Assessment of Soil Quality in Peruvian Andean Smallholdings: A Comparative Study of PCA and Expert Opinion Approaches

Abstract

Soil degradation poses a significant threat to the sustainability of agroecosystems, particularly in mountainous regions where environmental conditions are highly variable and management practices are often suboptimal. In this context, soil quality assessment emerges as a key tool for guiding sustainable land use and informing decision-making processes. This study aimed to develop and spatially evaluate a Soil Quality Index (SQI) tailored to the northeast sector of Jangas district, Ancash, Peru. A total of 24 soil indicators were initially considered and reduced using Spearman’s correlations to avoid multicollinearity. Depending on the weighting strategy applied, the final SQI configurations incorporated between 14 and 15 indicators. Two weighting strategies—Principal Component Analysis (PCA) and Expert Opinion (EO)—were combined with linear and non-linear (sigmoidal) scoring functions, resulting in four distinct SQI configurations. The spatial performance of each index was tested using Geographically Weighted Regression Kriging (GWRK), incorporating covariates like NDMI, elevation, slope, and aspect. The SQI constructed using PCA combined with non-linear scoring achieved the highest performance, effectively minimizing skewness and while achieving the highest predictive accuracy under GWRK. By contrast, although the EO-based index with linear scoring demonstrated similar statistical robustness, it failed to achieve comparable effectiveness in terms of spatial predictive accuracy. The SQIs generated offer a practical framework for local institutions to identify and prioritize areas requiring intervention. Through the interpretation of complex soil data into accessible, spatially explicit maps, these indices facilitate the targeted application of inputs—such as organic amendments in low-SQI zones—and support the implementation of improved management practices, including crop rotation and soil conservation, without necessitating advanced technical expertise.

1. Introduction

Soil quality is defined as “the capacity of a soil to function within ecosystem boundaries, to sustain biological productivity, maintain environmental quality, and promote plant, animal, and human health” [1]. SQIs function as a decision-making instrument by enabling the assessment of the sustainability and impact of soil management practices [2,3,4]. They also facilitate the systematic organization of commonly evaluated soil parameters, shifting the focus of soil quality from solely productivity to a more comprehensive approach centered on sustainable soil management [5]. The application of an SQI as a tool for assessing management practices has been explored across various spatial scales, including crop-level, experimental field, and regional contexts [6,7,8,9]. This approach is valuable because isolated measurements of soil properties may offer limited practical relevance to farmers when interpreted without contextual understanding. A comprehensive evaluation of soil conditions can empower farmers to devise effective management strategies for optimizing crop production [10].

Soil quality indices and indicators should be carefully selected to ensure their congruence with soil functions that cannot be directly measured [11]. The selection of indicators must be aligned with the specific soil functions of interest and the defined management objectives of the system to ensure a comprehensive evaluation of soil health and functional performance [12]. Developing a reliable SQI requires first identifying the most relevant indicators and then assigning weights that reflect their relative importance in representing particular soil functions [13]. In most cases, the formulation of an integrated SQI follows three main stages: determining the set of indicators, transforming these indicators into scores, and combining the scores to produce a single index value [8].

Soil properties frequently exhibit interdependencies; however, under diverse agricultural management practices, their influence on soil quality and productivity may vary significantly, complicating the interpretation of their effects across different management regimes [14]. An effective soil quality indicator should exhibit strong correlations with essential soil functions. In addition, it should be scientifically robust, methodologically precise, readily measurable, and responsive to variations in management practices [15]. In addition, their selection needs to reflect site-specific conditions, including geographic location, climatic regime, and management goals [16].

Given the collinearity of many soil attributes and the high demands of time and resources for exhaustive sampling, it becomes essential to define a Minimum Dataset (MDS) that preserves the most relevant information while eliminating redundancy [17,18]. Traditionally, the development of such datasets has relied heavily on expert judgment and statistical approaches, like PCA, to identify the most representative indicators [19]. A soil function–based approach incorporating EO offers a practical and reliable method for regional-scale assessments, provided that the derived SQI is validated against parameters aligned with specific management objectives [20]. On the contrary, a significance constraint of the PCA approach is its tendence to exclusively consider those Principal Components (PCs) that account for at least 5% of the total variance and have eigenvalues greater than 1.0. Consequently, this criterion may lead to the exclusion of important parameters that are critical for addressing specific management objectives [9].

In the context of SQI development, once the MDS is established, the selected indicators must be standardized to a common scale before integration into the index. Among the most widely used transformation methods are linear and non-linear (sigmoidal) scoring functions. The linear approach assumes a constant rate of change between an indicator and soil quality, offering simplicity and ease of interpretation, although it may fail to reflect threshold effects and non-linear ecological responses [21]. On the other hand, the sigmoidal approach models gradual changes at both low and high indicator values, with greater sensitivity near optimal ranges, thereby capturing ecological thresholds and diminishing returns more effectively [10]. While the latter can provide a more ecologically representative assessment, it demands additional parameterization and may be less intuitive for non-specialist users.

The choice of scoring function is therefore critical, as it not only determines how raw indicators are normalized—helping to mitigate issues such as non-normality and skewness [22]—but also influences the performance of subsequent spatial analyses. This becomes particularly important when SQI computation is integrated with spatial interpolation techniques, such as GWRK, which leverage the spatial structure of residuals to refine predictions [23,24] and generate high-resolution soil quality maps for more targeted land management decisions.

Traditional agricultural systems in the Peruvian Andes are highly vulnerable, largely due to the intrinsic fragility of their soils—marked by low fertility, high susceptibility to erosion, and limited water-holding capacity—conditions that are further aggravated by steep topography and intensive land use [25,26]. Climate change intensifies these challenges by increasing the frequency and severity of extreme events—such as droughts and frosts—that disrupt hydrological dynamics and exacerbate soil degradation processes [27]. Recent research indicates a progressive decline in soil organic matter and increased compaction across Andean landscapes, posing significant risks to the food security of Andean communities [28,29]. Smallholder farmers, who typically cultivate plots smaller than two hectares, often rely on empirical indicators—such as soil color and texture—to guide their management decisions, thereby limiting the integration of scientific knowledge into sustainable land use strategies.

Developing an SQI enables a comprehensive assessment of soil conditions, allowing for the identification of both limitations and potentials of the soil resource. This, in turn, facilitates more precise and effective agronomic decision making in regions with diverse topographic and edaphic conditions, such as Jangas, located in the Ancash region of Peru. A locally adapted SQI is critical for optimizing land use, conserving natural resources, and promoting long-term agroecosystem sustainability. In this study, four SQIs were developed based on two methodological frameworks (PCA and EO), each combined with two scoring functions: linear and non-linear (sigmoidal). These approaches were evaluated in terms of their statistical and spatial performance. Additionally, spatial modeling and mapping of soil quality were conducted using GWRK to explore how soil quality varies across the northeastern part of the Jangas district. This integrated methodology aims to identify the most reliable SQI formulation for supporting sustainable agricultural planning and decision making at the local scale.

2. Materials and Methods

2.1. Study Area

The study was conducted in the Jangas district, located in the Ancash region of Peru, within a tropical montane dry forest ecosystem. The area is primarily used for agricultural production, with key crops including avocado, maize, and beans. Geographically, the study area is located between latitudes 9°27′ and 9°23′ S and longitudes 77°36′ and 77°32′ W, covering an estimated surface area of approximately 2000 hectares (Figure 1). The study site is situated at elevations between 2865 and 2880 m. Climatic records from the Yungay weather station (9°8′30.79″ S; 77°44′59.91″ W; 2466 m) report an average annual temperature of 15.3 °C, with substantial variation between daily maximum and minimum values. The mean annual precipitation is approximately 616 mm, with a marked dry season from May to August, and two rainfall peaks, the main one occurring between January and March.

2.2. Soil Sampling and Analysis

Georeferenced soil samples were taken from 55 different sampling points. Within each sampling point, five sub-samples were extracted at a depth of 20 cm. Areas with atypical moisture conditions, plot edges, and zones with heavy traffic were excluded from sampling. The collected sub-samples were thoroughly homogenized to produce a single composite sample. Around 1 kg loose soil was collected, bagged, and stored under cool, dark conditions before transport to the laboratory. After air-drying, samples were sieved to 2 mm for subsequent analyses at INIA’s Soil, Water, and Foliar Laboratories network. The evaluated variables formed part of a comprehensive soil characterization based on standardized reference methods. Soil texture (sand, silt, and clay percentages) was determined using the Bouyoucos hydrometer method [30]. The pH was measured following EPA guidelines [31], while electrical conductivity (EC) was determined using the saturation extract method [32]. Soil organic carbon (SOC) was analyzed using NOM-021-RECNAT-2000 [30]. Available phosphorus for both neutral and acidic soils was measured using the Olsen method [30], and available potassium was measured according to Bazán [33]. Exchangeable cations (H⁺, Al⁺³, Ca⁺², Mg⁺², K⁺, and Na⁺) and micronutrient contents (Fe, Cu, Zn, and Mn) were also quantified [30]. Additional variables included particulate organic matter, determined using the wet sieving technique and divided into two size categories: fine particulate organic matter (fPOM, 0.053–0.25 mm) and coarse particulate organic matter (cPOM, 0.25–2.0 mm) [34]; labile soil organic carbon (POXC), measured using the KMnO₄ oxidation method [35]; aggregate stability in water, assessed using the successive sieving method [36]; as well as leaf biomass and litter production from the cover crops.

2.3. Soil Quality Evaluation

Evaluation of an SQI consists of three main stages: (1) identifying representative indicators from the complete set of measured soil indicators to establish the MDS; (2) transforming the MDS indicators into scores; and (3) integrating these scores into a single index [4,37].

2.3.1. Selecting the MDS

The selection of indicators was performed using two methods viz., PCA and EO methods.

Principal Component Analysis Approach

PCA is a statistical method used to reduce the dimensionality of a dataset by minimizing the number of variables while retaining most of the original variability. PCs that have higher eigenvalues are considered to be the most representative, accounting for the greatest proportion of variance within the data [4,21].

In this study, PCA was applied to the 24 measured soil indicators. The number of PCs was determined using the eigenvalue criterion, retaining those with values ≥1 for MDS identification. Varimax rotation was applied to the retained PCs to enhance their correlation with soil indicators through variance redistribution [8]. The selection of indicators was guided by the weighted loadings for each component, retaining those with absolute values exceeding 0.60 [4,38]. A multivariate correlation analysis was performed to assess redundancy and the degree of association among variables. For variable pairs showing strong correlations (r ≥ 0.60), the variable with the highest absolute factor loading was retained as the representative indicator. Conversely, when highly weighted variables were not significantly correlated—indicating distinct functional roles—all such variables were included in the MDS to preserve their contributions [39].

Expert Opinion Aproach

In this approach, the selection of primary soil properties was based on the criteria described by Lenka et al. [39], as well as on their recognized influence on soil fertility [40,41]. Indicators representing four key soil functions—(1) soil structure and water retention, (2) nutrient supply capacity, and (3) fundamental soil characteristics that may constrain productive use—were identified through expert judgment, relevant literature, and site-specific edaphic conditions (

Table 1. Soil functions, associated indicators, and their respective weights.

Function	Weight	Function Indicators	Weight	Scoring Function
Soil structural stability and water storage	0.35	SOC	0.20	More is better
		LAgre	0.10	More is better
		Sand	0.05	More is better
Nutrient supply function	0.30	POXC	0.0375	More is better
		fPOM	0.0375	More is better
		Ava_P	0.0375	More is better
		Ava_K	0.0375	More is better
		Fe	0.0375	More is better
		Cu	0.0375	More is better
		Zn	0.0375	More is better
		Mn	0.0375	More is better
Soil basic properties, potential to limit production	0.35	pH	0.10	Optimum is better
		EC	0.10	Less is better
		CEC	0.10	More is better
		Carb	0.05	Less is better

SOC: Soil organic carbon; LAgre: Large soil aggregates (2.00−0.25 mm); POXC: Permanganate oxidizable carbon, fPOM: Fine particulate organic matter; Ava_P: Available P; Ava_K: Available K; EC: Electric conductivity; CEC: Cation Exchange Capacity; Carb: Carbonates percentage.

).

2.3.2. Transformation of the MDS Indicators and Weight Assignment

Due to differences in measurement units among the selected soil parameters, values were standardized to a 0–1 scale using appropriate linear or non-linear transformation techniques. The scoring was based on each indicator’s relationship with soil fertility. Indicators that positively influence soil fertility were classified under a “more is better” function, while those with negative effects were assessed using a “less is better” approach. Indicators exhibiting both beneficial and detrimental effects depending on their levels were evaluated using an “optimum range” scoring method [5,42].

In the linear scoring approach, two types of functions were applied: “more is better” (Equation (1)) and “less is better” (Equation (2)):

S_L = (x − l)/(h − l)

(1)

S_L = 1 − ((x − l)/(h − l))

(2)

where S_L is the normalized linear score (0–1), x is the measured variable, l is the minimum, and h the maximum value [8].

In the non-linear scoring approach, a sigmoidal function was applied (Equation (3)) [5,10]:

S_NL = a/(1 + (x/x_o)^b)

(3)

where S_NL is the normalized nonlinear score (0–1), a is the maximum score (set to 1 in this study), x is the measured soil variable, x₀ is its mean value, and b is the slope (−2.5 for “more is better” and +2.5 for “less is better”) [43,44].

The indicator scores were integrated into indices using a weighted additive approach (Equation (4)) [4]:

$${SQI}_W ={\sum }_{n = 1}^{n}WiSi$$

(4)

where Si represents the score assigned to each indicator (whether obtained through linear or non-linear transformation), n denotes the total number of variables included in the index, and Wi corresponds to the weight assigned to each indicator.

In the PCA-based approach, the weighting of the indicators forming the MDS was determined according to the variance distribution obtained from the PCA results. Each PC accounted for a specific proportion of the total variance in the dataset. The weighting of each indicator in a PC was computed as the proportion of the variance it explained within that component. It was explained by the cumulative variance interpreted by all retained PCs with eigenvalues greater than 1, thereby reflecting the relative contribution of each indicator to the overall dataset variability [39].

Following the expert-based procedure, weighting values were attributed to the key soil functions (Table 1) to indicate their contribution to total soil performance. These weights were then proportionally distributed to the individual indicators within each function, as determined by expert judgment and supported by evidence from the scientific literature [9,39].

2.4. Spatial Mapping Using GWRK

To generate spatial predictions of the four developed IQS, the GWRK method was applied. GWRK is a hybrid geostatistical technique that integrates the strengths of both geographically weighted regression and ordinary kriging. This method consists of a deterministic component—modeled using geographically weighted regression, which captures spatially varying relationships between the target variable and a set of covariates—and a stochastic component, represented by the spatial interpolation of the regression residuals via kriging [23]. GWRK was implemented using the set of covariates listed in

Table 2. Environmental covariates used in the GWRK.

Covariate	Source
NDMI	Image collection “COPERNICUS/S2_SR”
Elevation	Shuttle Radar Topographic Mission (SRTM)
Slope	Shuttle Radar Topographic Mission (SRTM)
Aspect	Shuttle Radar Topographic Mission (SRTM)

. This approach allowed for the generation of maps for each IQS, capturing both spatial trends and local variations in soil quality across the study area.

The flowchart shown in Figure 2 summarizes the methodology used for the development of the four SQIs—EOLinear, EONLinear, PCALinear, and PCANLinear—and the subsequent spatial mapping. It outlines the sequence from soil sampling and laboratory analysis to indicator selection and weighting, using either the EO or PCA approach, each combined with linear or non-linear scoring methods. The resulting indices were then integrated into a spatial modeling process to generate high-resolution soil quality maps.

2.5. Statistical Analysis

The associations between the 24 soil parameters were examined using Spearman’s rank correlation coefficients [45]. A PCA was carried out to reduce the dimensionality of the data and to identify the MDS. Differences among the developed soil quality indices were evaluated using the non-parametric Friedman test, followed by Nemenyi’s post hoc test for pairwise comparisons [46]. All statistical analyses, including the implementation of GWRK, were performed using R software version 4.4.1.for Windows (R Core Team, Vienna, Austria, 2023) [47].

3. Results

3.1. Statistical Summary of Soil Properties

The parameters of interest included principal particles of soil (sand, silt, and clay), aggregate stability (in 2 sizes), and organic carbon due to their relevance in maintaining soil structure and enhancing water infiltration (see

Table 3. Statistical characteristics of parameters that maintain soil structure and water storage.

Variable	Min	Max	Mean	Median	Std. Deviation	CV	Kurtosis	Skewness
Sand (%)	36.70	79.06	57.49	56.98	7.82	13.59	1.32	0.12
Silt (%)	18.78	60.29	36.84	37.63	8.64	23.46	0.49	0.16
Clay (%)	1.46	22.46	5.66	2.94	5.33	94.09	1.83	1.08
LAgre (%)	8.50	83.88	42.89	44.51	20.84	48.60	−1.17	−0.04
SAgre (%)	5.24	59.61	28.30	28.46	13.22	46.73	−0.51	0.27
SOC (%)	0.12	3.31	1.27	1.04	0.84	66.01	−0.27	0.50

LAgre: Large aggregates (2.00−0.25 mm); Sagre: Small aggregates (0.25−0.053 mm), SOC: Soil organic carbon.

). The sand fraction in the soils under study varied between 36.70% and 79.06%, with an average content of 57.49% and an approximately symmetrical distribution (skewness = 0.12). The silt content ranged from 18.78% to 60.29%, with a mean of 36.84%, also displaying a near-symmetrical distribution (skewness = 0.16). By contrast, the clay content was markedly lower, averaging 5.66%, but exhibited substantial variability (CV = 94.09%) and a positively skewed distribution (1.08), suggesting the occurrence of occasional high values. In terms of soil structural composition, the proportion of large aggregates ranged from 8.50% to 83.88%, with a mean value of 42.89%. The distribution was symmetrical but platykurtic (kurtosis = −1.17), indicating a flatter-than-normal distribution. Small aggregates averaged 28.30%, with a slightly right-skewed distribution (skewness = 0.27). The soil organic carbon content was relatively low overall, with a mean of 1.27%. However, it was highly variable (66.01%) and exhibited moderate positive skewness (skewness = 0.50), reflecting a predominance of low-carbon samples with some locations showing elevated organic matter levels.

Spearman correlation analysis revealed significant relationships among soil physical and chemical properties (see Figure 3). A robust negative correlation was observed between sand and clay contents (−0.68, ***), indicating a compensatory relationship typical of soil textural components. Similarly, sand was moderately and negatively correlated with silt (−0.41, *). The proportion of large aggregates (LAgre) was strongly and negatively correlated with small aggregates (SAgre) (−0.79, ***), suggesting a clear structural differentiation. Additionally, LAgre showed a moderate positive correlation with SOC (0.45, **), highlighting the role of organic matter in promoting soil aggregation. By contrast, SOC did not exhibit significant correlations with texture components. Overall, the results suggested that while soil texture fractions are interrelated, the structural stability of the soil, particularly the formation of large aggregates, is more closely associated with organic carbon content than with texture.

Descriptive statistics for labile carbon fractions, available nutrients, and exchangeable cations revealed notable variability across soil samples (see

Table 4. Statistical characteristics of parameters related to nutrient availability.

Variable	Min	Max	Mean	Median	Std. Deviation	CV	Kurtosis	Skewness
POXC mg kg−1	203.64	1011.81	716.31	768.55	203.97	28.47	0.09	−0.51
cPOM (g kg−1)	0.62	11.32	2.74	2.24	1.78	64.91	11.32	1.94
fPOM (g kg−1)	1.68	28.15	4.75	4.10	3.64	76.63	32.38	3.39
Ava_P (mg kg−1)	2.37	44.02	16.68	13.45	10.59	63.46	−0.36	0.50
Ava_K (mg kg−1)	35.60	766.00	179.20	151.20	133.08	74.26	8.00	1.71
Ex_Ca (cmol(+) kg−1)	3.68	16.51	8.52	8.48	2.94	34.49	−0.29	0.21
Ex_Mg (cmol(+) kg−1)	0.13	3.03	1.17	1.10	0.56	47.88	3.10	0.80
Ex_Na (cmol(+) kg−1)	0.06	1.56	0.42	0.39	0.31	75.06	1.72	0.72
Ex_K (cmol(+) kg−1)	0.03	1.01	0.25	0.18	0.21	83.34	5.01	1.42
Fe (mg kg−1)	10.24	339.52	48.20	25.28	56.64	117.51	12.79	2.06
Cu (mg kg−1)	0.08	4.24	0.82	0.40	1.07	130.22	4.08	1.45
Zn (mg kg−1)	0.10	8.62	1.74	0.64	2.17	124.71	2.11	1.11
Mn (mg kg−1)	1.84	173.08	24.68	15.36	28.86	116.93	12.60	2.02

POXC: Permanganate oxidizable carbon, cPOM: Coarse particulate organic matter; fPOM: Fine particulate organic matter; Ava_P: Available P; Ava_K: Available K; Ex_Ca: Exchangeable Ca; Ex_Mg: Exchangeable Mg; Ex_Na: Exchangeable Na; Ex_K: Exchangeable K.

). The POXC ranged from 203.64 to 1011.81 mg kg⁻¹, showing moderate variability (28.47%) and a near-normal distribution. By contrast, both particulate organic matter fractions—cPOM and fPOM—exhibited high variability (64.91% and 76.63%, respectively) and highly leptokurtic distributions (kurtosis = 11.32 and 32.38, respectively), indicating strong data concentration near the mean and the presence of outliers. fPOM also showed marked positive skewness (3.39), reflecting a tail toward higher values. The available phosphorus and potassium levels showed considerable variation (CV > 60%). Ava_K, in particular, displayed a highly skewed and leptokurtic distribution, suggesting the influence of extreme values in some samples. Nonetheless, the majority of Ava_P and Ava_K values fell within ranges that are generally considered adequate for plant nutrition. Exchangeable cations, including Ca²⁺ and Mg²⁺, had moderate coefficients of variation (34.49% and 47.88%, respectively), whereas Na⁺ and K⁺ were more variable and right-skewed (CV > 70%). Micronutrient concentrations—Fe, Cu, Zn, and Mn—were especially variable, with CVs exceeding 100%; however, most values for these micronutrients also fell within levels considered adequate for soils, indicating generally favorable nutrient status despite the observed variability. All micronutrients exhibited highly skewed and leptokurtic distributions, indicating the presence of extreme concentrations in a subset of the samples. These findings suggested a heterogeneous distribution of micronutrients and labile organic matter in the soils, possibly influenced by localized management practices or parent material variation.

The Spearman correlation matrix revealed multiple significant associations among labile carbon fractions, macronutrients, exchangeable bases, and micronutrients (see Figure 4). A strong and highly significant positive correlation was observed between Ex_K and Ava_K (Spearman’s ρ = 0.94, ***), indicating a high consistency between these two potassium pools. The organic carbon fractions (POXC, fPOM, and cPOM) were weakly correlated with each other and showed only modest associations with nutrient variables. Micronutrient elements exhibited strong intercorrelations. For example, Cu was highly correlated with Zn (0.81, ***) and Mn (0.54, **), indicating co-accumulation or shared geochemical behavior. Zn was also positively associated with Mn (0.51, **). Among exchangeable bases, Ca was negatively correlated with Mn (−0.60, ***).

The results showed that the majority of these soils possessed optimal pH levels, were devoid of salinity issues, and did not exhibit sodium-related problems, making them well suited for cultivating a wide range of crops. Soil pH values ranged from 4.57 to 7.90, with an average of 6.77 and a low coefficient of variation (9.16%), indicating relatively consistent soil low acidity levels. The distribution of pH was slightly platykurtic and negatively skewed, suggesting a tendency toward lower pH values in some samples (

Table 5. Statistical characteristics of basic parameters, potential to limit production.

Variable	Min	Max	Mean	Median	Std. Deviation	CV	Kurtosis	Skewness
pH	4.57	7.90	6.77	6.76	0.62	9.16	1.85	−0.51
EC (dS m−1)	0.29	2.72	0.95	0.74	0.62	64.58	0.84	0.81
CEC (cmol(+) kg−1)	4.77	19.60	10.35	9.98	3.18	30.73	0.23	0.38
ESP (%)	0.53	11.64	4.19	3.18	3.05	72.81	−0.48	0.47
Carb. (%)	0.36	10.83	3.23	2.33	2.09	64.74	1.75	0.83

EC: Electric conductivity; CEC: Cation exchange capacity; ESP: Exchangeable sodium percentage; Carb: Carbonates percentage.

). Electrical conductivity values ranged from 0.29 to 2.72 dS m⁻¹, with a mean of 0.95 dS m⁻¹ and a high CV (64.58%), reflecting heterogeneity in soil salinity. Similarly, CEC varied from 4.77 to 19.60 cmol₍₊₎ kg⁻¹, with moderate variability (30.73%) and a near-normal distribution. Sodium exchange percentage and carbonate content (%) also displayed high variability. SEP values were slightly right-skewed and platykurtic, while the carbonate content showed a leptokurtic and positively skewed distribution, suggesting the presence of localized carbonate accumulation. These results indicated diverse soil chemical conditions across the sampled sites, with implications for nutrient availability and soil management.

Soil pH showed a moderately strong positive correlation with CEC (0.63) and carbonate content (0.56), indicating that higher pH levels are associated with increased nutrient retention capacity and carbonate accumulation. A strong positive correlation was also observed between carbonate content and CEC (0.72), suggesting a close linkage between soil alkalinity components and exchangeable base capacity. By contrast, EC and ESP showed no significant correlations with any of the other measured soil properties (Figure 5). These results highlighted the important interactions among pH, carbonates, and CEC in determining soil chemical fertility.

3.2. Principal Component Analysis

Based on the results of the Spearman correlation analysis among the initial set of variables (see Figure 3, Figure 4 and Figure 5), the number of indicators was reduced from 24 to 16 by applying an absolute correlation threshold of 0.60 in order to minimize multicollinearity. PCA extracted six components with eigenvalues greater than one, cumulatively explaining 74% of the total variance in the dataset (

Table 6. PCA output including eigenvalues, percentage of variance explained, and factor loadings of component matrix variables.

Principal Components	PC1	PC2	PC3	PC4	PC5	PC6
Eigenvalue	2.464	2.437	1.918	1.846	1.808	1.357
% variance	0.154	0.152	0.120	0.115	0.113	0.085
% cumulative variance	0.154	0.306	0.426	0.542	0.655	0.739
Weightage factor	0.21	0.42	0.58	0.73	0.89	1.00
Factor Loadings from the Rotated Component Matrix
Sand	−0.08	−0.06	−0.09	0.09	0.05	−0.91
Clay	0.82	0.05	−0.09	0.03	−0.23	0.17
LAgre	−0.16	0.77	0.16	−0.15	0.42	0.12
SOC	0.36	0.72	0.16	0.14	0.01	−0.25
POXC	0.12	0.60	0.13	0.26	0.24	0.06
fPOM	−0.24	0.05	0.17	0.16	0.77	0.11
cPOM	0.23	0.12	−0.15	−0.01	0.85	−0.17
Ava_P	0.13	0.15	−0.32	0.69	−0.15	−0.02
Ava_K	0.22	0.42	0.08	0.45	0.11	0.48
Ex_Mg	0.06	0.21	0.78	0.17	−0.12	0.09
Fe	0.78	0.19	−0.22	0.09	0.15	0.11
Zn	−0.05	0.79	−0.07	0.04	−0.18	0.19
pH	−0.51	0.16	0.36	−0.54	−0.07	0.27
EC	−0.04	0.10	0.23	0.78	0.25	−0.03
Carb	−0.16	0.03	0.79	−0.28	0.18	0.05
ESP	0.74	−0.04	0.45	0.06	0.04	−0.10

Indicators shown in bold were those selected in the corresponding principal component.

). PC1 accounted for the highest proportion of variance (15.4%), followed by PC2 (15.2%), PC3 (12.0%), PC4 (11.5%), PC5 (11.3%), and PC6 (8.5%). Varimax rotation of the factor loading matrix revealed distinct associations among variables within each component. PC1 was strongly influenced by clay content, Fe, and ESP. PC2 was associated with high loadings of Zn, LAgre, SOC, and POXC. PC3 was primarily defined by carbonate content and exchangeable magnesium (Mg), while PC4 reflected high loadings for EC and available phosphorus. PC5 was dominated by fractions of organic matter, specifically cPOM and fPOM. Finally, PC6 was solely characterized by sand content, which had a factor loading greater than 0.60.

3.3. Expert Opinion

The selection of properties for the minimum soil dataset was based on the available dataset, a consensus among the authors, and supporting literature. For indicators related to soil structure and water retention, sand content was chosen because it serves as a fundamental indicator of soil texture. The percentage of large stable aggregates was included due to its relevance in representing soil structural integrity. Organic carbon was also selected for its essential role in stabilizing aggregates and enhancing water-holding capacity. Regarding nutrient supply, the most reactive fractions of organic matter—POXC and fPOM—were prioritized, as they reflect the mineralization potential of soil organic matter. Additionally, direct indicators of macro- and micronutrient availability were included, such as P and K, along with Fe, Cu, Zn, and Mn. Among general soil chemical parameters, pH and EC were included for their influence on nutrient dynamics and plant growth. CEC was selected due to its central role in soil fertility and ion exchange processes, while carbonate content was considered for its importance as a limiting factor in nutrient availability and soil chemical behavior.

3.4. Indicator Scores

The MDS indicators were scored using both linear and non-linear functions, as described in Equations (1)–(3). The indicators were categorized into three functional classes based on their relationship with soil quality. The first group, categorized under a “more is better” function, included sand, clay, LAgre, SOC, POXC, fPOM, cPOM, Ava_P, Ava_K, Ex_Mg, Fe, Zn, Cu, Mn, and CEC, as higher values of these variables are generally associated with improved soil function and productivity. The second group followed a “less is better” function and included EC and Carb, given their potential negative impacts on soil quality when present in excess. The third group included pH, which was assessed using an “optimum” function, recognizing that both excessively low and high values can adversely affect soil health and crop performance.

3.5. Results of Weighted Soil Quality Index

In the PCA-based approach, weights for the 14 selected indicators were determined according to the proportion of variance each PC contributed, as presented in Table 5. As an example, the 21% variance explained by PC1 was distributed among the indicators with significant loadings—Clay, Fe, and ESP—by proportionally allocating this total variance based on the absolute values of their respective loading coefficients. The same procedure was applied to each of the six retained components. The integration of the index was based on the following formula:

$$\begin{array}{ll}{\mathrm{IQS}}_{\mathrm{PCA}}& ={0.074\mathrm{S}}_{\mathrm{Clay}}+{0.070\mathrm{S}}_{\mathrm{Fe}}+{0.066\mathrm{S}}_{\mathrm{ESP}}+{0.056\mathrm{S}}_{\mathrm{Lagre}}+{0.053\mathrm{S}}_{\mathrm{SOC}}+{0.058\mathrm{S}}_{\mathrm{Zn}}+\\ & {0.044\mathrm{S}}_{\mathrm{POXC}}+{0.079\mathrm{S}}_{\mathrm{Ex\_Mg}}+{0.083\mathrm{S}}_{\mathrm{Carb}}+{0.085\mathrm{S}}_{\mathrm{EC}}+{0.077\mathrm{S}}_{\mathrm{Ava\_P}}+{0.071\mathrm{S}}_{\mathrm{fPOM}}+\\ & {0.079\mathrm{S}}_{\mathrm{cPOM}}+{0.12\mathrm{S}}_{\mathrm{Sand}}\end{array}$$

(5)

Under the EO approach, 15 parameters were identified based on their importance to fundamental soil physical, chemical, and biological processes. The relative importance of each variable was quantified, and their respective weights are detailed in Table 2. The integration of the index was based on the following formula:

$$\begin{array}{ll}{\mathrm{IQS}}_{\mathrm{EO}}& ={0.20\mathrm{Ss}}_{\mathrm{OC}}+{0.10\mathrm{S}}_{\mathrm{LAgre}}+{0.05\mathrm{S}}_{\mathrm{Sand}}+{0.0375\mathrm{S}}_{\mathrm{POXC}}+{0.0375\mathrm{S}}_{\mathrm{fPOM}}+{0.0375\mathrm{S}}_{\mathrm{Ava\_P}}+\\ & {0.0375\mathrm{S}}_{\mathrm{Ava\_K}}+{0.0375\mathrm{S}}_{\mathrm{Fe}}+{0.0375\mathrm{S}}_{\mathrm{Cu}}+{0.0375\mathrm{S}}_{\mathrm{Zn}}+{0.0375\mathrm{S}}_{\mathrm{Mn}}+{0.10\mathrm{S}}_{\mathrm{pH}}+{0.10\mathrm{S}}_{\mathrm{EC}}\\ & +{0.10\mathrm{S}}_{\mathrm{CEC}}+{0.05\mathrm{S}}_{\mathrm{Carb}}\end{array}$$

(6)

3.6. Soil Quality Index

The SQI values derived from the four assessment methods—PCALinear, PCANLinear, EOLinear, and EONLinear—showed considerable variability across the 55 soil samples (

Table 7. Soil quality index by four methods in 55 soil samples.

	PCALinear	PCANLinear	EOLinear	EONLinear
Min	0.31	0.31	0.23	0.25
Max	0.73	0.80	0.67	0.72
Mean	0.43	0.47	0.43	0.46
Median	0.42	0.46	0.42	0.44
Std. deviation	0.071	0.088	0.087	0.101
CV	16.683	18.965	20.280	22.194
Kurtosis	5.142	2.374	0.362	−0.356
Skewness	1.666	1.171	0.971	0.881
Friedman Test	Friedman χ2 (df = 3)		p-value
	33.48		<0.001 ***
	Comparison		Nemenyi p-value
Nemenyi Test	PCALinear vs. PCANLinear		<0.001 (***)
	PCALinear vs. EOLinear		0.998 (ns)
	PCALinear vs. EONLinear		<0.001 (***)
	PCANLinear vs. EOLinear		<0.001 (***)
	PCANLinear vs. EONLinear		0.921 (ns)
	EOLinear vs. EONLinear		0.002 (**)

(**): p-value < 0.01; (***): p-value < 0.001; (ns): non significative.

). Mean values ranged from 0.43 (PCALinear and EOLinear) to 0.47 (PCANLinear), with median values following a similar trend. The highest SQI was recorded under the PCANLinear approach (0.80), while the lowest value was observed using the EOLinear method (0.23). The coefficient of variation was lowest for PCALinear (16.68%), suggesting greater consistency, and highest for EONLinear (22.19%), indicating more heterogeneity among samples. Distributional statistics showed that all methods exhibited positive skewness, particularly PCALinear (1.666), implying a concentration of lower index values with fewer higher value outliers. Similarly, the kurtosis values suggested a leptokurtic distribution in PCALinear (5.142), contrasting with the slightly platykurtic pattern in EONLinear (−0.356).

The Friedman test detected significant differences among the four methods (χ² = 33.48, p < 0.001). Post hoc comparisons using the Nemenyi test indicated that PCANLinear differed significantly from PCALinear and EOLinear. Significant differences were also found between EOLinear and EONLinear, while no significant difference was observed between PCALinear and EOLinear (Figure 6).

3.7. Spatial Mapping of Soil Quality Indices

The spatial distribution of SQI values derived from the four methodological approaches is illustrated in Figure 7. While the general point patterns were broadly consistent across methods, localized discrepancies were observed at specific sampling points. Notably, all four indices displayed a clear trend: lower SQI values tended to cluster in the southwestern part of the study area, while higher values were primarily found in the northern zone. This pattern suggested that, despite methodological variations, there was strong agreement among the indices in identifying areas of relatively high and low soil quality.

The GWRK performance of the four SQI calculation methods was evaluated using five statistical metrics (

Table 8. Performance metrics of GWRK models.

	R2	RMSE	MAE	AIC	CV_RMSE
PCALinear	0.588	0.0451	0.0242	−132.32	10.49
PCANLinear	0.736	0.0459	0.0233	−136.51	9.77
EOLinear	0.689	0.0478	0.0346	−134.58	11.12
EONLinear	0.688	0.0558	0.0435	−127.33	12.13

). The PCANLinear method demonstrated the highest predictive accuracy, with the greatest R² value (0.736), the lowest MAE (0.0233), and the lowest CV_RMSE (9.77), indicating a better fit and more reliable estimation across the dataset. Although PCALinear showed a slightly lower R² (0.588), it exhibited the lowest RMSE (0.0451) and a competitive AIC value (−132.32), suggesting a good balance between model fit and parsimony. The EOLinear method presented intermediate performance, with an R² of 0.689 and moderate error metrics. Conversely, the EONLinear method showed the weakest performance among the four, with the highest RMSE (0.0558), MAE (0.0435), and CV_RMSE (12.13), indicating lower precision and greater variability in its estimations. These results supported the use of PCANLinear as the most robust approach for SQI prediction in this study.

The spatial interpolation maps of adjusted SQI values generated using the four methodological approaches are shown in Figure 8. These maps reflect the extrapolation of SQI values across the study area based on the model performances summarized previously. As observed, the spatial distribution patterns were broadly consistent across all four methods, reinforcing the trends previously noted in Figure 7 with the raw point data. The PCALinear and PCANLinear maps (Figure 7A,B) exhibited particularly high spatial agreement, highlighting very similar patterns of SQI variation across the landscape. Both indicated higher soil quality values concentrated in the northern and northeastern portions of the study area, with a gradual decrease in quality toward the southwestern and central zones. Similarly, the EOLinear and EONLinear maps (Figure 7C,D) also presented comparable spatial patterns, albeit with slightly smoother transitions between classes and marginally broader areas showing intermediate quality values. These similarities within method pairs (PCA-based and EO-based) likely reflected the influence of the underlying indicator selection and weighting criteria, as well as their model-based performance metrics. This spatial coherence supported the reliability of the SQI indices for guiding site-specific land management and soil improvement strategies in the region.

4. Discussion

4.1. Soil Properties and Their Role in Soil Quality

The results obtained for the soil structural parameters indicated that the percentage of stable aggregates, in both large and small fractions, was more strongly related to SOC content than to the proportion of sand, silt, or clay [48,49,50]. This trend may be attributed to the low clay content observed in the study area (median = 2.94%), as clay is one of the primary physical agents involved in soil aggregation through mechanisms such as zeta potential, which influence the dispersion or aggregation of soil particles [51]. SOC plays a pivotal role in the formation of macroaggregates, as evidenced by studies by Zhou et al. [52], conducted under different soil management systems, and Wang et al. [53], conducted under various fertilization regimes. The percentage of strong organo-mineral bonds is higher in large aggregates, highlighting the binding strength of SOC, which can potentially reduce the bioavailability of the nutrients it contains [54]. Despite the relatively low levels of SOC observed in the studied soil, the positive effect on aggregate stability suggested that high concentrations of SOC may not be necessary to promote aggregation. Instead, the effectiveness of SOC appears to depend on the clay/SOC ratio, as reported by Soinne et al. [55], who found that a low clay/SOC ratio significantly enhanced aggregate formation. Conversely, small-sized stable aggregates demonstrated an absence of statistically significant relationships with any of the evaluated variables, with the exception of an inverse correlation with SOC. This phenomenon may be attributable to the influence of other factors, such as the specific mineralogy of the clay fraction, which plays a more prominent role in microaggregate formation [56].

The results obtained for POXC were consistent with the findings reported by various authors [35,57]. As one of the most active fractions of SOC, POXC showed a direct and significant relationship with SOC, which was also supported by other studies [58,59]. However, recent research has questioned whether POXC accurately represents the most labile fractions of SOC [60]. Regarding the different fractions of POM, the results for both fPOM (0.053–0.25 mm) and cPOM (0.25–2.0 mm) were consistent with those reported in other studies [61,62,63]. Nevertheless, in this study, neither fraction showed significant correlations with SOC or POXC. This contrasted with the results of other research, which reported significant associations [35,64,65]. One potential explanation is that those studies were conducted under controlled experimental conditions with defined treatments, which tend to reduce data variability. By contrast, our data were derived from a non-experimental context, characterized by high variability, elevated coefficients of variation, and the presence of extreme values (see Table 4).

Samaniego et al. [66] reported litter decomposition curves for the region’s primary crops, finding that nutrient release rates were generally rapid. This may explain why most of the assessed nutrients fell within adequate concentration ranges. On the other hand, the concentrations and significant correlations observed among the micronutrients may be associated with similar patterns previously documented by other authors [67,68,69,70]. All of these elements exist in cationic form in the soil solution, which leads to comparable interactions with the soil’s colloidal phase [40]. This assertion was further supported by the significant correlations observed between the micronutrients and exchangeable soil cations (see Table 4). In addition, given their low concentrations and high susceptibility to sorption and desorption processes [71], as well as potential interactions with the soil microbiome [72], these elements demonstrated high coefficients of variation and skewed distributions with long tails toward higher values.

Soil pH is a significant parameter that influences numerous chemical and biological processes. In this study, the pH showed significant correlations with CEC, calculated as the sum of exchangeable cations, thereby corroborating previous findings [73,74,75]. This relationship can be attributed to the fact that higher pH levels increase the number of negatively charged sites on the clay–humic complex, thereby enhancing the soil’s capacity to retain cations and increasing its CEC [40]. Additionally, soils with higher carbonate contents were significantly correlated with elevated pH values, likely due to the neutralization of H⁺ ions when carbonates react with the soil solution [76,77]. However, the low EC observed, along with its weak correlation with carbonate content, suggested that the presence of carbonates in these soils may originate from the parent material rather than from secondary precipitation via irrigation water—an accumulation mechanism reported in other environments [78].

Most of the evaluated indicators are commonly included in agricultural soil assessments. Some parameters, such as sand, silt, and clay contents, exhibit high temporal stability and are not easily influenced by management interventions [79]. By contrast, other indicators—such as soil SOC, EC, pH, CEC, and aggregate stability—are more responsive to medium-term management practices [80]. Finally, certain variables, such as the balance of exchangeable cations and the availability of micronutrients, can respond more rapidly to targeted soil fertilization strategies [81].

4.2. Comparative Analysis of SQI Construction

The development of SQIs using both the PCA- and EO-based approaches revealed distinct patterns in the selection and weighting of indicators, reflecting their underlying conceptual and methodological differences. In the PCA-based SQI, indicator weights were derived from the proportion of variance each contributed within their respective PCs. As a result, variables such as Sand (0.12), Ex_Mg (0.079), Carb (0.083), and EC (0.085) received relatively high weights. However, it has been observed that many of these variables are not typically representative of key soil functions, a phenomenon commonly observed in other studies. For instance, Damiaba et al. [82] reported high weights for Mg and moisture content, while Vasu et al. [9] found the sodium adsorption ratio and clay to dominate the PCA-derived SQI. Nonetheless, certain studies have shown that PCA places greater emphasis on functionally relevant indicators, such as SOC, macronutrients, and aggregate stability [5,83]. This suggests that the utility of PCA can vary depending on the dataset’s structure and soil conditions.

By contrast, the EO-based SQI prioritized indicators based on their relevance to essential soil functions, ensuring balanced representation across physical (e.g., LAgre and Sand), chemical (e.g., pH, EC, CEC, and Ava_K), and biological (e.g., SOC, POXC, and fPOM) domains. The highest weight was assigned to SOC (0.20), emphasizing its central role in supporting soil structure, nutrient retention, and microbial activity. Additionally, pH, EC, and CEC each contributed 0.10 to the index, thereby aligning with their well-established importance in regulating chemical balance and buffering capacity in agricultural soils. This function-oriented selection approach is strongly supported in the literature [10,39,84], and it allows for greater ecological interpretability and management relevance.

Agronomically significant indicators included in the EO approach—such as pH and Ava_K—were not retained in the PCA-based index. Their exclusion suggested either limited statistical variance across samples or collinearity with other variables, which would cause PCA to discard them during dimensional reduction. However, the omission of functionally essential variables represents a limitation of purely statistical selection methods. As shown in previous studies [4,85,86], PCA’s objectivity can sometimes come at the cost of ecological interpretability, particularly in systems where management goals dictate the need for certain indicator types, regardless of statistical loading.

The observed differences in SQI values between linear (PCALinear, EOLinear) and non-linear (PCANLinear, EONLinear) normalization approaches stemmed from the mathematical nature of their respective scoring functions. Linear normalization, using simple range rescaling (Equations (1) and (2)), tends to assign scores evenly across the data distribution, which can overemphasize the contribution of extreme values and underestimate those in proximity to the mean [87,88]. This leads to more skewed and peaked distributions, as evidenced in PCALinear.

By contrast, non-linear scoring through sigmoidal functions (Equation (3)) has been demonstrated to moderate the influence of extreme values and amplify mid-range differences. This result yields distributions that are more balanced and reflective of functional performance. This is illustrated by the flatter, less skewed distribution of EONLinear. Previous studies emphasized the superiority of non-linear transformations for environmental indices, noting that soil functions do not always improve proportionally in conjunction with increases in indicator values [5,10,89]. Consequently, scoring methods that incorporate thresholds or diminishing returns—such as sigmoid curves—more accurately reflect ecological realities.

This consideration is particularly important in edaphic datasets, which often exhibit high variability, skewness, and occasional values near or below detection limits [90,91]. These data characteristics can hinder statistical analyses and spatial interpolations, especially when using methods such GWRK that rely on residual modeling [92]. By attenuating the disproportionate effect of outliers and enhancing sensitivity in the mid-range, non-linear scoring facilitates the development of more robust soil quality indices and spatial representations.

The significant differences found in the Friedman and Nemenyi tests further supported that the adoption of scoring method—not just variable selection—can meaningfully alter soil quality assessments. In summary, while linear methods offer simplicity, non-linear approaches provide greater ecological realism and statistical robustness in SQI construction. They are particularly effective at capturing nuanced differences in mid-range soil quality and preventing the distortion caused by outliers.

4.3. Soil Quality Maps

A variable may exhibit statistical skewness or non-normality yet still display spatially coherent patterns that enhance model performance. Conversely, a variable with a statistically ideal distribution may lack spatial structure, rendering it less predictable in spatially explicit models. This distinction is especially relevant when employing GWRK, which relies on both spatial autocorrelation and covariate relationships to generate localized predictions [93,94]. For example, if the values are very uneven or have extreme outliers, the model may become less reliable in some areas. However, when there is a strong spatial pattern in the data, this can help the model stay consistent and accurate, even if the overall distribution is not ideal [95].

Despite the similar statistical robustness between PCANLinear and EONLinear, only PCANLinear achieved superior predictive performance when GWRK was implemented. The PCANLinear index (see Figure 8B) manifested the most extensive range of values and a more nuanced spatial structure, especially in the central-eastern part of the study area, which aligns with its superior performance. This map delineated a finer gradient of soil quality, particularly in transitional zones. By contrast, PCALinear (Figure 8A) displayed a more compressed value distribution, with a predominance of mid-range classes. This suggested that linear scoring may underrepresent variability at the extremes, leading to a loss of spatial detail in areas with marginal or exceptional soil conditions. The EOLinear (Figure 8B) and EONLinear (Figure 8D) maps shared some similarities with their PCA counterparts. However, they also exhibited increased spatial fragmentation and patchiness. In particular, EONLinear, despite its use of non-linear scoring, showed reduced spatial coherence and more noise in the southern region, which was consistent with its higher prediction error. As illustrated in Figure 7, the spatial distributions of the indicators derived from PCANLinear and EONLinear followed analogous trends However, minor discrepancies at specific points resulted in substantial variation in the performance of the GWRK model. Overall, the maps illustrated that non-linear scoring combined with PCA yielded the most consistent and spatially informative results, reinforcing the idea that both statistical distribution and spatial behavior must be considered when designing soil quality indices.

The objective of constructing SQIs is to create tools specifically designed for the management goals and environmental conditions of a study area. The application of an identical index—with similar indicators and weights—in diverse contexts can prove arduous and is frequently discouraged [96]. Nevertheless, this study provides important theoretical insights by demonstrating that the combination of PCA-based variable selection and non-linear normalization generates indices that are not only statistically robust but also ecologically interpretable. This integration of methodological rigor with ecological relevance strengthens the scientific basis for SQI development by showing how indices can more effectively represent non-linear responses and threshold dynamics in soil functions. From a practical standpoint, the framework developed also provides a transferable reference for agroecosystems with similar challenges, especially in regions marked by heterogeneity and scarce data resources. Because indicator selection is data-driven while remaining sensitive to local conditions, the approach can be readily adjusted to different regions without relying on a rigid or uniform scheme. Likewise, the use of non-linear scoring functions mitigates the influence of outliers and enhances the interpretability of mid-range variability, making the indices highly applicable to soils in tropical, arid, or degraded landscapes where such data features are common [97,98]. In this sense, the research contributes not only to advancing soil quality theory but also to providing a flexible and transferable tool for guiding sustainable land management in diverse agroecological contexts.

5. Conclusions

Among the four SQIs evaluated, the PCANLinear configuration exhibited superior performance, effectively reducing data skewness and kurtosis while achieving the highest predictive accuracy under GWRK. This finding underscores the benefits of objective weighting approaches such as PCA when combined with non-linear normalization. Nonlinear scoring functions were particularly effective in mitigating the influence of outliers and enhancing the representation of mid-range variability—an essential aspect analyzing highly dispersed edaphic datasets. SQIs should be integrated into public soil conservation policies, such as the development of targeted programs in the Ancash region that allocate subsidies for regenerative practices informed by edaphic quality maps and aligned with existing initiatives. This approach would promote sustainable land management aligned with local needs and climate resilience. Specifically, in areas with low SQI values (<0.4), farmers should be encouraged to apply organic amendments and mulching systems to improve soil conditions. Conversely, zones with high SQI values (>0.6) offer favorable conditions for the intensification of high-input crops, such as potatoes, alfalfa, and maize.