Study on the Factors Affecting the Humus Horizon Thickness in the Black Soil Region of Liaoning Province, China

Abstract

Understanding the spatial variability and driving mechanisms of humus horizon thickness (HHT) degradation is crucial for effective soil degradation prevention in black soil regions. The study compared ordinary kriging interpolation (OK), inverse distance weighted interpolation (IDW), and regression kriging interpolation (RK) using mean error (ME), mean absolute error (MAE), root mean square error (RMSE), and relative RMSE to select the most accurate model. Environmental variables were then integrated to predict HHT characteristics. Results indicate that: (1) RK was superior to OK and IDW in characterizing HHT with the smallest ME (11.45), RMSE (14.98), MAE (11.45), and RRMSE (0.44). (2) The average annual temperature (0.29), precipitation (0.27), and digital elevation model (DEM) (0.21) were the primary factors influencing the spatial variability of HHT. (3) The HHT exhibited notable variability, with an increasing trend from the southeast towards the central and northern directions, being the thinnest in the southeast. It was thicker in the northeast and southwest regions, thicker but less dense along the southern Bohai coast, thicker yet sporadically distributed in the northwest (especially Chaoyang and Fuxin), and thick with aggregated distribution over a smaller area in the northeastern direction (e.g., Tieling). These findings provide a scientific basis for accurate soil management in Liaoning Province.

1. Introduction

Black soil primarily refers to a type of soil that contains dark humic substances on its surface layer that had developed under grassland meadow vegetation in temperate and cold temperate zones [1]. Generally, black soils with thick and dark topsoil (mollic/chernic horizon) encompass various types of parent material, and include Phaeozems, Chernozems, Luvisols, Grayzems, Kastanozems, Cambisols, and Anthrosols in China [2]. Among them, Phaeozem is distinguished by its formation in temperate grassland-meadow conditions, with a thick mollic/chernic horizon and less calcium carbonate accumulation than Chernozems. Kastanozems differ in drier short-grass regions and might have a lighter color and varying organic matter [3]. The black soil area of Liaoning Province is located in the southern part of northeast China, with a total area of about 1.87 × 10⁶ hm², accounting for 10.07% of the total area, which is an important commercial grain production base in China [4].

Although the black soil in Northeast China had undergone reclamation for a brief duration, the exploitative management of these resources has accelerated soil degradation. In some areas, the humus horizon has disappeared, exposing the loess parent material. This progression from black soil to black loess, broken-skin loess, and final loess reflected the extent of erosion of the black soil [5,6,7]. Similar degradation problems have been reported in other regions, such as Southeast Moravia, Czech Republic [8], and southern Poland [9]. The degree of soil degradation was reflected in the succession of vegetation on sloping arable land, progressing from top-tier communities to secondary vegetation, miscellaneous grass communities, artificial vegetation, and ultimately barren. This degradation was primarily characterized by a reduction in the HHT, ultimately leading to a decrease in productivity [10]. Erosion can further degrade such soils, transforming them into less fertile units, such as weakly developed soils [11]. Therefore, understanding the spatial variability of the humus horizon and its driving mechanisms is crucial for formulating and implementing effective soil degradation prevention and control strategies, as well as ensuring national food security.

In recent years, the issues of “soil thinning, fertility decline, and soil hardening” have become a serious issue in the protection of black soils. The “fertility decline” refers to the decline of soil fertility, especially the decrease of organic matter content in humus horizon [10,12,13]. The “soil hardening” pertains to the deterioration of its physical structure due to mechanical compaction caused by anthropogenic activities, improper fertilizer management, accumulation of organic pollutants, and other factors. The hardening is primarily evident in the increase of soil bulk density [14,15,16]. The “soil thinning” is primarily a consequence of both the “fertility decline” and “soil hardening” processes [17]. On the regional scale, a large number of field investigations are needed to explore the situation of black soils, but the related research is rarely reported. At present, the variation of HHT in Liaoning Province is still unknown.

The HHT varied greatly across different regions, soil types, and stages of development due to their unique mineral composition, the combination pattern of formation factors, and the strength of interaction [17]. Although different soil-forming factors played distinct roles in the formation of humus horizon, the factors are interconnected. During the soil formation process, the parent material acted as the foundation, while the climate provided the necessary energy. Biological activity managed the material cycle and energy exchange, ultimately regulating the physicochemical properties [18]. Additionally, time, topography, and anthropogenic activities shaped the pace and direction [19]. Hence, the HHT varied significantly depending on the climatic zone, topographic features, and anthropogenic activities [19]. It is impractical to represent the HHT across the entire region using just one or a few profiles. Instead, it is crucial to study the overall variability of the HHT within the distribution area holistically, in order to facilitate sustainable utilization of black soil resources.

In soil science research, accurately characterizing the spatial distribution of soil properties is crucial. A statistical analysis was conducted to produce map of phosphorus (P) and potassium (K) data from 30 farmland sites, delving deeply into the relationship between the statistical characteristics of the data and the performance of various spatial interpolation methods. Their findings revealed that, after selecting an appropriate variogram model and applying logarithmic transformation to the data, the Kriging method outperformed the inverse distance weighting method on most data sets [20].

Yang et al. [21] investigated soil organic matter and compared the accuracy of ordinary, regression, and weighted regression kriging. They found that when auxiliary variables were considered, the interpolation accuracy of regression and weighted regression kriging was significantly higher than that of ordinary kriging, resulting in a more refined mapping effect [21]. Xu et al. [22] pointed out that the inverse distance weighting method has better applicability in soil erosion simulation research. Wang et al. [23] proposed a spatial random forest interpolation method with semi-variogram, which integrates the information of multi-dimensional auxiliary variables in spatial interpolation. Compared to the random forest method, distance-based random forest spatial prediction method, ordinary kriging method, and regression kriging method, this approach improves interpolation accuracy by more than 10%.

Each of these methods has its own advantages. Ordinary Kriging interpolation considers the spatial correlation between sample points and is suitable for soil properties with spatial continuity. Inverse distance weight interpolation is simple and well-suited for local interpolation tasks. Regression kriging interpolation combines the benefits of regression analysis and kriging interpolation, enabling it to simultaneously consider the relationship between soil thickness and environmental factors as well as spatial autocorrelation characteristics. Overall, these methods can effectively reveal the thickness distribution characteristics of the humus horizon in black soil.

In light of the achievements to date regarding the modeling of the HHT from around the world, including global models such as Soilgrid, the study utilizes multiple interpolation techniques to comprehensively investigate the spatial variability characteristics of HHT and its influencing factors within the black soil region of Liaoning Province. The aim is to provide a scientific basis for accurate soil management and land use.

2. Materials and Methods

2.1. The Study Area

Liaoning Province is located in the southern part of Northeast China (Figure 1). It belongs to the temperate monsoon climate zone. Due to the complex topography, the climate varies considerably across the province with long, cold winters and warm, humid summers (Figure 2). The eastern part is relatively wet while the western part is dry; the plains are windy and sunny. The average annual temperature is 5–10 ℃, decreasing gradually from the coast to the inland [4,10] (Figure 2). The black soil types are Phaeozems, Chernozems, Luvisols, Grayzems, Kastanozems, Cambisols, and Anthrosols according to the World Reference Base for Soil Resources [3].

2.2. Data Acquisition

2.2.1. Data Sources

(1)

The sampling data were collected from “Soil series of China—Liaoning Volume” [24] and “China Soil Species (Volume 2)” [25]. A total of 264 data were collected (Figure 3). The sampling period for most soil profiles was during 2008–2020.

(2)

The environmental variables were average annual temperature, land use type, DEM, vegetation type, geomorphology, and annual precipitation (Figure 2).

The DEM data (meter) in the topographic factors and the 90 m resolution DEM data were obtained from the Geospatial Data Cloud (http://www.gs−cloud.cn, 2 April 2024), an online platform that provides a wide range of geospatial data. Climate factor data (Average annual temperature and precipitation) were obtained from the Ares Environmental Science Data Platform (http://www.res−dc.cn, 3 April 2024).

2.2.2. Data Processing

(1)

The defining characteristic of soil in the black soil region is the existence of a black or dark black humus horizon, typically designated by letter A [2,3]. The thickness directly indicates the soil’s fertility and quality [2]. Therefore, in order to evaluate the HHT, the thickness of horizon A was selected as the measurement standard. Predicting the thickness of horizon A allows for a more precise assessment of the fertility status, providing crucial data support for agricultural production and soil management.

(2)

In ArcGIS, the reclassification tool can reclassify various data based on specific rules. Following reclassification, 122 samples were determined. To validate the delineated black soil region and assess its accuracy, 85 samples were randomly selected, while the remaining 37 samples were used as validation points. These validation points were chosen to evaluate the accuracy of the study results through comparative analyses.

2.3. Research Methods

2.3.1. Ordinary Kriging

OK is a statistically based spatial interpolation method that uses data from known sample points to predict attribute values in unknown regions [26].

The steps are as follows:

(1)

Calculation the distance and semi-variance: calculate the distance between known points is calculated to obtain the corresponding semi-variance.

(2)

Fitting the semi-variance function: a curve is chosen to fit the relationship between semi-variance and distance. This relationship is usually expressed through variational function models, including Gaussian, linear, spherical, damped sine, and exponential models.

(3)

Determine the range: a range is a specific distance within which attribute values are spatially correlated. Beyond this distance, the attribute values are no longer correlated.

(4)

Calculation of weights: for an unknown point, its distance from a known point within the variation is calculated and the semi-variance of the response is derived, which in turn back-calculates the weights (λ).

(5)

Performing interpolation: Using the weights (λ) and the attribute values of the known points, the attribute values of the unknown points are calculated by a linear unbiased and optimal estimation method. The formula is as follows.

$$Z_0={\sum }_{i=1}^n{\lambda }_i{Z}_i$$

(1)

where Z_i is the measured value at the ith position, λ_i is the unknown weight of the measured value at the ith position, Z₀ is the value of the attribute of the predicted position, and n is the number of measured values.

2.3.2. Inverse Distance Weight Interpolation

IDW is a spatial analysis method based on point data, which uses distance and weight to calculate the value of unknown position [27].

The IDW formula is as follows:

$$Z_{\left(x_0\right)}={\sum }_{i=1}^n{\lambda }_i{Z}_i$$

(2)

where Z_(x0) is the estimated value of position x₀, Z_i is the value of the ith known sample point, and λ_i is the weight of the ith known sample point. The weight λ_i is defined as:

$${\lambda }_i={\displaystyle \frac{d_{i_0}^{-p}}{{\sum }_{j=1}^n{d}_{i_0}^{-p}}}$$

(3)

where d_i0 is the distance between x₀ and the ith known sample point, while p is the power parameter to control the influence of distance. A larger value of p makes the influence of closer points stronger and vice versa. n is the total number of known sample points.

2.3.3. Regression Kriging Interpolation

RK is an advanced spatial interpolation method that combines regression analysis and ordinary kriging interpolation [28]. The steps are as follows:

(1)

Constructing regression model: Sample data and factors (Vegetation type, Land use type, Average annual temperature, Annual precipitation, DEM (m), and Geomorphology) were used to construct regression model, and extract the main trend of data.

(2)

Calculation residuals: Following the construction of the regression model, residuals are computed for each sample point. These residuals represent the discrepancies between the observed values and the predicted values derived from the regression model.

(3)

Ordinary kriging interpolation: The calculated residuals are interpolated using the ordinary kriging method to ensure the spatial correlation of the data is fully incorporated.

(4)

Combining trend terms and residuals: The final spatial interpolation results are obtained by integrating the trend terms derived from the regression model and the findings from the ordinary kriging interpolation analysis of the residuals.

(5)

A multiple regression model was used to fit the HHT distribution map to visualize the interpolation results and their spatial distribution.

The formula is as follows:

$$\mathrm{Y}={\beta }_0+{\beta }_1\left(x_1\right)+{\beta }_2\left(x_2\right)+{\beta }_3\left(x_3\right)$$

(4)

where Y represents the actual HHT at the site, β₀ is the constant term, and β₁, β₂, and β₃ are the estimated coefficients of the variables x₁, x₂, and x₃, respectively. The final predicted value is given by:

$$\mathrm{y}\left(p_i\right)={\sum }_{j=1}^n{\beta }_{j,\left(p_i\right)}{X}_{j,\left(p_i\right)}+{\epsilon }_{\left(p_i\right)}$$

(5)

where y(p_i) represents the precipitation value at spatial location P_i; β_j,(pi) are the coefficients of the regression model; X_j,(pi) denotes the jth independent variable at location p_i; ε(p_i) is the residual of the regression model, and n is the number of covariates, which in this case is three.

2.3.4. Random Forest Model

The Random Forest Model operate by training numerous independent decision trees, where each tree is built using a distinct subset of data and features [29]. These decision trees undergo a voting process to produce the ultimate prediction. The element of randomness in the training process gives Random Forest a strong resistance to fitting, making it perform well when dealing with datasets with a large number of features and high dimensionality (Figure 4) [30].

The random forest model was trained with a ratio of 0.8, utilizing 100 decision trees. The node splitting criterion was based on squared-error, with a minimum sample number of 2, and an unlimited maximum depth of the tree. The Random Forest algorithm was employed to construct a prediction model, using HHT as the dependent variable and the potentially major influencing factors (vegetation, land use type, mean temperature, annual precipitation, DEM, and geomorphological type) as independent variables, to provide an in-depth exploration of the environmental variables contributing to HHT.

2.3.5. Model Performance Assessment

To evaluate the prediction model’s accuracy, four error metrics were used in this study: ME, MAE, RMSE, RRMSE [30,31].

ME is an indicator used to evaluate the accuracy of a prediction model. It calculates the average value of the difference between the predicted value and the actual value. The average error provides information about the direction and magnitude of the prediction deviation, and the calculation formula is as follows:

ME:

$$M={\displaystyle \frac{1}{n}}{\sum }_{j=1}^n\left(Z_{oi}-Z_{pi}\right)$$

(6)

MAE is a measure used to evaluate the prediction accuracy of a model. It calculates the average of the absolute values of the differences between all predicted values and their corresponding actual values. The formula for calculating MAE is as follows:

$$M={\displaystyle \frac{1}{n}}{\sum }_{j=1}^n\left|Z_{oi}-Z_{pi}\right|$$

(7)

RMSE is a commonly used metric to measure the prediction error of a model. It is obtained by calculating the square root of the average of the squared differences between the predicted values and the actual values. In statistics, RMSE is widely used in regression analysis and provides a measure of prediction accuracy. The formula for calculating RMSE is:

$$R=\sqrt{{\displaystyle \frac{1}{n}}{\sum }_{j=1}^n{\left(Z_{oi}-Z_{pi}\right)}^2}$$

(8)

RRMSE is a metric used to evaluate the prediction accuracy of a model. It is calculated by standardizing the RMSE with the actual range of the observed values. This results in a dimensionless ratio, which facilitates comparison between different datasets or models. The formula for calculating RRMSE is as follows:

$$R={\displaystyle \frac{\sqrt{\frac{1}{n}{\sum }_{j=1}^n{\left(Z_{oi}-Z_{pi}\right)}^2}}{\frac{1}{n}{\sum }_{j=1}^n{\left(Z_{oi}-Z_{pi}\right)}^2}}$$

(9)

where: n represents the number of sampling points in the verification data subset; Z_oi and Z_pi represent the measured values and predicted values of the sampling points, respectively.

2.3.6. Cross Validation

Cross-validation is a statistical technique used to assess and enhance the efficacy of predictive models. This method evaluates a model’s predictive capability by partitioning the data into distinct, non-overlapping subsets—one designated for training and another for validation. The primary objective of cross-validation is to obtain a more consistent and precise estimation of model performance, thereby mitigating variations in evaluation outcomes due to disparate data splits [30,31].

In the present study, the dataset was randomly divided into a training set and a test set, adhering to a specific ratio (70% for training and 30% for testing), using straightforward cross-validation. Evaluation metrics are derived directly from the test set. If E_test represents an evaluation metric on the test set, then:

$$E_{Houdout}=E_{test}$$

(10)

2.3.7. Data Histogram

Histograms were constructed after processing the HHT data using Log transformation in ArcGIS 10.7 [21,28].

The difference between the mean and the median is an important indicator of whether the data follows a normal distribution. When the two are close or equal, the data usually show symmetry, suggesting that the data may obey a normal distribution. Kurtosis and skewness are two statistics that describe the distribution pattern of data. The skewness reflects the symmetry of the data distribution. If the data are symmetrical, the skewness is zero, indicating that the data distribution is similar to the normal distribution. If the skewness is greater than zero, the data distribution is right-skewed, with the peaks on the left more concentrated than those on the right. If the skewness is less than zero, the data distribution is left-skewed, with the peaks on the right more concentrated than those on the left. Kurtosis describes the steepness and tail thickness of data distribution. A kurtosis value of zero indicates that the data distribution has the same kurtosis as the standard normal distribution. When kurtosis is greater than zero, it is steeper than normal distribution, with more data clustered near the mean and thicker tails. When kurtosis is less than zero, the distribution is smoother than normal distribution, with more dispersed data.

2.3.8. Data Analysis

The Descriptive Statistics and Calculated Variable in SPSS Statistics 26 were used for data statistics. The ArcGIS 10.7 were used for plotting graphs.

3. Results

3.1. Statistical Characteristics of HHT

The mean and median values of HHT are 30.28 cm and 28.00 cm (

Table 1. The descriptive statistical eigenvalues of the original sample.

Name	Minimum (cm)	Maximum (cm)	Mean (cm)	Median (cm)	SD (cm)	Variance (cm)	Kurtosis	Skewness	CV
The humus horizon thickness	9.00	85.00	30.28	28.00	17.22	252.12	0.61	0.81	0.55

), respectively, which are basically equal, showing some symmetry.

The coefficient of Variation (CV) is 0.55 (Table 1). According to the grade of the CV, the HHT belongs to a high degree of variability, indicating that it is easily affected by the changes in the surrounding environment, and the stability is poor.

The skewness is −0.007 (Figure 5), indicating that the peaks of the data tend to be concentrated on the right side. The kurtosis is 2.34, indicating that the overall distribution of the image is more elevated, and the change is steeper compared with the standard normal distribution, which indicates that the distribution of the data has a thicker tail, i.e., there are more extreme values. This phenomenon may be indicative of an uneven distribution, i.e., the HHT varies considerably between different areas.

3.2. Spatial Distribution Characteristics of HHT

The IDW, RK considering environmental factors, and OK were used to characterize the spatial variation characteristics of HHT (Figure 6). It ranged from 10.20 to 84.55 cm (Figure 6a), 13.12 to 48.60 cm (Figure 6b), and 18.49 to 50.21 cm (Figure 6c), respectively. All results predicted by the three methods show that the HHT changes unevenly, which was consistent with the results showed by Histogram (Figure 5). After in-depth comparative analyses, all three interpolation algorithms can better reflect the distribution characteristics and are consistent with the actual situation.

Comparing and analyzing the predicted results with the actual HHT, it was observed that the results obtained by the IDW best matched the range of fluctuations in the actual observed values. However, the spatial distribution map generated by the IDW lacked smoothness (Figure 6a), which reduced the quality of its visualization. Especially in the region of high values and the dense area of interpolated points, it was easy to appear obvious “bull’s-eye” effect, i.e., the distribution pattern of surrounding high values around a low value point. On the other hand, the OK suffered from the same problem of poor transition smoothness (Figure 6c). Although the RK showed some deviation from the actual soil thickness (Figure 6b), the results showed a higher degree of smoothness and provided a better visualization.

3.3. Suitability Evaluation of Different Interpolation Methods

Soil samples (n = 264) were collected and distributed in various areas, but not all areas were belonged to black soils. After re-classification, 122 samples were screened out. The prediction results were independently validated against the HHT at the validation site to derive the errors (

Table 2. The prediction errors of IDW, OK, and RK results.

Name	ME (cm)	MAE (cm)	RMSE (cm)	RRMSE (cm)
IDW	13.81	13.81	18.08	0.56
OK	12.12	12.12	16.16	0.47
RK	11.45	11.45	14.98	0.44

). Comparing and analyzing the three different interpolation methods, the RK has the smallest ME (11.45 cm), RMSE (14.98 cm), MAE (11.45 cm), and RRMSE (0.44 cm). Conversely, the IDW displayed the lowest accuracy, with corresponding error metrics of 13.81 cm, 18.08 cm, 13.81 cm, and 0.56 cm, respectively, indicating a significant disadvantage. This indicated that the RK has the least difference between the predicted and actual values and is the best prediction results.

3.4. Factors Influencing the Humus Horizon Thickness

Vegetation and Land use type can reflect the land use situation. Average annual temperature and precipitation can reflect the climate state. Geomorphology and DEM determine the distribution and types of soil, and also affect the its formation process.

In a random forest model, the sum of the feature weights is one, which is a key metric to help understand the magnitude of each independent variable’s contribution to the model’s predictive power (

Table 3. The characterizations of weight values of different environmental variables.

Name	Contribution
Average annual temperature	0.29
Average annual precipitation	0.27
DEM	0.21
Land use type	0.12
Geomorphology	0.06
Vegetation type	0.05

).

According to the distribution of feature weights, the following results were obtained:

(1)

The independent variable, mean annual temperature, occupied the largest share in the model at 28.56%, indicating that mean annual temperature had the highest influence and was a key factor in model construction.

(2)

The weight of average annual precipitation was 27.38%, indicating that it also had an important influence.

(3)

DEM had a weighting of 21.12%, which was smaller than the mean annual temperature and precipitation, and had a relatively small degree of influence in the model.

(4)

The proportion of land use types is 12.00%, which has a moderate influence, which means that among the variables considered, it has certain explanatory power to predict the HHT.

(5)

The proportion of geomorphology was 5.70%, and the contribution was low. It showed that geomorphic factors were not the main factors.

(6)

The proportion of vegetation types was 5.20%, with the lowest contribution. This showed that the change of vegetation type was not the main influencing factor, or its influence was relatively weakened compared with other factors.

The weights of the three features (average annual temperature and precipitation, DEM) together accounted for 77.07%, implying that they were the most important features. Although the remaining three independent variables (land use and vegetation type, geomorphology) were relatively small, indicating that their contribution was not as significant as the first three features. Therefore, the average annual temperature and precipitation, DEM were the key influences, whereas land use and vegetation type, geomorphology, although also contributing, were relatively less influential.

4. Discussion

4.1. Characterizing the Spatial Variability of Humus Horizon Thickness

4.1.1. Selection of Interpolation Methods

The environmental variables include a variety of types, which were specifically classified as vegetation, and land use type, mean temperature and precipitation, DEM, and geomorphology. The presence of complex interactions between the variables brought some complexity. Therefore, in the experimental design, the complexity of environmental factors was fully considered, and as many representative typical profile data from the historical soil survey data were collected as possible. These samples were spatially distributed more evenly to ensure the representativeness and to guarantee the accuracy of the statistical inference. Three interpolation methods were used to characterize the spatial distribution of HHT.

In practice, Kriging interpolation is a geostatistical method based on spatial autocorrelation, which assumes that the data is continuous in space [32,33]. The spatial continuity of experimental data might be weakened by a variety of factors [34]. For example, topographic relief, climatic differences, anthropogenic, and other factors might lead to faults or inhomogeneities in the spatial distribution of data [35]. The topography of Liaoning Province is mainly composed of mountains in the east, Liaohe Plain in the middle and low hills and mountains in the west, while the Bohai Sea coast in the south is a long and narrow coastal plain [36]. In terms of climate, Liaoning Province experienced a gradual decrease in precipitation from the southeast to the northwest. The average annual precipitation in Dalian and other places in the southeast can reach more than 800 mm, while that in Chaoyang and Fuxin in the northwest is only 400–500 mm [36].

It was these diverse and variable environmental conditions that directly led to the uneven spatial distribution of HHT, so the OK had a relatively large errors in the results. RK combining environmental variables employed a variability function to describe the spatial variability between data points. By optimizing the variational function model, the spatial structure could be better captured, thus reducing the interpolation error. IDW focused on spatial distance and might ignore other factors [37], such as topography, climate, and human activities. Therefore, IDW was a poor choice when experimental data were affected by multiple environmental variables, while RK, which took into account the topographic complexity, climatic diversity, and human activities, was a more accurate method for predicting the spatial distribution of HHT.

4.1.2. Accuracy of Interpolation Methods

Combined with the RK of environmental variables, the variation function is used, which is the core of kriging interpolation and describes the spatial variability between data points [38]. By optimizing the variogram model, the spatial structure of data can be better captured, thus reducing the interpolation error [39]. Combined with the data characteristics, further analyses concluded that the RK combined with environmental variables was the most accurate interpolation method. Although, the RK had great deviation from the actual, the resulting spatial distribution maps showed high smoothness and provide better visualization effect [40]. This was because the RK not only considered the spatial correlation between sample points, but also combined the influence of environmental variables, so as to better presented the spatial distribution characteristics of the HHT. The result is consistent with the study of effective soil thickness distribution characteristics of arable land obtained by Liu et al. [41].

IDW is a distance-based interpolation method, which is suitable for data sets with strong spatial autocorrelation [42]. IDW interpolation mainly focuses on spatial distance, and ignore other factors that affect the HHT [43], such as topography, climate, and human activities. Although the IDW performed well in reflecting the range of fluctuations in actual observations (Figure 5), it suffered from a lack of smoothing, especially in areas of high values and in areas with a high density of interpolated points, where the “bull’s-eye” effect was likely to be evident [44]. This may be due to the fact that the IDW mainly relied on the imputation of the distance between the sample points, and gave a larger weight that were closer together. It led to the interpolation results being more affected by the local data, thus reducing the smoothness of the spatial distribution maps. This may affect the thickness distribution of humus horizon in black soils [45]. This result is consistent with the results obtained by the interpolation method of equal factors for the quality score of arable land studied by Yang et al. [46].

For the OK, when considering the spatial correlation between the sample points, the same weight was given with different distances, which led to the interpolation results being too smooth and could not well reflect the local change characteristics of the actual observations [47]. It has been demonstrated that the difference between the maximum and minimum values of the numerical result valuation is significantly smaller than the original data. This indicates that the OK has the characteristic of centralizing the distribution of interpolation results, leading to prediction results that are biased towards a larger error, which is also reflected in the present study [48].

In terms of error indicators (Table 3), the RK combined with environmental variables was superior to OK and IDW in ME, RMSE, MAE, and RRMSE. The results were consistent with previous research [49]. However, the IDW had the worst performance on these indicators, which was consistent with the findings of previous studies [50]. RK combined with spatial autocorrelation and the influence of environmental variables made up for the shortcomings of OK and IDW, and improved the interpolation accuracy [51]. Environmental variables, such as DEM, mean annual temperature and precipitation had important effects on spatial distribution [52]. Incorporating these variables into the model combined the strengths of geographic information and statistical modelling to produce smoother maps with lower errors and the highest prediction accuracy when dealing with the problem.

In conclusion, the RK combining environmental variables was the most accurate interpolation method due to its low error indicator performance and intuitive theoretical basis. The method had important practical value in practical application, especially in the fields of environmental science and geographic information system.

4.2. Characterization Results of Spatial Variability of Humus Horizon Thickness

The random forest model was used to analyze the effects of multiple environmental variables, identifying DEM, mean annual temperature and precipitation as key influences driving spatial variability of HHT. A more accurate spatial variability map was produced. It is as follows (Figure 7):

4.2.1. Characteristics of Spatial Distribution of Humus Horizon Thickness

The HHT in Liaoning province is uniform, which is consistent with the result in Figure 5. The HHT increases from the southeast towards the middle and north. The southeast region was the thinnest. However, the northeast and southwest regions exhibit relatively thicker horizons. Along the southern Bohai Sea coast, it is thick but not densely distributed. Chaoyang and Fuxin, located in the northwestern direction, are areas with thick humus horizons, but their distribution is relatively sporadic. Meanwhile, in the northeast direction, Tieling and other locations have thick humus horizons and are aggregated in distribution, albeit covering a relatively small area (Figure 7).

In the southeastern part, it was the thinnest, ranging from about 10 to 20 cm. Climatic factors were one of the important natural factors affecting it [53]. The humid climate and high precipitation with combination of agricultural land use produced strong water erosion, which took away the organic matter and nutrients from the soil, resulting in a decrease of soil thickness and a reduction in soil fertility [54]. In addition, the high humidity environment in the southeastern region enhanced the activity of soil microorganisms [55]. Although soil microorganisms were essential for the decomposition of organic matter and nutrient cycling, excessive microbial activity might also lead to excessive mineralization and decomposition, which accelerated organic matter depletion and affected soil fertility and structural stability [56]. Therefore, in this area, it was relatively thin.

The northeastern region, including the cities of Tieling and Fushun, is located in the upper reaches of the Liaohe River. The river flooding carried a large amount of sediment deposited on both sides of the riverbanks, providing a rich soil-forming parent material [57]. These deposits, which were rich in minerals and other nutrients, were the basis for the formation of fertile soil [2,57]. Tieling and other areas are located in the upper reaches of the Liaohe River, and the special geographical environment makes the river flow moderately, which is conducive to the deposition of alluvial sediments. In addition, many flood events in history had provided a large number of alluvial parent materials, which was an important material basis for the formation of black soil [57].

The northeastern part belongs to temperate continental monsoon climate, which is beneficial to the accumulation of organic matter and the formation of soil clay, thus promoting the development of humus horizon [58]. In addition, the natural vegetation is very lush. Dense forest can not only strengthen the soil through roots, but also provide rich organic matter for the soil under the forest, which is conducive to increasing soil fertility and improving soil structure [59]. Good forest cover also helps to maintain soil moisture, reduce evaporation losses, and regulate surface temperatures, providing a suitable environment for soil organisms, which in turn improves the overall quality and continued productivity of the soil [60].

The western part and the central-western Bohai Sea Rim belt, belonging to the northern temperate continental climate zone and the low mountainous and hilly areas respectively. These areas are subject to frequent invasions of warm air masses coming from the south and the dry-cold air from the Inner Mongolian Plateau, forming a semi-dry, semi-moist and prone-to-drying climatic feature [61]. Winters are cold and dry and summers are hot and rainy. The semi-dry and semi-humid climate is conducive to the growth of vegetation and the accumulation of organic matter. In the wet season, plants grow vigorously, which provides a rich source of organic matter. In the dry season, microbial activity is weakened and the decomposition rate of organic matter slowed down, contributing to the accumulation of organic matter. Moderate precipitation that does not lead to severe soil erosion due to excessive precipitation as in the southeast. Good vegetation coverage on the surface also plays an important role. Plants can prevent soil erosion by fixing the soil through their root systems, but the deadfall of vegetation can increase soil organic matter and improve soil fertility. Vegetation also intercepted precipitation and slowed the impact of rainfall on the soil surface, further protecting the soil structure. The alternating changes in low winter temperatures and high summer temperatures caused cyclical fluctuations in soil temperatures, which promotes the humification process of plant residues, increased the content of soil organic matter and created favorable conditions for the development of black soils [57,58,59].

4.2.2. Variability in the Spatial Distribution of Humus Horizon Thickness

The CV is an important statistical parameter in the in-depth analysis and assessment of soil properties, quantifying the degree of dispersion of values in a dataset [60,61], thus providing a measurement standard [61].

The detailed analysis of the CV revealed that the HHT showed high variability, typically indicating the significant variability. This difference might be caused by many factors, including the difference of natural conditions (such as topography, climate, vegetation, etc.) and anthropogenic activities (such as agricultural cultivation, land use change, urbanization expansion, etc.) [59,60,61].

The reason why the HHT shows high variability:

(1)

Environmental sensitivities: The humus horizon is highly sensitive to environmental changes and is susceptible to factors such as climate change and topographical differences. When environmental conditions change, soil organisms, microorganisms, enzymes, and other chemical substances are all affected in some way, thus influencing the black soil evolution process [61]. Liaoning Province has a temperate monsoon climate, with relatively high precipitation in the eastern coastal areas under the influence of the sea. The inland areas in the northwest have a gradual decrease in precipitation, and the variability of climate distribution leads to different degrees of erosion of the humus horizon [36,58]. Liaoning has a diverse topography, and as one of the natural geographic elements, it plays an important role in the redistribution of water and organic matter on the surface. Under different topographic conditions, there are significant differences in the migration, accumulation, and loss patterns of water and organic matter, which directly affect the formation and development of soil layers [59]. Liaoning Province has a rich variety of landform types, ranging from the mountainous hills in the east to the Liaohe Plain in the center to the Bohai Coastal Plain in the west, and these different topographic features have a profound impact on the distribution of water and organic matter [36]. In mountainous areas, the large slope and fast surface runoff rate easily lead to soil erosion [36], making it difficult for organic matter to accumulate in the surface layer, thus affecting the development of the humus horizon [36]. In contrast, in plain areas, the flat topography helps the accumulation of water and organic matter in the top layer of soil, which is conducive to the formation of a thicker humus horizon [59]. At the same time, there are also variations in precipitation across different regions. For instance, Dalian in the southeast receives abundant precipitation, while the northwestern region experiences relatively less. This precipitation gradient also influences soil moisture and organic matter content, further intensifying the spatial variability of HHT [6,54].

(2)

The soil-forming parent material of Liaoning Province is mainly influenced by its geological structure and rock types, and the soil formation conditions and history of different regions are different [36,60]. Black soils exhibit varying thicknesses across various regions, types, and development stages within the same region, due to differences in soil mineral composition, combination patterns of soil-forming factors, and interaction intensity [62,63]. The formation of soil is significantly influenced by the types of parent materials, which determine the initial material composition and texture of soil. The soil-forming parent materials are complex and diverse, primarily comprising loess, loess-like materials, alluvial deposits, and floodplain deposits [36]. Due to the differences in the type and distribution of these soil-forming materials, the soil properties are unevenly distributed in space. For example, thick humus horizon may be formed in loess and loess-like material sediment areas, while fertile paddy soil may be formed in alluvial areas for good drainage conditions.

(3)

Anthropogenic activity: Several anthropogenic factors play different roles leading to significant differences in the spatial distribution of humus thickness [36]. These factors include land-use practices and soil management measures. The Bohai Coast in the south and the plains in the north-central part of Liaoning Province are the main grain production bases. These areas generally adopt more scientific management methods in land use [1,6]. For instance, conservation tillage techniques, such as straw return and no-tillage sowing, are widely promoted in Northeast China. These techniques have effectively slowed down soil erosion and maintained the organic matter content and structural stability of the soil. These conservation tillage techniques reduce direct soil exposure and the impact of weathering. They also help maintain the HHT by increasing soil organic matter and enhancing the soil’s resistance to erosion and self-recovery. In contrast, the southeastern part is predominantly hilly. This results in a fragmented distribution of arable land, uneven implementation of management measures, and difficult management. Long-term agricultural cultivation and irrational land management practices have significantly impacted the humus horizon. Excessive ploughing, fertilizer use, and single crop planting patterns have damaged the soil structure and accelerated the decomposition of soil organic matter. This has led to the gradual thinning of the humus horizon. Furthermore, with the advancement of urbanization in southeastern part, arable land has been heavily used for industrial and residential construction. This not only reduces the area of arable land but also further destroys the soil structure through construction activities. The changes in land use have directly affected soil protection and recovery mechanisms, making it more difficult to maintain the originally weak humus horizon. Industrial emissions and improper treatment of wastewater may potentially lead to soil pollution, affecting the quality and fertility of the humus horizon.

The high spatial variability indicates the low stability of the humus horizon. This lack of stability may adversely affect the growth of crops, which require stable and fertile soils to ensure optimal growing conditions. It also implies that special attention must be given to soil protection measures in land management and planning to reduce the negative impact of environmental changes on the HHT and to ensure the sustainable use of soil resources.

5. Conclusions

(1)

In the study of predicting the spatial distribution of humus horizon thickness, the regression kriging generates relatively smooth images with good visualization and shows a significant advantage over other methods in terms of accuracy. Ordinary Kring may not be able to reach the accuracy level of regression kriging without introducing additional environmental variables. Inverse distance weight interpolation is too simple and relies only on the weights of distances, resulting in its inability to accurately reflect the actual situation in some cases.

(2)

The humus horizon thickness fluctuates greatly in spatial distribution and is easily affected by the changes of surrounding environment. Among the six environmental variables selected for this study, mean annual temperature, mean annual precipitation, and DEM are key influencing factors (with a contribution sum of 77.07%) of the humus horizon thickness.

(3)

The thickness of the humus horizon in Liaoning Province exhibits an increasing trend from the southeast towards the central and northern directions, with the thinnest horizons found in the southeast. Conversely, the northeast and southwest regions have relatively thicker horizons. Along the Bohai Sea in the south, the humus horizon is thicker but not densely distributed. Chaoyang and Fuxin, located in the northwestern direction, belong to the area with thicker humus horizons, although the distribution of black soil is scattered. Meanwhile, places such as Tieling in the northeastern direction have thick humus horizons and exhibit an aggregated distribution, albeit covering a relatively small area.