Magnetic susceptibility for characterizing areas with different potentials for sugarcane production

The objective of this work was to evaluate the use of magnetic susceptibility for characterizing the spatial variability of soil attributes and identifying areas with different potentials for sugarcane (Saccharum spp.) production. Samples were collected at 110 points (1 per 7 ha) in the layers of 0.00–0.20 and 0.20–0.40 m, to determine the magnetic susceptibility and physical and chemical attributes of the soil. Fiber content, sucrose polarization (POL), and sugarcane yield were determined in 33 points. The spatial variability model for magnetic susceptibility was 63 and 22% more accurate in delimiting soil potential for sugarcane production than soil physical and chemical attributes at the 0.0–0.2 and 0.2–0.4-m layers, respectively. The spatial variability map for magnetic susceptibility was strongly correlated with clay (0.83 and 0.89, respectively, for the layers) and sand contents (-0.84 and -0.88); moderately correlated with organic matter (-0.25 and -0.35), sum of bases (-0.46 and 0.37), cation exchange capacity (0.22 and 0.47), pH (-0.52 and 0.13), and POL (0.43 and 0.53); and weakly correlated with sugarcane yield (0.26 and 0.23). Magnetic susceptibility can be used to characterize the spatial variability of soil attributes and to identify areas with different potentials for sugarcane production.


Introduction
The second survey on the 2015/2016 sugarcane (Saccharum spp.) harvest in Brazil estimated national production to be 663.11 million tons (Acompanhamento…, 2015).The state of São Paulo, with 4.64 million hectares, is the primary sugarcane producer in the country.This shows the importance of having accurate and detailed information on soils, which can help to make decisions on crop management (Peluco et al., 2013).
The rapid and accurate characterization of the spatial variability of soil attributes can facilitate the development of effective, local management practices.Moreover, this information is used to develop global indexes on changes in land use (Rockström et al., 2009) and tools that enable sustainable agriculture production (Siqueira et al., 2010b;Peluco et al., 2015).
According to Legros (2006), soil mapping methods can be divided in: mesh mapping observations, which map the similarity of different pedons; free or categorical mapping, which uses concepts of soillandscape to set the locations of samples; mapping using geostatistical analysis; mapping using fuzzy logic; and mapping of the land use potential.Some researchers (McBratney et al., 2000) proposed hybrid mapping, involving different methods with geostatistics, in order to improve the quality of the spatial information maps.
However, the precision and accuracy of these procedures depend on the number of observation points (Siqueira et al., 2014), which are related to the costs of these mappings.Other factors, such as the time required to analyze soil properties and the environmental impacts caused by laboratory reagents, are also limitations to routinely perform these procedures.This has raised the need to develop alternative techniques for the indirect quantification of soil attributes (Siqueira et al., 2010a).
Some researchers have explored the possibility of representing the site-specific soil mapping of physicochemical attributes using electrical conductivity (Weller et al., 2007).However, the concentrations of iron oxides, abundant in tropical soils, can affect soil electrical properties (Wu et al., 2008).In Brazil, incipient researches have been carried out with geophysical tools for the indirect quantification of soil attributes (Matias et al., 2014).
Soil magnetic susceptibility is another technique with great potential for mapping and for indirect quantification of soil attributes (Resende et al., 1988;Camargo et al., 2014;Matias et al., 2014).According to Marques Jr. et al. (2014), it can be used to increase the accuracy in the delimitation of areas with different variability patterns of soil properties.Magnetic susceptibility measurements were used in the earliest experiments on the qualitative characterization of soils in the state of São Paulo, Brazil (Resende et al., 1988).Recently, Siqueira et al. (2010a) used it for the quantification of physical, chemical, and mineralogical attributes of soils with low iron content, i.e., less than 4% as Fe 2 O 3 .Magnetic susceptibility has also been used in sugarcane plantations in Brazil to identify areas with different potentials for soil CO 2 emission (Barrios et al., 2012;Leal et al., 2015) and with different capacity to endure vinasse application (Peluco et al., 2013), as well as to quantify adsorbed phosphorus (Peluco et al., 2015) and to map management units (Matias et al., 2015;Siqueira et al., 2015).However, few works focus on identifying the relationship between magnetic susceptibility and plant characteristics (Marques Jr. et al., 2014).
The objective of this work was to evaluate the use of magnetic susceptibility for characterizing the spatial variability of soil attributes and identifying areas with different potentials for sugarcane production.

Materials and Methods
The studied area (Figure 1) is located in Guariba, a municipality in the northeast of the state of São Paulo, Brazil (21°23'S, 48°11'W, at 674 m above sea level).The climate of the region is mesothermal with dry winters, Cwa, according to Köppen.The average annual rainfall is 1,400 mm, and the rainy period lasts from November to February.The natural vegetation consists of sub-acidophilic tropical forest plants, and the land is currently cropped with sugarcane, which is harvested mechanically without burning of crop residues.The area belongs to the geological province of the western plateau of the state of São Paulo, and its geological material is a combination of sandstones of the Bauru group of the Adamantina formation and of basalt of the Serra Geral formation (Mapa…, 1981).The soils are classified as Latossolo Vermelho-Amarelo distrófico (soil taxonomy; Typic Haplustox) and Latossolo Vermelho distrófico (soil taxonomy: Typic Haplustox).
The soil map (scale 1:10,000) was prepared by Centro de Tecnologia Canavieira, the Brazilian center for sugarcane technology (Carta…, 2008).Three production environments were identified for sugarcane production (Figure 1).In order to map these production environments, edaphic factors and the potential for agricultural production should be considered (Lepsch, 1987;Maule et al., 2001).The average of soil total iron content (Fe 2 O 3 , extracted by sulfuric acid) were 45.7 and 76.1 g kg -1 , respectively, at the 0.00-0.20 and 0.20-0.40-msoil layers.
Pesq. agropec.bras., Brasília, v.51, n.9, p.1349-1358, set.2016 DOI: 10.1590/S0100-204X2016000900034 A total of 110 points were sampled at both depths in 2007, after mechanical harvest, in an irregularly spaced grid, spanning an overall area of 770 ha.The sampling density was 1 sample per 7 ha (Figure 1).The highest sample density was observed in the northsouth direction, coinciding with the direction of greater variability in soil attributes (Camargo et al., 2014;Matias et al., 2015).The area was split into 33 plots, covering 24 ha in average.Sugarcane plants were assessed in each plot, also in 2007.
Soil samples were dried and sieved through a 2-mm mesh, prior to grain size and chemical analyses.The granulometric analysis was performed using the pipette method, using 0.1 N NaOH as a chemical dispersant under slow mechanical stirring for 16 hours (Claessen, 1997).Exchangeable bases (Ca, Mg, and K) and potential acidity (H + Al) were determined according to Raij et al. (1987).The organic matter (OM) content and pH were obtained as described in Claessen (1997), whereas the magnetic susceptibility was determined according to Siqueira et al. (2010a).The apparatus for magnetic susceptibility determination consists of an analytical balance, a magnet, a support for the magnet, and the sample port.The interaction of the minerals showing magnetic expression with the magnet generates a weight force measured in the analytical balance.This force is converted into magnetic susceptibility weight using a standard curve.Siqueira et al. (2010a) showed that this method is feasible and that its results correlate well with sensors of magnetic susceptibility (r = 0.94).
Thirty three samples of sugarcane segments were collected to evaluate the following plant properties: production, consisting in cane tons per hectare (CTH); polarization (POL), a measure of the sucrose content; and fiber content (Consecana, 2003).
Data processing was initially done with descriptive statistics, including mean, minimum, maximum, and coefficient of variation (CV).Soil attributes were analyzed in terms of their spatial dependence.The experimental semivariogram was estimated from the following expression: ), soil survey map scaled at 1:10.000 (B), production environment for sugarcane (Saccharum spp.) (C), and sketch of the sampling program (D).For the production environment, A is a high productivity environment, with yield >95 Mg ha -1 ; B is a high/medium productivity environment, with yield of 90-95 Mg ha -1 ; and C is a low productivity environment, with yield of 80-85 Mg ha -1 .
Pesq. agropec.bras., Brasília, v.51, n.9, p.1349-1358, set.2016 DOI: 10.1590/S0100-204X2016000900034 in which γ(h) is the experimental semivariance for a separation distance h; z(x i ) is the attribute value at the i th point; and N(h) is the number of pairs of points separated by distance h.The semivariogram model was adjusted based on cross-validations (mean and variance of residuals), external validation, and the coefficient of determination (R 2 ).The experimental semivariograms were modeled with 100 points.The rest of the points (n=10) were used to calculate the relative root mean square error (RRMSE), for external validation.This procedure is important to eliminate feedback in model validation.
The performance of the spatial interpolation method (ordinary kriging) for soil attributes was assessed from RRMSE values, which can be considered as an accuracy index of the model.The validation methods are important to compare the models for variables with different measurement units (Li & Heap, 2008), since they express their accuracy in terms of percentage.RRMSE was determined with the expression:  in which n is the number of values used in the validation; z(x i ) is the attribute value at the i th point; and ẑ(x i ) is the estimated attribute value at the i th point.Since plant attributes were assessed at a reduced number of points (33), the ordinary kriging interpolation was impossible.Therefore, the interpolation by inverse distance weighting, a non-geostatistical method, was used instead (Kravchenko & Bullock, 1999).The interpolation values for soil and plant attributes were estimated and formed a refined grid with a total of 893 points, regularly separated by 79 m.Similarities in the spatial distributions of the attributes were established by spatial correlation between magnetic susceptibility maps and soil or plant attributes.
The experimental semivariograms obtained were used to construct a scaled semivariogram according to Vieira et al. (1997): γ i sc (h) = γ i (h)/α i (i = 1, 2…m), in which γ i sc is the semivariance of the scaled semivariogram at distance h; γ i is the original semivariance at distance h; α is the scaling factor, in this case it is the variance of the attribute; and i is the number of attributes.

Results and Discussion
According to the classification proposed by Warrick & Nielsen (1980), the variability in pH values was low (CV=12%); in clay contents, total sand, and cation exchange capacity (CEC), it was moderate (12%<CV<24%); and in sum of bases (SB), OM, and magnetic susceptibility, it was high (CV>24%), at both evaluated depths (Table 1).Of these attributes, magnetic susceptibility was the one that exhibited the highest CV.This reflects the high sensitivity of this attribute to soil processes, which change continuously across the landscape.
All attributes showed spatial dependence structure, and data in the experimental semivariograms were fitted to spherical and exponential models (Table 2).The spherical model is better adjusted for variables with abrupt transitions along the area, while the presence of an asymptotic sill in exponential models is responsible for smoother transitions in space.All studied attributes exhibited a moderate to high degree of spatial dependence (DSD), as observed by Pesq.agropec.bras., Brasília, v.51, n.9, p.1349-1358, set.2016 DOI: 10.1590/S0100-204X2016000900034 Cambardella et al. (1994).Clay contents, pH values, OM contents, and CEC values had moderate DSD at both studied depths.The magnetic susceptibility attribute showed high DSD at 0.0-0.20 m, and moderate DSD at the higher depth.This suggests that the attribute is location-dependent and, therefore, related to the local soil formation processes.According to Maher & Thompson (1991), magnetic minerals store natural records of soil formation factors and processes.
The spatial variability of soil attributes may be affected by both intrinsic and extrinsic factors; the latter are usually related to soil management practices (Cambardella et al., 1994).In this regard, Oxisols are assumed to exhibit relatively uniform attributes across their profiles.However, an Oxisol cropped with sugarcane for more than ten years cannot be considered uniform if its physical and chemical attributes show spatial variability, irrespectively of land use.
The range parameter of the semivariogram indicates the DSD between samples (Table 2).In practical terms, it can be used to indicate the distance between samples in the field, which ranged from 630 to 1,700 m in the present study.The range for magnetic susceptibility was close to that of the other soil attributes, at both sampling depths.
At both depths, the exponential models were adjusted for the scaled semivariogram of soil attributes (Figure 2), and data on magnetic susceptibility were fitted to the spherical model.Marques Jr. et al. ( 2014) also found magnetic susceptibility to fit a spherical model.The range parameter of the scaled semivariogram, at the 0.0-0.2 and 0.2-0.4-mlayers, respectively, was 1,326 and 990 m for the soil physical and chemical attributes, and 1,450 and 1,650 m for magnetic susceptibility.This suggests that the spatial variability pattern of magnetic susceptibility can be used to examine the spatial variability of soil physical and chemical attributes.Matias et al. (2014) reported similar spatial variability pattern between magnetic susceptibility and soil physical and chemical attributes in the same area of study.
The nugget effect (C 0 ) may indicate lower error in the representation of spatial variability with the semivariogram model.The C 0 parameter of the scaled semivariogram for the soil physical and chemical attributes was 0.40 and 0.45, at the 0.0-0.2 and 0.2-0.4-mlayers, respectively.For magnetic susceptibility, the C 0 values were 0.15, at the 0.0-0.2-mdepth, and 0.35, at 0.2-0.4 m.Accordingly, error representation of the spatial variability of magnetic susceptibility was 63 and 22% [(1-C 0 shp/C 0 exp)100] smaller than that of the scaled semivariogram of the soil physical and chemical attributes at the 0.00-0.20 and 0.20-0.40-mlayers, respectively (Figure 2).
(2) SB, sum of bases; CEC, cation exchange capacity; OM, organic matter content; and MS, magnetic susceptibility (n=100).15-32% in soil texture are common in Brazilian laboratories.Due to the high relationship between magnetic susceptibility and physical, chemical, and mineralogical attributes of the soil -identified by linear correlations (Siqueira et al., 2010a;Camargo et al., 2014;Marques Jr. et al., 2014;Siqueira et al., 2014) or by the similarity between range values in the semivariograms or by the error in the representation of the spatial variability structure (smaller C 0 value) -, magnetic susceptibility can be used to improve the identification of boundaries between soil areas in the field.In addition, the lower cost in magnetic susceptibility determination allows analyzing a higher number of samples than in the conventional analyses.
With the exception of OM and CEC, few variations were observed in RRMSE at different depths in the studied attributes (Table 3).The largest uncertainties in OM at the 0.0-0.2-mdepth were related to the greater variability of this attribute at this depth, i.e., higher C 0 value (Table 2).CEC showed lower accuracy at the 0.2-0.4-mdepth, which agrees with the greater heterogeneity of this property in the deepest layer, that is, the lower range value.The RRMSE values found for magnetic susceptibility (32% for 0.0-0.2m and 33% for 0.2-0.4m) were similar to those reported by Li et al. (2007), for estimating the electrical conductivity of the soil.This reinforces the idea that magnetic susceptibility might replace electrical conductivity as a measure for indirect quantification of soil attributes in areas where this technique may have limitations.
Relating the spatial distribution of magnetic susceptibility with that of other soil attributes can be of great assistance towards understanding pedogenetic processes and obtaining useful information on soil formation conditions in specific environments.This can help, for example, to relate the magnetic properties of soil minerals to the environment (Maher & Thompson, 1991), which would help the proposition of accurate prediction models for other locations.
Magnetic susceptibility was especially well correlated with the soil clay content (Table 4).Matias et al. (2014) also found similarities between magnetic susceptibility and clay content maps in the same area of study, whereas Fontes et al. (2000) observed increased values of some magnetic properties with increasing clay contents.CEC exhibited positive spatial correlation with magnetic susceptibility, since both attributes are directly dependent on soil mineral composition.However, magnetic susceptibility was negatively correlated with SB, at the 0.00-0.20-mdepth.SB and base saturation (BS) are fertility attributes related to the clay fraction, but without magnetic expression.The variation of CEC in tropical soils is mainly caused by the variation of OM and iron oxide contents, such as hematite, goethite, and ferrihydrite (antiferromagnetic minerals) (Dearing, 1994).These minerals have low magnetic expression but are covariates of maghemite and magnetite, ferrimagnetic minerals, which have high magnetic susceptibility.Kaolinite and gibbsite are clay minerals and, as such, affect CEC, but they do not have magnetic expression (Dearing, 1994).The change in direction of the correlation of magnetic susceptibility with BS and pH with depth is associated with parent material.According to Camargo et al. (2014), the occurrence of mudstone intrusion in the  sandstone material is common in the studied region and increases the concentrations of magnetic minerals with depth.
Highly weathered soils, such as the studied Oxisol, contain large amounts of variably charged Al and Fe oxides and hydroxides.Therefore, correlating CEC with magnetic susceptibility can help estimate the variation potential of charge exchange capacity at a given location.This result is interesting, since it indicates that magnetic susceptibility can be used to identify soil areas with greater or lesser nutrient uptake and sensitivity to herbicide molecules.However, this would require prior characterization of the spatial variability of magnetic susceptibility, as well as the determination of its relationship with other soil attributes.
Sugarcane attributes exhibited positive correlation with magnetic susceptibility (Table 4).Marques Jr. et al. ( 2014), however, reported negative correlation between POL values and magnetic susceptibility.It should be noted that this qualitative attribute was better correlated with magnetic susceptibility than it was with cane yield (CTH), a quantitative attribute.Therefore, magnetic susceptibility shows potential to be used in the identification of sites where soil conditions favor the quality of sugarcane.
The soil Latossolo Vermelho-Amarelo distrófico showed lower magnetic susceptibility than the Latossolo Vermelho distrófico.Since the studied area is in a border region between sandstone and basalt, the increase in magnetic susceptibility observed in the Latossolo Vermelho distrófico was probably due to the relation of this soil type with basalt, which is a parent material with greater magnetic susceptibility than sandstone.Fontes et al. (2000) described the variability of magnetic susceptibility for different parent materials and concluded that magnetic susceptibility is always higher in soils derived from basalt, in comparison with the ones from sedimentary rocks.
Comparing the map of production environments (Figure 1) with those of magnetic susceptibility and plant attributes (Figure 3), it was possible to perceive potential for greater yields and processing quality with increasing SM values (>12 10 -6 m 3 kg -1 ).Therefore, magnetic susceptibility can be used to facilitate the acquisition of quantitative information from soil that can be used together with crop information for mapping purposes.However, more works are needed to clarify the relationship between these attributes.

Figure 3 .
Figure 3. Spatial distribution maps for soil attributes (A and B: MS, magnetic susceptibility; and C and D, clay) and sugarcane (Saccharum spp.) plant attributes (E, CTH, cane tons per hectare; F, Pol; and G, fibre).

Table 3 .
Relative root mean square error (RRMSE) used for external validation of soil attributes.

Table 4 .
Spatial correlation of the magnetic susceptibility map with soil and sugarcane (Saccharum spp.) attributes.