Bioscience Journal | 2023 | vol. 39, e39015 | ISSN 1981-3163 1 Cassio Rafael Costa DOS SANTOS1 , Augusto Takayuki MATSUNAGA2 , Luiz Rodolfo Reis COSTA2 , Mario Lima DOS SANTOS3 , Alberto Bentes BRASIL NETO4 , Richard Pinheiro RODRIGUES2 , Maria de Nazaré Martins MACIEL2 , Vânia Silva DE MELO2 1 Universidade Federal Rural da Amazônia, Capitão Poço, Pará, Brazil. 2 Soil Science Department, Universidade Federal Rural da Amazônia, Belém, Pará, Brazil. 3 Postgraduate student at Forestry Engineering Department, Universidade de Brasilia, Brasília, Distrito Federal, Brazil. 4 Instituto Federal de Educação, Ciência e Tecnologia do Pará, Santarém, Pará, Brazil. Corresponding author: Cassio Rafael Costa dos Santos cassio.santos@ufra.edu.br How to cite: DOS SANTOS, C.R.C., et al. Spatial variability of soil fertility under agroforestry system and native forest in eastern Amazonia, Brazil. Bioscience Journal. 2023, 39, e39015. https://doi.org/10.14393/BJ-v39n0a2023-62830 Abstract The usage of spatial tools might be helpful in the optimization of decision-making regarding soil management, with technologies that assist in the interpretation of information related to soil fertility. Therefore, the present study evaluated the spatial variability of chemical attributes of the soil under an agroforestry system compared to a native forest in the municipality of Tomé-açu, Eastern Amazon, Brazil. Soil samples were performed at 36 points arranged in a 55 x 55 m grid. The soils were prepared and submitted to analysis in order to determine pH in H2O, exchangeable calcium, magnesium, potassium and aluminium, available phosphorus, potential acidity, organic matter, bases saturation and aluminium saturation. For each soil attribute, the spherical, gaussian and exponential models were adjusted. After the semivariograms fitting, data interpolation for assessment of spatial variability of the variables was performed through ordinary kriging. The spherical and gaussian models were the most efficient models in estimation of soil attributes spatial variability, in most cases. Most of variables presented a regular spatial variability in their respective kriging maps, with some exceptions. In general, the kriging maps can be used, and we can take them as logistical maps for management and intervention practices in order to improve the soil fertility in the study areas. The results principal components indicate the need for integrated management of soil chemical attributes, with localized application of acidity correctors, fertilizers and other types of incomes, using the spatial variability of these fertility variables. Keywords: Geostatistics. Income optimization. Kriging maps. Precision agriculture and silviculture. Soil chemical attributes. 1. Introduction Agroforestry systems are a viable alternative for agricultural and forestry production to small and large farmers, since these systems present a diversification of production, in addition to contributing to the improvement of the characteristics of cultivated soils (Dhanya et al. 2013; Arévalo-Gardini et al. 2015; Laudares et al. 2017). Agroforestry crops have been widely used in Eastern Amazon, especially in the SPATIAL VARIABILITY OF SOIL FERTILITY UNDER AGROFORESTRY SYSTEM AND NATIVE FOREST IN EASTERN AMAZONIA, BRAZIL https://orcid.org/0000-0003-1039-8772 https://orcid.org/0000-0001-8057-9136 https://orcid.org/0000-0002-6458-2076 https://orcid.org/0000-0003-1679-9796 https://orcid.org/0000-0003-2815-6085 https://orcid.org/0000-0002-4837-6192 https://orcid.org/0000-0002-6458-2076 https://orcid.org/0000-0001-7230-9937 Bioscience Journal | 2023 | vol. 39, e39015 | https://doi.org/10.14393/BJ-v39n0a2023-62830 2 Spatial variability of soil fertility under agroforestry system and native forest in eastern Amazonia, Brazil Municipality of Tomé-açu, state of Pará, where agricultural producers have performed many planting experiments combining agricultural, forest and/or animal species (Bolfe and Batistella 2011). In this region, as well as in most of the Brazilian tropical region, agricultural and agroforestry systems are inserted in areas of soils with low natural fertility and high acidity. This means that many producers have to resort to ostensible use of fertilizers and liming. Thus, the knowledge regarding fertility conditions in which the cultivated soil is inserted is considerably important for planting planning. However, homogeneous assessment of soil quality can lead to sampling errors, because the attribut es of this soil can be highly influenced by the space variability (Rahman et al. 2013; Schwab et al. 2015; Rosemary et al. 2017). This is one of the main prerogatives of precision agriculture and precision silviculture, which consist of the rationalization of resources in order to optimize production, based on the concept of localized application of fertilizers and other kinds of inputs. Such practices can turn the correction of problems more precise and efficient, leading to an increased crop productivity. This makes it important to map the attributes that are intended to be managed in the planting area, among which the soil fertility attributes highlight (Fu et al. 2010; Suzuki et al. 2012; Vasu et al. 2017). The usage of tools based on spatial assumptions can be extremely helpful in the optimization of decision making with regard to soil management. This stands out the importance of developing technologies that help the interpretation of information regarding soil fertility based on the spatial dependence of such attributes (Zhang et al. 2010; Metwally et al. 2019). The geostatistics plays an important role as an efficient technique for soil fertility assessment, since the soil management and conservation in agroforestry systems and other kinds of soil coverages depends on the efficiency in the soil evaluation (Moshia et al. 2014). However, only a few studies have been developed with focus to spatial variability of soil atributes under agroforestry system aimin g to auxiliate the management of soil protection and fertilization (Silva et al. 2016; Silva et al. 2018; Panday et al. 2019). Therefore, the present study assessed the spatial variability of soil chemical attributes in an area of agroforestry system (AFS) compared to an adjacent forest (FOR), as a reference area, in the Municipality of Tomé-açu, Eastern Amazon, Brazil. 2. Material and Methods Study Area The study area belongs to Matsunaga Farm, located in the Microregion of Tomé-Açu, belonging to the Mesoregion of the Northeast of the State of Pará, Eastern Amazon, Brazil (2º 40’ 54’’S and 48º 16’ 11’’ W). The climate is humid mesothermal tropical, classified as Ami according to Koppen. The average annual temperature is 25 ºC. The average annual precipitation is 2250 mm, with a relative humidity of 80%. The predominant forms of natural vegetation in the area are the Dense Forest of the Low Plateaus and Dense Forest of the Plateaus, in addition to the preponderance of Secondary Forests, according to the classification proposed by IBGE (2012). The predominant soils in Tomé-açu are the Oxisols, with texture ranging from medium to clayey. The predominant reliafe is flat to smooth wavy (IDESP 2011). The soil from the study area is classified as Yellow Oxisol distrophic. The two study sites (forest and agroforestry system) are adjacent (Figure 1). For characterization purposes, 10 plots of 300 m² (30 x 10 m) were established, distributed in a completely randomized way in the Agroforestry (AFS) and Native Forest (FOR) systems, and the sampling of ten composite samples deformed at depth 0-0.2 m per system were made with the aid of a drill auger. Determinations of the granulometric fractions (sand, silt and clay) of the soil were carried out, with results ranging from soft to sandy sand to a depth of 0-0.2 m (Table 1). Management History and Areas Characterization The areas and their management history and characteristics are described below: Bioscience Journal | 2023 | vol. 39, e39015 | https://doi.org/10.14393/BJ-v39n0a2023-62830 3 DOS SANTOS, C.R.C., et al. Agroforestry system (AFS): Plantation of 15.73 ha, composed by Cocoa (Theobroma cacao L.), with 14 years of implantation and 4 x 4 m spacing; Amazonian mahogany (Swietenia macrophylla King), 25 years old and 8 x 8 m apart; and Coconut palm (Cocus nucifera L.) with 18 years of implantation and 10 x 10 m spacing. Previously, the area consisted of a black pepper plantation spaced in 1 x 1 m, which received an application of dolomitic limestone of approximately 2 t ha-1 and an application of NPK fertilizer mixture in the composition 10-28-20 of 2 t ha-1. All these fertilization practices were carried out in a pit. In the implementation of AFS, 1 t ha-1 of dolomitic lime (Total Neutralization Power: 90%) and 1 t ha-1 of phosphate fertilizer containing calcium, magnesium, and micronutrients (Yoorin) were applied. Both applications were performed only in the pit of the Cocoa plants. In addition, 1 t ha-1 of dolomitic lime, 0.6 t ha-1 of bone meal and 90 g of KCl per plant were applied in coverage, which is carried out annually. Before planting, the soil was prepared with a superficial cross harrow, using a disc harrow. In the area, a mowing is performed with a brush cutter attached to a tractor, every six months. The remaining vegetal material from the mowing is left on the agroforestry soil surface. Native Forest (FOR): secondary vegetation of 14.98 ha, predominantly from dense and mixed ombrophilous forest, with 30 years of natural regeneration. Figure 1. Location map with delimitation of agroforestry system (AFS) and native forest (FOR). Table 1. Granulometry and textural type of an Oxisol at depth 0-0.2 m under Agroforestry System and Native Forest, in Tomé-açu, Pará. Granulometric Fraction System Agroforestry System (AFS) Native Forest (FOR) g dm-3 Coarse Sand 506.8 656.5 Fine Sand 273.0 195.5 Silt 948.0 483.0 Clay 125.3 998.0 Textural Type Loamy Sand Sand Plots Spacialization and Soil Sampling The soil sampling was carried out at 72 collection points (36 points in each area) in February 2015, systematically distributed in a regular 55 x 55 m grid, totaling, approximately, 30 ha sampled (Figure 2). At each point, 5 simple samples were collected with distances of 5 m from the origin point, which were mixed and homogenized to obtain a soil composed sample. For soil sampling, a drill auger (for Bioscience Journal | 2023 | vol. 39, e39015 | https://doi.org/10.14393/BJ-v39n0a2023-62830 4 Spatial variability of soil fertility under agroforestry system and native forest in eastern Amazonia, Brazil deformed samples) was used at a depth of 0.2 m. At each collection point, the coordinates were marked in the UTM (Universal Transverse Mercator) format, considering Fuso 22M and Datum WGS 84. The GPS device used belongs to the Garmin Interface eTrex model. Figure 2. Samples spatialization for grid formation in A - Agroforestry System and B - Native Forest. Assessment of Soil Chemical Attributes For chemical analysis, the samples were separated and put to dry in the air and later sieved with a 2 mm mesh sieve, obtaining Air-dried Fine Sand (ADFS), in order to determine the following attributes: pH in H2O; exchangeable calcium (Ca), magnesium (Mg), potassium (K) and aluminum (Al); available phosphorus (P), potential acidity (H+Al); soil organic carbon for posterior calculation and determination of soil organic matter (S.O.M.). Bases saturation (V%) and aluminum saturation (m%) were also calculated based on the exchangeable cations and potential acidity. The soil analyses, calculations and determinations followed the methodology described by Embrapa (2009). Geostatistical Analysis An exploratory analysis of the variables was carried out using the following parameters: mean (m), variance (ℓ2), standard deviation (ℓ), coefficient of variation (CV%), skew-ness, kurtosis and normality test by Kormogorov-Smirnov significant at the 5% level. The variables that did not show normality by the test (p ≤ 0.05) were submitted to transfor-mation to normalization through fitting of Box-Cox model, using the Minitab 15 software. Based on the positioning of each soil sample, adjustment was carried out for three semivariogram models for each soil variable: Spherical model (Eq. 1), Gaussian model (Eq. 2) and Exponential model (Eq. 3), aiming to estimate the spatial variability of the chemical attributes of the soil. y(h)=C0+C1* [ 3 2 * h a - 1 2 * ( h a ) 3 ] (1) y(h)=C0+C1* [1-exp (-3* [ h a ] 2 )] (2) y(h)=C0+C1* [1-exp (-3* [ h a ])] (3) In which, y(h): Estimated variable; C0: nugget effect; C1: contribution C0+C1: sill, h: distance between the points, α: range, and exp: exponential. Bioscience Journal | 2023 | vol. 39, e39015 | https://doi.org/10.14393/BJ-v39n0a2023-62830 5 DOS SANTOS, C.R.C., et al. In order to verify which model obtained the best fitting, the following criteria were considered: lesser nugget effect and higher sill; higher spatial dependence degree (SDD) measured by the sill/nugget effect ratio, through the formula: SDD, expressed in percentage (%). The SDD was considered weak with 75% or more, moderate between 25% and 75%, and strong with less than 25% (Cambardella et al. 1994); the range could not be greater than the half of grid dimension (distance between the two most distant sampling points); smallest error, measured from the sum of residual squares (SQR); and higher determination coefficient (R²). To adjust the semivariogram models and choose the best model for each variable, the GS+ software. After calculating the semivariogram, the variables analyzed were interpolated to estimate the non - sampled area. To perform the estimation, ordinary kriging was used, assuming a linear association between the samples, since a systematic sampling was adopted in both study areas. From this estimate, an interpolation map was created for each of the variables, using ArcGIS 10.1 software. Among selected semivariogram models, only those with suitable values of R² (not extremely low) and SQR (not extremely high), with moderate or strong spatial dependence and with range lower than grid half dimension, were submitted to data interpolation data by kriging. Ordination Assessment The variables were also submitted to principal components analysis (PCA), through the formation of new components (axes) and usage of the two first components for creation of a graph with the variables’ vectorial disposition and sampling points dispersion from both study areas (AFS and FO R). PCA was performed with the aim of assessing the responsiveness of the variables in relation to the sampling points. 3. Results Variables Exploratory Assessment Most of averages were higher than their respective variances, which normally characterizes a more homogeneous distribution of the data (Table 2). The greatest variability (CV > 100%) was found for P in AFS. The vast majority of the studied attributes had a coefficient of variation considered moderate (60%> CV ≥ 12%). Exception is given to pH in H2O which, in both areas, presented a low coefficient of variation (CV <12%). Regarding normality, pH in H2O, Mg, H+Al, S.O.M. and m% variables showed normality in AFS. On the other hand, Ca, Al, P, K and V% were presented a distribution without normality. In FOR, Ca, Mg, K and V% had a non-normal distribution, turning necessary their transformation by Box-Cox model before the semivariogram assessment. Semivariograms Fitting Table 3 shows the results for cross validation of semivariogram estimates. In general, the values, specially Mean Standard Error, were considered suitable and satisfactory. All variables showed adjustment of semivariogram models with sill well established and a nugget relatively low, despite presenting considerable variation in the quality of the adjustments of such models. The spherical model and the Gaussian model were the ones that generated the best adjustments to the data based on the statist ical criteria established (Table 4). The spherical model showed the best fitting for 5 of the 9 soil fertility variables under agroforestry and 3 of the 9 soil variables under natural vegetation. Basically, pH in H2O, Ca, Al, H+Al and P were the variables best expressed by this model in AFS. In FOR, the variables Ca, K and V% showed better adjustments with the spherical model. The Gaussian model, in turn, showed a better fit in three soil properties in AFS (K, S.O.M. and V%) and FOR in 5 variables (pH in H2O, Mg, Al, K and m%). With regard to range (α), most of soil attributes presented suitable values, with exception to Al and S.O.M., which showed range higher than the half of grid size. Related to spatial dependence index (SDI), the models selected had satisfactory spatial dependence, in general. Most of soil chemical attributes showed strong spatial Bioscience Journal | 2023 | vol. 39, e39015 | https://doi.org/10.14393/BJ-v39n0a2023-62830 6 Spatial variability of soil fertility under agroforestry system and native forest in eastern Amazonia, Brazil dependence in AFS and FOR, with exception to some variables, which showed moderate spatial dependence. The R² in the AFS area ranged from 17.5 to 98.8% while the R² of the semivarogram models for FOR showed a variation from 23.3% to 76.2%. In AFS, the variables whose equations presented the highest R² were: Ca (98.8%), pH in H2O (97.5%) and P (92.4%), all with spherical model. In FOR, the attributes with the best coefficients of determination were: H+Al, expressed by the exponential model, with 76.2%; available P, better estimated by the exponential model, with 75.8%; and Ca, expressed by the spherical model, with logarithmic transformation, of 74.9%. S.O.M. semivariogram showed extremely low R² values and high SQR in both areas. Additionally, this variable presented excessive high values of range in FOR. m% also presented high values of SQR in both areas as well as Al in FOR. When comparing the quality criteria for the adjustment of each variable between the two areas under study, it is possible to note that, in general, the models adjusted for the AFS variables highlight in comparison to the same data for FOR. In general, AFS presented less nugget effect, greater range, higher level, greater spatial dependence, greater R² and less residual error. Table 2. Descriptive statistics of soil chemical attributes in an Agroforestry System (AFS) and a Native Forest (FOR). Attributes m ℓ ℓ2 CV% S K N Agroforestry System (AFS) pH in H2O 5.3886 0.1860 0.0346 3.45 -0.25 -0.68 p>0.15 Ca 0.6599 0.2805 0.0787 42.51 0.93 2.03 p<0.05 Mg 1.0501 0.4128 0.1704 39.31 0.24 0.12 p>0.15 K 0.0054 0.0024 0.00001 44.64 0.41 0.04 p<0.05 Al 0.4406 0.2026 0.0411 46.00 1.38 2.30 p<0.05 H+Al 4.3711 0.5556 0.3087 12.71 0.63 0.47 p>0.15 P 2.0110 2.0760 4.3110 100.00 1.53 1.12 p<0.05 S.O.M. 19.322 4.4970 20.2277 23.28 1.07 0.84 p>0.05 V% (%) 30.4700 13.260 175.8400 43.51 4.11 21.70 p<0.05 m% (%) 20.190 9.9600 99.2500 49.34 0.36 -0.60 p>0.05 Native Forest (FOR) pH in H2O 5.2854 0.2652 0.0703 5.02 0.01 -0.68 p>0.05 Ca 0.7494 0.4501 0.2026 60.05 1.71 2.57 p<0.05 Mg 0.4169 0.1528 0.0234 36.65 1.41 3.29 p<0.05 K 0.0059 0.0024 0.00001 41.35 1.63 2.44 p<0.05 Al 0.5884 0.2017 0.0407 34.28 -0.36 -0.45 p>0.15 H+Al 5.6280 1.0040 1.0070 17.83 0.04 -0.84 p>0.15 P 0.3609 0.1081 0.0117 29.94 0.30 -0.75 p>0.15 S.O.M. 23.897 3.5980 12.9460 15.06 0.34 -0.68 p>0.15 V% (%) 17.240 6.9900 48.8900 40.57 0.85 -0.17 p<0.05 m% (%) 35.750 14.7300 216.9200 41.20 -0.19 -0.41 p>0.15 In which, m: mean; ℓ2: variance; ℓ: standard deviation; CV%: coefficient of variation; S: skewness; K: kurtosis; N: normality; Ca, Mg, K, Al and H+Al given in cmolc dm-3; P given in mg dm-3 and; S.O.M. given in g kg-1. Bioscience Journal | 2023 | vol. 39, e39015 | https://doi.org/10.14393/BJ-v39n0a2023-62830 7 DOS SANTOS, C.R.C., et al. Table 3. Cross Validation data for the semivariogram models. Variables Model Standardized Mean Standardized Mean Square Root Mean Standard Error Agroforestry System (AFS) pH em H2O Spherical -0.034990 0.786700 0.28280 Ca Spherical -0.055900 1.136460 0.259470 Mg Exponential 0.007600 0.961400 0.392010 Al Spherical 0.007700 0.444700 0.497800 H+Al Spherical 0.016480 0.997480 0.637370 P Spherical 0.081690 4.473100 0.494930 K Gaussian 0.000005 0.010130 0.203807 S.O.M. Gaussian -0.016360 0.895860 3.852450 V% Gaussian -1.561400 59.295670 0.114960 m% Exponential 0.026000 0.927800 11.073500 Natural Forest (FOR) pH em H2O Gaussian 0.004970 0.918280 0.284400 Ca Spherical -0.043900 1.400198 0.279300 Mg Gaussian 0.001790 0.885160 0.187570 Al Gaussian 0.005210 1.010440 0.202540 H+Al Exponential -0.015170 0.757790 1.250600 P Exponential -0.026600 0.788440 0.120920 K Spherical -0.000001 0.000020 13.900000 S.O.M. Gaussian -0.006790 1.261140 2.881660 V% Spherical -0.143840 28.081900 0.221420 m% Gaussian 0.004980 0.799020 19.083260 Table 4. Fitting quality parameters of the selected models for each soil chemical attributes in an Agroforestry System (AFS) and a Native Forest (FOR). C0: nugget, C0 + C1: sill, SDI: spatial dependence index, α: range, SDD: spatial dependence degree (W: weak, M: moderate, S: strong), R²: coefficient of determination, SQR: residuals sum of square. Kriging Mostly, the kriging maps showed a suitable and satisfactory representation of the spatial variability, with exception for some maps. These tendencies may represent an error of variability estimation. The patterns of spatial distribution of attributes in AFS, it was noticed greater uneven spatial variability (Figure Attributes Selected Model C0 C0+C1 SDI (%) α (m) SDD R² (%) SQR Agroforestry System (AFS) pH in H2O Spherical 0.023 0.047 47.87 49.86 M 97.5 0.029 Ca Spherical 0.012 0.047 24.75 43.87 S 98.8 0.009 Mg Exponential 0.000 0.134 0.08 37.30 S 43.3 0.011 Al Spherical 0.023 0.193 11.95 45.01 S 74.3 0.127 H+Al Spherical 0.059 0.322 18.10 74.93 S 84.3 0.696 P Spherical 0.043 0.168 24.71 19.94 S 92.4 0.181 K Gaussian 0.000 0.039 0.26 56.10 S 78.6 0.000 S.O.M. Gaussian 8.540 17.090 49.97 113.90 M 17.5 165.000 V % Gaussian 0.002 0.010 18.18 44.38 S 42.0 0.004 m % Exponential 15.000 91.000 16.48 27.58 S 79.0 17035.0 Native Forest (FOR) pH in H2O Gaussian 0.056 0.125 44.44 225.00 M 44.2 0.005 Ca Spherical 0.022 0.045 44.89 0.10 M 74.9 0.009 Mg Gaussian 0.022 0.046 48.90 428.00 S 37.5 0.002 Al Gaussian 0.034 0.069 49.92 493.00 M 63.3 0.000 H+Al Exponential 0.383 1.712 22.37 181.70 S 76.2 0.362 P Exponential 0.001 0.013 10.58 83.50 S 75.8 0.000 K Spherical 710.000 14590.0 4.87 0.30 S 44.1 0.000 S.O.M. Gaussian 5.100 41.200 12.38 293.10 S 65.4 285.000 V % Spherical 0.014 0.028 49.82 0.10 M 57.0 0.002 m % Gaussian 104.200 208.500 49.88 0.10 M 44.4 106021.0 Bioscience Journal | 2023 | vol. 39, e39015 | https://doi.org/10.14393/BJ-v39n0a2023-62830 8 Spatial variability of soil fertility under agroforestry system and native forest in eastern Amazonia, Brazil 3 and 4). Unfortunately, the best models selected for S.O.M. and m%, for AFS and S.O.M. and m% for FOR showed exceptionally low values of R² and/or extremely high levels of SQR, as well as elevated values of range. Therefore, the semivariogram models selected for these variables were not able to be submitted to data interpolation by kriging. The predominant values of pH in H2O from AFS is ranging from 5.24 to 5.36 and from 5.36 to 5.48 (Fig. 3-A). Only a small part of the AFS area had pH values within 5.60 and 5.72, which fall within the pH range of 5.5 to 6.5. In FOR, most of the area has pH values between 4.95 and 5.14 (Fig. 4-A). Both Ca and Mg had a spatial distribution with considerable variations in AFS (Figure 3-B; Figure 3- B). The results of the potential acidity (H+Al), in turn, corroborate those results found for Al with regard to the spatial distribution in both study areas. FOR had its area predominantly occupied by H+Al values ranging from 5.68 to 6.3 cmolc dm-3 (Figure 4-E), whereas in AFS, this attribute showed predominance between 3.79 and 4.6 cmolc dm-3 (Figure 3-E). For V%, FOR showed a predominance of values between 11.2 and 15.6% (Figure 4-I), while AFS presented values occupying most of the area between 28.9 and 32.3% (Figure 4-I), with small patches of particular major and minor intervals. Ordination Assessment For the two first components, eigenvalues were higher than 1 and therefore were satisfactory (Table 5). Regarding percentage of variance, components 1 and 2 presented 67,87 % of accumulated variance, which is can be considered suitable to explain the relationship between the soil fertility attributes from both areas with their respective sampling points. The P, Mg, V% and pH were more responsive to sampling points belonging to AFS, while m%, Al, H+Al, S.O.M., K and Ca were more related to points from FOR (Figure 5). Mg, V% and pH presented more relationship among them and a negative relationship with m% and Al and, finally, H+Al, K, Ca and S.O.M. were highly related among each other and had a negative relationship with P. Table 5. Eigenvalues and variance values for principal components formed soil chemical attributes in an Agroforestry System (AFS) and a Native Forest (FOR). Components Eigenvalue Percentage of Variance (%) Accumulated Variance (%) 1 4,9679 49,679 49,679 2 1,8185 18,185 67,864 3 0,9217 9,217 77,080 4 0,8587 8,587 85,667 5 0,7957 7,957 93,623 6 0,2984 2,984 96,608 7 0,2155 2,155 98,762 8 0,0908 0,908 99,670 9 0,0252 0,252 99,922 10 0,0078 0,078 100,000 Bioscience Journal | 2023 | vol. 39, e39015 | https://doi.org/10.14393/BJ-v39n0a2023-62830 9 DOS SANTOS, C.R.C., et al. Figure 3. Kriging maps of soil chemical attributes in an Agroforestry System, notably: A – pH in H2O; B – Ca; C – Mg; D – Al; E - H+Al; F – P; G – K; H – V%. Bioscience Journal | 2023 | vol. 39, e39015 | https://doi.org/10.14393/BJ-v39n0a2023-62830 10 Spatial variability of soil fertility under agroforestry system and native forest in eastern Amazonia, Brazil Figure 4. Kriging maps of soil chemical attributes in a Native Forest, notably: A – pH in H2O; B – Ca; C – Mg; D - H+Al; E – P; F – K; G – V%. Bioscience Journal | 2023 | vol. 39, e39015 | https://doi.org/10.14393/BJ-v39n0a2023-62830 11 DOS SANTOS, C.R.C., et al. Figure 5. Principal componentes analysis, with vectorial disposition of soil chemical attributes from AFS and FOR. 4. Discussion Variables Exploratory Assessment The lowest mean values show the pattern of behavior of such soil attributes, which characterizes a preponderance of the spatial dependence regarding the influence of other environmental factors (Yamamoto and Landim 2013). This variation of data for most variables is considered important for the adjustment of spatial variation models. Santos et al. (2017) developed a research evaluating the spatial variability of soil fertility in a cocoa cultivation area in Ilhéus-BA, results similar to the present study, with a large majority of the attributes moderate variation coefficient (2.40