Topographic modulation of soil functional indicators in shaded coffee agroforestry systems: a multivariate and network-based approach

Abstract

Brazil is the world’s leading coffee producer and increasingly adopts shaded agroforestry systems to enhance sustainability. However, the influence of topography on soil functionality within these systems remains insufficiently understood. This study evaluated soil physical and chemical properties across slope positions (Upper, Middle, and Lower Thirds) and depths (0–60 cm) in a shaded coffee agroforestry system using multivariate statistics and Bayesian network modeling. Results revealed that upper slope positions exhibited greater macroporosity (15–20%) and lower bulk density (1.10–1.15 g cm⁻³), whereas lower slope positions accumulated higher total organic carbon (2.5–3.0%) and microporosity (28–32%). Principal Component Analysis indicated that topography modulated soil porosity and carbon distribution, with total organic carbon (TOC) positively correlating with nutrient availability and negatively with acidity. Bayesian network analysis identified TOC as the most influential attribute, displaying the highest expected influence (1.25) and strength (1.15), along with elevated centrality in conserved environments. These results demonstrate that TOC functions as a central integrator linking soil structure, chemistry, and fertility across topographic gradients. Overall, shaded coffee agroforestry enhanced soil quality and functionality, particularly in upper slope areas, underscoring its potential for sustainable land management in tropical landscapes.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-37724-3.

Introduction

Coffee cultivation (Coffea spp., Rubiaceae) is one of the most important perennial agricultural activities in tropical regions. Brazil, historically the global leader in both production and exports, harvested 51.36 million bags in 2016, accounting for 33.4% of the world’s supply, with exports exceeding 578,000 tons in the first months of 2017¹.

This coffee production chain supports approximately 8.4 million jobs and is economically strategic for small and medium-sized producers, especially in mountainous regions². However, production intensification under conventional systems, characterized by heavy use of synthetic fertilizers, herbicides, and mechanization on sloping terrain, has accelerated tropical soil degradation^3,4.

Soil degradation in conventional coffee systems is evidenced by declines in total organic carbon stocks, structural disruption, increased bulk density and water erosivity, acidification, nutrient leaching, and depletion of cation exchange capacity^5–7. These processes impair critical soil functions, including water retention, mineralizable nitrogen supply, and resilience to water erosion⁸. Soil function refers to the capacity of soil to perform essential processes such as nutrient cycling, water storage, and supporting biodiversity⁹, whereas soil quality is commonly assessed through indicators including organic matter content, porosity, and microbial activity¹⁰.

To address these issues, agroforestry systems with shaded coffee have been adopted by farmers seeking more sustainable production practices¹¹. In these systems, the presence of native tree species plays a key role in modulating soil biogeochemical and structural processes^12,13.

These trees directly enrich organic matter¹⁴, improve soil fertility, and stimulate microbial activity¹⁵, thereby increasing nutrient retention and carbon fixation in plants¹⁶. Additionally, the shade provided by these trees reduces temperature and evaporation, maintaining a more stable and favorable microclimate for coffee tree development¹⁷.

Furthermore, litter deposition in shaded agroforestry environments contributes to the formation of humic substances, aggregate stabilization through glomalin and mucilage production, and nutrient retention via organo-mineral interactions within colloidal microdomains^18,19. Glomalins are glycoproteins produced by arbuscular mycorrhizal fungi that enhance soil aggregate stability²⁰, whereas mucilages are polysaccharides that improve soil structure²¹. Organo-mineral interactions involve the binding of organic matter to mineral surfaces, protecting organic carbon from decomposition²². In highly weathered soils, such as Latosols and Argisols dominated by iron and aluminum oxides, these mechanisms are crucial for stabilizing soil organic matter²³.

Shading acts as a physical and biological barrier against water erosion through multiple interconnected mechanisms. Plant residues on the soil surface significantly reduce runoff, with studies reporting reductions of up to 90% under optimal agroforestry conditions^24–26. These protective mechanisms operate through residue accumulation at the soil surface, which attenuates raindrop impact and reduces soil detachment²⁷.

Biologically, residues promote the formation of larger soil aggregates and increased macroporosity, collectively enhancing water infiltration and reducing surface runoff. The protective mechanisms include litter accumulation, reduced soil surface sealing, and increased infiltration capacity^28,29. Additionally, shading modulates microclimatic conditions by reducing temperature fluctuations, thereby promoting water infiltration and stimulating functional microbial communities. This biological activity enhances nutrient mineralization and stabilizes biogeochemical cycles such as nitrogen (N) and carbon (C) cycling^6,30.

Topographic position is another key factor modulating soil functionality in coffee-growing systems. In mountainous tropical environments, toposequences influence the redistribution of sediments, water, and soil organic matter through leaching, sheet erosion, and surface runoff^31,32.

Upper slope regions frequently retain greater total organic carbon stocks^33,34, exhibit higher aggregate stability^35,36, and sustain increased microbial enzyme activity^37,38, whereas lower slope areas often experience compaction, accumulation of exchangeable aluminum^31,32, and reduced effective porosity³⁹, compromising water storage and root oxygenation.

Despite advances in understanding the benefits of agroforestry systems, substantial knowledge gaps remain regarding the integrated influence of shading and topography on soil functionality in tropical agroecosystems. Most current studies focus on isolated analyses of physical, chemical, or microbiological attributes and lack the systemic approach necessary to assess soil quality in these environments^40–42.

The fragmented thematic priorities and methodological approaches in soil research are well documented^43,44, particularly in tropical regions where comprehensive, integrated assessments remain scarce^42,45,46 and⁴⁷ highlight the lack of studies evaluating multiple variables simultaneously.

These integrated parameters provide sensitive indicators of soil quality. Among the most promising are stratified total organic carbon stocks, litter carbon-to-nitrogen ratios, activities of hydrolytic and oxidative enzymes (e.g., β-glucosidase, phosphatase), aggregate stability indices, and microbial biomass carbon⁴⁸. Such indicators enable assessment of the soil’s capacity to sustain essential ecosystem functions, especially in environments exposed to anthropogenic pressures and severe environmental constraints.

Therefore, this study aimed to analyze the interactions among physical, chemical, and biological soil attributes and their influence on total organic carbon (TOC) in shaded coffee agroforestry systems using network modeling and principal component analysis. Additionally, it evaluated the combined effects of shading by native trees and topographic position on physical, chemical, and microbiological soil attributes in agroforestry coffee plantations in the municipality of Taquaritinga do Norte, Pernambuco State, Brazil.

Material and methods

The study was carried out at the Várzea da Onça farm in the municipality of Taquaritinga do Norte, located within the Borborema Plateau geoenvironmental unit in Pernambuco State (Fig. 1), at the geographical coordinates 7° 53′17’’ South and 36° 5′33’’ West (decimal degrees: –7.88806, –36.0925). In the area, the predominant relief is called Brejo de Altitude, situated on the banks of the PE-130 highway, approximately 12 km from the PE-104 highway. The predominant soil type is a Typical Eutrophic Red-Yellow Argisol (Argissolo Vermelho-Amarelo Eutrófico Típico), with a medium to clayey texture and a prominent A horizon.

Fig. 1.

Location map of the study area at Várzea da Onça Farm in the municipality of Taquaritinga do Norte, Pernambuco State, Brazil. Note: The map was produced using QGIS version 3.34.4 (Prizren) (https://qgis.org), with vector data from IBGE (2022) and satellite imagery from Google Satellite (2025). Coordinate reference system: SIRGAS 2000.

Local vegetation consists of subperennial forest type and mountainous relief⁴⁹. Soil types in the region are predominantly anthropogenic due to historical land-use changes. To account for heterogeneity in soil nutrient variations across sampling sites, locations with comparable soil profiles were selected based on preliminary surveys.

Stability of soil background values was maintained through long-term (> 10 years) positioning of the agroforestry system. The study was conducted with the prior consent and authorization of the landowner of the Várzea da Onça farm, who granted permission for access to the area and execution of all field procedures before data collection.

The region’s climate is classified as Aw, a tropical climate with dry winters, deep and narrow valleys, and mountainous relief. Annual precipitation averages 721 mm, with the highest rainfall occurring between February and August. The mean annual temperature is 21 °C, and elevation ranges from 736 to 1.100 m⁵⁰.

Two areas were evaluated in the experiment: one with native vegetation and another with coffee plantations intercropped with native vegetation, hereafter referred to as shaded coffee. This system was subdivided into three plots: Upper Third (UT), Middle Third (MT), and Lower Third (LT).

Native vegetation: The experimental area covers 4.57 ha and consists of native tree species ranging from 20 to 35 m in height. Based on the floristic survey carried out, the following species were identified in the regenerating secondary vegetation: Tabebuia serratifolia (pau-d’arco), Anacardium occidentale L. (cashew), Cordia goeldiana (freijó), Calomys tener, Cecropia sp. (embaúba), Spondias purpurea (ciriguela), Ceiba glaziovii (barriguda), Acacia glomerosa, Copaifera langsdorffii (copaíba), Spondias mombin (cajá), Ficus sp. (gameleira), Guazuma ulmifolia (mutamba), Styrax leprosus, and Psidium australe Cambess (aracaeiro).

Additional regionally occurring species included Galezia gorazema Moq., Rubus sp. (Rosaceae), Inga subnuda Salzm. ex Benth., Caesalpinia leiostachya Benth. Ducke, Copaifera trapezifolia Hayne, Roupala cearaensis Sleumer, Cedrela sp., Terminalia sp., Oreopanax capitatum Decne et Planch. var. multiflorum (DC.) E. March, Manilkara rufula (Miq.) Lam, Aspidosperma pyricollum Müll. Arg., and Tabebuia avellanedae Lorentz ex Griseb., according to⁵¹.

Regarding shaded coffee (Fig. 2), the cultivation system consisted of rustic management with minimal forest alteration. The area was 25.59 ha and had been planted for more than ten years with coffee (Coffea arabica L). The percentage of shade provided by native vegetation was approximately 75%. Coffee seedlings were planted within a native vegetation area, with planting holes measuring 0.40 × 0.40 m in width and depth, spaced 2 m apart.

Fig. 2.

Experimental areas of shaded coffee cultivation in an agroforestry system: (a) Upper Third (UT), (b) Middle Third (MT), (c) Lower Third (LT), and (d) schematic representation of the experimental area.

The cultivation methods employed were organic fertilization with cattle manure and poultry litter, thinning with a hand-held brush cutter, and tree pruning. Fertilizers were sourced from local farms, with cattle manure containing approximately 1.5% N, 0.5% P, and 1% K, and poultry litter containing 3% N, 2% P, and 1.5% K. Agronomic practices included manual weeding, no irrigation, and pest control using neem oil sprays⁵². Pest control details followed Bhattarai et al.⁵³. Shade trees at each slope position consisted of similar native species to ensure comparable rhizosphere environments. Canopy density was estimated at 75% shade coverage using hemispherical photography⁵⁴.

Six trenches were excavated in each area, spaced 50 m apart, with dimensions of 1.5 × 1.5 m and a depth of 0.80 m across the three topographic positions. In each trench, three sampling points were established, and disturbed and undisturbed soil samples were collected for physical and chemical analysis at depths of 0–0.20, 0.20–0.40, and 0.40–0.60 m.

Disturbed samples for microbiological analyses were collected at 0–0.05 m depths. The 0–0.60 m sampling interval encompasses the A and upper B horizons, where nutrient dynamics are most active. This depth range was chosen based on literature evidence indicating significant variation in organic carbon and porosity within this range. The three sampling points per trench were collected based on preliminary power analysis and logistical constraints, providing adequate statistical power (> 0.80) for detecting medium to large effect sizes in MANOVA.

Soil granulometric analysis was carried out using the densimeter method⁵⁵. Soil bulk density was measured using the volumetric ring method described by⁵⁶, and total porosity was obtained using the direct method according to⁵⁷.

Pore-size distribution (Eq. 1) was calculated by analyzing the soil water retention curve (SWRC) after determining total porosity. Pore-size classification followed established protocols relating pore diameter to matric potential, with macropores (> 50 μm diameter) corresponding to drainage at low suctions^58,59. This approach is consistent with classical soil physics principles linking pore geometry to water retention behavior⁵⁹.

1

where Pt is the total porosity expressed in cm³ cm^–3, Vp is pore volume in cm³, and Vt is total sample volume in cm³.

Soil macroporosity was obtained at a tension of 1 kPa, as suggested by⁶⁰, according to Eq. 2:

2

where Ma is macroporosity in cm³ cm^–3, Vma is macropore volume in cm³, and Vt is sample volume in cm³.

Mesoporosity was determined after balancing the volume of water retained between 1 and 6 kPa, according to Eq. 3:

3

where Me is mesoporosity in cm³ cm^–3, Vme is mesopore volume expressed in cm³, and Vt is the total sample volume expressed in cm³.

Microporosity was obtained after balancing the volume of water retained between 6 and 1500 kPa, according to Eq. 4:

4

where Mi is microporosity in cm³ cm^–3, Vmi is micropore volume in cm³, and Vt is total sample volume in cm³.

The soil cryptopores were determined using the difference between the weight of the ring-soil set balanced at 1500 kPa and the weight of the ring-soil set dried in an oven at 105 ºC until constant weight, according to Eq. 5:

5

where Crypto is cryptoporosity, in m3 m-3, Vcripto is the cryptopore volume obtained from soil mass differences and converted to volume by dividing by water density (Da), and Vt is total soil volume, obtained by the volume of the volumetric ring (πr²h), expressed in m3.

In the chemical analysis, exchangeable cations (Ca²⁺, Mg²⁺, Na⁺, K⁺ and Al³⁺) were determined following⁵⁷, together with available phosphorus and extractable micronutrients (Mn, Fe, Zn, Cu, B and Mo) as operationally defined by the respective extractants. Potential acidity (H⁺ + Al³⁺), sum of bases (SB), potential cation exchange capacity (T), effective cation exchange capacity (t), base saturation (V) and aluminum saturation (m) were calculated according to⁵⁷. Total organic carbon was measured using potassium dichromate oxidation, and pH was determined in CaCl₂ solution.

Statistical analysis

Data were submitted to multivariate analysis of variance (MANOVA) using a Multivariate General Linear Model (GLM) with bootstrap resampling (n = 1000) to simultaneously evaluate the effects of topographic position (Upper, Middle, Lower Third, and Native Forest), soil depth (0–0.20, 0.20–0.40, and 0.40–0.60 m), and the interaction between these factors on different functional sets of soil physical attributes. A multivariate GLM was selected instead of a traditional ANOVA due to the presence of multiple intercorrelated dependent variables and the greater flexibility of GLM in terms of the shape of the distribution and residuals, being more robust in small or heteroscedastic samples. Bootstrap resampling mitigates overfitting by repeatedly sampling from the observed data to estimate sampling distributions. This reduces the dependence on model assumptions, providing more robust parameter estimates in small-sample contexts.

The model was adjusted using the gamma distribution and the identity link function, selected on the basis of the best performance of the model according to the Akaike Information Criterion (AIC). The assumption of homogeneity of covariance between the groups was met, as indicated by the Box M test. Multivariate normality was considered plausible based on the distribution of residuals and the robustness of the model with resampling.

All main effects and interactions were evaluated using multivariate statistics (Pillai’s trace, Wilks’ Lambda, Hotelling’s trace and Roy’s largest root). The significance of the effects was monitored by estimating the partial effect size (partial η²), interpreted according to⁶¹: small (η² > 0.01), moderate (η² > 0.06) and large (η² > 0.14). When significant main effects or interactions were identified, pairwise multiple comparisons were carried out with Bonferroni correction, which controls type I error in multiple tests and is recommended for samples with moderate variation. Bonferroni correction was preferred over false discovery rate methods because our study involves a moderate number of pairwise comparisons and Bonferroni provides stronger control of family-wise error rate, which is more appropriate for confirmatory analyses with moderate sample sizes.

The dependent variables of the physical attributes included in the models were: soil density, total porosity, macroporosity, mesoporosity, total microporosity and cryptoporosity. For the chemical attributes, the following were considered: pH in CaCl₂, available phosphorus, potassium, calcium, magnesium, sodium, potential acidity (H⁺ + Al³⁺), sum of bases, effective and potential cation exchange capacity, base saturation and aluminum saturation. The set of variables related to soil organic matter (SOM) included: labile carbon, light organic matter, carbon in fulvic acid, humic acid and humin, total carbon, total nitrogen, C/N ratio and the stocks of the respective fractions.

The adjusted means were estimated using the marginal means method, and multiple comparisons were carried out with Bonferroni correction. In addition, independent linear regression models were fitted for each soil depth, with the aim of investigating the intra-stratified variation of the attributes as a function of topographic position. The analyses were conducted in the R environment⁶².

Interactions by network analysis

Relationships among soil attributes were modeled using a partial correlation network estimated through a Gaussian Graphical Model (GGM), regularized with an ℓ₁ (LASSO) penalty and automated model selection based on the Extended Bayesian Information Criterion (EBIC) following⁶³. The tuning parameter γ was set to 0.5 to balance parsimony and model fit, eliminating spurious connections while favoring an interpretable network.

The network incorporated physicochemical and biological soil variables. Nodes represented soil attributes, and edges indicated partial correlations between pairs of variables, adjusted for the presence of all other variables.

To identify which attributes exert the greatest structural influence on the edaphic system, node centrality metrics were evaluated following⁶⁴. The metrics included strength (sum of absolute edge weights), proximity (inverse of the sum of the geodesic distances, indicating accessibility), and intermediation (frequency with which a node lies on minimum paths between variable pairs, indicating its role as a structural link). All centrality indices were standardized as z-scores.

Robustness of network estimates was assessed using nonparametric bootstrap resampling (n = 1000) to calculate confidence intervals for edges and centrality estimates, along with case-dropping bootstrap for stability evaluation.

The centrality stability coefficient (CS-coefficient) was calculated to determine the maximum proportion of cases that could be excluded without compromising, with 95% confidence, the stability of centrality rankings. Values greater than 0.25 were considered acceptable and values greater than 0.50 desirable, as recommended by⁶⁵.

Principal component analysis

The Principal Component Analysis (PCA) was used to synthesize variability and identify patterns associated with positions and soil depths. Variables were standardized (Z-score), and component retention followed the Kaiser rule (eigenvalues > 1), with Varimax rotation. Factor loadings |≥ 0.40| were considered significant. Visualization employed biplots of the first two components with 95% confidence ellipses for each position and depth. Analyses were conducted in R version 4.3.1⁶² using the RStudio IDE with FactoMineR, factoextra, and ggplot2 packages.

Results and discussion

Soil chemical attributes

For chemical attributes, MANOVA detected statistically significant main effects of topographic position, soil depth, and their interaction, indicating that spatial distribution in soil chemical attributes was influenced by both slope position and vertical profile stratification. The attributes evaluated included pH in CaCl₂, exchangeable aluminum (Al³⁺), potential acidity (H⁺ + Al³⁺), sum of bases (SB), cation exchange capacity at pH 7.0 (CEC), base saturation (V%), and aluminum saturation (m%).

Multivariate statistics were robust, with Pillai’s Trace = 2.436; F(18, 99) = 23.77; p < 0.001 and Wilks’ Lambda = 0.0002; F(18, 88) = 94.00; p < 0.001, indicating that the combined variation in chemical attributes differed significantly among the analyzed factors. The magnitude of Pillai’s Trace suggests a substantial proportion of variance in the dependent variables explained by the predictors, while the extremely low value of Wilks’ Lambda denotes minimal unexplained multivariate variance.

From an edaphic perspective, these results imply that toposequence position modulates chemical fertility gradients, with upper slope positions often exhibiting higher pH, lower Al³⁺ activity, and greater base saturation due to better drainage and reduced leaching of base cations⁶⁶. In contrast, lower slope positions tend to accumulate exchangeable aluminum and potential acidity, reflecting colluvial deposition and possibly greater water saturation, which may intensify base leaching and favor acidification⁶⁷. The significant interaction between slope position and depth reinforces that nutrient distribution patterns and acidity dynamics are not uniform across the profile, but rather shaped by the interplay between lateral and vertical transport processes in the landscape⁶⁸.

Large effect sizes for topographic position (F(18, 99) = 23.391; p < 0.001; partial η² = 0.810) and depth (F(12, 64) = 14.250; p < 0.001; partial η² = 0.728), combined with a significant interaction (F(36, 216) = 5.23; p < 0.001; partial η² = 0.466), indicate that vertical processes, such as base cation leaching, Al³⁺ mobilization, and organic matter mineralization, modulate the chemical contrasts imposed by topography. These results are in line with the conceptual models proposed by Lu et al.⁶⁹, who demonstrated that hydropedological gradients in agroecosystems influence soil functioning at the physicochemical and microbiological levels.

Specifically, zones with better drainage and greater organic matter retention exhibited enhanced representation of genes involved in nitrification, phosphatase activity, and organic matter decomposition, while hydromorphic zones with prolonged saturation showed reductions in these functions along with increases in genes linked to anaerobic pathways³³. In the present study, the contrasts between upslope and downslope positions in pH, CEC, and base saturation correspond to similar changes in microbial functional potential, reinforcing that topography-driven chemical patterns have cascading effects on biogeochemical processes and nutrient turnover⁷⁰.

At the univariate level, pH in CaCl₂ (Fig. 3) showed statistically significant differences with a high power of effect for the Upper Third (UT) in relation to the Middle Third (MT) (ΔM = 1.080; 95% CI [0.769; 1.391]; p < 0.001; d = 1.488), suggesting lower acidity in the highest slope positions. This was accompanied by a significant reduction in exchangeable Al³⁺ in UT compared with MT (ΔM = –1.800; 95% CI [–2.037; –1.563]; p < 0.001; d = –21.228) and in potential acidity (H⁺ + Al³⁺) (ΔM = –3.147; 95% CI [–3.603; –2.692]; p < 0.001; d = –19.282). Mechanistically, these patterns are consistent with reduced hydromorphic stress and improved aeration at upper slope positions, which limit Al³⁺ solubilization and acidification.

Fig. 3.

Adjusted means of soil chemical attributes by depth and topographic position, with horizontal error bars representing 95% confidence intervals of the marginal means.

Elevated positions also tend to accumulate more stable organic matter and reactive clay minerals, increasing the density of negative charges and enhancing cation retention^71,72. According to Długosz & Piotrowska-Długosz⁷³, this combination of organic and mineral surface activity not only elevates CEC but also sustains enzymatic activities involved in C and N turnover, reinforcing the chemical buffering capacity of these zones.

Consistently, UT exhibited significantly higher effective CEC (t) and potential CEC (T = SB + H⁺ + Al³⁺), with ΔM = 3.427 (95% CI [2.627; 4.228]; p < 0.001; d = 1.870) and ΔM = 4.775 (95% CI [4.075; 5.475]; p < 0.001; d = 1.803), respectively. These results may reflect a greater abundance of colloidal complexes and organo-mineral associations that stabilize negative charges and enhance base cation retention. Sujatha et al.⁷¹ and Yang et al.⁷⁴ reported similar positive relationships between CEC, organic carbon, and clay content, with elevated topographic zones favoring the accumulation of both.

The positive difference in base saturation (V%) in UT compared to MT (+ 48.53%; CI 95% [43.050; 54.020]; p < 0.001; d = 1.783) and the lower aluminum saturation (m%) (ΔM = –57.88%; 95% CI [–66.000; –49.775]; p < 0.001; d = –19.921) confirm that topography strongly exerts control over the composition of the exchange complex. In upper slopes, lower susceptibility to seasonal water saturation maintains aerobic conditions, enhancing nutrient cycling and exchangeable base retention⁷⁵. However, even in these more favorable zones, m% values exceeding 20% indicate persistent acidity constraints typical of highly weathered tropical soils.

In contrast, lower slope positions are likely affected by seasonal hydromorphia, which creates reducing conditions that mobilize Fe and Mn, increase Al³⁺ solubility, and immobilize phosphorus through precipitation or adsorption onto reduced iron surfaces. Crespo et al.⁷⁶ documented similar patterns in hillside systems in northeastern Brazil, whereas Lu et al.⁶⁹ demonstrated that such hydromorphic zones often exhibit reduced abundance of microbial functional genes associated with phosphorus solubilization and nitrification, decreasing nutrient bioavailability. In these contexts, management strategies such as targeted liming and controlled drainage could mitigate acidity and restore nutrient cycling efficiency⁷⁷.

Differences between UT and MT persisted at 0.20–0.40 m for pH in CaCl₂ (ΔM = 1.147; 95% CI [0.837; 1.458]; p < 0.001; d = 1.930), suggesting greater buffering capacity at elevated positions. This buffering effect may be due to stable organo-mineral complexes with high ligand-binding capacity that slow acid cation mobility⁷⁴. At 0.40–0.60 m, although mean pH remained slightly higher in cultivated upper slope soils, differences were not statistically significant (p > 0.05). This reflects a tendency toward chemical homogenization with depth, where ionic diffusion and vertical percolation dominate. Patrick et al.⁷⁸ attributed similar depth patterns to gradual downward migration of alkalinity from surface horizons.

CEC values (t and T) declined sharply with depth, with maxima at 0–0.20 m. At this depth, UT exhibited statistically higher values than MT, indicating a greater concentration of negatively charged colloids, likely linked to higher organic matter inputs and reduced erosion losses. These patterns are consistent with conservation-oriented systems where surface stability enhances nutrient retention and microbial biomass N⁷³. Potential acidity and Al³⁺ were also significantly lower at 0–0.20 m in UT relative to MT. In deeper layers, the trend toward lower acidity at upper positions persisted, even when not statistically significant, indicating weaker leaching and slower accumulation of exchangeable acidity in these profiles.

In the surface horizon, higher V% and lower m% in UT indicate a chemically more favorable environment, with greater availability of base cations and reduced Al³⁺ toxicity risk for roots⁷⁹. Eshetu et al.⁸⁰ emphasize that such favorable surface chemistry can amplify nutrient-use efficiency and support microbial functional networks. Similarly, Lu et al.⁶⁹ reported that high CEC and balanced base saturation in agroforestry soils correlate with denser and more resilient microbial co-occurrence networks.

Physical attributes of the soil

For the physical attributes of the soil, the MANOVA results indicated statistically significant effects of topographic position, depth and the interaction between the two. The multivariate model showed significant values in the Pillai and Wilks’ tests, with Pillai’s Trace = 1.541; F(12, 96) = 4.830; p < 0.001 and Wilks’ Lambda = 0.092; F(12, 96) = 5.32; p < 0.001, confirming significant multivariate differences between the different positions along the toposequence.

Specifically, the topographic position factor had a significant effect on the set of physical attributes (F(16, 96) = 4.83; p < 0.001; partial η² = 0.726), indicating that the spatial distribution of the soil’s physical structure can be modulated by the relief. The soil depth factor also significantly influenced the physical attributes (F(24, 96) = 2.48; p = 0.006; partial η² = 0.341), although to a lesser extent. The interaction between topographic position and depth was also significant (F(48, 96) = 2.112; p = 0.001; partial η² = 0.340).

At the univariate level, statistically significant effects in the different topographic positions were observed for microporosity between coarse aggregates (F(3, 36) = 22.06; p < 0.001; partial η² = 0.648), cryptoporosity (F(3, 36) = 16.396; p < 0.001; partial η² = 0.577) and total microporosity (F(3, 36) = 13.063; p < 0.001; partial η² = 0.521), all with high effect sizes (partial η² > 0.503). Other attributes also showed statistically significant effects as a function of the topographic position, with moderate to high effects: soil density (F(3, 36) = 8.65; p < 0.001; partial η² = 0.419), total porosity (F(3, 36) = 7.29; p = 0.001; partial η² = 0.378) and macroporosity (F(3, 36) = 6.181; p = 0.002; partial η² = 0.340).

These findings indicate that smaller diameter porous fractions are particularly sensitive to position in the landscape, due to variations in soil structural stability, intensity of use and organic matter dynamics. This relationship is corroborated by Sarmiento-Soler et al.⁸¹, who showed that microporosity varies significantly depending on the vegetation cover and the relief, with greater stability in shaded areas and at high altitudes—highlighting the direct role of topography and management in the structural organization of fine pores. In addition, in toposequences, soils in elevated positions tend to exhibit less accumulation of fine particles, greater drainage, intensified biological activity and high levels of organic matter⁸². These conditions, combined with less anthropogenic disturbance, favor not only the preservation of the physical structure, but also the development of more functionally efficient porosity.

These characteristics contribute to greater pore connectivity and aggregate integrity, especially in soils under perennial vegetation, where root activity plays a fundamental role in maintaining soil architecture⁸³. In addition, long-term organic fertilization promotes greater pore structure stability and pore connectivity, in contrast to management based on inorganic inputs⁸⁴.

The distribution of pore diameter is also impacted by the content and quality of organic matter, modulating the retention and movement of water and gases in the soil⁸⁵, while root growth tends to be more efficient in soils with greater macroporosity and lower penetration resistance, generally present in less compacted environments with greater structural stability (Zhou et al.⁸⁶). As for the effects of stratification, no statistically significant differences were observed for the physical attributes analyzed: soil density (F(2, 36) = 0.14; p = 0.866; partial η² = 0.008), total porosity (F(2, 36) = 0.81; p = 0.454; partial η² = 0.043) and macroporosity (F(2, 36) = 0.11; p = 0.900; partial η² = 0.006) showed no significant variation along the profile.

As for the Lower Third (LT), it showed statistically significant values for total porosity when compared to the Native Forest (NF), with an average difference of ΔM = −5.360 cm³ cm⁻³ (95% CI [−9.410; −1.310], p = 0.006; d = −0.920), which represents a reduction of approximately 10.5% in relation to the NF (Fig. 4). This result may be related to a restriction in total pore volume in the lower elevation areas, which is consistent with studies that identify greater compaction, lower porosity and lower biological activity in the lower positions of toposequences, which are generally subject to the deposition of fine particles and sediments³⁴. In addition, studies show that low-altitude areas have lower stocks of organic matter and lower quality litter, resulting in greater susceptibility to structural compaction⁸⁷.

Fig. 4.

Physical attributes of the soil in an agroecological system of shaded coffee, in different positions of the toposequence, Taquaritinga do Norte, Pernambuco state, Brazil.

In terms of macroporosity, the comparison between LT and NF indicated a mean difference of ΔM = −1.940 cm³ cm⁻³ (95% CI [−4.550; 0.660], p = 0.204; d = −0.920), representing a relative reduction of approximately 31.1%. Although not statistically significant, this reduction is consistent with the tendency toward structural compaction in lower-slope positions, where geotechnical pressure and reduced input of structuring organic residues can decrease aggregate stability⁸⁸.

Similar patterns were reported by Medeiros et al.⁸⁹, who observed decreases in void ratio and increases in bulk density under progressive applied pressures, indicating the susceptibility of surface layers to compaction. These findings reinforce the importance of management practices that enhance organic residue inputs, such as no-till systems, which, however, may require periodic mechanical interventions (e.g., scarification) to mitigate compaction over time⁹⁰. Furthermore, long-term adoption of no-till without corrective interventions can lead to subsurface compaction, as evidenced by Steponavičienė et al.⁹¹, necessitating subsoiling to restore structural quality and maintain productive sustainability.

Microporosity between aggregates was significantly higher in LT compared to NF (ΔM = 2.790 cm³ cm⁻³; 95% CI [0.520; 5.050], p = 0.011; d = 1.638), reflecting a relative increase of approximately 23.8%. This pattern may be related to changes in intermediate pore connectivity, possibly resulting from the reorganization of structural pores into isolated microcapillaries, limiting the preferential flow and redistribution of water in the profile. These findings are in line with observations by Nunes et al.⁹², who identified structural improvements in the pore distribution of soils under shaded coffee, especially in the meso and macro pore fractions, in contrast to open areas prone to compaction and pore segregation.

Principal component analysis—PCA

Principal Component Analysis (PCA) was applied to identify multivariate patterns between macronutrients (NT, K, Ca, Mg, TOC, C/N, P), evaluating the effects of topographic position. The first two components (PC1 and PC2) cumulatively explained 75.58% of the total variance (PC1: 56.86%; PC2: 18.72%), revealing distinct edaphic gradients.

PC1 emerged as the primary fertility axis, characterized by significant negative factor loadings for exchangeable bases (Ca²⁺: −0.443; Mg²⁺: −0.453; K⁺: −0.453) and total organic carbon (TOC: −0.392), with high spatial representativeness (cos² > 0.611 for Ca, Mg, TOC). This component clearly discriminated preferential deposition in the Lower Third (LT), where samples exhibited negative coordinates (PC1 = −4.05 ± 0.54), aligning with the hydrological redistribution model of fine sediments and organic matter. In contrast, the Middle Third (MT: PC1 + 0.32 ± 0.12) and Upper Third (UT: PC1 −1.16 ± 0.19) showed lower base saturation and organic content, consistent with erosion processes on convex slopes.

PC2 was an independent geochemical axis, dominated by available phosphorus (P) with a maximum negative factor load (−0.631) and high representation (cos² = 0.522). The segregation of P in relation to other nutrients (K⁺ = 0.241 in PC2) reflects its unique dynamics: low mobility due to preferential adsorption on Fe/Al oxides (constant > 10³ L-kg-^193,94;) and complexation with carboxylic groups of organic matter, ligand-promoted dissolution effect⁹⁵.

This selective retention is confirmed by the unstratified topographic distribution with intra-tract variability > 40%, de-characterizing the dominant hydrological control observed in PC1. In addition, the C/N ratio showed a distinct signature in PC2 (cos² = 0.137 vs. 0.081 in PC1), suggesting an association with local biogeochemical processes. The distribution of samples in multivariate space revealed patterns consistent with topographic influence.

The samples from the lower third (LT) were concentrated in regions with more negative PC1 values, indicating a greater accumulation of total nutrients, particularly exchangeable bases (Ca²⁺, Mg²⁺, K⁺) and organic carbon. This result is consistent with the convergence effect of fine materials and organic matter in the lower positions of the toposequence. In turn, the samples from the Middle Third (MT) and Upper Third (UT) tended to be distributed in regions with more positive PC1 values, suggesting lower nutrient accumulation and relative fertility.

The separation of phosphorus from other nutrients may be related to its low mobility in the soil and strong adsorption in specific mineral fractions, such as iron and aluminum oxides^93,94, as well as complexation with functional groups in organic matter. Therefore, PC2 does not represent a simple axis of organic matter quality, as previously suggested, but rather a geochemical gradient regulated by specific interactions between phosphorus and the edaphic matrix.

For the physical attributes (Fig. 5b), Principal Component Analysis (PCA) revealed that the first two axes (PC1 and PC2) cumulatively explained 60.19% of the total variance (PC1: 38.46%; PC2: 21.74%). PC1 defined a gradient of soil structural organization, characterized by pronounced negative factor loadings for Total Porosity (−0.637) and Macroporosity (−0.527), contrasting with a positive contribution from Total Microporosity (0.559).

Fig. 5.

Principal component analysis (PCA) projection of the samples on the plane defined by the first two principal components, (a) considering the soil macronutrients and (b) considering the soil physical attributes.

This result shows a combination of macropores and micropores, as evidenced by the high representativeness (cos²) of these variables in the Total Porosity (0.786) and Total Microporosity (0.604) components. Negative PC1 values were associated with soils with open porous architecture and could be attributed to the dominance of macropores, while positive values indicated a predominance of functional micropores for water retention, in line with the dual functionality model of porosity (Medeiros et al.⁸⁹).

The second component (PC2) shows a gradient of mineral compaction and intra-particle porosity, with a significant positive factor load for Particle density (0.801) and negative for Textural microporosity (−0.551). This axis shows a differentiation between soil fractions with a higher content of heavy minerals and smaller volumes of internal micropores, typical of materials with less aggregation or a more compact structure.

The representation of particle density (cos² = 0.697) and textural microporosity (cos² = 0.330) in PC2 indicates that this component synthesizes fundamental contrasts in the structural organization of the soil, reflecting both the influence of the mineral matrix and the dynamics of fine pores. These results are in line with the findings of Dexter⁹⁶, who demonstrated the intrinsic relationship between the particle density and the porosity architecture in mineral soils.

The expression of textural microporosity on this main axis corroborates the observations by Pires et al.⁹⁷, who highlight the determining role of pores < 30 μm in water retention and structural stability. The asymmetry in the magnitudes of cos² (greater for density than for microporosity) suggests, as proposed by Rabot et al.⁹⁸, that mineralogical attributes often exert greater control over the overall structural variability of the soil than textural parameters of the micro-scale porosity.

In multivariate space, the distribution of samples revealed patterns which are consistent with topographic position. The samples from the Upper Third (UT) were concentrated in regions with more negative PC1 and PC2 values, indicating greater macroporosity, greater total porosity and lower particle density. This behavior may be compatible with conditions of lower compaction and greater structural stability, favored by the higher relief, better drainage and lower degree of anthropogenic disturbance.

On the other hand, the samples from the Lower Third (LT) were predominantly located in regions with positive PC1 and especially PC2 values. These values reflect environments with higher mineral density and fewer structural macropores. However, the higher levels of total organic carbon (TOC) observed in the LT suggest that the colluvial accumulation of organic matter contributes to the development of water retention micropores, justifying its association with total microporosity in the PC1 component.

This pattern may be the result of a functionally distinct porous organization, modulated by processes of redistribution of organic and mineral material in areas of lower altitude, often subject to hydromorphism and accumulation of fine particles.

These results indicate that the physical structure of the soil along the toposequence can be modulated both by processes related to the position on the slope and the deposition of organic matter, which affects the distribution and function of the porosity on different slopes, management systems and morphopedological contexts⁹⁹. This topographic influence acts not only on the contribution of organic matter, but also on the degree of development of the profiles and the connectivity between the structural pores. Soils located on flatter surfaces, associated with sedimentary morphopedological compartments, tend to have a greater effective depth, continuous porosity and a more stable structure, while areas of greater slope and crystalline relief show shallower profiles, susceptible to compaction and porous disconnection¹⁰⁰.

In addition, the chemical attributes showed gradients of fertility and accumulation of organic matter also influenced by topographic position, with a higher concentration of nutrients in the lower segments of the slope, as observed by Zhou et al.⁶⁷.

Structure of edaphic interactions by network analysis

A visual inspection of Fig. 6 shows partial correlation networks that assess the conditional co-occurrence structure between soil physical, chemical and biological attributes across topographic positions, where blue edges indicate positive partial correlations and red edges indicate negative partial correlations, and edge thickness is proportional to the absolute association strength.

Fig. 6.

Interaction networks between soil physicochemical attributes in different positions of the toposequence, with panels (a) Upper Third, (b) Middle Third, (c) Lower Third and (d) Native Forest, where blue edges indicate positive partial correlations and red edges indicate negative partial correlations, and edge thickness is proportional to the absolute association strength.

In the Upper Third (Fig. 6a), the network analysis reveals a modular architecture with high functional integration among chemical attributes, characteristic of well-drained soils at higher topographic positions. The formation of a cohesive cation exchange core stands out, in which effective CEC (t) shows strong interdependence with exchangeable calcium (Ca²⁺; weight = 0.600) and potassium (K⁺; weight = 0.408), reflecting synergy in ion retention processes mediated by the permanent charge of clay soils (Brady and Weil¹⁰¹). This electrostatic cohesion suggests greater efficiency in fertility mechanisms, with chemical stability favored by drainage (Fontes¹⁰²).

This behavior may be related to total organic carbon (TOC) acting as a key integrating element, thus establishing positive connections with Ca²⁺ (weight = 0.268) through calcium-bridge complexation¹⁰³, association with pH (weight = 0.031) due to the buffering capacity of organic matter, and a negative relationship with K⁺ (weight = –0.044) due to competition for adsorption sites in 2:1 clays (Sparks¹⁰⁴).

The absence of direct links between TOC and exchangeable aluminium (Al³⁺) or phosphorus (P) indicates functional compartmentalization, suggesting that (i) aluminium toxicity is mitigated by high base saturation independent of organic matter, and (ii) phosphorus dynamics operate in subsystems dissociated from organic cycling¹⁰⁵. This pattern reflects the vertical stratification of biogeochemical processes in toposequences, where oxidative conditions favor the specialization of edaphic pathways (Lal¹⁰⁶).

In the Middle Third (Fig. 6b), cohesion of the chemical core persisted, with significant connections between TOC and Ca²⁺ (0.438), TOC and base saturation (V%) (0.376), and V% and aggregate microporosity (0.371). The interaction between TOC and total N was evidenced by the correlation between TOC and K⁺ (0.614), indicating an association between organic stocks and essential cation availability. The BSR, represented indirectly through TOC and pore attributes, maintained low connectivity. Negative relationships were evident, such as Al³⁺–K⁺ (–0.229), signaling chemical antagonism between acidity and nutrient availability in intermediate slope environments.

In the Lower Third (Fig. 6c), there is a reduction in the network density, reflected by greater topological dispersion and reduced inter-nodal cohesion compared to the Upper and Middle Thirds. Despite this general destructuring, the core of chemical interactions maintained its integrity, as evidenced by strong associations between Cation Exchange Capacity (CEC) and organic matter content (TOC, weight = 0.134) and between CEC and microporosity (weight = 0.181).

TOC exhibited differential connectivity, showing moderate associations with exchangeable calcium (Ca²⁺, weight = 0.390) and pH (weight = 0.183), but no connections with available phosphorus (P) and exchangeable aluminum (Al³⁺).

This pattern may demonstrate functional fragmentation of the edaphic processes at lower altitudes, where traditional fertility mechanisms, P cycling, and Al³⁺ neutralization are dissociated from organic matter dynamics^107,108. Exchangeable potassium (K⁺) showed topological isolation, without significant connections with acid–base variables (pH, Al³⁺), structural attributes (microporosity), or organic matter indicators (TOC).

This structural disconnection indicates that K⁺ dynamics are decoupled from main edaphic controls in the landscape, reflecting intrinsic restrictions of hydromorphic environments. This structural isolation of potassium may be associated with edaphic processes characteristic of hydromorphic environments: water saturation promotes intensified leaching, drastically reducing K⁺ retention in the soil matrix¹⁰⁹, while ionic competition at exchange sites allows Al³⁺ and H⁺ to suppress potassium adsorption^110,111. In addition, microbial activity is further compromised, limiting the biogeochemical cycling of the nutrient (Fageria and Stone¹¹²).

For the Native Forest (Fig. 6d), the network structure displayed higher connection density and topological complexity. The core involving CEC (t), Ca²⁺, and base saturation (V%) was maintained, with edge weights of 0.434, 0.625, and 0.581, respectively. TOC was highly connected to P (0.581), Ca²⁺ (0.390), and Al saturation (–0.442), suggesting greater integration between the chemical functionality and organic matter quality in systems under perennial cover.

Also, for the Native Forest, the negative edge between Al³⁺ and K⁺ (weight = –0.229) may be related to a functional antagonism relationship: in contexts of greater acidity (greater presence of Al³⁺), exchangeable K⁺ availability tends to be lower. This association may be related to competition between cations for the exchange complex and to aluminum toxicity, which can inhibit root uptake of K⁺, as observed in acid tropical systems.

Conversely, the positive link observed between Al³⁺ and Ca²⁺ (weight = 0.398) suggests that, in certain conserved edaphic contexts, functional coexistence of acidic and basic cations is possible. In experiments related to similar management, this association may result from joint adsorption processes on colloidal surfaces or simultaneous release during mineral weathering. Previous studies have shown that ions such as Ca²⁺ and Al³⁺ can share adsorption sites on clays, manganese oxides, and aluminosilicates, indicating synergistic interaction on reactive surfaces^113–115. In addition, coprecipitation mechanisms or ionic exchanges involving Ca–Al-containing compounds, such as lamellar double hydroxides, may contribute to this functional association¹¹⁶.

The positive connection between pH and available phosphorus (P) (weight = 0.085) reflects the known dependence of P availability on soil acidity¹¹⁷. At higher pH (less acidic), phosphorus fixation by iron and aluminum oxides is reduced, increasing its concentration in the soil solution¹¹⁸. Although moderate in weight, this edge indicates a persistent relationship even after controlling for Al³⁺, TOC, and Ca²⁺ effects, denoting a conditional association between pH and P, characteristic of chemically stable soils under perennial vegetation³⁴.

The higher connections observed in the upper segments and in the native vegetation indicate greater structural cohesion and functional capacity, whereas the Lower Third networks reveal vulnerability to fragmentation, likely induced by hydromorphia, compaction, or surface runoff³⁴. The connectivity between TOC and Ca²⁺, and its absence with Al³⁺, reinforces the hypothesis that organic matter acts as a structural axis for chemical interactions in conserved environments, as also suggested by Mabagala and Mng’ong’o¹¹⁹.

As shown in the weighting matrix (Fig. 7), the Upper Third (UT) and Native Forest (NF) exhibited predominant connections among key variables such as Total Organic Carbon (TOC), Calcium (Ca²⁺), Base Saturation (V%) and Effective CEC, which indicates a more integrated and functional chemical structure. These connections support the central role of organic matter as a mediator of chemical and structural interactions, possibly favored by greater inputs of plant residues and reduced mechanical disturbance in these areas.

Fig. 7.

Centrality measures for soil physicochemical attributes in networks constructed for different toposequence positions: Upper Third (red), Middle Third (green), Lower Third (blue), and Native Forest (purple). The metrics presented are: Closeness, Betweenness, Strength, and Expected Influence. Note: Centrality metrics were computed for each network and standardized (z-score) within each topographic position to allow comparison across variables. Colors represent topographic positions: UT = Upper Third, MT = Middle Third, LT = Lower Third, NF = Native Forest. Metrics shown are Closeness, Betweenness, Strength, and Expected Influence.

In contrast, the network in the Lower Third (LT) showed reduced connection density and a predominance of weaker associations, especially involving TOC, Ca²⁺, and macroporosity, indicating a possible functional fragmentation of the attribute network. This disconnection may be attributed to processes such as hydromorphy, surface compaction, and lateral runoff, which reduce the continuity between physical and chemical processes, compromising soil structural functionality.

In addition, there was little or no connection between TOC and Al³⁺ in all four positions, indicating a functional disconnection between organic matter and potentially toxic ionic species, consistent with the stabilizing role of organic matter in conserved environments. This pattern reinforces that, even under conditions of partial degradation, organic matter tends to maintain functional selectivity in interactions with cations, prioritizing base cations such as Ca²⁺ and suppressing bonds with Al³⁺, as evidenced by previous studies.

Conclusions

This study demonstrates that topographic position is a primary driver of soil physical and chemical heterogeneity in shaded coffee agroforestry systems, with slope position explaining over 70% of soil variability. Upper slope positions exhibited superior structural quality, characterized by higher macroporosity and lower bulk density, whereas lower slope positions showed organic carbon accumulation and dominance of microporosity, reflecting distinct pedological processes along the toposequence.

Network analysis identified total organic carbon (TOC) as the central integrator of soil functions, displaying the strongest connectivity with base saturation, cation exchange capacity, and structural attributes across all topographic positions. This finding reinforces the role of TOC as a multifunctional regulator in hillside coffee systems, influencing not only nutrient cycling but also physical structure and chemical buffering capacity.

These results have direct implications for coffee agroforestry management. Farmers should consider topographic position when implementing soil conservation practices, with erosion control measures prioritized in upper slopes and organic matter management emphasized in lower slopes. The identification of key soil indicators through network analysis provides a framework for efficient soil monitoring programs, focusing resources on parameters with the greatest influence on overall soil functionality. The shaded agroforestry approach demonstrated measurable advantages over conventional systems, particularly in maintaining cation exchange capacity and reducing aluminum toxicity in upper slope positions.

Study limitations include the single-site scope, cross-sectional design, and absence of direct measurements of microbial functional dynamics. Future research should expand to multiple sites and climatic zones, incorporate microbial profiling to elucidate biological mechanisms underlying TOC-mediated processes, evaluate economic feasibility of topography-specific interventions, and assess system resilience under projected climate change scenarios. These advances would strengthen the scientific foundation for precision management in tropical coffee agroecosystems.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

C.M.G.C. (Cristiane Maria Gonçalves Crespo) – Investigation; Data curation; Formal analysis; Writing – original draft. V.C.P. (Victor Casimiro Piscoya) – Conceptualization; Methodology; Supervision; Writing – review & editing. R.C.P.M. (Robson Carlos Pereira de Melo) – Investigation; Fieldwork; Data curation. L.M.P. (Ludmilla Morais Pereira) – Formal analysis; Visualization; Writing – review & editing. L.D.V.S. (Luiz Diego Vidal Santos) – Conceptualization; Methodology; Statistical analysis; Writing – original draft; Writing – review & editing; Project administration. F.S.R.H. (Francisco Sandro Rodrigues Holanda) – Funding acquisition; Resources; Supervision. A.P. (Alceu Pedrotti) – Resources; Supervision; Writing – review & editing. S.A.B.S. (Savanna Alice Botelho da Silva) – Investigation; Laboratory analyses; Data curation. M.C.F. (Moacyr Cunha Filho) – Formal analysis; Statistical modeling; Validation. R.N.A.F. (Renisson Neponuceno de Araújo Filho) – Conceptualization; Methodology; Supervision; Writing – review & editing.

Funding

This work was supported by the Coordination for the Improvement of Higher Education Personnel (CAPES) and the National Council for Scientific and Technological Development (CNPq).

Data availability

The datasets generated and/or analysed during the current study are available in the Open Science Framework (OSF) repository at 10.17605/OSF.IO/CY5DU.

Declarations

Competing interests

The authors declare no competing interests.