The aim of this work is to investigate how the spatial variability of soil
properties and soil erodibility (
Soil erosion is a significant economic and environmental problem worldwide as a driving force affecting landscapes (Zhao et al., 2013). It is a very dynamic and complex process, characterized by the decline of soil quality and productivity, as it causes the loss of topsoil and increases runoff (Lal, 2001; Yang et al., 2003). Furthermore, soil erosion often causes negative downstream impacts, such as sedimentation in rivers and reservoirs, decreasing their storage volume as well as lifespan (Pandey et al., 2007; Haregeweyn et al., 2013).
One of the main causes of soil loss intensification around the world is associated with land-use change (Leh et al., 2013). The relationship between different land use and soil susceptibility to erosion has attracted the interest of a variety of researchers (Yang et al., 2003; Cerdà and Doerr, 2007; Blavet et al., 2009; Biro et al., 2013; Wang and Shao, 2013), who have shown the impact of changes in vegetation cover and agricultural practices on soil properties and therefore on overland flow. Generally, cultivated lands experience the highest erosion yield (Cerdà et al., 2009; Mandal and Sharda, 2013). In the Mediterranean regions, in combination with these anthropogenic factors, climate change has amplified the concerns about soil erosion since it is expected that there will be an increase of dry periods followed by heavy storms with concentrated rainfall (Nunes et al., 2009).
Some models have been developed to predict soil loss and sediment delivery.
The Revised Universal Soil Loss Equation (RUSLE) is the most used empirical
equation for modeling annual soil loss from agricultural watersheds (Renard
et al., 1997). The susceptibility of soil erosion and land degradation depends
largely on various inherent soil properties, namely chemical, physical,
biological and mineralogical properties (Cambardella et al., 1994;
Pérez-Rodríguez et al., 2007). However, according to the RUSLE model only
some of the soil's properties define soil erodibility (
Spatial variability in soils occurs naturally as a result of complex interactions between geology, topography and climate. Moreover the spatial variability of soil properties, which influence soil susceptibility to erosion, is highly related to anthropogenic factors particularly in cultivated lands (Paz-González et al., 2000; Wang and Shao, 2013). Thus, information on the spatial variability and the interactions between soil properties is essential for understanding the ecosystem processes and planning sustainable soil management alternatives for specific land uses (Pérez-Rodríguez et al. 2007; Ziadat and Tamimeh, 2013).
Classical statistics and geostatistics methods have been widely applied in studies about spatial distribution of soil properties (Pérez-Rodríguez et al., 2007, Tesfahunegn et al., 2011). Geostatistical techniques based on predictions and simulations have been used to describe areas where predicted information is established by a limited number of samples (Goovaerts, 1997). Geostatistics provides tools for analyzing spatial variability structure and distribution of soil properties and evaluating their dependence (Panagopoulos et al., 2014).
The biplot methodology provides an added value for analyzing spatial variability of soil properties. This multivariate statistical technique allows the graphical representation of a large data matrix (Gabriel, 1971), whereby it is possible to interpret the relations between individuals (samples) and between variables, as well as between both. Biplot can also indicate clustering of units with close characteristics, showing inter-unit distances as well as displaying variances and correlations of the variables (Gallego-Álvarez et al., 2013). The HJ-Biplot permits not only the analysis of the behavior by sample but also the determination of which variable is responsible for such behavior (Garcia-Talegon et al., 1999), allowing a visual appraisal to establish relations between soil properties and land uses.
The construction of the Alqueva dam in a semiarid area of southern Portugal
created one of the largest artificial lakes in Europe. Taking advantage of
water availability from the reservoir, this Mediterranean region has been
subjected to land-use conversion from the native montado grassland to
intensive agricultural uses. Land-use conversion from the native ecosystem
to agriculture may alter physical, chemical and biological soil properties,
which consequently may increase soil erosion and siltation in the reservoir.
Soil erosion in the area has to be carefully evaluated in order to undertake
sustainable soil management measures. Therefore, the aim of this study was
to evaluate the effects of cultivation practices on some chemical and
physical soil properties and on soil erodibility (
Located in the semiarid Alentejo region of Portugal, at the Guadiana
River, the Alqueva reservoir (8
Location of the study area at the Alqueva dam watershed in Portugal.
The study experimental site (farm “Herdade dos Gregos”), located in the surrounding area of the reservoir (Fig. 1), is a private property with 900 ha. The landscape is characterized by its hilly topography with significant altitude variations (mainly between 100 and 250 m). The bedrock of the study area is rocky, and, according to the World Reference Base for Soil Resources (FAO, 2006), the two types of soil in this area are Haplic Luvisols (LVha) and Lithic Leptosols (LPli). This farm was selected to include a diversity of land uses, including native montado grassland and more intensive land uses, with irrigation, namely olive tree orchard and lucerne cultivation. Direct pumping from Alqueva reservoir is done on this private property since it is near the reservoir.
The typical landscape in the Alentejo region is the montado native
grassland, an agrosilvopastoral system characterized by savannah-like,
low-density woodlands with evergreen holm oaks (
Taking advantage of the water availability, another land use (with 33.5 ha)
is an irrigation area (pivot sprinkler irrigation system) on which lucerne
(
Another irrigated land use consists of an olive tree plantation (57.5 ha), which is done in strips. This cultivation has a drip irrigation system, is fertilized once every 2 years and is ploughed once a year to decrease weed competition. The olive orchard is located in the low elevations of the farm (150–186 m), and it is on the side of the reservoir (Fig. 1). The slope varies from 0 to 14.2 %.
Since the objective was to study the relation between soil properties and the
Soil erodibility factor (
Data were subjected to classical analysis using SPSS 17.0 software to obtain descriptive statistics, namely the mean, minimum and maximum; standard deviation (SD); coefficient of variation (CV); and skewness of each parameter.
Soil data were introduced in the ArcGIS environment, and geostatistical
analyses were performed using Geostatistical Analyst tool, in other to
examine spatial distribution of soil properties. Prior to using geostatistics to
obtain prediction maps, a preliminary analysis of data was done to check
data normality and global directional trends. Skewness is the most common
statistical parameter to identify a normal distribution that is confirmed with
skewness values varying form
The geostatistical methodology is based on the creation of a semivariogram
(SV), a graphical representation (Eq. 2) that describes how samples are
related to each other in space, and it is based on
Ordinary Kriging (OK) was selected as a geostatistical method. OK is considered one of the most accurate interpolation techniques which assumes that variables close in space tend to be more similar than those further away (Goovaerts, 1999).
Using the Geostatistical Analyst tool (ArcGIS) and selecting the OK methods, a semivariogram was created for each measured property. In the Kriging method different semivariogram models can be used (e.g., spherical, exponential) and the selection is usually performed by employing the cross-validation technique, which permits the evaluation of the prediction accuracy. Cross validation was executed to investigate the prediction performances through the statistical values, as the mean error (ME) or root-mean-square standardized error (RMSSE), which results from comparing the estimated semivariogram values and real observed values. Additional semivariogram parameters were analyzed to better understand the spatial structure and dependence of each variable. Nugget is the variance at distance zero and reflects the sampling error. Sill is the semivariance value at which the semivariogram reaches the upper bound and flattens out after its initial increase; it is the variance in which the samples are no longer spatially related to the study area.
Once the cross-validation process was completed, interpolation maps of spatial distribution, for each soil variable, were produced according to the semivariogram model selected, in the ArcGIS software.
HJ-Biplot represents a matrix, without assumptions related to its probabilistic distribution, permitting a graphic representation of the geometric data structure, representing the data set (samples and variables) variability. The prefix “bi” is due to a simultaneous representation of the matrix rows and columns, searching for the maximum representation quality possible, at the same scale (Martín-Rodríguez et al., 2002; González-Cabrera et al., 2006; Gallego-Álvarez et al., 2013).
A data matrix
With the
On the HJ-Biplot graphic representation, the points represent individuals
(samples) and the vectors represent variables (in this case, chemical and
physical soil properties). To interpret and discuss the graphs obtained with
this methodology, it is essential to be aware of the following (Gallego-Álvarez et al., 2013):
The distance between points represents the variability and can be interpreted
as similarity or dissimilarity; i.e., the close samples have similar
behaviors. The angle formed by variable vectors is interpreted as correlation; i.e., small
angles between variables represent similar behaviors with high positive correlations,
and the obtuse angles that are almost a straight angle are associated with variables with
high negative correlations; i.e., the cosine value of the angles represents the correlation
between variables. The proximity of individual points and variable vectors means high preponderance;
in other words the closer a point is to a variable vector, the more important this sample
is to explain this variable. The length of the vector represents the variable's variability; the longer the
vector, the higher this variability.
Descriptive statistics of soil properties and parameters of the fitted variogram models and the cross-validation results.
*Transformation for normal distribution.CV – coefficient variation; Min – minimum; Max – maximum; VFS – very
fine sand; N – nitrogen; OM – organic matter; EC – electrical
conductivity; HC
The descriptive statistics of soil properties are given in the first part of Table 1. All measured parameters varied considerably within the areas (different land uses) as indicated by the coefficient of variation (varies from 4.2 to 70.2 %). Nitrogen (N) and organic matter (OM) show the highest variation values, especially for cultivated fields (lucerne cultivation and olive orchard), which can be explained with the lack of homogeneous fertilization or tillage practices applied to soil in these areas.
The skewness results, which vary from
These mean results show significant differences between land uses for all the properties analyzed. From the particle size distribution reported in Table 1, the soils are mostly sandy loam, formed mainly of sand, followed by silt and low quantities of clay. However, there are some differences between land use areas that can be explained by soil type. The LPli soils are characterized by a thin layer (about 10 cm), in that case upon a schist rock, justifying the higher clay content at the montado grassland. The LVha soils in the lucerne cultivation and the olive orchard are characterized by a loam or sandy loam layer (first 20 cm) with good drainage over clay-enriched subsoil (upon a basic crystalline rock), explaining the lower values of clay and fine sand, especially in the olive orchard. Despite the same soil type, soil texture is different between lucerne and the olive orchard, which can be justified by land use. The lucerne is a more intensive cultivation (intensive irrigation, tillage and continuous cultivation; fertilizers and lime application), involving conditions that promote changes in the soil weathering and moisture and, consequently, in soil texture (Yimer et al., 2008). On the other hand the soil between olive trees is kept without vegetation for most of the year and can explain the clay drainage to a sub-layer.
Montado shows the highest content of OM (5.22 %), whereas lucerne and olive fields show the lowest values (2.08 and 2.10 %, respectively). Other studies suggest that OM is higher in no-tillage soils compared to minimum tillage that increases aeration (Celik, 2005). Tillage mixes the subsoil with topsoil; after soil erosion, the nutrients are easily leached and the surface becomes poor in nutrients (Al-Kaisi and Licht, 2005). As for OM, the highest value of N nutrient occurs in the montado (0.19 %) and the lowest values in lucerne (0.11 %) and the olive orchard (0.10 %), which is related to the tillage practice that is frequently employed in these last two land uses, while in the montado grassland the cattle enriches the soil.
Soil EC values (Table 1) were similar when comparing the montado grassland
(0.100
The soil pH was significantly higher in the lucerne cultivation land (7.1) compared to the montado grassland (5.9) or in the olive tree orchard (5.5) (Table 1). The soil pH in the lucerne was greater due to lime application to increment the soil pH in that area. Lucerne's optimum pH for production is between 6.5 and 7.2, and lime application has been found to produce a significant improvement in nodulation of lucerne (both number and dry weight of nodules per plant) (Grewal and Williams, 2001).
Saturated hydraulic conductivity (HC) values were greater in the lucerne
area (5.95
As a result, the
Model selection for each soil property was based on the nugget, sill, ME and the RMSSE presented in the second part of Table 1 (Geostatistics).
Three-dimensional perspective of the trends in the input data sets.
Nugget is low in most soil properties studied, implying strong spatial dependence. The nugget-to-sill ratio is used to define spatial dependence of soil properties: if the ratio is < 0.25, there is strong spatial dependence; if it is 0.25 to 0.75, there is moderate spatial dependence; and if the ratio is > 0.75, spatial dependence is weak (Cambardella et al., 1994). As shown in Table 1 the ratio values indicate the presence of high to moderate spatial dependence for all soil parameters (values between 0 and 0.64). In general, there is stronger spatial dependence in montado (low nugget-to-sill ratio), which can be explained with the non-existence of extrinsic factors, such as management cultivation practices, that influence soil properties, and soil is left as it is for permanent pasture.
Cross validation facilitated the selection of the best-fit semivariogram for an interpolation map, which could provide the most accurate predictions. Closer values of the ME to 0, and closer values of the RMSS to 1, suggested that the prediction values were close to measured values (Wackkernagel, 1995). Most of the soil properties were best fitted with an exponential model, particularly in the montado area and olive orchard, whereas in lucerne the Gaussian, circular and stable semivariogram models were used.
Prediction map of very fine sand (VFS), total nitrogen (N),
organic matter (OM) and soil erodibility (
The interpolation maps obtained with geostatistics are useful to better understand spatial variability and its influences. The variability of spatial soil properties can be influenced by natural factors (such as particle-size composition and topography) and anthropogenic factors (such as land cover or management practices) (Tesfahunegn et al., 2011). Sometimes, the effect of some factors is at least 1 order of magnitude greater (as topography or soil type) than the land use. So, as mentioned trend analysis was performed to study the existence of directional trends caused by these factors with large scale of variation, and it is shown in Fig. 2. Global trend exists if a curve that is not flat (i.e., a polynomial equation) can be fitted to the data (for example for total N in montado or very fine sand (VFS) in the olive orchard). These trends were identified for part of the soil properties and for different land uses (Fig. 2). The strongest influence of a directional trend was identified from southeast to the northwest, which could be associated with the topography (Fig. 1) since the altitudes increase in accordance with these direction. So, trend removal is crucial to create more accurate prediction maps in order to justify an assumption of normality.
The interpolation maps for some studied soil properties are shown in Fig. 3. Through looking at the VFS distribution, it was noticed that the higher fractions of these particles (Fig. 3) were measured at low altitudes or on flat slopes such as the valley (see elevation in Fig. 1). This can be explained by erosion–deposition processes because these particles are easily detached and transported by water.
The highest percentages of N and OM were found on montado, as discussed previously. These two properties present similar distributions for all land uses. The nitrogen existing in the soil is mostly organic, and the inorganic forms (ammonium and nitrate) are easily leached or assimilated by plants. So, when OM breaks down due to mineralization, the N fraction decreases (Varennes, 2003). There were higher values in montado because the soil is not frequently tilled as it is in the other land uses. In the lucerne cultivation and the olive orchard, the variation of OM and N can be explained by inadequate management practices (e.g., inadequate fertilization rates, tillage, irrigation rates, seed rates).
Figure 3 illustrates the interpolation map for the
When looking for natural vs. anthropogenic impact on the
The HJ-Biplot representation matrix of soil samples and studied variables.
The HJ-Biplot representation matrix of soil properties is showed in Fig. 4. It was observed that the dominant axis (axis 1) takes 35.83 % of the total inertia (information) of the system. With both dimensions, an accumulative inertia of 61.04 % was achieved. Regarding this graphic representation, it was observed that samples were grouped according to the land use. The montado samples were close to OM, N and clay vectors, showing their preponderance to be a characterization of these variables. The lucerne samples were important to describe the pH and silt content. On the other hand the olive samples were more disperse but related to EC, permeability class, sand, VFS and K.
The variables demonstrating a more positive correlation were OM and N, as previously noticed. Clay and silt were also positively correlated, but they were negatively correlated with sand, as expected, because soils with more sand have less clay and/or silt.
Through the matrix representation it was detected that soils with more sand have higher EC (olive orchard), although EC normally increases with the percentage of clay. This may be explained by the addition of fertilizers, as previously discussed, that can contribute to an EC increase. These results for EC show low variability between land uses, revealing a low cation exchange capacity of these soils. This is frequently caused by intensive soil mobilization (Paz-González et al., 2000).
Permeability class increases as the HC
Nevertheless, the properties more positively correlated with the
Hierarchical clusters representation of soil samples and studied variables.
Figure 5 shows the hierarchical cluster representation. Using HJ-Biplot
methodology and the aggregation tool
Therefore, the cluster analysis is convenient to identify the effect of different land use and management on soil properties and consequently on soil erosion. On the other hand, the cluster analysis could support the delineation of zones according to soil properties, and subsequently according to erosion susceptibility, which could be used for site-specific soil management recommendations.
This study demonstrated that the variability of soil properties and the
Therefore, in the surrounding area of the Alqueva reservoir, the ongoing change in land use and soil management practices can have a significant effect on chemical and physical soil properties. As a result, this affects the soil erodibility index, intensifying the risk of erosion. The increase of soil loss in the watershed might have a significant impact on a reservoir's ability to store water, reducing its lifespan.
Knowledge of soil spatial variability is fundamental for environment management and can help in the sustainable use of the resource soil. The prediction maps produced with geostatistics are an important monitoring tool, showing the exact position in the field of the specific soil properties. The HJ-Biplot methodology was demonstrated to be useful in gaining a better understanding of how soils properties were correlated and allowed not only a determination of the behavior by sample but also a conclusion as to which variable is responsible for such behavior. The simultaneous use of HJ-Biplot with geostatistics allows this information to be found on the map, which has important theoretical and practical significance for precision agriculture. Facing the intensification of cultivation in the surrounding area of the reservoir, site-specific soil management and careful land use planning are needed to take into account the spatial variability of soil properties, delineating management zones, variable fertilization management, irrigation scheduling, conservation practices and other efforts.
The authors want to thank the Portuguese Foundation for Science and Technology for its support in this research project (PTDC/AAC-AMB/102173/2008 and SFRH/BD/69548/2010) and the Research Center for Spatial and Organizational Dynamics (CIEO) for having provided the conditions to publish this work. Edited by: A. Cerdà