- Copyright © by the Soil Science Society of America, Inc.
We used a recent ground-penetrating radar (GPR) methodology, early-time amplitude analysis, with the goal of monitoring changes in soil water content (SWC) in response to irrigation in clayey soils. We hypothesized that early-time analysis could be used to monitor changes in SWC in clay-rich soil where ground wave and reflection-based GPR methods traditionally fail. An overnight irrigation experiment was performed in a 20- by 14-m section of natural grassland at the Samford Ecological Research Facility in southeastern Queensland, Australia. Both GPR reflection surveys and ground wave velocity analysis were ineffective at the site due to the signal attenuation associated with the clay-rich soil. We collected daily GPR and time-domain reflectometry (TDR) data sets during a 5-d period in August 2014, with soil samples collected for gravimetric analysis at the conclusion of data collection. The GPR data display a clear response of the early-time signal amplitude to changes in SWC. The GPR data sets exhibit a strong correlation with SWC, as measured by TDR and gravimetric analysis of soil cores, which is consistent with the dependence of GPR early-time amplitude on relative permittivity. The results suggest that the early-time method can be used to obtain spatially distributed information on subsurface moisture content in clay-rich soils.
- AEA, average envelope amplitude
- GPR, ground-penetrating radar
- GWC, gravimetric water content
- SWC, soil water content
- TDR, time-domain reflectometry
- WARR, wide-angle reflection and refraction
Understanding the spatial distribution of soil moisture is critical to climate modeling, agricultural water management, and research on biological and chemical processes in the vadose zone. Recent reviews by Vereecken et al. (2008) and Robinson et al. (2008) explored the importance of soil moisture measurements in detail. However, traditional methods of estimating soil moisture have significant limitations. Methods such as time-domain reflectometry (TDR) and soil sampling are both invasive methods that can interfere with the processes being studied. Additionally, both are point measurements that cannot easily provide dense, spatially expansive estimates of soil moisture. On the other hand, large-scale estimates of soil water content (SWC) such as those collected by cosmic ray probe (Franz et al., 2013) or satellite (Kerr et al., 2010) do not easily allow analysis of field-scale processes.
Ground-penetrating radar (GPR) has long been established as a tool for measuring SWC variation at the field scale (Huisman et al., 2003) using reflection data (van Overmeeren et al., 1997; Lunt et al., 2005), ground wave data (Galagedara et al., 2003; Grote et al., 2003; Huisman et al., 2001), and borehole transillumination data (Cassiani et al., 2004; Rucker, 2011). It provides a noninvasive way to efficiently collect data across large areas, which can be used to estimate the SWC in the vadose zone. However, both reflection and ground wave analyses have limitations. Traditional reflection analysis requires detection of a reflector at a known depth to calculate the two-way travel time, although in certain instances this requirement can be circumvented (Gerhards et al., 2008). The ground wave method requires the operator to find an antenna offset where the direct ground wave is not interfered with by the air wave or any reflections, regardless of changes in the SWC or location at the field site. Both methodologies are hampered by highly conductive soils, such as those with a high clay content, which attenuate the signal.
The recently proposed early-time methodology (Pettinelli et al., 2007, 2014) provides an alternative way to estimate SWC variation at the field scale using GPR. The methodology includes an investigation into changes in the amplitude of the combined air and ground waves to estimate changes in the electromagnetic properties of the near surface. The so-called average envelope amplitude (AEA) statistic allows researchers to quantitatively analyze these changes in amplitude. It is also sometimes referred to as the “average envelope” (Pettinelli et al., 2014) or “envelope amplitude” (Di Matteo et al., 2013; Ferrara et al., 2013). The early-time method can be applied when there is superposition of the air and ground waves. Additionally, because the method does not rely on reflections, no near-surface reflector is required. These benefits mean that the early-time method may provide a solution to clay-rich field sites, where GPR is not commonly utilized due to the unavoidable effects of excessive signal attenuation.
In this study, we used early-time GPR to monitor changes in SWC using time-lapse measurements before and after an irrigation event at a clay-rich field site. We hope to show that the early-time methodology can be applied in a highly attenuating environment at the field scale. This would provide a way to use GPR in clay-rich field sites, where it is traditionally considered a nonviable technology.
Overview of Ground-Penetrating Radar
A GPR measurement typically involves two antennae: a transmitting antenna emits a short-pulsed electromagnetic signal, and transmitted and reflected energy from the subsurface are then measured by the receiving antenna. The signal travels through the air and subsurface at different velocities, depending on the relative dielectric permittivity (εr) of the materials, which is the ratio of the absolute permittivity of the material to the permittivity of a vacuum. Ground-penetrating radar is a valuable tool for soil moisture studies because the water content of a soil has a significant effect on its εr. The εr of fresh water is around 80, while for air and for mineral soil grains it is 1 and around 4, respectively. Small changes in water content can thus lead to significant variations in εr (Van Dam and Schlager, 2000). Additionally, water can have a wide range of conductivities depending on its salt content, which affects signal attenuation (Davis and Annan, 1989). Therefore, a change in water content will significantly affect the electromagnetic properties of the subsurface.
The εr is related to the wave velocity bywhere v is the wave velocity, c0 is the speed of light, μr is the relative magnetic permeability of the material, μ = μrμ0, where μ0 is the magnetic permeability in a vacuum, μr is 1 for most common Earth soils, and P is a loss factor:where σ is the conductivity of the material, ω is the angular frequency, and ε is represented by ε = εrε0, where ε0 is the permittivity of a vacuum. Ground wave and reflection-based methods use Eq.  for estimation of εr by measuring the travel time of the GPR signal that travels directly from the transmitter to the receiver through the shallow subsurface (Grote et al., 2003) and the signal that reflects off subsurface interfaces where there is a εr contrast, respectively. The permittivity value can then be converted into an estimate of the SWC by use of a pedotransfer function (Van Dam, 2014).
The usefulness of traditional GPR methods is limited in high-conductivity material, such as clay-rich soils or materials with saline pore fluids. The GPR attenuation coefficient (α) is related to σ and ε by
In highly conductive soils, the GPR signal will often attenuate to very low amplitude before it reaches the receiver.
GPR Early-Time Method
Early-time amplitude analysis, a methodology recently proposed by Pettinelli et al. (2007), provides an alternate way to extract information on changes in SWC from common-offset GPR surveys. A benefit of the method is that the antenna offset does not need to be large. This means that the method works with commercial bistatic common-offset antenna equipment with inseparable antennae. The method allows the collection of amplitude information sensitive to permittivity and/or conductivity variations at shallow depth. In early-time analysis, the earliest part of the signal in a GPR reflection survey—the combination of the ground and air waves—is analyzed without regard to whether they are separated in time, an often restrictive requirement for the ground wave method (Di Matteo et al., 2013). By analyzing changes in the amplitude attributes of the early-time signal from both modeled and field data, Di Matteo et al. (2013) were able to map near-surface variations in εr. They suggested that the method is highly capable of mapping SWC variations when field calibration can be performed using discrete point measurements.
The early-time method has been used in a number of recent studies to determine its efficacy. Pettinelli et al. (2014) built a polyvinyl chloride tank and filled it with a layer of gravel at the bottom and a layer of river sand above. Water pipes in the gravel layer permitted raising the water level incrementally, allowing them to monitor changes in the early-time signal as the water approached the surface. They concluded that (i) the results of their GPR measurements were in line with previous numerical modeling done by Pettinelli et al. (2007) and Di Matteo et al. (2013), (ii) the average envelope amplitude of the early-time signal and dielectric constant (i.e., SWC) are inversely correlated, and (iii) the thickness of the portion of the subsurface affecting the early-time signal is on the same order as the GPR signal wavelength in the materials they tested.
The amplitudes of the air wave (Aair-wave) and ground wave (Aground-wave) are determined by (Di Matteo et al., 2013)wherewhere μ0 is the magnetic permeability, ε0 is the dielectric permittivity of a vacuum, σ is the soil electrical conductivity, A0 is the amplitude of the ground wave in a vacuum, and S is the antenna separation. The exponential term in Eq.  accounts for the evanescent portion of the ground wave that propagates, decaying, above the surface (Annan, 1973; Di Matteo et al., 2013). One drawback to the early-time methodology is that it is difficult to determine to what extent changes in the amplitude of the early-time signal are caused by changes in permittivity vs. changes in conductivity, although the effect of each property on the early-time signal has been analyzed independently in a controlled setting (Comite et al., 2016). This is because the relative amplitude of the direct signal is affected both by the conductivity of the subsurface, as shown in Eq. , and by the permittivity of the subsurface, as seen in Eq. [4–6]. However, it has been suggested that permittivity plays a larger role in affecting the early-time amplitude, and it has been shown that changes in conductivity do not affect the goodness of correlation between early-time amplitude and changes in permittivity (Di Matteo et al., 2013). Comite et al. (2016), however, suggested that envelope amplitude as a statistic for relating the GPR signal to relative permittivity may be negatively impacted by high-conductivity materials. They proposed an alternate statistic, which they refer to as the “carrier frequency amplitude,” as a potentially more robust alternative.
The Samford Ecological Research Facility (SERF) is a 51-ha property in Queensland, Australia (Fig. 1). The property is used for a wide variety of ecological, engineering, environmental, and educational programs, with a strong focus on understanding greenhouse gas emissions (Karan et al., 2016). In support of these goals, a large portion of the property is maintained as a typical Australian subtropical grassland, with grain grown and harvested on an annual basis. The soil at the site is clay rich, with up to 40% clay content in some horizons. This means that traditional transmission–reflection GPR methods to characterize the site are likely to be ineffective. The combination of a need for spatial soil moisture data and a clay-rich soil was a strong motivation for testing the early-time method at this site.
This study focused on an area in the northwest corner of SERF’s grassland, where there is a relatively flat area near a stream, which provided water for an irrigation experiment. We selected a 20- by 14-m portion of this area for irrigation and geophysical measurements.
Upon selecting the field site, we performed a series of geophysical measurements to improve our understanding of the subsurface. We collected a GPR wide-angle reflection and refraction (WARR) survey (Fig. 2). We also collected GPR common-offset measurements (Fig. 3) over an irrigated patch using a PulseEKKO system equipped with a SmartCart to determine the optimal GPR settings for the site. The GPR data were collected using a bistatic system with 200-MHz antennae at 1-m separation, with a pulser voltage of 400 V. The SmartCart was equipped with an odometer wheel to accurately position measurements every 5 cm as the GPR moved along the transect. For both WARR and common-offset measurements we used a 0.2-ns sampling interval.
Figure 2 shows the results of the WARR survey collected at the field site on 11 August, prior to the main experiment. There is a clear air wave in both the raw and gained data, while the ground wave only appears after dewow and gain were applied—we used a manual linear gain increasing from 2.5 dB at time zero to 15 dB at 30 ns—and still attenuates to the point of being difficult to pick by the maximum offset. No reflections are visible in the data down to 30 ns.
Figure 3 shows the results of the common-offset survey collected on 22 July. The direct signal at the start and end of the line is a combination of the overlapping air and direct ground waves. A 1- by 1-m segment in the center of the line was wetted with 120 L of water, causing the direct ground wave to slow down and separate from the air wave. However, this separation cannot be maintained across a wide range of SWC values, and the first break is difficult to identify. No reflections are evident down to 30 ns.
Field Instrumentation and Data Collection
One week prior to data collection, 3 cm of rain fell at the field site. We created a tarp and placed it over the area of our geophysical investigation, but it is likely that some water flowed under the tarp and infiltrated at the edges of our 20- by 14-m grid.
The timeline of measurements and irrigation can be seen in Table 1. The grid was irrigated on the night of Day 1 using a garden sprinkler. The sprinkler was set along the edge of the grid, spraying water into the grid in a 180° arc, to a distance of approximately 8 m. The sprinkler applied water at a rate of approximately 2.5 cm h−1 at a 3-m radius from the sprinkler, decreasing in rate both closer to and farther away from the sprinkler. The sprinkler ran for approximately 8 h, starting at 4:33 PM. In addition to the sprinkler irrigation, three 1- by 1-m box infiltrometers were placed in the grid (Fig. 1) and filled with 100 L of water at 4:00 PM on Day 1. The water was allowed to completely infiltrate before the infiltrometers were removed from the grid. This provided an alternate type of wetted area: a small area of high water content surrounded by dry soil, as opposed to the large swath of land irrigated by the sprinkler.
For the main experiment, we collected GPR measurements at 10 AM daily from Day 1 until Day 5. Each data set consisted of 15 lines, 20 m long, at 1-m spacing (Fig. 1). Data collection took approximately 1 h. We used the same SmartCart GPR setup to collect these data sets as for the 2 July common-offset measurements, except with a 0.5-m antenna separation instead of 1 m to ensure that the air and ground waves would overlap at all levels of saturation.
We collected TDR, in conjunction with the GPR data sets, at 10 AM using a HydroSense CS620 soil water measurement system with a two-rod probe with 12-cm rods. The system determines the period average output from the probe, τ, in milliseconds. This varies based on the εr of the subsurface and is therefore affected by changes in the SWC. A total of 101 TDR points were taken each day, focused around the locations selected for collecting soil samples. These locations were selected to provide information on the widest possible range of wetting conditions: completely dry, within the box infiltrometers, and at 3 and 6 m from the sprinkler. The TDR data collection took approximately 90 min.
At the end of the final day of measurements, soil samples were collected from the 5- to 10- and 15- to 20-cm depths at 24 locations throughout the field. We selected these depths to avoid the significant root zone in the grassland soil, as well as to mitigate the effects of evapotranspiration during the course of the day. The soil samples were collected using a standard size soil sampling ring and promptly bagged. They were refrigerated on return to the laboratory, within 3 h of their collection. A standard oven-drying methodology was applied to calculate the gravimetric water content (GWC) of the 48 soil cores (Klute, 1965).
The GPR surveys contained approximately 400 measurements along each line. Before processing, a moving average was applied to the data in blocks of seven. Each averaged data point thus represents a portion of the transect approximately 0.35 m in length and overlaps 30 cm of the data points before and after it. The data were averaged to remove small-scale variation (outliers) that probably resulted from challenges with ground coupling of the antennae in the thick grassland.
A Hilbert transform was performed on the first positive half cycle (Fig. 4) of each of the averaged measurements, according to Di Matteo et al. (2013), using MATLAB’s built-in hilbert() function. We passed each trace’s averaged GPR data directly into the hilbert() function. The Hilbert transform is described bywhere is the Hilbert transform of the function x(s), and the integral is the Cauchy principle value integral. The Hilbert transform provides the imaginary part of the complex GPR trace. The acquired GPR signals represent the real portion of the trace. The Hilbert transform allows calculation of the envelope of the trace (Fig. 4), also known as the instantaneous amplitude, by taking the absolute value of the transformed GPR signal (Taner et al., 1979; Ferrara et al., 2013).
A custom function was used to extract the first positive half cycle of each measurement, which was shown by Di Matteo et al. (2013) to provide the greatest signal/noise ratio of any portion of the GPR early-time signal. We took the absolute value of the Hilbert transform of the measurements and the integral of the result divided by the unit length of the integral. The resulting value is the AEA, which was then inverted to be consistent with the work of Di Matteo et al. (2013) and Pettinelli et al. (2014). The AEA−1 data were then compared from day to day and with the soil core and TDR data sets, with an increase in AEA−1 expected to correspond to an increase in GWC. The AEA−1 data are in unitless “amplitude units,” meaning that the GPR data show a relative change in GWC rather than an absolute GWC value (Pettinelli et al., 2007, 2014; Di Matteo et al., 2013). Hislop (2015) proposed an alternate methodology that may estimate an absolute value of GWC from the early-time signal.
We derived a site-specific empirical relationship between our TDR measurements of τ and our GWC values based on the 5- to 10-cm soil cores. This equation was used to convert all TDR measurements of τ into GWC values (in %). The TDR values used for the calibration ranged from 11 to 22% GWC, which encompasses most of the saturation levels observed during the study.
Results from Time-Domain Reflectometry
The results from TDR data collection are shown in Fig. 5. The background TDR data set collected on Day 1 shows a mean and median GWC of 13% within the grid. A few areas are around 15% GWC, probably due to the effects of the storm that took place the prior week. The first data set after irrigation, on Day 2, shows an increase in the GWC in the wetted areas, both the area irrigated by sprinkler and the area irrigated by the box infiltrometer at grid coordinate (9,17). The area irrigated by the sprinkler ranged from 13 to 20% GWC, peaking between 2 and 3 m from the sprinkler. The wetted area at grid coordinate (9,17) measured 19% GWC. The data set collected on Day 3 is very similar to the Day 2 data set. On Day 4, the GWC values began to decrease, with the area irrigated by sprinkler ranging from 12 to 18%, but with much more of the area measuring in the 16% range than on Days 2 or 3. On Day 5, the area near the sprinkler increased to around 23% GWC.
Early-Time GPR Analysis and Soil Sampling
Figure 6 shows the results of the early-time GPR analysis. The plots show the AEA−1 of the first positive half cycle of the traces derived from averaging blocks of seven traces. The Day 1 data set, the background measurement, averaged an AEA−1 value of 5.91 × 10−5. There were a few higher AEA−1 areas around the edge due to rainwater leaking under our tarp during the previous week’s rainstorm. On Day 2, the first data set collected post-irrigation, the values in the unirrigated region were mostly unchanged from Day 1. In the area of the irrigation, the AEA−1 values jumped to the range of 6.6 × 10−5 to 1.26 × 10−4. The AEA−1 values remained elevated throughout the remainder of the study period, as infiltration occurred slowly in the clay-rich soil. However, the average AEA−1 value of the field site on Day 2 was 7.10 × 10−5, which decreased to 6.52 × 10−5 by Day 5.
The results from gravimetric analysis of the soil cores collected at the 5- to 10- and 15- to 20-cm depths can be seen in Fig. 7 compared with our GPR AEA−1 data. The 5- to 10-cm samples ranged from 8 to 22% GWC, and the 15- to 20-cm samples ranged from 9 to 19% GWC. The TDR output τ, which represents travel time of the signal along the TDR rods, was converted to GWC by correlating the data with the 24 5- to 10-cm-depth soil core measurements collected on Day 5 of the experiment. The obtained relationship is linear:and has an R2 value of 0.85. There is no evidence in the data to suggest that a nonlinear equation would be more appropriate within the range of GWC used to obtain Eq. . Because we do not use the equation to extrapolate to higher or lower GWC values, a linear relationship is appropriate.
Discussion of Background Measurements
Our initial measurements at the field site, taken a month before our experiment, demonstrated the ineffectiveness of traditional methods of using GPR to measure the SWC in clay-rich soils. The WARR survey (Fig. 2) showed that the ground wave method cannot be used to determine SWC at this site due to the significant attenuation caused by the high clay content of the soil. The ground and air waves do not separate until antenna offsets where the ground wave has attenuated to the point that it is difficult to accurately pick (Fig. 2b). These results suggest that during wet soil conditions with higher signal attenuation, use of the ground wave method would have been even more challenging. No reflections are visible beneath the direct signal, rendering the reflection method ineffective as well. This is further evidenced by the common-offset data collected at the site prior to the irrigation experiment (Fig. 3). In contrast, the early-time method appears to be a viable GPR methodology to map the spatial variability in shallow SWC at this site.
Discussion of Early-Time Method
Cross-plots of the GPR AEA−1 data (Fig. 7) with independently obtained soil core GWC values show the expected relationship (Fig. 7a and 7b), supporting the argument that the early-time method provides a viable way to use GPR in clay soils to map GWC variations. In the irrigated areas, a significant jump in AEA−1 is observed, in line with the results of previous researchers (Pettinelli et al., 2014; Ferrara et al., 2013). We also see a decrease in the AEA−1 values by the end of the experiment, probably indicative of a small amount of evapotranspiration and infiltration of water to below the GPR region of influence.
Unfortunately, the early-time methodology has not been demonstrated to distinguish changes in conductivity from changes in dielectric permittivity in a field setting. The irrigated water may have had a different electrical conductivity than the pre-irrigation groundwater, and the increase in water content would also cause an increase in the bulk relative permittivity of the unsaturated zone. Electrical resistivity monitoring in conjunction with GPR could potentially help isolate changes due to dielectric permittivity. Despite this limitation, the good correlations in Fig. 7 suggest that GPR AEA−1 values could be converted to GWC values using a calibration equation.
The TDR data show an increase in GWC in the sprinkler-irrigated region from Day 1 to Day 2. These data confirm the trend seen in the GPR data: increasing subsurface water content corresponding with increasing AEA−1 (Fig. 7c). The TDR data correlate better with the GPR data than do the soil core data. This is potentially because the TDR measured the top 12 cm of the subsurface, whereas the soil cores were taken from discontinuous, 5-cm sections of the subsurface, none of which began at the surface. However, the fact that there are many more TDR measurements (101 per day) than soil core measurements (24 collected in 1 d) probably improved the correlation between the GPR and TDR data as well. The better correlation between GPR AEA−1 and TDR data exists even if we use the volumetric water content estimate provided from the probe, and thus does not directly result from our linear fitting of TDR to soil GWC values.
The soil sample data also confirm the viability of the early-time method in this environment, as well as giving some information toward the depth of investigation of the method. Where the soil core GWC increased, we saw an increase in the AEA−1 of the GPR signal. This correlation is stronger with the 5- to 10-cm soil samples (Fig. 7a) than with the 15- to 20-cm soil samples (Fig. 7b), which suggests that the early-time signal of our antennae in this soil is primarily sensitive to the top 15 cm (approximately 1/8 of the GPR wavelength) of the subsurface (although there is still a positive, weaker correlation between the GPR and 15- to 20-cm soil sample data).
This irrigation experiment demonstrates the viability of the early-time GPR method for estimating the spatial variability in SWC in clay soils at the field scale, where other GPR methods fail. The WARR analysis and common-offset ground wave analysis were unable to provide a reasonable method to measure the SWC at our field site. Irrigated areas of the field site corresponded with increases in GPR AEA−1. What percentage of the change in AEA−1 was due to changes in permittivity vs. conductivity is unclear. However, our data show a strong correlation between AEA−1 and independent GWC measurements from both TDR and gravimetric soil sample analysis. The GPR data correlate best with the TDR data and the 5- to 10-cm soil core data. This suggests that our GPR measurements are most sensitive to the top 15 cm of the subsurface. Continuous soil cores from a larger range of the subsurface would improve our understanding of the depth of investigation of GPR at this site. The early-time method opens up new avenues of research using GPR in clay-rich soils and can benefit from further laboratory and field investigation.
The National Science Foundation and Australian Academy of Science funded this research through an East Asia and Pacific Summer Institutes for US Graduate Students grant. The Institute for Future Environments at the Queensland University of Technology (QUT) as well as at the Samford Ecological Research Facility supported this research by providing access to equipment and facilities. Marcus Yates, David Rowlings, and Nicholas Josephs provided invaluable assistance with the field work. Remke Van Dam acknowledges support from Peter Grace of the Healthy Ecosystems and Environmental Monitoring group at QUT. Jason Muhlbauer, PhD candidate at University of Tennessee, Knoxville, assisted with the creation of the figures. Finally, we would like to thank our peer reviewers, including Dale Rucker, David Nobes, and an anonymous reviewer, for their constructive feedback that greatly improved our paper.
- Received March 31, 2016.
- Accepted July 25, 2016.
This is an open access article distributed under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)