A single-polarization, vertical wave propagation transmitted-vertical wave propagation received (VV), terrestrial imaging radar and time domain reflectometry (TDR) equipment were used to monitor the fluctuations of water content within two thin (15.24 centimeters) compacted clay test sections (each 30.48-m long, by 15.34-m wide). Radar imagery was obtained in 2013 on March 22, June 26, July 7 and July 10 using a Gamma Remote Sensing Portable Radar Interferometer version II (GAMMA Remote Sensing and Consulting AG, Bern, Switzerland). In situ observations were acquired hourly using TDR probes. The background, methodology, and comparison of the results of the remote sensing and in situ measurements are presented. Specifically, two remote sensing data reduction algorithms were considered. The water content results obtained from these algorithms were compared with the water content values derived from correlations with the dielectric permittivity values measured by the TDR probes. Key findings include: 1) there were differences between the values obtained from the data reduction methods, 2) there were differences between the values obtained from the remote sensing methods and the values obtained from the in situ method, and 3) the identification of additional avenues of research. The remotely sensed measurements were, on average, higher than the in situ measurements (the average volumetric water content values were 0.2 and 0.5 depending on the method, as obtained via the remotely sensed methods; the average volumetric water content value in the compacted native soil test sections was 0.22, as obtained via the in situ method).
The use of satellite- and or aerial-based active microwave remote sensing platforms is common in the fields of geoscience and remote sensing. In these fields, significant research has been conducted on the measurement of soil water content values for large geographical areas using active microwave remote sensing devices and technologies (Ulaby, 1974; Dubois et al., 1995; Wegmuller, 1997; and Wagner, 1998). Although these technologies have provided researchers with the ability to obtain data for a large spatial extent or a “swath”, the data are limited in terms of spatial and temporal resolution (approximately 20 m and 35 days, respectively). While such limitations are acceptable for broad-scale climatological, geological, or biological research, the limited spatial and temporal resolution may prevent the use of these advanced technologies for detailed qualitative and quantitative investigations by geotechnical engineers. To overcome these limitations, a single-polarization, portable, terrestrial, Ku-band, real aperture radar system was tested to obtain high spatial and temporal resolution measurements of volumetric soil water content and surface deformation. The single polarization included a vertical wave propagation transmitted-vertical wave propagation received (VV) configuration as intrumented in the Gamma Remote Sensing Portable Radar Interferometer version II (GPRI-II), which can provide spatial resolution for data at approximately one meter, surface deformation of one centimeter, and an acquisition rate on the order of one minute. These highly detailed acquisition patameters are sufficient for small-site engineering applications.
The specifications and previous use of the GPRI-II system to detect and quantify ground surface movements induced by slope failures and expansive clay materials are contained in Conte (2012), Coffman and Garner (2012), Conte and Coffman (2013), and Garner and Coffman (2014). Although the Ku-band GPRI-II radar was optimized to collect high resolution, sub-centimeter measurements of ground surface displacement (via the use of interferometric techniques), the purpose of this work was to evaluate its use to obtain values for the in situ volumetric water content of the soil. Utilizing the data collected from the GPRI-II device and data processing techniques developed by Wegmuller (1997) and Wagner (1998), values of volumetric soil water content can be obtained remotely and non-destructively. The use of the Ku-band does have a limited penetration depth of the incident radiation and therefore the depth for observations is limited. However, the data collection and processing techniques developed for this project are also applicable for other lower frequency terrestrial-based radar systems. Specifically, by using remote sensing, values of volumetric soil water content may be obtained without installing probes or collecting soil samples.
Time domain reflectometry (TDR) data were used to validate the post-processed volumetric soil water content data that were collected using the GPRI-II. Specifically, pointwise in situ TDR measurements were simultaneously collected at the same site that the GPRI-II measurements were collected; the University of Arkansas Cato Springs Research Center (UACSRC) was the geological site. TDR was chosen because it is a technique accepted by the geotechnical and geological engineering community for determining the volumetric water content for soils, it is capable of continuous data collection, it does not disturb the soil surface after the initial probe installation, and the variation in volumetric water content measurements, as obtained from data collected from the GPRI-II and TDR probes, both correspond to changes in the dielectric permittivity of the soil.
Two thin compacted clay test sections (30.48-m long, 15.34-m wide, and 0.15-m thick) were constructed adjacent to the UACSRC building at the UACSRC site. The roof of the building provided an elevated vantage point (incidence angle of approximately 80 degrees) for the GPRI-II. The two sections were constructed from a native sandy clay material. One of the sections was also amended with six percent, by dry weight, sodium bentonite (WyoBen Enviro-Plug No. 8) prior to compaction to increase the expansive behavior and the hydraulic retention. The construction methodology, soil placement window, and site layout are described in Coffman and Garner (2012) and Garner and Coffman (2014). The sections were compacted at water contents ranging from 18 to 25 percent gravimetric water content for the amended test sections and 17 to 24 percent gravimetric water content for the unamended test sections. Each test section was constructed on top of a 0.05-m thick freely draining sand base and a 1-m by 7-m control section was included in each of the pads. The control sections consisted of a native sandy clay within the bentonite amended pad and a bentonite amended sandy clay within the native sandy clay pad. Vegetation was not allowed to grow on the test sections because the test sections were constructed for engineering observation purposes. The UACSRC project site was concurrently used to evaluate the use of two remote sensing techniques (Radio and Light Detection and Ranging (RADAR and LIDAR) to monitor surface deformation caused by expansive clay behavior. The results of that research are documented in Garner and Coffman (2014).
In situ instrumentation (tensiometers, TDR probes, and thermocouples in the air and soil) was installed in each of the sections. Specifically, 14 Campbell Scientific CS-610 (Campbell Scientific, 2012) three lead TDR probes (7 probes per pad) were installed into the two thin compacted clay test sections to obtain volumetric water content measurement. The probes were excited and recorded (hourly) using a Campbell Scientific TDR-100 time domain reflectometer device, three Campbell Scientific SDMX-50 multiplexers, and a Campbell Scientific CR-10X datalogger. The placement of the in situ instrumentation and surrounding infrastructure, including the location of the GPRI-II observation point and the locations of the sensors within the test sections, are discussed in Coffman and Garner (2012) and Garner and Coffman (2014). A graphical schematic of the UACSRC project site is presented in Fig 1.
The value of dielectric permittivity for dry soil particles, as obtained from microwave frequency excitation, ranges from three to eight, whereas the dielectric permittivity value of water is 80 and the dielectric permittivity air is unity (Hanson and Peters, 2000; European Space Agency, 2014). Fundamentally, similar variations in soil dielectric permittivity, as caused by changes in the volumetric water content of the soil, can be measured either from the GPRI-II backscatter or from the TDR waveform, albeit at different frequencies. Because the dielectric permittivity value for water is high, the volume of water within the soil is the dominant influence on the overall dielectric permittivity of the porous media. Water is the constituent that enables the extraction of soil water content measurements from the radar backscatter data or from the TDR waveform data. The relationship between the complex dielectric permittivity and water content is frequency dependent and has been presented in Topp et al. (1980) and Njoku and Entekahbi (1994) for the frequencies utilized in TDR and radar testing, respectively.
Active Microwave Remote Sensing
The interaction of microwave waves with a soil surface, as employed in polarimetric and non-polarimetric imaging radars, is dependent on the dielectric properties of the reflecting surface, the surface texture of the soil surface, the polarization of incident and backscatter radiation, and the local angle of incidence at the soil surface. Specifically, the intensity of the returned electromagnetic (EM) energy is proportional to the volumetric water content of the soil (e.g., the intensity value for each of the pixels increases with increasing water content). Regions of standing water act as a specular “mirror” reflector and therefore do not return energy to the sensor. While previous researchers have used radar technology to analyze water content of the soil using ground-based systems (e.g., Ulaby, 1974; Sarabandi et al., 1994), the systems most commonly implemented in the geoscience fields are orbital-based synthetic aperture radars (e.g., Dubois et al., 1995; Wegmuller, 1997; Wagner, 1998). For ground- or orbital-based systems, the volumetric water content (θ) is measured by determining the influence of water within the soil (via dielectric permittivity) on the radar backscatter coefficient (σ).
Polarimetric methods are underpinned by electromagnetic scattering theory. The advantage of determining the volumetric water content by using polarimetric equipment is that the absolute water content of the soil can be computed from a single pass, as opposed to the repeat pass techniques used in non-polarimetric methods. Specifically, the water content of the soil is obtained by comparing the reflection received from two of the four different polarizations. These include the co-pol horizontal wave propagation transmitted–horizontal wave propagation received (HH) and vertical wave propagation transmitted–vertical wave propagation received (VV), or cross-pol horizontal wave propagation transmitted–vertical wave propagation received (HV) and vertical wave propagation transmitted–horizontal wave propagation received (VH). Common methods to determine the volumetric soil water content by employing polarimetric radar data include the Small Perturbation Method (SPM) introduced by Rice (1951), the Integral Equations Method (IEM) introduced by Fung et al. (1992), and the Advanced Integral Equations Method (AIEM) introduced by Chen et al. (2003). As a note, the polarimetric methods are included here for completeness, but were not used for this study because a single-polarization (VV) radar was used.
The non-polarimetric methods introduced by Wegmuller (1997) and Wagner (1998) are change detection algorithms in which the change in reflectivity of the soil (change in the backscatter coefficient and received power as per Eq. 1) is correlated to temporal changes in the values of water content and saturation, respectively. The primary advantage of these methods, over the polarimetric methods, is the assumption that the effects of the physical site parameters, or the backscatter intensity, do not change between scenes and therefore cancel out. The observed changes in backscatter intensity are only attributable to changes in the dielectric permittivity of the soil, as shown by Wagner (1998) and presented as Eq. 2 and Eq. 3. The assumption of constant parameters for surface roughness, surface autocorrelation length, and local incident angle (for a terrestrial radar reoccupying an observation point) allows for reduced data collection and data processing demands. The principle assumption of the Wagner (1998) method is that the variations in pixel backscatter intensity are only a function of water content and therefore the maximum and minimum pixel brightness values will correspond to the highest and lowest levels of volumetric water content. For pixel values that are between the aforementioned maxima and minima values, the volumetric water content is obtained by linear interpolation. Key limitations of the Wagner (1998) method include the assumption of unchanged local ground conditions between observations (neglection of temporal decorrelation) and the requirement that the soil surface was observed at the maximum and minimum saturation levels. Furthermore, a measurement of in situ porosity is required to obtain the absolute volumetric water content values using the Wagner (1998) method.
In Eqs. 1 through 3, σ is the radar cross-section (backscatter coefficient) of the target, Pr is the power received, λ is the wavelength, R is the slant range distance between the radar and the target, Pt is the transmitted power, Aeff is the effective antenna area, S is the soil saturation, σ(80°,t) is the backscatter coefficient at a given time (t) and at an incident angle of 80 degrees, σdry(80°,tdry) is the backscatter coefficient at a corresponding time (tdry) when the soil is completely dry, and σ wet(80°,twet) is the backscatter coefficient at a corresponding time (twet) when the soil is completely wet (completely saturated). Equation 1 was modified from Ulaby et al. (1986); Eqs. 2 and 3 were modified from Wagner (1998):
Wegmuller (1997) developed an empirical correlation between changes in the backscattered intensity values (logarithmic) and changes in the volumetric water content values: where θ is the volumetric water content, Δσ(80°,t) is the difference in corrected backscatter coefficient as corrected for an incident angle of 80 degrees, σ2(80°,t) and σ1(80°,t) are the backscatter coefficients of temporally separated returns. Values of volumetric water content of the soil are then calculated for pixels in each scene by considering a reference image with values of known water content for corresponding pixels (Wegmuller, 1997):
In Eq. 6, θ0 is the known volumetric water content in the reference image, σ0(80°,t0) is the reference backscatter coefficient corresponding to the θ0 condition.
Wegmuller (1997) further extended the empirical relationship based on satellite observations at test sites in the United States and Europe : where i is the intercept value of the empirical backscattering (dB)–volumetric soil water content (%) relationship. The Wegmuller (1997) method allows for the determination of the absolute volumetric water content values within a reference image. The disadvantage of the Wegmuller (1997) method is that the method requires the collection of in situ volumetric soil water content information and temporally registered radar imagery. However, to avoid the need to collect in situ instrumentation, Wegmuller (1997) proposed using observation data that were captured during periods of below freezing conditions.
Time Domain Reflectometry
TDR systems operate by sending individual stepped voltage increases, with a fast rise time (<300 ps according to Evett, 2003), through the center conductor of a coaxial cable into each probe (typically three wire probes) located within the soil. The variation in travel time is a function of the wave propagation speed in the unshielded portion of the probe, because the physical dimensions and material properties of the cabling and probes are known and are unchanged.
Although the dielectric permittivity for soil is a complex number, the real component of the complex number does not contribute to the electric loss term over the range of frequencies employed by TDR systems (Nemmers, 1998). Therefore, only the imaginary component is utilized. By combining Eq. 9 and 10 to form Eq. 11, the dielectric permittivity of the soil can be obtained (Evett, 2003): where vp is the velocity factor setting, v is the velocity of propagation through the shielded cable, c0 is the speed of light in a vacuum, ε is the dielectric permittivity of the cable shielding, μ is the magnetic permeability of the dielectric material (equal to unity in free space), L is the length of the unshielded probe lead, tt is the travel time within the unshielded probe lead, εa (or Ka) is the apparent dielectric permittivity of the soil, and La is the apparent length of the unshielded probe lead.
Specifically, the dielectric permittivity is obtained from Eq. 11 and the Tangent Method (Evett, 2003 and Nemmers, 1998). Multiple relationships have been proposed to utilize the Ka value obtained from the waveform to determine the soil volumetric water content (θ); the most commonly employed relationship is Topp et al. (1980):
The Topp et al. (1980) equation is considered to provide acceptable results for most soil types, however it is possible to generate a specific relationship by conducting tests on compacted samples with known volumetric water contents, as presented in Take et al. (2007).
Methods and Procedures
To compare the results obtained from the GPRI-II and the TDR techniques, standardized data acquisition and processing procedures were developed and followed. These procedures included: radar data collection, radar data processing, TDR apparent dielectric permittivity–volumetric water content relationship development, TDR equipment installation, TDR data collection, and TDR data processing. Each procedure is discussed in this section.
Radar Data Collection
Radar images of the test sections at UACSRC were captured from the roof of UACSRC, using the GPRI-II. To ensure the observation area and observation geometry were identical for the respective observations, the GPRI-II was repositioned over a survey monument located on the roof of UACSRC for subsequent scans. Scans with a 95-degree field of view were acquired at a rate of 5 degrees per second using a 250 microsecond chirp from the GPRI-II. Specifically, each scan was acquired from −15 degrees to 80 degrees with the basis of rotation being the power pole located approximately 700 meters to the southeast of the project site, located at an angle of rotation of 37 degrees (as observed using a telescopic optical sight affixed to the side of the GPRI-II instrument). Radar imagery was collected during periods of high and low volumetric water content for use in the Wegmuller (1997) and Wagner (1998) equations. Radar imagery collected during late June 2013, after allowing several weeks for the soil surface to desiccate with daily high temperatures exceeding 32° Celsius and limited precipitation (less than 2.5 cm of rainfall), was used as the Wegmuller (1997) reference imagery. The late June 2013 imagery was also used to obtain “minimum” pixel brightness values for use in Wagner (1998) data processing. During the same period, the volumetric water content of the soil was obtained using manually collected soil samples and procedures according to ASTM D2216 (2012) and ASTM D854 (2012). Radar imagery was also collected during and after a period of significant precipitation (greater than 2.5 cm of rainfall in a 24 hour period) to provide “maximum” pixel brightness values for the Wagner (1998) data processing.
Radar Data Processing
Radar data were processed in both Linux and Windows software environments. Pre-processing was conducted in Linux using commercially available Gamma Remote Sensing software (Gamma Remote Sensing, 2010) to convert the raw radar data into single look complex data (SLC, 64-bit complex data comprised or two strings of 32-bit floating point data for the real and imaginary components) and subsequently multi-look intensity images (MLI, 32-bit floating point data). The MLI files were then exported to Windows for further processing using MATLAB (Mathworks, 2012).
Using MATLAB, the individual MLI files for each day, which typically included seven scans collected during a 10 minute period, were temporally averaged to increase the signal-to-noise ratio. The averaged MLI images were transferred back to Linux and processed, following the Wegmuller (1997) procedure, using a pre-programmed function (soil_moisture.c) in the Gamma Remote Sensing software. A reference map of the same size as the MLI images was also created in MATLAB. The reference map contained the in situ water content data that was acquired using the aforementioned sampling techniques and the reference image was also incorporated into the soil_moisture.c software. The averaged MLI images were also processed in MATLAB following the Wagner (1998) procedure. After processing, all of the images were visualized and compared using MATLAB.
Utilizing the Wagner (1998) method, the minimum and maximum backscatter amplitude, or intensity, for each individual pixel were determined by comparing each of the individual averaged images. The minimum and maximum intensity corresponded to the lowest and highest volumetric soil water contents, respectively, and these values were used to create a reference maximum (wet) and minimum (dry) intensity image. For each pixel, a soil saturation value was calculated using Eq. 3 rendering a soil saturation image by referencing the wet and dry images. A volumetric water content image was then obtained by multiplying the pixel values in the soil saturation image by the porosity of the soil, which was 0.37. The description of the porosity measurement, as obtained from the TDR calibration, is described in the next section. A 3×3 pixel moving average was also applied to the processed data to reduce the variance in the image. All individual dates were then processed sequentially using the same technique. Because the values obtained from the Wagner (1998) method represent saturation, all of the pixel values were multiplied by the porosity of the soil to obtain volumetric water content values.
The GPRI-II platform has significant advantages, in terms of simplicity of processing, when using the Wagner (1998) method; the method was originally developed for use with the European Space Agencies (ESA) European Remote Sensing 1 (ERS-1) satellite platform. Specifically, the pre-processing to correct the backscatter coefficient for variations between multiple satellite passes, including ascending/descending orbits, different incident angles, and amalgamation of data from different antennas, as mentioned in Wagner (1998), is avoided for the GPRI-II system because a constant angle of incidence of 80 degrees, which was maintained for all scans. Additionally, information was only compared for images collected using the same antenna (e.g., data obtained from the lower antenna was compared to data obtained from the lower antenna for subsequent observations). Therefore, the only correction applied to the backscatter coefficient data prior to the computation of water content was the transformation of the respective pixel intensity values from a linear scale to a decibel scale (as shown in Eq. 2).
TDR Apparent Dielectric Permittivity (Ka) – Volumetric Water Content (θ) Relationship
A relationship between the TDR obtained apparent dielectric permittivity and volumetric water content was created for the soil used within the test sections. Soil samples were compacted following the procedures outlined in American Society of Testing and Materials (ASTM) standard D698 (ASTM D698, 2012) with the following deviations from the standard: mold size, compaction energy, and number of lifts. The samples were compacted into a 0.342-m long by 0.152-m diameter aluminum mold (6.23×10−3 m3 mold volume instead of the 9.43×10−4 m3 mold volume) to accommodate the 0.3-m long CS-610 probes. Each sample was compacted in nine lifts (3.81 cm thick) using 25 blows per lift, resulting in a compaction effort of 75 percent of the standard Proctor effort (450 kN-m/m3) to resemble the field compaction effort. Twelve samples, including eight calibration samples and four validation samples, were compacted at volumetric water contents between 12 and 35 percent, which were obtained from the phase diagram of each sample. After compaction, a TDR probe was installed into each of the compacted samples and TDR waveforms were collected. The apparent dielectric permittivity was obtained from the TDR waveform uisng Eq. 11 and the Tangent Method as presented in Fig. 2.
TDR Installation and Field Data Collection
The TDR probes were hydraulically pushed horizontally into the side of a manually excavated trench using a custom jig (at mid-depth of each of the sections at each location). After installation, the performance of each TDR probe was verified to ensure that each probe was properly installed and that the probe leads were not in contact with one another. The datalogger was then programmed to remotely capture and store the waveform data from each probeevery hour. The raw data file was exported from the datalogger at regular weekly intervals as a comma delineated ASCII text file and the data were processed in MATLAB. The volumetric water content was obtained for each location at each time using Eq. 11, the Tangent Method, and the soil specific relationship.
Volumetric Water Content: Single-Polarimetic Remote Sensing and TDR Methods
Examples of volumetric water content values from the GPRI-II and processed using the Wagner (1998) and Wegmuller (1997) methods, are presented in Figs. 3, 4, and 5. The first set of images (July 7 and July 10) was obtained with the Wagner (1998) method. The second set of images (also from July 7 and July 10) was obtained using the Wegmuller (1997) method. In both cases, the initial image (July 7) was taken during and after a heavy rainfall event approximately 2.5 cm of precipitation (NOAA, 2012), while the subsequent images (July 10) were taken after three days with air temperatures exceeding 32° C (NOAA, 2012). Desiccation of the soil body was expected and observed between the first and second set of radar observations. The third set of images on March 22 was obtained prior to significant rainfall or desiccation and represents partially saturated conditions.
Using the Wagner (1998) method, average values of degree of saturation for dates March 22, July 7, and July 10 for the amended section were 0.56, 0.72, and 0.58, respectively; for the unamended section the saturations were 0.36, 0.52, and 0.41, respectively. These saturation values corresponded to average volumetric water content values of 0.21, 0.21, 0.27 and 0.13, 0.15, 0.20 for the amended and unamended sections, respectively. By contrast, average volumetric water content values, using the method proposed by Wegmuller (1997) produced 0.31, 0.15, 0.27 for the amended section and 0.37, 0.19, 0.28 for the unamended section.
Several factors are responsible for deviations between the values obtained from the two remote sensing data reduction methods. First, the Wagner (1998) method is a measurement of soil saturation and volumetric water content measurements were obtained by multiplying the saturation by a constant porosity value. Although a calibrated constant value of porosity was used, porosity values typically vary. Additional measured values of in situ porosity, at different locations on the surface, should have been used to calibrate the model to account for spatial variability of porosity. Second, empirical coefficients that match reflectance with volumetric water content are used within the Wegmuller (1997) method. Although the default, empirically derived coefficients in the Gamma Remote Sensing software program were developed using sites in the United States, it is not certain if these coefficients were appropriate for the soil type used in this study. Site specific values of radar cross-section and the corresponding measured values of volumetric water at different locations on the surface should have been used to calibrate the model. Additionally, the data were not corrected for contribution from the thermal noise, although the effects of this contribution is not known and may not be able to be characterized. For the site considered, use of the Wagner (1998) method is recommended because of the empirical constants within the Wegmuller (1997) method needing to be calibrated for: 1) satellite-based systems operating at lower frequencies (C-Band), 2) significantly different viewing geometries, and 3) specific sites in the U.S. and Europe.
As shown in Fig. 6, average volumetric water content values of 0.22 (amended pad) and 0.23 (unamended pad) were obtained on July 10 using the TDR technique. The TDR-obtained volumetric water content values are within 0.1 m3·m−3 of the remotely sensed measurements. Discrepancies between the remotely sensed and in situ measurements are attributed to aforementioned factors and the differences in sampling depths (0 to 1.7 cm for the GPRI-II and 8 to 12 cm for the TDR data). Specifically, these discrepancies are attributed to reduced infiltration depth and high surface evaporation rates. However, by utilizing in situ measurements of the surface soil properties (volumetric water content or porosity), the remote sensing measurements may be calibrated to match the in situ measurements.
Remote sensing, when coupled or uncoupled with pointwise in situ measurements, offers many potential benefits to both geotechnical researchers and practitioners. The use of ground- based systems (such as the GPRI-II) can eliminate or alleviate many of the disadvantages associated with satellite- or aerial-based technologies. Specifically, the use of ground-based systems can allow for the collection of data covering a large spatial extent, with high spatial and temporal resolution, and with a relatively low marginal cost per observation, as compared to costs associated with satellite imagery, aerial imagery, or additional pointwise TDR measurement equipment that would be required for the same spatial resolution.
A correlation between the volumetric water content values obtained from the in situ measurements and the two remote sensing, single-polarization, data reduction methods was observed. Specifically, the average volumetric water contents obtained using the different methods were within ten percent of each other. However, based on the observed differences between the two remote sensing methods and between the remote sensing and in situ methods, additional research is required to: 1) couple the in situ measurements, 2) develop model calibrations, and 3) incorporate vegetation and temperature corrections.