An Assessment of Landscape Evolution Through Pedo-Geomorphological Mapping and Predictive Classification Using Random Forest: A Case Study of the Statherian Natividade Basin, Central Brazil

Abstract

Understanding the relationship between geological and geomorphological processes is essential for reconstructing landscape evolution. This study examines how geology and geomorphology shape landscape development in central Brazil, focusing on the Natividade Group area. Sentinel-2 and SRTM data were integrated with geospatial analyses to produce two key maps: (i) a pedo-geomorphological map, classifying landforms and soil–landscape relationships, and (ii) a predictive geological–geomorphological map, based on a machine learning-based prediction of geomorphic units, which employed a Random Forest classifier trained with 15 environmental predictors from remote sensing datasets. The predictive model classified the landscape into six classes, revealing the ongoing interactions between geology, geomorphology, and surface processes. The pedo-geomorphological map identified nine pedoforms, grouped into three slope classes, each reflecting distinct lithology–relief–soil relationships. Resistant lithologies, such as quartzite-rich metasedimentary rocks, are associated with shallow, poorly developed soils, particularly in the Natividade Group. In contrast, phyllite, schist, and Paleoproterozoic basement rocks from the Almas and Aurumina Terranes support deeper, more weathered soils. These findings highlight soil formation as a critical indicator of landscape evolution in tropical climates. Although the model captured geological and geomorphological patterns, its moderate accuracy suggests that incorporating geophysical data could enhance the results. The landscape bears the imprint of several tectonic events, including the Rhyacian amalgamation (~2.2 Ga), Statherian taphrogenesis (~1.6 Ga), Neoproterozoic orogeny (~600 Ma), and the development of the Sanfranciscana Basin (~100 Ma). The results confirm that the interplay between geology and geomorphology significantly influences landscape evolution, though other factors, such as climate and vegetation, also play crucial roles in landscape development. Overall, the integration of remote sensing, geospatial analysis, and machine learning offers a robust framework for interpreting landscape evolution. These insights are valuable for applications in land-use planning, environmental management, and geohazard assessment in geologically complex regions.

1. Introduction

A complex interplay of geological, geomorphological, pedological, and climatic factors controls landscape evolution, with variations in rock types and structural configurations playing a fundamental role in terrain morphology and differential erosion [1]. These processes shape present-day topography and reflect long-term geological and environmental changes [2]. In the South Tocantins State region, the geological history of the Natividade Group significantly influences modern geomorphological features, affecting landform distribution, drainage patterns, and terrain stability.

This study investigates the geological and geomorphological controls on landscape evolution in the Natividade Basin, located in southeastern Tocantins State, central Brazil. This region is of particular interest due to the presence of gold mines dating back to the 18th century, which have played a significant role in local economic development and remain important to this day. The proximity of these mines to urban areas highlights the need for a detailed understanding of the geological and geomorphological factors shaping the landscape, as they influence terrain stability, which can be exemplified by risks such as landslides and erosion. These factors are crucial for effective land-use planning and risk management in the region.

The Natividade Group consists of a Paleo-Mesoproterozoic metasedimentary sequence with a maximum depositional age of 1776 Ma [3]. The basin evolved during the late Paleoproterozoic (Statherian), possibly extending into the early Mesoproterozoic, within the context of the northern external zone of the Neoproterozoic Brasília Belt [4,5]. Its formation was influenced by thermo-flexural subsidence, leading to the deposition of mixed siliciclastic–carbonate platforms and shallow-water turbidite facies. Geological models suggest that the Almas Block represented a high-paleorelief area that separated the Araí Basin to the south from the Natividade Basin to the north [3,6], a configuration that is still reflected in the present-day landscape.

To investigate the relationship between the geological framework and landscape evolution, this study was guided by the following research questions: (1) How do the lithological and structural characteristics of the Paleoproterozoic basement, the Natividade Group, and the Sanfranciscana Basin influence the distribution of landforms, drainage patterns, and topographic variation in southeastern Tocantins? (2) Can remote sensing and geospatial analysis effectively delineate pedo-geomorphological units and predict geological–geomorphological patterns in a Paleo-Mesoproterozoic sedimentary basin? (3) To what extent does the modern landscape reflect the depositional environments and tectonic configuration of the Natividade Basin during the Statherian period, and how is it related to the underlying Paleoproterozoic basement and the overlying Upper Cretaceous sediments?

By correlating geological and geomorphological characteristics with present-day landscape features, this study aims to enhance the understanding of landscape evolution in the Natividade region. Using Sentinel-2 and SRTM (Shuttle Radar Topography Mission) imagery combined with machine learning techniques, particularly the Random Forest algorithm, it proposes a methodological framework for mapping geomorphological patterns and terrain dynamics. The approach is based on two geospatial products: (i) a pedo-geomorphological map, which derives information on soil distribution from geomorphic units and slope variations, and (ii) a predictive geological–geomorphological map, incorporating multiple remote sensing indices.

Understanding the interactions between geological structures, landforms, and soil development is a central concern in pedo-geomorphology, which has been significantly advanced in recent years through the use of geospatial technologies, such as remote sensing, GIS, and machine learning techniques [7]. These advancements allow for more detailed and accurate mapping of landscape features, offering a more comprehensive understanding of landscape evolution. In this context, the present study not only provides insights into the evolution of the Natividade landscape in central Brazil but also introduces a rapid and cost-effective technique that can be applied in other regions with similar pedogenetic and geomorphological conditions, complementing traditional field mapping and supporting environmental and land-use planning.

2. Geological and Environmental Context

2.1. Geological Setting

The Natividade Group is a sedimentary sequence related to the Proterozoic basins in the Northern Brasília Belt, spanning from the Paleoproterozoic (Statherian) to the early Neoproterozoic (Tonian), with a maximum depositional age of 1776 Ma [3,8]. It can be included within the context of the Veadeiros Supergroup. The study area is situated within the Tocantins Structural Province (Figure 1) [9], which includes the Brasília, Araguaia, and Paraguay orogenic belts [10,11].

In Brazil, the Amazonian and São Francisco cratons are partially covered by rift, sag, and rift–sag-type volcano-sedimentary units deposited during the Paleoproterozoic and Mesoproterozoic. These basins are generally located at the margins or interiors of these cratonic landmasses and are commonly surrounded by Neoproterozoic fold belts (Figure 1a,b).

The northern segment of the Brasília Belt is characterized by a general NE–SW structural trend and an overall east-to-southeast vergence. Its external zone, where the study area is located, comprises a stack of sedimentary sequences deformed against the western margin of the São Francisco Craton, including significant exposures of its sialic basement [12,13,14,15,16,17,18].

During the evolution of the Brasília Belt basement, an accretionary orogeny occurred, involving the amalgamation of micro-blocks along the western margin of the São Francisco Craton between 2.5 and 2.2 Ga [19,20,21]. According to [22], all pre-Neoproterozoic rocks in the central–northern part of the Brasília Belt are grouped within the Goiás Massif. In this context, the study area exposes rocks of the Almas and Aurumina terranes [21].

The Almas Terrane consists of an amphibolite facies greenstone–TTG (tonalite–trondhjemite–granodiorite) association [23,24,25]. The Almas Greenstone Belt (2206 ± 13 Ma [26]), which hosts the Vira Saia and Paiol gold mines, is preserved as narrow belts along TTG batholith margins and is represented by the Riachão do Ouro Group (RO) [21,23,27]. Bordering the Almas Terrane, the Aurumina Terrane is characterized by peraluminous granite and tonalite/granodiorite emplaced between 2.11 and 2.16 Ga [25].

Figure 1. Study area context: (a) location of the study area in relation to the Brasiliano/Pan-African cratons; (b) study area (red rectangle) within Brazil, situated in the Neoproterozoic Tocantins Province [16]; (c) detailed view of the study area, highlighting the Natividade Group and its subdivisions (adapted from [6,17,28] and modified from [18]).

Figure 1. Study area context: (a) location of the study area in relation to the Brasiliano/Pan-African cratons; (b) study area (red rectangle) within Brazil, situated in the Neoproterozoic Tocantins Province [16]; (c) detailed view of the study area, highlighting the Natividade Group and its subdivisions (adapted from [6,17,28] and modified from [18]).

In the Statherian (~1.78 Ga), crustal extension mechanisms led to thermo-flexural subsidence, creating accommodation space for the deposition of the Natividade Group, which is classified as a sag-type basin. According to [3], the paleogeography of the crystalline basement controlled the deposition of the Natividade Group, particularly the high paleorelief of the Almas Block, which facilitated gravitational flows in the southeast and the deposition of siliciclastic and carbonate sediments directly over the crystalline basement [6,17,28]. In the present study, we adopted the terminology proposed by [6], who described eleven sedimentary rock types in the Natividade Group, grouped into four rock assemblages corresponding to specific depositional conditions:

(i) Sand–silt–carbonate assemblage—mixed platform environment with simultaneous siliciclastic and carbonate deposition; (ii) sand–conglomerate assemblage—shallow turbidite environment, related to mass flow controlled by the paleorelief of the source area; (iii) sand assemblage—internal platform in backshore and foreshore conditions; (iv) silt–clay assemblage—external siliciclastic in an open-marine platform with primarily fine-grained deposition.

Finally, in the northeastern portion of the study area lies the Phanerozoic cover of the São Francisco Craton, defined as the Sanfranciscana Basin. This cover is predominantly composed of continental sedimentary rocks, with a minor presence of alkaline volcanic rocks to the south. The basin’s origin is related to the isostatic rearrangement during the Paleozoic, with reactivations in the Mesozoic and neotectonic activity in the Cenozoic [29].

In this region, the Sanfranciscana Basin is represented by the Urucuia Group (Upper Cretaceous), mainly composed of sandstone. This group is subdivided into the Posse and Serra das Araras formations, each with distinct depositional characteristics. The Posse Formation includes two facies: one formed by aeolian deposits from dry dune fields and another by interwoven fluvial deposits accumulated primarily in channels. In contrast, the Serra das Araras Formation consists of braided fluvial deposits characterized by sand and gravel sedimentation [29]. These sedimentary rocks form plateaus with flat tops and steeply eroded slopes [30].

2.2. Environmental Characterization

The study area comprises the Brazilian Cerrado biome, which represents the Neotropical savanna vegetation of central Brazil. The climate is classified according to [31] as C2wA’a’ (dry sub-humid with moderate water deficiency), with an average annual rainfall of approximately 1500 mm.

This biome is characterized by a diverse mosaic of ecosystems, including tropical grasslands, savannas often referred to as Cerrado stricto sensu, and seasonal forests, regardless of their floristic composition [32]. In the carbonate hilltops of the Natividade Group, within the Cerrado biome, a Tropical Deciduous Forest, also known as a “dry forest” [33], can be found. The environmental heterogeneity of the Cerrado, marked by a wide range of vegetation types, topographical features, and climatic conditions, makes it an ideal setting for applying soil spatial predictive models [7].

Additionally, the Cerrado biome exhibits significant latitudinal (n) and altitudinal variability. Its broad geographic distribution across diverse erosion surfaces (r), including lowlands (<300 m), plains, and extensive plateaus (900–1600 m) [34], contributes to substantial climatic diversity (c) [35,36,37].

According to the geomorphological mapping of Tocantins State [34], the study area encompasses six major geomorphological units: the Upper Tocantins Depression (37.06%), characterized by a gently undulating terrain; the Natividade Ridge (27.87%), marked by rugged topography; the Mangabeiras Stepped Relief (21.65%); the Dissected Plateau of Tocantins (12.46%); the Fluvial Plains (0.85%); and the Stepped Plateaus of the Western Chapadão of Bahia (0.10%) [38] (Figure 2).

Considering this geomorphological framework, the distribution of soils in the region reflects strong topographic control. In this text, soil classes are primarily presented according to the Soil Taxonomy system, with corresponding equivalents from the World Reference Base for Soil Resources (WRB/FAO) classification.

In a regional context, the predominance of rugged morphologies, such as stepped plateaus, dissected plateaus, and ridges, favors the formation of young soils, like Entisols (Leptosols; WRB/FAO) (30.47%) and Inceptisols (Cambisols; WRB/FAO) (2.52%). Oxisols (Ferralsols; WRB/FAO) (23.66%) are commonly associated with Plinthic subgroups (Plinthosols; WRB/FAO) (26.95%) in various geomorphological units, particularly in plateau regions, where conditions promote intense pedogenetic processes. Another significant soil group is Ultisols (Acrisols, Lixisols, and Alisols; WRB/FAO) (12.82%), predominantly found in the Upper Tocantins Depression. Finally, Entisols (Aquents, Andepts, and other subgroups in the Soil Taxonomy) and Gleysols (WRB/FAO) (3.58%) are less common and typically occur in Fluvial Plains [38].

Locally, the relief is distinguished in the region by the Natividade Ridge Complex, which is marked by elongated elevations with a sparse vegetation cover and is the main geographical reference in the south of Tocantins State. Quartzite and limestone are the main rocks supporting the highlands. The entire ridge complex is north–south oriented, which is coherent with the main tectonic direction. Because of the relief pattern, this geomorphological compartment is preserved from human occupation.

The lowlands occurring along the sides of the high plateaus show a planar-to-undulating-relief pattern, gentle slopes, and a small total amplitude of the stream valleys and are predominantly linked to schist, phyllite, and granitic rocks. Due to the relief features, this region is predominantly used for extensive cattle ranching.

The main specific landforms identified in the region include hogbacks (associated with metasedimentary successions), trapezoidal facets (along regional fault zones), landforms typical of sandstone tablelands, such as mesas, buttes, and pinnacles, as well as epikarst features (related to the main limestone lenses).

Finally, throughout the Quaternary, compressive stresses associated with neotectonic activity may have reactivated ancient geological structures (ranging from the Paleoproterozoic to the Cretaceous), thereby influencing relief compartmentalization and morphostructural evolution. Karstic landforms, such as dolines, likely developed as a result of intensified subterranean drainage and differential dissolution of carbonate lithologies [39,40,41,42].

Moreover, the transition from a predominantly semi-arid Cretaceous climate to intertropical humid conditions during the Cenozoic enhanced pedogenetic processes, facilitating the formation of more developed and deeply weathered soils, as well as the genesis of planation surfaces [40].

3. Materials and Methods

Figure 3 illustrates the methodological framework applied to the present study, detailed in the subsequent sections. This framework was developed to address the specific research objectives.

3.1. Geomorphological and Geological Classification Map Using Random Forest (Sentinel-2 and SRTM)

3.1.1. Sentinel-2 and SRTM Data

This study utilized Sentinel-2 (Level L2A) and SRTM data for geomorphological and geological classification, considering the specificities of each dataset according to the research objectives. Sentinel-2 Level L2A images, which underwent atmospheric correction, were selected to ensure greater accuracy in spectral analysis. The chosen images correspond to the scene from 13 September 2025, selected due to minimal cloud cover, ensuring optimal data quality. The data were obtained directly from the open-access Copernicus website, with standard spatial resolutions of 10 m for visible and near-infrared bands and 20 m for red-edge and short-wave infrared bands.

Spectral analysis focused on selecting bands 4 (red; 665 nm); 3 (green; 560 nm); 2 (blue; 490 nm); 8A (Narrow NIR; 865 nm); 11 (SWIR-1; 1610 nm); and 12 (SWIR-2; 2190 nm) due to their relevance for geological studies. The SWIR-1 (1610 nm) and SWIR-2 (2190 nm) bands are sensitive to the presence of clays and hydrothermal alteration processes, while the Narrow NIR band (865 nm) helps differentiate between soil and vegetation [43,44,45].

In addition to individual bands, spectral indices were calculated to enhance specific geological features, including the normalized difference water index (B3 − B8/B3 + B8), which is sensitive to soil and vegetation moisture [46]; the iron oxide index (B4/B2), used to highlight the presence of iron oxides [47]; and the clay index (B11/B12), associated with clay minerals [48]. Additionally, the normalized difference vegetation index was computed using the red (Band 4) and near infrared (Band 8) bands to evaluate the vegetation cover ref. [49], which can influence erosion.

The combination of these bands and spectral indices enabled a detailed characterization of the study area’s geology and geomorphology, contributing to the interpretation of erosional processes and landform features.

The Shuttle Radar Topography Mission (SRTM) data were obtained from the Topodata/INPE platform [50], which provides pre-processed digital elevation models with improved spatial resolution and corrections for terrain artifacts. These data were essential for generating geomorphometric variables relevant to this study [51].

Pre-processing included filling depressions (sinks) using the Fill Sinks tool within the SAGA extension in QGIS [52]. This step ensures a continuous surface, preventing artificial interruptions in drainage pathways. The same tool was applied to generate the flow direction, which serves as the basis for delineating drainage networks and understanding terrain connectivity. Subsequently, terrain derivatives were computed using another SAGA plugin, including slope, aspect, and total curvature maps, which provide insights into the terrain inclination, orientation, and concavity/convexity [53]. The flow accumulation model was generated, which was later used to derive the topographic wetness index (TWI), an indicator of potential water concentration zones and their influence on surface dynamics.

The erosion sensitivity index (ESI) was calculated by combining the slope and total curvature derived from the SRTM data with the NDVI obtained from Sentinel imagery. This integration of topographic and vegetation factors provides a more comprehensive assessment of erosion susceptibility, capturing both terrain dynamics and vegetation cover influence [54,55].

3.1.2. Random Forest Classification

The Random Forest classifier, a supervised machine-learning algorithm, integrates decision trees with an ensemble learning approach to improve classification accuracy. Three key parameters were selected for Random Forest: Ntree (the number of trees to grow), Mtry (the number of variables considered at each node split), and variable importance (the contribution of each variable or band to model performance). The optimal values for Ntree and Mtry were determined iteratively by minimizing the mean square error (MSE).

In this study, we applied the Random Forest algorithm using the Dzetsaka classification tool in QGIS.

Sentinel and SRTM imageries were loaded into the Dzetsaka interface, and multiple spectral signatures were captured for each geological/geomorphological category. The number of trees was set to 100, while the maximum depth, variables per split, and other parameters were left at their default settings. The classification was successfully performed.

Sentinel-2 and SRTM imagery were employed to analyze and estimate the geological and geomorphological influences on the modern landscape of the Paleo-Mesoproterozoic Natividade Basin, based on their ability to capture spectral and topographic variations. Fifteen predictive variables were employed, including Sentinel-2 bands 2, 3, 4, 8, 11, and 12, as well as spectral indices, such as NDWI, NDVI, clay index, and iron oxide index. Additionally, terrain derivatives derived from SRTM data, including slope, aspect, the erosion sensitivity index, and the topographic wetness index, were incorporated, as they effectively characterize the region’s current landscape. All input raster layers were normalized and resampled to a 100 m cell size to ensure consistency across datasets, which corresponds approximately to a cartographic scale of 1:100,000.

To generate training data, a geological–geomorphological features map based on the combination of a bibliographic review, field campaigns, and satellite image interpretation was produced. From this map, five geological/geomorphological classes of interest were defined. Using the methodology of [56], 100 random points per class were generated, and raster values corresponding to geological/geomorphological features were extracted for these points using ArcGIS 10.3 ArcToolbox tools.

For classification, QGIS v3.24 was used. The classification was conducted with the Scikit-learn library and the Dzetsaka plugin, which facilitated the supervised classification of Sentinel-2 and SRTM imagery using the Random Forest classifier.

3.1.3. Accuracy Assessment

An accuracy assessment is crucial for correcting mapping errors, such as misclassifications at categorical boundaries, where continuous features are assigned to discrete classes. It also helps identify variations in input data, differences in classification methodologies, and analyst bias, improving the reliability of the final classification.

To assess accuracy, a confusion matrix (error matrix) was applied, which compared the predicted class labels against the reference data. The Dzetsaka plugin in QGIS includes an option to split the training dataset for model evaluation. A total of 70% of the data were allocated for model training, and 30% was retained for validation, allowing for an independent assessment of classification quality [56]. The confusion matrix organizes classification results by displaying correctly classified instances along the main diagonal, while omission and commission errors appear in the off-diagonal elements. Two key metrics were derived from this matrix: overall accuracy (OA), which measures the proportion of correctly classified instances across all classes, providing a general estimate of classification performance, and the Kappa index (κ), which accounts for agreement beyond chance, offering a more robust evaluation of classification accuracy by considering both correct and incorrect classifications across all classes. These accuracy metrics ensure a reliable assessment of how well the classification model represents the geological and geomorphological features of the study area.

3.2. Pedo-Geomorphological Map

One of the most relevant models for soil prediction is SCORPAN, developed by [57]. This soil spatial predictive function establishes quantitative relationships between soil properties and environmental covariates, encompassing geology, geomorphology, pedology, and other relevant factors [7]. SCORPAN builds upon the CLORPT model [58], which defines soil formation based on climate (C), organisms (O), relief (R), parent material (P), and time (T). SCORPAN extends this concept by incorporating additional variables: soil information (S) and spatial location (N). These factors are primarily represented through rasterized images obtained via remote/proximal sensing or geoprocessing-derived data.

The relationship between landforms and soils at different spatial scales allows for the inference of soil unit distribution through digital topography analyses [59]. Additionally, pedoforms are intrinsically linked to slope variations, which influence soil formation and landscape evolution [2,60,61,62,63,64]. Given these interdependencies, the methodology of [1] was applied to map the pedoforms of the Natividade Group region. This approach defines pedoforms based on the integration of geomorphological units and slope classes, followed by their interpretation in relation to regional soil distribution.

The first stage involved the generation of the geoform map, which was created by combining geomorphological units with slope classes. To achieve this, a spatial overlay analysis was performed using the raster calculation tool in ArcGIS 10.8.2. A summation operation was applied to merge the geomorphological unit map with the slope classification map, ensuring that all slope classes were represented within the geomorphological units identified in the study area. This step allowed for a more refined delineation of terrain variations and their potential influence on pedogenesis.

To ensure the reliability of the final maps, a field validation procedure was carried out, guided by methodological rigor rather than quantitative metrics. The validation focused on assessing the consistency between mapped pedoform classifications and real-world landscape conditions through direct observations at representative locations across the study area. These sites were selected based on their expression of key geomorphological and pedological features, informed by remote sensing analysis and previous survey data. This process aimed at qualitatively verifying the coherence between the mapped units and field evidence. Although the validation did not produce a statistical accuracy index, the field campaign provided consistent empirical support for the reliability and internal coherence of the final pedoform map, reinforcing its capacity to represent the spatial organization of landforms, soils, and lithological settings in the Natividade Group region.

By integrating digital terrain analysis, remote sensing data, and field validation, this methodological approach provides a robust framework for understanding the landscape evolution of the study area and its relationship with geological and geomorphological processes.

4. Results

Two main maps were generated to analyze the landscape evolution of the Natividade region: a pedo-geomorphological map and a predictive geological–geomorphological map. The first highlights the relationship between relief and soil distribution, while the second reveals geological and structural patterns shaping the modern landscape. Together, these maps provide a comprehensive view of the geomorphological and geological influences on terrain evolution.

4.1. Pedo-Geomorphological Map (PGM)

The pedo-geomorphological map was produced by combining geomorphological units with slope classes, allowing for a detailed analysis of terrain variation and its relationship with soil distribution [1] (Figure 4). This process resulted in the identification of nine pedo-geomorphological units, each corresponding to a soil association and suborders derived from the six main soil types shown in Figure 2. The spatial distribution of these units revealed distinct patterns, where flatter areas are predominantly associated with deeper, more developed soils, while steeper terrains are linked to less-developed soils, influenced by erosion and geological resistance [65]. These patterns reflect the underlying lithology, where quartzite-rich areas are characterized by rugged relief and thin soils, while sequences richer in silt and clay, as well as Paleoproterozoic basement rocks, support more stable and thicker soil profiles.

Pedo-geomorphological Unit 1—This unit represents landforms associated with flat reliefs (0 to 5°), exhibiting small, circular morphologies that are barely discernible on maps, and is typically associated with lagoons and hydromorphic soils. It is related to impermeable basement rocks of the Almas Terrane, which consist of an amphibolite facies greenstone–TTG association.

Pedo-geomorphological Unit 2—This unit includes landforms associated with flat reliefs (0 to 5°), particularly within the geomorphological unit of the dissected Tocantins plateau. It is related to areas with more intensive agricultural activity and the occurrence of well-developed soils, such as Oxisol and Plinthic subgroups. These soils are commonly found in the northwestern portion of the area, associated with the basement of the Aurumina Terrane, and are characterized by peraluminous granite and tonalite/granodiorite.

Pedo-geomorphological Unit 3—This unit represents landforms associated with gentle reliefs (0 to 5°), allowing for the formation of deeper soils, such as Oxisol and Plinthic subgroups. However, shallower soils also occur, such as Lithic Entisol and Inceptisol. These soils are found in the Natividade Group and are especially related to the association of silt–clay and sand–silt–carbonate assemblages. Their occurrence is common in the northwestern portion of the area, related to the basement of the Aurumina Terrane, and they are characterized by peraluminous granite and tonalite/granodiorite.

Pedo-geomorphological Unit 4—This unit is characterized by a flat relief, with slopes ranging from 0 to 5°, and is associated with the Patamares das Mangabeiras compartment. In this context, the occurrence of soils is related to Entisol, specifically, the quartz sandy suborder, due to their predominantly sandy texture, and the Lithic subgroups of Entisol, which are characterized by shallow soils over a hard bedrock or weathered material [66]. The presence of these soils reflects the influence of the Phanerozoic cover of the Sanfranciscana Basin (Urucuia Group), where sandy sediments play a key role in the formation of the landscape.

Pedo-geomorphological Unit 5—This predominant relief class is flat (0° to 5°), with the most prominent geomorphological feature being the Upper Tocantins Depression. The soils are well developed and deep, with an association of Ultisol, Plinthosol subgroups, and Oxisol, indicating well-drained areas that favor intense weathering processes. Geologically, the region is primarily underlain by the Paleoproterozoic basement, particularly, rocks from the Almas Terrane. The Aurumina Terrane, which occurs in this unit, is characterized by peraluminous granite and tonalities rocks.

Pedo-geomorphological Unit 6—This unit encompasses landforms with gentle to moderate slopes (5° to 15°), strongly associated with drainage networks and valley systems. Geomorphologically, these areas are linked to fluvial plains, the dissected Tocantins plateau, and the Natividade Ridge. The predominant soils are Aquic Entisols, often accompanied by other Entisol suborders, with occasional occurrences of more developed soil classes. Geologically, this unit is closely associated with recent fluvial deposits from the Sanfranciscana Basin and exhibits a distinct morphological alignment with the Natividade Group, while its connection to the crystalline basement is less pronounced.

Pedo-geomorphological Unit 7—This predominant relief class is gently sloping to moderately steep (5° to 15°), with the possibility of occurring in areas of steeply undulating (15° to 30°) terrain. This unit is predominantly found in more rugged portions of the geomorphological unit of the Upper Tocantins Depression and is frequently associated with areas near the Natividade Ridge. Geologically, it is found in the structurally more complex portions of the Paleoproterozoic basement, especially near the contact with the Natividade Group. Due to the relatively low level of detail in the soil mapping, this unit may theoretically occur in various soil associations, such as Ultisols and Plinthic subgroups. However, based on field observations and remote sensing product analysis, the predominant occurrence is of Entisol and Inceptisol. When it occurs within the Natividade Group, this unit is typically found on relatively flatter hills (undulating relief) and is associated with the sand–silt–carbonate and sand–conglomerate assemblages. In this context, it is related to the internal portions of Pedo-geomorphological Units 8 and 9.

Pedo-geomorphological Unit 8—This unit is predominantly characterized by steep slopes (15° to 30°) and is primarily located at the boundaries of the Natividade Ridge, marking significant geomorphological breaks, especially in the transition to the Upper Tocantins Depression. The predominant soils are an Entisol association of Typic Udorthents and Lithic Udorthents, reflecting the direct influence of topographic characteristics and geological context. Geologically, this unit is found at the abrupt contact between the Natividade Group and the Paleoproterozoic basement. The landscape is also associated with large carbonate lenses, indicative of the sand–silt–carbonate assemblage, and quartz-rich portions from both the sand–silt–carbonate and sand–conglomerate assemblages. The presence of shear zones and high slope gradients further accentuates its steep terrain, emphasizing the ruggedness and variability of this unit.

Pedo-geomorphological Unit 9—This unit is mainly found on hilltops and crests within the Natividade Ridge, with steeper slopes (15° to 30°), in some areas exceeding 30°, classifying it as mountainous. The soils are Entisols, particularly Lithic Udorthents, with exposed bedrock, highlighting more erodible and less-developed soil profiles compared to Unit 8. Unlike Unit 8, which lies at the boundary and involves more complex interactions with various lithologies, Unit 9 is strongly associated with quartzite from the sand–conglomerate assemblage, a rock highly resistant to weathering and erosion.

4.2. Predictive Geological–Geomorphological Map (PGG Map)

The geological–geomorphological predictive map was generated using the Random Forest algorithm, incorporating 15 predictive layers derived from SRTM and Sentinel-2 imagery. These layers included the slope; topographic wetness index (TWI); total curvature; erosion; aspect; NDVI; NDWI; clay index; iron oxide index; and spectral bands B2, B3, B4, B8A, B11, and B12 (Figure 5). This combination of variables enabled a detailed analysis of geological and geomorphological patterns in the study area.

To evaluate the reliability of the classification, the overall accuracy and the Kappa index were calculated, resulting in values of 52% and 0.4, respectively. These metrics indicate moderate agreement between the predicted and reference data, reflecting both the potential and the limitations of predictive mapping in geologically complex terrains.

As highlighted by [67], the use of remote sensing data in areas with high geological variability often yields moderate accuracy levels, which are considered acceptable in geological and geomorphological studies due to the complexity of terrain patterns and challenges in distinguishing between similar landform classes.

A crucial step in this process was the construction of the regional geological–geomorphological map (Figure 6), which aided as the primary reference dataset for training the predictive model. This map was developed based on the interpretation of high-resolution satellite imagery, topographic data, and extensive fieldwork (Figure 6). It provides a comprehensive representation of the geological formations, geomorphic features, and structural elements that characterize the landscape. To generate training data for the predictive model, 100 random points were extracted for each geological–geomorphological unit identified in this reference map.

The analysis identified six distinct geological–geomorphological units in the study area, shown on the predictive geological–geomorphological map (Figure 7), each reflecting specific characteristics related to geology, topography, and spectral patterns observed in satellite images.

PGG Unit 1—This unit is associated with the Natividade Group, occurring in areas of high topography with steep slopes, indicating its resistance to erosional processes and connection to more resistant geological structures. It is also related to drainages and other areas with topographic variations.

PGG Unit 2—This unit is located in the northeastern portion of the study area and is linked to the Sanfranciscana Basin (Urucuia Group) and is associated with a high density of spectral patterns related to land management, suggesting the influence of depositional processes and anthropogenic activities.

PGG Unit 3—This unit corresponds to the Almas Terrane and is associated with pasture areas and spectral patterns indicative of anthropogenic activities.

PGG Unit 4—This unit corresponds to a major drainage area in the region, reflecting spectral patterns associated with water. Geologically, it corresponds to the Almas and Aurumina Terranes, displaying structural patterns characteristic of this unit.

PGG Unit 5—This unit corresponds to the Riachão do Ouro Group Greenstone Belt (Almas Terrane Greenstone) and is associated with a moderate density of spectral patterns related to land use and management, reflecting an area of low relief.

PGG Unit 6—This unit is also linked to the Almas Terrane and exhibits a spectral pattern characteristic of vegetation, indicating a greater natural vegetation cover compared to the other geological–geomorphological units.

These units reflect the interaction between geological and geomorphological processes in the region, providing insights into landscape evolution and identifying patterns related to geological structuring and surface processes.

The resulting geological–geomorphological predictive map reflects the complex interplay between geology, geomorphology, and landscape evolution in the region. By integrating spectral, topographic, and environmental variables, it enhances the understanding of spatial patterns and their geological significance. To assess the reliability of the classification, the overall accuracy and the Kappa index were calculated, yielding values of 52% and 0.4, respectively. These metrics indicate moderate agreement between the predicted and reference data, highlighting both the potential and the challenges of predictive mapping in geologically complex terrains (Figure 7).

4.3. Pedo-Geomorphological and Geological–Geomorphological Groups

In the study area, pedo-geomorphological units were grouped based on slope classes, highlighting the relationship between geology, geomorphology, and pedology. These groups were defined within three main relief classes, with each class being further associated with specific PGM (pedo-geomorphological map) and PGG (predictive geological–geomorphological map) units. This approach integrates geomorphological features, slope gradients, and soil types, offering a comprehensive understanding of the landscape’s evolution and its controlling factors (Figure 8).

Group 1 encompasses units characterized by flat to gently sloping terrains (0–5°) and is associated with stable landscapes and deep, well-developed soils (PGG Units 2, 3, 4, 5, and 6). It includes PGM Units 1, 2, 3, 4, and 5, which are linked to well-developed soils, such as Oxisol, Ultisol, and Plinthic subgroups, and is associated with bedrocks from the Almas and Aurumina Terranes, as well as less-developed soils, like Quartzipsamments, which are associated with the Sanfranciscana Basin (Urucuia Group).

Group 2 comprises areas with gently undulating to undulating reliefs (5–15°) and is strongly associated with drainage networks and valley systems, featuring a mix of soil development stages (PGG Unit 1). This unit includes less-developed soils, such as Aquic Entisols (PGM Unit 6) and Entisols and Inceptisols (PGM Unit 7). Geologically, these areas are associated with the Sanfranciscana Basin, the Natividade Group, and the crystalline basement.

Group 3 encompasses units with steep to mountainous terrains (15–30° and >30°). It includes PGM Units 8 and 9 and is predominantly characterized by Udorthent soils. These areas feature significant rock outcrops composed of quartzites and substantial carbonate lenses, especially along the ridges and edges of the Natividade Ridge, in contact with Paleoproterozoic basement rocks (PGG Unit 1).

5. Discussion

To enhance the discussion of the fundamental aspects of this study, two main axes were considered: (1) the relationship between the generated maps and landscape evolution and (2) the influence of geology and geomorphology on landscape development.

5.1. The Relationship Between the Generated Maps and Landscape Evolution

The interaction between pedology and geomorphology can be analyzed at multiple levels, as soils and landforms are influenced not only by the topography and slope but also by their relative position within the landscape [61,68]. In this study, we aimed to investigate the landscape evolution of the region encompassing the Natividade Group in central Brazil by integrating Sentinel-2 and SRTM imagery with geomorphological and geological analyses.

Through the pedo-geomorphological/landscape map, we identified and interpreted nine distinct pedoforms/landforms, each associated with specific soil types in the region. These units reflect variations in reliefs, drainage patterns, and lithology, providing valuable insights into the long-term landscape dynamics. The spatial distribution of these pedoforms suggests a strong correlation between soil development, geomorphological processes, and the underlying geological framework [1]. This mapping approach enhances our understanding of how different environmental factors have interacted over time to shape the current landscape configuration.

The pedoforms identified in this study reflect the intricate relationship between bedrock composition and soil development, demonstrating how soils serve as key indicators of landscape evolution. Different rock types influence soil formation processes, dictating the mineralogical composition, weathering rates, and nutrient availability. In turn, these soils record the long-term geomorphological changes that have shaped the region. In general, more resistant lithologies, such as quartz-rich metasedimentary rocks, are often linked to shallow, poorly developed soils, whereas less-resistant rocks, like phyllite and schist, and granitic rocks tend to produce deeper, more chemically altered soils.

As presented in the Results Section, the pedo-geomorphological units were organized into three groups based on relief characteristics. This classification highlights the interplay between topography, geology, and soil development, contributing to a deeper understanding of the processes shaping the landscape (Figure 8).

By analyzing the spatial distribution of pedoforms, it is possible to infer past and present geomorphological dynamics. The presence of well-developed soils in stable, low-relief areas suggests long periods of weathering with minimal erosion, while thinner soils in steeper terrains indicate active denudation. In this sense, the soil distribution serves as a geological–geomorphological archive, recording the landscape’s transformation over time and reinforcing the importance of integrating pedological studies into broader geomorphological analyses.

The hills and ridges in the region show rounded shapes which indicate the evolution of old mountains. Even the lower flattened areas are covered by thick soil regolith that corroborates long-term geomorphological evolution.

The continuous plinthosol horizon, commonly associated with lithified iron oxide crusts, also indicates an ancient geomorphological evolution. This layer is preserved at the 500 to 550 m elevation surface and may represent a former regional phreatic surface, where fluctuations in the water table were responsible for changes in iron oxidation. Iron mobilization is related to the ion valence, which is mobile at 2+ and tends to precipitate when oxidized to 3+.

The predictive map identified six distinct classes, each representing key aspects of the contemporary landscape. These classes reflect the dynamic interaction between geological formations and geomorphological processes, illustrating how subsurface structures and surface features are manifested today. By integrating the predictive modeling results with field observations and geological data, this map offers valuable insights into how ancient geological processes continue to influence current landforms and surface patterns.

The accuracy index obtained in this study was moderate, indicating that while the predictive geological–geomorphological mapping was effective, it is not entirely representative of the landscape’s complexity. This result highlights that the relationship between geology and landscape evolution is significant but not absolute, as other environmental and climatic factors also play essential roles in shaping the terrain [39].

Although the predictive model successfully mapped the geological–geomorphological units, some limitations suggest that additional attributes could improve classification accuracy. For instance, geophysical data, such as magnetometer and gamma ray spectrometry, could provide valuable insights into subsurface variations that influence surface processes. However, the absence of high-resolution geophysical maps for the study area prevents their integration into the analysis.

Despite these limitations, the generated map effectively captured the major geological–geomorphological units, reinforcing the strong influence of geological structures and lithology on landscape evolution. The correspondence between mapped units and known geological features confirms the approach’s reliability in identifying broad landscape patterns and their geological controls.

5.2. Influence of Geology and Geomorphology on Landscape Evolution

In this sense, it is possible to understand that the role of geology and geomorphology in landscape evolution is fundamental, as these factors dictate the physical characteristics and spatial arrangement of landforms over time. Geology, through the distribution of rock types and structures, directly influences the formation and modification of landforms, while geomorphology governs the processes of erosion and sedimentation and the development of surface features. Together, they shape the topography, soil formation, and drainage patterns that define the landscape.

Based on the model by [3], the Almas Block (crystalline basement), particularly the Conceição do Tocantins region, is considered to be a high paleorelief area due to the following factors: the deposition of the Natividade Group, which contains shallow water turbidites, required elevated source areas to control sedimentation. Furthermore, these data are corroborated by the current landscape, where the predominance of carbonates and mass flow deposits (sand–conglomerate assemblage) in the southern portion of the Natividade Group, near the Almas Block, suggests a shallower sea depth to the south of the basin. To the north, the predominance of fine-grained terrigenous sediments and the absence of flow deposits indicate a deeper basin (Figure 8).

It is worth mentioning that the high paleorelief area was only identified through magnetometer data and geological interpretations, indicating tectonic stability during the Rhyacian, following the amalgamation of the Almas–Natividade Terrane [3]. In the current landscape, these regions consist of flat surfaces with a higher proportion of deep soils and intense agricultural activity.

Another important factor in the current relief of the region was imposed by the orogeny related to the Neoproterozoic fold belts, which originated the Brasília Belt. This area is characterized by the northern segment, which presents a NE-SW structural trend and an overall east to southeast vergence, which also strongly influences the current landscape, especially with the topographic highs of the Serra do Natividade, which have undergone significant shortening due to tectonic inversion.

Finally, the last major geological event recorded in the northeastern portion of the area was the development of the Sanfranciscana Basin, with significant reflections in the current landscape configuration. This Phanerozoic cover is predominantly composed of continental sedimentary rocks, which, due to the high proportion of quartz, promote the predominant occurrence of soils related to Entisols (sandy suborder), typically in higher flat reliefs when compared to the basement.

Thus, the interaction between these elements is particularly important in regions with complex geological histories, such as the Natividade region. Here, the influence of the underlying geological formations, such as the Natividade Group, Almas–Aurumina terranes and Urucuia Group, has a lasting impact on the region’s geomorphology, contributing to its present-day geomorphology. Additionally, the combination of these factors with climate and vegetation further refines the landscape, creating patterns of soil distribution, landforms, and drainage that reflect both past and ongoing processes of landscape evolution.

By integrating remote sensing data with field-based geological and geomorphological studies, this research highlights how the evolution of the landscape in the Natividade region is shaped not only by the geological framework but also by the dynamic interplay between these natural processes. The results underscore the importance of understanding these relationships in order to better manage and plan for land use, environmental conservation, and risk mitigation in areas with complex geological and geomorphological settings.

In a general view, the current land uses in the region are compatible with the landscape suitability, where the major agricultural and grasslands are developed in lowland, flat reliefs with thick soil cover. The main farm preservation areas and wildlife reserves are distributed in the high relief regions with hills and ridges.

The analysis of land stability showed lowlands with flattened relief pattern areas with the most stable terrains. However, some highlands even with high slopes also show high geotechnical stability due to the substrate related to quartzite and metaconglomerate, which are the most resistant rocks in the region.

The presence of complex substrates with different rock types, including silicified quartzite, resulted in a high fit of geology controlling geomorphology, as can be observed in the Chapada dos Veadeiros and Federal District regions, central Brazil [68,69].

6. Conclusions

The study area has undergone significant transformations throughout geological time, beginning with the amalgamation and stabilization of the Almas Terrane during the Rhyacian, which led to the formation of a pronounced paleorelief. This was followed by the deposition of the Natividade Group, related to the Statherian Taphrogenesis process, and later by tectonic inversion during the Neoproterozoic. Subsequently, the deposition of the Phanerozoic cover of the Sanfranciscana Basin occurred. This complex geological history has had a profound influence on the current landscape, which has been further shaped by ongoing processes related to climate, vegetation, and topography.

The regional directions and relief trends are strongly influenced by the geological structures at depth, including folds axis, fault, and thrust planes. The ridges, continuous hills, and parallel valleys are northeast-aligned (North 20 to 30° East), which is the same direction of the folds and faults related to the Neoproterozoic orogenesis.

The integration of geological, geomorphological, and pedological data was essential for understanding landscape evolution in the study area. By combining remote sensing data (SRTM and Sentinel-2) with field observations, two key cartographic products were generated: the pedo-geomorphological map and the predictive geological–geomorphological map. It is worth noting that the model demonstrated moderate classification accuracy, with an overall accuracy of 52% and a Kappa index of 0.4, values that are considered acceptable given the geological complexity of the terrain.

The pedo-geomorphological map revealed a clear relationship between geomorphological features and soil distribution, with thicker soils predominating in flatter areas linked to the Almas–Aurumina Terrains and the Urucuia Group and shallower soils found in regions associated with the Natividade Group. In contrast, the predictive geological–geomorphological map, generated using the Random Forest algorithm and validated through fieldwork, identified six distinct geological–geomorphological units, each reflecting patterns shaped by geological formations, topography, and surface processes.

Although the maps showed moderate accuracy, they offer valuable insights into the interactions between geological, geomorphological, and pedological factors, providing actionable insights for land-use planning and regional geotechnical assessments. The integration of remote sensing techniques with field data proved to be an effective method for mapping and analyzing the factors influencing terrain development, underscoring the continuing role of geological structures and geomorphological features in soil development.

Finally, this study contributes to the broader field of pedo-geomorphology by demonstrating how geospatial tools and geomorphological, geological, and pedological interpretations can be integrated to analyze landscape evolution and soil distribution in geologically complex regions. The methodological framework adopted here, combining machine learning (Random Forest), remote sensing, and field validation, can serve as a reference for similar studies worldwide, particularly in regions with comparable soil-forming factors (CLORPT). This approach has the potential to assess landscape evolution through pedo-geomorphological mapping and predictive classification, enabling faster, cost-effective, and reproducible studies. It could also contribute to guiding policy decisions for environmental management.