Brasil
Revista Brasileira de Ciência do Solo
Tabla de contenido Anterior Actual Siguiente
Tabla de contenido
Division - Soil in Space and Time | Commission - Pedometrics • Rev. Bras. Ciênc. Solo 50 • 2026 • https://doi.org/10.36783/18069657rbcs20250171 linkcopy

Open-access Mapping pedogenetic processes in an Amazonian watershed by fuzzy logic applied to multiscale generalized geomorphometrics

SCIMAGO INSTITUTIONS RANKINGS

ABSTRACT

Soil-forming factors determine the conditions of soil-forming processes, and their temporal overlap results in soil properties within the pedon. These pedons can be described in terms of soil classes. However, the relationships are complex; on the one hand, different processes can result in similar properties, and on the other hand, the same process can result in different soil classes depending on their prevalence and the extent of changes in soil properties within the pedon. The multiscale influence of topography on soil distribution has two general aspects: the overlay of pedological processes that occurred at different times, and driving forces in the present time, determined by the sum of forces that are better correlated with one, several, or many geomorphologic scales. The hypothesis tested in this study is whether the multiscale analysis of topography contributes to the definition of preferential environments for the occurrence of pedogenetic processes such as gleization, elutriation, and clay translocation, rather than the spatiotemporally widespread distribution of ferralitization. The objective of this study was to map the soil classes in two watersheds at a scale of 1:100,000. The basins of this study are homogeneous in terms of geological basement, with a predominance of sandstones, comprising the Alter do Chão formation. To achieve this objective, fuzzy set models were applied using a multiscale geomorphometric database, generalized from the SRTM digital elevation model, to predict the distribution of soil classes. Fuzzy membership rules were defined for each pedological process, based on pedological knowledge from soil surveys along transects, to define the membership curves and scales of the covariates. The maps were defuzzified, defining the soil class of each pixel. The maps were evaluated using confusion matrices, based on validation points defined by a conditioned Latin hypercube. With generalized covariates, the overall accuracy was 84 %, compared with 74 % for the mapping that considered only the covariates in the original scale. Considering the multiscale covariates, the following user and producer accuracies were obtained: GXbd (88 %, 78 %); undifferentiated group PAd and LAdarg (60 %, 60 %); undifferentiated group LAd and LAdpsa (88 %, 92 %). The results demonstrate that knowledge-based models that observe the multiscale influence of topography on soil formation processes can improve the prediction of soil classes in intermediate-level soil surveys.

Keywords
knowledge-based models; soil-landscape relationship; geomorphometric upscaling; soil forming process; soil survey

INTRODUCTION

The soils of the Amazon Basin are predominantly formed by strong chemical weathering processes typical of a warm, moist climate acting on a sedimentary geological basement over a long period of stability in environmental conditions (Schaefer et al., 2017). As a result, the long-term pedological processes are the ones that are most spatially distributed, with emphasis on ferralitization. However, other processes can become prevalent or significant and drive important changes in a given portion of soil under specific environmental conditions, such as gleization, elutriation, and clay translocation (Kämpf and Curi, 2012). Therefore, to understand soil distribution patterns, it is necessary to identify and delineate the preferred environments for the occurrence of the different pedological processes. In this sense, quantification methods have made considerable progress in predicting soil processes and related soil horizon variations across landscapes (Hartemink et al., 2020).

Topography is the main variable to define environments where different pedological processes can occur (Florinsky, 2016). The soil-landscape relationship is based on the concepts of the catena and hillslope models (Milne, 1935; Huggett, 1975) anon the essential soil science concepts established by Simonson (1959): soil-forming factors, soil-forming processes, and resulting soil properties. When considering geomorphic development and the multiscale topography influences, two general aspects can be highlighted: overlay of pedological processes that occurred at different times (Wysocki and Schoeneberger, 2011); and driving forces, determined by the sum of forces better correlated with one, several, or many geomorphologic scales (Florinsky, 2016). The classical mental models of pedologists generally include, even if tacitly, observations at different levels of relief generalization in the construction of soil-landscape rules, at various stages of knowledge development (Bui, 2004), including features resulting from geomorphic evolution and the flexible selection of scales to determine terrain parameters.

In digital soil mapping applications, process-based modeling approaches have recently received attention, despite greater inherent complexity compared to models based on property prediction (Ma et al., 2019; Lozbenev et al., 2021). The challenges include site-specific dynamics, high uncertainty, the impossibility of direct observation under uncontrolled field conditions, the need to infer pedological processes from soil properties, and ambiguity in the relationship between properties and processes. Despite this, process-based approaches can help gain new insights into soil processes by observing mapping sites, thereby promoting feedback between digital soil mapping and pedology. The fuzzy set approach to knowledge-based modeling (McBratney and Odeh, 1997; Zhu, 1997) for process-based soil mapping is especially useful for handling uncertainties, qualitative or semi-quantitative information, and multiple membership rules that can translate complex environmental conditions.

To produce digital soil maps of the Amazon region at a high level of resolution, the pedological processes in each pedoenviroment need to be modeled with a level of detail useful for a low-cost soil survey approach, resulting in soil class maps at the subgroup categorical level. Furthermore, this pathway of digital soil mapping contributes to a better understanding of soil pedogenesis in the landscape, which is a significant challenge for soil science in a pedometric context (Arrouays et al., 2020; Wadoux et al., 2021).

We tested the hypothesis that multiscale analysis of topography contributes to defining preferential environments for the occurrence of pedogenetic processes such as gleization, elutriation, and clay translocation, rather than the spatiotemporally widespread distribution of ferralitization. This study aimed to incorporate multiscale geomorphometry into a knowledge-based fuzzy modeling framework using a multiscale-generalized covariable database (Araújo et al., 2021) and an upscaling operation, detailed in the methodology section. This methodology brings to the pedometric perspective the understanding that the soil-landscape relationship occurs through complex, multiscale interactions, allowing the formalization of scale selection based on expert knowledge.

MATERIALS AND METHODS

The procedures described in this section were performed using the open-source software QGIS 3.16; SAGA GIS 2.3; GRASS GIS 7.8; and R Programming 3.5 (Conrad et al., 2015; GRASS, 2019; QGIS, 2019; R Core Team, 2019).

Study area

The study was conducted in the Iripixi Lake (ILW) and Caipuru Lake (CLW) watersheds, with an area of 27,137 and 28,315 ha respectively, located in the Trombetas River Basin, in Oriximiná-Pará, Eastern Amazon (Figure 1).

The Alter do Chão formation is a Cretaceous sedimentary deposit (CPRM, 2008). Locally, in western Pará, the formation process of Alter do Chão took place in a depositional environment of a sinuous fluvial system, resulting in the record of a succession of sandstones, conglomerates and mudstones (Mendes et al., 2012). It was formed approximately 135 million years ago and has evolved in accord with tectonic motion ever since (Somoza and Ghidella, 2012). Later, in the context of the beginning of the Andean uplift, another depositional formation buried the Alter do Chão formation in the Paleocene, approximately 55 million years ago. The movement resulting from the epeirogenesis of the South American continent and variations in the global sea level caused the dissection and almost complete destruction of this Paleogene Detrital-Lateritic Cover in the Oligocene, 30 million years ago. In the final phase of geological evolution, from the Miocene, approximately 10 million years ago, continental movements related to the advanced stages of the Andean uplift and variations in the global ocean level due to glaciations reduced the base level and promoted the dissection of the Alter do Chão formation.

From the Quaternary onward, 1.4 million years ago, the region experienced period of geological stability (Somoza and Ghidella, 2012). In this sense, we can infer that geomorphological and pedological contemporary evolutionary dynamics have been developing since then. However, during this period, there were at least three major reductions in global ocean levels that may have initiated erosive processes, soil rejuvenation, and reorganization of drainage patterns (Bridgland, 2022). The geomorphic units of these watersheds are classified as a homogeneous dissection with coarse drainage density and weak incision depth (IBGE, 2008, 2009).

thumbnail of Figure 1
Figure 1
Study area location in the eastern Amazon.

Topography covariates

At the proposed mapping scale, topographic factors are the main drivers of soil variation, and in this study, the Shuttle Radar Topography Mission Digital Elevation Model 30 m (SRTM DEM 30 m) (Farr et al., 2007) was used to represent these factors. The corrections made in the SRTM DEM were to fill sinks and reduce the effects of deforestation using the estimated canopy addition method (Brochado, 2015). The topography information was upscaled and organized into generalized multiscale geomorphometric variable groups, as detailed in the next section.

Multiscale geomorphometric generalization

Multiscale geomorphometric generalization (MGG) is an upscaling operation based on cartographic concepts of map generalization (Li and Openshaw, 1993). This approach can be applied to any geomorphic variable, including elevation models, landform units, and primary and secondary derivatives. This operation results in variables at different scales, grouped according to the criteria required for the analysis. In this sense, the formalization of the desirable scales of analysis and modeling occurs both in their definition and in the group arrangements.

For the MGG operation, vector and raster representations require different upscaling approaches, as each has specific scale-transformation issues due to its mathematical structure. Furthermore, it is necessary to have a unique reference for the scales to ensure compatible representation of geomorphic features in both types of variables, thus allowing joint interpretation. In this study, the concept of the minimum mappable area for soil surveys (IBGE, 2015) was used to define pixel sizes relative to the cartographic scale. The detailed descriptions for each of the four scales used are in table 1. The area equivalence between raster and vector is calculated using a 5 × 5 pixel grid, a conservative parameter for determining geomorphic features (Florinsky, 2017). These geomorphic covariables were obtained by upscaling methods, as illustrated in figure 2, in which local averages of the DEM and subsequent derivatives were used in the covariable calculation, and these were adjusted using the inner search radius for geomorphons. Such methods correspond to cartographic generalization applied to general geomorphometry and specific geomorphometry, respectively (Zinck, 2016).

Thumbnail
Table 1
Correspondence between scale and pixel size for Multiscale Geomorphometric Generalization (MGG), using the concept of minimum mappable area
thumbnail of Figure 2
Figure 2
Methodology flowchart of MGG for the topography covariables.

Soil survey and classification

The soil survey was conducted to acquire empirical data for model construction. Soil profile sampling was carried out in three transects, covering the lower, middle and upper portions of the hydrographic basins (Figure 1). An independent sampling dataset was used for validation, and additional observation points were analyzed in a random-stratified arrangement, distributed according to the terrain covariates in the original scale using a cost-constrained, conditionally constrained Latin hypercube method (Roudier et al., 2012; Silva et al., 2014). The watersheds were sampled at 18 and 20 points, respectively, a sufficient density for high-intensity reconnaissance soil surveys and compatible with 1:100,000-ale soil maps (IBGE, 2015).

The morphological description of the soil profile was performed to a depth of 2 m (Santos et al., 2015) at selected positions along the transects and at the extra validation points. Soil samples were collected from the horizons and their subdivisions for complementary physical and chemical analyses, necessary for discriminating soil classes (Teixeira et al., 2017).

The Brazilian Soil Classification System (SiBCS) is a hierarchical system that covers all soils in the national territory (Santos et al., 2018). The categorical level structure determines the classification into orders, suborders, large groups, subgroups, families, and series. The prevailing pedological process determines the order, and minor environmental and pedological characteristics define lower categorical levels. In this sense, the soil classification of a pedon could vary despite the occurrence of the same pedological processes. The main criteria used for pedon classification relate to the level of evolution of soil characteristics and properties, produced by one or another soil process. In this study, soil profiles were classified to the 4th categorical level within the subgroups.

Fuzzy modeling and production-evaluation of soil maps

Fuzzy logic modeling is an approach capable of dealing with uncertainties, as the relationship between elements and fuzzy sets is expressed by degrees of membership (Nicoletti and Camargo, 2004). Fuzzy logic modeling was used in several studies related to knowledge-driven soil mapping (McBratney and Odeh, 1997; Menezes et al., 2013, 2014; Silva et al., 2014; Ma et al., 2019).

In this study, the fuzzy modeling process and soil map production was carried out as shown in figure 3. The definition of intervals, the shape of the pertinence curves, and the scales of the considered covariates were established based on knowledge acquired during the soil survey along transects. The pertinence curves represent favorable environmental conditions for the occurrence of the target pedogenetic processes of the study. The pedogenetic processes considered were gleization, elutriation, clay translocation, and ferralitization. Defuzzification and soil class map production were performed by selecting the highest pertinence value and using the classification provided in the SiBCS. The soil class maps were evaluated against validation points using a confusion matrix, with kappa index, global accuracy, and producer and user accuracies reported.

The main hypothesis of this study was that, in a dissection modeled on a sedimentary geological formation, topography, expressed through generalized multiscale covariates of slope, curvature, and moisture content (landscape factor), controls the redistribution of moisture via water flow along hillslopes (landscape process). This, in turn, controls the processes of clay translocation, elutriation, and gleization (soil-forming processes), resulting in the differentiation of pedons (soil properties) and the distribution of soil classes.

thumbnail of Figure 3
Figure 3
Methodology flowchart of Fuzzy Modeling with Covariable Scale Selection.

RESULTS

The survey of soils in the transects allowed for the construction of a mental model of the soil-landscape relationship, explaining the link between topographic, landscape processes, landscape properties, pedogenetic processes and resulting soil profiles, as shown in table 2.

Thumbnail
Table 2
Correspondence between Landscape and Soil according to the nested triad Factor-Process-Property

Ferralitization is the most widespread pedogenetic process in the landscape, due to the long period of weathering that has occurred in this area of geological stability. Considering the geomorphological evolution of a dissection model in this geological context, it is possible to conclude that the weathering mantle formed prior to erosive dissection, resulting in a widespread distribution of Oxisols that were eventually reorganized into differentiated profiles. It should be noted that the predominance of sandstones produced soils with high contents of the sand fraction, close to the threshold between Latossolo Amarelo Distrófico típico/Typic Hapludox and Latossolo Distrófico psamítico/ Arenic Hapludox. In this sense, both profiles were mapped as an undifferentiated group (LAd + LAdpsa).

Elutriation is the most relevant process for the reorganization of profiles in the mapped basins, occurring primarily under high runoff conditions. This selective erosion process, which carries mainly finer particles under conditions of forest vegetation cover, occurs in the high, convex portions of the slope due to the greater potential energy of water in these positions. The elutriation process gradually removes clay from the upper layers, reducing it in the profile, and eventually produces a textural gradient. Clay translocation is a relevant but localized process for the reorganization of profiles in the mapped basins, occurring primarily under high percolation conditions. This process of vertical movement of clays in the profile has a low distribution because the geological basement has a reduced clay content. Therefore, the differentiation that results in textural gradients arising from clay accumulation in the lower layers is restricted to areas with intense water movement. In these places, it was possible to notice the presence of weak clay films, which is the main indication of the predominance of clay translocation in such profiles. For both processes, the current stage of such evolution may or may not meet the requirements for classification as an Argissolo. In this sense, the Argissolo Amarelo Distrófico típico and Latossolo Amarelo Distrófico argissólico are mapped as an undifferentiated group (PAd + LAdarg).

The process of gleization occurs relatively quickly, unlike the three processes described above. In this sense, the reorganization of the profiles resulting from water saturation in a redoximorphic environment is strongly correlated with the currently saturated zones. Therefore, gleization is predominant in the bottoms of flat valleys and in low areas under the influence of the water table. For this process, the soils are mapped as Gleissolo háplico Tb Distrófico típico (GXbd).

Soil fuzzy modeling was done for each soil process, resulting in one or more pertinence curves, and all of these curves are sigmoidal, with the adjustment points presented in table 3. Additionally, covariates with different generalization scales were selected for application in the same curves, as shown in table 4, and the generalization geomorphometric effects were exemplified in a 3D view (Figure 4). Therefore, this methodological structure allowed for the formalization of decisions based on the pedologist knowledge regarding the selection of scales and relevance rules, and for the compartmentalization of the models according to relief forms.

Thumbnail
Table 3
Landform segmentation and Fuzzy pertinence rules for gleization, elutriation, clay translocation and ferralitization process
Thumbnail
Table 4
Generalized geomorphometric covariables selection for pertinence rules
thumbnail of Figure 4
Figure 4
Three-dimensional (3D) view of geomorphic covariates at original and generalized scales.

Soil wetness conditions (dry, wet, or waterlogged) were principal aspects in the fuzzy models; therefore, the topographic wetness index (TWI) was selected as the main geomorphometric covariable for distinguishing the occurrence conditions of soil processes. Other covariates were selected as important determinants of the dynamics of humidity in the landscape, including the potential energy compared to the base level; tendencies to gain or lose speed; and converging or diverging flows. In this sense, the covariates’ slope, plan curvature, profile curvature, and relative slope position were used.

The gleization process occurs in the valleys. In these segments, the process occurs at high levels of soil moisture (TWI >9), that is, under conditions of humidity accumulation and in positions in the landscape that make drainage difficult. The adoption of generalized covariates for valley segmentation and moisture content at 1:75,000 and 1:50,000 scales, respectively, produced maps in which the probable occurrence of gleization in small valleys and headwaters was reduced. On the other hand, it favored the appearance of significant gleization spots in the broad valleys of the main channels, as seen in figure 5a.

thumbnail of Figure 5
Figure 5
Three-dimensional (3D) view of fuzzy membership maps for soil processes: (a) gleying, (b) eluviation, (c) clay translocation, and (d) ferrallitization.

The elutriation process occurs on summits, ridges, shoulders, spurs and slopes. In these segments, two pertinence rules were constructed to encompass preference situations. In rule 2, the most likely areas are flat and convex areas, with low and intermediate wetting rates, and at relative slope positions between the middle and upper thirds. In rule 3, the model indicates that the highest probability occurs in conditions of high slope, especially between 8 and 15 %, and that it is smooth, progressively increasing in areas with higher soil wetness. That is, conditions with water content and energy availability, promoting the occurrence of surface runoff. The adoption of generalized covariates for geomorphons segmentation, slope, TWI, and curvatures, at 1:75,000 and 1:100,000 scales respectively, produced maps where the probable occurrence of elutriation presented higher indices in upper convex portions, considering the large features in the basin, with emphasis on the interfluves of basins and subbasins (Figure 5b).

The clay translocation process occurs in valleys, slopes, footslopes and hollows. In these segments, the process is more likely to occur when associated with high levels of humidity, without saturation; in portions of flow convergence, with concave plan curvatures; and in portions of flow deceleration, with concave profile curvatures. That is, conditions with significant water flow, converging laterally and losing kinetic energy, primarily promote vertical flow. The adoption of generalized covariates for geomorphic segmentation, TWI, and curvatures at 1:75,000 and 1:100,000 scales, respectively, produced maps in which the probable occurrence of clay translocation in footslopes and low concave positions appears as rounded spots with higher values in the central areas. When using only geomorphometric covariables at the original scale, clay translocation does not occur in these zones, but only with a noise-like pixel distribution (Figure 5c).

The ferralitization process is widespread across the landscape and is more likely to occur associated with the lowest slope inclinations, bellow 10 %; and dry or intermediate wetness conditions, especially with TWI <7. That is, conditions that promote the deepening of the weathering mantle. The adoption of generalized covariates for slope and TWI at a 1:100,000 scale produced maps showing the probable occurrence of ferralitization, considering the large features in the basin, with lowest values in dissection areas of interfluves and sub-basins, as seen in figure 5d.

The soil class maps were produced for original scale and multiscale generalized geomorphometrics (Figure 6), and resulted in one pure soil unit and two undifferentiated groups: GXbd, with a gley horizon (n = 9); PAd + LAdarg, (n = 5); LAd + LAdpsa, with a Bw horizon (n = 25). The adoption of generalized covariates in the fuzzy model rules produces more relevant patches with PAd+LAdarg and GXbd. In this sense, the prevalence of soil patches was observed at large scale on ridges (convex high geoformations) and footslopes (hollows). On the other hand, the GXbd soil patches occurred in flat, larger valleys.

thumbnail of Figure 6
Figure 6
Soil class maps produced by fuzzy modeling with (a) original scale geomorphometric and (b) multiscale generalized geomorphometric covariates.

The evaluation of soil class maps reveals increases in general accuracy and user and producer accuracies for the minority of soil classes with the use of generalized geomorphometric covariates (Table 5). The overall accuracy was higher in models that used generalized covariates, but this result should be interpreted with caution given the class imbalance. It is important to highlight the increase in map reliability in the less prevalent classes, as reflected in the highest values for the user’s accuracy of GXbd and PAd+LAdarg. Despite some restrictions on the use of the Kappa Index (Pontius and Millones, 2011), it is worth noting the significant increase in this measure, from fair agreement to substantial agreement. This pronounced effect is justified, in part, by the prevalence of Latossolos and Latossolos psamíticos, making these classes more prone to "random hits". When this characteristic is considered, through the Kappa index, the benefits obtained in the minority classes are highlighted - an important fact for a joint interpretation with the accuracy indices.

Thumbnail
Table 5
Accuracy evaluation for soil classification at the 4th categorical level of SiBCS

DISCUSSION

Using the high-intensity resolution 1:100,000 mapping scale proposed in this study, topography is the main source of variation. In this sense, digital soil mapping at this scale can contribute to understanding soil-landscape relationships, specifically regarding soil property patterns resulting from hillslope hydrology.

Textural gradients

Textural differentiation, resulting in textural gradients in the soil profile, is a central aspect for mapping soils in watersheds located on the Alter-do-Chão geological formation. In this pedoenvironment, this characteristic may originate from two distinct pedological processes (Kämpf and Curi, 2012), namely: clay translocation, or lessivage, and elutriation. Clay translocation refers to the suspended movement of particles of the clay fraction, mainly fine clays <0.002 mm, oxides and organic compounds, inside the soils. Clays remain in suspension and are transported through the macropores, where the flows are sufficiently rapid. Deposition occurs where dispersion and transport are less effective. Mobilization occurs by crumbling, aggregate separation, wetting-drying processes, or dispersion. Elutriation is a process of removing fine particles by selective erosion, which is common in tropical forests. Elutriation is caused by dripping, dispersion, and surface flow of water, which removes fine material into the drainage system. It may be associated with the removal of fine silty material and its absence in the B horizon.

The formation of soil textural gradients in the mapped watersheds does not have its predominant origin in the clay translocation process in most of the slope positions, except in the lower concave ones. This pattern is supported by the tendency of high percolation in this segment of the landscape (Wysocki and Schoeneberger, 2011), indicating the occurrence of clay translocation in these environments, even from a geological base poor in clays. Therefore, we can assume that the elutriation process is a widespread mechanism in different positions in the landscape, which would explain the significant occurrence of profiles with textural gradients in areas with higher runoff (Möller and Volk, 2015).

Glei horizon

Soil that is saturated, or with very high levels of moisture, creates an environment where the influx of O2 from the atmosphere is very difficult, due to the absence of free pore space. The absence of O2 favors the reduction of complex organic compounds and the growth of anaerobic microorganisms that use metals as electron acceptors. In this way, Fe ions are reduced and released from oxides, then migrate into the soil solution. Therefore, Fe loss zones turn grayish, eventually forming Glei horizons. These ions in solution migrate by diffusion to environments of lower redox potential, where they are oxidized and precipitate (Kämpf and Curi, 2012).

The Fe accumulation zones, marked by concretions or mottles, require thousands of years for their formation. However, the time required to form a well-developed glei horizon is only a few years or decades(Kämpf and Curi, 2012). Considering this geomorphological time scale, this time lapse can be understood as "present time". In this sense, it is expected that the patterns of occurrence of soils with a gley horizon are adjusted to the current disposition of the valleys, drainage and water table.

Scale in soil-landscape relationships

The investigation of soil-landscape relationships occurs indirectly in the mapping approach proposed in this study. The direct evidence is the soil profile, but the phenomenon studied is the spatial distribution of pedogenetic processes. Therefore, it is necessary to identify the relationship between such elements in a theoretical scheme that encompasses not only the factor-property relationship. In this study, we analyze the results from the perspective of Simonson's theoretical model (Simonson, 1959), proposed as a basis for soil-landscape analysis through the nested triad of factor-process-property (Lozbenev et al., 2021).

The parameter variation tested in the model is the scale of the landscape factor. The generalization of landscape units, represented by geomorphons, occurred at a scale of 1:75,000 for all processes. This is because this level of detail presented the most consistent features with the expert interpretation, as observed in fieldwork and in geographic information systems. For rougher, more heterogeneous reliefs, a 1:50,000-scale subdivision might better reflect the local dynamics of hillslopes (Möller and Volk, 2015; Pachepsky and Hill, 2017).

The ideal prediction scale for soil properties, which ultimately can be grouped into soil classes, varies depending on the property in question. Therefore, this study built each model from covariates at different scales, more suitable for analyzing hillslopes in relation to the respective landscape properties. Almost all the variables were selected from the generalized scales 1:75,000 and 1:100,000. This can be explained by the prevalence of pedological phenomena at these scales and by the nature of the evolution of the local geomorphological landscape. This landscape features gentle undulations and results from a dissection process in a geologically stable environment. These results agree with a study conducted in the plains and smooth valleys of western Romania (Dornik et al., 2022). For slope gradient, general curvature, profile curvature, and convergence index, coarse and medium scales were generally identified as optimal, with optimal scales over 700 m. The short-range variation was only for slope gradient for clay and sand content, and by convergence index for porosity within the restrictive horizon and sand content. Still, the results of this study proved that more accurate maps could be obtained when predictors are optimally scaled, as compared to using unscaled or all multiscale predictors.

A study that mapped soils in Andalusia, Spain, concluded that the relevant scales do not differ only in the horizontal domain, but also in the vertical domain across the soil profile (Rentschler et al., 2022). They hypothesize that there are depth gradients in spatial and structural dependencies. The results showed that using topsoil features at small to intermediate scales does not increase model accuracy. In contrast, subsoil models benefit from all scales, from small to large. For soil surface mapping conducted at the Krui River, Australia, Random Forest was used with DEM at 5, 25, 30, and 90 m resolution to predict SOC (Gibson et al., 2021). They observed, using large-scale maps, that SOC predictions are not affected by topography because overriding climatic factors dominate. The topography and DEM scale were found to have greater influence on SOC at local scales, and modeling requires high-resolution DEMs and sampling data to capture hillslope-scale variability.

It is important to note that elutriation is a surface process, whereas clay translocation and gleization are subsurface processes. In this sense, in the studied landscape, the coarser scales of terrain analysis showed greater correlation with the observed soil properties. We can infer that the use of covariates at multiple scales, generalized from a digital 30 m elevation model, produced better predictive effects than the original scales or monoscale generalization. This result agrees with the conclusions from soil mapping using the "mixed scaling" method and deep learning, carried out in Piracicaba, Brazil; Meuse, Netherlands; and Rhine, Germany (Behrens et al., 2018). These authors observed that the proposed mixed-scaling method yields the most accurate results when using multi-scale covariates and could allow more intuitive interpretation of the datasets.

In a later study on the same mapping areas, the concept of information horizon was proposed, and attention is drawn to the possibility of meaningless covariates being identified as important for the establishment of soil patterns (Behrens and Viscarra Rossel, 2020). “The information horizon is the range (or scale) above which interpretations are no longer possible, because beyond the horizon, relationships lack structure”. They suggest several important recommendations for models to be interpretable: only domain-relevant and structurally related predictors must be used, the scales of the predictors must be aligned with the range of soil spatial dependence, and structural and spatial dependence should be analyzed separately. In this sense, the fuzzy modeling pertinence rules have a pedological meaning, maintaining a structural relationship with the target variables of the soil mapping carried out in the hydrographic basins of the Alter-do-Chão formation. On the other hand, further activities should focus on the analysis of the spatial dependence of soil properties.

CONCLUSIONS

The analysis of scale in the soil-landscape relationship brings a series of complicating factors that deserve attention for methodological development, namely: (a) the establishment of unambiguous relationships between landscape properties, pedological processes and soil properties; (b) optimization of topography generalization methods, with emphasis on the preservation of relevant features; (c) the application of interpretable modeling methods capable of dealing with quantitative, semi-quantitative and qualitative information, whether driven by data or knowledge.

Mapping of pedogenetic processes and soil classes was carried out with a knowledge-oriented approach, using soil survey observations to establish fuzzy pertinence curves and to select the scale of the geomorphometric covariates. The use of generalized covariates, at different scales according to the pedogenetic process in question, resulted in maps of soil classes with greater accuracy: the general accuracy increased from 74 to 84 %, user accuracy increased from 50 % to 88 % and from 20 to 60 % for GXbd and PAd+LAd arg, respectively, and the Kappa index went from 0.4 to 0.69, signifying substantial agreement. Therefore, generalized geomorphometry covariable selection improves the accuracy of soil class maps.

Nested triad factor-process-property proved to be a useful theoretical model for the soil-landscape relationship, even in the complex observation and association of pedogenic processes with related soil properties in field conditions. Regarding the pattern of occurrence of pedogenetic processes associated with landscape properties, it is worth noting that (a) gleization has a well-defined position at the bottom of the valleys, (b) elutriation is the prevalent process resulting in texture gradients, widespread on slopes, and (c) clay translocation has a minor effect, restricted to some lower concave positions.

How to cite:

Araújo CF, Souza NM, Oliveira Junior RC, Beldini TP. Mapping pedogenetic processes in an Amazonian watershed by fuzzy logic applied to multiscale generalized geomorphometrics. Rev Bras Cienc Solo. 2026;50:e0250171. https://doi.org/10.36783/18069657rbcs20250171

DATA AVAILABILITY

The data will be provided upon request.

REFERENCES

  • Araújo CF, Junior RCO, Beldini TP. Multiscalar geomorphometric generalization for soil-landscape modeling by random forest: A case study in the Eastern Amazon. J Geogr Inf Syst. 2021;13:434-51. https://doi.org/10.4236/jgis.2021.134024
    » https://doi.org/10.4236/jgis.2021.134024
  • Arrouays D, McBratney A, Bouma J, Libohova Z, Richer-de-Forges AC, Morgan CLS, Roudier P, Poggio L, Mulder VL. Impressions of digital soil maps: The good, the not so good, and making them ever better. Geoderma R. 2020;20:e00255. https://doi.org/10.1016/j.geodrs.2020.e00255
    » https://doi.org/10.1016/j.geodrs.2020.e00255
  • Behrens T, Schmidt K, MacMillan RA, Viscarra Rossel RA. Multi-scale digital soil mapping with deep learning. Sci Rep. 2018;8:15244. https://doi.org/10.1038/s41598-018-33516-6
    » https://doi.org/10.1038/s41598-018-33516-6
  • Behrens T, Viscarra Rossel RA. On the interpretability of predictors in spatial data science: The information horizon. Sci Rep. 2020;10:16737. https://doi.org/10.1038/s41598-020-73773-y
    » https://doi.org/10.1038/s41598-020-73773-y
  • Bridgland D. The role of geomorphology in the Quaternary. In: Burt TP, Goudie AS, Viles HA, editors. The history of the study of landforms or the development of geomorphology. Volume 5: Geomorphology in the Second Half of the Twentieth Century. London: Geological Society; 2022. p. 109-20. https://doi.org/10.1144/M58-2021-14
    » https://doi.org/10.1144/M58-2021-14
  • Brochado GT. Atenuação do efeito do desflorestamento em dados SRTM por meio de diferentes técnicas de interpolação [dissertation]. São José dos Campos: Instituto Nacional de Pesquisas Espaciais; 2015.
  • Bui EN. Soil survey as a knowledge system. Geoderma. 2004;120:17-26. https://doi.org/10.1016/j.geoderma.2003.07.006
    » https://doi.org/10.1016/j.geoderma.2003.07.006
  • Conrad O, Bechtel B, Bock M, Dietrich H, Fischer E, Gerlitz L, Wehberg J, Wichmann V, Böhner J. System for automated geoscientific analyses (SAGA) v. 2.1.4. Geosci Model Dev. 2015;8:1991-2007. https://doi.org/10.5194/gmd-8-1991-2015
    » https://doi.org/10.5194/gmd-8-1991-2015
  • Companhia de Pesquisa de Recursos Minerais - CPRM. Mapa geológico do estado do Pará. Escala 1:1.000.000. Belém: CPRM; 2008.
  • Dornik A, Cheţan MA, Drăguţ L, Dicu DD, Iliuţă A. Optimal scaling of predictors for digital mapping of soil properties. Geoderma. 2022;405:115453. https://doi.org/10.1016/j.geoderma.2021.115453
    » https://doi.org/10.1016/j.geoderma.2021.115453
  • Farr TG, Rosen PA, Caro E, Crippen R, Duren R, Hensley S, Kobrick M, Paller M, Rodriguez E, Roth L, Seal D, Shaffer S, Shimada J, Umland J, Werner M, Oskin M, Burbank D, Alsdorf D. The shuttle radar topography mission. Rev Geophys. 2007;45:RG2004. https://doi.org/10.1029/2005RG000183
    » https://doi.org/10.1029/2005RG000183
  • Florinsky IV. An illustrated introduction to general geomorphometry. Prog Phys Geogr. 2017;41:723-52. https://doi.org/10.1177/0309133317733667
    » https://doi.org/10.1177/0309133317733667
  • Florinsky IV. Influence of topography on soil properties. In: Digital terrain analysis in soil science and geology. 2nd. ed. London: Elsevier; 2016. p. 265-344. https://doi.org/10.1016/C2010-0-65718-X
    » https://doi.org/10.1016/C2010-0-65718-X
  • Gibson AJ, Hancock GR, Bretreger D, Cox T, Hughes J, Kunkel V. Assessing digital elevation model resolution for soil organic carbon prediction. Geoderma. 2021;398:115106. https://doi.org/10.1016/j.geoderma.2021.115106
    » https://doi.org/10.1016/j.geoderma.2021.115106
  • Geographic Resources Analysis Support System - GRASS DT. GRASS Software - Version 7.8. [internet]; 2019. Available from: https://grass.osgeo.org/news/2019_09_15_grass_gis_7_8_0_released/
    » https://grass.osgeo.org/news/2019_09_15_grass_gis_7_8_0_released/
  • Hartemink AE, Zhang Y, Bockheim JG, Curi N, Silva SHG, Grauer-Gray J, Lowe DJ, Krasilnikov P. Soil horizon variation: A review. Adv Agron. 2020;160:125-85. https://doi.org/10.1016/bs.agron.2019.10.003
    » https://doi.org/10.1016/bs.agron.2019.10.003
  • Huggett RJ. Soil landscape systems: A model of soil Genesis. Geoderma. 1975;13:1-22. https://doi.org/10.1016/0016-7061(75)90035-X
    » https://doi.org/10.1016/0016-7061(75)90035-X
  • Instituto Brasileiro de Geografia e Estatistica - IBGE. Manual técnico de pedologia. 3. ed. Rio de Janeiro: IBGE; 2015.
  • Instituto Brasileiro de Geografia e Estatistica - IBGE. Manual técnico de geomorfologia. 2. ed. Rio de Janeiro: IBGE; 2009.
  • Instituto Brasileiro de Geografia e Estatistica - IBGE. Estado do Pará: Mapa de geomorfologia. Rio de Janeiro: IBGE; 2008.
  • Jasiewicz J, Stepinski TF. Geomorphons - a pattern recognition approach to classification and mapping of landforms. Geomorphology. 2013;182:147-56. https://doi.org/10.1016/j.geomorph.2012.11.005
    » https://doi.org/10.1016/j.geomorph.2012.11.005
  • Kämpf N, Curi N. Formação e evolução do solo (pedogênese). In: Ker JC, Curi N, Schaefer CEGR, Vidal-Torrado P, editors. Pedologia: Fundamentos. Viçosa, MG: Sociedade Brasileira de Ciência do Solo; 2012. p. 207-302.
  • Li Z, Openshaw S. A natural principle for the objective generalization of digital maps. Cartogr Geogr Inform Syst. 1993;20:19-29. https://doi.org/10.1559/152304093782616779
    » https://doi.org/10.1559/152304093782616779
  • Lozbenev N, Yurova A, Smirnova M, Kozlov D. Incorporating process-based modeling into digital soil mapping: A case study in the virgin steppe of the Central Russian Upland. Geoderma. 2021;383:114733. https://doi.org/10.1016/j.geoderma.2020.114733
    » https://doi.org/10.1016/j.geoderma.2020.114733
  • Ma Y, Minasny B, Malone BP, Mcbratney AB. Pedology and digital soil mapping (DSM). Eur J Soil Sci. 2019;70:216-35. https://doi.org/10.1111/ejss.12790
    » https://doi.org/10.1111/ejss.12790
  • McBratney AB, Odeh IOA. Application of fuzzy sets in soil science: Fuzzy logic, fuzzy measurements and fuzzy decisions. Geoderma. 1997;77:85-113. https://doi.org/10.1016/S0016-7061(97)00017-7
    » https://doi.org/10.1016/S0016-7061(97)00017-7
  • Mendes AC, Truckenbrod W, Rodrigues ACRN. Análise faciológica da Formação Alter do Chão (Cretáceo, Bacia do Amazonas), próximo à cidade de Óbidos, Pará, Brasil. Rev Bras Geocienc. 2012;42:39-57.
  • Menezes MD, Silva SHG, Mello CR, Owens PR, Curi N. Solum depth spatial prediction comparing conventional with knowledge-based digital soil mapping approaches. Sci Agric. 2014;71:316-23. https://doi.org/10.1590/0103-9016-2013-0416
    » https://doi.org/10.1590/0103-9016-2013-0416
  • Menezes MD, Silva SHG, Owens PR, Curi N. Digital soil mapping approach based on fuzzy logic and field expert knowledge. Cienc Agrotec. 2013;37:287-98. https://doi.org/10.1590/S1413-70542013000400001
    » https://doi.org/10.1590/S1413-70542013000400001
  • Milne G. Some suggested units of classification and mapping particularly for East African soils. Soil Res. 1935;4:183-98
  • Möller M, Volk M. Effective map scales for soil transport processes and related process domains - Statistical and spatial characterization of their scale-specific inaccuracies. Geoderma. 2015;247-248:151-60. https://doi.org/10.1016/j.geoderma.2015.02.003
    » https://doi.org/10.1016/j.geoderma.2015.02.003
  • Nicoletti MC, Camargo HA. Fundamentos da teoria de conjuntos fuzzy. São Carlos: EdUFSCar; 2004.
  • Pachepsky Y, Hill RL. Scale and scaling in soils. Geoderma. 2017;287:4-30. https://doi.org/10.1016/j.geoderma.2016.08.017
    » https://doi.org/10.1016/j.geoderma.2016.08.017
  • Pontius RG, Millones M. Death to Kappa: Birth of quantity disagreement and allocation disagreement for accuracy assessment. Int J Remote Sens. 2011;32:4407-29. https://doi.org/10.1080/01431161.2011.552923
    » https://doi.org/10.1080/01431161.2011.552923
  • QGIS DT. QGIS Geographic Information System; 2019. Available from: https://www.qgis.org/
    » https://www.qgis.org/
  • R Development Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2019. Available from: http://www.R-project.org/
    » http://www.R-project.org/
  • Rentschler T, Bartelheim M, Díaz M, Bonilla Z, Teuber S, Schmidt K, Behrens T, Scholten T. Contextual spatial modelling in the horizontal and vertical domains. Sci Rep. 2022;12:9496. https://doi.org/10.1038/s41598-022-13514-5
    » https://doi.org/10.1038/s41598-022-13514-5
  • Roudier P, Hewitt AE, Beaudette DE. A conditioned Latin hypercube sampling algorithm incorporating operational constraints. In: Minasny B, Malone BP, McBratney AB, editors. Digital soil assessments and beyond - Proceedings of the 5th Global workshop on digital soil mapping 2012, Sydney, Australia. London: CRC Press; 2012. p. 227-31. https://doi.org/10.1201/b12728-46
    » https://doi.org/10.1201/b12728-46
  • Santos HG, Jacomine PKT, Anjos LHC, Oliveira VA, Lumbreras JF, Coelho MR, Almeida JA, Araújo Filho JC, Oliveira JB, Cunha TJF. Sistema brasileiro de classificação de solos. 5. ed. rev. ampl. Brasília, DF: Embrapa; 2018.
  • Santos RD, Santos HG, Ker JC, Anjos LHC, Shimizu SH. Manual de descrição e coleta de solo no campo. 7. ed. Viçosa, MG: Sociedade Brasileira de Ciência do Solo; 2015.
  • Schaefer CEGR, Lima HN, Teixeira WG, Vale JF, Souza KW, Corrêia GR, Mendonça BAF, Amaral EF, Campos MCC, Ruivo MLP. Solos da região amazônica. In: Curi N, Ker JC, Novais RF, Vidal-Torrado P, Schaefer CEGR, editors. Pedologia - Solos dos Biomas Brasileiros. Viçosa, MG: Sociedade Brasileira de Ciência do Solo; 2017. p. 75-111.
  • Silva SHG, Owens PR, Menezes MD, Santos WJR, Curi N. A technique for low cost soil mapping and validation using expert knowledge on a watershed in Minas Gerais, Brazil. Soil Sci Soc Am J. 2014;78:1310. https://doi.org/10.2136/sssaj2013.09.0382
    » https://doi.org/10.2136/sssaj2013.09.0382
  • Simonson RW. Outline of a generalized theory of soil genesis. Soil Sci Soc Proc. 1959;23:152-6. https://doi.org/10.2136/sssaj1959.03615995002300020021x
    » https://doi.org/10.2136/sssaj1959.03615995002300020021x
  • Somoza R, Ghidella ME. Late Cretaceous to recent plate motions in western South America revisited. Earth Planet Sci Lett. 2012;331-332:152-63. https://doi.org/10.1016/j.epsl.2012.03.003
    » https://doi.org/10.1016/j.epsl.2012.03.003
  • Teixeira PC, Donagemma GK, Fontana A, Teixeira WG. Manual de métodos de análise de solo. 3. ed. rev e ampl. Brasília, DF: Embrapa; 2017.
  • Wadoux AMJC, Heuvelink GBM, Lark RM, Lagacherie P, Bouma J, Mulder VL, Libohova Z, Yang L, McBratney AB. Ten challenges for the future of pedometrics. Geoderma. 2021;401:115155. https://doi.org/10.1016/j.geoderma.2021.115155
    » https://doi.org/10.1016/j.geoderma.2021.115155
  • Wysocki DA, Schoeneberger PJ. Geomorphology of soil landscapes. In: Huang PM, Li Y, Sumner ME, editors. Handbook of soil science: Properties and processes. Boca Raton, Florida Chemical Rubber Company Press; 2011. p. 1-26.
  • Zhu A-X. A similarity model for representig soil spatial information. Geoderma. 1997;77:217-42. https://doi.org/10.1016/S0016-7061(97)00023-2
    » https://doi.org/10.1016/S0016-7061(97)00023-2
  • Zinck JA. The geomorphic landscape: Criteria for classifying geoforms. In: Zinck JA, Metternicht G, del Valle HF, Angelini M, editors. Geopedology. Cham: Springer International Publishing; 2016. p. 77-99. https://doi.org/10.1007/978-3-319-19159-1_6
    » https://doi.org/10.1007/978-3-319-19159-1_6

*

Corresponding author: e-mail: cauan.araujo@ufopa.edu.br

Editor:

José Miguel Reichert https://orcid.org/0000-0001-9943-2898 and Waldir de Carvalho Junior https://orcid.org/0000-0002-8619-0989

Publication Dates

  • Publication in this collection
    31 Aug 2026
  • Date of issue
    2026

History

  • Received
    30 Sept 2025
  • Accepted
    19 Feb 2026
Creative Common - by 4.0
This is an Open Access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
location_on
Sociedade Brasileira de Ciência do Solo Sociedade Brasileira de Ciência do Solo, Departamento de Solos - Edifício Silvio Brandão, s/n, Caixa Postal 231 - Campus da UFV, CEP 36570-900 - Viçosa-MG, Tel.: (31) 3612-4542 - Viçosa - MG - Brazil
E-mail: sbcs@sbcs.org.br
rss_feed Acompañe los números de esta revista en su lector de RSS