5.1. Conformities and Discrepancies between the Different Soil Moisture Methods
We derived SM data at the DWD test site in Munich using two continuous measurement methods; firstly by GPS sensors, which were investigated for the first time and secondly by FD ECH2O probes. Further, we manually measured surface SM with FD ThetaProbes and gravimetric probes in a weekly to bi-weekly basis. Moreover, we continuously modelled SM by PROMET.
The agreement of all results from the different methods is high, and all data follow the overall course of dry and wet events during the observation period of almost 1.5 years. The increase in SM after the occurrence of precipitation events with infiltration into the soil as well as long dry periods are covered by all methods. However, as described in Section 2.4
, the comparability of these methods is limited due to different sampling volumes and different vertical validity ranges within the soil column. According to the results, differences in the SM readings regarding the different methods can be explained by differences in the SM conditions in the upper 10 cm. For a better understanding, the described bulk soil column of 10 cm can be divided into two layers, firstly in approximately the first 5 cm as the direct air and soil interface and secondly in a slightly deeper layer of 5 to 10 cm soil depth.
The gravimetric and ThetaProbe measurements are taken in the uppermost 4 to 5 cm. It is apparent that during dry periods, this layer dries out more than the 5 to 10 cm layer, as the latter is covered by the surrounding soil particles and has no direct exchange with the atmosphere. In the findings, it can be clearly seen that the two manual methods reach the lowest SM values compared to all other methods during these dry periods (see Figure 6
and Table 3
). During wet periods they reach a similar maximum than the other methods because the water percolates quite fast from the surface to the 5 to 10 cm layer in the soil column.
O measurements, which were derived in a layer depth of 5 to 10 cm, show less dynamics but the highest mean value of all methods (see Table 3
). In this vertical soil range, soil water is stored for a longer time, the SM content is more stable and influences from the atmosphere are less pronounced.
SM simulated by PROMET considers the entire 10 cm layer. The SM evolution over the entire period is highly correlated with the ECH2O SM readings derived for the 5 to 10 cm soil depth. However, the SM dynamic range of PROMET is higher, which can be explained by the fact that PROMET also covers the highly dynamic surface processes but shows less dynamic than the manual surface measurements derived by ThetaProbe or gravimetric probes, which only consider the upper centimetres of the soil. It is therefore reasonable that the dynamics of PROMET occur between ECH2O and the manual methods.
PROMET simulations and GPS measurements consider the same vertical range of the uppermost 10 cm and are therefore most suitable for comparison. Positively, the SM curves of GPS and PROMET fit very well, reach the highest correlation coefficient and a very low RMSE and have a similar dynamic range as well as similar minimum, maximum and mean values (see Figure 5
and Figure 6
and Table 3
). However, GPS is slightly more dynamic than the modelled data, especially during dry periods. One reason for this might be that the determined soil texture in the laboratory could be slightly different or inhomogeneous compared to the conditions in the soil plot, which would have an influence on the parameterisation of the PROMET model as well as the Dobson four-component model applied for the GPS approach. This in turn has an influence on the SM calculations with both methods. Secondly, as the GPS measurements undergo reflection, refraction and attenuation processes, which are non-linear, especially in case of high SM gradients within the first 10 cm, the GPS calculations might slightly under- or overestimate SM. This is the case if, e.g., the upper 5 cm are drier or wetter, than the lower 5 cm. This might be the reason why during dry periods, GPS is more similar to the ThetaProbe and gravimetric measurements. In summary, we state that the GPS sensors are more sensitive to the dynamics at the soil surface, which is positive in terms of interpreting events directly at the interface of air and soil.
5.2. Advantages and Limitations of GPS Soil Moisture Measurements
This study successfully demonstrates that low-cost GPS sensors are highly capable of quantitatively and continuously retrieving bulk SM of the upper soil layer in a non-destructive manner. A main advantage is that the GPS antennas are installed below the soil column of interest, in this case the upper 10 cm, which means that the soil above the antennas is unaffected by the sensors after the installation. In contrast, gravimetric measurements are accurate but they are highly destructive as for each sample, soil has to be extracted. Regarding FD and TDR probes, their prongs might influence the soil sampling volume during installation. This could cause destruction of the soil structure that could result in an increased pore volume that can fill up with air or water or compacted soil [58
], which is not the case when using the GPS measurement system.
Further, soil is a quite heterogeneous medium. Soil texture, pore size distribution, the soil density, and the existence of large air and water pockets or larger heterogeneities like small rocks might vary largely even at small scales. This makes SM measurements quite difficult and explains the effect that sensors installed next to each other might possibly measure different SM values [58
]. As the GPS sensors cover the entire soil column above the antennas, which integrate over potential soil inhomogeneities, uncertainties are reduced due to larger sampling volumes. In contrast, most permanently installed FD or TDR sensors only have small sampling volumes at a certain soil depth.
Moreover, we assume the advantage that the GPS sensors are sensitive to the surface SM. Especially the uppermost centimetres of a soil column, at the interface between air and soil, are subject to a high dynamic in SM due to the direct influence of precipitation as well as intense radiation during dry periods. All these processes can be resolved with high sensitivity by the GPS sensor measurements.
It is also considered as an advantage that the GPS signals are broadcasted in the microwave L-band. For SM observations in general, that band is most suitable because the penetration depth of the electromagnetic waves reaches several centimetres even for moist soils. For dry soils, it can even reach several meters [11
]. As the L-band is widely and successfully used for SM retrievals from microwave measurements, e.g., from SMOS or SMAP, SM data derived by GPS use the same band and rely on a similar physical basis. Therefore, GPS-retrieved SM measurements could have a high potential as ground truth validation data for L-band microwave products. However, passive microwave systems derive SM through emission and the active microwave systems, and GNSS-R use signal reflections, and the GPS approach presented in this study is mainly based on signal strength attenuation.
As this method was only tested on a bare soil field, no quantitative statements about the SM retrieval capabilities of this GPS SM approach can be made for vegetation covered soils. However, we assume that this approach has high potential for soils covered, e.g., by agricultural crop plants like wheat, barley or even maize. In this case, an additional antenna above the vegetation would have to be installed to extract signal strength information from the plant water content, which has to be considered in the calculation in relation to the reduced signal strength at the antennas placed within the soil. Forest stands, however, might be too reductive for the GPS signals to also extract the underlying soils’ SM.
Nevertheless, the vertical range of the soil column that can be investigated is limited, as at some point the GPS signal strength becomes too weak to be tracked continuously. The maximum installation depth might, however, differ for different soil textures and SM contents. For example, for high elevation angles (>70 degrees), the mean C/N0 values received above the soil at GPS 1, reached approximately 47 dBHz. At a GPS antenna installation depth of 10 cm at GPS2 and GPS3 and for this specific soil texture, we always received strong enough GPS signals without interruption. During quite dry periods, with an SM content of 20 m3·m−3, the mean C/N0 values of the high elevation angles reached approximately 36 dBHz, whereas they declined to approximately 29 dBHz during wet days, exemplarily at an SM content of 35 m3·m−3. However, an antenna installation depth below approximately 20 cm for this soil type, especially under wet soil conditions, makes potentially no practical sense, because the GPS signals would be attenuated excessively before reaching the GPS antennas. This could result in a reception of too weak GPS signals, which are then unreliable, or even longer signal interruptions or no signals at all. We tested this during a short experiment with an SM content of 35 m3·m−3 at a depth of 20 cm, where the C/N0 declined to approximately 19 dBHz considering the high elevation angles; during even wetter conditions the signals were, however, not continuous anymore. Applying this approach potentially to vegetation-covered soils at deeper laying root-zones, it could be difficult or even impossible to derive continuous SM information regarding this specific soil type.
A further limitation regarding this measurement approach might be a too constrained hemispherical coverage, e.g., due to tall buildings and trees in the surroundings or extremely steep mountain slopes and deeply cut valleys, which could cover the direct line of sight of too many GPS satellites to the receiver. However, other than GPS positioning algorithms, our approach to measure SM does not depend on a minimum number of four satellites to find a solution. Theoretically, the signal of one single GPS satellite for each time step would be sufficient.
Last but not least, as for every retrieval algorithm, a wrong parameterization might lead to imprecise results. To obtain an impression of its sensitivity, we carried out a sensitivity analysis in the next section.
5.3. Sensitivity Analysis of GPS Soil Moisture Measurements
According to Dobson et al. [40
] and Hallikainen et al. [65
], we investigated the sensitivity of the retrieval algorithm for soil texture and soil temperature, which were identified as the most important physical soil parameters for this approach. In addition, we conducted a sensitivity assessment of the soil density because this parameter exhibited a rather large variability in our soil sampling. For this sensitivity analysis, we applied Equations (1)–(9) analogous to the original and correct soil parameterization.
Regarding different soil textures with different sand, silt and clay contents, the same SM values are based on different values of the complex permittivity of moist soil [40
]. The soil texture is mainly represented by the physical soil parameter
, which was set to the value of 49 m2
for the observed soil at the DWD test site in Munich. Lower values for
represent soils with lower clay and higher sand contents, whereas higher values represent higher clay and lower sand contents. We increased and decreased
= 29 m2
represents, for example, the textural class sand and
= 69 m2
silt loam. As shown in Figure 7
a, the changes in
have quite a large impact on the SM results. If the soil is assumed to be too sandy, SM is underestimated, whereas if it is assumed too clayish, SM is overestimated. In case of an overestimation, the dynamic range increases and in case of an underestimation, it decreases. According to these findings, great effort should be taken to obtain reliable soil texture measurements in the laboratory.
In general, the GPS SM retrieval is based on the complex permittivity of water, which is temperature dependent in the microwave range [40
] (see also Table 2
). For this study, we used separate soil temperature measurements to correctly adapt the Dobson four-component dielectric mixing model. In case no soil temperature readings are available, uncertainties in SM calculations might occur due to wrong soil temperature assumptions. To address this, we additionally calculated for each time step the SM with a fixed soil temperature assumption of 15 °C, which is approximately the average annual soil temperature at the DWD test site. Figure 7
b illustrates the difference between the assumed fixed soil temperature and the results with the correct soil temperature. Especially during the colder winter periods, the SM curves deviate the most with a rather high overestimation of up to 8%. However, during warm periods especially in summer, the deviation, expressed as an underestimation, is negligible or less than 2.5%. In case no soil temperature data are available, we recommend an assumption of the soil temperature based on air temperature measurements, if they are available, e.g., from stations nearby. The dynamic range of the air temperature measurements at our test site was only slightly higher than of the soil temperature readings at 5 cm soil depth, which is negligible for the SM results.
To obtain an impression of how soil density variations affect the SM calculations, we decreased and increased the actual measured soil density of
= 1.3 g·cm−3
leading to densities of 1.17 g·cm−3
and 1.43 g·cm−3
leading to densities of 1.04 g·cm−3
and 1.56 g·cm−3
. Figure 7
c shows that these changes in soil density have no severe impact on the SM results. An increase in soil density leads to a slight increase in the SM curve, whereas a decrease in soil density leads to a slight decrease in the SM curve. Though the soil density measurements might not be absolutely accurate, it is still possible to retrieve reasonable SM values.