Soil moisture and electrical conductivity relationships under typical Loess Plateau land covers

Vegetation changes that are driven by soil conservation measures significantly affect subsurface water flow patterns and soil water status. Much research on water consumption and sustainability of newly introduced vegetation types at the plot scale has been done in the Loess Plateau of China (LPC), typically using local scale measurements of soil water content (SWC). However, information collected at the plot scale cannot readily be up‐scaled. Geophysical methods such as electromagnetic induction (EMI) offer large spatial coverage and, therefore, could bridge between the scales. A noninvasive, multicoil, frequency domain, EMI instrument was used to measure the apparent soil electrical conductivity (σa) from six effective depths under four typical land‐covers (shrub, pasture, natural fallow, and crop) in the north of the LPC. Concurrently, SWC was monitored to a depth of 4 m using an array of 44 neutron probes distributed along the plots. The measurements of σa for six effective depths and the integrated SWC over these depths, show consistent behavior. High variability of σa under shrub cover, in particular, is consistent with long term variability of SWC, highlighting the potential unsustainability of this land cover. Linear relationships between SWC and σa were established using cumulative sensitivity forward models. The conductivity–SWC model parameters show clear variation with depth despite lack of appreciable textural variation. This is likely related to the combined effect of elevated pore water conductivity as was illustrated by the simulations obtained with water flow and solute transport models. The results of the study highlight the potential for the implementation of the EMI method for investigations of water distribution in the vadose zone of the LPC, and in particular for qualitative mapping of the vulnerability to excessive vegetation demands and hence, unsustainable land cover.


INTRODUCTION
Landscape alterning, such as conversion of natural ecosystems to agricultural lands, or application of soil conservation measures as revegetation for preventing land degradation, has a significant effect on soil water dynamics. The conversion of natural vegetation to croplands with shallow rooting systems can increase water levels in unconfined aquifers and mobilize salts to groundwater (Hancock et al., 2008;Kurtzman & Scanlon, 2011;Radford et al., 2009;Scanlon et al., 2009). Afforestation or revegetation, where trees, grass, and shrubs are replanted, are related to depletion of soil water and reduction in groundwater recharge fluxes (Adane et al., 2018;A. Allen & Chapman, 2001;Bai et al., 2020;Gates et al., 2011;T. Huang et al., 2013;Ouyanga et al., 2021;Scott & Lesch, 1997;Zhang et al., 2008). Various factors are attributed to the disturbance of the soil water status, such as high water demand, larger water-holding capacity of forest soils, deep roots, climate variability, and planting of vegetation in an inadequate environment (Barbeta et al., 2015;Cramer et al., 1999;Y. Jia & Shao, 2014;Rodriguez-Iturbe et al., 2001;Lazo et al., 2019). However, the effect on water yield by revegetated areas is debatable and depends on different conditions (van Dijk & Keenan, 2007). Therefore, there is a growing interest in development of monitoring methodologies to improve our knowledge of these processes (Krause et al., 2015;Robinson et al., 2008).
The soil water content (SWC) comprises information regarding the interaction between climate, vegetation, and soil (Rodriguez-Iturbe et al., 2001;Vereecken et al., 2014). However, SWC is spatially and temporally highly variable (Western et al., 2002). Remote sensing of SWC can provide valuable spatial information of SWC but only on the top few centimeters of the soil; other methods, such as time domain reflectometry and neutron probes, are limited in their support volume. In contrast, geophysical methods, such as ground penetrating radar, electromagnetic induction (EMI), and electrical resistivity, can be used for monitoring subsurface water and solute dynamics at a range of temporal and spatial scales (Binley et al., 2015).
The link between soil electrical conductivity (σ) and SWC has been the focus of attention for some time. Gardner (1898) first proposed the use of electrical conductivity for inferring SWC. Although σ is strongly influenced by SWC, it is also affected by other factors, such as soil texture, temperature, and pore water electrical conductivity (e.g., Friedman, 2005), necessitating the development of local (site-specific) relationships between σ and SWC. Binley and Slater (2020) provided a comprehensive analysis of the properties and states of soil that influence electrical conductivity. In Section 2, we discuss the relationship between σ and SWC in detail, and in the context of the current study.

Core Ideas
• An increasing trend in apparent electrical conductivity (ECa) corresponds with the increase in soil water content (SWC) under four typical landcovers. • Linear relationships between SWC and ECa were established using the cumulative sensitivity model. • Elevated pore water conductivity effects the relationship between conductivity and SWC.
The EMI method measures the apparent bulk electrical conductivity of the soil (σ a ), which is the depth weighted average value of the σ, with no requirement to establish any contact with the soil surface. The apparent conductivity is an integrated measurement of electrical conductivity that is governed by the depth-sensitivity pattern of the specific measurement. Electromagnetic induction is a relatively mobile technique allowing the measurement of σ a over large scales (Abdu et al., 2008;Robinson et al., 2012). Doolittle and Brevik (2014) review the use of EMI measurements for qualitative mapping of soil properties and soil water processes. A number of studies have illustrated the potential and challenges of the EMI method for estimation of SWC over large areas by establishing relationships between σ a and SWC (Altdorff et al., 2018;Calamita et al., 2015;Martínez et al., 2020;Martini et al., 2017;Nagy et al., 2013;Robinson et al., 2012). Although the σ a -SWC relationship can indicate the integrated state of the soil water, a detailed description of the soil water state with depth is limited (Corwin & Rhoades, 1982;Hendrickx et al., 2002). Modern EMI devices are manufactured with multiple coils and multiple frequencies, enabling the simultaneous measurement of σ a from multiple effective depths. This permits the inversion of the measured σ a values in order to obtain the "real" soil conductivity, σ. Previous studies suggested a number of approaches to establish the σ-SWC relationship under field conditions for different soil types (J. Huang et al., 2016Huang et al., , 2017. They used σ values derived from inversion of the σ a data and related these to observed SWC values. The major drawback of the inversion solution is nonuniqueness, that is, multiple solutions for the same dataset. To encourage unique solutions and reduce some uncertainties, various approaches were suggested, such as regularization or joint inversions of geophysical datasets (Constable, 1987;. Recently, Robinet et al. (2018) reported on difficulties to invert σ a for the establishment of in situ σ-SWC relationships. Instead, they utilized a σ a forward modeling approach to develop field-based σ-SWC relationships.
Given the potential value of EMI for mapping variation in soil water and the need to understand the effect of land  (Liu & Shao, 2016;Zhao et al., 2017). Liu and Shao (2016) showed that the vegetation type significantly controls the vadose zone water dynamics. Furthermore, Zhao et al. (2017) analyzed a 10-yr record of soil water variability under different land covers and revealed high temporal variability (coefficients of variation up to 40% to depths of 4 m) under purple alfalfa and peashrub covers, which reflect the significant water demands by these vegetation types. Earlier studies (e.g, J. Li et al., 2008) have shown that water uptake under these vegetation types can extend to several meters' depth. From the investigation of Zhao et al. (2017), the millet and soybean (and fallow) land covers seem to be the most sustainable in this environment. Therefore, the first objective of this study was to explore the capability of using σ a , measured by EMI, to assess water sustainability of particular land covers. The second objective was to explore the σ-SWC relationships in the deep vadose zone under the different land covers. Most previous soil water-EMI studies have targeted relatively shallow variation in electrical conductivity; here, we study variation in soil water and σ to depths of 4 m.

Study site
This study was conducted at the Shenmu Soil Erosion and Environment Experimental Station (38˚47′46″ N, 110˚21′55″ E) on the northern LPC. The mean annual air temperature is 8.4˚C, the annual reference evapotranspiration (ET 0 ) is 1,020 mm, and the average annual precipitation is 437 mm, 70% of which falls from July through October (climate records are presented in Supplemental Material). Significant soil erosion driven by wind and rainfall in this region motivated the implementation of a large-scale vegetation restoration, the "Grain to Green" project, to improve soil stability Y. Jia & Shao, 2014). Since 1999, many farmlands were converted into forest and grassland, mainly in areas where slopes exceed 15˚ (Liang et al., 2015). Throughout the replantation project, nonindigenous and indigenous vegetation were introduced to the region . The study site was established to understand the effect of introducing different cover types in the LPC. Experimental data has indicated that the nonindigenous vegetation appear to have excessive demands on soil water, keeping the soil under dry conditions and limiting soil water replenishment, in addition to reducing F I G U R E 1 Layout of the four plots in the Shenmu Research site. The lower part of the photograph is the downslope. The black dots in the schematic show the locations of neutron probe and electromagnetic induction measurements aquifer recharge. Therefore, the sustainability of some introduced land cover types is in question (Liu & Shao, 2016;Zhao et al., 2017). Four adjacent plots (61 × 5 m) were established in 2004 on slopes with a uniform gradient (12-14˚) ( Figure 1). To test the effect of different vegetation types on the dynamics of soil water, three vegetation covers were introduced: "shrub" (Korshinsk peashrub), "grass" (purple alfalfa), and "crop" (2-yr rotation of millet and soybean). A "fallow" plot was also created. This was cultivated until 2004, and subsequently abandoned with no further disturbance. Different vegetation types grow over this plot. For the crop plot, the soybean-millet were sowed during May and harvested in October. After harvest, the crop plot emained clear of vegetation until following May. Both crops were fertilized with 120 kg N ha −1 and 60 kg P 2 O 5 ha −1 annually, following the recommendation of the local agriculture service. The peashrub were planted at a planting spacing of 70 × 70 cm, then left alone to grow naturally, and alfalfa were planted with a row spacing of 50 cm in 2004. The aboveground parts of the alfalfa were cut in the beginnings of July and October every year. Note that the plots are rainfed and no irrigation is applied. In order to maintain consistency with previous studies at the site, we adopt the same labelling here: shrub (SL), grass (GL), fallow (FL), and crop (CL) (Figure 1).
Neutron-probe access tubes to 4-m depth were installed along 11 points in the centerline of each plot, at 5-m intervals ( Figure 1). A previous study (Liu & Shao, 2016) presented analyses of soil samples at the site, indicating similarity in soil physical properties between the plots. The soil is a Calcaric Regosol (Food and Agriculture Organization-United Nations F I G U R E 2 Profiles of particle size distribution for two locations at the site, showing little spatial variability in textural properties. The locations of the shrub and fallow plots are shown in Figure 1 Education, Scientific and Cultural Organization), developed from low fertility loess. The soil has weak cohesion, high infiltrability, low water retention, and is prone to erosion (Fu et al., 2010). The soil texture is composed of 11−14% clay, 30−45% silt, and 45−51% sand (Liu & Shao, 2016) and can be classified as loam. Figure 2 shows example particle size distribution data from two 3-m deep sampling points at the site. The texture profiles show remarkable similarity over 3-m depth; from these and other profiles measured at the site, the soil texture spatial variability is insignificant. As part of a regional deep vadose investigation, a borehole was drilled to bedrock at 60-m depth in Shenmu (X. Jia et al., 2018). Further, there were observations of bulk density from samples extracted from the deep vadose zone (Supplemental Figure S2). These observations reveal an increase in bulk density over the top 4 m of the profile.

Data collection
Soil water content and σ a measurements were carried out during 3 d in August and September 2017 ( Figure 3). All measurements of SWC and σ a were conducted at each of the four plots, at the 11 locations in the centerline of each plot. The SWC measurements were made using a CNC503DR Hydro probe neutron probe (Beijing Super Power Company). Neutron counts were taken at an interval of 0.1 m in the upper 1 m and at 0.2-m intervals over 1-4 m. Thus, in total there are 3,300 SWC measurements. Apparent electrical conductivity measurements were made using the CMD Explorer (GF Instruments) EMI device, positioned at 1 m above ground F I G U R E 3 Daily rainfall between July 2017 and October 2017.
The arrows indicate when the soil water content and electromagnetic induction surveys were conducted level and orthogonal to the neutron probe tube. The instrument is 5-m long and has a 10-kHz transmitter coil and three receiver coils at different spacing from the transmitter (1.48, 2.82, and 4.49 m). The accuracy of measurement is ±4% at 50 mS m −1 (GF Instruments). The instrument is used in two types of coil orientation: horizontal coplanar (HCP) and vertical coplanar (VCP). Thus, the EMI device allows the collection of σ a from six different effective depths. In total, there are 792 measurements of σ a . Field tests were conducted to confirm negligible effect of the neutron probe access tube on the measurements when carried out 1 m above ground level. If EMI measurements are made at ground level and assuming relatively uniform electrical conductivity, it is normal practice to assume that the cumulative sensitivity patterns can be expressed, for VCP and HCP orientation, as follows (McNeill, 1980): F I G U R E 4 Cumulative sensitivity functions for vertical coplanar (VCP) and horizontal coplanar (HCP) orientations with instruments located 1 m above ground level. Arrows are positioned at the depth of investigation for a given coil spacing (s) and where s is the transmitter receiver coil spacing (1.48, 2.82, or 4.49 m).
In Equations 1 and 2 the cumulative sensitivity will be, by definition, unity at the ground surface. As discussed by Morris (2009), measurements made with the coils above ground level result in a modified cumulative sensitivity pattern, as shown in Figure 4 for measurements made 1 m above ground level. As is common practice for EMI measurements, adopting a definition of the depth of investigation (DOI) as the depth over which 70% of the signal is sensitive, then for the VCP orientations, we can compute a DOI of 2.7, 3.4, and 4.5 m for the three-coil spacing, and a DOI of 3.1, 4.6, and 6.9 m for the HCP orientation ( Figure 4).

Establishment of a relationship between SWC and σ
The development of a relationship between SWC and σ is required in order to convert the observed EMI data to SWC. Numerous models have been developed to relate σ to SWC. Many originate from early oil reservoir studies (e.g., the wellestablished approaches of Archie [1942] and Waxman and Smits [1968]); several approaches have targeted soils (most notably, Rhoades et al. [1976]). Models range from purely empirical, to semiempirical and physics-based. Laloy et al.
(2011) documents a valuable comparison of a range of models for soils, using the term "pedo-electrical" model to differentiate this from the classical petrophysics terminology.
Despite the range of approaches, the general structure of a σ-SWC model is that there should be a conducting term for the pores and a parallel contribution from conduction along the particle surface ("surface conduction"), which is intuitively linked to the proportion of fine particles, often based on clay content (see, for example, Revil and Glover [1998]). Laloy et al. (2011) show, from their comparison, that a volume averaging approach, used by , was the most effective at fitting their experimental data. This model can be written as follows: where F is the formation factor, σ f is the fluid electrical conductivity, θ is the SWC, ϕ is porosity, n is a parameter that is controlled by the texture of the media, and σ s is the surface electrical conductivity. The formation factor (F) is also a function of the soil texture and porosity, typically expressed as ϕ −m , where m is the commonly named cementation exponent. A number of studies have shown that a simple linear relationship can be established between water content and electrical conductivity (e.g., Calamita et al., 2012;Michot et al., 2003;Robinet et al., 2018), which is clearly equivalent to assuming n = 1 in Equation 3. Following this, we may write: where, if adopting Equation 3, the coefficients are as follows: To convert the σ from Equation 4 to σ a , the forward solution of the cumulative sensitivity model is utilized, following the approach of Robinet et al. (2018). The EMI instrument measures the bulk σ a , which, using the cumulative sensitivity functions in Equations 1 and 2, is related to σ(z). Assuming a series of layers, where the middle of each layer is the SWC depth measurement, with conductivity σ i (i = 1,2,3. . . M), the apparent conductivity for a given coil spacing (s) and orientation, can be expressed as follows: where M is the lowest layer. In this study, we have SWC observations to 4-m depth, and so the value of σ M is assumed to represent the electrical conductivity at greater depths.
The approach adopted involved taking, for all land cover types, measurements of SWC at 25 depths, and converting these to six apparent conductivities (three coil spacings, two orientations) for the 11 locations on three dates using a given value of a and b in Equation 4. The optimum values of a and b that minimize the RMSE of a sample size N, are given by the following: where σ a (obs) are the observed apparent conductivities and σ a (predicted) are the predicted values for a given a and b. The optimization was carried out using the fminsearch function that is available on the Matlab optimization toolbox (Math-Works, 2015). This function uses the Nelder-Mead simplex algorithm (Lagarias et al., 1998).

Unsaturated water flow and solutes transport modelling
For the current study, there are no measurements of pore water electrical conductivity. To address this, the Richards equation and the advection-dispersion equation were used to simulate the accumulation of chloride in the vadose zone of the four land covers.
We implemented a calibrated unsaturated water flow model that was calibrated to long term data measured at the study site (Bai et al., 2020). For a detailed description of the model calibration and validation results, see Bai et al. (2020). The unsaturated water flow is described by the Richards equation, as follows: where ψ is the matric potential head (L), θ is the volumetric water content (L 3 L −3 ), t is time (T), z is the vertical coordinate (L), K(ψ) (L T −1 ), the unsaturated hydraulic conductivity function, is a function of the matric potential head, and S is a root water-uptake sink term (L 3 L −3 T −1 ). The Richards equation was solved numerically by using the Hydrus 1D code (Šimůnek et al., 2008). Simulation of the root water uptake rate (the sink term) was conducted according to the model suggested by Feddes et al. (1978) (2020) were implemented in the model. The advection-dispersion equation was applied to describe the unsaturated chloride transport in the unsaturated zone of the different land covers, as follows: where C Chloride (M L −3 ) is chloride concentration in the pore water solution, D (L 2 T −1 ) is the hydrodynamic dispersion coefficient, and q (L T −1 ) is the water flux. Turkeltaub et al. (2018) suggested a representative value of 7.5 cm for the longitudinal dispersivity in the LPC. This value was calculated according to sampled chloride and nitrate vadose zone profiles across the LPC. Atmospheric boundary conditions with a surface layer (assuming zero for ponding depth at the soil surface) were prescribed at the upper boundary (land surface) as rain, leaf area index (LAI), potential evapotranspiration (ET 0 ), rain chloride concentrations, and the minimum allowed pressure head at the soil surface (Šimůnek et al., 2008) at a daily temporal resolution. To estimate the potential ET 0 values, reference evapotranspiration (ET ref ) values were multiplied with the single crop coefficients (K c ). The K c values for millet (crop), grass, alfalfa (shrub), and bare soil were based on R. Allen et al. (1998). The chloride concentration in the rain was 1.7 mg L −1 (T. Huang et al., 2013).
The vertical root density distributions for the different covers were implemented according to the root profiles that were published by Bai et al. (2020). For the crop plot, a linear root distribution was assumed through ∼50-cm depth (Bai et al., 2020). Under the grass and the shrub plots, the roots were distributed over 400 and 270 cm, respectively (Bai et al., 2021).
For the root distribution profiles, see Supplemental Figure S2 (Bai et al., 2020). The increase in LAI during the growing season for millet, grass, and alfalfa was estimated with the model of Leenhardt et al. (1998), where the increase in LAI is assumed a function temperature according to the following: where LAI max is the maximum LAI of the crop, T i (˚C) is the sum of temperature at the inflection point of the curve, and b is a curvature parameter. The LAI max and the b parameters were estimated using the temperature database and reported LAI curves (

Spatiotemporal variability of SWC
In Figure 5, SWC profiles for all the survey dates are shown. The movement of a drying front can be seen between the first two survey dates, followed by subsequent wetting in the third survey (following the late August rainfall event). The profiles show similarity for a given land cover type (limited horizontal variability was observed along the slope) and also the reduced soil water content at depth for the grass and shrub cover type due to the greater water demands of such cover and the deep root penetration, which is estimated to be greater than 4-m depth (Zhao et al., 2017). These are consistent with the longterm study at the site of Zhao et al. (2017), which also showed that water percolates to deeper parts of the vadose zone under the crop cover compared with the other land covers. The significant differences in SWC between the land covers, which are subjected to the same climatic conditions, and uniform soil texture (Figure 2), highlights the potential negative effect on SWC due to the planting of vegetation that is unsustainable in the LPC region (Fang et al., 2016;Liu & Shao, 2016;Zhao et al., 2017). Figure 6 summarizes the SWC data for the three survey dates, adding further illustration of the effect of land cover type on soil water availability.

σ a measurements
The apparent conductivity measurements are summarized as box and whisker plots in Figure 7. The vertical coplanar and horizontal coplanar configurations with similar depths of investigation show consistency. The plots indicate an increasing conductivity with depth across all land cover types and a clear contrast in apparent conductivity for the four land covers, particularly for the measurements over greater depths. There is a clear similarity between land cover contrasts in SWC ( Figure 6) and apparent conductivity (Figure 7), particularly when we compare the shrub and grass cover with the fallow and crop cover. Robinson et al. (2008) reported on similar variability in σ a for different vegetation species. They related the ranking in σ a values to the relationship between plant communities and soil types. However, the plots in the current study are adjacent, and major differences in soil texture are not observable ( Figure 2). Therefore, it can be assumed that the ranking of σ a is probably dominated by the water conditions in the vadose zone and influenced by the water demand of the vegetation cover. We note that some discrepancy between crop and fallow cover might be related to the fertilizer application for the crop (Zhao et al., 2017). Similar observations were reported elsewhere (Calamita et al., 2015). Nevertheless, the σ a values obtained at the crop and fallow are generally higher to those obtained on the shrub and grass plots, which are known to experience bigger demands on soil water status.
Further interpretation was suggested in previous studies regarding the statistics of the σ a values (Calamita et al., 2015;Robinson et al., 2008). For the following interpretation, two assumptions are made: (a) the σ a measurements reflect the soil water conditions (as was shown above) and (b) vegetation under optimal conditions would show a low CV of the SWC (Robinson et al., 2008;Zhao et al., 2017). Robinson et al. (2008) showed empirically that highly skewed σ a distributions and high CV values indicate that vegetation grows outside their optimal environment. The long-term investigation (>10 yr) of SWC time series measurements at the study site by Zhao et al. (2017) revealed a decreasing trend in the CV of SWC as follows: CL < FL < GL < SL. Similarly, a high CV was calculated for the σ a measurements under the shrub cover (Table 1). Thus, following the presented analysis, we observe the same ranking of variation in apparent conductivity for the deeper measurements (see Table 1). Based on their observations of SWC, Zhao et al. (2017) concluded that the Korshinsk peashurb is not sustainable in terms of SWC F I G U R E 5 Soil water content profiles in the four plots on the three survey dates. The solid line is the median profile; the shaded region shows the 1st and 3rd interquartile range

-σ-SWC relationship
The measurements obtained in the current study enabled us to explore relationships between SWC and σ at the study site. As stated earlier, the approach involved compiling an aggregate dataset for the site, rather than applying the model search for different cover types, because there is likely to be a limited range of the data to perform the latter. Table 2 reports the linear coefficients a and b (Equation 4) obtained using the F I G U R E 7 Box and whisker plots of the apparent electrical conductivity (σ a ) measurements from six effective depths, which were obtained over the different land covers. The horizontal line shows the median soil water content, the box shows the 2nd and 3rd quartile range, and the whiskers show the 1st and 4th quartiles. DOI, depth of investigation; HCP, horizontal coplanar; VCP, vertical coplanar F I G U R E 8 Estimated vs. observed σ a for all crop cover types using the relationships in Table 2. The black line in each plot is the 1:1 relationship. DOI, depth of investigation; HCP, horizontal coplanar; VCP, vertical coplanar optimization process adopted here. The fit for each model is similar, approximately 1 mS m −1 , which is within the accuracy of the instrument. However, power law models were also tested, and these models did not provide any further improvement in performance, which is in line with previous studies (Calamita et al., 2012;Michot et al., 2003). In addition, Robinet et al. (2018) noted that a better linear relationship between σ a and soil moisture could be obtained by using σ a observations from their deeper sensed EMI configuration. Figure 8 shows the model fit for the six coil orientations, plotted to differentiate the four cover types. The grass and shrub cover data show the greatest departure from the 1:1 apparent conductivity, particularly at greater depths. This may be related to the relatively high salinity conditions that might prevail under these cover types due to elevated evapotranspiration. Figure 9 shows the variation in σ-SWC relationship parameters with depth, using a nominal depth as that at which the cumulative sensitivity function CS(z) equals 0.5, that is, the depth over which 50% of the EMI measurement is sensitive. Note that this "halfdepth" is a nominal depth used for illustration, although it is sometimes used to guide EMI survey design (Morris, 2009). A consistent increase with depth in both a and b is seen for both coil orientations. From Equation 5, an increase in a could account for (a) increase in pore water conductivity, (b) reduction in porosity, and (c) increase in cementation exponent, m. An increase in b can also be attributed to a reduction in porosity and an increase in m, in addition to an increase in surface conductivity. The observations of bulk density reveal an increase in bulk density over the top 4 m of the profile (Supplemental Figure S2). Assuming a particle density of 2.65 g cm −3 , this equates to a reduction in porosity from 0.50 at 0.5-m depth to 0.44 at 4.5-m depth, that is, a reduction by 10%. Assuming a cementation exponent m = 2 because most porous sediments have cementation exponents between 1.5 and 2.5 (Cai et al., 2017), such a reduction in porosity can only account for a 30% increase in a. Therefore, it would appear that pore water conductivity variation with depth is a primary driver of the change in model coefficients with depth.
Developing relationships between soil water content and electrical conductivity is constrained by the influence of a range of properties, making the use of universal models somewhat limited without local calibration. GF Instruments report that the measurement accuracy for the CMD-Explorer is ±4%, and the measurement accuracy of the CNC503DR Hydro neutron probe is also reported to be ∼4%. The RMSE values of all the models are ≤10% than the mean of the measurements. Furthermore, the R 2 and the RMSE values that were reported here are comparable to previously published calibrated F I G U R E 9 Variation in σ-soil water content relationship parameters with depth. CS(z), cumulative sensitivity function models (Calamita et al., 2015;Coppola et al., 2016;Robinet et al., 2018;Robinson et al., 2012;Tromp-van Meerveld & McDonnell, 2009). Therefore, for the dataset studied here, a linear σ-SWC model was considered to be suitable. Although we recognize that given a wider range of soil water, a more nonlinear function may be suitable (e.g., Robinet et al., 2018). Despite this, our results show that qualitative mapping of the effect of soil water reduction from excessive crop water uptake is potentially feasible in the LPC.

Accumulation of chloride in the vadose zone
Simulated and observed SWC are plotted in Supplemental Figures S4 and S5. Note that the soil hydraulic functions and root vertical distributions were prescribed according to Bai et al. (2020), and no further adjustments were conducted. The RMSE and MAPE were similar and low for all the plots (Supplemental Figure S5), while the NSE value was different for each plot and showed higher efficiency for the crop and grass plots (Supplemental Figure S5). These results were comparable to the analysis presented by Bai et al. (2020). Thus, the model can be considered to adequately describe the SWC dynamic under the investigated plots (Bai et al., 2020). By including the longitudinal dispersivity in the model, the transport of chloride (of rainfall origin) in the vadose zone under the different covers is revealed. Figure 10 presents the calculated chloride concentrations at the end of the model runs (20 Sept. 2017). The simulated chloride concentrations under the alfalfa are nearly two times higher compared with the fallow and six times that with the crop (millet) (Figure 10). Previous studies in the LPC reported soil profile information that are comparable to the simulated chloride. T. Huang et al. (2013) showed an intensive accumulation of chloride under alfalfa (about 6.5 times higher than under rain-fed winter wheat crop). Additional studies (Gates An earlier study by Hilhorst (2000) suggested that under dry conditions, the σ a measured by EMI is more affected by the increase of pore water conductivity and less closely associated to SWC. Furthermore, in semiarid areas, the climatic forcing has a major effect on deep drainage. The level of deep drainage intensity would define the build-up of salts and their distribution in the vadose zone (Scanlon et al., 2010). Recently, several studies have indicated that the pore water conductivity distribution in the vadose zone should be considered when establishing an in situ σ-SWC relationship in semiarid areas (Cassiani et al., 2016;Moreno et al., 2015). However, currently there are no reported field studies of in situ simultaneous measurements of SWC, σ a , and pore water conductivity under semiarid conditions. The build-up of salts and associated soil salinity in the LPC vadose zone has received surprisingly little attention.

CONCLUSION
The measurements of SWC in deeper parts of the vadose zone at large scales is challenging. Geophysical methods such as the EMI approach might facilitate a bridge between processes observed locally and at larger scales. Here, EMI was applied to measure apparent electrical conductivity over six effective depths in four plots covered by typical land cover types (shrub, grass, fallow, and crop) in the north of the LPC. Soil water content was measured with neutron probes from the ground surface to a depth of 4 m. The unique loess environment in the LPC, with its characteristic deep soils and relatively insignificant soil variability, reduces the effect of soil texture variation on EMI readings to minimum. Moreover, for this particular study, soil textural variation is insignificant and can be neglected. The similarity of the soil texture between all plots enabled a focus of investigation on the potential influences of different cover types on the spatiotemporal variability of SWC and apparent electrical conductivity. An increasing trend in σ a values, SL < GL < FL < CL, corresponds with the increase in average SWC in the plots. Moreover, σ a values that were measured in the shrub covered plot show a relatively high variability, which is consistent with documented variability of SWC for soils under this vegetation, indicating unsustainable water conditions in the vadose zone.
Linear relationships between soil water content and specific-depth σ under the different land covers were established. The σ values were estimated using the SWC observations, assuming a linear relationship between these variables. The analysis reveals a change in model parameters with depth. Textural variation is apparently negligible (to 3-m depth, at least); however, such variation in model parameters may be attributed, in part, to changes in bulk density. Increases in pore water electrical conductivity are hypothesized as a primary cause of the depth dependency of the σ-SWC model parameters. Simulations of chloride profiles support the hypothesis that contrasts in pore water electrical conductivity could exist under different crop types. Elevated pore water conductivity beneath shrub and grass covers would imply even greater significance of the soil water content because these two cover types exhibit lower apparent conductivity than the other two cover types. To improve SWC prediction from EMI observations, pore water conductivity should be measured. Nevertheless, the results presented here illustrate how excessive water demands of Korshinsk peashrub and purple alfalfa at the study site are revealed by their lower apparent conductivity and (for the case of the shrub cover at least) their high variation in apparent conductivity. Our EMI dataset reveals an immense potential for mapping-qualitatively at leastareas of the LPC that are vulnerable to excessive vegetation demands, and hence, unsustainable land cover.

A C K N O W L E D G M E N T S
This work was supported by the UK Natural Environment Research Council (grant ref NE/N007409/1 awarded to Lancaster) and the National Natural Science Foundation of China (41571130081). We appreciate Dr. Yuanzhi Wu from Linyi University for her generosity in providing the particle size distribution data.