Discrimination of geomorphic surfaces with multivariate analysis of soil attributes in sandstone-basalt lithosequence 1

The geomorphic surface concept allows interrelationship among various branches of soil sciences, such as geology, geomorphology and pedology. This association enhances the understanding of spatial soil distribution through landscape, pointing out the soil attributes behavior, which are mainly related to stratigraphy and relief forms. Therefore, this study aims to apply multivariate statistics to categorize geomorphic surfaces in sandstone basalt lithosequence, so as to provide a basis for soil assessment in similar areas. The study area is located in Pereira Barreto County, SP, Brazil. An area of 530 hectare was selected, where three geomorphic surfaces (I, II and III) were located and mapped. In this area, 134 soil samples were collected at depths of 0.0-0.2 m and 0.8-1.0 m below ground surface. Sand, silt and clay contents were determined, pH in CaCl2 solution, OM, P, Ca, Mg, K, Al and H+Al contents were also evaluated. Based on the results, univariate, multivariate analysis of variance, cluster and principal-component analysis were performed in order to compare the three geomorphic surfaces. The univariate statistical analysis of soil attributes was not efficient enough to categorize the three geomorphic surfaces. By using the physical and chemical soil properties, the multivariate statistical techniques enabled the differentiation of the three groups of soil natural bodies which were equivalent to the same three mapped geomorphic surfaces (GS). These results are interestingin order to demonstrate the feasibility of the numerical classification use on geomorphic surfaces to assist the soil mapping.


INTRODUCTION
The digital mapping of soil classes generally starts with soil profile description organizing the soil classes at a taxonomic level in a particular classification system.The current methodology treats soil classes as 'labels' and their prediction only considers the minimization of the misclassification error.Soil classes at any taxonomic level have taxonomic relationships between each other, and in some instances the errors in prediction of certain classes are more serious than the others (MINASNY; MCBRATNEY, 2007).
In this sense, some authors (CAMPOS et al., 2007;CUNHA et al., 2005;SANCHEZ et al., 2005;TERAMOTO et al., 2001) have been using geomorphic surfaces to assist in more accurate transition lines identification between the involved regions, and help in understanding of greater or lesser variability space areas.
Conceptually, geomorphic surfaces are land portions defined by geographic boundaries and located within time and space (DANIELS et al., 1971;RUHE, 1956).The knowledge and practice of these soil study concepts enable the performance of spatial variability studies and pedological assessments.In addition, it consists in an instrument to predict pedological features from still unknown areas (MOTTA et al., 2002a).
Hence those studies on soil variability and its geomorphological attributes are aid tools in pedology studies, since they do not consider the pre-established taxonomic limits, but rather follow soil limits as natural bodies.Thus, they improve interpretations in assessments for land suitability studies, capacity use, managing zone establishment and etc (CUNHA et al., 2005).
A tool that has been used in such research is the multivariate analysis.The use of multivariate statistical techniques associated with geomorphic surface concepts make it possible to observe the soil attributes variation, thus consisting of an attempt to reduce error and to understand the sequences of pedogenetic processes, and clarifying the participation and importance order of soil variables (YEMEFACK et al., 2005).The use of this techniques will categorize clusters in such a way that error rate can be classified as minimal, thus providing important information to give accurate interpretation of land use planning (VASELLI et al., 1997), landscape understanding, soil attributes (FU et al., 2004;SENA et al., 2002;SOUZA et al., 2006), behavior as well as its spatial distribution, studies on soil genesis and classification (GOMES et al., 2004).Siqueira et al. (2010) proposed the use of the soil landscape model and multivariate analysis to identify potentially productive areas in landscape for citrus orchard.
This study aimed to apply multivariate statistical analysis to discriminate between the geomorphic surfaces in a sandstone/ basalt transition located in Pereira Barreto region, São Paulo State, Brazil, in order to support soil assessment over similar areas.

MATERIAL AND METHODS
The study area is located in Pereira Barreto County, northeast São Paulo State, in geographical coordinates of 20 o 41'15" S latitude and 51 o 03'45" W longitude.The region has savanna climate (Aw), according to Köppen's classification, presenting wet summers and dry winters.It lies on a land with relatively low, a plain, and the temperature ranges between 21.2 and 26.8 Celsius degrees (°C) and an annual rainfall of 1.128 millimeters (mm).
The area is currently under transition management from pasture to sugarcane cultivation and includes the geomorphological province of the Western Plateau, with the Latosol predominance distributed downhill on linear and convex profiles.
On the flat hilltop surface, a typical medium texture dystrophic Red Latosol (Haplustox) is found, whose original material proceed mainly from sandstone belonged to Santo Anastácio Formation, gradually changes downslope into a clayey texture, Eutroferric Red Latosol (Eutrustox), that is mainly originated from the products of basalt alteration from Serra Geral Formation.
A 530 hectares area was mapped using GPS unit device and the geomorphic surfaces were identified and delimited according to criteria proposed by Ruhe (1956) and Daniels et al. (1971).Three geomorphic surfaces (I, II and III) were located and mapped (Figure 1).A number of one hundred and thirty-four soil samples were collected from these geomorphic surfaces at 0.0-0.2m and 0.8-1.0m depths, in grid shape, to the effect of confirming the occurrence of soil classes.
The soil samples were collected and classified according to criteria established by Embrapa (2006), as typical dystrophic Red Latosol (Haplustox) in geomorphic surface I; typical eutrophic Red Latosol (Eutrustox) in geomorphic surface II; typical eutrophic Red Latosol, typical eutrophic Litholic Neosol (Orthents) and chernozemic eutroferric Red Latosol (Eutrustox) in geomorphic surface III (Table 1 and Figure 1).Silt, sand and clay contents and also organic matter content were determined by the pipette method according to the Empresa Brasileira de Pesquisa Agropecuária (1997) methodology.Active acidity (pH) was potentiometrically determined in CaCl 2 by using a ratio 1:2.5 soil to CaCl 2 solution measuring the potential acidity (H+Al), according BS -base sum; V -base saturation; OM -organic matter, Fe 2 0 3 -total ferric iron to Raij et al. (2001).Phosphorus (P), calcium (Ca), magnesium (Mg) and postassium (K) were extracted from the soil by an ion-exchanging resin (RAIJ et al., 2001).
Using results from samples collected in the different geomorphic surfaces, univariate (ANOVA) and multivariate (MANAVA) analyses of variance were performed conjointly with predefined values for contrast in order to compare them.
The conjoint action of granulometrical (fine and coarse sand, silt and clay) and chemical (pH in CaCl 2 , P, OM, Ca, Mg, K and H+Al) attributes were also evaluated by multivariate statistics, principal-component analysis and cluster (MORRISON, 1967;SNEATH;SOKAL, 1973) to discrimination the geomorphic surfaces.
The original data were standardized aiming to minimize the effects of the various measuring scales.At this phase, a convertion to normalized scores (distributed, with an average in 0.0 and the standard deviation was 1.0), in order to reach this result it must be subtracted the average and divided by the standard deviation (FERREIRA, 2008).
The principal components analysis, in order to obtain a larger set of linear combination variables, would preserve the most information provided by the original variables (fine sand, coarse sand, silt, clay, pH in CaCl 2 , P, OM, Ca, Mg, K and H+Al).Due to this, there was a selection of original attributes leading to a smaller set of attributes that had preserved the information from original attributes and reducing the two principal components (PC1 and PC2), through which the identified units were represented in a bi-dimensional graphic.The used criteria in selecting the principal interpreted components was on the percentage of variance explained.According to Carvalho et al. (2004) choose the first components that accumulate a variance explained percentage of about 70%.The correlation matrix was composed of 11 variables measured at 67 points.Based on the most important attributes for PC1 at the 0.0 -0.2 m depth (clay, silt and calcium) and 0.4 -0.6 m depth (clay fine sand and calcium) the group analysis was used to construct a dendrogram.
The cutoff for the dendrogram which defines the number of groups was obtained by "watching" method, where the researcher specifies the level of grouping for convenience (ALBUQUERQUE, 2005, BARROSO;ARTES, 2003, SNEATH;SOKAL, 1973).It was chosen as the cutoff the mean euclidean distance (4.5).Single linkage cluster was used to obtain other sequential, agglomerative, hierarchical, non-superposed groups expressing the results by means of hierarchical-scheme graphs or dendograms.The similarity coefficient used for cluster analysis (enabling the dendograms design) was the mean Euclidean distance between the studied geomorphic surfaces.The data were processed in Statistica software version 7.0.

RESULTS AND DISCUSSION
The averages of physical and chemical soil attributes located on the three geomorphic surfaces are presented in Table 2.The chemical soil attributes such as pH, calcium, magnesium, and potassium have shown an increasing trend towards the more rejuvenated geomorphic surfaces on transects (geomorphic surfaces III).This reflects the soil source material influence mainly in 0.8-1.0m depth.Similar results were found by Cunha et al. (2005) in sandstone to basalt transition soils.
The soil clay content increases from I to III geomorphic surfaces, which is associated with source material variation and the weathering action emerges as Montanari et al. (2010).For the coarse and fine sand means the behavior is, of course, contrary to this trend (Table 2).Anjos et al. (1998), while studying the soil genesis and their relationships with the landscape in southeastern Brazil, concluded that geomorphic surfaces define weathering rates and the degree development of Solum and the flowing behavior of the water, which coordinate not only the illuviation but also the cations accumulation processes.
When individually analyzed by the univariate analysis of variance (ANOVA) at both depths, the sand and silt attributes have presented the same behavior identified only in two geomorphic surfaces.However, the clay attribute has shown significative differences among those three geomorphic surfaces (Table 3).For t h e c h e m i c a l a t t r i b u t e s ( p H , O M , P, K , C a , M g , H + A l ) , a t both depths, there was no clear discrimination between the geomorphic surfaces under study.Therefore, it was not possible to confirm the occurrence of three geomorphic surfaces by using this method (Table 3).
It has been observed that the sand content presented increasing behavior from I to III geomorphic surfaces presenting a sandy texture in Dystrophic Red Latosol (Haplutox) and Eutrophic Red Latosol (Eutrustox) (Table 1).This information has helped in geomorphic surface distinction, since those raised sand contents come from original material, in this case the geomorphic surface I located over the sandstone and the geomorphic surface III over the basalt.
The results from the multivariate analysis of variance (MANAVA) for granulometric attributes (clay, silt, and sand) have presented significant differences among the geomorphic surfaces for all tested contrasts (Table 4) at 0.0-0.2m and 0.8-1.0m depths, thus differentiating three environments in agreement with the three previous identified geomorphic surfaces.For chemical attributes, it was observed that, at 0.0-0.2m depth, there was significant difference for all tested contrasts (Table 4).With regards to 0.8-1.0m depth, a significant difference was observed for all contrasts, except for GS I vs. GS II contrasts, which did not present significant difference (Table 4).According to Webster and Oliver (1990) in the MANAVA analysis are used statistical tests with multiple variables to investigate the likehood ratio test of the hypothesis (Wilks) or null hypothesis of no treatment effects (Roy, Hotelling-Lawley e o de Pillai).The tests differentiate themselves due to the used criteria to evaluate the treatments' diference: trace (Hotelling-Lawley e Pillai) and largest root (Roy e Wilks), being Wilks' test the most used in multivariate analysis of variance.These tests results have shown that multivariate analysis, independent of the used test, it is more efficient in distinguishing landscape compartments than the univariate statistics (YEMEFACK et al., 2005).
In the pedological assessment, univariate statistical criteria are usually used in order to establish taxonomic limits in the discrimination and separation of soil classes.For Hudson (1992) and Young and Hammer (2000), these taxonomically pre-established limits are considered to be artificial.On the other hand, the cluster strategy based on multivariate statistics allows for more complete information concerning soils distinction in the conceptual sense of the natural body (CUNHA et al., 2005;YOUNG;HAMMER, 2000).

Table 4 -
The p-values from the multivariate analysis of variance (MANAVA) tests regarding to soil granulometric (fine sand, coarse sand, silt e clay) and chemical attributes (pH in CaCl 2 , P, OM, Ca, Mg, K and H+Al) on different geomorphic surfaces at 0.0-0.2m and 0.8-1.0m depths The results of the grain and chemical attributes analysis, main components at 0.0-0.2m 0.8-1.0m depths are presented in Table 5.The first principal component can be interpreted as a physical and chemical quality index of environment.Thus, PC1 and PC2, together, explain 74.77% of the total variance in 0.0-0.2m depth and 67.67% of the total variance in 0.8-1.0m depth (FIG.2).Sanchez-Maranón et al. (1996) and Splechtna and Klink (2001) found similar results working with the same soil attributes, mention that PC1 and PC2 explain about 60% of the total soil variation.

Contrasts
In the 0.0-0.2m depth, attributes that most contributed to the first component (PC1) were: clay, silt and calcium, and in the 0.8-1.0m depth, the attributes of greatest contribution to PC1 were: sand, calcium and clay (TAB.4).There was a negative correlation between the attributes of fine and coarse sand with PC1.Manlay et al. (2000), studying the relationships among the soil abiotic factors has also observed a negative correlation for the sand fraction.Thus, the attributes of greatest influence on the surface and subsurface horizons were calcium and clay, corroborating to the results found by Motta et al. (2002b), whom had studied the occurrence of "macaúba" (a native Brazilian palm) in Minas Gerais State and its attributing relationship with soil and vegetation.
Discrimination of geomorphic surfaces with multivariate analysis of soil attributes in sandstone -basalt lithosequence  The Figure 3 confirms the relationship among the (classification) of different land classes, when located within the same geomorphic surface.Analyzing the correlation values of clay, silt and calcium attributes (TAB.5) with the PC1 axis (FIG.2a), it can be observed that the soils located to the right of these attributes are soils developed from basalt.This same pattern occurred in 0.8-1.0m depth.Ogg et al. (2000) has mentioned that the groups follow the occurrence logic in the landscape.In this study, the soil occurrence logic follows the pattern of geomorphic surfaces.The results of cluster analysis are demonstrated in Figure 5. Groups created based on the three most important attributes for PC1 (Table 2) validated the limit classification of soil natural bodies in the landscape (HUDSON, 1992), because the boundaries of these bodies concur with the geomorphic surfaces boundaries mapped in the field.Webster and Oliver (1990) confirm the effectiveness of cluster analysis when examining and separating classes of geomorphic surfaces and soil incorporated within these areas.
According to Fu et al. (2004), the idea that the multivariate statistic allows the viewing of variability within a minimal group and maximum variability among groups is applied in this study.In studying the relationship between topography and plant diversity, by means of multivariate statistics, it was observed that the landscape allowed the distinction of groups with different variability.
Analyzing the clusters, it was observed that within a Euclidean distance of 4.5 there was an even number of groups in the 0.0-0.2m depth (Figure 3a) and in 0.8-1.0m depth (Figure 3b).This indicates a consistency in both depths, reinforcing the concept of Latosol (Oxisol) concerning the homogeneous depth distribution of clay (EMPRESA BRASILEIRA DE PESQUISA AGROPECUÁRIA, 2006).
According to Adams et al. (1992), cluster analysis in soil studies favors the organization of similarity degree; therefore, its use is also indicated for taxonomic purposes.According to Young and Hammer (2000), a more detailed soil study using cluster analysis may show important

CONCLUSIONS
1.The individual mean test comparison of the soil attributes were not efficient to discriminate the three mapped geomorphic surfaces; 2. The use of multivariate statistical techniques (cluster and principal component analysis) enabled the separation of three groups of soil natural bodies that were equivalent to the mapped geomorphic surfaces; 3. The identification of geomorphic surfaces should be used to assist in soil surveys in order to better map the precise boundaries between different soil types or areas with different patterns of soil attributes.

Table 1 -
Description of the area soil profiles

Table 5 -
Correlation of soil attributes between the first two principal components and classification of attribute scores according to their contribution