Soil physicochemical properties and terrain information predict soil enzymes activity in phytophysiognomies of the Quadrilátero Ferrífero region in Brazil

Catena 199 (2021) 105083

Contents lists available at ScienceDirect

Catena journal homepage: www.elsevier.com/locate/catena

Soil physicochemical properties and terrain information predict soil ´tero Ferrífero region enzymes activity in phytophysiognomies of the Quadrila in Brazil Anita Fernanda dos Santos Teixeira, S´ergio Henrique Godinho Silva, Teotonio Soares de Carvalho, Aline Oliveira Silva, Amanda Azarias Guimar˜aes, Fatima Maria de Souza Moreira * Federal University of Lavras, Soil Science Department, P.O. Box 3037, Zip Code 37200-900 Lavras, MG, Brazil

A R T I C L E I N F O

A B S T R A C T

Keywords: Portable X-ray fluorescence Soil enzymes prediction Relief Soil quality Prediction models

Soil enzymes act in biogeochemical cycles of elements and are indicators of soil quality since they rapidly reflect changes of the environmental conditions. Moreover, enzymes are related to soil physicochemical properties, but their spatial distribution has been rarely evaluated. The hypothesis of this work is that soil properties related to fertility and texture (F), total contents of chemical elements obtained by portable X-ray fluorescence (pX) spectrometry and terrain attributes (T) can be used as predictor variables to soil enzyme activity, along with phytophysiognomy and season information. The objective of this work was to predict soil enzymes activity and assess its spatial variability in the most common phytophysiognomies of the Quadril´ atero Ferrífero mineral province, in Brazil. Soil samples were collected in four phytophysiognomies during both dry and humid seasons. Activity of β-glucosidase, acid phosphatase, alkaline phosphatase, urease, and hydrolysis of fluorescein diacetate (FDA) was determined. Phytophysiognomy, season, F, T, and pX, were used together or separately to predict the enzymes activity through conditional random forest algorithm and the accuracy was assessed via leave-one-out cross validation. The generated models were accurate, with coefficient of determination (R2) varying from 0.63 (FDA by pX) to 0.82 (β-glucosidase by F). F variables were more important for the predictions, while pX variables were more important for predicting acid phosphatase and urease. The accurate models using T variables allowed the generation of maps showing the enzymes variability along the phytophysiognomies. This approach can accelerate the determination of soil enzymes activity across the landscape.

1. Introduction ´tero Ferrífero region is one of the largest mineral prov­ The Quadrila inces in the world and one of the main areas for Fe mining in Brazil. In this region, the variation of soils is remarkable and closely related to the underneath lithology, mostly causing the formation of soils presenting very high contents of Fe and Mn (Carvalho Filho et al., 2010). These unique soil conditions influence the regional biodiversity by promoting

the occurrence of several endemic species (Skirycz et al., 2014). This fact along with the environmental changes caused by Fe mining activities ´tero Ferrífero region a biodiversity hotspot, requiring makes Quadrila further studies about the relationships between the existing soils and organisms (Castro et al., 2017; Silva et al., 2018; Teixeira et al., 2017). Soil organisms perform key environmental functions (Schaetzl and Anderson, 2005; Resende et al., 2014; Singh et al., 2018), such as decomposition of residues and xenobiotics, nutrient cycling and

Abbreviations: pXRF, portable X-ray fluorescence spectrometer; F or f, soil fertility and texture attributes; pX or px, soil contents of chemical elements obtained by portable X-ray fluorescence spectrometry; T, terrain attributes; FDA, hydrolysis of fluorescein diacetate; fNtotal, N content in soil; Prem, remaining P; fH+Al, po­ tential acidity; CEC, cation exchange capacity; ft, effective cation exchange capacity fT, potential cation exchange capacity at pH 7; fm, aluminum saturation; fV, base saturation; fOM, soil organic matter; twi, Topographic Wetness Index; hillsh, Hillshade; channbl, Channel Network Base Level; csc, Cross-sectional Curvature; longcurv, Longitudinal Curvature; mrrtf, Multi-resolution Ridge Top Flatness; relslop, Relative Slope; valleydep, Valley Depth; vertdis, Vertical Distance to Channel Network; LOOCV, Leave One Out Cross-Validation; R2, coefficient of determination; RMSE, root mean square error; MAE, mean absolute error. * Corresponding author. E-mail addresses: [email protected] (S.H.G. Silva), [email protected] (F.M. de Souza Moreira). https://doi.org/10.1016/j.catena.2020.105083 Received 10 February 2020; Received in revised form 26 October 2020; Accepted 1 December 2020 0341-8162/© 2020 Elsevier B.V. All rights reserved.

A.F. dos Santos Teixeira et al.

Catena 199 (2021) 105083

maintenance of the biogeochemical cycles that guarantee the continuity of life on the planet (Plante, 2007). In this context, soil enzymes pro­ duced by microorganisms have a key role. They are responsible for accelerating multiple processes related to nutrient cycling and catalysis of reactions. Moreover, soil enzymes activity has been considered an indicator of soil quality, since it is rapidly affected by environmental changes, e.g. humidity and temperature variations such as those caused by different seasons, soil physicochemical properties, vegetation cover (or phytophysiognomy) and structure of the microbial community (PazFerreiro and Fu, 2016). However, methods for determination of soil enzymes activity are time-consuming and expensive, which encourages the search for alternative methods of evaluation of this key soil micro­ biological property. Soil enzymes activity can be positively or negatively affected by several soil properties, such as those related to fertility, moisture, texture, and chemical composition. Furthermore, some chemical ele­ ments, such as some heavy metals, can be toxic to the microorganisms producing these enzymes (Gianfreda et al., 2002; Mounissamy et al., 2017; Wang et al., 2007). With the exception of studies related to contaminated areas, few investigations have been conducted aiming to evaluate the simultaneous influence of different chemical elements on soil enzymes activity, probably due to the costly and time consuming conventional methods for quantification of soil elemental contents (dos Santos et al., 2016; Wang et al., 2007). However, modern tools, such as the portable X-ray fluorescence (pXRF) spectrometer, have proved suc­ cessful for determining multiple elemental contents in a few seconds, with minimal sample preparation and without generation of chemical residues (Weindorf et al., 2014; Ribeiro et al., 2017). These results have been increasingly adopted not only for soil characterization, but also for predicting different soil properties across the globe (Benedet et al., 2020; Mancini et al., 2019; Rawal et al., 2019, Silva et al., 2017; Teixeira et al., 2018; Weindorf et al., 2012; Zhu et al., 2011), although very few studies were conducted regarding soil microbiological properties (Weindorf et al., 2018). In the last decades, the number of works dealing with mapping soil physical, chemical, and biological properties largely increased and, to that end, several techniques have been used (Duda et al., 2017; Forkuor et al., 2017; Hengl et al., 2017; Liu et al., 2012; Malone et al., 2017; Pelegrino et al., 2018; Qu et al., 2018; Spohn et al., 2013; Vasu et al., 2017). Under such mapping approach, spatial characterization of soil enzymes activity throughout an area of interest, using solely pXRF data could be used, but with some limitations, since the results of each pXRF analysis only corresponds to the sampling places (punctual results) (Duda et al., 2017; Mancini et al., 2020). This would require soil sam­ pling at several places followed by interpolation of pXRF results in order to represent the continuous variability of elemental contents throughout an area of interest and, hence, allow the establishment of their spatial relation with soil enzyme activity. Besides pXRF data, terrain features have been more commonly used for mapping soil properties based on soil-topography relationships (Adhikari et al., 2013; Hengl et al., 2017; Machado et al., 2019). Moreover, terrain attributes, i.e., representation of topography features in a continuous raster type (pixel-based) format, such as slope gradient, curvature and topographic wetness index, may be strongly correlated with the variation of the target variable (e.g., soil enzymes activity) throughout the study area due to the continuous nature of such data, as opposed to pXRF data. For instance, Teixeira et al. (2019) found strong correlations between terrain attributes, e.g. topographic wetness index and slope gradient, and soil microbiological properties, e.g., urease, basal respiration and number of spores of mycorrhizal fungi. This study provided a better comprehension of the variability of soil microbiolog­ ical properties across a tropical landscape. Therefore, after establishing these relations, maps showing the spatial variability of soil microbio­ logical properties, such as enzyme activities, could be created combining terrain, soil physicochemical properties, pXRF and environmental fac­ tors (e.g. phytophysiognomy and warm/humid or cold/dry seasons)

data that are related to enzymes activity. Since conventional methods of quantification of soil enzymes activ­ ity are time-consuming and expensive, the hypothesis of this work is that contents of chemical elements obtained by pXRF, terrain attributes, phytophysiognomy (i.e. vegetation type) and season (warm/humid or cold/dry) can be used as predictor variables of soil enzyme activity along with soil physicochemical properties related to fertility and texture. The objective of this work was to assess the spatial distribution ´tero of soil enzymes activity in four phytophysiognomies of the Quadrila Ferrífero mineral province, in Brazil, evaluating the importance of soil physicochemical properties, pXRF, terrain, phytophysiognomy and season information to generate accurate predictions. 2. Material and methods 2.1. Study area This study was conducted in Brumadinho and Nova Lima counties, ´tero Ferrífero region, in the state of Minas Gerais, located in the Quadrila Brazil. The climate of the region is Cwa (humid subtropical climate, with ¨ppenrainy and warm summer and cold and dry winter) according to Ko Geiger classification (Alvares et al., 2013). Mean annual rainfall ranges from 1100 to 1420 mm and rainfalls are more intense from September to March. Mean annual temperature ranges from 19 to 22 ◦ C, with tem­ peratures ranging from lesser than 15 ◦ C in winter (June to September) and greater than 29 ◦ C in summer (December to March) (Silva et al., 2018). The areas chosen for this study represent the natural vegetation types occurring in the region (Cerrado, Ironstone Outcrops and Atlantic For­ est) in addition to an area under rehabilitation after mining activities, all of them containing contrasting characteristics and making up the socalled phytophysiognomies. In the county of Brumadinho, the study area is located at approximately 800 m of altitude, with mean slope gradient of 20.3%, covering 17 ha, and the geology is characterized by the presence of granite, related to Metamorphic Complexes, composed of granite-gneissic terrains of Archaean age. In Nova Lima, the study area is located at approximately 1300 m of altitude, covers 13.8 ha with mean slope gradient of 14.8%, and the geology is characterized by the presence of superficial iron-rich crusts called “lateritic canga” (Coelho et al., 2017). The four vegetation types with contrasting characteristics (phyto­ physiognomies) were described by Castro et al. (2017), Coelho et al. (2017) and Silva et al. (2018) as follows (soils classified according to both Keys to Soil Taxonomy (Soil Survey Staff, 2014) and Brazilian Soil Classification System (Santos et al., 2018), the latter in parenthesis): Ironstone Outcrops (known as “Canga”, containing grasses and small trees) on Typic Plinthaquox (Plintossolo P´etrico), Neotropical Savanna (known as “Brazilian Cerrado”) on Typic Plinthaquox (Plintossolo P´etrico), Atlantic Forest on a toposequence of Rhodic Haplustox (Latossolo Vermelho), Typic Haplustox (Latossolo Vermelho-Amarelo), ´plico) (Fig. 1). The rehabilitated and Typic Dystrustept (Cambissolo Ha area revegetated with grass contains a human-modified soil type, here called anthropic soil, because both Soil Taxonomy and Brazilian Soil Classification systems do not encompass soils with these characteristics (this soil is equivalent to Anthrosols in World Reference Base soil clas­ sification system (IUSS Working Group WRB, 2014). 2.2. Soil sampling Soil sampling was conducted in both cold/dry and warm/rainy seasons. In each phytophysiognomy, 10 composite samples were collected in two transects per season, making up a total of 80 samples (10 samples × 4 phytophysiognomies × 2 seasons) (Fig. 1 of Supple­ mentary Material). The transects were approximately 50–70 m apart from each other. Variations of these distances in some phytophysiogn­ omies were caused by factors like inaccessibility of the area and absence 2

A.F. dos Santos Teixeira et al.

Catena 199 (2021) 105083

Fig. 1. Location of the soil sampling at the Ferrous Technology Center (Miguel˜ ao - Nova Lima) and the C´ orrego do Feij˜ ao Mine (Brumadinho) in Minas Gerais State, Brazil.

of soil at the pre-determined sampling place. Each composite sample was composed of 5 samples collected at 0–20 cm depth, as follows: one sample collected at a central position of the transect, two samples at 5 m and two samples at 10 m from the central position to the east and west directions to capture such east–west vari­ ability (see Supplementary Material for further details about the sam­ pling design) (Silva et al., 2018). The geographic coordinates of the central position of each composite sample were recorded using a GPS device and this position is represented in Fig. 1. To improve soil samples characterization, soil moisture was determined at sampling in both seasons (Table 1). For that, samples were weighted at sampling and after being oven-dried at 105 ◦ C during 24 h. Thus, soil moisture (%) was calculated as follows: soil moisture (%) = 100 × (weight at sampling – dry weight)/weight at sampling.

The following soil properties were determined: total N content (fNtotal) (Joergensen and Brookes, 1990), exchangeable contents of Ca2+, Mg2+, and Al3+, (Mclean et al., 1958); pH in water (1:2.5); available contents of P, K, Fe, Zn, Mn and Cu (Mehlich, 1953); remaining P (Prem), i.e., the amount of a solution containing P that is not adsorbed by soil particles, commonly used in Brazil (Alvarez and Fonseca, 1990); potential acidity (H + Al) (Shoemaker et al., 1961); available S (Hoeft et al., 1973) and B (Raij et al., 2001); effective cation exchange capacity (CEC) (t), potential CEC at pH 7 (T), aluminum saturation (m), base saturation (V) (Alvarez et al., 1999), and soil organic matter (OM) ob­ tained by oxidation with potassium dichromate in acidic medium (Walkley and Black, 1934). Sand, silt, and clay contents were obtained through soil texture determination by the pippete method (Bouyoucos, 1951). Mean values of soil fertility and texture corresponding to the composite samples and used in this study were determined by Castro et al. (2017) and Silva et al. (2018). Soil chemical elemental contents were obtained through a pXRF Bruker® model S1 Titan LE (50 kV e100 μA X-ray tubes). For such analysis, samples were air-dried, sieved to 2 mm, and scanned in trip­ licate, during 60 s, at Trace (dual soil) mode, using the GeoChem soft­ ware. The accuracy of the equipment was assessed by scanning reference materials and comparing the pXRF results with the contents of the 2710a and 2711a samples certified by the National Institute of Standards and Technology (NIST) and a check sample (CS) provided by the equipment manufacturer. The recovery values (% of recovery = 100 × obtained

2.3. Soil analysis The activity of the following enzymes was evaluated (methods pre­ sented in parentheses): β-glucosidase (Eivazi and Tabatabai, 1988), acid phosphatase, alkaline phosphatase (Eivazi and Tabatabai, 1977), and urease (Keeney and Nelson, 1982; Tabatabai and Bremner, 1970), in addition to total enzyme activity by hydrolysis of fluorescein diacetate (FDA) (Dick et al., 1996). The activity values of the soil enzymes, in mg substrate kg-1h− 1, used in this study were determined and reported by Silva et al. (2018).

Table 1 Soil moisture at sampling in the different phytophysiognomies and seasons (n = 10 per phytophysiognomy and season). Phytophysiognomies Neotropical savanna Ironstone Outcrops Atlantic Forest Rehabilitated Area Neotropical savanna Ironstone Outcrops Atlantic Forest Rehabilitated Area

Mean

Minimum

Maximum

Standard deviation

——————————————————————Dry season——————————————————————————— 19.33 14.09 24.43 3.81 7.72 3.24 15.54 3.42 20.89 16.84 22.56 1.90 11.05 5.51 25.84 5.90 —————————————————————Rainy season—————————————————————————— 20.67 8.62 36.34 8.73 7.64 3.51 9.91 2.08 27.08 23.18 29.96 2.10 20.08 13.88 38.05 6.79

3

A.F. dos Santos Teixeira et al.

Catena 199 (2021) 105083

content/ certified content) for the elements used in this study was calculated. The elements used in this study were those detected by pXRF in at least one out of the four phytophysiognomies. The recovery values (%) for such elements following the order of samples 2710a/2711a/CS are, respectively: Al (78.8/68.9/88.3%), As (0/0/0), Bi (0/0/0), Ca (36.2/42.9/0), Ce (0/0/0), Cl (0/0/0), Cr (0/124.7/0), Cu (84.0/77.2/ 94.3), Fe (75.8/67.6/89.3), K (55.6/45.9/84.3), Mn (69.9/61.2/80.5), Nb (0/0/0), Ni (0/115.0/90.2), P (411.8/577.9/0), Pb (108.2/106.3/ 105.6), Rb (88.3/91.1/0), Si (57.4/49.3/90.5), Ta (0/0/0), Ti (77.7/ 72.0/0), V (0/0/0), Y (0/0/0), Zn (94.2/77.8/0) and Zr (105.0/0/0). Zero values indicate either that the element was not present in the certified material or that pXRF could not detect it.

and, hence, can be used for predictions of soil enzymes activity for the four phytophysiognomies and for the two seasons. The importance of all the predictor variables was determined for the models containing F + pX + T data to assess those variables that can better explain soil enzymes activity. This importance is calculated ac­ cording to the changes of the accuracy of the models when a certain variable is left out of the model (Breiman, 2001). The greater the reduction of accuracy when a variable is removed from the model, the greater the importance of that variable for the model (Liaw and Wiener, 2002). All models were validated by the Leave One Out Cross-Validation (LOOCV) method of the caret package (Kuhn et al., 2018). This method repeatedly creates models for the prediction of a variable of interest, leaving one sample out of each of these models. Then, this sample is later used to assess the accuracy of the generated model. LOOCV is commonly used when dealing with a small number of samples, e.g. from 40 to 120 (Chakraborty et al., 2016), since it provides the best estimate about the performance of the prediction model created (Mar­ tens and Dardenne, 1998). The values of the coefficient of determination (R2), root mean square error (RMSE), mean absoluteerror (MAE) and predictive square corre­ lation coefficient or also named cross-validated R2 (Q2, calculated as 1 – (predictive error sum of squares/total sum of squares)) were calculated by comparing the values estimated by the model with the observed (real) values. The prediction values delivered by the best model per soil enzyme activity (greatest R2 and Q2 and smallest RMSE and MAE) were also plotted in 1:1 graphs for comparison with the observed values. Moreover, the spatial variability of the enzymes activity was mapped per phytophysiognomy using the models created based on T data. For that, the values of the terrain attributes were extracted per pixel and associ­ ated with each phytophysiognomy and season. Then, they were inserted into the best prediction model delivered by the data set containing only T variables (in addition to season and phytophysiognomy information) for determination of soil enzymes activity in the four phytophysiogno­ mies and for both seasons. In this procedure, the raster and rgdal packages of the R software were used to generate the maps of soil en­ zymes activity per phytophysiognomy and season, providing an over­ view of the spatial and temporal variability of soil enzymes activities within the study areas.

2.4. Terrain attributes The terrain attributes used in this study were derived from an Alos Palsar digital elevation model (spatial resolution of 12.5 m) obtained from the digital platform of the Alaska Satellite Facility (https://vertex. daac.asf.alaska.edu). The terrain attributes slope, topographic wetness index (twi) (Beven and Kirkby, 1979), aspect, hillshade (hillsh), channel network base level (channbl), cross-sectional curvature (csc), longitu­ dinal curvature (longcurv), multi-resolution ridge top flatness (mrrtf), multi-resolution valley bottom flatness (mrvbf), relative slope position (relslop), valley depth (valleydep) and vertical distance to channel network (vertdis) were obtained using the software System for Auto­ mated Geoscientific Analysis (SAGA) GIS v 2.1.4 (Conrad et al., 2015). These are terrain attributes commonly used for prediction of soil classes and properties (Adhikari et al., 2013; Pelegrino et al., 2016; Arrouays et al., 2017; Hengl et al., 2017; Menezes et al., 2020). The values of these terrain attributes were extracted at the sampling sites (central position) and used on the models for the prediction of soil enzymes activity. 2.5. Data analysis The software R (R Core Team, 2019) was used for fitting the models through the partykit package (Zeileis and Hothorn, 2014) and the con­ ditional inference random forest (cforest) algorithm, which creates random forests from unbiased regression trees (Hothorn et al., 2006; Strobl et al., 2007). Random forests are an ensemble method that combines several individual regression or classification trees, being a powerful statistical tool (Liaw and Wiener, 2002). However, in this method, the variable importance calculation is affected by the number of variables and their range of values. In these conditions, cforest function has shown better results than random forest, since it provides unbiased variable selection for the development of individual trees (Strobl et al., 2007). The parameters for cforest modeling were ntree = 1000, mtry = square root of the number of predictor variables (variable according to the data set) and nodesize as the default of the partykit package (Zeileis and Hothorn, 2014). For modeling and validation of β-glucosidase, acid phosphatase, alkaline phosphatase, urease and FDA, the following variables were used as predictors: soil organic matter, soil fertility and texture (F); terrain attributes (T); pXRF data (pX); phytophysiognomy; and season. The sets of predictor variables F, T and pX were used separately or in combina­ tion, encompassing seven data sets: a) pX; b) F; c) T; d) F + pX; e) F + T; f) T + pX; and g) F + pX + T. In addition to the seven combinations of variables and based on both Millard and Richardson (2015) and Silveira et al. (2019), an eighth data set was created with the 15 most important variables for predicting each enzyme activity (importance of each var­ iable for the models was determined by the cforest algorithm) to test if the utilization of the most important variables would improve the pre­ dictions. Moreover, a ninth data set containing all variables that pre­ sented positive importance values was used for predictions. All the nine prediction models contained phytophysiognomy and season as predictor variables. Moreover, all the nine data sets presented 80 samples (10 samples per phytophysiognomy × 4 phytophysiognomies × 2 seasons)

3. Results 3.1. Variables importance The importance of all the predictor variables, i.e. soil physico­ chemical properties, terrain, pXRF, season and phytophysiognomy data, is presented in Table 2. The most important variables for the β-gluco­ sidase prediction were phytophysiognomy, fAl, fMg, fB and potential cation exchange capacity (fT) (“f” subscript was used in front of the variables to distinguish soil fertility and texture properties from those obtained by pXRF, which were written using a “px” subscript). Besides phytophysiognomy, only variables related to soil fertility were among the most important ones to predict this enzyme activity. For the acid phosphatase prediction, the most important variables were pxFe, pxMn, phytophysiognomy, pxAs and clay content (Table 2). Contrarily to β-glucosidase, which presented soil fertility properties as the most important predictor variables, for acid phosphatase prediction, the total contents of pxFe, pxMn and pxAs obtained by pXRF in addition to clay content and phytophysiognomy were the most important variables. Other pXRF variables also appeared among the five most important ones for alkaline phosphatase (pxCu, phytophysiognomy, fAl, pxZn and fm) and FDA (season, fB, fS, pxTi and pxAs) prediction (Table 2). Similarly to the FDA prediction, season was also the most important variable for urease prediction, followed by pxTa, phytophysiognomy, pxPb and pxFe (Table 2). Considering the 25 variables occupying the first five positions of 4

A.F. dos Santos Teixeira et al.

Catena 199 (2021) 105083

Table 2 Importance of the variables (increase in Mean Squared Error) from the most important at the top to the least important at the bottom for the prediction of the hydrolysis of fluorescein diacetate (FDA) and the activity of the soil enzymes β-glucosidase, acid phosphatase, alkaline phosphatase and urease. Var.*

β-glucosidase

Phytop. fAl fMg fB fT fK fV pxZr pxCu pxMn pxZn pxAs pxFe vertdis fPREM Season pxNb pxCr fm fNtotal pxCe fSB pxAl2O3 relslop fZn pxSiO2 fClay channbl ft fpH fCa pxRb fP fSilt hillsh mrvbf pxK2O fMn fS fHAl pxCaO slope fFe pxY aspect pxPb pxV mrrtf pxBi fOM fSand fCu pxTa csc longcurv pxNi pxCl pxTi pxP2O5 valleydep twi

799.6 621.9 462.8 449.9 429.8 393.9 366.1 353.3 336.4 313.9 304.6 298 297.9 259.8 256.5 240.2 226.1 203 187.2 186.4 182.5 177.5 175.8 175.8 175.4 173.3 163.3 147.1 141.8 138.1 119.6 115.2 115.1 115.1 112.7 106.3 95 92.6 91.2 85.1 81.5 79.1 78.9 78.7 73.9 59.7 40.3 35.7 26.6 22.9 22.5 21.9 21.3 15.3 13.4 6.7 0 0 − 8.3 − 15.2 − 18.9

*

Var.

Acid phosphatase

Var.

Alkaline phosphatase

Var.

FDA

Var.

Urease

pxFe

526.5 498.7 497.2 401.5 370.6 348 288.1 267.6 141 134.7 134.6 130.7 112.8 111.4 107.7 106.3 97.5 96.5 94.3 87.3 86.2 80.6 80.1 78.8 75.9 70.7 68.8 68.5 54.9 52.7 52.3 49 48.4 45 43.9 36 35 29.1 25.3 21.8 19.3 17.3 15.8 12.7 8.8 8.6 3.6 2.6 0.1 0 0 0 0 0 − 1.8 − 7.3 − 8.2 − 11.4 − 13.7 − 24.6 − 47.2

pxCu

1314.2 1264.8 967.4 773.6 732.3 679.8 660.2 586.9 564.9 537.8 526.8 513.4 422.7 412.2 393.1 374.1 365.5 361.2 353.7 352.9 347.6 275.9 266.6 265.7 244 237.7 235.4 224.6 199.6 184.5 183.1 171.5 156.1 123 110.8 105.4 101.5 82.1 69.1 67.6 65.9 49.9 45.2 43.7 39.5 27.4 10.5 9.9 6.1 5.2 2.6 0 0 0 0 − 3.7 − 7.1 − 17.4 − 21.5 − 24.8 − 34.6

Season fS fB pxTa pxFe pxAs pxSiO2 fPREM pxAl2O3 pxPb pxZn channbl pxBi fOM pxCu fm fNtotal fCu pxK2O fZn pxY pxNi ft fClay relslop pxCe pxRb pxMn pxZr fP fHAl Phytop. fAl fpH vertdis fT mrvbf fV longcurv pxNb fMg pxCl pxCaO fSB valleydep pxCr fCa fFe fMn fK pxTi fSand slope csc mrrtf pxV hillsh twi aspect fSilt pxP2O5

3.6 1.6 1.6 1.4 1.3 1.3 1.1 1.1 1 1 0.8 0.8 0.8 0.7 0.6 0.6 0.6 0.6 0.6 0.6 0.5 0.5 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.2 0.2 0.2 0.2 0.2 0.1 0.1 0.1 0.1 0.1 0.1 0 0 − 0.1 − 0.1 − 0.1 − 0.1 − 0.2 − 0.2 − 0.2 − 0.2 − 0.7

Season pxTa Phytop. pxFe pxPb pxSiO2 pxAl2O3 pxMn pxRb ft pxZn pxBi pxNi fB hillsh fS fm fPREM relslop pxAs channbl fClay fCu pxK2O fZn fP fAl pxCu pxNb pxZr pxCl pxTi fSand pxY pxP2O5 vertdis slope fNtotal fHAl fT fOM pxCe fMn fK twi fpH pxCaO fFe fCa pxV fV mrvbf csc aspect fMg fSilt fSB pxCr longcurv valleydep mrrtf

9.2 7.8 6.6 5.9 5.9 4.7 4.5 4.2 4.1 3.4 3.1 3.1 2.9 2.6 2.6 2.5 2.3 2.2 2.2 2.1 1.6 1.6 1.5 1.3 1.3 1.2 1.1 1 1 0.9 0.9 0.9 0.8 0.7 0.7 0.6 0.6 0.5 0.5 0.5 0.3 0.3 0.1 0.1 0.1 0 0 0 − 0.2 − 0.2 − 0.3 − 0.4 − 0.4 − 0.4 − 0.6 − 0.6 − 0.8 − 0.8 − 1.1 − 1.4 − 2

pxMn

Phytop. pxAs fClay pxTa pxZn pxCu fPREM pxPb fpH pxAl2O3 vertdis ft pxCr fSand fm pxCe pxSiO2 relslop fAl pxY fZn pxBi fNtotal pxK2O channbl fOM fMg pxTi fT pxV fSB pxCl pxZr pxCaO longcurv fV fK hillsh fP pxRb slope valleydep fS fCa pxP2O5 fFe fB fMn fHAl mrrtf csc pxNi Season twi mrvbf aspect fSilt pxNb fCu

Phytop. fAl pxZn fm ft fClay fOM fpH pxTa fNtotal valleydep pxCr pxFe pxY fCu relslop pxAl2O3 vertdis channbl pxCe pxAs fCa pxRb pxBi pxMn pxPb slope fT mrvbf pxNi pxSiO2 fV pxK2O pxNb pxTi hillsh fMg fFe fS fSand pxZr pxCl fMn fPREM fZn pxP2O5 fP Season fB csc pxCaO fHAl mrrtf twi fK fSB pxV aspect fSilt longcurv

Var. – Variable. “f” and “px” subscripts refer to variables related to soil fertility/texture and pXRF, respectively.

greatest importance for the five enzymes activity prediction, pXRF variables were responsible for 10 of them, while soil physicochemical properties represented nine. Phytophysiognomy was not among the five most important variables only for FDA prediction, while season was present twice among the most important ones. Terrain variables did not appear among the most important variables for the models probably because they are constant throughout the year, as opposed to the en­ zymes activity.

3.2. Prediction of soil enzymes The overall results for soil enzymes prediction yielded R2 values ranging from 0.63 (FDA) to 0.82 (β-glucosidase) considering all data sets (Table 3). However, considering only the best R2 per soil enzyme, values ranged from 0.73 (FDA and urease) to 0.82 (β-glucosidase). The best results were delivered by the data sets including soil fertility and texture data (F data set) alone or in combination with pXRF data (pX data set). In general, when F variables were not included as predictors, the results 5

A.F. dos Santos Teixeira et al.

Catena 199 (2021) 105083

Table 3 Root mean square error (RMSE), coefficient of determination (R2), mean absolute error (MAE) and predictive squared correlation coefficient (Q2) for the prediction of hydrolysis of fluorescein diacetate (FDA) and the activity of the soil enzymes β-glucosidase, acid phosphatase, alkaline phosphatase and urease through models with varying predictor variables: soil fertility and texture (F), portable X-ray fluorescence (pX), terrain attributes (T), the 15 most important variables (15+), and all variables with positive importance (All+). RMSE FDA β-glucosidase Acid Phosphatase Alkaline Phosphatase Urease R2 FDA β-glucosidase Acid Phosphatase Alkaline Phosphatase Urease MAE FDA β-glucosidase Acid Phosphatase Alkaline Phosphatase Urease Q2 FDA β-glucosidase Acid Phosphatase Alkaline Phosphatase Urease

F

F þ pX

F þ pX þ T

FþT

pX

pX þ T

T

15þ

All þ

1.26 13.89 11.80 18.73 2.21

1.27 14.45 12.43 18.41 2.18

1.26 14.48 12.26 18.69 2.20

1.27 14.09 12.21 18.98 2.19

1.39 16.72 13.16 19.27 2.26

1.37 16.43 12.85 19.11 2.28

1.40 15.86 13.85 21.73 2.32

1.26 17.14 12.82 18.71 2.15

1.27 14.49 12.59 18.47 2.17

0.73 0.82 0.78 0.76 0.72

0.70 0.80 0.75 0.76 0.71

0.71 0.80 0.76 0.76 0.71

0.72 0.82 0.73 0.75 0.73

0.63 0.72 0.72 0.74 0.69

0.65 0.74 0.73 0.74 0.69

0.69 0.79 0.71 0.68 0.72

0.69 0.70 0.73 0.75 0.73

0.70 0.81 0.74 0.76 0.72

0.96 11.29 9.65 14.87 1.68

1.03 11.77 10.27 14.73 1.70

1.01 11.80 10.26 14.80 1.70

0.99 11.47 1.70 15.17 1.68

1.13 13.54 10.95 15.50 1.75

1.12 13.33 10.85 15.29 1.77

1.08 12.55 11.24 16.96 1.83

1.00 13.67 10.77 15.13 1.67

1.02 11.78 10.59 14.73 1.67

0.17 0.69 0.61 0.60 − 0.01

0.26 0.69 0.65 0.69 0.23

0.23 0.70 0.65 0.70 0.22

0.17 0.69 0.65 0.62 0.11

0.04 0.57 0.41 0.69 0.12

0.09 0.60 0.62 0.68 0.06

− 0.30 0.43 0.61 0.31 − 0.14

0.26 0.58 0.66 0.68 0.17

0.24 0.69 0.66 0.70 0.23

were slightly worse than those obtained by using F variables. However, for all the enzymes, data sets containing only terrain data as predictors (T data set) achieved R2 values very close to those obtained by the best data set per enzyme (differences of R2 values from 0.02 to 0.08). Greater R2 values were found for acid phosphatase and alkaline phosphatase predictions using pXRF data set (0.72 and 0.74, respec­ tively) than those obtained by the T data set (0.71 and 0.68, respec­ tively), although they were slightly worse than F data set (0.78 and 0.76, respectively). These results indicate that both terrain and pXRF data can provide prediction accuracies similar to those delivered by F data, but requiring less time than that needed for obtaining F data. The lowest R2 was obtained for the FDA model using only pXRF data (R2 = 0.63). Values of Q2 were lower than R2, but in general they followed the trend of R2 values. These values ranged from − 0.30 (FDA) to 0.70 (β-glucosidase and alkaline phosphatase) in agreement with the lowest and the greatest R2 values. β-glucosidase, acid and alkaline phosphatases presented Q2 values greater than 0.5 for almost all data sets. This value is considered adequate to proof the capacity of the prediction models to deliver accurate results when using leave-one-out cross validation (Hosseini et al., 2015). The RMSE and MAE values behaved similarly to each other and contrarily to R2 and Q2, as expected (Table 3). Considering all enzymes activity, the highest values of these parameters were achieved by the models for alkaline phosphatase prediction, while acid phosphatase presented lower values of RMSE and MAE. The selection of predictor variables by importance for enzymes prediction (both most important variables with positive importance and the 15 most important variables for predicting each enzyme) did not improve the results compared to the data sets that used all predictors, contrary to the findings of (Millard and Richardson, 2015). Based on the lowest RMSE and the greatest R2, in this order of importance, the best data set for the prediction of each soil enzyme activity was determined: F for FDA (RMSE = 1.26, R2 = 0.73), for β-glucosidase (RMSE = 13.89, R2 = 0.82) and for acid phosphatase (RMSE = 11.80, R2 = 0.78); and F + pX for both alkaline phosphatase (RMSE = 18.41, R2 = 0.76) and urease (RMSE = 2.18, R2 = 0.71). Although pX and T data sets alone or com­ bined could not outperform the predictions using F data, in most cases their results were similar to the ones achieved using F data.

The predicted values determined by the best models for each soil enzymes activity were compared with the observed values (Fig. 2). The observed vs. predicted plots for FDA (F data set), β-glucosidase (F data set) and urease (F + pX data set) were closer to the trend line, showing the good accuracy of models as demonstrated in Table 3. 3.3. Mapping the spatial variability of soil enzymes activity Although the models generated with only T data set were not the best, their good results indicate they can also provide accurate pre­ dictions of soil enzymes activity. For instance, the R2 values generated from the predictions using only terrain data ranged from 0.68 (alkaline phosphatase) to (0.79 (β-glucosidase). Besides, these variables are continuously distributed throughout the study area, which allows mapping the spatial variability of soil enzymes activity through the prediction models based on terrain data. Furthermore, since the models can be used for predictions in both warm/humid and cold/dry seasons by simply informing to the models the season of interest, maps of the temporal variability of soil enzymes activity can be created. The models generated only with terrain attribute data were used for enzymes and FDA mapping in the phytophysiognomies evaluated in the two seasons. As an example of this approach, the maps showing the spatiotemporal variability of β-glucosidase are presented in Fig. 3, since the predictions for this soil enzyme activity were the best ones using terrain data (R2 = 0.79, RMSE = 15.86, MAE = 12.55, Q2 = 0.43). Due to the length of the manuscript, the maps of the other enzymes activity and FDA are presented in the Supplementary Material. First, the maps per phytophysiognomy showed different values, with Neotropical Savanna and Atlantic Forest presenting greater values of β-glucosidase than Ironstone Outcrops and Rehabilitated Area. The same trend occurred for the other four microbiological attributes. Moreover, the range of values varied between phytophisiognomies and between seasons. For β-glucosidase, greater values were found during the warm/ rainy season as well as for FDA. The opposite trend was observed for urease. Conversely, season had little influence on the values of acid and alkaline phosphatases. This approach can discriminate areas with varying soil enzymes activity, providing support for further in­ vestigations about this variability within an area of interest. 6

A.F. dos Santos Teixeira et al.

Catena 199 (2021) 105083

Fig. 2. Comparison between observed values and values predicted by the best set of predictors considering the different prediction models tested for hydrolysis of fluorescein diacetate (FDA) and the activity of the soil enzymes β-glucosidase, acid phosphatase, alkaline phosphatase and urease. F = soil fertility and texture properties; T = terrain attributes; pX = elemental contents obtained by portable fluorescence X-ray spectrometer.

4. Discussion

and reported significant changes of their activity according to phyto­ physiognomy and season. While phytophysiognomy appeared among the three most important variables for the prediction of β-glucosidase, acid phosphatase, alkaline phosphatase and urease, it was not important for FDA prediction, which is used as a measure of general enzymatic activity. The phytophysiognomy and the type of vegetation cover pro­ vided more influence on soil enzymes activity than climate and topog­ raphy in Mediterranean shrublands (Mayor et al., 2016), confirming the importance of these variables for the models developed herein. Simi­ larly, He et al. (2020) reported that soil enzymes activity was greater influenced by the type of vegetation than by climatic conditions. Tan et al. (2014) reported strong variations in soil enzymes activity ac­ cording to the type of land use.

4.1. Variables importance and prediction of soil enzymes activity Among the variables used in the models, phytophysiognomy, soil fertility and season are already well known for influencing soil micro­ biology attributes (Alkorta et al., 2017; Bünemann et al., 2018; Ravin­ ˇ ´ dran and Yang, 2015; Silva et al., 2018; Stone et al., 2015; Stursov a et al., 2016). This study demonstrated that the soil enzymes activity here evaluated are differently influenced by these factors, according to the different importance of phytophysiognomy, soil fertility and season for the prediction models (Table 2). Moreover, Silva et al. (2018) evaluated the statistical differences between enzymes activity in this same area 7

A.F. dos Santos Teixeira et al.

Catena 199 (2021) 105083

Fig. 3. Spatial variability of the predicted values of β-glucosidase activity based on the model generated with terrain attributes (T) for the Ironstone Outcrops, Rehabilitated Area, Neotropical Savanna and Atlantic Forest phytophysiognomies in both warm/humid (rainy) and cold/dry seasons.

8

A.F. dos Santos Teixeira et al.

Catena 199 (2021) 105083

For FDA prediction, season appeared as the most important variable, followed by available S and B contents (F data set). Season was also the most important variable for urease prediction. By representing the dif­ ferences of temperature and humidity along the year, especially during the warm/humid summer and the cold/dry winter, season has consid­ erable importance for soil enzymes activity and caused statistically ´tero Ferrífero region (Silva differences in their activity in the Quadrila et al., 2018). Soil moisture, which is variable according to the season, was found to be correlated to soil microbiological and biochemical ac­ tivity along with soil types (Borowik and Wyszkowska, 2016). The variables importance demonstrated that the enzymatic activities vary in different intensities when phytophysiognomy and/or season change, as evidenced by Silva et al. (2018). Also, this analysis discriminated the importance of each variable used as predictor, e.g., the importance of pXRF and soil fertility and texture data (Table 2). In addition to phytophysiognomy and season, variables related to soil fertility and texture commonly appeared among the most important ones in the models for predicting soil microbiology attributes. The variables of the F data set among the five most important ones for the predictions include cation exchange capacity, clay content, available S and B, exchangeable Mg and Al3+ saturation. These variables along with other soil physicochemical properties such as electrical conductivity, pH, organic carbon, soil texture and nutrient contents influence soil enzymes activity (Bünemann et al., 2018; Forkuor et al., 2017; Hengl et al., 2017); consequently, they improved the accuracy of the prediction models. Rodríguez-Loinaz et al. (2008) found strong relationships be­ tween soil fertility properties and β-glucosidase, acid and alkaline ´z˙ yło and phosphatases and urease in forests from Spain. Moreover, Ro Bohacz (2020) found positive correlations between enzyme activity and nutrient contents in a Podzol from Poland. Overall, pXRF variables presented considerable importance for the models occupying at least two positions among the five most important variables for the models, except for B-glucosidase prediction. This equipment has been successfully used for rapid and low cost soil char­ acterization due to the relations between total elemental data and soil physicochemical properties (Tavares et al., 2020, Mukhopadhyay et al., 2020, Wan et al., 2020, Silva et al., 2020, Benedet et al., 2020, Qu et al., 2019, Rawal et al., 2019, Duda et al., 2017). Regarding relations to soil microbiological properties, studies involving pXRF data are still initi­ ating, although other proximal sensors, such as the visible near-infrared (Vis-NIR) diffuse reflectance spectroscopy, have succeeded to predict soil biological attributes (Weindorf et al., 2018). ´tero As the evaluated phytophysiognomies are located in the Quadrila Ferrífero, a region with highly weathered soils derived from parent materials with high concentrations of Fe and Mn (Carvalho Filho et al., 2010), the mineralogy of these soils is rich in Fe and Mn oxides, known for their high P fixing capacity (K¨ ampf et al., 2012; Resende et al., 2019, 2014). We believe that the importance of these elements obtained by pXRF was caused by their relation to the mineralogy of the soils of the region. The influence of these oxides has been observed for phosphatase and urease, promoting the abiotic polymerization of phenolic com­ pounds, with the adsorbed enzyme molecules presenting greater activ­ ities when exposed to those oxides (Gianfreda et al., 2002). Mancini et al. (2019) demonstrated that pXRF is capable of detecting even slight chemical variations of parent materials via analysis of the soils derived from them, which also relates to soil mineralogy. This fact explains why the importance of these elements obtained by pXRF was greater than the available contents of these same elements for the predictions. Another explanation for the influence of these elements on acid phosphatase is the P fixation ability of these soils. As the soils of the region are generally acidic and have a high P fixation capacity (Motta et al., 2002), the higher acid phosphatase activity may be related to the lower P availability in these environments, which is also a reflection of their mineralogy (Resende et al., 2014). Conversely, for alkaline phosphatase prediction, pxCu was the most important variable. This may be related to the effect of Cu on the

decrease of alkaline phosphatase activity, which was already demon­ strated in studies in areas under influence of this element (Mounissamy et al., 2017; Wang et al., 2007). The importance of pxTa for urease and pxAs for FDA predictions is probably related to their toxic effect to the microbial population, being able to reduce the enzymatic activity (Li et al., 2016; Nadimi-Goki et al., 2018; Oladipo et al., 2014). Regarding the accuracy of the predictions, the best results were delivered by models using soil fertility and texture data (F data set) as predictor variables (Table 2). Soil fertility, which is related to the increment of nutrients in soils, has proved strongly correlated with en­ zymes activity since several enzymes participate of nutrient cycling processes, such as those of carbon, nitrogen and phosphorus cycles (Adetunji et al., 2017, Ullah et al., 2019). Despite F data set, the pXRF and terrain variables could also yield adequate predictions, indicating they are alternative data to predict soil enzymes activity. In this sense, Comino et al. (2018) could predict β-Glucosidase, alkaline phosphatase and acid phosphatase activities via infrared spectroscopy and Partial Least Squares algorithm, although the results were lesser accurate than the ones found in this current work using pXRF data and cforest algo­ rithm. Similarly, Weindorf et al. (2018) used Vis-NIR spectroscopy data and machine learning algorithms for the prediction of different soil biological attributes, reducing the time needed for such evaluations. Although these works are still initiating, requiring further tests under different conditions, they demonstrate that proximal sensors can become an important tool for studies related to the assessment of soil microbiological activity (Comino et al., 2018). In general, the models using terrain data also presented adequate performance, although slightly worse than the models including F data set. For instance, the R2 values for the predictions using only terrain data ranged from 0.68 to 0.79, while these values using only F data varied from 0.72 to 0.82. Since terrain variables are easily obtained and represent the variability of data continuously, they have been widely applied to mapping soil properties based on the relations between them and topography (Adhikari et al., 2013; Arrouays et al., 2017; Machado et al., 2019; Silva et al., 2016). This same approach was applied in this work as an example for spatial variability of soil enzymes activity (see Sections 3.3 and 4.2). 4.2. Spatial variability of the soil enzymes activity The spatial analysis of soil enzymes activity showed the stability of the acid and alkaline phosphatases values at the different seasons (Supplementary Material). The stability of the phosphatase had been previously reported by Lopes et al. (2018), who observed that acid phosphatase did not present significant variations over 5 years of eval­ uations conducted in Brazilian Cerrado. They argue that this stability is probably due to the fact that phytophysiognomy have reached a certain equilibrium stage. Stability of acid phosphatase was also reported by ˜o et al. (2020) in different Brazilian soils, including some soils Araga similar to the ones evaluated in our study. Thus, the differences in phosphatase activity are more evident between phytophysiognomies or when there are changes in relation to the equilibrium condition. In this same work of Lopes et al. (2018), the authors reported that urease was the only enzyme that had its activity reduced in the rainy season, similarly to the findings of our work (Supplementary Material). Fan et al. (2019) also reported reduced urease activity in the rainy season, prob­ ably due to the more intense leaching of nutrients in this season pre­ venting its activity. To relate microbiological attributes, such as enzymes, to soil quality has become an increasing trend worldwide (Spohn et al., 2013). In this sense, the spatial variability maps of soil enzymes activity, mainly those created based on easily accessible data, e.g. terrain data, are an impor­ tant alternative for assessing soil quality, since these attributes can vary under slight changes of the environmental conditions, such as those caused by soil physicochemical properties, season and terrain features (Paz-Ferreiro and Fu, 2016; Teixeira et al., 2019) (Fig. 3 and 9

Catena 199 (2021) 105083

A.F. dos Santos Teixeira et al.

Supplementary Material). Tan et al. (2014) used a geostatistical approach for studying the spatial variability of soil enzymes activity, including urease and phosphatase, correlating these results with soil quality in China. This current study demonstrated that relating terrain features with soil microbiological attributes variability may contribute for spatial variability analyses of these attributes, as reported by Teixeira et al. (2019) along a toposequence in Brazil. Moreover, punctual and accurate predictions of soil enzymes activity can be achieved using phytophysiognomy, season, soil physicochemical properties, total elemental contents obtained by pXRF and terrain data. These predictions may not only accelerate, but also facilitate the determination of soil enzymes activity, enabling inferences of soil quality based on these soil microbiological attributes. This approach presented herein intends to demonstrate the feasibility of predicting soil enzymes activity as a first attempt on this matter under tropical conditions. However, the authors encourage further investigations on this aspect, especially under different environmental conditions and encompassing a larger number of samples.

Denmark. Soil Sci. Soc. Am. J. 77, 860–876. https://doi.org/10.2136/ sssaj2012.0275. Alkorta, I., Epelde, L., Garbisu, C., 2017. Environmental parameters altered by climate change affect the activity of soil microorganisms involved in bioremediation. FEMS Microbiol. Lett. 364 https://doi.org/10.1093/femsle/fnx200. Alvares, C.A., Stape, J.L., Sentelhas, P.C., de Moraes Gonçalves, J.L., Sparovek, G., 2013. K¨ oppen’s climate classification map for Brazil. Meteorol. Zeitschrift 22, 711–728. https://doi.org/10.1127/0941-2948/2013/0507. Alvarez V., V.H., Fonseca, D.M., 1990. Definition of phosphorus doses for the determination of the maximum phosphate adsorption capacity and for greenhouse trials. Rev. Bras. Cienc. do Solo 14, 49–55. Alvarez V., V.H., Novais, R.F. de, Barros, N.F. de, Cantarutti, R.B., Lopes, A.S., 1999. Interpretation of the results of soil analysis. In: Ribeiro, A.C., Guimar˜ aes, P.T.G., Alvarez V., V.H. (Eds.), Recommendations for the Use of Corrective and Fertilizers in Minas Gerais - 5th Approach. Soil Fertility Commission of the State of Minas Gerais, Viçosa, MG, pp. 25–32. Arag˜ ao, O.O.S., Oliveira-Longatti, S.M., Souza, A.A., Jesus, E.C., MerloOliveira, M.N.E.P., Moreira, F.M.S., 2020. The effectiveness of a microbiological attribute as a soil quality indicator depends on the storage time of the sample. J. Soil Sci. Plant Nutr. 20, 2525–2535. https://doi.org/10.1016/j.geoderma.2020.114212. Arrouays, D., Lagacherie, P., Hartemink, A.E., 2017. Digital soil mapping across the globe. Geoderma Reg. 9, 1–4. https://doi.org/10.1016/j.geodrs.2017.03.002. Benedet, L., Faria, W.M., Silva, S.H.G., Mancini, M., Guilherme, L.R.G., Demattˆe, J.A.M., Curi, N., 2020. Soil subgroup prediction via portable X-ray fluorescence and visible near-infrared spectroscopy. Geoderma 365, 114212. https://doi.org/10.1016/j. geoderma.2020.114212. Beven, K.J., Kirkby, M.J., 1979. A physically based, variable contributing area model of ` base physique de zone d’appel variable de basin hydrology / Un mod` ele a l’hydrologie du bassin versant. Hydrol. Sci. Bull. 24, 43–69. https://doi.org/ 10.1080/02626667909491834. Borowik, A., Wyszkowska, J., 2016. Soil moisture as a factor affecting the microbiological and biochemical activity of soil. Plant, Soil Environ. 62, 250–255. https://doi.org/10.17221/158/2016-PSE. Breiman, L., 2001. Random forests. Mach. Learn. 45, 5–32. Bouyoucos, G.Q., 1951. A recalibration of the hydrometer method for making analysis of soils. Agron. J. 43, 434–437. Bünemann, E.K., Bongiorno, G., Bai, Z., Creamer, R.E., De Deyn, G., de Goede, R., Fleskens, L., Geissen, V., Kuyper, T.W., M¨ ader, P., Pulleman, M., Sukkel, W., van Groenigen, J.W., Brussaard, L., 2018. Soil quality – A critical review. Soil Biol. Biochem. 120, 105–125. https://doi.org/10.1016/j.soilbio.2018.01.030. de Carvalho Filho, A., Curi, N., Shinzato, E., 2010. Relaç˜ oes solo-paisagem no Quadril´ atero Ferrífero em Minas Gerais. Pesqui. Agropecu´ aria Bras. 45, 903–916. https://doi.org/10.1590/S0100-204X2010000800017. Castro, J.L. de, Souza, M.G., Rufini, M., Guimar˜ aes, A.A., Rodrigues, T.L., Moreira, F.M. de S., 2017. Diversity and efficiency of rhizobia communities from iron mining areas using cowpea as a trap plant. Rev. Bras. Cienc. do Solo 41, 1–20. https://doi.org /10.1590/18069657rbcs20160525. Chakraborty, S., Weindorf, D.C., Michaelson, G.A.R.Y.J., Ping, C.L., Choudhury, A., Kandakji, T., Acree, A., Sharma, A., Wang, D., 2016. In-situ differentiation of acidic and non-acidic tundra via portable X-ray fluorescence (PXRF) spectrometry. Pedosphere 26, 549–560. https://doi.org/10.1016/S1002-0160(15)60064-9. Comino, F., Aranda, V., García-Ruiz, R., Ayora-Ca˜ nada, M.J., Domínguez-Vidal, A., 2018. Infrared spectroscopy as a tool for the assessment of soil biological quality in agricultural soils under contrasting management practices. Ecol. Indic. 87, 117–126. https://doi.org/10.1016/j.ecolind.2017.12.046. Conrad, O., Bechtel, B., Bock, M., Dietrich, H., Fischer, E., Gerlitz, L., Wehberg, J., Wichmann, V., B¨ ohner, J., 2015. System for automated geoscientific analyses (SAGA) v. 2.1.4. Geosci. Model Dev. 8, 1991–2007. https://doi.org/10.5194/gmd-81991-2015. Coelho, M.R., Vasques, G.M., Tassinari, D., Souza, Z.R., Oliveira, A.P., Moreira, F.M.S., 2017. In: Solos do Quadril´ atero Ferrífero sob Diferentes Coberturas Vegetais e Materiais de Origem. Embrapa Solos, Rio de Janeiro, RJ. Dick, R.P., Breakwell, D.P., Turco, R.F., 1996. Soil enzyme activities and biodiversity measurements as integrative microbiological indicators. In: Doram, J.W., Jones, A.J. (Eds.), Methods for Assessing Soil Quality. Soil Science Society of America, Madison, pp. 247–272. dos Santos, J.V., Var´ on-L´ opez, M., Fonsˆ eca Sousa Soares, C.R., Lopes Leal, P., Siqueira, J. O., de Souza Moreira, F.M., 2016. Biological attributes of rehabilitated soils contaminated with heavy metals. Environ. Sci. Pollut. Res. 23, 6735–6748. htt ps://doi.org/10.1007/s11356-015-5904-6. Duda, B.M., Weindorf, D.C., Chakraborty, S., Li, B., Man, T., Paulette, L., Deb, S., 2017. Soil characterization across catenas via advanced proximal sensors. Geoderma 298, 78–91. https://doi.org/10.1016/j.geoderma.2017.03.017. Eivazi, F., Tabatabai, M.A., 1988. Glucosidases and galactosidases in soils. Soil Biol. Biochem. 20, 601–606. https://doi.org/10.1016/0038-0717(88)90141-1. Eivazi, F., Tabatabai, M.A., 1977. Phosphatases in soils. Soil Biol. Biochem. 9, 167–172. https://doi.org/10.1016/0038-0717(77)90070-0. Fan, Z., Lu, S., Liu, S., Li, Z., Hong, J., Zhou, J., Peng, X., 2019. The effects of vegetation restoration strategies and seasons on soil enzyme activities in the Karst landscapes of Yunnan, southwest China. J. For. Res. https://doi.org/10.1007/s11676-019-009590. Forkuor, G., Hounkpatin, O.K.L., Welp, G., Thiel, M., 2017. High resolution mapping of soil properties using remote sensing variables in South-Western Burkina Faso: a comparison of machine learning and multiple linear regression models. PLoS One 12, e0170478. https://doi.org/10.1371/journal.pone.0170478.

5. Conclusions Soil physicochemical properties, total elemental contents obtained by pXRF and terrain attributes along with phytophysiognomy and sea­ son information provided the generation of accurate prediction models for soil enzymes activity. Predictions of soil enzymes activity based on soil fertility and texture provided the best results (R2 ranging from 0.63 to 0.82), while predictions based on pXRF and terrain data were also accurate, but slightly worse than those using soil fertility and texture data. Adequate predictions based on terrain data allowed mapping the spatial variability of soil enzymes activity, providing a better overview of their variability across each studied phytophysiognomy in different seasons of the year. This approach does not intend to propose the replacement of conventional analyses of soil enzymes activity; it only demonstrates that it is possible to predict soil enzymes activity and reduce the number of laboratory analyses needed for a more detailed characterization of their activity both punctually or spatially over a large area across the landscape. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgement The authors are grateful for the project [CRA- RDP- 00136- 10 ` (FAPEMIG/ FAPESP/ FAPESPA/ Vale SA)], the Fundaç˜ ao de Amparo a ˜o de Pesquisa do Estado de Minas Gerais - FAPEMIG, the Coordenaça Aperfeiçoamento de Pessoal de Nível Superior - Capes and the Conselho ´gico - CNPq for Nacional de Desenvolvimento Científico e Tecnolo financial support and scholarships granted, and to Luiz Roberto Guimar˜ aes Guilherme, for lending us the pXRF. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi. org/10.1016/j.catena.2020.105083. These data include Google maps of the most important areas described in this article. References Adetunji, A.T., Lewu, F.B., Mulidzi, R., Ncube, B., 2017. The biological activities of β-glucosidase, phosphatase and urease as soil quality indicators: a review. J. Soil Sci. Plant Nutr. 17, 794–807. https://doi.org/10.4067/S0718-95162017000300018. Adhikari, K., Kheir, R.B., Greve, M.B., Bøcher, P.K., Malone, B.P., Minasny, B., McBratney, A.B., Greve, M.H., 2013. High-resolution 3-D mapping of soil texture in

10

A.F. dos Santos Teixeira et al.

Catena 199 (2021) 105083

Gianfreda, L., Rao, M.A., Sannino, F., Saccomandi, F., Violante, A., 2002. Enzymes in soil: properties, behavior and potential applications. Developments in Soil Science. 301–327. https://doi.org/10.1016/S0166-2481(02)80027-7. He, Q., Wu, Y., Bing, H., Zhou, J., Wang, J., 2020. Vegetation type rather than climate modulates the variation in soil enzyme activities and stoichiometry in subalpine forests in the eastern Tibetan Plateau. Geoderma 374, 114424. https://doi.org/ 10.1016/j.geoderma.2020.114424. Hengl, T., Leenaars, J.G.B., Shepherd, K.D., Walsh, M.G., Heuvelink, G.B.M., Mamo, T., Tilahun, H., Berkhout, E., Cooper, M., Fegraus, E., Wheeler, I., Kwabena, N.A., 2017. Soil nutrient maps of Sub-Saharan Africa: assessment of soil nutrient content at 250 m spatial resolution using machine learning. Nutr. Cycl. Agroecosystems 109, 77–102. https://doi.org/10.1007/s10705-017-9870-x. Hoeft, R.G., Walsh, L.M., Kenney, D.R., 1973. Evaluation of various extractants for available soil sulfur. Soil Science Society American Proceeding, Madison. Hothorn, T., Hornik, K., Zeileis, A., 2006. Unbiased recursive partitioning: a conditional inference framework. J. Comput. Graph. Stat. 15, 651–674. https://doi.org/ 10.1198/106186006X133933. Hosseini, R., Newlands, N.K., Dean, C.B., Tekemura, Al, 2015. Statistical modeling of soil moistuer, integrating satellite remote-sensing (SAR) and ground-based data. Remote Sens. 7, 2752–2780. https://doi.org/10.3390/rs70302752. IUSS Working Group WRB, 2014. World Reference Base for Soil Resources 2014, update 2015 - International soil classification system for naming soils and creating legends for soil maps. World Soil Resources Reports No. 106. FAO, Rome. Joergensen, R.G., Brookes, P.C., 1990. Ninhydrin-reactive nitrogen measurements of microbial biomass in 0.5 m K2SO4 soil extracts. Soil Biol. Biochem. 22, 1023–1027. https://doi.org/10.1016/0038-0717(90)90027-W. K¨ ampf, N., Marques, J.J., Curi, N., 2012. Mineralogia de Solos Brasileiros. In: Pedologia Fundamentos. SBCS, Viçosa, MG, p. 343. Keeney, D.R., Nelson, D.W., 1982. Nitrogen organic forms, in: Page, A.L. (Ed.), Methods of Soil Analysis: Chemical and Microbiological Properties. American Society of Agronomy/Soil Science Society of America, Madison, pp. 643–698. Kuhn, M., Wing, J., Weston, S., Williams, A., Keefer, C., Engelhardt, A., Cooper, T., Mayer, Z., Kenkel, B., Team, R.C., Benesty, M., Lescarbeau, R., Ziem, A., Scrucca, L., Tang, Y., Candan, C., Hunt, T., 2018. Package ‘ caret.’. Li, Z., Ma, T., Yuan, C., Hou, J., Wang, Q., Wu, L., Christie, P., Luo, Y., 2016. Metal contamination status of the soil-plant system and effects on the soil microbial community near a rare metal recycling smelter. Environ. Sci. Pollut. Res. 23, 17625–17634. https://doi.org/10.1007/s11356-016-6958-9. Liaw, A., Wiener, M., 2002. Classification and Regression by randomForest. R News 2, 18–22. Liu, F., Geng, X., Zhu, a. X., Fraser, W., Waddell, A., 2012. Soil texture mapping over low relief areas using land surface feedback dynamic patterns extracted from MODIS. Geoderma 171–172, 44–52. https://doi.org/10.1016/j.geoderma.2011.05.007. Lopes, A.A.C., Sousa, D.M.G., dos Reis, F.B., Figueiredo, C.C., Malaquias, J.V., Souza, L. M., Carvalho Mendes, I., 2018. Temporal variation and critical limits of microbial indicators in Oxisols in the Cerrado, Brazil. Geoderma Reg. 12, 72–82. https://doi. org/10.1016/j.geodrs.2018.01.003. Machado, D.F.T., De Menezes, M.D., Silva, S.H.G., Curi, N., 2019. Transferability, accuracy, and uncertainty assessment of different knowledge-based approaches for soil types mapping. Catena 182, 104134. https://doi.org/10.1016/j. catena.2019.104134. Malone, B.P., Styc, Q., Minasny, B., McBratney, A.B., 2017. Digital soil mapping of soil carbon at the farm scale: A spatial downscaling approach in consideration of measured and uncertain data. Geoderma 290, 91–99. https://doi.org/10.1016/j. geoderma.2016.12.008. Mancini, M., Silva, S.H.G., Teixeira, A.F. dos S., Guilherme, L.R.G., Curi, N., 2020. Soil parent material prediction for Brazil via proximal soil sensing. Geoderma Reg. 22, e00310. https://doi.org/10.1016/j.geodrs.2020.e00310. Mancini, M., Weindorf, D.C., Chakraborty, S., Silva, S.H.G., dos Santos Teixeira, A.F., Guilherme, L.R.G., Curi, N., 2019. Tracing tropical soil parent material analysis via portable X-ray fluorescence (pXRF) spectrometry in Brazilian Cerrado. Geoderma 337, 718–728. https://doi.org/10.1016/j.geoderma.2018.10.026. Martens, H.A., Dardenne, P., 1998. Validation and verification of regression in small data sets. Chemometr. Intell. Lab. 44, 99–121. ´ Mayor, A.G., Goir´ an, S.B., Vallejo, V.R., Bautista, S., 2016. Variation in soil enzyme activity as a function of vegetation amount, type, and spatial structure in fire-prone Mediterranean shrublands. Sci. Total Environ. 573, 1209–1216. https://doi.org/ 10.1016/j.scitotenv.2016.03.139. Mclean, E.O., Hedleson, M.R., Bartlett, R.J., Holowaychuk, D., 1958. Aluminium in soils: I. Extraction methods and magnitud clays in Ohio soils. Soil Sci. Soc. Am. Proc. 22, 382–387. Mehlich, A., 1953. Determination of P, Ca, Mg, K, Na and NH4. North Carolina Soil Testing Division, Raleigh. Menezes, M.D., Bispo, F.H.A., Faria, W.M., Gonçalves, M.G.M., Curi, N., Guilherme, L.R. G., 2020. Modeling arsenic content in Brazilian soils: what is relevant? Sci. Total Environ. 712, 136511 https://doi.org/10.1016/j.scitotenv.2020.136511. Millard, K., Richardson, M., 2015. On the importance of training data sample selection in random forest image classification: a case study in peatland ecosystem mapping. Remote Sensing 7, 8489–8515. https://doi.org/10.3390/rs70708489. Motta, P.E.F., Curi, N., Siqueira, J.O., Van Raij, B., Furtini Neto, A.E., Lima, J.M., 2002. Adsorption and forms of phosphorus in latosols: influence of mineralogy and use. Rev. Bras. Ciˆ encia do Solo 26, 349–359. https://doi.org/10.1590/S010006832002000200008. Mounissamy, V.C., Kundu, S., Selladurai, R., Saha, J.K., Biswas, A.K., Adhikari, T., Patra, A.K., 2017. Effect of soil amendments on microbial resilience capacity of acid

soil under copper stress. Bull. Environ. Contam. Toxicol. 99, 625–632. https://doi. org/10.1007/s00128-017-2173-8. Mukhopadhyay, S., Chakraborty, S., Bhadoria, P.B.S., Li, B., Weindorf, D.C., 2020. Assessment of heavy metal and soil organic carbon by portable X-ray fluorescence spectrometry and NixProTM sensor in landfill soils of India. Geoderma Reg. 20, e00249 https://doi.org/10.1016/j.geodrs.2019.e00249. Nadimi-Goki, M., Bini, C., Wahsha, M., Kato, Y., Fornasier, F., 2018. Enzyme dynamics in contaminated paddy soils under different cropping patterns (NE Italy). J. Soils Sediments 18, 2157–2171. https://doi.org/10.1007/s11368-017-1830-1. Oladipo, O.G., Olayinka, A., Awotoye, O.O., 2014. Ecological impact of mining on soils of Southwestern Nigeria. Environ. Exp. Biol. 12, 179–186. Paz-Ferreiro, J., Fu, S., 2016. Biological indices for soil quality evaluation: perspectives and limitations. L. Degrad. Dev. 27, 14–25. https://doi.org/10.1002/ldr.2262. Pelegrino, M.H.P., Silva, S.H.G., de Menezes, M.D., da Silva, E., Owens, P.R., Curi, N., 2016. Mapping soils in two watersheds using legacy data and extrapolation for similar surrounding areas. Ciˆencia e Agrotecnologia 40, 534–546. https://doi.org/ 10.1590/1413-70542016405011416. Pelegrino, M.H.P., Weindorf, D.C., Silva, S.H.G., de Menezes, M.D., Poggere, G.C., Guilherme, L.R.G., Curi, N., 2018. Synthesis of proximal sensing, terrain analysis, and parent material information for available micronutrient prediction in tropical soils. Precis. Agric. 1–21 https://doi.org/10.1007/s11119-018-9608-z. Plante, A.F., 2007. Soil biogeochemical cycling of inorganic nutrients and metals. In: Soil Microbiology, Ecology and Biochemistry. Elsevier, pp. 389–432. https://doi.org/10 .1016/B978-0-08-047514-1.50019-6. Qu, M., Chen, J., Li, W., Zhang, C., Wan, M., Huang, B., Zhao, Y., 2019. Correction of insitu portable X-ray fluorescence (PXRF) data of soil heavy metal for enhancing spatial prediction. Environ. Pollut. 254, 112993 https://doi.org/10.1016/j. envpol.2019.112993. Qu, M., Wang, Y., Huang, B., Zhao, Y., 2018. Spatial uncertainty assessment of the environmental risk of soil copper using auxiliary portable X-ray fluorescence spectrometry data and soil pH. Environ. Pollut. 240, 184–190. https://doi.org/ 10.1016/j.envpol.2018.04.118. R Core Team, 2019. R: A language and environment for statistical computing. van Raij, B., Andrade, J.C., Cantarella, H., Quaggio, J.A., 2001. An´ alise química para avaliaç˜ ao da fertilidade de solos tropicais. Instituto Agronˆ omico de Campinas, Campinas. Ravindran, A., Yang, S.-S., 2015. Effects of vegetation type on microbial biomass carbon and nitrogen in subalpine mountain forest soils. J. Microbiol. Immunol. Infect. 48, 362–369. https://doi.org/10.1016/j.jmii.2014.02.003. Rawal, A., Chakraborty, S., Li, B., Lewis, K., Godoy, M., Paulette, L., Weindorf, D.C., 2019. Determination of base saturation percentage in agricultural soils via portable X-ray fluorescence spectrometer. Geoderma 338, 375–382. https://doi.org/ 10.1016/j.geoderma.2018.12.032. Resende, M., Curi, N., Rezende, S.B. de S.B., Corrˆ ea, G.F.G.F., Ker, J.C.J.C., 2014. Pedologia: Base para distinç˜ ao de ambientes, 6a ediç˜ ao. ed, Pedologia: Base para distinç˜ ao de ambientes. Editora UFLA, Lavras. Resende, M., Curi, N., Rezende, S.B., Silva, S.H.G., 2019. Da rocha ao solo: enfoque ambiental, 1st ed. Editora UFLA, Lavras. Ribeiro, B.T., Silva, S.H.G., Silva, E.A., Guilherme, L.R.G., 2017. Portable X-ray fluorescence (pXRF) applications in tropical Soil Science. Ciˆencia e Agrotecnologia 41, 245–254. https://doi.org/10.1590/1413-70542017413000117. Rodríguez-Loinaz, G., Onaindia, M., Amezaga, I., Mijangos, I., Garbisu, C., 2008. Relationship between vegetation diversity and soil functional diversity in native mixed-oak forests. Soil Biol. Biochem. 40, 49–60. https://doi.org/10.1016/j. soilbio.2007.04.015. R´ oz˙ yło, K., Bohacz, J., 2020. Microbial and enzyme analysis of soil after the agricultural utilization of biogas digestate and mineral mining waste. Int. J. Environ. Sci. Technol. 17, 1051–1062. https://doi.org/10.1007/s13762-019-02522-0. ´ de, Lumbreras, J.F., Santos, H.G. dos, Jacomine, P.K.T., Anjos, L.H.C. dos, Oliveira, V.A. Coelho, M.R., Almeida, J.A. de, Filho, J.C. de A., 2018. Sistema Brasileiro de Classificaç˜ ao de Solos, 5. ed., re. ed. Embrapa, Brasília, DF. Schaetzl, R.J., Anderson, S., 2005. Soil: Genesis and Geomorphology, first ed. Cambridge University Press, New York. Shoemaker, H.E., McLean, E.O., Pratt, P.F., 1961. Buffer methods for determining lime requirement of soils with appreciable amounts of extractable aluminum1. Soil Sci. Soc. Am. J. 25, 274. https://doi.org/10.2136/sssaj1961.03615995002500040014x. Silva, A.O., Costa, A.M., Teixeira, A.F. dos S., Guimar˜ aes, A.A., Santos, J.V., Moreira, F. M. de S., 2018. Soil microbiological attributes indicate recovery of an iron mining area and of the biological quality of adjacent phytophysiognomies. Ecol. Indic. 93, 142–151. https://doi.org/10.1016/j.ecolind.2018.04.073. Silva, S.H.G., Menezes, M.D. De, Owens, P.R., Curi, N., 2016. Retrieving pedologist’s mental model from existing soil map and comparing data mining tools for refining a larger area map under similar environmental conditions in Southeastern Brazil. Geoderma 267, 65–77. https://doi.org/10.1016/j.geoderma.2015.12.025. Silva, S.H.G., Teixeira, A.F. dos S., Menezes, M.D. de, Guilherme, L.R.G., Moreira, F.M. de S., Curi, N., 2017. Multiple linear regression and random forest to predict and map soil properties using data from portable X-ray fluorescence spectrometer (pXRF). Ciˆencia e Agrotecnologia 41, 648–664. https://doi.org/10.1590/1413-7054201741 6010317. Silva, S.H.G., Weindorf, D.C., Pinto, L.C., Faria, W.M., Acerbi Junior, F.W., Gomide, L.R., de Mello, J.M., de P´ adua Junior, A.L., de Souza, I.A., Teixeira, A.F. dos S., Guilherme, L.R.G., Curi, N., 2020. Soil texture prediction in tropical soils: A portable X-ray fluorescence spectrometry approach. Geoderma 362, 114136. https://doi. org/10.1016/j.geoderma.2019.114136. Silveira, E.M.O., Silva, S.H.G., Acerbi Junior, F.W., Carvalho, M.C., Carvalho, L.M.T., Scolforo, J.R.S., Wulder, M.A., 2019. Object-based random forest modelling of

11

A.F. dos Santos Teixeira et al.

Catena 199 (2021) 105083 Microbiological indicators of soil quality under native forests are influenced by topographic factors. An. Acad. Bras. Cienc. 91, e20180696 https://doi.org/10.1590/ 0001-3765201920189696. Teixeira, A.F. dos S., Weindorf, D.C., Silva, S.H.G., Guilherme, L.R.G., Curi, N., 2018. Portable X-ray fluorescence (pXRF) spectrometry applied to the prediction of chemical attributes in Inceptisols under different land uses. Ciˆencia e Agrotecnologia 42, 501–512. https://doi.org/10.1590/1413-70542018425017518. Ullah, S., Ai, C., Huang, S., Zhang, J., Jia, L., Ma, J., Zhou, W., He, P., 2019. The responses of extracellular enzyme activities and microbial community composition under nitrogen addition in an upland soil. PLoS One 14, 1–19. https://doi.org/ 10.1371/journal.pone.0223026. Vasu, D., Singh, S.K., Sahu, N., Tiwary, P., Chandran, P., Duraisami, V.P., Ramamurthy, V., Lalitha, M., Kalaiselvi, B., 2017. Assessment of spatial variability of soil properties using geospatial techniques for farm level nutrient management. Soil Tillage Res. 169, 25–34. https://doi.org/10.1016/j.still.2017.01.006. Walkley, A., Black, I.A., 1934. An examination of the Degtjareff method for determining soil organic matter and a proposed modification of the chromic acid titration method. Soil Sci. 37, 29–38. Wan, M., Hu, W., Qu, M., Li, W., Zhang, C., Kang, J., Hong, Y., Chen, Y., Huang, B., 2020. Rapid estimation of soil cation exchange capacity through sensor data fusion of portable XRF spectrometry and Vis-NIR spectroscopy. Geoderma 363, 114163. https://doi.org/10.1016/j.geoderma.2019.114163. Wang, Y., Shi, J., Wang, H., Lin, Q., Chen, X., Chen, Y., 2007. The influence of soil heavy metals pollution on soil microbial biomass, enzyme activity, and community composition near a copper smelter. Ecotoxicol. Environ. Saf. 67, 75–81. https://doi. org/10.1016/j.ecoenv.2006.03.007. Weindorf, D.C., Chakraborty, S., Moore-Kucera, J., Li, B., Fultz, L., Acosta-Martinez, V., Li, C., 2018. Advanced modeling of soil biological properties using visible near infrared diffuse reflectance spectroscopy. Int. J. Bioresour. Sci. 5, 1–20. https://doi. org/10.30954/2347-9655.01.2018.1. Weindorf, D.C., Bakr, N., Zhu, Y., 2014. Advances in portable X-ray fluorescence (PXRF) for environmental, pedological, and agronomic applications. Adv. Agron. 128, 1–45. https://doi.org/10.1016/B978-0-12-802139-2.00001-9. Weindorf, D.C., Zhu, Y., Haggard, B., Lofton, J., Chakraborty, S., Bakr, N., Zhang, W., Weindorf, W.C., Legoria, M., 2012. Enhanced pedon horizonation using portable Xray fluorescence spectrometry. Soil Sci. Soc. Am. J. 76, 522–531. https://doi.org/ 10.2136/sssaj2011.0174. Zeileis, A., Hothorn, T., 2014. partykit: a toolkit for recursive partytioning. J. Mach. Learn. Res. 16, 3905–3909. Zhu, Y., Weindorf, D.C., Zhang, W., 2011. Characterizing soils using a portable X-ray fluorescence spectrometer: 1. Soil texture. Geoderma 167–168, 167–177. https:// doi.org/10.1016/j.geoderma.2011.08.010.

aboveground forest biomass outperforms a pixel-based approach in a heterogeneous and mountain tropical environment. Int. J. Appl. Earth Obs. Geoinf. 78, 175–188. https://doi.org/10.1016/j.jag.2019.02.004. Singh, N., Singh, J., Singh, K., 2018. Small at size, big at impact: microorganisms for sustainable development. In: Microbial Bioprospecting for Sustainable Development. Springer Singapore, Singapore, pp. 3–28. https://doi.org/10.1007/978-981-130053-0_1. Skirycz, A., Castilho, A., Chaparro, C., Carvalho, N., George, T., Siqueira, J.O., 2014. Canga biodiversity, a matter of mining. Front. Plant Sci 5, 1–9. https://doi.org/ 10.3389/fpls.2014.00653. Soil Survey Staff, 2014. Keys to soil taxonomy, 12th ed. United States Department of Agriculture Natural Resources Conservation Service. Spohn, M., Carminati, A., Kuzyakov, Y., 2013. Soil zymography - A novel in situ method for mapping distribution of enzyme activity in soil. Soil Biol. Biochem. 58, 275–280. https://doi.org/10.1016/j.soilbio.2012.12.004. Stone, M.M., Kan, J., Plante, A.F., 2015. Parent material and vegetation influence bacterial community structure and nitrogen functional genes along deep tropical soil profiles at the Luquillo Critical Zone Observatory. Soil Biol. Biochem. 80, 273–282. https://doi.org/10.1016/j.soilbio.2014.10.019. Strobl, C., Boulesteix, A.L., Zeileis, A., Hothorn, T., 2007. Bias in random forest variable importance measures: Illustrations, sources and a solution. BMC Bioinformatics 8. https://doi.org/10.1186/1471-2105-8-25. ˇ ˇ Stursov´ a, M., B´ arta, J., Santrůˇ ckov´ a, H., Baldrian, P., 2016. Small-scale spatial heterogeneity of ecosystem properties, microbial community composition and microbial activities in a temperate mountain forest soil. FEMS Microbiol. Ecol. 92, fiw185. https://doi.org/10.1093/femsec/fiw185. Tabatabai, M.A., Bremner, J.M., 1970. Arylsulfatase activity of soils1. Soil Sci. Soc. Am. J. 34, 225. https://doi.org/10.2136/sssaj1970.03615995003400020016x. Tan, X., Xie, B., Wang, J., He, W., Wang, X., Wei, G., 2014. County-scale spatial distribution of soil enzyme activities and enzyme activity indices in agricultural land: implications for soil quality assessment. Sci. World J. 2014, 1–11. https://doi. org/10.1155/2014/535768. Tavares, T.R., Molin, J.P., Nunes, L.C., Alves, E.E.N., Melquiades, F.L., de Carvalho, H.W. P., Mouazen, A.M., 2020. Effect of x-ray tube configuration on measurement of key soil fertility attributes with XRF. Remote Sens. 12 https://doi.org/10.3390/ rs12060963. Teixeira, A.F. dos S., Kemmelmeier, K., Marascalchi, M.N., Stürmer, S.L., Carneiro, M.A. C., Moreira, F.M. de S., 2017. Arbuscular mycorrhizal fungal communities in an iron mining area and its surroundings: Inoculum potential, density, and diversity of spores related to soil properties. Ciˆencia e Agrotecnologia 41, 511–525. https://doi. org/10.1590/1413-70542017415014617. Teixeira, A.F. dos S., Silva, J.S., Vilela, L.A.F., Costa, P.F., Costa, E.M.D., Guimar˜ aes, A. A., Santos, J.V.D., Silva, S.H.G., Carneiro, M.A.C., Moreira, F.M.S., 2019.

12