IntroductionTop
The Argentinean pampas is a vast plain region of about 50 Mha and it is considered one of the most suitable areas for grain crop production in the world (Satorre & Slafer, 1999). However, on its southern portion (Flooting Pampas), the predominant soils of the region belong to the great group Natraquoll and Natraqualf (Soil Survey Staff, 2010). These soils exhibit a distinctive characteristic, which is the presence of a natric horizon (Btn), locally called ‘‘sodic’’ (Soil Survey Staff, 2010). Also, they have a poorly developed drainage system, normally situated in flat landscapes, with a strong textural contrast between horizons and with halomorphism and hidromorphism processes (Taboada, 2006). For these reasons these soils are used for livestock production (Vazquez et al., 2001). Furthermore, they are managed extensively and homogeneously, which in turn can reduce the system sustainability. A way to improve this type of management could be based on site specific agriculture technologies, improving soil-plant interactions knowledge and efficient production factors usage at farm and paddock level (Serrano et al., 2013).
Previous research has shown that the amount of soil variability across a farm and within a field of agricultural soils (high productivity) is of key importance for determining potential benefits of adopting precision farming (King et al., 2005; Bullock et al., 2009). However, relatively little is known about the degree of within-field spatial variation in soils used for livestock production (Serrano et al., 2013). Typically, soil sampling of the field and mapping, comprises grid-sampling and mapping approach as well as laboratory work. This is impractical at the farming scale because it is labor intensive, time consuming and expensive (King et al., 2005; Peralta et al., 2013). Therefore, it is desirable to find other more rapid and low cost means of obtaining information for detailed soil mapping (King et al., 2005). Measurements of apparent soil electrical conductivity (ECa) can be intensively recorded in an easy and inexpensive way, being one of the most reliable techniques to characterize within-field variability of edaphic properties (Moral et al., 2010; Peralta et al., 2015).
There are two types of electrical conductivity sensors currently on the market to measure soil ECa in the field. The first type of sensor (contact method) uses electrodes, in the shape of coulters that make contact with the soil to measure the electrical conductivity. The second type of sensor (non-contact method) is based on the principle of electromagnetic induction and does not contact the soil directly (Moral et al., 2010). ECa has been frequently used in the establishment of soil management zones and in the inference of several edaphic physicochemical properties and their respective spatial variation (Sudduth et al., 2005; Peralta et al., 2013). In agricultural soils, ECa has been used to characterize soil salinity (Rhoades et al., 1989); soil texture (Sudduth et al., 2003); soil depth (Peralta et al., 2013); soil moisture (Hossain et al., 2010); soil organic matter (OM) (Corwin & Lesch, 2005a) and cation exchange capacity (Kitchen et al., 2000). However, various authors have shown inconsistent relationships between ECa and soil characteristics, probably due to the fact that ECa is influenced by complex site dependent soil properties interactions (Corwin et al., 2003; Sudduth et al., 2005). Some studies have shown that ECa values are related to soil properties variability in extensive livestock production systems and are also related to pasture productivity (Serrano et al., 2010, 2014a,b). However, there is no information on the degree of within-field variability of the edaphic properties in Natraquoll and Natraqualf soils. Knowledge of these variations is essential if one intends to analyze the potential benefits of adopting a site specific approach to grassland and pasture field management in these soils.
The main objective of this study was to determine whether soil ECa is a potential estimator of the edaphic variability in Natraquoll and Natraqualf soils, which are characteristic of many livestock production systems around the world.
Materials and methodsTop
Experimental site
This study was conducted at Balcarce, in the southeast of the Buenos Aires Province, Argentina (37°45´ S, 58°18´ W; mean annual rainfall: 930 mm; mean annual temperature: 13.7°C) (Figure 1). The experiment was established in a paddock of 16 ha that sustained a permanent pasture dominated by Thinopyrum ponticum (Podp.) Liu & Wang. The site contains various soil series: Chelforó (Typic Natraqualf), Las Armas (Typic Natraquoll) and Tandileofú series (Mollic Natraqualf) (Soil Survey Staff, 2010). These soils are characterized by a clay loam texture (0-0.30 m).
|
|
Geophysical surveys
Data collection using the Veris 3100
Soil ECa measurements were made using the Veris 3100® sensor system (Fig. 2), at a low soil moisture content. The Veris 3100 device has six disc-shaped metal electrodes (coulter), which penetrate approximately 6 cm into the soil. One pair of electrodes passes electrical current into the soil, while the other two pairs measure the voltage drop. The measurement depth is based on the distance between the emitting and receiving coulter-electrodes. The system can be set up to work in configuration A (0-0.30 m) or B (0-0.90 m). Configuration A comprises the inside coulters (2, 3, 4, 5) and voltage is measured between the innermost ones (3 and 4). In configuration B, the four outside coulters (1, 2, 5, 6) include the 0-0.90 m deep measurement, and the voltage gradient is measured between coulters 2 and 5. Output from the Veris data logger reflects the conversion of resistance to conductivity (1/Resistivity = Conductivity). In this work, the ECa was measured at 0-0.30 m because 80% of the pasture roots are found at this depth (Doll & Deregibus, 1986). The Veris 3100 sensor was pulled across each field behind a pick-up truck (Fig. 2), taking simultaneous and geo-referenced ECa measurements in real-time with a differential GPS with sub-meter measurement accuracy and configured to take a satellite position once per second. The differential GPS was installed over the Veris 3100. On average, travel speeds through the field mapping ranged between 7 and 11 km/h, corresponding to about 2–3 m spacing between measurements in the direction of travel. For ease of maneuvering, the field was traversed in a series of parallel transects spaced from 15 to 30 m intervals, because a spacing greater than 30 m generates measurement errors and information loss (Farahani & Flynn, 2007). Elevation data were collected at the same time as the ECa data, using a differential GPS (vertical accuracy of 3-5 cm).
|
|
Geophysical data analysis
The structure of the ECa and elevation were quantified using geostatistics analysis, which were estimated as Isaaks & Srivastava (1989):
where y* (h) is the semivariogram that expresses the variation of the semivariance with the relative distance between the measured data; z (xi) is the measured sample value at sample points xi, in which there are data at xi; and xi + hN(h) is the total number of sample pairs within the distance interval h.
The semivariogram shows the decrease of spatial correlation between two points in space when the separation distance increases. The adjusted semivariograms were used to interpolate the ECa and elevation data using ArcGIS Geostatistical Analyst (ArcGIS v9.3.1, ESRI, Redlands, CA, USA), by means of ordinary kriging after checking geo-statistical common assumptions (Isaaks & Srivastava, 1989). A final 10 m × 10 m grid cell size was chosen because it reflects the scale of variability associated with the ECa and elevation measurements (Kitchen et al., 2005; Peralta & Costa, 2013).
Geographically weighted regression (GWR) is a technique for exploratory spatial data analysis. In linear regression, it is assumed that the relationship being modeled holds globally in the study area, but in many situations this is not necessarily true. The GWR provides the means for modeling such relationships (Brunsdon et al., 2002). A GWR tool (ArcGIS v9.3.1, ESRI, Redlands, CA, USA) was used to analyze the regional relation between elevation and ECa. It is possible that one set of variables provides a good model for a part of the studied area, but at the same time it may be unsatisfactory for other parts; GWR will adjust the relationship coefficients in order to reflect the regional variation (Serrano et al., 2010; Terrón et al., 2011).
Soil sampling
Data collection strategy
Based on geophysical surveys a grid composed by 12 points was sampled at a depth of 0-0.30 m. Each point represents the spatial variability of the plot. Each composite soil sample (three subsamples) was stored in a plastic bag and air dried in the laboratory. The following soil parameters were obtained: i) soil organic matter content (OM), using the Walkley & Black (1934) method; ii) pH, using a glass electrode at a 1:2.5 soil/water ratio suspension; iii) the electrical conductivity of the saturation extract (ECext), following the Chapman (1965) method; iv) available phosphorous (P) was determined according to the Bray & Kurtz (1945) method; and v) anaerobically incubated nitrogen (Nan), following the Echeverría et al. (2000) method.
Data analysis
Descriptive statistics were determined for elevation, ECa, and soil properties. Georeferenced buffers of 15 m (Peralta & Costa, 2013) were created around each soil sampling point using ArcGIS 9.3.1 (ArcGIS v9.3.1, ESRI, Redlands, CA, USA) and the ECa mean was calculated within the buffer areas. Linear regressions were computed between this ECa mean and soil properties with PROC REG (SAS Inst., 2007).
Principal component analysis (PCA) was used to examine the relationship between soil properties and to estimate which of these exert a greater influence on ECa. Principal components (PCs) become new, independent and random variables that can be used to identify which studied soil properties influence ECa. Any PCs with an eigenvalue > 1 explains a significant soil property variance (Peralta & Costa, 2013) and therefore were used in a stepwise-regression procedure (SAS Institute, 2007) to determine if there was a significant relationship between the PCs and ECa. The stepwise-regression procedure repeatedly alters the model by adding or removing the PCs predictor until the significance level of the last one is above 0.15. When the PCs remaining in the regression model accounted for >50% of the ECa measurement variability, the eigenvectors (loading factors) were examined and the soil properties in the PCs ranked according to the amount of variability explained by the PCs. Soil properties with loading factors <0.4 were not considered key latent variables because they did not substantially influence the relationship between the PC groups and the ECa.
Results and discussionTop
Data exploratory analysis
The measured soil properties, elevation, and ECa are summarized on Table 1. Accordingly standard criteria suggested by Wilding et al. (1994) some soil properties manifested high variation coefficients, especially OM, P and ECext (43%, 72 and 61%, respectively), whereas a relative stability was registered for Nan and pH (less than 20%). High variation coefficients of soil properties normally indicate high spatial variability and consequently suggest the convenience of site-specific management (Moral et al., 2010).
Table 1. Descriptive statistics of saturation extract electrical conductivity (ECext), pH, organic matter (OM), available phosphorus (P) and anaerobically incubated nitrogen (Nan) of the 12 sampled points
The ECa and elevation surfaces are shown in Figure 3. ECa showed substantial spatial variability for this particular field, ranging from 1.8 to 162.1 mS/m with a variation coefficient of 92% (Table 1). On the contrary, the elevation range is rather small (2.2 m) revealing a smooth topography with gentle slopes (Taboada, 2006) and consequently with low variation coefficients (Table 1). Elevation has a direct influence on soil forming processes and on soil water movement, and in consequence in salinity distribution within a paddock (Corwin & Lesch, 2005b). For this reason, elevation and ECa can be correlated (Tarr et al., 2005; Peralta & Costa, 2013). In this case, a visual inspection indicates higher ECa values on elevation depressions despite the low elevation variability (Fig. 3), as also described by Officer et al. (2004) and Serrano et al. (2010) in agricultural and livestock aptitude soils, respectively. Therefore, elevation was negatively associated with ECa throughout the field, due to the fact that higher ECa values are observed in lower areas. These results can be explained by the fact that the soils in the depressed areas used for livestock are usually clasified as Natraqualfs soils (Batista et al., 2005), which are characterized by a high solute concentration (Soil Survey Staff, 2010).However, the relationship between ECa and elevation varied spatially. The GWR analysis allowed the delineation between areas with a strong and a low relation between elevation and ECa (Fig. 4).
|
|
|
|
Relationships among ECa and soil properties
Regression analysis
The relationships between soil properties and ECa are shown in Table 2. The pH and ECext were positively associated with ECa. These high correlations are expected because they reflect the influence of salts and pH on the measured ECa and because these properties are highly correlated (Corwin et al., 2003; Peralta & Costa, 2013). Salts concentration and pH increased soil solution conductivity and is consistent with findings in previous studies (Rhoades et al., 1989; Kaffka et al., 2005). These results also agree with those reported by Peralta et al. (2013) in agricultural soils of the Argentinean pampas.
Table 2. Models describing the relationships between ECa and saturation extract electrical conductivity (ECext), pH, organic matter (OM), available phosphorus (P) and anaerobically incubated nitrogen (Nan)
Our results showed that significant and negative correlation coefficients were found between ECa and OM (Table 2). This may be due to the fact that in these soils, the areas with thin superficial horizon, and in consequence with lower OM content, also have the highest solute concentrations (Batista et al., 2005), which increases the measured ECa. On the contrary, in agricultural soils of the Argentinean pampas, a direct association was established between ECa and OM (Peralta et al., 2013).
The P showed a weak association with ECa (Table 2). This nutrient is less positively correlated, but still significant (p<0.05). Jung et al. (2005) mentioned that the low association between ECa and P is attributable to the influence of the fertilization method (band application) usually used in the Argentinean pampas (Simón et al., 2013). On the contrary, no association was established between ECa and Nan (Table 2). This behavior may be explained by variation and low concentrations of N. These results agree with those reported by Peralta & Costa (2013).
Principal component analysis and PC-stepwise regression
A significant relation was found between some soil properties (pH, OM, ECext and P) and ECa (Table 2). However, due to the co-linearity of the independent variables, mutivariate statistical methods that include PC analysis are more appropriate to evaluate the relation between soils properties and ECa (Moral et al., 2010; Peralta & Costa, 2013).
Table 3 shows the three first PCs. These PCs had a cumulative variance of more than 95%. The first PC (PC1) explained 69% of the total variance and was positively influenced by pH and ECext, and negatively by OM (loading factors>0.4) (Table 3). On the other hand, the second PC (PC2) and third PC (PC3) only explained 20 and 10% of the total variance respectively. PC2 was highly related to Nan, whereas PC3· was related to P (Table 3).
Table 3. Key principal components (PCs), eigenvalues, cumulative variance, loading factors for each soil property and regression model resulting from the principal component stepwise regression analysis
PCs with an eigenvalue greater than 1 explain a significant amount of the variance present in the soil properties (Peralta & Costa, 2013). In this case only PC1 had an eigenvalue greater than 1 (Table 3). Confirming this, the PC-stepwise regression analysis only retained PC1 (Table 3). Therefore, ECext, pH and OM were considered key latent variables because they substantially influence the relationship between the PC1 and the ECa (loading factors>0.4) (Table 3). Conversely, as previously mentioned, PC2 and PC3 showed a more intense relationship with Nan and P (Table 3). Nevertheless, these PCs were not retained in the PC-regression model. Figure 5 shows the spatial distribution of PC1. The sites with lower values of PC1 correspond to sectors of the field where the ECext and the pH are low and the OM is high.
|
|
As conclusion, identification of regression models that were able to account for a large portion (50%) of the variability in soil ECa would indicate situations where this parameter could be used successfully to measure soil properties (Heiniger et al., 2003). Our model explained slightly more than 69% of the total variation of the ECa measured. Therefore, our results provide evidence that soil ECa is useful in identifying sites with different pH, ECext and OM (loading factors>0.4) in soils used for livestock. Thus, ECa can be used as a support tool to implement site-specific management in permanent pastures.
This study shows that soil pH, OM and ECext have a reasonably strong spatial correlation with the ECa of the soil. The use of geo-electric sensors in the particular type of soil of the studied site can be promising for the nutritional management of pastures. This will enable increased economic, environmental and energy efficiency. It also allows mapping the soil at field scale with a low input of resources.