Observation of

Volumetric Water Content and Reflector Depth

with Multichannel Ground-Penetrating Radar
in an Artificial Sand Volume
Jens S. Buchner, Alexander Kühne, Benny Antz, and Kurt Roth Ute Wollschläger
Institute of Environmental Physics UFZ
Heidelberg University, Germany Helmholtz Centre for Environmental Research
email: [email protected] Leipzig, Germany

Abstract—The performance of the multichannel ground- is necessary [14]. Such presumptions are not requested for the
penetrating radar (GPR) method introduced by [1] to simul- multichannel GPR method established by [1] which allows to
taneously estimate reflector depth and average volumetric soil simultaneously infer large-scale layer geometries and average
water content is tested on an artificial soil volume. The test
site consists of several layers of sand which are partly wa- soil water content.
ter saturated. Volumetric water content and reflector depth In this study, the applicability and accuracy of the mul-
deduced from the GPR measurements are compared with in tichannel GPR method introduced by [1] is investigated on
situ measurements. It is shown that both reflector depth and an artificial test bed. It consists of several layers of sand
volumetric water content can be reconstructed with an accuracy of known geometry and it is water saturated at the bottom.
of about 0.1 m and 0.03 . . . 0.04, respectively. Possibilities for
improving the multichannel GPR method are identified in terms Volumetric water content is independently monitored by an
of the employed ray path model and the measurement setup. array of time domain reflectometry (TDR) sensors buried in
the soil at known depths.
Index Terms—Ground-Penetrating Radar, Surface GPR, Mul- We show that the multichannel GPR method can recover
tichannel GPR, Subsurface Geometry, Volumetric Water Con- the layer geometry as well as the volumetric water content
tent
with sufficient accuracy demonstrating the method to be well
suited for this type of application.
I. I NTRODUCTION
The vadose zone plays an important role in the global II. M ATERIALS AND M ETHODS
hydrological cycle as it is the highly dynamic zone located A. Multichannel GPR
between the groundwater and the atmosphere. To understand
the hydrological processes in the vadose zone, knowledge In this study we employ the multichannel GPR method [1]
about its layer geometry [2] as well as the water content to infer the average volumetric water content of distinct soil
distribution and dynamics is crucial [3], [4].
To date, many geophysical methods are available to
a1 a2 a4
non-invasively explore the shallow subsurface [5]. Ground-
penetrating radar is well established in hydrogeophysical a3
applications to measure the vadose zone water content dis-
tribution and dynamics [6]–[8] as well as the subsurface T R T R
architecture [9], [10].
The traditional single channel GPR is hardly applicable
to obtain an accurate simultaneous observation of layer ge- p2
p1 p4
ometries and water content over large distances with a high
density of measurements, however. Multi-offset methods like p3
common midpoint (CMP) measurements [11] or based on that
reflection tomography coupled with prestack depth migration
[12] yield both quantities. But, large-scale applications of these
methods involve a high measurement effort. Thus, usually Fig. 1. Multichannel GPR measurement setup: Two bistatic GPR antennas
either the average water content is estimated by using reference are connected at a fixed distance to achieve four different antenna separations
reflectors [13] or ground truth information on the layer depth (a1 to a4 ) leading to four different ray paths Fig. (p1 to p4 ) [1].

978-1-4577-0333-1/11/$26.00©2011 IEEE
layers, reflector depth, and reflector inclination angle. All three
quantities are obtained from common offset multichannel GPR
measurements and are measured continuously and simultane-
ously along a line.
The main principles of a multichannel GPR measurement
are similar to the other quantitative time domain surface GPR
methods: an electromagnetic signal is emitted into the subsur-
face. There, it is eventually reflected due to abrupt changes in
the dielectric permittivity distribution which typically occur at
layer interfaces. The reflected signal is detected by a receiving
antenna and its travel time can be deduced. The simultaneous
estimation of reflector depth, average soil dielectric permittiv-
ity, and reflector inclination angle is realized with at least three
independent measurements conducted using different antenna Fig. 2. Measurement site: In the middle the GPR measurement line is
separations (Fig. 1). Two or more bistatic GPR antennas are indicated by the gray band. The different sand types occurring at the ground
connected in a row to achieve multiple antenna separations surface can be distinguished by the different coloring. At the back left the
automatic weather station and the pumping well are visible.
using all available transmitters and receivers. The whole setup
is moved along the survey line and takes measurements at
predefined trace intervals. With that, for each channel, a
common offset measurement along the same GPR line is
retrieved. The radargrams from the various channels are then
shifted laterally such that the midpoints of the transmitting
and the receiving antennas coincide. Hence, narrow common
midpoint gathers are obtained for each measurement point
along the survey line.
The evaluation of the CMP measurements is done in several
steps: First, the time-zero offset, determined from indepen-
dent measurements in air, is subtracted. Second, reflectors
appearing at a minimum of three antenna separations are
identified and picked. Then, at every measurement point along
the line one travel time t(an ) per antenna separation an is
given. Employing a least-squares optimization procedure the Fig. 3. Layer geometry of the soil volume. The picture shows the y-z plane
reflection hyperbola at x = 2.1 m with exaggerated y-axis. The black area at the bottom denotes
√ the gravel layer. On top of it the three different types of sand are indicated in
ε p shades of gray. The black dots mark the positions of TDR and temperature
t(a) = cos(α) 4d2 + a2 (1) sensors positioned in the y-z plane with x = 0.7 m.
c0
is fitted to the CMP data and dielectric permittivity ε, reflector
depth d, and reflector inclination α are estimated. To account polyethylene foil. The bottom of the cube is filled with a gravel
for the limited horizontal resolution of the antenna [15] data layer (grain size: 9 . . . 32 mm). On top of it several layers of
from measurement points in the close surrounding might be quartz sand, all with a grain size smaller than 2 mm but of
included as well. Note that this procedure assumes that the slightly different textures, are installed with mirror symmetry
space above the reflecting interface is homogeneous without with respect to the y-z plane (Fig. 2). The actual geometry of
any interface in between. Uncertainty estimates are provided the sand layers is depicted in Fig. 3.
by the least-squares fit [16]. The estimates represent the σ The sand volume is instrumented with 32 TDR probes (three
confidence interval of the probability distribution for d and ε rod, 20 cm) and temperature sensors (PT 100) to measure
from the hyperbola model. volumetric water content by point observations (Fig. 3). TDR
The volumetric water content is calculated via the complex and temperature measurements are recorded by an automatic
refractive index model (CRIM) according to [17]. The tem- weather station (Campbell Scientific) at 30 min intervals. The
perature dependency of the dielectric permittivity of water is TDR traces are evaluated with a two point evaluation method
corrected using a first order exponential model [18]. which retrieves the signal travel time. Based on a previous
calibration of the TDR probes in water and air, the travel time
B. Measurement Site is transferred to dielectric permittivity values. The temperature
The ASSESS-GPR test site (Fig. 2) near Heidelberg, Ger- corrected volumetric water content is then calculated exactly
many, is a cuboid (4 m*19 m*2 m) confined with concrete in the same way as for the GPR measurements (Sect. II-A).
side walls and floor with the upper side open to atmosphere. Since the multichannel GPR method is specifically applica-
The floor and the inner walls are covered by a water proof ble to the vadose zone, the soil volume is kept in a partially
saturated state. Due to precipitation, water enters the system 5
from the upper boundary and ponds on the floor of the cuboid 10
which is sealed by the foil. This leads to a vertical division 15

Time [ns]
into a fully saturated part and a partially saturated section
20
on top of the water table. The water table is controlled by
25
periodical pumping in a previously installed pumping well
(Fig. 2). In the unsaturated part of the soil volume, the water 30
content distribution is determined by the different soil textures 35
of the various sand types. Most importantly, characteristic
5 10 15
jumps occur at layer boundaries which lead to reflections of
the GPR signals.
To achieve this situation, the sands were selected based 5
on differences in their hydraulic properties ([19], [20]) which 10
were estimated for several sands in preceding multi-step out- 15

Time [ns]
flow experiments which had been conducted in the laboratory.
20
Based on this information, the sands were selected with the
25
aim to induce maximal volumetric water content contrasts at
the layer boundaries. 30
35
C. Measurements and Evaluation
The GPR measurements presented in this study were carried 5 10 15

out on 15th of July 2010. Two shielded bistatic antenna pairs

(Ingegneria dei Sistemi S.p.A., Italy) with a nominal center 5
frequency of 400 MHz were employed. Measurements were 10
carried out on the line with x = 2.1 m (Fig. 2) and y = 1 . . . 18 15
Time [ns]

m. We performed three measurements with the setup sketched

20
in Fig. 1. Signal emissions were triggered every 0.05 m by
a measuring wheel. For each of the three measurements, the 25

separation between the two antenna boxes was enhanced by 30

40 cm resulting in a data set consisting of radargrams with 35
7 different antenna separations for the multichannel analysis.
The radargrams used for evaluation are shown in Fig. 4. 5 10 15
y [m]
For evaluation, a dewow filter was applied and two con-
tinuous reflections were picked (first zero crossing with an
Fig. 4. Radargrams corresponding to ray path p2 with a2 = 0.66 m (upper),
accuracy of 0.03 ns) in each radargram. To obtain absolute a2 = 1.06 m (middle), and a2 = 1.46 m (lower). Picks are indicated by
travel times the time zero was estimated from direct wave the black lines. By comparison with Fig. 3, the major reflections occurring
measurements in air. Problems with the time zero estimation at layer boundaries can be identified. Weak reflections in-between are most
likely caused by compression of the sand during the construction process.
and instrumental issues caused data incoherence for the differ- From t = 3 ns to t = 7 ns, t = 7 ns to t = 10 ns, and t = 9 ns
ent ray paths. Thus, only three antenna separations (ray path to t = 13 ns the direct waves are found in each radargram. In the lower
p2 with a2 = 0.66 m, a2 = 1.06 m, and a2 = 1.46 m) part (27 to 37 ns, 28 to 38 ns, and 29 to 39 ns) the reflections from the
gravel layer and the concrete floor are detected. For x ≥ 16.0 m we find a
could be used for the final estimation of reflector depth and steep reflection stemming from the end walls. To better the depicting quality,
volumetric water content. For the actual hyperbola fitting all the amplitude was cut at 5%, 13%, and 15% of the maximum (positive and
data points in an interval of 0.2 m around the point of interest negative) amplitude.
were included in the analysis.
III. D ISCUSSION GPR signal quality caused by clutter or interfering reflections.
The results for reflector depth and volumetric water con- This indicates a general limitation of the current method. It
tent in comparison with in situ observations are depicted in can only be applied if the wavelet can be identified. Hence,
Fig. 5. The multichannel GPR method yields both quantities geometrical features like synclines (e.g. at y ≈ 11 m) cannot
simultaneously for each measurement point along the GPR be reconstructed if they cause interference as it is the case
line. Hence, the reflector depth and water content at one here (Fig. 4).
measurement point and reflector stem from one evaluation. The uncertainty estimates depicted for the GPR results
Since two reflectors were evaluated (as indicated by the were retrieved from the optimization algorithm (Sect. II-A)
picks in Fig. 4) one average volumetric water content value is without including any uncertainties in the travel times. For
given for the space above the upper as well as the lower re- the volumetric water content they were calculated by Gaussian
flector. The data gaps in the GPR results are due to insufficient error propagation.
 0.0 layers above the upper and lower reflector, respectively.
GPR
0.2 In Situ For both the upper and the lower reflector one finds lateral
0.4 variations in the estimated reflector depth. These did not occur
0.6 in the layer geometry during the construction of the soil
0.8 volume. As can be seen in the lower part of Fig. 5, the water
1.0
d [m]

content values show unexpected lateral variations as well. But,

1.2 high values of water content correspond to low values of
1.4
reflector depth. This indicates that the retrieved reflector depth
1.6
and water content values are anti-correlated. A finding that
1.8
is supported by the results for the correlation coefficient of
2.0
reflector depth and permittivity obtained by the optimization
2.2
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 procedure: more that 80% of all correlation coefficients are
y [m] smaller than −0.9. An investigation of a synthetic dataset
0.10 has shown that a periodic disturbance of the surface topology
GPR Lower Reflector
GPR Upper Reflector leads to similar results. Hence, the variations are most likely
In Situ Lower Reflector caused by artificial disturbance: Due to the pulling of the
Volumetric Water Content [−]

0.08 In Situ Upper Reflector

antenna, the sand surface got wavy. This strong influence of
surface disturbance can be reduced by including more antenna
0.06
separations that cover a larger range of values.
0.04 B. Volumetric Water Content
For the interpretation the volumetric water content deduced
0.02 from the in situ and multichannel GPR observations (Fig. 5)
several important facts have to be taken into account:
0.00 • When calculating the volumetric water content from
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
dielectric permittivity using a volumetric mixing formula
y [m]
(CRIM), the porosity, the dielectric permittivity of the
Fig. 5. Results of the multichannel GPR observation compared to in pore matrix have to be known. Here, the porosity was
situ measurements. The upper and lower graph shows reflector depth and deduced from volumetric soil samples providing an av-
volumetric water content, respectively. The in situ values for volumetric water
content are calculated by linear interpolation and subsequent averaging of all
erage value of 0.37. For all types of sand the relative
sensors above the reflector at the point of interest. dielectric permittivity was assumed to be 5, respectively.
The temperature correction was performed individually
for each in situ measurement. In contrast, for the GPR
A. Reflector Depth measurement the mean of all temperature measurements
above each reflector was used. This causes a maximal
Comparing the reflector depth results, an accuracy of about deviation in water content of 0.0003.
0.1 m is found on average. Considering the accuracy of the • The in situ measurements are taken at x = 0.7 m while
in situ measurements (0.05 m) and the uncertainty estimates the GPR measurements were carried out at x = 2.1 m.
retrieved from the multichannel GPR measurements, this result Although the site was constructed to achieve mirror sym-
is judged to be acceptable. metry it is not necessarily valid to assume homogeneity of
While the best results are obtained for the flat parts, still the water content along the x-axis. Specifically, because
a structural over or under estimation of the reflector depth textural heterogeneities in the soil might be present from
is observed for the upper and lower reflector, respectively. the construction process.
This has two most likely reasons: First, the accuracy of the • Most importantly, the particular averaging volumes of
time zero estimation is limited because the wavelet shape multichannel GPR and the in situ measurements have
changes between measurement in air and on-ground due to the to be considered. For multichannel GPR this volume is
ground coupling of the antennas. Second, the lower reflector is approximately represented by the ray coverage of each
evaluated with a one-layer ray path model although two layers CMP measurement. In contrast, the averaging volume of
are given. Hence, the travel time may for instance be under- the TDR measurement just includes the close surrounding
estimated by the model which would result in a compensation on the order of some centimeters of the rods [21].
by overestimation of reflector depth and dielectric permittivity The TDR in situ measurements show that the average water
by the algorithm. content above the lower reflector is higher than above the
Considering the dipping parts of the reflector, the depth upper reflector. Because of the presence of the water table at
accuracy is significantly worse than for the flat parts. Again, the bottom of the cuboid and the given hydraulic properties,
this might be caused by an insufficient model representation an increase of water content with depth is expected. This
of the ray path: for y ≤ 7.5 m (Fig. 3) we find two and three increase in water content with depth is not observable in the
GPR results. The contrary is given for most of the data points. R EFERENCES
However, this is hardly significant since the results are subject [1] H. Gerhards, U. Wollschläger, Q. Yu, P. Schiwek, X. Pan, and K. Roth,
to the lateral variability discussed above. “Continuous and simultaneous measurement of reflector depth and
Obviously, the lateral variability of the water contents average soil-water content with multichannel ground-penetrating radar,”
Geophysics, vol. 73, no. 4, pp. J15–J23, 2008.
inferred from the GPR measurements is not represented by [2] A. Bronstert and E. Plate, “Modelling of runoff generation and soil
the given uncertainty estimates. This can be understood since moisture dynamics for hillslopes and micro-catchments,” Journal of
the data are subject to an artificial disturbance (Sect. III-A). Hydrology, vol. 198, no. 1-4, pp. 177–195, 1997.
[3] H. Vereecken, J. Huisman, H. Bogena, J. Vanderborght, J. Vrugt, and
Accounting for the lateral variability, the results displayed J. Hopmans, “On the value of soil moisture measurements in vadose
in Fig. 5 show that under the given conditions the volumetric zone hydrology: A review,” Water Resources Research, vol. 44, no. 8,
water content can be observed with an absolute accuracy pp. 1–21, 2008.
[4] A. W. Western, S.-L. Zhou, R. B. Grayson, T. A. McMahon, G. Blöschl,
of 0.03 . . . 0.04 using the multichannel GPR method. When and D. J. Wilson, “Spatial correlation of soil moisture in small catch-
comparing this result to the absolute values for the volumetric ments and its relationship to dominant spatial hydrological processes,”
water content one has to account for the high sensitivity of the Journal of Hydrology, vol. 286, no. 1-4, pp. 113–134, 2004.
[5] D. Robinson, A. Binley, N. Crook, F. Day-Lewis, T. Ferre, V. Grauch,
water content for low dielectric permittivity values ([17] equ. R. Knight, M. Knoll, V. Lakshmi, R. Miller et al., “Advancing process-
14): Relative errors of dielectric permittivity lead to unlimited based watershed hydrological research using near-surface geophysics: a
relative errors of volumetric water content since it goes to vision for, and review of, electrical and magnetic geophysical methods,”
Hydrological Processes, vol. 22, no. 18, pp. 3604–3635, 2008.
zero. Under these premises, we consider the accuracy and the [6] J. Huisman, S. Hubbard, J. Redman, and A. Annan, “Measuring soil
matching of the results shown in Fig. 5 to be very good. water content with ground penetrating radar: A review,” Vadose Zone
Journal, vol. 2, no. 4, pp. 476–491, 2003.
IV. S UMMARY AND C ONCLUSIONS [7] S. M. J. Moysey, “Hydrologic trajectories in transient ground-
penetrating-radar reflection data,” Geophysics, vol. 75, no. 4, pp.
The multichannel GPR method provides continuous infor- WA211–WA219, 2010.
mation about reflector depth and volumetric water content [8] S. Lambot, E. C. Slob, M. Vanclooster, and H. Vereecken, “Closed loop
GPR data inversion for soil hydraulic and electric property determina-
along a measurement line. This method was applied at a test tion,” Geophysical Research Letters, vol. 33, no. 21, p. L21405, 2006.
site consisting of an artificial soil volume which is monitored [9] A. Neal, “Ground-penetrating radar and its use in sedimentology:
by TDR measurements at various depths. Due to its specific principles, problems and progress,” Earth-Science Reviews, vol. 66, no.
3-4, pp. 261–330, 2004.
design, the site is a good test bed that provides independent [10] R. L. V. Dam, “Landform characterization using geophysics - recent
measurements which can be compared to the results of the advances, applications, and emerging tools,” Geomorphology, vol. In
GPR measurements. Press, Corrected Proof, pp. –, 2010.
[11] R. J. Greaves, D. P. Lesmes, J. M. Lee, and M. N. Toksöz, “Velocity vari-
The comparison of measured and estimated reflector depths ations and water content estimated from multi-offset, ground-penetrating
show a consistence of the data ranging within 0.1 m. Differ- radar,” Geophysics, vol. 61, no. 3, pp. 683–695, 1996.
ent averaging volumes and measurement locations weakened [12] J. H. Bradford, “Measuring water content heterogeneity using multifold
GPR with reflection tomography,” Vadose Zone Journal, vol. 7, no. 1,
the volumetric water content results. Nevertheless, the water pp. 184–193, 2008.
content was found to be recovered by the GPR observations [13] K. Grote, S. Hubbard, and Y. Rubin, “GPR monitoring of volumetric
within an absolute accuracy of 0.03 . . . 0.04. water content in soils applied to highway construction and maintenance,”
The Leading Edge, vol. 21, no. 5, pp. 482–485, 2002.
Various possible improvements of the multichannel GPR [14] U. Wollschläger and K. Roth, “Estimation of temporal changes of
method were indicated by the results. First, the time zero volumetric soil water content from ground-penetrating radar reflections,”
determination should be optimized to increase the accuracy Sensing and Imaging: An International Journal, vol. 6, pp. 207–218,
2005.
in the estimation of the signal travel time. Second, the rep- [15] D. Daniels, Ground Penetrating Radar 2nd Edition, 2nd ed., D. Daniels,
resentation of multiple layers and higher order terms of layer Ed. The Institution of Electrical Engineers, London, United Kingdom,
dipping in the ray path model should be implemented. Third, 2004.
[16] U. Wollschläger, H. Gerhards, Q. Yu, and K. Roth, “Multi-channel
an inclusion of more antenna separations in the measurement ground-penetrating radar to explore spatial variations in thaw depth
setup to increase the number of independent measurements and moisture content in the active layer of a permafrost site,” The
would improve the accuracy of the measurement as well. Cryosphere, vol. 4, no. 3, pp. 269–283, 2010.
[17] K. Roth, R. Schulin, H. Flühler, and W. Attinger, “Calibration of time
Multichannel GPR was demonstrated to be a reliable and domain reflectometry for water content measurement using a composite
accurate method for the observation of reflector depth and dielectric approach,” Water Resources Research, vol. 26, no. 10, pp.
volumetric soil water content. Since it is non-invasive and 2267–2273, 1990.
[18] U. Kaatze, “Complex permittivity of water as a function of frequency
fast to apply it is a favorable method to observe vadose zone and temperature,” Journal of Chemical & Engineering Data, vol. 34,
architecture and water content at a scale of meters up to no. 4, pp. 371–374, 1989.
kilometers. [19] Y. Mualem, “A new model for predicting the hydraulic conductivity of
unsaturated porous media,” Water Resources Research, vol. 12, no. 3,
pp. 513–522, 1976.
ACKNOWLEDGMENTS [20] M. T. Van Genuchten, “A closed form equation for predicting the
Financial support for this research is provided by the hydraulic conductivity of unsaturated soils,” Soil Science Society of
America Journal, vol. 44, no. 5, pp. 892–898, 1980.
Deutsche Forschungsgemeinschaft (DFG) through project RO [21] D. A. Robinson, S. B. Jones, J. M. Wraith, D. Or, and S. P. Friedman, “A
1080/8-2. review of advances in dielectric and electrical conductivity measurement
We would like to thank the reviewers for their profound in soils using time domain reflectometry,” Vadose Zone Journal, vol. 2,
no. 4, pp. 444–475, 2003.
comments that helped to improve the quality of the paper.