Four different soil samples (Table 1) were used in order to estimate the absorption of THz radiation at different frequencies. They were selected with respect to variation in important soil parameters, namely organic matter (OM) and particle size distribution (soil texture) and with respect to the physical bulk density. Differences between the soil samples were important to induce contrasts for the THz approach. Three samples were natural soils collected in Potsdam, Germany (soils 1 to 3 in Table 1). One was an artificial soil, composed to obtain a sample with medium organic matter content (soil 4 in Table 1). The samples were ground, sieved to 2 mm and air dried. The air dried samples contained a certain amount of water as shown in Table 1, which may be relevant for THz attenuation. All samples were exposed to the same humidity conditions during the measurement campaign. They were analyzed for organic matter and water content (Table 1) according to VDLUFA [20–22].

Particle size distribution was analyzed by sieving with different meshes. Figure 1 shows the distribution of particle sizes over eight classes.

Figure 1 shows a change in the tendency at particle sizes of 25 μm to 63 μm and 355 μm to 500 μm for all four soil samples. The smallest particle size was related to the amount of organic material in the soil samples (see OM in Table 1), except for Soil 4.

Bulk density of soils may be another influencing factor on THz attenuation. Therefore, bulk density was measured with a wedge sample holder and a measuring cup. Results of 10 repetitions are summarized in Table 2.

For the evaluation of transmission greater than 1 THz, an optically pumped molecular gas laser was used. This laser operates at discrete frequencies between 0.7 THz and 5.2 THz, and it has an output power of up to 10 mW. The transmission measurements were done at frequencies of 1.627 THz and 2.523 THz. The measurements showed no sufficient transmission of THz radiation at this frequency range.

THz frequencies from 340 GHz to 360 GHz were generated with an yttrium iron garnet (YIG)-oscillator operating at 11–12 GHz as fundamental source. Multiplier diodes were used to upconvert the signal of the YIG oscillator (VDI-TX-S119, Virginia Diodes, Inc., Charlottesville, VA, USA). The power was emitted from a horn antenna with a Gaussian beam shape of 9° divergence at approximately 1 mW output power. The beam was optically modulated with a chopper wheel with frequencies ranging from 23 Hz to 27 Hz and then amplified by a lock-in amplifier. A lens made of TPX^® was used to focus the emitted power on the sample [Figure 2(a,b)]. A second lens was used to collect the power transmitted through the soil sample and to focus it onto a Golay cell detector (Tydex, St. Peterburg, Russia). The transmitted power was amplified with a lock-in amplifier (SR850, Stanford Research Systems, Inc., Sunnyvale, CA, USA), which in turn was referenced to the optical chopper. Data acquisition was done automatically with a computer.

Two setups were tested for exposing the samples to the THz radiation. In the first setup, the soil was placed in a Teflon sample dish. After putting the soil into the dish, the sample was carefully compacted, and its thickness was measured [Figure 2(a)]. The second setup employed a wedge sample holder of 6 cm height (y-direction), 6 cm length (x-direction) and of varying width from 0 cm to 2 cm (z-direction). The walls of the wedge sample holder were made of 2 mm high density polyethylene (HDPE) plates. This second setup was designed to reduce the complexity in determining the soil absorption coefficient. Based on the constant thickness of the walls and their plain surfaces absorption and refection coefficients of the wedge sample holder were independent from the position of the THz beam as shown in Figure 3. This was also true for the interface to the soil sample, because the diameter of the divergent beam was much larger than the soil particles.

The transmitted power received by the Golay cell detector, Pt, is given by: (1) P t ( d ) = P e T atm T 1 T 2 e − α d (2) P t ( d ) P e = T atm T 1 T 2 e − α dwhere P_e is the power emitted from the source, T_atm is the atmospheric transmission, T₁, T₂ are the transmission of the HDPE windows of the sample holder, d is the thickness of the soil at the measuring position and α is the absorption coefficient of the soil. The transmitted electromagnetic wave must pass all materials in series and therefore the transmission coefficients must be multiplied in Equation (1).

From this the absorption coefficient follows as: (3) ln ( P t ( d ) P e ) = α d − ln ( T atm T 1 T 2 )

The atmospheric transmission and the transmission of the HDPE windows are constant and appear as an offset to the Equation (3). The Equation (3) explains why the linear regression can be used to determine the absorption coefficient and why the unknown reflection-coefficients do not affect the soil dependent absorption coefficient. The soil samples were compacted by knocking the wedge on a table. The soil density in the wedge varied from high to low from the bottom to the top of the wedge, caused by the soil’s own weight. The same protocol was applied to all soil samples. The wedge was scanned with a xy-positioning stage ([Figure 2(b)] (LIMES 90, OWIS GmbH, Staufen, Germany). The increment in x- and y-direction was set to 1 mm while the focus of the beam was adjusted to the wedge mid-position in the z-direction. Lock-in integration time at each grid point of the xy-array was 1 s. By this we obtained an image of 60 columns and 60 rows summing up to 360 pixels.

To investigate the ability of visualizing organic material with high water content embedded in air dried soils, a piece of a carrot as well as a garden bean were prepared in a dish (70 mm × 70 mm) [Figure 4(a)]. Possible signal variations caused by sample preparation, measurement day, and tilt of the wedge mount were minimized by filling the wedge with different soil samples at the same time. This layering of soils is shown in Figure 5. Additionally, a piece of asparagus was placed in the wedge sample holder as shown in [Figure 4(b,c)].

The absorption coefficients were derived from the wedge-scanning approach (second setup) by linear regression. Only data from the mid-region of the image were used. Data from the boarders were omitted due to distortion effects at the edges. Every column of data, collected along the y-direction, corresponded to a known, constant thickness of the sample. The sample thickness was 2 cm at the x-position (6 cm) for the wedge used, and so the absorption coefficient (in decibels per millimeter) needed to be multiplied by a factor of three. For regression analysis, data were averaged along the columns and then related to the respective sample thicknesses [Equation (4)]. Only those columns were used that have more than 20 values exceeding the least significant bit of the analog to digital converter. Linear regression models were fitted to the data by means of ordinary least squares. The slope of the regression line is a direct estimate of the absorption coefficient as expressed in decibels per millimeter. This can be explained as follows:

No offset zeroing was needed for the absorption of the HDPE wedge wall material;

The following equations were used: (4) mean ( x ) = ∑ i = 1 n I ( x ; y i ) n (5) S ( x ) = 10 ⋅ log [ mean ( x ) ] (6) S ( x ) = offset + α x / 3 (7) d = x / 3where I(x;y) is the signal intensity at the xy-position in the image from the lock-in amplifier output, and S(x) are the mean values along each column (y_i) on a decibel scale. At each day of measurement, temperature and humidity were recorded. All values, including the frequencies used, are listed in Table 3.

The small amount of soil sample area in the 25 mm × 25 mm dish caused poor repeatability among the reproductions of sample preparations (data not shown). The high fluctuations of the signal intensity are shown in Section 3.3. The main reasons for the poor repeatability included variations in the soil sample, an insufficient thickness preparation procedure with respect to the stamp and variations in the emitter power on different days. However, some differences for the transmitted signal intensity among the soil samples were detectable. The second approach that used the wedge sample holder was more efficient because all sample thicknesses were measured in one scan, and the influence of the soil density was averaged for the y-height. Therefore, the dish setup was only used with a 75 mm × 75 mm dish to image the buried test objects in the soil (Figure 4).

The wedge sample holder has the benefit that no position adjustments or calibrations are needed to determine the absorption coefficients. However, the usability of this setup must be evaluated. Therefore, the empty wedge sample holder was measured at three frequencies in order to estimate its influence on the transmitted signal intensity. For all three frequencies, an undistorted center region was visible (Figure 6).

The artifacts at the corners, as shown in Figure 6, are caused by the frame of the wedge holder, which shadowed parts of the divergent beam. The transmission at the three frequencies as measured in the central region of the holder is slightly different. However, this appears as an offset in the analysis and does not affect the attenuation coefficients of the soil samples.

As shown in Table 3, nearly equal frequencies were used to measure the local transmitted signal intensity. The resulting images, shown in Figure 7, indicate why averaging along the y-direction was required. One can observe strong local signal variations. Thus, a single point measurement could be misleading and would not be adequate for characterizing these difficult soil samples.

The blue regions in the acquired images (Figure 7) represent a measured intensity of zero, meaning that the signal intensity was below the detection limit of the setup in this configuration. Because zero cannot be a permissible value on the decibel scale, all zeros were rejected and replaced by missing values. The mean value and standard deviation for each column are plotted in Figure 8. Every plot was adjusted with an additional offset to separate the graphs and to increase visibility. Lengths of plots vary in Figure 8 for the different soil samples, because only those signal amplitudes were included that are based on more than 20 nonzero measurements in a column.

Figure 8 shows representative plots of the signal absorption for the four soil samples measured with three different frequencies. The results varied with the size and position of the used inner xy-array of the data points. Therefore, the estimated absorption coefficients are inaccurate as shown by the repeated measurements at 340 GHz in Table 4.

Thus they should not be used for precise predictions. However, the differences between the absorption coefficients were obvious. These results demonstrated the feasibility of discriminating soils by THz radiation. With a more sensitive and stable setup as well as a mare reproducible sample preparation, quantitative measurements may be possible. A well defined calibration protocol is required. With respect to the error of estimate of the absorption coefficient (Table 4), higher absorption can be assumed with higher measuring frequencies for the first three soil samples (Table 4). This rule is not applicable to the THz analysis of the fourth soil sample. We may conclude that spectral information in the THz region contains additional information about different absorption mechanisms. To verify this in further experiments more stable measurements over a wider frequency range are required. The influence of water, carbon, nitrate and other soil ingredients needs to be analyzed in more detail. The amount of water and carbon increased in the same manner for the first three soil samples (Table 1). Therefore, the mechanism that dominates the absorption effect cannot be defined with this preparation [Figure 9(b,c)]. Moreover, the influence of the particle size distribution may be responsible for the different spectral behavior of the fourth soil sample. The first three soil samples had the same sequence for the value of the absorption coefficient and the amount of large and small particles, and the sequences were inversed for the middle particle size. Soil 4 had the same middle value as Soil 2, but they had a different distribution for small and large particle densities (Figure 1). The water content of Soil 3 and Soil 4 was nearly identical, and so water content was not the reason for the different spectral behavior [Figure 9(a)].

With respect to the densities (Table 2) of the four soil samples, the signal absorption is not dominated by the density, because Soil 1 has the highest density but the lowest absorption. Therefore, the absorption is dominated by the conducting and dielectric properties of the soil ingredients and not by the density. As previously mentioned, one disadvantage of the setup in the present study is the poor reproducibility of the absorption coefficients. Table 4 shows the results of the first measurements and the three repeated measurements for 340 GHz. The difference between the measured results demonstrates the difficulties in reproducing the sample preparation and alignment. Humidity changes may have also caused the signal variations. However, humidity changes did not dominate the effect because humidity was higher for the second measurement of Soil 2 in which the determined absorption coefficient was lower. The difference for Soil 3 may have been caused or amplified by the restricted number of useful data points, resulting in unreliable quantitative results. The alignment difficulties may be overcome in the future with two modifications of the setup. First, implementing a snap-in adapter for the wedge sample holder may allow for higher reproducibility of the mechanical alignment, although the problem of refilling the sample holder in a reproducible manner would not be solved. However, this could be solved with a second modification by using a reference sample that has to be measured together with the soil sample in the wedge at the same time. These modifications should allow better quantitative measurements in future experiments. This new approach is demonstrated in Figure 10 with the first three soil samples. The beam divergence caused a triangle overlay in the signal between the soil samples. Therefore, a larger wedge sample holder or a smaller beam divergence is required.

The ability to focus the THz beam has the advantage of a high spatial resolution of localized absorption causes, thus, enabling imaging. Visualizing with high spatial resolution in terms of images is an important application for a sensor setup. The potential of this visualization method is shown in Figure 10 with an image of three different soil samples in the wedge sample holder.

The different signal absorption of the three soil samples was visualized by the image colors and was highly detectable. The signal absorption in the bottom region was an edge effect from the overlay of the divergent beam and the aluminum part of the sample holder. Under this setup condition, the Lock in amplifier was overdriven for the non-soil region, allowing better signal resolution in the soil region. Thus, the signal amplitude was still detectable for the highest sample thickness.

The imaging capability of THz radiation was also demonstrated in the second approach. The ability to find and analyze buried vegetables is also an important instrumentation for horticulture, to find and analyze asparagus for example. The THz approach may also be a solution for the difficulties encountered in analyzing buried objects because of the high spatial resolution and soil penetration possibilities.

The high water content of vegetables caused signal absorption for the transmitted THz wave. The blue regions in Figure 11 indicate the positions of the carrot and garden bean. The garden bean had less signal absorption because of its smaller size and lower water content as compared to the carrot. The image in Figure 11 is overlaid with some type of interference, which demonstrates the poor absorption coefficient analysis under the dish approach. Figure 12 shows the absorption image of the asparagus prepared in the wedge sample holder [see also Figure 4(b,c)].

The position of the asparagus was localized using a higher signal absorption, and the hand-drawn dimension of the asparagus was manually added to Figure 12. The signal amplitude, dynamic range and beam divergence must be optimized for better spatial resolution and information content. However, the imaging capability of the THz approach was clearly demonstrated, thus suggesting advantages for future applications in horticulture experiments. High-oriented structures in the asparagus may also induce different signal absorption for different polarizations of the transmitting signal, which may be used to raise or insert a new contrast in the image in future applications.