Geodetector–Geographically Weighted Regression Integrated Analysis of Factors Controlling Selenium Distribution in Farmland Topsoil: A Case Study of Xin’an Town (Linli County, Hunan, China)

Abstract

Selenium (Se) is an essential trace element for humans, and agricultural soils are a major source of dietary Se. Therefore, identifying the key environmental drivers of Se in farmland is crucial for evaluating the resource base for Se-rich agriculture and improving human health. Although soil Se distribution and its controlling factors have been widely investigated, quantitative assessments of soil Se in small-scale farmland systems under humid monsoon conditions remain limited. Sampling sites were designed to represent different geological types, soil types, and topography, and 314 farmland topsoil (0–20 cm) samples were collected. Total Se was determined after complete HNO₃–HClO₄ wet digestion and quantified by HG–AFS (AFS–830), with certified reference materials showing recoveries of 95.3–101.2%. The spatial patterns were mapped using ordinary kriging. Geographically weighted regression (GWR) and Geodetector were used to explore the impact of environmental factors (geological type, precipitation, etc.) on soil Se from both local and overall perspectives. The findings reveal a mean total soil Se of 1.76 mg/kg (95% CI: 1.540–1.974), and 91.40% (n = 287) of soil samples were classified as Se-rich (0.4–3 mg/kg). Organic matter (OM), elevation, slope, and the topographic wetness index (TWI) exhibited non-stationary spatial relationships with Se. The spatial variation trend of precipitation corresponds with the local R² values between Se and elevation, indicating that precipitation may strengthen the association between elevation and Se distribution. Geological type and rainfall were identified as key driving factors affecting soil Se content within the study area, particularly through their interactions with OM. Overall, the synergistic effects of geological type, precipitation, and OM are responsible for the accumulation of Se in the agricultural soils of Xin’an Town.

1. Introduction

Selenium (Se) is an essential trace element that plays a vital role in maintaining human health, preventing diseases, and delaying the aging process. It helps to activate tumor suppressor genes, which facilitate the repair of damaged cells, and provides strong anti-cancer effects [1]. Additionally, Se has antioxidant properties that neutralize free radicals and reduce the accumulation of peroxides, protecting cell membranes from oxidative damage. It also enhances the immune system by forming metal–Se protein complexes, which inhibit toxic heavy metals such as As, Hg, and Cd [2]. Nevertheless, the relationship between Se and human health is complex because of its dual biological effects [3]. Both Se deficiency and excess can cause serious health issues. For example, Se deficiency can cause various diseases, including Keshan disease, Kashin–Beck disease, and leukodystrophy. Conversely, excessive Se intake may result in severe and potentially fatal consequences, including skin lesions, neurological disorders, hair loss, and brain edema [4].

The average daily Se intake for humans largely depends on the Se content in food, which is determined by the Se levels in agricultural soils. Therefore, the concentration of Se in these soils indirectly affects the amount of Se ingested by humans. However, the global average Se content in soil is just 0.4 mg/kg [5]. Meanwhile, approximately 15% of the population (500–1100 million) is facing health risks associated with Se deficiency [6]. China is a typical Se-deficient country, with 39–61% of residents consuming daily Se levels below the WHO/FAO-recommended value (26–34 μg/day) [7]. Therefore, it is crucial to investigate the spatial distribution and impact factors of Se content in agricultural soils to help the population meet their Se needs.

Se in the soil is influenced by a variety of factors, including parent material, climate, soil physicochemical properties, soil type, topography, human activities, and biological processes [8,9,10]. The natural concentration of Se in the soil is directly linked to the elemental composition of the underlying bedrock. Therefore, the parent material largely determines the Se content of the soil [11]. The bedrock of high-Se soils is generally rich in Se. For instance, the Se-rich soil found in Ziyang, China, developed from carbonaceous slate and black shale (rich in Se) of the Ediacaran to early Cambrian [12]. Similarly, in northeastern Brazil, Se-rich parent rocks are considered a major large-scale factor influencing soil Se concentrations [13]. Rainfall is also a key factor in regulating soil Se levels. Natural and anthropogenic activities release substantial amounts of Se into the atmosphere, which is subsequently deposited into the soil through rainfall (wet deposition). In addition, the migration and solubility of Se are greatly affected by soil physicochemical properties. Organic matter (OM) exhibits a strong binding affinity for Se, with 40–50% of total Se in soil bound to OM [14]. Soil pH regulates the migration of Se by altering its redox state, which indirectly affects Se concentrations [9]. Soils of different types exhibit distinct properties, which subsequently affect the variations in Se content. In Northeast China, black soils and dark-brown earth are characterized by high OM content, which enhances Se accumulation in these soils [15]. Topography affects the distribution of soil Se by controlling the movement of minerals, water, and energy. In an area with severe soil erosion, the soil Se content may not be high. Human activities (e.g., fertilization, industrial emissions, and mining) and biological processes (especially microbial transformation and plant uptake) can also modify soil Se, although these effects are often localized.

In recent years, researchers have examined the spatial distribution of soil Se and its environmental drivers across diverse climatic and geological contexts. In arid to semi-arid regions where alkaline soils are widespread, Liu et al. identified parent material as a primary factor affecting topsoil Se [16]. In the black soil region of Northeast China under a temperate continental climate, Gong et al. showed that depositional environments can strongly impact Se spatial patterns [17]. Liu et al. demonstrated that in subtropical monsoon regions with a low-Se geochemical background derived from Mesozoic continental clastic rocks, OM exerts a stronger influence on Se variability [9]. However, most existing studies are based on large-scale surveys and often focus on arid climates or low-Se geological settings. In township-scale cropland systems with intense rainfall and heterogeneous geological types, quantitative evidence identifying the dominant controlling factor affecting soil Se spatial differentiation remains limited. In addition, many studies rely on a single analytical method, making it difficult to simultaneously consider both overall driving factors and local variations. Conventional global models (e.g., OLS and global spatial regression) often assume spatial stationarity and may mask local heterogeneity, while exploratory spatial statistics (e.g., Moran’s I) describe clustering but do not quantify the explanatory power of multiple drivers or their interactions. Geodetector can quantify factor effects and interactions at the global scale but provides limited insight into local spatial variations. Geographically weighted regression (GWR) can capture spatially non-stationary relationships but is less suitable for globally ranking key drivers and their interactions. Due to variations in natural background, Xin’an Town may exhibit pronounced small-scale spatial heterogeneity. Therefore, this study integrates Geodetector and GWR, leveraging their complementary strengths to analyze the driving mechanisms of Se distribution in farmland topsoil.

Based on 314 cultivated topsoil samples in Xin’an Town, this study integrates GWR and Geodetector to discuss the influence of various environmental factors (geological type, precipitation, soil type, pH, OM, elevation, slope, and topographic wetness index (TWI)) on soil Se. The research objectives are to (1) investigate the spatial distribution of Se; (2) quantify the effects of environmental factors on Se distribution from multiple perspectives and identify the key driving factors; and (3) provide explanations for the mechanism of the variations in the spatial distribution of Se in agricultural soils. This study provides transferable evidence and an analytical framework for regions with similar climates and geological types, while directly informing cropland management and resource allocation.

2. Materials and Methods

2.1. Study Area

Xin’an Town is located in the northwest of Linli County, Hunan Province, China (111°24′–111°33′ E, 29°35′–29°43′ N; Figure 1). It borders Shimen County to the west and Lixian County to the east, encompassing a total area of 59.59 km². The town experiences a humid monsoon climate transitioning from mid-subtropical to northern subtropical, characterized by distinct seasons and moderate temperatures, with an average annual temperature of 16.4–17.8 °C, and a frost-free period of 336.4 days. Xin’an Town is situated on a piedmont plain, with a topographical characteristic of higher elevations in the western and northern regions and lower elevations in the eastern and southern regions. The geological types of the town include Quaternary, Cretaceous, Permian, Triassic, Devonian, and Silurian, with the predominant geological type being Quaternary, developed in river valley regions (Figure 1b). Land use in Xin’an Town is dominated by agricultural land and construction land. The main soil types are paddy soil, fluvo-aquic soil, and red soil (Figure 1c). Xin’an Town has fertile land, a strong industrial base, and abundant mineral resources, including coal, limestone, and iron ore.

2.2. Sample Collection and Pretreatment

The survey comprehensively considered factors such as parent material, topography, geomorphology, and soil type in the design of sampling points. The samples were collected using S-shaped or X-shaped sampling methods. A total of 314 topsoil samples (0–20 cm) were obtained from farmland in Xin’an Town (Figure 1a), and the average sampling density was 13 topsoil samples/km².

After sample collection, debris such as weeds, plant roots, gravel, bricks and fertilizer clumps were removed. The samples were then thoroughly dried in a drying chamber maintained at 55 °C. The dried samples were processed using a jaw crusher to a particle size of 0.84 mm (20 mesh) for direct pH analysis. Subsequently, the samples were finely ground to 0.074 mm (200 mesh) using a multi-head agate planetary ball mill for soil elemental content analysis.

2.3. Chemical Analysis and Quality Control

Total selenium (Se) in soil was determined according to the Analytical Methods for Regional Geochemical Samples (DZ/T 0279.14—2016) using hydride generation–atomic fluorescence spectrometry (HG–AFS) [18]. A 0.5 g soil sample was moistened with a small volume of deionized water, and 10 mL HNO₃ (1.42 g/mL) was added. The sample was digested under low-temperature heating. After the reaction subsided, 2 mL HClO₄ (1.67 g/mL) was added, and the mixture was heated at 100 °C for 30 min. Heating was continued until white fumes appeared. While hot, 5 mL HCl (1.19 g/mL) was added. After standing briefly, 10 mL HCl (HCl: H₂O = 1:1) was added to meet hydride generation requirements. The digest was quantitatively transferred to a 50 mL volumetric flask, amended with 5 mL of ferric salt solution (10 mg/mL), diluted to volume with deionized water, and mixed thoroughly. Se was quantified using an atomic fluorescence spectrometer (AFS–830; Jitian, Beijing, China). Notably, this procedure measures total Se following complete digestion and does not differentiate Se valence states, chemical species, or bioavailable fractions. Therefore, the results are reported solely as total soil Se concentrations.

Se standard stock and working solutions were prepared using an external calibration approach. Briefly, 20 mL HCl (HCl: H₂O = 1:1) was added to a 50 mL volumetric flask. Aliquots of 0.00, 0.50, 1.00, 2.00, 5.00, and 10.0 mL of the Se working standard solution were then transferred into separate flasks. After the addition of 5 mL ferric salt solution, each flask was brought to volume with deionized water, mixed thoroughly, and allowed to stand for approximately 30 min. The Se fluorescence intensity in each sample solution was measured and recorded. A calibration curve was constructed by linear regression of fluorescence intensity.

Quality control (QC) was applied throughout field sampling and laboratory analysis. Field duplicate samples accounted for 3% of the total samples. In the laboratory, 5% of samples (n = 17) were randomly selected for blind duplicate analyses. Blank solution measurements were performed for every 15 soil samples. In the process of sample determination, soil reference materials (GBW07423 (GSS–9), GBW07430 (GSS–16), etc.) were selected for quality control. The Se detection limit was 0.002 µg/g. Recoveries for certified reference materials ranged from 95.3% to 101.2%, with an RSD < 5%, indicating good analytical accuracy and precision. Because HG–AFS can be influenced by matrix effects and acidity variability, potential interferences and instrumental drift were monitored using blanks, reference materials, and duplicate samples. Batches that did not meet QC criteria were reanalyzed or recalibrated according to established protocols.

2.4. Data Source

The geological map for the study area is derived from the regional geological map of the Changde sheet (1:250,000). Digital Elevation Model (DEM) data with a resolution of 30 m were acquired from the China Geospatial Data Cloud (http://www.gscloud.cn/, accessed on 15 January 2025). The terrain factors (elevation, slope, topographic wetness index (TWI)) were processed and calculated using ArcGIS 10.7 (Esri, Redlands, CA, USA). The TWI and slope are expressed in Equations (1) and (2), respectively. Soil data were sourced from the China Soil Science Database (http://vdb3.soil.csdb.cn/ accessed on 25 January 2025). Mean annual precipitation was obtained from https://www.resdc.cn/ (accessed on 10 March 2025). The data were extracted and spatially interpolated to generate the precipitation distribution map. Terrain data were extracted using the ArcGIS 10.7 “Extract Values to Points” tool. A small number of sampling points failed to return valid values and were coded as −9999. Consequently, the effective sample size for analysis of soil Se across topographic factors was smaller than the total number of samples collected. Given that missing values were rare (2 for elevation, 2 for slope, and 3 for TWI), the influence on parametric statistical results was considered negligible.

$$\mathit{TWI}=\mathit{ln} {\displaystyle \frac{\alpha }{\mathit{tan} \beta }}$$

(1)

where α is the upslope contributing area per unit length of the contour, used to estimate how water accumulates in an area with elevation differences; β denotes the slope at a given point:

$$\begin{gathered} \beta = \mathit{arctan} \sqrt{p^2 +{ q}^2} \\ p = {\displaystyle \frac{\mathit{∆}z}{\mathit{∆}x}}; q = {\displaystyle \frac{\mathit{∆}z}{\mathit{∆}y}} \end{gathered}$$

(2)

where Δz is the elevation difference in the vertical direction; Δx and Δy are the differences in distance in the horizontal direction.

2.5. Semi-Variance Function

The semi–variance function is a method used in geostatistics to analyze and examine the spatial variability in regional variables. It can provide input parameters for performing ordinary kriging interpolation and is considered the foundation of accurate kriging interpolation. In the process of semi-variance analysis, the semi–variogram models are used to fit the semi–variance function and describe the variability in spatial data. The semi–variogram models include spherical, exponential, Gaussian, and linear models. Selection of the optimal model follows the principle of the maximum fitting coefficient (R²) and the minimum residual sum of squares (RSS). The nugget (C₀) is the sum of the measurement error and micro-scale variability at the origin, which indicates the variation caused by non-sampling intervals, known as random variation. The sill (C₀ + C) is the height reached by the semi–variogram when it first attains a stationary state, representing the maximum variation in the spatial system, which includes the total variation due to both random and structural factors. The nugget-to-sill ratio (C₀/(C₀ + C)) represents the proportion of random variation to the total variation and is considered an indicator of the spatial correlation of the element.

2.6. Geodetector

Geodetector is a statistical approach designed to detect spatial heterogeneity and identify the driving forces [19]. The core concept of the method is to evaluate the similarity in the spatial distribution of two variables based on the spatial stratified heterogeneity. Geodetector excels at analyzing categorical data. It can also perform statistical analysis on ordinal, ratio, and interval data through appropriate discretization, possessing the ability to simultaneously detect both numerical and qualitative data.

Geodetector consists of the factor detector, the interaction detector, the ecological detector, and the risk detector. In this study, Se is used as the dependent variable, and environmental factors (geological type, soil type, precipitation, pH, organic matter (OM), elevation, slope, and TWI) are used as independent variables. The factor detector and interaction detector are employed to quantitatively analyze the impact of individual environmental factors and their interactions on Se content distribution. The expression for the factor detector is as follows:

$$\begin{gathered} q = 1 - {\displaystyle \frac{{\mathit{\sum }}_{h=1}^L{N}_h{\delta }_h^2}{N{\delta }^2}} = 1 - {\displaystyle \frac{\mathit{SSW}}{\mathit{SST}}} \\ \mathit{SSW} ={\mathit{\sum }}_{h=1}^L{N}_h{\delta }_h^2; \mathit{SST} = N{\delta }^2 \end{gathered}$$

(3)

where h (h = 1, …, L) is the strata of independent variable (X) or dependent variable (Y); N_h and N are the number of units in the h-th stratum and the total study area, respectively; ${\delta }_h^2$ and δ² are the variances of Y in the h-th stratum and the total study area, respectively. SSW and SST refer to the within sum of squares and the total sum of squares, respectively. The value of q is in the range of [0, 1], where q = 0 suggests no coupling between X and Y, and q = 1 indicates that Y is entirely determined by X. A higher q value reflects that the independent variable explains the dependent variable more effectively, whereas a lower q value indicates weaker explanatory power.

Interaction detection is utilized to examine the effects of interactions between two risk factors Xs. In detail, it analyzes whether the interaction between risk factors X₁ and X₂ enhances or diminishes the explanatory power for Y, and to what extent. It also assesses whether X₁ and X₂ have independent effects on Y. There are five types of interaction, as shown in

Table 1. Interaction types between factors.

Interaction Types	Criterion
Nonlinear-weaken	$q(X_1\cap X_2) \mathit{<} \mathit{Min}(q\left(X_1\right),q\left(X_2\right))$
Uni-weaken	$\mathit{Min}\left(q\left(X_1\right),q\left(X_2\right)\right) \mathit{<} q\left(X_1\mathit{\cap }{X}_2\right) \mathit{<} \mathit{Max}\left(q\left(X_1\right),q\left(X_2\right)\right)$
Bi-enhance	$q(X_1\mathit{\cap }{X}_2) \mathit{>} \mathit{Max}(q\left(X_1\right),q\left(X_2\right))$
Independent	$q(X_1\cap X_2)=q\left(X_1\right)+q\left(X_2\right)$
Nonlinear-enhance	$q(X_1\cap X_2) \mathit{>} q\left(X_1\right)+q\left(X_2\right)$

.

Based on discretization methods in ArcGIS 10.7 and classification methods in relevant standards, we conducted discretization tests on continuous data for the independent variables. The different classification methods for continuous data may impact the results. Classification effectiveness was evaluated using the q-statistic from Geodetector, where a higher q value indicates a better discretization method [20]. The classification method with the maximum q value was selected for impact factor analysis. The optimal classification methods for each impact factor are summarized in

Table 2. The classification results of impact factors.

Factor	1	2	3	4	5	6	Classification Method
Precipitation (mm)	2039.49–2058.65	2058.66–2072.28	2072.29–2090.84	2090.85–2107.60	2107.61–2164.63	–	Natural breaks
pH	4.63–6.90	6.91–7.59	7.60–7.88	7.89–8.28	–	–	Quantile
OM (g/kg)	≤10	10–20	20–30	30–40	>40	–	[9]
Elevation (m)	16–56	57–98	99–153	154–210	211–345	–	Natural breaks
Slope (°)	0–4.61	4.62–9.95	9.96–16.95	16.96–25.43	25.44–46.98	–	Natural breaks
TWI	3.46–6.01	6.02–7.84	7.85–20.13	–	–	–	Quantile
Geological type	Qp2b	Qhal	Qhlal	K2l	T1d	P2m + q + l	[21]
Soil type	Fluvo-Aquic soil	Red soil	Paddy soil	–	–	–	Chinese Soil Classification System

.

2.7. Geographically Weighted Regression

Geographically weighted regression (GWR) is a spatial regression technique widely utilized to explore spatial relationships and reveal spatial non-stationarity [22]. It calculates local regression coefficients through regression models at each sampling point, with the regression coefficients of various impact factors varying according to spatial location. Unlike traditional regression models, GWR can quantitatively identify the effects of factors on the dependent variable across different directions and degrees, thereby better accounting for spatial heterogeneity and non-stationarity with higher accuracy [23]. The GWR model is represented by Equation (4):

$$y_i ={ \beta }_0({\mu }_i, v_i) +\sum _{k=1}^m{\beta }_k({\mu }_i, v_i)x_{ik} +{ \epsilon }_i$$

(4)

where y_i represents the dependent variable at location i, (u_i, v_i) denote the geographic coordinates of point i, β₀(u_i, v_i) is the intercept of the model at location i, β_k(u_i, v_i) indicates the local regression coefficient of the independent variable at location i, x_ik is the value of the independent variable at location i, and ε_i is the error associated with the model at location i.

In this study, Se is used as the dependent variable, and environmental factors (precipitation, pH, OM, elevation, slope, and TWI) are used as independent variables. For model construction, the Akaike information criterion (AIC) was used to determine the optimal bandwidth for model fitting between soil Se content and environmental factors. Additionally, the kernel function radius was fixed, and the Gaussian kernel was selected as the kernel function. The local fitting results of the GWR model were visualized as local coefficients of determination (R²).

2.8. Statistical Analysis

Data organization and preliminary statistical analysis were conducted using Excel 2010 (Microsoft, Redmond, WA, USA). IBM SPSS Statistics 27 (IBM, Armonk, NY, USA) was employed to conduct a descriptive statistical analysis of the data. The Kolmogorov–Smirnov (K–S) test, which is a non-parametric statistical method, was used to verify the normality of the data. Semi-variance function fitting and spatial distribution prediction were performed by GS⁺ 9.0 (Gamma Design Software, Plainwell, MI, USA) and ordinary kriging in ArcGIS 10.7, respectively. The Kruskal–Wallis H (K–W H) test was employed to assess the differences in Se content between various geological and soil types. To normalize the data, the logarithm of each parameter was calculated. The relationships between continuous variables and Se content were evaluated using Pearson’s linear correlation analysis (p < 0.05) and the GWR model, implemented in Origin 2021 (OriginLab, Northampton, MA, USA) and the ArcGIS 10.7 Spatial Modeling toolbox, respectively. The impact factors of soil Se were quantitatively evaluated by Geodetector, using the Excel-based Geodetector software available at http://www.geodetector.cn (accessed on 16 March 2025). The drawing of various maps in the study area was carried out by using ArcGIS 10.7, Origin 2021, and Adobe Photoshop 2021 (Adobe Systems Incorporated, San Jose, CA, USA).

3. Results and Discussion

3.1. Se Content in Topsoil

The selenium (Se) content in the agricultural soils of Xin’an Town ranges from 0.37 to 26.14 mg/kg, with an average of 1.76 mg/kg (Figure 2a). This concentration is 4.69 times the national background value (0.29 mg/kg) and 3.14 times the level in agricultural soils in Hunan Province (0.56 mg/kg) [24,25]. According to the standards established by the China Geological Survey, Se-rich soil is defined by a Se content of at least 0.4 mg/kg [26]. The average Se content in the agricultural soils of Xin’an Town is 4.4 times higher than this threshold, indicating that soils in Xin’an Town have significant Se-rich characteristics. The coefficient of variation (CV) is used to reflect the uniformity of element distribution in the soil. A smaller CV suggests that the element is more uniformly distributed spatially. Wilding categorized the CV into three levels: weak variation (CV less than 10%), moderate variation (CV between 10% and 100%), and strong variation (CV greater than 100%) [27]. Soil Se content showed strong variation (CV = 111%) in the study area. The results indicate significant differences in Se content across the agricultural soils in Xin’an Town, which might be affected by external factors. Soil Se concentrations were non-normally distributed according to the K–S test (p < 0.01). After logarithmic transformation, the distribution more closely approximated normality (Figure 2b), indicating that the log-transformed soil Se concentrations met the requirements for geostatistical modeling. To benchmark our results against comparable studies, we compiled summary statistics (mean, median, and range) of total Se concentrations in surface soils across diverse global regions (Supplementary Table S1). Across these studies, mean Se concentrations were generally <1 mg/kg, but values exceeded 2 mg/kg in localized high-concentration areas, indicating pronounced regional heterogeneity in soil Se.

3.2. Spatial Distribution of Se in Topsoil

Semi-variogram model fitting was conducted using GS⁺, and the results are presented in

Table 3. Parameters of semi-variogram model for soil Se.

Model	C0	C0 + C	C0/C0 + C	Range/m	R2	RSS
Gaussian	0.064	1.521	0.958	16,680	0.990	3.065 × 10−3
Linear	0.002	0.394	0.994	5270	0.926	0.023
Spherical	0.002	1.063	0.998	21,100	0.921	0.025
Exponential	0.001	1.729	0.999	63,300	0.909	0.029

. Among the models tested, the Gaussian model demonstrated the best fit, with the highest R² value (0.990) and the lowest RSS (3.065 × 10⁻³⁾. Therefore, the Gaussian model was selected as the optimal model to analyze the spatial distribution of Se in the agricultural soils of Xin’an Town. A nugget effect of less than 25%, between 25% and 75%, and greater than 75% represents strong spatial correlation, moderate spatial correlation, and weak spatial correlation, respectively [28]. Under the Gaussian model, the nugget effect value of the study area is 0.958, indicating weak spatial correlation. This result implies that there is high spatial variability in soil Se within the study area, largely driven by random factors. It may be related to the fact that the study area is farmland, which is more susceptible to human activities.

Based on the Gaussian model, the Kriging method was employed to interpolate the Se content. The Se content in the agricultural soil of Xin’an Town is unevenly distributed due to the differences in parent materials, topography, and other influencing factors. The areas with high Se content are predominantly found in the northwest of the study area, whereas the areas with low Se content are concentrated in the eastern part (Figure 3a). Overall, the Se content in the study area decreases gradually from west to east. Following the classification criteria for geochemical classes of soil Se, the Se content in the agricultural soils of Xin’an Town is classified as follows: Se-sufficient (0.175–0.4 mg/kg), Se-rich (0.4–3.0 mg/kg), and Se-excessive (>3.0 mg/kg) [26]. As illustrated in Figure 3b, no soils in the study area are categorised as Se-marginal (0.125–0.175 mg/kg) or Se-deficient (<0.125 mg/kg). The majority of the study area (91.40%) is Se-rich area. The Se-excessive area comprises 8.40% of the total and is mainly situated in the northwest corner. The Se-sufficient area is minimal, accounting for just 0.6% of the total, and is essentially negligible. The Se content in the agricultural soil of Xin’an Town is concentrated between 0.40 and 3.00 mg/kg (Se-rich). This indicates that farmland in Xin’an Town may have the potential to cultivate naturally Se-rich crops under appropriate development conditions.

3.3. Relationship of Geological Type, Soil Type and Precipitation with Se Content in Soils

The parent material is the foundational substance for soil formation, significantly influencing the elemental composition and physical properties of the soil. Due to the different rock types and mineral compositions within various geological backgrounds, the Se content of soils formed through natural weathering shows significant variation. Here, the Se content in the soil of various strata is summarized (

Table 4. Statistical parameters of topsoil Se content according to geological type, soil type, and mean annual precipitation.

Topsoil Se	Max (mg/kg)	Min (mg/kg)	Median (mg/kg)	Mean (mg/kg)	SD
Qp2b	5.26	1.01	2.25	2.40	0.81
Qhal	2.15	0.45	0.68	1.02	0.52
Qhlal	5.70	0.37	1.22	1.28	0.48
K2l	9.37	1.54	4.19	4.80	2.65
T1d	12.33	2.34	7.34	7.34	7.06
P2m + q + l	26.14	3.77	14.96	14.96	15.82
Fluvo-aquic soil	2.08	0.45	0.60	0.75	0.43
Red soil	1.54	0.46	0.84	0.95	0.55
Paddy soil	7.90	0.37	1.42	1.65	0.92
2000 < MAP < 2050	2.35	0.37	1.17	1.19	0.31
2050 < MAP < 2100	5.70	0.45	1.27	1.27	0.58
2100 < MAP < 2150	12.33	1.33	2.54	3.16	2.02
2150 < MAP < 2200	26.14	2.34	8.80	11.52	10.21

and Figure 4a). The results of the K–W H test show that there is a significant difference (p < 0.01) in soil Se content between the different geological types. The soil Se content in the Middle Permian (Maokou Formation, Qixia Formation, and Liangshan Formation) is the highest, with an average of 14.96 mg/kg, followed by the Daye Formation and Luojingtan Formation. Research has found that the weathering of Se-rich rocks can result in the development of Se-rich soil [29]. During the Permian period, intense tectonic activity and frequent large-scale extinction events resulted in the formation of large quantities of black rocks. The Middle Permian (Maokou Formation, Qixia Formation, Liangshan Formation) is predominantly composed of black carbonaceous shale, carbonate rocks, and sandstone. The black rock series are major geological bodies of Se enrichment on Earth [30]. However, Xia et al. found that carbonate rocks and sandstone are typical low-Se rocks when comparing Se content between different rock types [31]. It is worth noting that carbonate rocks undergo extensive weathering during soil formation. The substantial dissolution and loss of carbonate rocks result in a decrease in the volume of weathering residue. The weathering products comprise clayey particles with high maturation and sticky texture, which contribute to retaining Se from the parent rocks in the weathering residue. Similarly, Qin et al. observed that although the Se content of carbonate rocks is relatively low in the Karst Area of Guangxi, China, the soil developed from carbonate rocks can exhibit higher Se levels [32]. Therefore, the high Se content in the soil from the Middle Permian (Maokou Formation, Qixia Formation, Liangshan Formation) substrate is most likely attributed to the inheritance from Se-rich black carbonaceous shale and the concentration effect of carbonate rocks during weathering [33].

The lowest Se content was observed in soils developed from Holocene alluvial deposits and Holocene lacustrine alluvial deposits, which were 1.02 mg/kg and 1.28 mg/kg, respectively. Previous research suggests that the older the parent rock strata, the more favorable it is for Se enrichment [34]. The Holocene is the youngest period in geological history. In addition, the soil of Holocene alluvial deposits and Holocene lacustrine alluvial deposits is generally formed by alluvial and fluvial processes. During soil formation, intense leaching causes a substantial loss of Se. Consequently, soils derived from Holocene alluvial deposits and Holocene lacustrine alluvial deposits exhibit relatively low Se content.

The stratigraphic distribution and the spatial distribution of Se content in the study area are compared in Figure 5a. It is evident that the distribution pattern of low soil Se content coincided with the southern region of Holocene alluvial deposits (Figure 5a). This observation indicates that soil inherits the geochemical characteristics of its parent rock. However, the characteristics of Se content in the north of Holocene alluvial deposits are inconsistent with stratigraphic changes, suggesting that the parent material alone cannot account for the differences in soil Se content. Se content in the study area was also influenced by other factors. For Holocene alluvial deposits, the Se content was clearly higher in areas with high precipitation (Figure 5b). A comparison between Figure 5a and Figure 6a indicates that precipitation in the north of the Holocene alluvial deposits exceeds that in the south, suggesting that rainfall is an important factor contributing to the differences in soil Se content between these areas.

Table 4 presents the statistical parameters of Se content for different soil types, categorized according to the Chinese Soil Classification System (CSCS). The boxplot of Se content for various soil types is shown in Figure 4b. The results of the K–W H test indicate a significant difference in soil Se content between the various soil types (p < 0.01). The average Se content of different soil types is ranked as follows: paddy soil (1.65 mg/kg) > red soil (0.95 mg/kg) > fluvo-aquic soil (0.75 mg/kg). Soil properties are different for different soil types. Fluvo-aquic soil belongs to a sandy soil with low organic matter (OM) content. Due to the influence of a shallow water table, it has a strong leaching effect and is prone to Se loss. In contrast, paddy soil has been in a waterlogged state for a long time, which enhances its ability to retain water and nutrients. The soil formed under long-term waterlogging conditions has a higher content of OM and clay particles, which is helpful for the immobilization of Se. Furthermore, red soil is characterized by abundant Fe–Al oxides, high development, and low alkalinity. In the study area, the mean Se content of fluvo-aquic soil is lower than that of paddy soil and red soil, in line with the findings of Jiang et al. and Yue et al. [35,36].

Regardless of changes in geological type or soil type, Se content in soils from areas with higher precipitation was consistently greater than that in areas with lower precipitation (Figure 4). In the study area, the average annual precipitation ranges from 2040 to 2159 mm (Figure 6a). Within the narrow precipitation range, soil Se content was calculated at different precipitation levels using 50 mm intervals. The results indicate that an increase in precipitation significantly enhances the average Se content in soils (Figure 6b). This finding further suggests that precipitation is a critical factor influencing soil Se content.

3.4. Relationship Between Soil Physicochemical Properties and Se Content

Soil pH is a fundamental physicochemical indicator that plays an important role in constraining the physical, chemical, and biological reactions occurring in the soil. The appropriate pH can increase the quality and yield of crops, whereas excessively acidic or alkaline soils may disrupt soil structure [24]. The pH in the study area ranges from 4.63 to 8.28, averaging at 7.21, and the soil is predominantly alkaline (Figure 7a). Many studies have reported a negative correlation between soil Se content and pH [37,38,39]. This relationship is attributed to the fact that in acidic soil environments, Se exists primarily in the form of selenite (Se⁴⁺), which has limited mobility and is readily adsorbed or complexed by oxides, clay minerals, and OM in the soil [40]. In an alkaline soil environment, selenate (Se⁶⁺) is often the predominant form. Selenate (Se⁶⁺) has strong bioactivity and high solubility, making it easily absorbed by plants, which leads to a decrease in Se content in the soil [41]. In this study, we found that pH is weakly negatively correlated with soil Se (Figure 7c), although this correlation is statistically significant (p < 0.05, r = −0.14). The local R² values for Log10(pH) and Log10(Se) are also generally low (Figure 7b), indicating that pH explains little spatial variability in Se. Previous studies suggest that pH influences Se primarily by regulating bioavailable fractions rather than total Se. In paddy soils from Luxi County, Xu et al. reported that pH primarily regulates Se adsorption–desorption and Se(IV)–Se(VI) interconversion, rather than exerting simple linear control on total Se [42]. In addition, the study area is dominated by paddy soils that readily develop reducing conditions. In this environment, redox-driven reduction and immobilization may dominate Se retention, thereby attenuating the apparent influence of pH [43,44]. A regional-scale study indicated that pH generally has limited independent explanatory power for spatial variability in total Se content in soil and is rarely a primary controlling factor [45]. However, it may help explain the generally low local R² values in this study. Notably, Pearson’s correlation analysis summarizes the overall linear trend. Figure 7d shows a slight increase in Se when pH ranges from 5 to 7, indicating that Se trends vary across pH intervals, further weakening the overall linear correlation.

Soil OM is a class of special, complex, and stable macromolecular organic compounds formed by organic residues under the action of microorganisms. As a typical biophilic element, Se tends to accumulate in soil with high OM content and strong bioactivity. The OM content in the study area ranges from 11.17 to 54.40 g/kg (Figure 8a). There is a significant positive correlation between OM and Se (Figure 8c). This relationship is primarily attributed to the adsorption and immobilization of Se by OM. Initially, Se is converted into forms that plants can directly absorb from the soil. After plants absorb and accumulate Se, the decomposition of plant residues results in Se combining with humus to form organic-bound Se, which usually has low mobility. Therefore, Se is effectively adsorbed in the soil. Furthermore, an increase in soil OM facilitates the formation of soil aggregates, which enhances the specific surface area of the soil. As a result, more adsorption sites are available for Se, enabling it to bind more closely with soil components such as OM, clay minerals, and iron oxides, thereby becoming immobilized. The combined effects of adsorption and fixation can maintain the stability of Se in the soil, making it less susceptible to leaching or transport. Furthermore, Se content increased significantly with rising OM levels at OM < 35 g/kg. However, the increasing trend of soil Se content was slowed down at OM > 35 g/kg, and a sudden decrease in soil Se content was observed when OM reached 40–45 g/kg (Figure 8d). This result is in agreement with the findings of Liu et al. and Pang et al. [46,47]. The observed trend may be due to the volatilization of organoselenium compounds resulting from microbial activity [48]. In environments rich in OM, active microbial processes enhance biogeochemical cycling, which can reduce local soil Se accumulation through two pathways: by promoting the formation and volatilization of organic Se compounds [49] or by increasing dissolved organic carbon (DOC) levels, which facilitates the formation of soluble Se-DOM complexes, thereby enhancing Se migration, redistribution, or leaching [50]. Soil OM content in the Holocene alluvial deposits is lower than that in other regions, while the local R² of Log₁₀(Se) and Log₁₀(OM) is generally higher (Figure 8b), leading to the lowest soil Se content in the Holocene alluvial deposits (Figure 2a). Gong et al. also observed a similar phenomenon [51].

3.5. Relationship Between Topography and Se Content

Topography indirectly regulates the properties and composition of soil by influencing soil formation processes and vegetation distribution. Elevation and slope are fundamental topographic factors that influence physical transport on the Earth’s surface and are used to reflect the degree of terrain undulation. The topographic wetness index (TWI) is used to evaluate the water accumulation capacity at a given location, thereby reflecting the soil’s moisture. In this study, three fundamental topographic parameters—elevation, slope, and TWI—were extracted to explore the influence of topography on soil Se.

According to the classification standards of Chinese geomorphologic types, the study area is divided into plains (<200 m) and low-relief mountains (200–500 m) [52]. The low-relief mountains are predominantly located in the southern part of the study area, while the majority of the remaining regions consist of plains (Figure 9a). A significant positive correlation (r = 0.542, p < 0.05) is observed between elevation and Se in Figure 9c, in line with Liu et al. [9]. Moreover, soil Se content is higher in low-relief mountainous areas (200–500 m) compared with plains (<200 m) (Figure 9d), consistent with the findings of Xu et al. regarding the impact of topography on soil Se concentrations in Yongjia County [53]. However, Shang et al. reported that Se was transported to the low-lying areas through scouring action by surface runoff, which led to a gradual decrease in soil Se content with increasing elevation [24]. The spatial variation trend of precipitation in the study area is highly consistent with the distribution of local R² values for Se and elevation based on geographically weighted regression (GWR) (Figure 5a and Figure 9b). Specifically, precipitation decreases progressively from west to east, and the local R² values of Log₁₀(Se) and Log₁₀(elevation) show a gradual decline. This suggests that rainfall may enhance the effect of elevation on the spatial distribution of Se. Therefore, we speculate that this discrepancy may be attributed to the fact that Se-rich rocks are developed in low-relief mountainous areas, which serve as the source region. Rainfall transports and accumulates Se-rich weathered materials to the foot of the mountain, which affects the movement of Se, leading to higher Se concentrations in the low-relief mountainous areas compared with the plains.

In accordance with the comprehensive control of soil and water conservation general rule of planning, the slope of a study area is classified as a gentle slope (<5°), moderate slope (5–15°), steep slope (15–25°), and sharp slope (>25°) [54]. Steep slopes and sharp slopes are primarily distributed in the south of the study area (Figure 10a). As illustrated in Figure 10c, Se exhibits a significant positive relationship with the slope (r = 0.279, p < 0.01). With the steeper slope, the soil Se content is higher (Figure 10d). This finding is inconsistent with Yan et al., who reported that severe soil erosion is most often associated with steep slopes [55]. Greater slope steepness increases erosion intensity, promoting particle-bound Se transport and reducing soil Se content. The difference is likely due to backflow phenomena in steep slope areas, where water flow is impeded or rebounds below steeper regions. This leads to the deposition of dissolved Se or suspended particles, thus increasing the Se content in soils on steep slopes. Furthermore, the development of Se-rich rocks in steep and sharp slope areas significantly enhances Se content in the surrounding soils. The high R² values of Se and slope are concentrated in regions with greater OM content (49–54.4 g/kg) (Figure 8a and Figure 10b), suggesting that under the influence of OM, areas with steeper slopes are more effective in retaining Se, thereby strengthening the correlation between slope and Se.

The TWI was proposed by Kirkby in 1975 to quantitatively describe the spatial distribution of soil moisture in a region [56]. A higher TWI indicates a higher soil moisture content. In gentle slope areas, water flow velocity is slow, which typically facilitates water accumulation. Consequently, a high TWI is often found in areas with gentler slopes (Figure 11a). The TWI exhibits a significant negative correlation with soil Se content (r = −0.221, p < 0.01) (Figure 11c). There is an upward trend in Se content when the TWI is between 15 and 20, but we only collected seven soil samples in this range, which may lead to randomness in the results (Figure 11d). Higher soil moisture can increase surface water flow, which in turn enhances soil leaching and leads to a decrease in soil Se content. Therefore, Se content in the soil decreases with the increase in the TWI. The local R² values for Log₁₀(Se) and Log₁₀(TWI) are higher for the Daye Formation and the Middle Permian (Maokou Formation, Qixia Formation, and Liangshan Formation) (Figure 11b), likely due to the Se-rich geological characteristics of these strata. As data on soil texture and porosity were unavailable, we could not determine whether the stronger local correlations observed in certain strata are driven by differences in leaching. Future studies should integrate the TWI with indicators such as soil texture and porosity to better elucidate the underlying mechanisms. Overall, the mechanisms driving this relationship warrant further investigation.

3.6. Driving Factors Influencing the Spatial Distribution of Topsoil Se

3.6.1. Factor Detection

Factor detection is used to assess the extent to which the independent variable (X) explains the spatial variation in the dependent variable (Y). It is the preferred method when applying Geodetector to characterize the spatial differentiation of elements. The significance test is a key step in conducting factor detection [57]. The test results indicate that geological type, precipitation, soil type, OM, and elevation significantly influence the spatial distribution of Se at the 0.001 level, while pH, slope, and the TWI show significance at the 0.05 level. This suggests that all environmental factors selected for this study have a significant impact on Se content, confirming the reliability of the factor detection results. Among all factors analyzed, geological type (stratum) has the greatest impact on soil Se content (q = 0.48) (

Table 5. Factor detection results for Se in topsoil.

	Geological Type	Soil Type	Precipitation	pH	OM	Elevation	Slope	TWI
q	0.48 ***	0.13 ***	0.43 ***	0.07 *	0.20 ***	0.30 ***	0.12 *	0.03 *
p	0.000	0.000	0.000	0.010	0.000	0.000	0.030	0.040

Note: ***: p < 0.001; *: p < 0.05.

). The black carbonaceous shale developed in the Middle Permian (Maokou Formation, Qixia Formation, Liangshan Formation) provides abundant Se resources for the agricultural soils in Xin’an Town, and the soil formed through the weathering of carbonate rocks creates a favorable environment for Se preservation. Previous research has generally recognized that parent material plays a fundamental role in determining soil Se levels [51]. The factor detection results further support this conclusion. In addition to geological type, precipitation showed relatively strong explanatory power (q = 0.43), ranking second only to geological type. This indicates that precipitation plays a key role in reshaping the spatial distribution pattern of soil Se. According to the statistics, approximately 13,000 to 19,000 tons of Se cycle in the troposphere each year [58]. Se in the atmosphere reaches up to 2000 tons, and the majority enters terrestrial systems via wet deposition (precipitation) [59].

The q values of elevation, OM, and soil type fall within the range of 0.1 to 0.3, indicating that these factors are secondary drivers influencing the distribution of Se in the study area. Although pH, slope, and the TWI passed the significance test, their factor detection results exhibit a high p-value and low q value. The q values for pH and the TWI are below 0.1, which means that these parameters have little influence on soil Se content in farmland of Xin’an Town. Among the topographic parameters, elevation has the greatest impact on Se. Some studies have found that the relationship between pH and Se is not distinct, but a low pH does contribute to the adsorption of Se anions in soil, which aligns with the findings of this study [60,61].

3.6.2. Interaction Detection

Interaction detection can evaluate the influence degree of each pair of factors on Se under interaction. The effect of individual factors is not independent, and interactions exist between them (Figure 12). The interaction of any two factors has a greater impact on Se than the effect of a single factor, indicating either nonlinear enhancement or bi-enhancement. The Se content in the agricultural soils of Xin’an Town is influenced by a combination of factors, including geological type, precipitation, OM, elevation, and soil type. The q values for geological type and precipitation after interactions with other factors are 0.51–0.61 and 0.48–0.63, with an average of 0.56 and 0.53, respectively. These results show that geological type and precipitation are key drivers influencing soil Se distribution in the study area. The interaction values (q) of OM with geological type and precipitation are both more than 0.6, further confirming that in areas characterized by Se-rich bedrock, high precipitation, and abundant OM, the Se content in soil tends to be more enriched. The synergistic effects of geological type, precipitation, and OM are responsible for the accumulation of Se in the agricultural soils of Xin’an Town. Consistent with these results, comparable studies worldwide frequently identify geology/parent material, precipitation, and OM as key drivers of spatial variability in soil Se (Supplementary Table S1), supporting the reliability of the drivers identified in this study.

4. Conclusions

This study investigates the concentration and spatial distribution of Se in the topsoil of farmland in Xin’an Town and systematically analyzes the effects of geological type, precipitation, soil type, soil physicochemical properties, and topography on soil Se concentrations. The results reveal that the agricultural soil in Xin’an Town exhibits a higher degree of Se enrichment compared with the Chinese background value, with 91.40% (n = 287) of the soil samples classified as Se-rich. The high soil Se content in the Middle Permian (Maokou Formation, Qixia Formation, and Liangshan Formation) is related to the development of Se-rich black carbonaceous shale. The distribution of Se in the topsoil is closely associated with mean annual precipitation. The geographically weighted regression (GWR) results indicate that organic matter (OM), elevation, slope, and the topographic wetness index (TWI) exhibit a non-stationary spatial relationship with Se. Precipitation may strengthen the association between elevation and Se distribution. The findings from Geodetector reveal that geological type and rainfall are key driving factors affecting soil Se content in the study area, particularly through their interactions with OM. The low Se content observed in the Holocene alluvial deposits is primarily attributed to low OM, insufficient rainfall, and non-Se-rich parent rock. Se enrichment in Xin’an Town farmland soil is primarily associated with the combined effects of geological type, precipitation, and OM. In summary, the farmland in Xin’an Town represents a resource basis for developing Se-enriched agriculture. As this study did not assess Se speciation or bioavailability, further validation through measurements of bioavailable Se fractions and crop uptake is required to determine feasibility. This study provides a reference for investigating factors influencing soil Se in the farmland of humid monsoon regions within similar geological settings.