SPATIAL VARIABILITY STRUCTURE OF THE SURFACE LAYER ATTRIBUTES OF GLEYSOLS FROM THE COASTAL PLAIN OF RIO GRANDE DO SUL

The spatial variability structure of soil attributes in a certain area might influence the semivariogram fitting model and, consequently, the attribute behavior mapping in this area leading to different decisions regarding crop management. This study aimed to identify, characterize and quantify the spatial variability of chemical attributes and the clay content in the superficial layer of a Gleysoils mapping unit (MU) at reconnaissance scale in the coastal plain of Rio Grande do Sul, through descriptive statistics and geostatistics and compare the results taking into consideration the existence of three Gleysoils mapping units at semi-detailed scale through the scaled semivariogram technique. A 403 ha area located in the Rio Grande do Sul Coastal Plain, in the city of Jaguarão was sub-divided into three mapping units (GL-mo, GL-mo.lv and GL-lv), a sampling grid with 403 points, 100 m far one from another was established. In a 5 m radius around each sampling point, 10 sub-samples of disturbed soil were collected from the 0-0.20 m layer, making up a soil compound sample, and the following attributes were determined for each sample: pH in water, organic carbon, phosphorus, potassium, sodium, calcium, magnesium, aluminum, potential acidity and clay content. The cation Exchange capacity (pH=7.0) and base saturation were also calculated. The identification, characterization and quantification of the spatial variability of attributes from the soil Ap horizons were carried out through descriptive statistics and geostatistics, considering the mapping unit at the reconnaissance scale and the three units at the semi-detailed scale. In the geostatistics analysis, the scaled semivariogram technique was employed aiming to compare the spatial variability structure for each soil attribute in the total area and in the three MUs at the semi-detailed scale. Regarding the descriptive statistics, the Ap horizon attributes behavior in GL-lv was similar to that in the total area of the soil layer under analysis; however, when considering the spatial coordinates, the spatial variability structure of the GL-mo.lv attributes was the one that best described the attributes variability in the total area. The scaled semivariogram technique revealed that the spatial behavior of the attributes pH and exchangeable sodium was similar, regardless of the evaluation scale adopted or the factor used for the scaled semivariogram.


INTRODUCTION
Learning about the spatial variability of soil attributes, at the field or water basin level, is essential for a better farming management practice as well as to evaluate the farming impact on the environment (CAMBARDELLA et al., 1994).In Brazil, however, most of its territory has only information generated in low intensity exploratory or reconnaissance surveys, made up by highly heterogeneous mapping units (associations), which are not enough for this kind of evaluations (STRECK et al., 2008).
Criticism to the qualitative character of the conventional pedologic surveys, based on a mental model (soil-landscape relation), allied to the advent of new technologies (remote sensing, GPS, SIG, digital elevation model, etc.) led to the development of new quantitative methods to carry out the soil pedologic survey taking into consideration its spatial distribution.McBratney et al. (2003) pointed out that in response to these criticisms, quantitative methods have been used to describe, classify and study the soil spatial distribution pattern more objectively.The authors also emphasize that these methods are categorized within a soil science emerging area known as Pedometry, and the ones Spatial variability structure… BITENCOURT, D. G. B. et al. most commonly used to analyze the soil spatial distribution are geostatistics, classical statistics and the combination of both.
Geostatistics has been used successfully to evaluate, identify and map the spatial variability of soil attributes (LARK, 2012;TRANGMAR et al., 1985) and its tools have been employed in Precision Farming aiming at soil sampling, attributes estimate and mapping, fertilizers variable application rate, among others (OLIVER, 2010).Some authors have suggested to consider the soil mapping units in its interpretation as advantageous (COUTO et al., 1997;ROGOWSKI;WOLF, 1994).Wu et al. (2008) used geostatistics to estimate the concentration of four heavy metals (copper, zinc, lead and cadmium) in three soil mapping units.Those authors concluded that the kind of soil was one of the factors that most affected these metals concentration and, therefore, the metal spatial variability would be better characterized if the soil mapping units were taken into consideration.Liu et al. (2006), comparing ordinary kriging combined with soil map design, concluded that the soil map conventional design can be used to improve its attributes spatial interpolation.Duffera et al. (2007), studying the relation between vertical and horizontal spatial variability of physical attributes within three mapping units in a 12 ha area (scale 1:2400), found that the spatial variability structure of some physical attributes (texture, water available to the plants and soil resistance to penetration) were captured by the units but others (soil density, total porosity and water conductivity in saturated soils) were not.Nielsen et al. (1996) carried out a detailed soil survey (scale around 1:15.000) from a 100 ha area within a single mapping unit at the reconnaissance scale, identifying seven mapping units, named 'pedotops' (mapping unit including a taxonomic unit at its lowest level) by the authors.After that, they measured the water infiltration rate in the soil at 293 sites (separated by 60 m intervals) and estimated it by using regression equations based on the soil superficial layer texture.The authors concluded that in the situation in which the units ('pedotops') were not considered, there was no correlation between the measured values and those estimated for the infiltration rate; however, when they were considered, the data correlation became quite high (r 2 = 0.936).
The hypothesis of this study is that the mapping units associated to the scale of soil survey might influence the semivariogram fitting model and, consequently, the spatial variability structure, leading to different decisions regarding crop management.In this sense, the objective of this study was to identify, characterize and quantify the spatial variability of chemical attributes and clay content in the superficial layer of a Gleysoil mapping unit at the low intensity reconnaissance scale, used with flooded rice in the Coastal Plain of Rio Grande do Sul, through descriptive statistics and geostatistics and to compare the results taking into consideration the existence of three Gleysoil mapping units at the semi-detailed scale through the scaled semivariogram technique.

MATERIAL AND METHODS
The study was carried out in a 403 ha area, inside the "Formiga" mapping unit in the Reconnaissance Survey in the State of Rio Grande do Sul (BRASIL, 1973;STRECK et al., 2008) (Figure 1).The area is located in Granja Bretanhas farm, in the Coastal Plain of Rio Grande do Sul, municipality of Jaguarão -Brazil.The area central point geographical coordinate is 32°32'45"S and 53°05'45"W and the relief is plain, with altitude ranging from 5 to 7 m above the sea level.According to the Köppen climate classification, it has a Cfa climate, with a maritime subtropical environment, sub-humid summer and humid or super-humid climate throughout the other seasons (MOTA, 1983).
Throughout the 1999 to 2005 crops, the crop system adopted in the study area was flooded rice followed by soybeans.In the years 1998 and 1999, the area was systematized and since then the nontillage system has been employed.In the experimental area, the sampling grid had 403 points established, spaced 100 m, with the aid of a GPS (Figure 2).In a 5 m radius around each sampling point, 10 sub-samples of disturbed soil from the 0-0.20 m layer were collected in order to produce a compound sample from each sampling point.Each soil sample had the following attributes determined: clay content, pH in water, organic carbon, exchangeable phosphorus, potassium, sodium, aluminum, calcium and magnesium and potential acidity.The cation exchange capacity (pH=7.0)and the base saturation were also calculated.All methodologies are described by Tedesco et al. (1995).
The identification, characterization and quantification of spatial variability of the soil superficial horizon (Ap) attributes were carried out through descriptive statistics and geostatistics, considering the total area (low density reconnaissance mapping unit) and, also, considering its semi-detailed mapping units.For the exploratory analysis of data, the following statistics were calculated: arithmetic mean, maximum value, minimum value and coefficient of variation.The Wilding;Drees (1983) criterion was adopted to classify the soil attribute variability based on the coefficient of variation: low -CV ≤ 15 %, moderate -15 < CV ≤ 35 % and high variability -CV > 35 %.The Kolmogorov-Smirnov (MASSEY, 1951) adherence test was applied at 5% significance level, in order to verify the normality of the data set, aiming to help the selection of the semivariance estimator for the geostatistical analysis.
Two semivariance estimators were used in the geostatistics analysis: the classical estimator (MATHERON, 1962), when the variable under study presented normal distribution, and the Cressie;Hawkins (1980) robust estimator, when it did not.The experimental and theoretical semivariogram adjustments with the respective parameters (nugget effect "Co", sill "C+Co" and range "a") were carried out based on the ordinary least square method implemented in the software R. The spatial dependence degree (SDD) of each attribute [SDD=(Co/C+Co)*100] was classified according to Cambardella et al. (1994) as: strong SDD ≤ 25 %; moderate -25 % < SDD ≤ 75 %; and weak -SDD > 75 %.The scaled semivariogram technique proposed by Vieira et al. (1997), in which each experimental semivariance value was divided by the most suitable scale factor (sampling variance or the adjusted theoretical model sill), was applied aiming to compare the spatial variability structure of each soil attribute in the total area and its mapping units.
All the statistical procedures were carried out aided by the software R, and the descriptive statistics was used with the Rcmdr packet (FOX, 2005), the adherence tests employed the fBasics packet (WUERTZ, 2012) and the experimental and theoretical semivariograms were built with the geoR packet (RIBEIRO JÚNIOR; DIGGLE, 2001).In order to compare the results of this study considering the existence of different areas, it was used the scaled semivariogram technique proposed by Vieira et al. (1997).
According to the Soil Fertility and Chemistry Committee (CQFSRS/SC, 2004), the pH and V% average values in the total area and the GLmo and GL-lv (Table 1) soils mapping units were classified as high, while in GL-mo.lvthey were considered medium; organic matter content (carbon content x 1.724) are in accordance with the medium classification for the GL-mo unit and low for the remaining ones (< 2.5 %); the Ca and Mg average content are classified as high in all cases; regarding CTC pH 7.0, it can only be considered high in the GL-mo (> 15.0 cmol c dm -3 ).
Considering total area and the soil mapping units, the phosphorus average content (Table 1) regarding the clay class (class 3-21 to 40%) is in the agronomy interpretation band regarded as low; while the potassium average content referring to the CTC (pH 7.0) (from 5.1 to 15.0 cmol c dm -3 ), is in the high band in the total area and GL-mo and medium in GL-lv and GL-mo.lv.In general (Table 1), it could also be observed that the majority of medium content of the attributes under analysis were found in GL-mo, following a tendency of reduction in the average content as a function of the systematization plan adopted in the area, since the larger volume of soil section was observed in GLmo.lv.Regarding the Na high average content observed in Table 1, it is believed to result from the use of saline water to irrigate the area, which comes from the Mirim Lake.
By analyzing data dispersion around the average (Table 1), expressed by the coefficient of variation (CV), the pH and V values in the total area were verified to present low variability (CV ≤ 15 %), while the C, Ca, Mg, S, H+Al, CTC, Clay and K data variability is classified as moderate (15 < CV ≤ 35 %), according to the classification proposed by Wilding;Drees (1983).While the Na (CV = 37.6 %) and P (CV = 49.0 %) data dispersion is considered high (CV > 35 %).The same behavior can be observed regarding the data set evaluated in GL-lv (Table 1).On the other hand, the data sets evaluated in the soil mapping units GL-mo and GL-mo.lvpresented distinct behavior in relation to the CV behavior, and the data low variability was predominant in GL-mo.lv(50% of the soil attributes evaluated) and the moderate data variability in GLmo (8 out of 12 data sets evaluated).
By analyzing the data in Table 1, it was seen that the P data sets presented the highest CV values when compared to the remaining sets, varying from 37.8 % (GL-mo.lv) to 54.2 % (GL-lv).Such results indicate that, for the dimensioning of samples with non spatial statistical methodology, when considering the total area as an experimental unit, sample sizes can be over or under estimated in some units, that is, units with high variability might have an underestimate of the sample size while more homogeneous units might be using a sampling effort above the necessary to represent the area.Descriptive statistics also indicates that the use of uniform management in the whole experimental area might lead to results which do not meet the agricultural needs of the soil under study, that is, the need to apply Precision Farming is verified and, consequently, the analysis of the soil mapping units influence in the soil attributes spatial variability.
Based on Kolmogorov-Smirnov adherence test and considering a 5% p value for probability (Table 1), the pH, Ca, Mg, Na, H+Al, Clay and P data distribution in the total area and in the GL-lv was not verified to tend to normality, confirming the attribute behavior previously seen in relation to data variability around the average.In GL-mo, the Na, H+Al, Clay and P data distribution did not follow normal behavior, while in GL-mo.lv the pH, Na and Clay values did not tend to normality, indicating that these attributes presented distribution localized in the area (GREGO et al., 2006) and that the arithmetic average cannot be considered as the distribution center (NIELSEN;WENDROTH, 2003).
The spatial variability structure of all attributes in the total area and in the soil mapping units was evaluated through the construction of experimental isotropic semivariograms.The adjustment parameters "Co", "C+Co" and "a" for the theoretical models were only obtained for the attributes that presented semivariograms in which the intrinsic hypothesis (semivariogram with a defined sill in the sampling space evaluated) was confirmed (Table 2).For data that followed normal distribution (Table 1), the semivariances were calculated via Matheron (1962) estimator, while for data that did not, they were calculated through the Cressie;Hawkins (1980) robust estimator, according to Webster;Oliver (2007).Table 2 shows that the experimental semivariograms were adjusted, for most of the soil attributes, to the spherical model regardless of the evaluation scale under analysis (Table 2).The spatial dependence degree of the attributes under evaluation in the total area was classified as moderate (25 % < SDD ≤ 75 %), while in the soil mapping units GL-mo, GL-lv and GLmo.lv most of their SDD was classified as strong; the degree of spatial dependency is the relationship, in percentage terms, of the nugget (Co) to sill (Co + C), i.e. the relationship between the unexplained variance or random, caused by measurement errors or the micro-variability of the property under study which cannot be detected in the sample scale used by the total variance of the sample for stationary data (TRANGMAR et al., 1985).The magnitude of the nugget variance is important in kriging because it sets a lower limit to the size of estimation variance and, therefore, to the precision of the interpolation (TRANGMAR et al., 1985).Thus, kriging and the mapping carried out when considering total area must present lower precision in relation to the individual areas mapping.This fact is very relevant for the management systems that involve precision farming technology.According to Cambardella et al. (1994), variables that present strong SDD (SDD ≤ 25 %) are more influenced by soil composition characteristics such as original material, relief, climate, organisms and time.Thus, the semidetailed mapping units, which are more homogeneous compared to the reconnaissance mapping unit, with fewer inclusions of different types of soils, tend to have lower variability of their physical and chemical properties, thereby influencing the degree of spatial dependence of the soil properties.
The comparison of spatial variability structures of the attributes pH, Mg, Na, H+Al, V, Clay, P and K in the total area and in the soil mapping units was carried out through the scaled semivariogram technique (Figure 3), and in the scaled semivariogram estimated via Matheron (1962), the data sample variability was adopted as scale factor and in the scaled semivariograms built by the Cressie;Hawkins (1980) estimator, the sill "C+Co" (Table 2) was adopted as scale factor.
When Figure 3 is analyzed, the attributes pH and Na are observed to have shown higher coalescence of scaled semivariance values when compared to the other attributes, indicating the possibility of adjustment of an only semivariogram model regardless of the factor used for the scaling and the evaluation scaling under analysis, thus, characterizing more similarity in the spatial dependency structure.Montanari et al. (2012) pointed out that the adjustment of an only semivariogram model would provide an only sampling scheme for these attributes.The Mg, H+Al, V, Clay, P and K attributes spatial variability in GL-mo.lv were observed to be similar to the scaled semivariogram model structures in the total area, this result is not in accordance with the descriptive statistics result (Table 1), but indicates the distribution of high and low values in the total area similar to that in the soil mapping unit GLmo.lv, altering the magnitude of values and, therefore, influencing the descriptive analysis.
The greatest difference in behavior between scaled semivariograms can be seen through the potential acidity (H+Al) as a function of the soil mapping units, followed by the semivariogram behavior for clay content (Figure 3) indicating that estimate errors when considering total area might be more accentuated for these attributes.This more marked difference for these attributes can derive from the cut and fill performed in the surface layer by the land leveling, which brought to the surface the lower parts of the A horizon or even the top of the B horizon in some parts of the area, with different physical and chemical properties compared to the natural surface more organic horizon.

CONCLUSIONS
There is no correspondence in the attributes spatial distribution and structure between the semidetailed mapping units and the reconnaissance mapping unit in which they are included.
The use of non-spatial statistics in the management decisions might lead to unsuitable recommendations to the crop if spatially different soil mapping units are present in the area.
The scaled semivariogram technique allows the identification of similar or different spatial behaviors between the soil mapping units, contributing to the attributes sampling plan.

ACKNOWLEDGEMENTS
To Granja Bretanhas for the general support throughout the development of the study, to CAPES and CNPq for the Master's and research productivity grants.

Figure 1 .
Figure 1.Location of the study area and the semi-detailed survey in relation to the soil reconnaissance survey in the State of Rio Grande do Sul (STRECK et al., 2008), municipality of Jaguarão, Rio Grande do Sul.

Figure 2 .
Figure 2. Georeferenced points in the study area and its homogeneous areas.

Table 1 .
Results of applying descriptive statistics to the soil attributes under evaluation in the total area and in the soil mapping units classified as Mollic Gleysol (GL-mo), Luvic Gleysol (GL-lv), and Luvic Mollic Gleysol (GL-mo.lv).

Table 2 .
Experimental and theoretical semivariograms with respective adjustment parameters (nugget effect Co, sill C+Co and range a) and residual sum of squares (RSS) and the spatial dependence degree (SDD) of the attributes evaluated in the total area and in the soil mapping units.