 1971). A commercial soil EC sensor using the Wenner array configuration has been applied in many precision agriculture experiments for fast field measurement. However, due to the interactions between the capacitive and conductive characteristics of soils, this sensor is incapable of separating the main factors affecting the bulk soil EC–soil water content and salinity (Lund et al., 1998).



View larger version (31K):
[in this window]
[in a new window]
  		Fig. 1. Four-electrode Wenner array.

 
Interactions between the capacitive (dielectric) and the conductive behaviors of soils can be explained partially by the complexity of electric conductance in soils, involving conductance through the solid, liquid, and gas phases. Several models describing the pathways for electrical conductance within soil have been established through many years of experiments (Rhoades and Van Schilfgaarde, 1976; Rhoades et al., 1989). The latest model considered pathways through the solid phase, which is due to exchangeable cations adsorbed to layer silicates, and the liquid phase, which is due mainly to the ions, including cations and anions of salts dissolved in water. Pathways for the liquid phase are comprised of series-coupled pathways through fine pores and separated pathways through large pores. Therefore, volumetric soil water content in these pathways directly affects the electrical conductance, indicating a strong effect of the dielectric behaviors of soil on its conductive behaviors. On the other hand, Campbell (1990) reported that ionic conduction could make the real part of relative permittivity rather complicated, especially at low frequencies, indicating a strong effect of the conductive behaviors of soil on its dielectric behaviors.

Time domain reflectometry (TDR) sensors have been used to simultaneously measure soil water content and bulk soil EC (Topp et al., 1980). Volumetric water content is measured from the time consumed for the first reflection of a TDR pulse signal traveling through a probe immersed in soil, whereas the bulk soil EC is measured from the amplitude of the reflection signal. When the amplitude of the first reflection is used, EC is always overestimated, because this is a measurement of the transient response, which reflects mainly the high-frequency behavior (dielectric behavior) of the soil. It has been recommended that the EC measurement be taken at a time much longer than the TDR pulse rise time, so that the measurement would reflect the low-frequency (close to DC) steady-state response, where the capacitive effect is minimized (Topp et al., 2000). The effect of ionic conduction on the measurement of water content was also observed on TDR devices. For soils with high EC, energy of the TDR signal tends to dissipate substantially due to conductive loss. As a result, the reflection becomes difficult to detect (Heimovaara, 2001). Thus, water content measurement using TDR is not recommended for soils with high salinities (Topp et al., 2000). Simultaneous measurement of water content and salinity using TDR has attracted much attention during the recent years. Efforts included improvement in the probe and hardware design and study of the waveforms using time-domain dynamic analysis, aiming at more effective separation of the dielectric and conductive behaviors (Robinson et al., 2003).

Effects of the conductive behaviors on the capacitive behaviors of soils can be minimized only at a sufficiently high frequency range. Gaudu et al. (1993) claimed that the electrical conductance of soils has a negligible effect on their capacitive water-content sensor operating at 40 MHz. Gardner et al. (1998) found a low-frequency (<50 MHz) capacitive water-content sensor sensitive to soil EC. Eller and Denoth (1996) operated a capacitive water-content sensor at 32 MHz and noticed a slight reduction in accuracy due to EC in a wet organic soil. Robinson et al. (1998) operated a capacitance probe in the frequency range of 80 to 150 MHz. Topp et al. (1980) used the TDR method to measure soil water content using step signals with a bandwidth of 1 MHz to 1 GHz. To avoid the effect of EC on the measurement of the real part of permittivity, Hilhorst (1998) measured the real and imaginary parts of relative permittivity independently using a frequency-domain sensor at 20 MHz.

Effects of the capacitive behaviors on the conductive behaviors of soils can be minimized at low frequencies. Hilhorst (1998) stated that soil EC can be measured more accurately than dielectric permittivity at low frequencies because, at these frequencies, electric currents in soil due to conductance are higher than those due to capacitance. Rhoades and Van Schilfgaarde (1976) measured soil salinity using a Wenner array sensor at frequencies of 10 to 20 Hz.

This study intended to develop a sensor and a frequency-response method to simultaneously measure soil volumetric water content and salinity within a relatively low frequency range of 1 Hz to 15 MHz. This frequency range was selected based on hardware limitations. Specific objectives of this study were to (i) design a sensor that is capable of measuring both the capacitive and conductive behaviors of soils, (ii) design the hardware and software needed for signal acquisition, processing, and analysis within the 1-Hz to 15-MHz frequency range, and (iii) establish calibration models through multivariate analyses of the trends of the frequency-response data to quantitatively separate the effects of volumetric water content and salinity.


   MATERIALS AND METHODS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Soil Sample Preparation
A silt loam soil (16% sand, 67% silt, 17% clay) with a pH of 7.5, CEC of 20 cmol kg–1, and organic C content of 10 g kg–1 was used. Soil samples with different water contents were prepared in covered containers. Before adding water, soil samples were oven-dried at 105°C for 24 h and then ground to pass through a 2-mm sieve. Known weights of water were added to the soil samples to achieve desired water contents. Measurements were initiated at least 24 h after the water was added to allow the sample to reach equilibrium. The actual water contents were reexamined and recorded after each test by weighing. The intended gravimetric water contents were 10, 15, 20, and 23%. Actual gravimetric water contents slightly deviated (±1% max.) from these values. Volumetric water contents were calculated from the measured values of gravimetric water contents using a bulk density of 1.25 g cm–3, which was maintained for all soil samples used in the experiment.

For each of the four levels of salt concentration (0, 0.68, 1.366, and 2.03 g kg–1 soil), a calculated amount of NaCl was dissolved in water and added to the soil. After the experiment, soil samples from the containers were sent to the Soil Testing Laboratory at Kansas State University to determine soluble salt concentrations (soluble cations) of saturated paste extracts. Because the soil was not washed with NaCl to replace all exchange sites with Na, soluble salts also included contributions from Ca2+, Mg2+, and K+. The actual measured soluble salt concentrations were 0.110, 0.722, 1.129, and 1.752 cmol kg–1, respectively. These values were used as the salinity levels in this study.

Solution EC of soil samples was estimated by (i) dividing mass-based salinity (cmol kg–1) by gravimetric water content, and (ii) converting to solution EC using the conversion factor provided by Rhoades et al. (1999) (1 cmol L– 1 {cong} 1 dS m–1). These values were used as the actual solution EC, together with frequency-response data, to establish models for simultaneously predicting volumetric water content and solution EC. Solution EC estimated using this procedure ranged from 0.5 to 16.5 dS m–1.

A preliminary test was conducted to select a bulk density that was achievable at all the intended water contents for the soil. Bulk densities of 1.0, 1.05, 1.1, 1.15, 1.2, 1.25, 1.3, and 1.35 g cm–3 were tested and 1.25 g cm–3 was finally selected for the experiment. A universal material testing machine (Instron 4506, Instron Corp., Canton, MA) was used to compress the soil samples to the desired density. Soil compression was accomplished by an aluminum piston mounted on the Instron machine. The position of the piston was fixed and the container was moved upwards by the machine (Fig. 2). The total amount of soil needed in each container was previously calculated based on the designated water content for that container, the total depth of the sample (70 mm) in the container, the geometry of the container, and the desirable bulk density (1.25 g cm–3). After the soil sample was weighed, it was evenly divided into seven subsamples by weight. After each soil subsample was added to the container, the container was slowly moved upwards 10 mm. Thus, after all seven subsamples were added, the bulk density of the sample was approximately 1.25 g cm–3.



View larger version (132K):
[in this window]
[in a new window]
  		Fig. 2. Control of soil sample density.

 
Sensor Design and Experimental Setup
The prototype sensor was designed based on the four-electrode Wenner-array structure (Fig. 1). The four electrodes were made in the shape of circular discs with a diameter of 100 mm. The discs were made of brass alloy. They were assembled on a wooden square shaft and were separated using plastic spacers. The space between adjacent discs was 15 mm (Fig. 3). The disc-type design was intended for field applications, where the discs may be mounted on the three-point-hitch of a tractor or pulled by a truck to rotate freely on soil surface, as in the case of a commercial, Wenner-array-based soil EC sensor (Lund et al., 1998).



View larger version (97K):
[in this window]
[in a new window]
  		Fig. 3. The sensor used in the experiment.

 
The relatively large area of discs and the relatively small separation between the discs were selected with an intention to enlarge the capacitive effect of the sensor. This is a deviation from the traditional Wenner array sensors, which were designed to measure only the conductive behaviors of soils. Penetration depth and electrode size were considered unimportant in design of the traditional Wenner array sensors (Rhoades and Ingvalson, 1971).

Sinusoidal current signals were supplied to the outer electrodes of the Werner array. These signals were generated by a function generator (Hewlett-Packard, Palo Alto, CA, model 33120A) followed by a voltage-controlled-current-source circuit. The amplitude of the controlling signals was maintained constant (5 V) and the frequency varied from 1 Hz to 15 MHz. The root-mean-square (RMS) values (Vrms) and the average values (Vavg) of the output voltage were measured from the inner electrodes of the Werner array by a duel-beam, mixed-signal oscilloscope (Hewlett-Packard, model 54645D). Both the function generator and the oscilloscope were interconnected with a PC computer through the General Purpose Interface Bus (GPIB) (Fig. 4).



View larger version (14K):
[in this window]
[in a new window]
  		Fig. 4. Functional diagram of the experimental system.

 
A LabVIEW (National Instruments, 2000) program was written to control the system. The program was capable of adjusting the frequency of the sinusoidal signal generated by the function generator in the range of 1 Hz to 15 MHz at 2-s intervals. The LabVIEW program also controlled the oscilloscope to measure, display, and record the RMS (Vrms) and average (Vavg) values of the output voltage signals from the sensor.

Experimental Design
An experiment was designed to observe the frequency responses of the sensor at combinations of four water contents and four salinities. For these combinations, 16 cylindrical containers were used. These containers were made of plexiglass with an inner diameter of 218 mm and a depth of 130 mm. For each water content-salinity combination, the sensor penetrated into the center of the soil samples to depths of 10, 20, 30, and 40 mm, respectively, using the Instron machine. A ruler was mounted on the wall of each container to indicate the penetration depth (Fig. 2). For each water content-salinity combination, the test was replicated twice.

For each test, Vrms and Vavg values of the sensor output were measured at 45 frequencies, which included 31 frequencies at even intervals on a logarithmic frequency scale between 1 Hz and 1 MHz and 14 frequencies at even intervals on a linear frequency scale between 2 and 15 MHz. The 45 Vrms and 45 Vavg values from each test form a frequency-response data. With 64 combinations of four water contents, four salinities, four penetration depths and two replicates for each combination, the total number of frequency-response data was 128 (Fan, 2002).

Data Analysis
The frequency-response data were used to build quantitative calibration models through the partial least squares (PLS) regression procedure (Beebe and Kowalski, 1987). The PLS procedure was performed using GRAMS/32 (Galactic Industries Corp., 1996), a spectroscopic software package that combines data importing, processing, viewing, organizing, and accessing capabilities.

Data Preprocessing
The mean-center procedure available in the GRAMS software was used to preprocess the frequency-response data. This procedure calculated the mean frequency-response of all 128 frequency-response data and then subtracted the mean from each data. In addition, the mean values of water content, salinity, and penetration depth of the 128 samples were also calculated and subtracted from individual values. This procedure was designed to remove the common portion of variations from the data to allow more efficient computation.

Partial Least Squares Calibration
Partial least squares is a statistical multivariate calibration procedure that is widely applied in spectroscopic analyses for chemical compositions. Being a multivariate procedure, PLS provides the ability to predict multiple components of interest simultaneously.

The PLS procedure simultaneously estimates underlying factors (loading factors or eigenvectors) that represent the variation patterns (trends) in both the spectral data R (frequency-response data in this study) and the concentration values C (water content, salinity, solution EC, and penetration depth in this study). These loading factors are used to define a subspace in R that better models C. This is accomplished by using the columns of C matrix (sample values for individual concentrations) to estimate the loading factors for R matrix. At the same time, the columns of R matrix (responses of samples at individual frequencies) are used to estimate the loading factors for C matrix. The resulting models are shown in Eq. [5] and [6].


[5]

[6]
where R equals spectral data—the frequency-response data (Vrms and Vavg measured at 45 frequencies using the sensor developed in this study), C equals concentration values—water content, salinity, solution EC, and penetration depth values, P equals loading factors for the spectral (frequency-response) data, Q equals loading factors for the concentration values (water content, salinity, solution EC, and penetration depth), T and U equal scores assigned to P and Q, respectively, and E and F equal errors associated with modeling R and C, respectively.

The factors for R and C are associated through the following relationship:

[7]
where u and t equal the column vectors of U and T matrices, respectively, and {epsilon} equals errors associated with the u–t relationship.

In Eq. [7], b is termed the inner relationship between u and t and is used to calculate subsequent loading factors if more than one loading factor is necessary to describe the variation. Substituting Eq. [7] into Eq. [6] modifies U and , consequently updates C. (Beebe and Kowalski, 1987).

Using both spectral and concentration information to determine the loading factors is the main difference between the PLS and other statistical multivariate procedures, such as multiple linear regression and principal component regression. This feature makes the PLS prediction models more robust for complex spectral data. On the other hand, PLS models are linear models, which are suitable for chemical composition analyses because of the linear relationship between constituent concentration and energy absorption (Beer-Lambert Law) at specific wavelengths. However, a linear model may not provide the best prediction for this study, because a linear relationship between the frequency-response data at specific frequencies and the concentrations to be measured—water content, salinity, solution EC, and penetration depth—may not exist.

The number of loading factors used in the final calibration model was selected through a cross-validation procedure. The first step was to remove a sample, including a spectral (frequency-response) data and corresponding concentration values (water content, salinity, solution EC, and penetration depth), from the entire training data set, which contains all the spectral data and corresponding concentration values used for training the calibration model. Next, calibration models with different numbers of loading factors, starting from one, were established using the remaining training data set following the PLS procedure. The resulting models were then used to predict the concentration values of the removed sample. This procedure was repeated for each sample within the training data set. The errors obtained from all models with the same number of loading factors were then used to calculate the "prediction residual error sum of squares" (PRESS) as a function of the number of loading factors. From the trend of PRESS, an optimum number of loading factors can be selected. A number of loading factors was considered optimum if the PRESS reached a minimum at this number and increased steadily as the number further increases, indicating an overfitting of the model by introducing factors that contain noises, which are not related to the concentration of interest. Once the optimum number of loading factors is determined, the entire frequency-response training data set can be used to develop the final prediction model following the same PLS procedure discussed above (Galactic Industries Corp., 1996).

Partial least squares calibration models were established using the frequency-response data. Separate models were established to simultaneously predict volumetric water content and salinity, or volumetric water content and solution EC, using data obtained at individual penetration depth—10, 20, 30, and 40 cm. A model was also developed using the frequency-response data obtained at all four depths to simultaneously predict water content, salinity, and penetration depth.


   RESULTS AND DISCUSSION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
CONCLUSIONS
 REFERENCES
 
Samples of the frequency-response data are shown in Fig. 5, 6, and 7. The Vrms values shown in these figures are the average values of two replications. In general, both higher water contents and higher salinities resulted in lower voltage output of the sensor, indicating greater EC. The abrupt rise in frequency responses around 10 MHz was caused by insufficient bandwidth of the current-source circuit. To evaluate the usefulness of the 10- to 15-MHz portion of the data in the prediction models, loading factors selected by the PLS procedure were examined. For both water content and salinity, the first loading factors assigned lower weights to the 10- to 15-MHz frequency range. However, weights assigned to the 10- to 15-MHz portion in the second loading factors were in general higher than those assigned to the lower frequency region, especially for salinity prediction. Thus, frequency-response data measured in the entire 1-Hz to 15-MHz range were used in the models.



View larger version (24K):
[in this window]
[in a new window]
  		Fig. 5. Frequency responses of the sensor at four volumetric water contents, 0.11 cmol kg–1 salinity, and the 40-mm penetration depth.

 


View larger version (24K):
[in this window]
[in a new window]
  		Fig. 6. Frequency responses of the sensor at four salinities, 0.19 m3 m–3 volumetric water content, and the 40-mm penetration depth.

 


View larger version (20K):
[in this window]
[in a new window]
  		Fig. 7. Frequency responses of the sensor at four penetration depths, 0.29 m3 m–3 volumetric water content, and 1.752 cmol kg–1 salinity.

 
Figure 5 also showed that, at the water content of 0.29 m3 m–3, the signal was rather unresponsive to frequency. This trend exists at other salinity levels but is more distinguished at the lowest salinity level, as shown in the figure. At the greatest water content, the resistive impedance of soil is much smaller than its capacitive impedance. Thus, the capacitive effect is minimized and the EC of soil becomes frequency-independent within the frequency range of 0 to 10 MHz.

Penetration depth also affected the sensor's frequency response (Fig. 7). The voltage output of the sensor decreased with penetration depth, which apparently was a result of reduced resistance due to enlarged contact area between the soil and the electrodes.

Performances of the calibration models developed at individual, as well as all penetration depths on predicting volumetric water content and salinity are compared in Table 1. The models developed for 10- and 20-mm depths had low R2 values and high RMS errors, indicating poor prediction abilities. It is believed that one of the main causes for the poor performance was large contact resistance between the electrodes and soil. When the electrodes were not pushed to a sufficient depth, the contact between the electrodes and soil may be loose, resulting in increased resistance and large measurement errors.


View this table:
[in this window]
[in a new window]
  		Table 1. Performances of calibration models for simultaneously predicting water content and salinity trained at different penetration depths using all frequency-response data.

 
The enhanced capacitive effect of the sensor may also have contributed to the better performance of the models developed for greater depths. The capacitance between adjacent electrodes of the sensor can be considered a parallel connection of two capacitors: the portion of the electrodes above the soil surface form the first capacitor, which uses air as the dielectric medium; the portion of the electrodes within the soil form the second capacitor, which uses soil as the dielectric medium. Deeper penetration of the sensor into soil enlarges the area of the plates of the second capacitor, thus, enlarging its capacitance. When the sensor is not pushed into soil deeply, the capacitive effect of the sensor may be trivial compared with the conductive effect due to the low dielectric permittivity of air, and the capacitive and conductive effects of soil become difficult to separate.

To validate this assumption, the capacitance and resistance between two adjacent disks on the sensor were measured in soils using a multiple-frequency LCR meter (Hewlett-Packard, Palo Alto, CA; model 4275A). Measurements were done in combinations of two volumetric water content levels (0.12 and 0.29 m3 m–3) and two salinity levels (0.11 and 1.752 cmol kg–1) at two penetration depths (10 and 40 mm). Capacitance and resistance were read at 10 frequencies, from 10 kHz to 10 MHz. The resistance measured at the 10-mm depth (70 {Omega}–2 k{Omega}) was 4 to 12 times the resistance measured at 40 mm (17–550 {Omega}). At both depths, a lower resistance was found in a wetter soil. On the other hand, the capacitance measured at the 40-mm depth (70 pF–0.045 µF) was 2 to 25 times the capacitance measured at 10 mm (25 pF–0.01 µF). At both depths, the highest capacitance was found in a wetter and more saline soil at the lowest measurement frequency. Water content played a major role in both capacitance and resistance. These measurements clearly demonstrated that penetrating the sensor into soil allowed the measurement of both capacitive and conductive behaviors of soil. Moreover, the values of capacitance and resistance, and the amounts of change in these two variables were all at the measurable level.

The best prediction models were found at the 30-mm penetration depth, at which R2 values of 0.89 and 0.83 and RMS errors of 0.020 m3 m–3 and 0.249 cmol kg–1 were achieved for predicting water content and salinity, respectively (Fig. 8 and 9). In general, the models developed for predicting water content performed better than those for salinity. Again, this may be explained by the fact that water content has a strong effect on both the capacitive and conductive behaviors of soil, whereas salinity mainly affects the conductive behavior. Furthermore, the effect of water content on conductivity may be stronger than that of salinity, especially at high water contents, because an appreciable amount of electrolytes in saline soils can be dissolved to conduct electric current only with the presence of water. The strong effect of water on soil EC increased the difficulty in predicting salinity from EC measurement when the water content changes.



View larger version (22K):
[in this window]
[in a new window]
  		Fig. 8. Actual and predicted volumetric water contents at the 30-mm penetration depth.

 


View larger version (21K):
[in this window]
[in a new window]
  		Fig. 9. Actual and predicted salinities at the 30-mm penetration depth.

 
It is interesting to notice that the number of loading factors selected through the cross-validation procedure decreased when the sensor penetration depth increased. A minimum number (2) was reached at the 30-mm depth. The excessive number of loading factors selected at the 10-mm depth clearly indicates the effect of noise on the recognition of major variation patterns. Two loading factors selected for models established using data with two major factors influencing the variation pattern should be viewed as a good sign for variable separability.

Table 1 also reports the performance of models established to simultaneously predict volumetric water content and soil solution EC. These models did not overperform the models developed for simultaneous prediction of water content and salinity. This result indicates that mass-based salinity was better related to bulk properties (as detected by voltage responses of the four-electrode sensor) than solution EC over a range of water contents and salinity. It also seems to have agreed with the following statement of Rhoades et al. (1999): "ECa (bulk EC) is primarily a measure of the content of dissolved electrolyte present in a unit-volume of soil."

The model established using data taken at all depths (Table 1) predicted volumetric water content reasonably well. However, adding the third factor, the penetration depth, ruined the ability to predict salinity. On the other hand, the reasonably high R2 value (0.65) and the relatively low RMS error (6.7 mm) (not shown in Table 1) for predicting the penetration depth indicated the importance of maintaining a constant penetration depth when simultaneously measuring water content and salinity using this sensor. The fact that more loading factors were recommended by the cross-validation procedure (10, 6, and 5 for water content, salinity, and depth, respectively) indicated the difficulties in identifying the variation patterns caused by the three main factors.

Two outliers were found in Fig. 9. These are the two replicate measurements at the highest salinity level (1.752 cmol kg–1) and the lowest water content (0.12 m3 m–3), representing dry and saline condition. These two measurements were considered outliers, because this water content was very close to a threshold of 0.1 m3 m–3, below which it becomes inappropriate to infer salinity from bulk EC measurements using a Wenner array (Rhoades et al., 1999). The PLS calibrations were conducted with the two outliers removed. This resulted in improved R2 value (0.91) and reduced RMS error (0.173 cmol kg–1) for salinity prediction (Fig. 10). Performances of the models predicting water content were basically not changed (Table 2).



View larger version (21K):
[in this window]
[in a new window]
  		Fig. 10. Actual and predicted salinities with two replicate frequency-response data collected at the highest salinity (1.752 cmol kg–1) and the lowest water content (0.12 m3 m–3) removed.

 

View this table:
[in this window]
[in a new window]
  		Table 2. Performances of calibration models for simultaneously predicting water content and salinity trained at different penetration depths. Frequency-response data collected at the highest salinity (1.752 cmol kg–1) and the lowest water content (0.12 m3 m–3) were not used.

 
Traditional Wenner-array sensors have been repeatedly proven effective in measuring soil salinity (Rhoades and Ingvalson, 1971; Rhoades and van Schilfgaarde, 1976). However, this measurement can be made accurately only under a relatively constant water content (Rhoades and Ingvalson, 1971). To examine the effectiveness of the sensor developed in this study in measuring salinity under a constant water content, we developed four PLS models using frequency-response data obtained at four fixed volumetric water contents, respectively, and obtained good fits (R2 = 0.943–0.986 and RMSE = 0.070–0.148 cmol kg–1) between the predicted and actual salinities (Fig. 11). These results demonstrate that the sensor designed in this study is still capable of measuring salinity accurately as the traditional Wenner-array sensors did. Similar models were established to predict only volumetric water content at fixed salinity levels. The R2 values of 0.969 to 0.994 and RMS errors of 0.004 to 0.011 m3 m–3 were achieved by these models (Fig. 12).



View larger version (28K):
[in this window]
[in a new window]
  		Fig. 11. Performances of calibration models established for predicting salinity at four fixed volumetric water levels and a fixed penetration depth (30 mm).

 


View larger version (28K):
[in this window]
[in a new window]
  		Fig. 12. Performances of calibration models established for predicting volumetric water content at four fixed salinity levels and a fixed penetration depth (30 mm).

 
It should be pointed out that the sensor developed in this study is not different from the traditional Wenner-array sensor when used within a low frequency range (e.g., 1–300 Hz). Therefore, the sensor can still be used to directly measure bulk soil EC, a comprehensive index that reflects the effects of both salinity and water content on soil conductivity, based on the output voltage signals and the sensor geometry. On the other hand, the sensor differs from traditional Wenner-array sensors in that it is capable of dealing with both conductive and capacitive effects within a relatively wide frequency range so that the effects of salinity and water content can be separated.

The frequency used in this study was from 1 Hz to 15 MHz, which is lower than the frequencies used in many other sensors designed for measuring the dielectric behaviors of soils. Many studies have shown that the capacitive behaviors of soils in this frequency range are still affected by the conductive behaviors. The difference between this sensor and other dielectric-based sensors is that this sensor uses multiple frequencies within the 1-Hz to 15-MHz range, whereas other sensors use only a single frequency. The use of multiple frequencies permitted detection of trends in the frequency-response data using various pattern-recognition algorithms to assist in separation of the conductive and dielectric effects.

While higher frequencies may further improve the ability of the sensor to simultaneously measure water content and salinity, reducing the bandwidth would reduce the hardware requirement and cost. Another factor to be considered for frequency selection is the transmission line effect. When high frequency signals are applied, the length of cables and the spacing between electrodes would have to be short to avoid signal reflection and phase shift. This would set a limit to frequency allowed for field-size sensors.

The disc shape design for the sensor was to allow the sensor to rotate behind a tractor or a truck for fast field measurement. However, this design did not consider deep penetration of the sensor into soil. Changing the sensor to a chisel type or a knife type design would allow easier penetration. The knife design also would allow larger electrode sizes to enlarge the capacitance. In fact, cutting into soils would provide a tighter contact between the sensor electrodes and soil, especially when the sensor is in a constant move within the soil. This would greatly reduce the contact resistance. To maintain the penetration depth, a depth-control mechanism, either mechanical or electronic type with an embedded microcontroller, would help improve the measurement accuracy.

It should be pointed out that only one type of soil has been tested for this sensor and the calibration models established in this study were not further validated using a separate frequency-response data set. In addition, factors studied in this study only included soil water content and salinity. The purpose of this study was to demonstrate the potential of the sensor and the frequency-response method in simultaneous measurement of multiple soil properties. Further experiments have been planned to test the sensor on different types of soils and to develop fully validated prediction models.

For TDR sensors, separation of capacitive and conductive measurements is accomplished by measuring two variables in the transient and steady state step-response data, respectively. This is a time-domain approach for dynamic analysis. For the sensor designed in this study, separation is accomplished through dynamic analyses of the frequency-response data. As it is well known in the system analysis field, system dynamics can be analyzed in time and frequency domains, or both. This paper only reported studies on the gain part of frequency response. We are currently studying the phase part of the frequency response, hoping this addition would provide more information to allow us to solve for more unknowns.


   CONCLUSIONS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
REFERENCES
 
In this study, a soil dielectric permittivity sensor was designed based on the traditional Wenner array, four-electrode configuration. The sensor deviated from the traditional Wenner array design by (i) enlarged electrode size, (ii) reduced spacing between electrodes, and (iii) allowing the electrodes to penetrate into soil so that the capacitive effect of the dielectric material (soil) can be detected.

Hardware and software were developed for laboratory tests of the sensor. Sinusoidal current signals with a constant amplitude and 45 frequencies ranging from 1 Hz to 15 MHz were supplied to the sensor. Soil samples with four water-content and four salinity levels were prepared. The bulk density for all samples was strictly controlled to 1.25 g cm–3. For each water content and salinity combination, tests were conducted at four controlled penetration depths. Frequency responses of the sensor (gain only) at various combinations of water content, salinity, and penetration depth were recorded.

Partial least squares calibration models were developed using the frequency-response data. The following conclusions can be drawn from these tests:

  1. The calibration models in general demonstrated a better prediction ability for water content than for salinity.
  2. Calibration models established to predict salinity at fixed volumetric content and depth and models established to predict volumetric water content at fixed salinity and depth all performed well.
  3. The best calibration models for simultaneously predicting volumetric water content and salinity were found at the 30-mm penetration depth, where R2 values of 0.88 and 0.83 were achieved for predicting water content and salinity, respectively. The RMS errors for predicting water content and salinity at this depth were 0.02 m3 m–3 and 0.25 cmol kg–1, respectively.
  4. With measurement outliers at the highest salinity and the lowest water content removed, the R2 value and the RMS error of the model predicting salinity at all volumetric water content levels were improved to 0.91 and 0.173 cmol kg–1, respectively. The performance of the model predicting water content was unchanged.
  5. The depth to which the sensor penetrates into soil has a strong effect on the measurement. To obtain better calibration models for both water content and salinity, a penetration depth of >20 mm is required for the specific geometry of the sensor designed in this study.
  6. The frequency range (1 Hz–15 MHz) used in this study allowed partial separation of the effects of water content and salinity.
  7. Measurement of capacitance and resistance in soil samples showed that the sensor provided measurable changes in both resistance and capacitance within the water content and salinity ranges tested.
  8. Laboratory test results indicated that the sensor and the data analysis method developed in this study—using multivariate analysis to analyze the patterns of the sensor's frequency response—has a potential for simultaneous measurement of soil water content and salinity. However, numerous difficulties, including contact resistance, depth control, and the effect of soil type, will need to be addressed to greatly improve the accuracy for salinity measurement.


   NOTES
TOP
 NOTES
ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS AND DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Contribution number 03-54-J of the Kansas Agricultural Experiment Station, Manhattan, KS.

Received for publication August 7, 2003.


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




This article has been cited by other articles:


Home page
 Soil Sci.Home page
A. Nadler
Comments on "Field Calibration of Water Content Reflectometers"
Soil Sci. Soc. Am. J., June 28, 2005; 69(4): 1356 - 1357.
[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 (2)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Zhang, N.
Right arrow Articles by Loughin, T. M.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Zhang, N.
Right arrow Articles by Loughin, T. M.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Zhang, N.
Right arrow Articles by Loughin, T. M.
Related Collections
Right arrow Water Content
Right arrow Other Geophysical Methods
Right arrow Soil Physics

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