2.1 Study area

The study was carried out in Lezíria de Vila Franca, a peninsula of alluvial origin surrounded by the Tejo and Sorraia rivers, and the Tejo estuary, located 10 km north-east of Lisbon, Portugal, as shown in Fig. 1. Soils in this region have fine to very fine texture and are classified as Fluvisols in the northern part and as Solonchaks in the southern part, according to the Harmonized World Soil Database (Fischer et al., 2012). Climate is temperate with hot and dry summers, according to the Köppen classification. Daily measurements of precipitation, mean temperature, and reference evapotranspiration recorded during the study period at the meteorological station represented by the blue circle in Fig. 1b are shown in Fig. 2. Land use in this area (of about 130 km²) is constituted by irrigated annual crops in the northern part and mainly by rainfed pastures in the southern part. Irrigation is ensured by an infrastructure that covers most of the area, collecting surface water at the confluence of the two rivers. The irrigation water has low salinity with electrical conductivity typically below 0.5 dS m⁻¹ and a sodium adsorption ratio below 1 (mmol_c L⁻¹)^0.5. The area exhibits a north–south soil salinity gradient which influences the distribution of land-use types and which is probably due to the regional distribution of the marine fraction of sediments and to the saline influence of the estuary on groundwater in the southern part.

Four locations were chosen in the study area, as presented in Fig. 1b, with numbers 1 to 4. Locations 1, 2, and 3 are cultivated with annual rotations of irrigated herbaceous crops in spring and annual ryegrass (Lolium multiflorum) in the autumn, with ploughing usually once a year. During the study years (2017 and 2018), the spring crop at location 1 was tomato drip irrigated, and at locations 2 and 3 it was maize irrigated by centre pivots. Location 4 is a rainfed spontaneous pasture that has not been ploughed at least in the last 10 years. During the study period, location 1 was irrigated from 12 April to 23 July 2017 and from 30 May to 23 September 2018; location 2 was irrigated from 17 June to 11 October 2017 and from 24 May to 22 September 2018; and location 3 was irrigated from 17 May to 10 September 2017 and from 6 June to 17 September 2018. Groundwater level is shallow, as expected in an estuarine environment, and has saline characteristics. In the southern part of the study area, closer to the estuary, the depth and salinity of groundwater are influenced by tidal variation.

Figure 1(a–b) Location of the study area in Portugal, showing the main geographical features and the four locations; (c) details of the four locations showing the EM38 transects and the soil sampling sites © Google Earth.

Figure 2Distribution of daily precipitation (P), reference evapotranspiration (ET), and mean temperature (T) recorded at the meteorological station located in the study area during the study period.

Download

2.2 Electromagnetic induction data acquisition and inversion

EMI data were acquired using the EM38 instrument (Geonics Ltd, Mississauga, Canada). Technology of this instrument is based on two coils, one transmitting the electromagnetic signal and the other receiving it, distanced 1 m apart from each other inside the instrument case. The position of these coils can be controlled by placing the instrument in a vertical position relative to the soil surface – horizontal dipole mode (the coils stand in the horizontal position), which provides a maximum depth of investigation of 1.5 m – or in a horizontal position relative to the soil surface – vertical dipole mode (the coils stand in the vertical position), which provides a maximum depth of investigation of 0.75 m. EM38 surveys were done on five dates at locations 1 and 4 and on six dates at locations 2 and 3, during the period of May 2017 to October 2018. Measurements on the first two dates were continuously acquired at each location, along a 100 m transect, using a GPS (Rikaline 6010, with 5 m position accuracy) for registration of the position. Subsequent EMI measurements were acquired at each location, along a 20 m transect. The middle point of each 20 m transect was coincident with the medium point of each previous 100 m transect. Measurements were acquired at positions 1 m apart along the 20 m transects (Fig. 1c), overlapping the medium section of the 100 m transects. EC_a was collected at two heights from the soil surface (0.15 and 0.4 m) in the horizontal and vertical dipole orientations, which was ensured by placing the EM38 on a cart built specifically for this purpose. The cart has two shelves to accommodate the instrument, one at 0.15 m from the soil surface and the other at 0.40 m from the soil surface. Inversion of EC_a data to obtain σ was carried out using a 1-D laterally constrained inversion algorithm (Monteiro Santos et al., 2011). The EC_a responses of the model were calculated through forward modelling based on the full solution of the Maxwell equations (Kaufman and Keller, 1983). The subsurface model used in the inversion process consisted of a set of 1-D models distributed according to the position of the EC_a measurements. The subsurface model at each measurement position was constrained by the neighbouring models, allowing the use of the algorithm in regions characterized by high conductivity contrast. An Occam regularization (De Groot-Hedlin and Constable, 1990) based approach was used to invert the EC_a data. All EC_a data, collected at the four locations, were inverted by applying a five-layer earth initial model with electrical conductivity of 100 mS m⁻¹ and a fixed layer thickness of 0.30 m. To run the algorithm, several parameters were selected, such as the type of inversion algorithm, the number of iterations, and the smoothing factor (λ) that controls the roughness of the model. The optimal inversion parameters for the present conditions were obtained in previous studies for the study area (Farzamian et al., 2019).

Table 1Description and equations of the statistical indicators used to evaluate the prediction ability of the regional calibration in this work.

^* n is the total number of data; mEC_e is measured EC_e; pEC_e is predicted EC_e; the upper bar represents the mean of the indicated data; s² is the variance of the indicated data.

Download Print Version | Download XLSX

2.3 Soil sampling and laboratory analysis

Soil samples were collected at the same time as EMI surveys along the transects, as shown in Fig. 1c. At each sampling site, five soil samples were collected at 0.3 m increments, from a depth of 0.15 to 1.35 m, as a representation of topsoil (0–0.3 m), subsurface (0.3–0.6 m), upper subsoil (0.6–0.9 m), intermediate subsoil (0.9–1.2 m), and lower subsoil (1.2–1.5 m), to monitor water content and EC_e. In the laboratory, water content was obtained using the gravimetric method and then converted to volumetric water content (θ – m³ m⁻³) after bulk density (g m⁻³) determination from undisturbed 100 cm³ soil samples. EC_e was measured with a conductivity meter (WTW 1C20-0211 inoLab) in the extract collected from the soil saturation paste obtained from 300 g of air-dry soil samples, according to the methods described by Richards (1954). In this study, the soil is classified according to its EC_e level as non-saline (EC_e<2 dS m⁻¹), slightly saline (2–4 dS m⁻¹), moderately saline (4–8 dS m⁻¹), highly saline (8–16 dS m⁻¹), and severely saline (>16 dS m⁻¹), according to the terminology proposed by Barrett-Lennard et al. (2008).

2.4 Prediction of EC_e from time-lapse EMCI

A regional calibration to predict EC_e from σ was previously developed for the study area, resulting in the linear equation EC_e=0.03σ–1.05 (Farzamian et al., 2019). This calibration was termed “regional” because the equation was obtained using all EC_e and σ data collected at four locations in the study area. Farzamian et al. (2019) tested the regional and location-specific calibrations, verifying that they have comparable prediction ability. However, the regional calibration can be used at any new location in the study area, within the range of measured EC_e, which makes it highly suitable for mapping salinity in the study area. The regional calibration was based on data collected during May and June 2017 and was validated using a leave-one-out cross-validation method with good results (RMSE = 2.54 dS m⁻¹ in the 0–37 dS m⁻¹ range). The detailed calibration and cross-validation procedures are described in Farzamian et al. (2019).

In the present study, the regional calibration was used to predict EC_e from time-lapse EMCI (pEC_e). The predicted EC_e and EC_e measured from soil samples (mEC_e), collected at the same time as the EMI surveys, were used as an independent dataset for the validation of the regional calibration. The validation was performed by calculating the root mean square error (RMSE), the coefficient of determination (R²) between the measured and predicted EC_e, Lin's concordance correlation coefficient (CCC), and the mean error (ME). Description of these statistical indicators and the equations used to calculate them are shown in Table 1. Calculations were done using global data and also using data discriminated by date of measurement (in this case we considered dates when measurements were done at the four locations – January, June, and October 2018).

To further explore the spatial and temporal sensitivity of the regional calibration and to investigate whether the regional calibration can be used to assess soil salinity variability at different locations and depths, we calculated the RMSE between the deviations of mEC_e (dmEC_e) and the deviations of pEC_e (dpEC_e) for each soil depth and location. These deviations were calculated by using the following equations:

$$\begin{array}{}\text{(5)}& {\displaystyle }{\mathrm{dmEC}}_{{\mathrm{e}}_i}={\mathrm{mEC}}_{{\mathrm{e}}_i}-\stackrel{\mathrm{‾}}{{\mathrm{mEC}}_{\mathrm{e}}},\text{(6)}& {\displaystyle }{\mathrm{dpEC}}_{{\mathrm{e}}_i}={\mathrm{pEC}}_{{\mathrm{e}}_i}-\stackrel{\mathrm{‾}}{{\mathrm{pEC}}_{\mathrm{e}}},\end{array}$$

in which the means were calculated with the values referent to each date of measurement for each soil depth and location. RMSE was then calculated between dmEC_e and dpEC_e according to Eq. (1).

3.1 Temporal variation of measured θ and EC_e

Figure 3 shows the variation of θ and EC_e with time at the sampling site located in the middle of each transect (Fig. 1c), at locations 1 to 4. At location 1, θ increases with depth and the lower subsoil (1.2–1.5 m) is permanently saturated within the study period. In the more superficial layers until 0.9 m depth, the influence of rainfall, evapotranspiration, and irrigation is noticeable. For instance, in the topsoil, θ peaks in January 2018 and lowers during the dry seasons, because drip irrigation during the dry seasons has a localized effect and there is high water uptake by the crop. At location 2, unlike the other locations, the lower subsoil is unsaturated. The influence of rainfall, evapotranspiration, and irrigation is also noticeable. At locations 3 and 4, θ also increases with depth and the intermediate and lower subsoil layers are permanently saturated.

Regarding EC_e, at location 1 the values observed are always below 1 dS m⁻¹, except for the topsoil in September and October 2018, which is probably due to fertigation practices during the irrigation period. At location 2, EC_e generally increases with depth. All layers show a peak in June and July 2018, probably due to fertigation practices. At location 3, EC_e reaches higher levels than at the previous locations, exceeding 4 dS m⁻¹, which is the generally accepted threshold for the classification of saline soils. Location 4 presents the highest EC_e of all locations. At the topsoil the values are below 4 dS m⁻¹ but increase consistently with depth to about 50 dS m⁻¹ in the lower subsoil. The increase in EC_e during June 2018 can be due to the influence of saline groundwater.

Figure 3Volumetric water content (θ – m³ m⁻³) and electrical conductivity of the soil saturation extract (EC_e – dS m⁻¹) in the topsoil (0–0.3 m), subsurface (0.3–0.6 m), upper subsoil (0.6–0.9 m), intermediate subsoil (0.9–1.2 m), and lower subsoil (1.2–1.5 m), measured in the sampling site located at the middle of each transect, at locations 1 to 4, during the study period. Each circled number refers to each location. Crosses refer to the dates when there were EC_e measurements but no σ measurements, due to adverse field conditions.

Download

3.2 Time-lapse EMCIs

Figure 4 shows the obtained EMCIs at locations 1 to 4 for each date of the EMI surveys. Globally, σ ranges from 19.44 to 1431.57 mS m⁻¹, with the lowest values at location 1 and the highest at location 4. A general increasing trend of σ is quite evident from the north to the south, accompanying the previously known soil salinity gradient. In addition, σ increases with depth at locations 2, 3, and 4. At location 1, σ ranges spatiotemporally from 19.44 to 128.08 mS m⁻¹. At location 2, σ ranges from 28.02 to 469.39 mS m⁻¹, with the highest values at depth. A similar pattern of σ is evident at locations 3 and 4. However, a greater range of σ is seen at location 3, with values from 36.23 to 706.32 mS m⁻¹. Location 4 exhibits the largest variations of σ, ranging from 48.57 to 1431.57 mS m⁻¹.

Figure 4Time-lapse electromagnetic conductivity images (EMCIs) for locations 1 to 4.

Download

3.3 Prediction of EC_e using the regional calibration

Figure 5 shows EC_e predicted with the regional calibration versus the measured EC_e and the 1:1 line, with points identified in terms of date of measurement (Fig. 5a) and depth of measurement (Fig. 5b). Table 2 shows the statistical indicators obtained using global data, i.e. data collected at all locations, from July 2017 to October 2018, and the statistical indicators for each date. The validation of the regional calibration using global data resulted in a RMSE of 3.14 dS m⁻¹ and a R² of 0.88, which indicates satisfactory prediction ability, given the large range of EC_e (52.35 dS m⁻¹). High global Lin's CCC of 0.94 shows agreement between measured and predicted EC_e. The ME is −1.23 dS m⁻¹, indicating that the regional calibration generally overestimates EC_e. Figure 5a and b show that the points are generally scattered around the 1:1 line, and it is not possible to identify variations depending on the date or depth of the measurement. In order to analyse the prediction ability at each location, Fig. 5c and d display an enlargement of the lower left parts of the previous figures, displaying EC_e values below 15 dS m⁻¹. Figure 5c and d show differences in the prediction ability according to the location, namely at locations 2 and 3, where EC_e is generally overestimated. At location 2, EC_e is more overestimated in deeper soil layers (Fig. 5d), which is likely due to the clay content that consistently increases with depth at this location, while it is rather uniform or declines with depth at the other locations (Farzamian et al., 2019). At location 3, EC_e is also overestimated, most likely due to the influence of θ and cation exchange capacity (Paz et al., 2019a), which are higher on average compared to locations 2 and 4. Finally, the EC_e ranges of location 4 and of the lower subsoil are similar to the EC_e range of global data, showing dominance of location 4 and of lower subsoil data on the calibration. On the other hand, the statistical indicators discriminated by date of measurement (in this case we considered only the dates when measurements were done at the four locations – January, June, and October 2018), shown in Table 2, reveal that the prediction ability does not vary significantly when comparing the statistical indicators of the three dates. These results suggest that spatial variability of data has a much stronger influence on the prediction ability of the regional calibration than temporal variability of data.

Figure 5Plots of predicted EC_e versus measured EC_e and the 1:1 line, obtained for locations 1 to 4, identified in terms of date of measurement (a) and depth of measurement (b). Plots (c) and (d) show enlargements of the lower left parts of plots (a) and (b), respectively.

Download

Table 2RMSE, ME, Lin's CCC, R², minimum, maximum, and range of EC_e, and the number of data used to calculate these statistical indicators, discriminated in terms of global and date of measurement.

Download Print Version | Download XLSX

Table 3 shows the RMSE between dmEC_e and dpEC_e obtained for each soil depth and each location. The RMSE is 1.04 dS m⁻¹ at the topsoil, which is relatively high given the small range of deviation variability (2.18 dS m⁻¹). This indicates that it is difficult to estimate soil salinity variability at topsoil from time-lapse EMI data and regional calibration. We attribute the weak prediction ability at topsoil to the small range of EC_e variability and to the larger variability of other soil properties (e.g. θ and temperature) at topsoil, which are due to different irrigation schemes and cultivated crops at each location. The results of RMSE show better prediction ability at the subsurface and very good prediction ability at upper, intermediate, and lower subsoil, suggesting that the time-lapse EMI data and regional calibration can be used to assess soil salinity changes in these soil layers. On the other hand, the RMSE results for each location show that the prediction ability is poor for monitoring of soil salinity variability at location 1 (1.06 dS m⁻¹) and location 2 (2.00 dS m⁻¹) given the small range of deviation variability at these locations. This is not surprising as the soil is non-saline at location 1, with a very small range of EC_e (0.35–1.89 dS m⁻¹), and slightly saline at location 2, with a small range of EC_e (0.91–3.86 dS m⁻¹). Thus, other soil properties such as θ and clay content have a larger impact on the spatiotemporal variability of σ. In addition, larger variations of EC_e and σ at location 4 dominated the regional regression calibration, further limiting the application of the regional calibration for monitoring the small variability of soil salinity at locations 1 and 2. The RMSE results are better to some extent at location 3 and very good at location 4, suggesting that the time-lapse EMI data and regional calibration can fairly predict soil salinity variability at location 3 and very well at location 4.

This spatial sensitivity of the regional calibration can be improved by studying new locations across the study area to include a wider variability of soil properties and ranges of EC_e in the regression calibration. On the other hand, longer observation periods and more frequent EMI surveying and soil sampling, as well as monitoring of other soil dynamic properties that influence σ (i.e. θ, soil temperature, level and salinity of groundwater), and finding ways to quantitatively account for their impact on time-lapse EMCIs, can improve the temporal sensitivity of regional calibrations.

Table 3RMSE between dmEC_e and dpEC_e, range of dmEC_e, and the number of data used to calculate RMSE, discriminated in terms of depth of measurement and location.

Download Print Version | Download XLSX

3.4 Generation of soil salinity cross sections from time-lapse EMCI

Figure 6 shows the soil salinity cross sections (EC_e predicted using the regional calibration) at locations 1 to 4 for each date of the EMI surveys, categorized into six salinity classes, ranging from non-saline to severely saline. The measured EC_e and the groundwater level at the sampling site located in the middle of each EMI transect are also shown.

Figure 6Cross sections of soil salinity (predicted EC_e) for locations 1 to 4, with representation of measured EC_e (in circles) and groundwater level (blue triangles) at the sampling sites located in the middle of each transect. Note that in June 2018 at location 3 and in July 2017 at location 4 there was no soil sampling.

Download

The salinity cross sections for location 1 show that the soil is generally non-saline, with slightly saline zones on all dates except for October 2018. These saline zones occur in the top soil layers until 0.9 m depth (topsoil, subsurface and upper subsoil) and represent an overestimation of the soil salinity when compared to the measured EC_e of the sampling point (which is invariably non-saline). This overestimation tendency is in agreement with Fig. 5d, where the very low range of spatiotemporal variations of soil salinity at this location can also be observed. In such conditions, other soil properties, such as θ, dominate the small variations of σ, and therefore the ability to predict salinity from σ at this location was reduced as discussed in Sect. 4.3. Our previous studies with both location-specific and regional calibrations tested at this location showed similar results (Farzamian et al., 2019).

At location 2 the salinity cross sections show an increase in salinity with depth from non-saline at the topsoil to highly saline in the lower subsoil, with the exception of July 2018, where the entire soil profile is moderately saline. The increase in soil salinity in the upper soil layers in July 2018 can be attributed to fertigation practices for the maize cultivation that introduced salts into the soil profile. The salinity cross sections also show the overestimation of salinity occurring mainly in the deeper soil layers, which agrees with the results presented in Fig. 5d and discussed in Sect. 4.3.

At location 3 soil salinity is well predicted in May 2017 but tends to be slightly overestimated on the remaining dates, especially in July 2018. The salinity cross sections show that salinity increases with depth, reaching severely saline in May 2017 and October 2017. This can be due to the influence of the saline groundwater (as seen in Fig. 3, the intermediate and lower subsoil layers are permanently saturated). The groundwater level is above 1.5 m in January 2018, although the salinity of the deeper soil layers (>0.9 m) decreases compared to May and October 2017, which could be due to washing of the profile by rainfall. The increase in soil salinity in the upper soil layers in July 2018, similarly to location 2 on the same date, can be attributed to fertigation practices for the maize cultivation.

At location 4 the trend of increasing salinity with depth is accurate on all dates, but it tends to be slightly underestimated. The salinity cross sections show that salinity increases from non-saline in topsoil to severely saline in lower subsoil. This is probably related to the saline groundwater level above 1.5 m. During the dry period of the year, salinity of the lower subsoil reaches the highest values (June 2018).

Comparison of the salinity cross sections between locations confirms the previously known north–south soil salinity spatial gradient of the study area; that is, from location 1 to location 4, soil salinity generally increases. Soil salinity dynamics at each location reveals fluctuations in time related to the input of salts and water either through irrigation, precipitation, or groundwater level and salinity. Location 1 tends to have non-saline characteristics, which can be attributed to good-quality irrigation. In addition, this location is far from the estuary, making it less prone to the presence of saline groundwater. At locations 2 and 3, the salinity cross sections show an increase in soil salinity in the upper layers during the dry season (when irrigation occurs), which decreases in the following months with increased rainfall (Fig. 2). At location 4, an increment of salinity along the entire profile is visible during the dry season. This is likely due to the influence of the saline groundwater and capillary rise along the profile.