Multi-Scale Integration and Distribution of Soil Organic Matter Spatial Variation in a Coal–Grain Compound Area

Abstract

Soil organic matter (SOM) scale effects are critical for crop growth and food security, especially in coal–grain complexes. However, few studies describe the spatial variation in SOM and its influencing factors at different sampling scales. Here, geostatistical theory and mathematical statistical methods were adopted to analyze the spatial variation characteristics of and structural differences in SOM in the coal mining subsidence area at Zhaogu No. 2 Mine at different sampling scales. The results showed that SOM varied spatially at large, medium, and small scales, and the coefficients of variation were 28.07%, 14.93%, and 14.31%, respectively, which are moderate values. The characteristic functions of the SOM content at different sampling scales differed, and the spatial structure scale effect was obvious. The spatial distribution of the SOM content fitted by the multiscale fitting model method was generally the same as the spatial distribution law of the SOM content fitted by the single scale kriging interpolation method; however, in terms of the detailed expression and spatial distribution of small-scale SOM content, the fitting model method was more accurate, and the accuracy increased by 36%. At the different sampling scales, sample size and soil type had specific effects on the SOM spatial distribution. These results provide research concepts and technical countermeasures for improving food security and the ecological environment in the coal–grain complex and help ensure sustainable agricultural lands.

1. Introduction

China has abundant coal resources. At present, the coal resource reserves that have been identified and surveyed account for approximately one-third of the world’s total coal reserves, and the annual output exceeds 3 billion tons, making China first in the world in terms of coal output [1,2,3]. The large-scale coal resource development also has led to many surface environmental damage problems, such as surface collapse, excavation losses, and compression, which substantially harmed the surface ecological environment and the life of mining residents with adverse impacts on the atmospheric carbon and nitrogen cycle [4,5,6,7]. The main coal mining method in China is borehole mining, which readily causes large-area surface subsidence above the goaf and produces a large number of cracks and fracture surfaces [8,9,10]. Coal mining has destroyed a large amount of land and vegetation, causing severe damage to general farmland [11,12]. Currently, the area of land subsidence caused by coal mining is more than 0.45 × 106 hm², and coal mining continues to increase. A large area of surface subsidence leads to spatial variation in cultivated soil nutrients in mining areas, resulting in reduced soil quality, and further affecting grain yield and straining the human–land relationship in mining areas [13,14,15,16]. Immense destruction of cultivated land has severely affected the ecological environment of coal–grain compound area and restricted the sustainable development of the social economy in these areas. Therefore, it is necessary to improve the quality of cultivated land in coal–grain compound area [17,18,19,20].

Soil is a basic natural resource that humans depend on for survival. It is gradually formed by a series of physical and chemical processes of the parent material under specific water, heat, and biological conditions [21,22,23,24]. Under the joint action of natural factors and human factors, soil forms a spatiotemporal continuum of nonuniform changes with high spatial variability. In an area with the same soil type, there are also significant spatial differences in soil characteristics, which is called spatial variation in soil [25,26,27,28,29]. As one of the most general attributes of soil resources, the abundance and deficiency of soil characteristics directly affect the quality of soil and thus the growth and reproduction of vegetation and crops [30]. High spatial variability in soil properties in coal–grain complex [31].

In addition, spatial variation of soil characteristics is greatly influenced by external factors, especially in soil fertility and quality [32,33,34]. Although the soil organic matter (SOM) content in soil is minimal, it is an indispensable part of the soil that provides various nutrients necessary for crop growth and development and is a key indicator for evaluating soil fertility [35,36]. The SOM can promote the soil’s ability to retain and supply fertilizers and can continuously improve cultivated land soil structure, enhance soil biological activity, and improve soil nutrient levels [37,38,39,40]. Therefore, it is important to study the SOM content and its spatial variation characteristics in coal–grain complex areas to strengthen ecological environmental protection and food security in mining areas [41,42].

Accurate information on the spatial variation in SOM is an important symbol of not only soil quality but also carbon storage in terrestrial ecosystems [43,44]. Thus, many of information on the content of SOM in space and time is needed. However, the accurate measurement of SOM content requires the acquisition of a large number of soil samples and thus considerable labor and material resources and time. It was found that due to the different scales of soil sampling, there are great differences in spatial variation, which is important for the management of precision agriculture production programs. In addition, soil fertilization and similar activities can be inconvenient [45,46]. Therefore, SOM samples need to be collected at different scales for better planning and management. In soil science, the scale effect mainly refers to the dependence of spatial variation in soil characteristics on sampling scale; that is, a certain spatial distribution feature of soil characteristics can only show a change rule at a specific sampling scale [47]. To better describe the spatial variability in soil characteristics and improve the accuracy of research results, most scholars have clearly noted that scale should be used to address practical problems. At present, most studies on the scale effect of spatial variation in soil properties focus on the comparison of soil properties at different scales. Paterson et al. [48] studied the variation law of spatial variation in soil characteristics with scale change in Australia and found that approximately 50% of the spatial variation occurred within the range of less than 10 km and 30% within the range of less than 1 km. Within that study area, the spatial variability in soil texture increased with increasing depth, the spatial variability in soil characteristics was highly random at small scale, and the randomness gradually decreased with increasing study scope. Jenerette et al. [49] used the landscape scale method to compare the degree of dependence of soil spatial variability to sampling range, as well as conducted an uncertainty analysis related to land use types and the regional scale of Phoenix. The results show that the spatial variation in soil total nitrogen and organic matter was very different at different scales. Suriyavirun et al. [50] studied the differences in soil characteristics and the spatial distribution of microorganisms at the field scale and found that soil characteristics and microorganisms were significantly different at different sampling scales. In summary, there are many studies on the spatial variability in soil properties, and they mainly focus their research on the spatial variability in soil properties at a single scale. Information on how to study the spatial variation in multiscale soil properties, how to obtain and apply multiscale spatial information, etc., is lacking, especially for soil organic matter.

Therefore, the aims of the present study are (1) to study the spatial variation characteristics and structural differences in SOM at different sampling scales; (2) to construct a multiscale fit model of the semivariance function of SOM and determine the distribution of SOM under the multiscale fit model; and (3) to analyze the main influencing factors of the spatial structure of SOM at different sampling scales and for land reclamation and ecological reconstruction in coal–grain composite areas.

2. Materials and Methods

2.1. Study Area

The study area is located in the Zhaogu No. 2 Mine in northwestern Henan Province, China (112.53°–113.63° E, 34.8°–35.5° N), as shown in Figure 1. The region is 55 km away from Jiaozuo city, 37 km away from Xinxiang city in the southeast, and 10 km away from Huixian city in the northeast. The study area is on the piedmont plain; the mountains present are primarily the Taihang Mountains; the ground elevation is generally 72–85 m; the lowest part is the coal mining subsidence area. The land is fertile; and the natural slope is 4–6%.

The study area is semiarid and semihumid. Spring is windy and rainy, summer is hot and rainy, autumn is cool, and winter is cold and dry. January is the coldest month; July is the hottest month. The average annual precipitation is approximately 589.1 mm, and the evaporation is 2039 mm. From July to September, there is substantial precipitation. The average monthly precipitation is 182.3 mm. The soil types at the test site were cinnamon soil, moisture soil, and paddy soil. The study area is rich in coal resources, and the identified coal reserves total approximately 1.2 billion tons. The main planted crops are wheat, rice, corn, etc., and the main economic crops are oilseeds, mushrooms, etc. Figure 2 shows the status of the research area. The total land area is 69.09 km², including 39.38 km² of cultivated land, 0.022 km² of dry land, 10.21 km² of irrigated land, and 0.079 km² of forested land. Villages, towns, and roads cover an area of 7.389 km². The surface area is 0.052 km², and the subsidence area is 1.97 km². The coal mining method in the study area is well mining, which is easy to cause a large area of surface subsidence above the goaf and produce a large number of cracks and fracture surfaces.

2.2. Soil Sampling

In this study, soil sampling was carried out in April 2022 in the coal–grain complex area of the Zhaogu No. 2 Mine, as shown in Figure 3. Grid nesting was implemented in the sample area, and 181 SOM samples were obtained. In order to obtain ideal experimental results, the principle of randomness and equivalence is followed during sampling. According to the sampling requirements, the design of sampling points conforms to the principles of comprehensiveness, representativeness, objectivity, feasibility, and continuity. Each sample site was 10 m in diameter, and soil samples were collected from 8–12 drill holes at a depth of 0–20 cm within the center of the circle and its radius. The soil samples representing this point were mixed. To study the scale effect of spatial variation of soil characteristics and explore the effect of multiscale fitting methods and different models on improving the accuracy of spatially estimating SOM, the actual situation of the study area was observed, the maximum sampling interval was 1000 m, and there were 72 sampling points. The distance of 1000 m was L scale. The representative coal mining subsidence area was selected as the first-level intensified area, and the sampling interval was 500 m, and there were 76 sampling points. The distance of 500 m was M scale. The same method was used as the interval of 250 m to generate a secondary sampling area with 79 samples. The distance of 250 m was S scale. Geographic coordinates of sampling points were recorded with handheld GPS. The sampling distribution is shown in Figure 3a. When collecting soil samples densely, the principles of randomness and fairness were used to uniformly select dense grids for soil sample collection, and sampling occurred using the quincunx shape. The SOM content was determined by the potassium diphosphate volumetric external heating method.

2.3. Research Methods

2.3.1. Semivariogram Model

Semi-variance function, also known as semi-variance function, is a key function in the study of soil variability in geostatistics. It reflects the spatial variation structure of variables in the study area, and can describe both structural changes of regionalized variables and random changes. It plays an extremely important role in determining the number of soil samples and the calculation of kriging interpolation. Semivariograms have been widely used in geostatistical analysis to evaluate the spatial dependence of soil properties based on regionalized variable theory and intrinsic assumptions [51]. In this paper, a variogram was used to study the soil organic matter content, and the semivariogram is expressed as:

$$\gamma \left(h\right)=\frac{1}{2N\left(h\right)}{{\displaystyle \sum _{i=1}^{N\left(h\right)}\left[Z\left(xi\right)-Z\left(xi+h\right)\right]}}^2$$

(1)

where $\gamma \left(\mathrm{h}\right)$ is the semivariogram, $h$ is the spatial length; $Z\left(xi+h\right)$ and $Z\left(xi\right)$ represent the index measured data of the position of point $xi+h$ and $xi$, respectively; and $N\left(h\right)$ is the number of pairs of sample points separated by the lag distance $h$.

2.3.2. Multiscale Nested Model

The nested structure is the sum of multiple semivariance functions [51], and its expression is as follows:

$$\gamma (h)={\gamma }_0(h)+{\gamma }_1(h)+{\gamma }_2(h)+\cdot \cdot \cdot +{\gamma }_i(h)={\displaystyle \sum _{i=0}^n{\gamma }_i(h)}$$

(2)

In the formula, $\gamma (h)$ represents the microscopic change, which is the spatial structure that cannot be reflected at the minimum sampling scale, that is, the structural variance in the scale fit structure, and ${\gamma }_i(h)$ represents the same theoretical model or different theoretical models, which are reflected at different scales out of the spatial structure.

2.3.3. Model Test Method

This study adopts the $I$ value test method, which is a comprehensive method for testing the best fit of the semivariogram function.

$$I=\overline{{\left[\hat{Z}(x_i)-Z(x_i)\right]}^2}\times \left[P\times \left|1-1/{\left\{{\left[\hat{Z}(x_i)-Z(x_i)\right]}^2/\hat{\sigma }(x_i)\right\}}^2\right|+(1-P)\right]$$

(3)

$$P=\{\begin{array}{lll}0.1& & 0\le \overline{{\left[\hat{Z}(x_i)-Z(x_i)\right]}^2}<100\\ & & \\ 0.2& & \overline{{\left[\hat{Z}(x_i)-Z(x_i)\right]}^2}\ge 100\end{array}$$

(4)

where $P$ is the empirical parameter, and the smaller the value of $I$ is, the better the fitting effect of the semivariance function. $\hat{Z}(x_i)$ and $Z(x_i)$ are the predicted and measured values at the $x_i$ position, respectively.

The K-S test method is to compare the cumulative frequency distribution of sample data with a specific theoretical distribution. If the difference between the two values is small, the sample belongs to a specific distribution [42]. The basic principle of the K-S test is as follows: Assume that $H_0$ represents that the sample obeys a specific distribution, $H_1$ represents that the sample does not obey a specific assumption, $F_0(x)$ represents the theoretical distribution function, $F_n(x)$ represents the cumulative frequency function of any set of random samples, and $D$ is the difference between $F_0(x)$ and $F_n(x)$. The maximum value is defined as $D=\max \left|F_n(x)-F_0(x)\right|$, when the value of $D$ is small, the $H_0$ hypothesis is accepted, otherwise the $H_1$ hypothesis is accepted.

2.4. Data Processing and Analysis

SPSS 20.0 was used to analyze the maximum, minimum, and average values of SOM content at different scales and to conduct the K-S normal distribution test (SPSS, Chicago, USA) [3]. If p < 0.05, it was accepted as significant. The variability in SOM was divided into three levels: low variability (CV < 15%), medium variability (15% ≤ CV ≤ 35%), and high variability (CV > 35%) [52].

In this study, GS+ 9.0 was used to conduct the semivariogram structure analysis and anisotropy analysis on SOM at three sampling scales and to determine the optimal semivariogram function theoretical model. The spatial structure of the SOM at the three sampling scales was cumulatively fitted, and the semivariogram function model was tested for optimality using the I value method. The spatial interpolation of the SOM was carried out by using the scale fit model method.

3. Results

3.1. SOM Content at Different Scales

Outliers refer to the existence of large or small values in the research data, and they are mainly caused by the sampling process and test analysis process. The existence of outliers will cause changes in the semivariance function. Before the spatial data analysis of soil properties, it is necessary to analyze the outliers, identify them, and eliminate them. In this study, three times the mean plus (minus)the standard deviation was used to test the specific value, and if the value was not within this range, then the maximum or minimum value was used instead.

Table 1. Statistical characteristics of soil organic matter.

Item	Scales a	Samples	Max	Min	Mean	SD	Median	Kurtosis	Skewness	K-S (p)	CV (%) b
SOM (g·kg−1)	L	72	26.35	10.26	15.92	3.66	15.61	0.341	0.043	0.866	28.07
	M	76	24.26	10.92	14.94	2.93	14.32	0.613	−0.291	0.916	14.93
	S	79	23.61	11.13	13.62	2.01	13.38	0.695	−0.306	0.969	14.31

a. L, M, and S represent large, medium, and small scales, respectively. b. Coefficient of variation.

shows the spatial distribution characteristics of soil organic matter at L, M, and S scales. The normal distribution test results for the SOM data at the three sampling scales were 0.866, 0.916, and 0.969, respectively, reaching a significant level (K-S > 0.05), which was consistent with the normal distribution law. This result indicated that the combination of sampling points at the L, M, and S scales did not change the overall distribution characteristics of SOM, and the three scales could be combined into one study. The average, maximum, and minimum values of the SOM at the three scales were different to some extent. The average values of the SOM at the L, M, and S scales were 15.92 g/kg, 14.94 g/kg, and 13.62 g/kg, respectively. The median values of the SOM were 15.61 g/kg, 14.32 g/kg, and 13.38 g/kg, respectively. The mean values and median values of the SOM at each sampling scale were close to each other, and the mean values and median values showed a decreasing trend with decreasing sampling scale, indicating that the influence of the specific values decreased. As the sampling scale decreased, the absolute values of the kurtosis coefficient and skewness coefficient increased gradually, while the coefficient of the standard deviation and variation decreased gradually, indicating that the concentration degree of data at the M scale and S scale was higher than that at the L scale; in addition, the variation was smaller. The coefficients of variation at the M and S scales were 14.93% and 14.31%, respectively, and the coefficient of variation at the L scale was 28.07%, indicating that the SOM at all three scales exhibited moderate variation.

3.2. Spatial Structure Analysis of SOM at Different Scales

When studying soil characteristics, it is necessary to consider the variation in regionalized variables in different directions. If regionalized variables have different properties in different directions, then they are referred to as anisotropic; otherwise, they are referred to as isotropic. Anisotropy is relative, while anisotropy is absolute. Anisotropy reflects the difference in the semivariance function in different directions, and isotropy is a special form of anisotropy [53]. Some scholars have shown that at least 4–8 variogram maps should be calculated to determine whether soil properties are isotropic or anisotropic by comparing the magnitudes of the principal and secondary variograms. If the ratio of the principal variogram to the secondary variogram is 1 or close to 1, then it is isotropic, indicating that regionalized variables have the same or similar changes in different directions [54]. Anisotropy needs to be converted to isotropy according to the linear proportion relationship.

The semivariogram function of the SOM at different scales was analyzed through the ArcGIS geostatistical analysis module, the semivariogram cloud view was used, the search direction and angle were automatically changed, the change in the semivariogram function was observed, and GS+ was combined to analyze the regionalization variables in different directions. This function can be determined by the change direction diagram of the range (that is, the change diagram of the range change with the direction). The anisotropy ratio of the variables at different scales and directions was 1 or close to 1 (

Table 2. Anisotropy ratio of SOM at different scales.

Scales	Main Variable Direction/°	Main Range/m	Secondary Range/m	Anisotropy Ratio
L	90	2460	2460	1.000
	135	2496	2473	1.009
	180	2460	2460	1.000
	225	2314	2438	0.949
M	44	1126	1167	0.965
	89	1253	1197	1.047
	134	1083	1106	0.979
	179	1324	1289	1.027
S	142	621	621	1.000
	187	656	643	1.020
	232	621	621	1.000
	277	661	673	0.982

), so the SOM content was isotropic at the L, M, and S scales.

Under isotropic conditions, the structural properties of the SOM at different scales were fitted, and the theoretical semivariance function and test parameters are shown in

Table 3. Semivariance function models at different scales and with different fitting parameters.

Item	Scales	Model	Determination Coefficient	Residual	Range (m)	C0	C0 + C	C0/C0 + C%
SOM	L	Exponential	0.431	0.160	2460	0.207	0.413	50.12
	M	Exponential	0.462	0.126	1135	0.106	0.357	29.70
	S	Spherical	0.397	0.182	621	0.132	0.402	32.84

. The spatial variation structure of the SOM at the L and M scales was the best fitting effect of the exponential model, while the spherical model was the best at the S scale. With the increase in sampling interval, there were some differences in range, nugget value, and nugget coefficient, and they all showed an increasing trend; the structural variance in SOM at the M scale was 0.251, and the structural variance of SOM at the S scale was 0.27, with a difference of 0.019. At the same time, the difference between the ranges of the two scales was also small, so the spatial structure of the SOM at the M and S scales can be expressed as that at the M scale. When sampling, this value can be combined with the work, labor, financial resources, etc. The sampling points at the M scale were used to study the spatial variation in the SOM at the S scale. At the L scale, the nugget coefficient and nugget value were significantly greater than those at the M and S scales, and the nugget coefficient reached 50.12%, exceeding 50%, indicating that regional factors such as topography and soil type were the main factors affecting the spatial variation in large-scale SOM. The nugget coefficients at the M and S scales were relatively small, indicating that at medium and small scales, human and natural factors such as land use and fertilization management were the main factors affecting the distribution of SOM. From the L scale to the M scale, each parameter changed substantially, indicating that with the decrease in the sampling scale, new influencing factors appeared, and some influencing factors covered at the L scale appeared at the M scale. The above research shows that increasing the sampling scale can reduce the workload, and the large scale overlapped with the small scale to some degree; however, the accuracy of the analysis of the SOM spatial structure was reduced. Therefore, it is more meaningful to use a nested model of different scales for research.

3.3. Multiscale Integration of SOM

In geostatistics, the theoretical model of the variation function adopted by the kriging interpolation method requires that the regionalized variables must be isotropic. When the research variables are anisotropic, they need to be converted to isotropic in advance. In this study, the spatial variation characteristics of the SOM at different scales were isotropic, so there was no need to convert from anisotropy.

According to the theory of multiscale fit, the structural variance is the sum of the structural variances of different scales, and the nugget value of the new model is the spatial variation that cannot be expressed at the smallest scale. In this study, three scales were selected based on the differences in the spatial structures of the different scales. The method of spatial structure accumulation was used to study the fitting model, and the expression of the fitting structure of the SOM semivariance function at the L, M, and S scales was obtained according to Formula (5):

$$r(h_3)=\{\begin{array}{cc}0.132& h=0\\ 0.132+0.27(\frac{3}{2}\cdot \frac{h_3}{621}-\frac{1}{2}\cdot \frac{h_3{}^3}{{621}^3})+0.251(1-e^{\frac{-h}{1135}})+0.206(1-e^{\frac{-h_{}}{2460}})& 0<h<621\\ 0.402+0.251(1-e^{\frac{-h}{1135}})+0.206(1-e^{\frac{-h_{}}{2460}})& h>621\end{array}$$

(5)

3.4. Accuracy Verification

The semivariance function model at the three scales was fitted by the nested structural model, and the kriging interpolation method was used for the spatial estimation, namely, the multiscale nested model interpolation; in addition, the prediction accuracy of the multiscale nested model method and the ordinary kriging interpolation method at a single scale was compared. This study used the I value to test the theoretical semivariogram function model. This method is an index for testing the optimality of the theoretical semivariogram function formed by the combination of the cross-validation method and the estimated variance test method. A smaller I value indicates that the theoretical semivariogram model fits better. At the same time, P is one of the indexes for determining the optimality of the theoretical model. The smaller the value is, the better the theoretical semivariance function model. It can be seen from

Table 4. Comparison of the accuracy of different interpolation methods for SOM.

Item	Interpolation	I Value	P
SOM	Ordinary Kriging	2.369	2.406
	Fit Model	0.561	0.529

that the I value of the ordinary kriging interpolation method is 2.369, the I value of the fitted model interpolation method is 0.561, and P is also smaller than that of the ordinary kriging method, which shows that the interpolation accuracy of the multiscale fitted model method is higher compared to the ordinary kriging method. This method can better reveal the spatial variation of SOM.

3.5. Spatial Distribution of SOM

After integrating the three scale semivariance function models, the L-scale general kriging interpolation method and the multiscale embedding model method were used to estimate the SOM content in the study area, the spatial distribution map of SOM was drawn, and the interpolation results of the two methods were compared. Figure 4a shows the general kriging spatial interpolation diagram of the SOM at the L scale, and Figure 4b shows the spatial interpolation diagram of the nested model. The spatial distribution characteristics of the SOM fitted by the scale-nesting model were generally consistent with the distribution characteristics at sampling scale L. The content of the SOM in the coal mining subsidence area was relatively low; the farther away from the coal mining subsidence area, the higher the content of SOM was, and the content of SOM tended to be stable at the boundary of the mining area. This result mainly occurred because the coal mining subsidence area is most affected by underground coal mining, and the surface limit subsidence results in surface cracks, subsidence, severe soil and water loss, and losses in soil nutrients to the surrounding area; thus, of the that in the mining areas, the SOM content the coal mining subsidence area was the lowest. The farther the distance from the mining subsidence area, the less underground mining influences SOM and the more stable the content of organic matter.

In terms of expressing the spatial variability in SOM at the M and S scales, the integrated model method is more detailed. At a small scale, there was little difference in the variation in topography, soil texture, and tillage management; thus, the diffusion of SOM contents in all directions was more uniform, and the contour lines were smooth, with a tendency to form clumps. At a large scale, the spatial distribution of the SOM contents showed a zonal characteristic overall due to the influence of soil hydrothermal fertilizer, coal mining subsidence, farming methods, and other factors. Through scale integration, the spatial distribution pattern of SOM was more refined. The fertilization capacity should have further increased in the central, western marginal, and northwestern regions where the SOM content was low. At the same time, a sustainable application of organic and inorganic fertilizers should be emphasized, and the proper use of tillage systems such as intercropping and rotation should be adopted to improve the soil quality. The content of soil organic matter was relatively high in the north, west, and southeast of the study area, so the soil should maintain its fertility retention ability.

In this study, three scales were nested to collect SOM samples, the sampling points were locally coded, and the overall spatial distribution was unbalanced. Therefore, the nested model and the ordinary kriging method were used to carry out spatial interpolation at the L scale in the sample area, and the ordinary kriging standard error prediction map was drawn. The kriging standard error prediction of the SOM at the L scale is shown in Figure 5a, and the kriging standard error prediction of the SOM using the nested model is shown in Figure 5b. The standard prediction error was mainly caused by the original error in the sample data and the uncertainty caused by the uneven distribution of the sample points in the spatial interpolation.

Comparing the estimation accuracy of the two methods can provide a reference and basis for the design of a sample point scheme for studying soil nutrient spatial variation. It can be seen from Figure 3 that the estimation accuracy of nested value model interpolation had a wide variation range and was highly sensitive to distance variation. At the S sampling scale, the estimation accuracy of the fitting model method was significantly higher than that of the ordinary kriging method. With the increase in the sampling scale, the accuracy gradually decreased. The interpolation accuracy of the ordinary kriging method increased, and when the L-scale sampling distance was reached, the interpolation accuracy value of the fitted model method was similar to that of the ordinary kriging method. Therefore, on the basis of the L-scale sample points, the multiscale fitting structure model obtained interpolation results with higher estimation accuracy.

From the standard error prediction graph, it can be seen that the farther away from the sampling point, the greater the interpolation error, and the smallest error around the sampling point. And at the boundary of the study area, the error is the largest. It shows that there is a marginal effect on the accuracy of spatial interpolation of soil organic matter, because the sampling points in the boundary area are relatively sparse, resulting in a decrease in the estimation accuracy at the boundary; while the sampling points in the non-boundary area are denser, the prediction error is smaller.

Table 5. Statistical analysis of the SOM interpolation results by different methods.

Methods	Max	Min	Mean	CV(%)
Measured value	26.35	10.26	15.92	28.07
Ordinary Kriging	24.19	12.36	15.76	24.03
Fit model	26.93	11.13	15.92	26.56

shows the statistical results of the two methods of interpolation for SOM, and the resulting value was greater than the measured value. The maximum and minimum values for the SOM obtained by the nested model interpolation method were greater than the measured values, the coefficient of variation was smaller than the measured values, and the average value was equal to the measured values. Thus, the kriging interpolation method had a certain smoothing effect on the data, and in comparison to the ordinary kriging method, the fit model interpolation method resulted in a higher degree of agreement between the predicted value of SOM and the measured value, indicating that the interpolation effect of the fit model method was due to the grid method.

4. Discussion

In this paper, the SOM content was generally the lowest in the coal mining subsidence area, and the farther away from the coal mining subsidence area, the higher the SOM content was; in addition, the spatial variability in the SOM content was moderate. Soil texture, slope size, land use type, microbial content, underground coal mining, artificial fertilization, irrigation, and other factors all affected the spatial distribution of the SOM. In Panagea’s study, the results showed that under different management measures, the change in soil organic carbon was not obvious, and an increased soil water-holding capacity could be obtained by reducing soil disturbance and increasing fertilization of SOM [55]. Chiara Piccini et al. studied the spatial variability in SOM in the Abruzzo region of central Italy. The results showed that soil organic carbon storage and SOM content in hilly areas were relatively low, and soil management methods should be improved to ensure the sustainable development of agricultural land [56].

4.1. Effect of Sample Size on the Spatial Variability in SOM

This study found that the number of sample sampling points had a specific impact on the spatial variability in SOM. Too many sampling points will result in wasted resources, but fewer sampling points will not be representative and cannot reflect the spatial distribution characteristics and variability in SOM. In this study, 72 soil samples were obtained at the L scale, 76 at the M scale, and 79 at the S scale, and a total of 227 soil samples were obtained. A correlation analysis was conducted on the number of SOM samples in the study area and the coefficient of spatial variation, as shown in Figure 6. Within a specific range of soil samples, with the increase in the number of soil samples, the spatial variation coefficient of the SOM content increased, and the spatial variation in the SOM content showed a positive correlation with the number of soil samples. The results show that the spatial variation in SOM can be better expressed by using the three scale sets as statistical units and sampling points between 150 and 180.

4.2. Effect of Soil Type on SOM

There were three main soil types in the study area, namely, brown soil, tidal soil, and paddy soil, as shown in

Table 6. Changes in SOM content in different soil types.

Type of Soil	Number (a)	Mean (g/kg)	Standard Deviation	CV(%)
cinnamon soil	73	20.14	3.32	15.61
moisture soil	68	10.16	2.93	30.26
paddy soil	59	18.29	2.35	11.44

. There were significant differences in the SOM content among the different soil types. Ninety-six samples were collected from the brown soil, and the average SOM content in this soil was 20.09 g/kg. Seventy-three soil samples were collected from the tidal soil, and the average content of SOM in this soil was 10.26 g/kg. There were 58 soil samples collected from the paddy soil, and the average content of SOM in this soil was 17.41 g/kg. The average SOM content of the different soil types occurred in the order of brown soil > paddy soil > tidal soil. The variation coefficient of the SOM content in the different soil types occurred in the order of tidal soil > brown soil > paddy soil, and the spatial variation coefficient of the SOM in the tidal soil was the largest.

It can be seen from Figure 7 that the SOM content in the moisture soil was 16.0–19.5 g/kg, accounting for approximately 55% of the overall distribution. The SOM content in the cinnamon soil was 19.5–25.5 g/kg, accounting for approximately 47% of the total distribution, and the SOM content was greater than 23.0 g/kg and 13.0–16.0 g/kg, accounting for approximately 38% of the total distribution. The SOM content in the paddy soil was approximately 70% of the overall distribution.

4.3. Other Factors Affecting SOM

Related studies have found that natural and human factors account for more than 75% of the impacts on the spatial variation in SOM. The SOM content is often correlated with soil properties and is influenced by both physical and chemical processes. The results of this study are consistent with those of previous studies and provide a systematic analysis of the factors affecting the spatial distribution of SOM [57]. Coal mining can cause the surface to subside, resulting in a large number of fractures [42]. In this study, there were many different sized cracks in the study area, which affected soil respiration and then the spatial distribution of SOM. Figure 4 shows the spatial distribution characteristics of the SOM in the coal mining subsidence area. The farther away from the underground mining area, the higher the SOM content was; the SOM content was lower in the mining area and surrounding areas because the mining area was most affected by coal mining, which resulted in ground subsidence, cracks, water accumulation, salinization, etc., [58]. Cracks promote soil respiration, changing soil temperature, and moisture, leading to increased microbial decomposition capacity [59]. At the same time, underground coal mining leads to a decrease in soil water content, which also affects the content of SOM. In addition, underground coal mining leads to water and soil erosion, the loss of soil nutrients from high-lying areas to low-lying areas, severe soil erosion. Due to the low groundwater level in the Zhaogu mining area, coal mining has led to an increase in the surface water level, resulting in a large amount of accumulated water in the subsidence area [19], as shown in Figure 2b. Waterlogging in the subsidence area will increase soil viscosity, reduce soil respiration.

In addition, different land-use patterns and fertilization management also have had impacts on the content of SOM, as shown in Figure 2e. The SOM content of the cultivated land was relatively high, while the soil organic matter content of the sandy land and mining areas was relatively low. Farmers’ management methods implemented on different cultivated lands have also caused differences in soil organic matter content. Farmers use substantial energy managing irrigated land and high-quality farmland through efforts such as scientifically controlling the amount of organic fertilizer application, crop rotation, and returning straw to the field, to reduce the loss of SOM and other nutrients; investment in the management of dry land and subsidence areas involves less energy; and few people fertilize and manage abandoned cultivated land, so the soil fertility of abandoned cultivated land is low. The organic fertilizer applied by farmers in irrigated land is basically purchased in accordance with the recommendation of formula fertilization, and the fertilizer is suitable for the disease of the land. Therefore, the proportion of fertilizer applied in the study area tends to be reasonable. Overall, different management methods related to fertilization and fertilizer conservation have different effects on different cultivated lands.

Yang Qiyong et al. [60] used geostatistical methods to explore the spatial variation characteristics of farmland soil properties at county and township scales. The results showed that, except for total nitrogen, the spatial autocorrelation of soil properties decreased as the sampling scale decreased for other soil indicators, and the coefficient of variation increased. Qin Jingtao et al. [61] took the sandy loam area in northern Henan as the research area to study the spatial variation of farmland soil moisture content at different scales, and found that the larger the sampling scale, the lower the confidence level of the normal distribution of soil moisture content. The overall performance was that the standard deviation and coefficient of variation of soil moisture content gradually increased with the increase in sampling scale. The larger the sampling scale, the stronger the spatial variability of moisture content. However, it should be pointed out that most of the previous studies focused on the spatial distribution characteristics of soil properties at different scales, and there were fewer reports on using scale nesting model to study soil properties. Therefore, the multi-scale integration method was used in this study to study the spatial distribution characteristics of SOM in the coal–grain composite area. It can not only improve the accuracy of interpolation, but also save manpower, material, and financial resources. At the same time, the results can provide reference for farmers to apply fertilizer.

5. Conclusions

This study considered the SOM in the coal mining subsidence area of the Zhaogu No. 2 Mine as the research object, used three sampling scales (L, M, and S) to study the spatial scale variability in SOM, and analyzed the spatial variation structure of the SOM at the different sampling scales. The results showed that (1) SOM had spatial variability at different scales, and at all scales, the spatial variability was moderate. The spatial structure at the M and S scales was expressed at the M scale, and the distribution of soil sampling points at the M scale was considered to be approximate to the spatial variation in SOM at the S scale. (2) The spatial distributions of the SOM simulated by the single-scale ordinary kriging interpolation method and the fitted model interpolation method were generally consistent, but the fitted model interpolation method was more accurate in expressing the spatial variability in the medium- and small-scale organic matter. The I-value test results showed that the interpolation results of the multiscale fitting model method improved the reliability of the application. (3) Under different sampling scales, the size of the sample had a certain impact on the expression of the spatial variation in SOM; soil type could affect the spatial distribution of soil organic matter. However, this study only discusses the multi-scale nested model method based on the small-scale area of the mining area, and its application in the large regional scale needs to be further verified. In addition, the results of this study provide a theoretical basis for ecological protection and agricultural sustainable development in coal–grain complex areas.