Geoscience Reference
In-Depth Information
wetting front through dry soil. If not for interactions of the capillary fringe, the horizontal bound-
ary between wet and dry soil would always offer an excellent radar reflective plane. For example, a
falling or retreating free-water edge is much more difficult (if not impossible) to delineate with GPR
than a stagnant or rapidly rising water table due to the more gradual, elongated dielectric change
remaining above the saturation zone, as residual free moisture is still draining. Fortunately, a large
body of saturated soil does have a differentiating signature from nearby drier soil, due to it being
a much more electrically conductive medium. The wetter region, being a much higher conductive
body than the surrounding drier soil, significantly weakens the return signal, and thus less energy is
reflected back to the surface. Extremely high gain, when applied to a very weak (or no) signal, will
highlight the naturally occurring background noise, giving a characteristic telltale static pattern.
When observing this return waveform in oscilloscope format, the later-returning component (or the
lower tail finale) of the waveform, which is normally somewhat steady in its structured movements
(Figure 26.1a), will dance erratically when passing over saturated soil (Figure 26.1b). Within the
radargram, adjacent strong horizontal reflectors at the same depth as that of the saturation zone will
become masked or obliterated in static as the antenna passes over a large saturated region, only to
reappear when reentering a drier region (Figure 26.2). A static pattern that suddenly appears and
masks adjacent strong horizontal reflectors (or a strong uniformity signal) suggests passing over a
highly conductive wetter zone. In this instance, a rectangle of static will fill the radargram from the
saturated zone downward in its entirety.
Subsurface morphology and environmental conditions are widely variable. Subsurface anoma-
lies can generate reflections that mimic the signature of saturated soil, as can also the instrument
settings and the individual survey protocol technique of the operator. Surface observations, site
experience, and calibrations along with limited ground truth probing can assist substantiating that
the GPR signatures are indeed originating from saturated soil.
Employing GPR to trace diminutive pathways of preferential flow is therefore very challeng-
ing, as the channels conducting water primarily do so under seasonal wet soil conditions, having
also saturated the surrounding soil, and are thereby masked. One approach is to map the channels
directly by artificially applying a significant quantity of water to the subsurface during very dry soil
conditions, and to track the distinct wetting front as it progresses outward beneath the surface.
26.2 CASe hIStoRy
This case study highlights mapping subsurface lateral preferential flow in Major Land Resource
Area 134 (MLRA 134—Southern Mississippi Valley Silty Uplands), which extends along the Mis-
sissippi River from southern Illinois to northern Louisiana. These highly productive agricultural
lands are primarily in row-crop production and represent thousands of hectares that formed in the
loess-covered Tertiary-aged Claiborne and Wilcox geologic formations (Hardeman, 1966). At one
research site, an interface between the loess (wind-blown silt) and the alluvial clay covering the
underlying paleosol (an ancient seabed of deep stratified sands) forms a distinct textural discontinu-
ity for perching water at a below-surface depth of approximately 2 m. The objective of the project
was to develop a noninvasive tool that illustrates the rate and direction of the preferential flow that
perches and moves atop this interface.
26.2.1 s
u R v e y
P
R of t of c of l
Conducting a high-resolution GPR survey generates copious data. Gigabytes of GPR data per
hectare must be recorded while traveling the lengthy, closely spaced traverses that are required in
detecting preferential flow pathways. Thus, it is implausible to employ high-resolution GPR map-
ping over entire watersheds if relying upon manual radargram interpretations (Figure 26.3). For this
reason, we employ a neural-network (NN) classifier to automatically segment patterns based upon
