Soil Water Extraction Monitored Per Plot Across a Field Experiment Using Repeated Electromagnetic Induction Surveys

Soil water (θ) dynamics are important parameters to monitor in any field-based drought research. Although apparent electrical conductivity (ECa) measured by electromagnetic (EM) induction has been used to estimate θ, little research has shown its successful application at the plot-scale for evaluating crop water use. An EM38 conductivity meter was used to collect time-lapse ECa data at the plot scale across a field cropped with 36 different chickpea genotypes. An empirical multiple linear regression model was established to predict θ measured by neutron probes and depth-specific electrical conductivity (σ) generated by a 1-D EM inversion algorithm. Soil water dynamics and movement were successfully mapped with a coefficient of determination (R2) of 0.87 and root-mean-square-error of 0.037 m3 m−3. The rate of soil drying varied with depth and was influenced by chickpea growth stages and genotypes. The results were also used to evaluate the differences in soil water use and rooting depths withinand across-plant species and during the growth stages. Coupled with physiology measurements, the approach can also be used to identify mechanisms of drought tolerance in the field and screening for effective water use in crop breeding programs.


Introduction
Monitoring soil water content at different depths and its dynamics can be used to determine plant water use and hence, water use efficiency.The ability to do this at the plot level on a meaningful scale has been considered impractical by plant breeders [1], so that selection for drought-tolerant and more effective water use in the field has been limited [2].Measuring soil volumetric water content (θ) in the laboratory using the thermogravimetric method is labour-intensive [3].Indirect methods to estimate θ in the field have been used, such as neutron probes, time domain reflectometry, capacitance probes, and electrical resistivity tomography [4,5], but these instruments need to be installed into the soil profiles, which is time-consuming and expensive, and hinders their potential applications for estimating θ at depths, across larger spatial extents and hundreds of plots.In addition, some agricultural practices (e.g., ploughing) may not be possible if sensors are permanently installed in the soil (especially at shallow depth).
Recently, non-invasive electromagnetic induction (EMI) instruments have been successfully used in the field to estimate and map θ within the root-zone and vadose zone [6,7].Estimation of θ with EMI is possible because the measured soil apparent electrical conductivity (EC a ) is directly

Geonics EM38 Configuration
An EM38 (Geonics Ltd., Mississauga, ON, Canada) conductivity meter provides ECa measurements in the root zone.The depths of exploration (DOE) of the EM38 are 1.5 m in the vertical (EM38v) and 0.75 m in the horizontal (EM38h) modes of coil orientation when the instrument is placed on the ground surface [26].When the height of the instrument is raised, ECa decreases.ECa data collected at multiple heights can be used to improve modelling of the shallow earth structure [26].This was the approach of various previous studies [27][28][29] and it was adopted in this study.

Time-Lapse EM38 Surveys
To establish an empirical calibration model to predict θ dynamics, ECa data were collected across 20 plots in the irrigated section on 1 day before (1 October 2017) as well as 2 (5 October 2017), 3 (6 October 2017), 4 (7 October 2017), and 5 (8 October 2017) days after a 25-mm irrigation event (Figure 1).ECa data were also collected across 20 rainfed plots on the same days (Figure 1).ECa data were measured in both EM38v and EM38h modes and at five different heights (i.e., 0 m, 0.2 m, 0.4 m, 0.6

Geonics EM38 Configuration
An EM38 (Geonics Ltd., Mississauga, ON, Canada) conductivity meter provides EC a measurements in the root zone.The depths of exploration (DOE) of the EM38 are 1.5 m in the vertical (EM38v) and 0.75 m in the horizontal (EM38h) modes of coil orientation when the instrument is placed on the ground surface [26].When the height of the instrument is raised, EC a decreases.EC a data collected at multiple heights can be used to improve modelling of the shallow earth structure [26].This was the approach of various previous studies [27][28][29] and it was adopted in this study.

Time-Lapse EM38 Surveys
To establish an empirical calibration model to predict θ dynamics, EC a data were collected across 20 plots in the irrigated section on 1 day before (1 October 2017) as well as 2 (5 October 2017), Soil Syst.2018, 2, 11 4 of 17 3 (6 October 2017), 4 (7 October 2017), and 5 (8 October 2017) days after a 25-mm irrigation event (Figure 1).EC a data were also collected across 20 rainfed plots on the same days (Figure 1).EC a data were measured in both EM38v and EM38h modes and at five different heights (i.e., 0 m, 0.2 m, 0.4 m, 0.6 m and 0.8 m).EC a data were collected close to the centre of each plot where an aluminium access tube was installed.
To compare the water use of chickpea associated with different treatments and genotypes, repeated EM38 surveys were also measured on all 288 plots at the same heights and 1 (23 October 2017), 3 (25 October 2017), 9 (31 October 2017) and 12 (3 November 2017) days after a rainfall event (~12.6 mm of water).
Temperature has previously been reported to influence the temporal and spatial pattern of data suggesting the need for temperature correction [12,22,[30][31][32][33][34][35].Based on our previous experiment [36], we inferred that the change of soil temperature in a Vertosol with large specific heat capacity and covered with an enclosed crop canopy would be less than 2 • C at 0-0.2 m during our experiment and the change would decrease with increasing depth.This would lead to a shift of EC a of approximately 3.8% in the field due to diurnal soil temperature fluctuation at 0-0.2 m and a shift of less than 3.8% below 0.2 m.Based on the minor impact of this estimation, we chose not to correct the EC a based on soil temperature.
Because diurnal drifts of EMI instruments due to ambient temperature have also been reported [36,37], preliminary EC a measurements were taken before the irrigation event at three selected locations at different times of the day when the temperature ranged between 20 and 35 • C. Diurnal drifts were found to be negligible (<1 mS m −1 ) once the EM38 meter was calibrated according to the user manual prior to the survey.All the EM38 surveys in this study were completed within this temperature range and therefore, the measured EC a data were assumed to be stable.

Collection of Neutron Probe Measurements
To calibrate the EM38 EC a data, neutron probe readings were also measured in the 20 rainfed plots and the 20 irrigated plots on the respective days when the EM38 surveys were carried out.The neutron probe readings (CPN ® 503DR Hydroprobe, CPN International, Concord, CA, USA) were collected in the access tube at 0.1, 0.2.0.3, 0.4, 0.5, 0.6, 0.8, 1.0 1.2 and 1.4 m depths.A pre-defined universal calibration formula was employed to convert the neutron counts to soil volumetric water content (θ): θ = (Neutron counts − 7863)/182.9 (1)

Inversions of EM38 Data
Because the 40 calibration plots were randomly located across the whole field (Figure 1), a 1-D inversion algorithm was employed.The inversion algorithms [38] were embedded in the EM4Soil Version 3.02 (EMTOMO 2017, http://www.emtomo.com/).In brief, the algorithm aimed at generating estimates of depth-specific electrical conductivity using the cumulative function of EM field in the soil [26].This study applied an initial model which included 11 layers with the middle depths of the first 10 layers equaling to 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.8, 1.0, 1.2, 1.5 m and the 11th layer at infinity.

Predicting θ Using an Empirical Model
An empirical calibration model was required to convert σ to θ.Given the dependence of θ on σ and the soil depth (Figure 2a,b), a multiple linear regression (MLR) model was fitted to predict θ using σ and the soil depth.The model parameters were fitted using ordinary least squares with JMP 10.2 (SAS Institute Inc., Cary, NC, USA).Once the model was fitted, it was applied to the whole field to predict θ at different depths across the 288 plots over the 11-day period after the rainfall event.

Estimating θ Dynamics in 3-Dimensions
To better visualise and interpret θ dynamics in 3-D across the field, the change in predicted θ (∆θ) generated by MLR at various depths across the field was calculated based on the first day after irrigation.The water flux (J Depth , mm) within different depth intervals down to 1.2 m (negligible

Estimating θ Dynamics in 3-Dimensions
To better visualise and interpret θ dynamics in 3-D across the field, the change in predicted θ (∆θ) generated by MLR at various depths across the field was calculated based on the first day after irrigation.The water flux (J Depth , mm) within different depth intervals down to 1.2 m (negligible changes of θ below 1.2 m) and between 0 and 1.2 m across the rainfed and irrigated fields were also calculated.

Comparing Water Use of Different Genotypes
To compare the water use of various chickpea genotypes, we fitted a linear mixed effect model as follows: Herein, ∆θ was the change of θ within 0-1.2 m across the 288 plots during the first 11 days after the irrigation event.The fixed effects included the following variables: X and Y were the distance between the plots as shown in Figure 1; Genotype contained 36 values and were labelled from 1 to 36 due to confidentiality issue; Treatment had two values (i.e., rainfed and irrigated); Genotype × Treatment referred to the interaction between genotype and treatment.η referred to the spatial random effect and was modelled with a semi-variogram; ε was the error.In this study, the parameters of the linear mixed model were fitted using the restricted maximum likelihood (REML) [39].The geoR package [40] in R was used to fit the model parameters.
In this study, the mean water flux (J Depth , mm) within different depth intervals was also calculated for several selected genotypes (with largest and smallest fixed effects).These values would be used to infer the rooting depths and root activities during the experiment.Herein, we did not fit a linear mixed effect model for every genotype because we did not have a sufficient number of sites to estimate a variogram.

Correlation Between θ and EC a
Figure 2 shows the plot of measured θ at two depths versus EC a on different days.Figure 2a,b show the plots of measured θ at 0.1-0.2m versus EC a measured by EM38h and EM38v, respectively, before irrigation (close to permanent wilting point).In general, EC a increased with increasing θ but the correlations were weak (R 2 = 0.26 and 0.29).Not surprisingly, similar patterns were observed between irrigated and rainfed plots because the soils in both plots were dry. Figure 2c,d show the plots of measured θ at 0.4-0.5 m versus EC a measured by EM38h and EM38v, respectively, before irrigation.Similar but slightly stronger correlations were found between θ with EC a measured by EM38h and EM38v (R 2 = 0.30 and 0.40).
Figure 2e-f show the plots of measured θ at two depths versus EC a measured by EM38h and EM38v and 2 days after irrigation (close to field capacity).In this case, the correlations between θ with EC a measured by EM38h and EM38v were much stronger (R 2 = 0.72, 0.83, 0.74 and 0.75) than those identified when the soil was dry.The increase in correlation was mainly because there was a broader range in measured θ as well as EC a values when the soil became wetter.
It was also worth noting that the correlations between θ and EC a were different for different depths and different times (soil moisture conditions).Some researchers [23] reported a similar relationship between θ and measured EM38 EC a in Vertosols, and they also found that the relationship varied with soil depth.The difference in this study was because ECa represented the depth-average soil electrical conductivity with different measurement modes (EM38h and EM38v) corresponding to different DOEs.Therefore, different regression models were required to predict q at different depths using the same set of EC a data.Based on the results, it is necessary to apply the inversion algorithms to convert EC a to depth-specific σ and use it for modelling θ.

Correlation Between θ with σ and Soil Depth
Figure 3a shows the plot of measured θ versus estimated σ across 40 calibration plots over 4 days.In general, σ increased with increasing θ.More importantly, it was worth noting that the overall correlation between θ and σ was stronger than those calculated between θ and EC a at different depths and on different days (Figure 2).This was not unexpected because the depth-specific σ has a similar Soil Syst.2018, 2, 11 7 of 17 measuring volume compared to θ while EC a represented the depth-average soil electrical conductivity.Some researchers [16] identified a curve-linear relationship between θ and σ in homogenous loamy soils, which supports these findings.Herein, the soil was similarly uniform but had a medium to high clay content (i.e., Vertosols).
Figure 3b shows the plot of measured θ versus the soil depth.In general, θ increased with the soil depth, which is to be expected because Vertosols normally have a high field capacity (>0.5 m 3 m −3 ) and permanent wilting point (>0.4 m 3 m −3 ).Vertosols can hold a large amount of water at depths even when the shallow soils become drier due to evaporation and transpiration.
Soil Syst.2018, 2, 11 7 of 17 measuring volume compared to θ while ECa represented the depth-average soil electrical conductivity.Some researchers [16] identified a curve-linear relationship between θ and σ in homogenous loamy soils, which supports these findings.Herein, the soil was similarly uniform but had a medium to high clay content (i.e., Vertosols).
Figure 3b shows the plot of measured θ versus the soil depth.In general, θ increased with the soil depth, which is to be expected because Vertosols normally have a high field capacity (>0.5 m 3 m −3 ) and permanent wilting point (>0.4 m 3 m −3 ).Vertosols can hold a large amount of water at depths even when the shallow soils become drier due to evaporation and transpiration.

The Empirical Model of θ
Given the strong correlation between θ with σ and soil depth, a multiple linear regression (MLR) model was established.The model parameters are shown in Figure 3c. Figure 3c also shows 10-fold cross-validation results of the established MLR model.In general, the model performance was good given that Lin's concordance [41] was 0.93, root-mean-square-error (RMSE) was 0.037 m 3 m −3 and mean-error (ME) was 0.000 m 3 m −3 , respectively.Given the good performance of the MLR model, a non-linear regression model was not fitted and compared in this study.
Figure 3c also shows that the MLR model tended to over-estimate θ values < 0.1 m 3 m −3 , which were mostly found in the topsoil.This divergence between measured and predicted θ is most likely caused by the poor performance of the Neutron probe calibration formula for topsoil due to the radiation escaping from the soil surface [42].In addition, this over-estimation of θ could be due to the uncorrected shift of ECa due to soil temperature, particularly within 0-0.2 m.
It should be noted that apart from θ, a number of soil properties also influence ECa, including clay content, soil salinity, the electrical conductivity of the soil solution, and soil temperature [6,22,31,43].In this study, the coefficient of variation (CV) of EM38v and EM38h measured at the ground surface across the 288 plots was small prior to rainfall (5% and 8%, respectively) and it is expected that the CV values will be smaller if there was no crop in the field.These suggest that the

The Empirical Model of θ
Given the strong correlation between θ with σ and soil depth, a multiple linear regression (MLR) model was established.The model parameters are shown in Figure 3c. Figure 3c also shows 10-fold cross-validation results of the established MLR model.In general, the model performance was good given that Lin's concordance [41] was 0.93, root-mean-square-error (RMSE) was 0.037 m 3 m −3 and mean-error (ME) was 0.000 m 3 m −3 , respectively.Given the good performance of the MLR model, a non-linear regression model was not fitted and compared in this study.
Figure 3c also shows that the MLR model tended to over-estimate θ values < 0.1 m 3 m −3 , which were mostly found in the topsoil.This divergence between measured and predicted θ is most likely caused by the poor performance of the Neutron probe calibration formula for topsoil due to the radiation escaping from the soil surface [42].In addition, this over-estimation of θ could be due to the uncorrected shift of EC a due to soil temperature, particularly within 0-0.2 m.It should be noted that apart from θ, a number of soil properties also influence EC a , including clay content, soil salinity, the electrical conductivity of the soil solution, and soil temperature [6,22,31,43].In this study, the coefficient of variation (CV) of EM38v and EM38h measured at the ground surface across the 288 plots was small prior to rainfall (5% and 8%, respectively) and it is expected that the CV values will be smaller if there was no crop in the field.These suggest that the soils are relatively homogenous across the field and these factors are not likely to influence the accuracy of the MLR model established in this study.However, it should be noted that MLR model is site-specific and may not be readily applicable to other areas because of the variation of soil properties in space and time [35,44,45].

Spatial Distribution of Model Residuals Across the Field
Figure 4 shows the contour plots of the spatial distribution of the MLR model residuals (measured θ-predicted θ) at two depth intervals (0.1-0.2 m and 0.4-0.5 m) across the field before and 2 days after irrigation.In general, the residuals were negative (over-estimation) in the rainfed plot and positive (under-estimation) in the irrigated plot.This was most likely because θ in the rainfed plot was larger than that in the irrigated plot due to the low evapotranspiration rate of the mature chickpeas in the rainfed plot (refer to Figures 1 and 2).Furthermore, it was noted that the residuals in the irrigated plot were more negative at 0-0.1 m than at 0.4-0.5 m, which can be due to either the inaccurate estimation of θ from neutron probe measurements close to the ground surface or the uncorrected drift of EC a due to soil temperature fluctuations within 0-0.2 m.
Soil Syst.2018, 2, 11 8 of 17 soils are relatively homogenous across the field and these factors are not likely to influence the accuracy of the MLR model established in this study.However, it should be noted that MLR model is site-specific and may not be readily applicable to other areas because of the variation of soil properties in space and time [35,44,45].

Spatial Distribution of Model Residuals Across the Field
Figure 4 shows the contour plots of the spatial distribution of the MLR model residuals (measured θ-predicted θ) at two depth intervals (0.1-0.2 m and 0.4-0.5 m) across the field before and 2 days after irrigation.In general, the residuals were negative (over-estimation) in the rainfed plot and positive (under-estimation) in the irrigated plot.This was most likely because θ in the rainfed plot was larger than that in the irrigated plot due to the low evapotranspiration rate of the mature chickpeas in the rainfed plot (refer to Figures 1 and 2).Furthermore, it was noted that the residuals in the irrigated plot were more negative at 0-0.1 m than at 0.4-0.5 m, which can be due to either the inaccurate estimation of θ from neutron probe measurements close to the ground surface or the uncorrected drift of ECa due to soil temperature fluctuations within 0-0.2 m.

Predicted θ Dynamics Across the Field
Because the day-to-day change in θ was small, the change in θ (∆θ) of a specific day was plotted compared with the first day after the rainfall event.Figure 5 shows the ∆θ at various depths across the 288 plots during day 1 and day 3 (2-day period) after the rain.
In general, θ was constant at the depth interval of 0.1-0.2m for most parts of the field (∆θ: −0.01-Because the day-to-day change in θ was small, the change in θ (∆θ) of a specific day was plotted compared with the first day after the rainfall event.Figure 5 shows the ∆θ at various depths across the 288 plots during day 1 and day 3 (2-day period) after the rain.
Soil Syst.2018, 2, 11 9 of 17 from the soil as they approached physiological maturity compared with the immature chickpea in the irrigated plots, which were still actively taking up water.Similar patterns in ∆θ were observed at other depths but with a few notable differences (Figure 5b-f).In the northern and central rainfed plots, soils at 0.5-1.0m depths became wetter.This indicated that water infiltrated down the profiles, probably as preferential flows via the cracks in the Vertosols.In the irrigated plots and the southern end of the rainfed plots, soils were drying faster with increasing depth.This suggested that the chickpea root system was more active at these deeper depths when extracting water from the soils.Figure 6 shows ∆θ across the 288 plots during day 1 and day 9 (8-day period) after the rain.In the rainfed plots, soils dried out at 0.1-0.2m depth, remained unchanged at the depth of 0.3-0.4m and became wetter below 0.5 m.This indicated that the evapotranspiration rate of chickpea in the rainfed plots was very low so that the surface drying was mostly caused by evaporation and the subsoil wetting was due to the infiltration of water with time.In the irrigated plots, soils were drying out throughout the whole profile indicating water uptake by chickpea roots across the profile.
Figure 7 shows ∆θ across the 288 plots during day 1 and day 12 (11-day period) after the rain.Compared to the first day after irrigation, θ in the rainfed plots remained unchanged.In the irrigated plots, soil continued to dry out throughout the whole profile.In particular, more water was extracted from the deeper depths (below 0.6 m) compared with the shallow depths (Figure 7d-f).This was not In general, θ was constant at the depth interval of 0.1-0.2m for most parts of the field (∆θ: −0.01-−0.01m 3 m −3 ) except a few plots in the rainfed side of the field that became slightly wetter (∆θ: 0.01-−0.03m 3 m −3 ) while a few plots in the irrigated half became slightly drier (∆θ: −0.03-−0.01m 3 m −3 ) (Figure 3a).This is consistent with the chickpea genotypes in the rainfed section extracting less water from the soil as they approached physiological maturity compared with the immature chickpea in the irrigated plots, which were still actively taking up water.
Similar patterns in ∆θ were observed at other depths but with a few notable differences (Figure 5b-f).In the northern and central rainfed plots, soils at 0.5-1.0m depths became wetter.This indicated that water infiltrated down the profiles, probably as preferential flows via the cracks in the Vertosols.In the irrigated plots and the southern end of the rainfed plots, soils were drying faster with increasing depth.This suggested that the chickpea root system was more active at these deeper depths when extracting water from the soils.
Figure 6 shows ∆θ across the 288 plots during day 1 and day 9 (8-day period) after the rain.In the rainfed plots, soils dried out at 0.1-0.2m depth, remained unchanged at the depth of 0.3-0.4m and became wetter below 0.5 m.This indicated that the evapotranspiration rate of chickpea in the rainfed plots was very low so that the surface drying was mostly caused by evaporation and the subsoil wetting was due to the infiltration of water with time.In the irrigated plots, soils were drying out throughout the whole profile indicating water uptake by chickpea roots across the profile.Based on the results presented in Figures 5-7, it can be concluded that soil water content at the plot scale can be extremely dynamic and wetter/drier areas at one time step could either persist or change in a different direction at the next time step.This is reasonable because the crop water use is highly dynamic, depending on varieties, developmental stages and even previous soil moisture status (hysteresis effect) [46].To validate these assumptions, sensor-based real-time monitoring of leaf turgor pressure and stem water status coupled with plant physiological analysis [47] can be used in the future to value-add to the time-lapse 3-D images of soil water status generated using repeated Figure 7 shows ∆θ across the 288 plots during day 1 and day 12 (11-day period) after the rain.Compared to the first day after irrigation, θ in the rainfed plots remained unchanged.In the irrigated plots, soil continued to dry out throughout the whole profile.In particular, more water was extracted from the deeper depths (below 0.6 m) compared with the shallow depths (Figure 7d-f).This was not unexpected because θ values at the depth of 0-0.6 m were close to the permanent wilting point of the clay soils (~0.15-0.20 m 3 m −3 , figures not shown) while θ in the subsoil was still high (~0.25-0.30m 3 m −3 , figures not shown), so there was still available water for uptake by the chickpea genotypes.

Water Balance Across the Field
Averaged soil water flux (J depth , mm) at different depths and within the top 1.2 m of the soils is presented in Table 1.J depth values deeper than 1.2 m were not shown because θ was relatively constant during the whole experimental period.A net amount of 1.9 mm of water was added to the rainfed plots and 17.0 mm of water was removed from the irrigated plots over 11 days (day 1 to day 12) considering 12.6 mm of water was added to the soil via rainfall (Table 1).Table 1 also shows that water was first (days 3-9) removed from the shallow soil and then (days 9-12) from the deeper soil.Based on the results presented in Figures 5-7, it can be concluded that soil water content at the plot scale can be extremely dynamic and wetter/drier areas at one time step could either persist or change in a different direction at the next time step.This is reasonable because the crop water use is highly dynamic, depending on varieties, developmental stages and even previous soil moisture status (hysteresis effect) [46].To validate these assumptions, sensor-based real-time monitoring of leaf turgor In addition to water use, genotypic and growth-dependent variation in root system traits were detected.Table 3 shows the mean values of water flux (J Depth , mm) within each depth interval for several genotypes during the 11-day period.Genotypes 6 and 34 were used to present the genotypes that were actively extracting water during the experiment and genotypes 36 and 10 were used to present the genotypes that showed least water extraction from the soil.
Mature genotype 6 (rainfed) extracted most of the water between 0.9 m and 1.5 m (<−1.0 mm).By comparison, immature genotype 6 (irrigated) extracted water throughout the whole soil profile with the maximum water extraction between 1.1 m and 1.5 m (<−1.5 mm).This was not the case for genotype 34, which extracted water throughout the whole soil profile during both immature and mature stages.In terms of genotype 36, most of the water was extracted at the soil surface (0-0.2 m) during the mature stage and at the depth of 1.0-1.1 m during the immature stage.This was similarly the case for genotype 10, which extracted most of the water within 0-0.2 m during mature stage and within 0.7-0.8m during the immature stage.differences in water use over short durations and small changes in soil moisture content, as well as different depths.The method has the potential to be used over the entire season to measure and map patterns in water use, as well as to identify genotypic and growth-dependent variation in total soil water uptake and rooting depths.Coupled with plant physiological measurements, the approach can be used to estimate root depths, to identify mechanisms of drought tolerance in the field and to screen for effective water use in crop breeding programs.

Figure 1 .
Figure 1.(a) Location of the experimental field in the Plant Breeding Institute, The University of Sydney, Narrabri, New South Wales, Australia; (b) Locations of the chickpea genotypes and the rainfed (closed circles) and irrigated (open circles) treatments.Note: the red circles indicate plotswhere access tubes were installed, which were used for calibrating the empirical model for predicting soil water dynamics; the dash lines in Figure1b indicated

Figure 1 .
Figure 1.(a) Location of the experimental field in the Plant Breeding Institute, The University of Sydney, Narrabri, New South Wales, Australia; (b) Locations of the chickpea genotypes and the rainfed (closed circles) and irrigated (open circles) treatments.Note: the red circles indicate plots where access tubes were installed, which were used for calibrating the empirical model for predicting soil water dynamics; the dash lines in Figure 1b indicated boundaries of the irrigated and rainfed plots.
Soil Syst.2018, 2, 11 10 of 17 unexpected because θ values at the depth of 0-0.6 m were close to the permanent wilting point of the clay soils (~0.15-0.20 m 3 m −3 , figures not shown) while θ in the subsoil was still high (~0.25-0.30m 3 m −3 , figures not shown), so there was still available water for uptake by the chickpea genotypes.

Table 2 .
Estimated fixed effect coefficients for different chickpea genotypes under rainfed and irrigated conditions.Note: negative coefficients indicate greater water extraction.Note the change in the ranking of the genotypes with the different treatments.