Spatial Dependence of SO₂ Emissions and Energy Consumption Structure in Northern China

Abstract

China has made achievements in SO₂ emissions reduction in recent years. However, the emissions of SO₂ in northern China remain high, which need to be reduced. To effectively control SO₂ emissions in northern China, this paper from the perspective of the coordinated treatment of air pollution discusses the impact of energy consumption, economic development, and environmental regulation on SO₂ emissions in 14 provinces and regions by the Spatial Lag Model (SLM), Spatial Error Model (SEM), and Spatial Durbin Model (SDM). The study shows that (1) there is an obvious spatial dependence between SO₂ emissions and energy consumption; (2) the increase in the scale of industry enterprise can exacerbate SO₂ emissions in local and adjacent regions; and (3) the consumption of electricity suppresses SO₂ emissions in the local region, and increases SO₂ emissions in adjacent regions, which indicated that the electricity transmission can transfer the emissions of SO₂. Therefore, in the treatment of SO₂, it is necessary to fully consider the characteristics of SO₂ transfer in the electric power industry.

1. Introduction

The economy of China is growing at a relatively high rate, which has caused a lot of natural resource consumption and SO₂ emissions [1,2]. Additionally, the emissions of SO₂ have caused heavy damage to human health and social development. In recent years, the Chinese government has adopted many methods to reduce SO₂ emissions, such as energy structure adjustment, desulfurization treatment, and making achievements in SO₂ reduction [3]. However, it should be noted that China is still the third largest SO₂ emitter, and the SO₂ emissions in provinces and regions have enormous differences. According to the provincial SO₂ emissions distribution in 2011 and 2017 (Figure 1), it can be found that the emissions of SO₂ in northern China are still significantly high [4,5].

Generally, the northern region in China includes 14 provinces (autonomous regions and municipalities) including Beijing, Tianjin, Hebei, Shanxi, Inner Mongolia, Liaoning, Jilin, Heilongjiang, Shandong, Shaanxi, Gansu, Ningxia, Xinjiang, and Qinghai (because Henan Province does not cover the whole province in the “the Winter Clean Heating Plan for Northern China”; therefore, this paper does not analyze the economic data of Henan Province). Compared with other provinces and regions in China, the energy structure in northern China is dominated by fossil energy, which leads to a large amount of SO₂ [6,7]. In addition, northern regions are heating in winter, which undoubtedly consumes more energy and produces more SO₂ emissions [5]. To combat air pollution in northern China, the government has issued a series of policies, such as the “three-year plan on defending the blue sky”. How to balance energy consumption and SO₂ emissions has become an important issue related to the sustainable development of the regional economy and society in northern regions. Therefore, the aims of this paper are to study the spatial correlation between SO₂ emissions and energy consumption in the northern regions of China, and put forward suggestions for SO₂ emissions reduction.

In recent years, the research on SO₂ involve the measurement of emissions [8,9,10], the design of air quality monitoring network [11,12], and the treatment policy discussion [13,14,15]. However, the most concerning of previous studies is the identification of the influencing factors [16,17]. It has been found that the economic growth [2], industrial upgrading [18], and many other factors can affect the emissions of SO₂ [19,20,21,22]. Some studies have discussed the impact of energy consumption on SO₂ emissions [23], and found that energy consumption and regional energy intensity are important factors affecting SO₂ emissions [24,25,26]. However, those studies did not research the spatial relation of SO₂ emissions. Therefore, based on previous studies, this paper focuses on the spatial correlation between energy consumption and SO₂ emissions.

In the spatial correlation of SO₂ emissions, previous studies have discussed the spatiotemporal evolution [27,28]. Some studies believe that the transfers of SO₂ emissions tended to flow from developed countries to developing countries [29]. SO₂ also flows within China. Researchers found SO₂ emissions outsourced from highly developed coastal provinces are transferred to less developed inland provinces while SO₂ emissions from under-developed provinces like Henan and Hebei to the energy intensive provinces including Shanxi and Inner Mongolia have seen a significant increase [30]. Thus, it is necessary to focus on the SO₂ emissions reduction in the northern region.

In recent years, the spatial correlation of air pollutant emissions has been found [31,32,33,34]. Therefore, scholars began to explore the spatial correlation of SO₂ emissions [35]. The spatial econometric model is the common method to study spatial correlation, which has been used to analyze environmental research and development (R&D) [36], high-tech industry development [37], and the share of the industry [4,38] on the spatial effect of SO₂ emissions. However, the above studies mostly discussed the spatial correlation of SO₂ from the perspective of economic development and industrial development, ignoring the spatial influence of energy consumption. In addition, studies use a random effect eigenvector spatial filtering approach examining the spatial effect of coal consumption and oil consumption on SO₂ emissions in China [39]. However, to achieve sustainable development, China has made great efforts to develop renewable energy, natural gas, which has changed the energy structure. However, those studies did not discuss the impact of power generation and natural gas consumption on SO₂. Therefore, based on the previous studies and the new changes in energy structure, this paper discusses the impact of energy consumption on SO₂ emissions from the perspective of spatial correlation.

In general, as far as it is known, there is little literature dedicated to studying the spatial dependence of SO₂ emissions and energy consumption. Based on the research status, the purpose of this paper is to research the spatial correlation between SO₂ emissions and energy consumption in the 14 provinces and autonomous regions in northern China. This study is conducive to clarifying the spatial relationship between SO₂ emissions and energy consumption, and promoting the coordinated treatment of air pollutants.

2. Data and Methods

2.1. Variable Selection and Data Sources

The SO₂ emissions are the total SO₂ emissions in the region in a year, whose unit is measured in tons, and it is expressed in pollution. Additionally, the data of SO₂ emissions are obtained from the National Bureau of Statistics of China.

Energy consumption is always accompanied by SO₂ emissions. With the rise of the national economy, residents have increased their demand for energy consumption. The consumption of different energy sources has a great impact on SO₂ emissions. This paper selects the four different energy consumption indicators of coal consumption, crude oil consumption, natural gas consumption and electricity consumption as the energy consumption structure index, which are expressed by coal, crude, natgas, and electric, respectively, and the units are in Percentages. The detailed definition of the indicators is shown in

Table 1. Variable description.

Variable	Variable Name	Variable Meaning	Unit
pollution	SO2 emissions	Total SO2 emissions in the region	Ton
coal	Coal consumption structure	Regional coal consumption as a percentage of total energy consumption	%
crude	Crude oil consumption structure	Regional crude oil consumption as a percentage of total energy consumption	%
natgas	Natural gas consumption	Regional natural gas consumption as a percentage of total energy consumption	%
electric	Electric consumption structure	Regional electricity consumption as a percentage of total energy consumption	%
firm	The scale of industrial enterprises	The ratio of the number of industrial enterprises above the regional scale to the national average	%
reg	Environmental regulation	Proportion of regional environmental pollution control investment in fiscal general budget expenditure	%
cgdp	GDP per capita	Regional GDP divided by the resident population of the region	yuan

.

There are many control variables that affect SO₂ emissions. This article comprehensively considers the indexes selected by previous research, and the factors that directly or indirectly affect SO₂ emissions [40,41,42]. Therefore, three control variables are selected: GDP per capita, the scale of industrial enterprises, and environmental regulation. Among them, the scale of industrial enterprises refers to the ratio of the number of industrial enterprises above designated size in the region to the national average, and environmental regulation refers to the proportion of regional environmental pollution control investment in the fiscal general budget expenditure. The specific model variables are shown in Table 1. Additionally, the data from 2011 to 2017 are all obtained from the database of the National Bureau of Statistics of China and the EPSDATA “https://www.epsnet.com.cn (accessed on 21 January 2023)”.

2.2. Spatial Autocorrelation Test Model

To investigate whether the energy consumption structure has a spatial effect on SO₂ emissions, the spatial autocorrelation of SO₂ must first be analyzed. To test whether there is spatial autocorrelation among regional variables, generally according to the spatial autocorrelation Moran I index [43,44,45], the calculation formula is as follows (1):

$$\mathrm{Moran} I=\frac{{\sum }_{i=1}^n{\sum }_{j=1}^n{W}_{\mathit{ij}}\left(Y_i-\overline{Y}\right)\left(Y_{\mathrm{j}}-\overline{Y}\right)}{S^2{\sum }_{i=1}^n{\sum }_{j=1}^n{W}_{\mathit{ij}}}$$

(1)

where, $S^2={\sum }_{i=1}^n{\left(Y_i-\overline{Y}\right)}^2$, $\overline{Y}=\frac{1}{n}{\sum }_{i=1}^n{Y}_i$, $Y_i$ is the observation value of the region $i$, and $n$ is the total number of regions. $W_{ij}$ is a binary adjacency space weight matrix, which uses adjacency criteria or distance criteria to define the mutual adjacency relationship of spatial objects. The adjacency criterion is 1 between two regions, and 0 otherwise. In the selection of the spatial weight matrix, the most common geospatial weight is adopted; that is, it is set according to whether the spaces are adjacent. Adjacent areas are assigned “1” and other areas are assigned “0”. The matrix is defined as follows (2):

$$W_{\mathit{ij}}=\{\begin{array}{c}1 \mathrm{Regions} i \mathrm{and} j \mathrm{are} \mathrm{adjacent}\\ 0 \mathrm{Regions} i \mathrm{and} j \mathrm{are} \mathrm{not} \mathrm{adjacent}\end{array}$$

(2)

Generally, the value of Moran I ranges from −1 to 1. A value greater than 0 indicates that there is a positive spatial correlation between regions; that is, there is a spatial clustering phenomenon. The larger the value, the stronger the positive correlation. The value less than 0 indicates a negative spatial correlation, that is, a spatial exclusion phenomenon. The value equal to 0 means that the distribution of an economic variable and location is independent of each other.

2.3. Spatial Econometric Model

Spatial econometrics is an important branch of econometrics, which is used to deal with spatial correlation and spatial heterogeneity in panel data regression models. There are currently three types of spatial econometric models: the Spatial Lag Model (SLM), the Spatial Error Model (SEM), and the Spatial Durbin Model (SDM). These three models correspond to different ways of setting spatial interaction effects.

(1)

Spatial Lag Model

If the spatial interaction effect or spatial autocorrelation comes from substantial correlations, such as energy consumption structure, the scale of industrial enterprises, environmental regulation, and per capita GDP, it can be analyzed by adding the spatial lag factor of the dependent variable. The SLM mathematical formula is as follows (3):

$$Y=\rho \mathit{WY}+X\beta +\epsilon$$

(3)

where, $Y$ is the explanatory variable, $X$ is the exogenous explanatory variable matrix, $\beta$ is the parameter vector of $X$, and $\rho$ is the spatial lag regression coefficient. $\epsilon$ represents a random error term. $W$ is the spatial weight matrix (N × N order matrix, and N is the number of regions).

(2)

Spatial Error Model

In the model setting process, it is likely that some variables related to the interpreted variables are omitted (the variables are hidden or cannot be accurately quantified), and these variables have spatial autocorrelation, and random errors may impact the space spillover effect between regions. Therefore, in some cases, ignoring the spatial autocorrelation of errors can also cause bias in the model setting. The mathematical formula of the spatial error model is as follows (4) and (5), respectively:

$$Y=X\beta +u$$

(4)

$$u=\lambda \mathit{Wu}+\epsilon$$

(5)

where, $u$ is the error term, and $\lambda$ is the spatial error regression coefficient.

(3)

Spatial Durbin Model

The disadvantage of the spatial lag model and the spatial error model is that the spatial pattern of the data may not be explained by only endogenous interaction effects or perturbations with autocorrelation. Instead, it may require the use of endogenous interaction effects, exogenous interaction effects, and error terms with autocorrelation. The mathematical expression of the spatial durbin model is as follows (6):

$${Y=\rho \mathit{WY}+X\beta }_i{+\mathit{WX}\delta }_i+\epsilon$$

(6)

where, $W$ is the spatial weight, ${\beta }_i$ is the parameter vector of $X$, ${\delta }_i$ is the spatial autocorrelation coefficient of exogenous variables, and $\epsilon$ is a random perturbation term that satisfies the normal independent and identical distribution. In the calculation of the model, all values are logarithmic to avoid singular matrices from affecting the results. The type of spatial econometric model is commonly determined by the LM Error test, LM Lag test, Wald test, and LR test. The model selecting process is shown in Figure 2. As the test methods are mature, they will not be introduced in detail.

3. Results and Discussion

3.1. The Results of Spatial Autocorrelation Test

According to the SO₂ emissions from the provinces of northern China from 2011 to 2017, the corresponding Moran I index, and the test results are calculated using STATA16.0 software in combination with Equation (1), which are shown in

Table 2. Province-wide Moran I Index.

Year	Moran I	Z Value	p Value
2011	0.274	1.808	0.035
2012	0.255	1.711	0.044
2013	0.231	1.582	0.057
2014	0.206	1.453	0.073
2015	0.211	1.485	0.069
2016	0.272	1.895	0.029
2017	0.222	1.547	0.061

. It can be seen that the Moran I index of SO₂ emissions in the provinces and regions is positive, and the SO₂ emissions are stable at the significance level of 10%. It shows that the SO₂ emissions of various provinces have a significant positive autocorrelation relationship in the spatial distribution in the year under investigation.

SO₂ emissions have a spatial clustering feature. The spatial characteristics of SO₂ emissions can be vividly illustrated by a scatter plot of the Moran I index. To reflect the changes of local spatial characteristics in each province, this paper draws a scatter plot of the Moran I index of each province from 2011 to 2017. The scatter diagram and saliency diagram of the specific SO₂ Moran I are shown in Figure 3. Where, the numbers in the figure indicate: 1: Beijing, 2: Tianjin, 3: Hebei, 4: Shanxi, 5: Inner Mongolia, 6: Liaoning, 7: Jilin, 8: Heilongjiang, 9: Shandong, 10: Shaanxi, 11: Gansu, 12: Qinghai, 13: Ningxia, and 14: Xinjiang.

According to Figure 3, high-high (H-H) and low-low (L-L) type regions dominate, with most of the provinces and regions clustered in quadrants I and III. In 2011, five provinces and regions were located in quadrant I: Hebei, Shanxi, Liaoning, Shandong, and Shaanxi, showing a positive autocorrelation cluster of high-high (H-H). Additionally, five provinces and regions, including Beijing, Tianjin, Gansu, Qinghai, and Xinjiang, are located in the third quadrant and exhibit low-low (L-L) spatial autocorrelation. In 2013, Xinjiang shifted from the third quadrant to the fourth quadrant. In 2016, Heilongjiang and Ningxia shifted from the second quadrant to the third quadrant, and Shaanxi shifted from the first quadrant to the second quadrant. In 2017, four provinces and regions, including Hebei, Shanxi, Liaoning, and Shandong, located in the first quadrant exhibit a high-high (H-H) positive autocorrelation cluster. While six provinces, including Beijing, Tianjin, Gansu, Qinghai, Ningxia, and Heilongjiang, are located in the third quadrant and exhibit low-low (L-L) spatial autocorrelation.

Combining the Scatter plot of the Moran I index of SO₂ emissions from 2011 to 2017, it can be seen that Jilin and Inner Mongolia have been in quadrants II and IV, the rest of the provinces have basically been in quadrants I and III with spatial correlation. Inner Mongolia has a tendency to shift to the first quadrant, Shaanxi has some fluctuations and a tendency to shift to the third and first quadrants, and Jilin has a tendency to shift to the third quadrant. In summary, the SO₂ emissions in the 14 northern provinces and regions have obvious spatial distribution characteristics, and it is extremely necessary to conduct a spatial analysis of SO₂ emissions.

3.2. The Results and Discussion of Model Solving

3.2.1. The Results of Model Solving

Through the data analysis, it is necessary to study the spatial characteristics of SO₂ in 14 provinces in northern China. This article uses the economic data of these regions from 2011 to 2017 and applies the spatial panel measurement method and MATLAB software to determine the model type. According to the model selection process in Figure 2, the LM lag Test, LM error Test, Robust LM lag Tests, and Robust LM error Tests are needed. Additionally, the test results are shown in

Table 3. LM test results.

Test	Value	p-Value
LM lag test	0.35	0.034
robust LM lag test	1.1022	0.064
LM error test	0.5403	0.032
robust LM error test	1.2925	0.056

.

From the data in Table 3, it can be seen that the regression results of the LM lag test and LM error test show that they pass the test at the 10% significance level. It is indicated that the spatial econometric model can be used for analysis.

From the spatial measurement test results in

Table 4. Space test results.

Test	Statistics	p-Value
Hausman	32.51	0.003
Wald spatial lag	22.5533	0.002
LR spatial lag	21.7338	0.0028
Wald spatial error	24.9988	0.0007
LR spatial error	24.2687	0.001

, it can see that both the Wald spatial lag test and the Wald spatial error test passed the 1% significance level, indicating that the SDM model cannot be simplified into SLM and SEM models. Then, according to the Hausman test, the fixed effects or random effects are evaluated. The results of Hausman test shown that the fixed effect model should be adopted.

To make the analysis rigorous, the spatial fixed, time fixed, and double fixed effect models of SLM, SEM, and SDM have been established. All calculation results are shown in

Table 5. Results of SLM model calculation of 14 provinces in northern China from 2011 to 2017.

Variable	Space-Fixed	Time-Fixed	Double-Fixed
ln(coal)	−0.236805	0.606774 *	0.046755
ln(crude)	−0.279980 ***	−0.279638 ***	−0.331107 ***
ln(natgas)	−0.111302	0.117752	−0.054648
ln(electric)	−0.027053	−0.439097 **	−0.217198
ln(firm)	0.340663 ***	0.738655 ***	0.670068 ***
ln(reg)	−0.695054 ***	−0.063302	−0.100911
ln(cgdp)	−0.775865 ***	−1.610074 ***	−1.207676 ***
W*dep.var	0.356973 ***	0.126988 *	0.060370 *
log-likelihood	13.319502	−9.0310913	49.991949
R-squared	0.9537	0.9238	0.9769
sigma2	0.0494	0.0751	0.0263

Note: *, **, and *** indicate that the statistics are significant at the significance level of 10%, 5%, and 1%, respectively.

,

Table 6. Calculated SEM model results of 14 provinces in northern China from 2011 to 2017.

Variable	Space-Fixed	Time-Fixed	Double-Fixed
ln(coal)	0.065146	0.578399 *	0.032080
ln(crude)	−0.296668 ***	−0.267374 ***	−0.361589 ***
ln(natgas)	0.019678	0.125235	−0.065204
ln(electric)	−0.412520 ***	−0.435818 **	−0.164065
ln(firm)	0.536658 ***	0.710759 ***	0.719561 ***
ln(reg)	−0.269814 **	−0.089276	−0.090760
ln(cgdp)	−1.169465 ***	−1.483610 ***	−1.314848 ***
spat.aut.	0.750999 ***	0.161986 *	0.076073 *
log-likelihood	26.834229	−9.5254691	50.356996
R-squared	0.8933	0.9213	0.9767
sigma2	0.0309	0.0758	0.0262

Note: *, **, and *** indicate that the statistics are significant at the significance level of 10%, 5%, and 1%, respectively.

and

Table 7. Calculated SDM model results of 14 provinces in northern China from 2011 to 2017.

Variable	Space-Fixed	Time-Fixed	Double-Fixed
ln(coal)	−0.009327	0.554319 *	0.032917
ln(crude)	−0.318253 ***	−0.278919 ***	−0.324738 ***
ln(natgas)	0.027080	0.121777	0.013251
ln(electric)	−0.559424 ***	−0.649843 ***	−0.483385 ***
ln(firm)	0.579426 ***	0.746949 ***	0.660460 ***
ln(reg)	−0.206039 *	0.137562	−0.051362
ln(cgdp)	−1.064286 ***	−1.594257 ***	−1.092002 ***
W*ln(coal)	−0.042288	−0.222599	−0.116486
W*ln(crude)	0.159295 **	−0.110297	−0.173175 **
W*ln(natgas)	−0.228624 **	0.002353	−0.151666 *
W*ln(electric)	0.770627 ***	0.267088	0.532241 ***
W*ln(firm)	−0.376940 ***	0.131731	0.307019 ***
W*ln(reg)	−0.215083	−0.560945 **	−0.014312
W*ln(cgdp)	0.630607 **	0.301205	−0.595836 *
W*dep.var	0.572989 ***	0.200941 *	0.147971 *
log-likelihood	37.645102	−2.6162142	62.491359
R-squared	0.9736	0.9337	0.9824
sigma2	0.0242	0.0607	0.0161

Note: *, **, and *** indicate that the statistics are significant at the significance level of 10%, 5%, and 1%, respectively.

.

Additionally, the data in Table 5, Table 6 and Table 7 shows that the SDM model has the highest R², the largest log-likelihood, and the smallest sigma²; meaning that the goodness of fit and log likelihood are high, and the errors are low. Therefore, among the three models, the SDM model should be selected as the theoretical model for the spatial distribution of SO₂ emissions in 14 provinces in northern China. In the SDM model, from the point of the goodness of fit and log-likelihood value, the double fixed model is the best. Therefore, the results of this paper are based on SDM’s double fixed model.

3.2.2. The Discussion of Model Solving Results

From the results of the SDM model in Table 7, the value of W*dep.var is 0.147971 and passes the 10% significant level test, which indicates that the spatial spillover effect of SO₂ is significant among the 14 provinces and regions in northern China. Additionally, the direct and indirect effects of the variables on SO₂ are shown in

Table 8. The Direct, indirect and total effects of variable on SO₂ in double fixed model SDM Model.

--	ln(coal)	ln(crude)	ln(natgas)	ln(electric)	ln(firm)	ln(reg)	ln(cgdp)
Direct Effects	0.0323	−0.3264 ***	0.0115	−0.5135 ***	0.6604 ***	−0.0613	−1.1201 ***
Indirect Effects	−0.1164	−0.1731 **	−0.1516 *	0.5322 ***	0.3171 ***	−0.0143	−0.5917 *
Total Effects	−0.0865	−0.5122 ***	−0.1332	0.0352	0.9846 ***	−0.0735	−1.7137 **

Note: *, **, and *** indicate that the statistics are significant at the significance level of 10%, 5%, and 1%, respectively.

based on the double fixed model SDM Model. The main results of this paper are as follows.

First, it should be noted the impact of the proportion of natural gas and electricity on SO₂. The research results show that natural gas has an insignificant impact on local SO₂ emissions but has an inhibitory effect on adjacent regions, while electricity reduces SO₂ emissions in the location, but increases SO₂ emissions in adjacent regions. There is a large amount of power transmission between regions in China. Beijing and Tianjin, as load centers, consume a large amount of power from other provinces and regions. As it is known, the SO₂ of the electric power industry mainly comes from power production, and power consumption usually does not produce SO₂. Thus, the power transmission leads to the negative effect of power consumption on the SO₂ emissions in local regions and the positive effect in adjacent regions. This result proves that the SO₂ emissions transfer through power transmission. Based on this research result, this paper suggests that the transmission of SO₂ emissions caused by power transmission should be strengthened in the control of SO₂ emissions.

Second, the direct and indirect effects of the scale of industrial enterprises on SO₂ are positive, and pass the 1% significant level test, which indicates that an increase in the scale of industrial enterprises will exacerbate SO₂ missions in the local region and other regions. Industrial enterprises are an important economic source for regional development and an important source of SO₂ emissions. In recent years, however, regions have increasingly focused on local industrial SO₂ prevention and control, taking measures, such as installing desulfurization towers and building desulfurization ponds to reduce SO₂ emissions. To further control SO₂ emissions, localities should deepen industrial reform and industrial transformation, and accelerate the elimination of highly polluting traditional industries.

Third, the proportion of oil and GDP per capita will significantly inhibit SO₂ emissions in local and adjacent regions. The consumption structure of crude oil has a negative impact on SO₂ emissions, which indicates that the crude oil production process in these regions has been greatly improved and desulfurization facilities have been retrofitted in a timely manner. GDP per capita can reflect the economic level of a region. The more economically developed a region is, the more advanced its economic development is, the more thorough it is in eliminating backward production capacity, and, thus, the less SO₂ is emitted into the air. As GDP per capita rises, the standard of living rises, and people gradually pursue healthier lifestyles. People will buy eco-friendly products and the local government will have more money to invest in environmental protection and incentives for eco-friendly product development. This in turn will reduce the demand for highly polluting products and reduce SO₂ emissions.

Finally, the impact of the coal consumption structure and environmental regulation investment on SO₂ emissions is insignificant. Coal is used as the main energy source in China over a long period of time, and the proportion of coal in energy consumption is relatively stable, which leads to the insignificant impact of coal consumption on SO₂ emissions. Taking Hebei Province as an example, according to the statistics of Hebei Provincial Bureau of Statistics, the proportion of coal consumption in the total energy consumption in Hebei Province is 89.09% in 2011, and 86.05% in 2017, which only changed by 3.04%. The environmental regulation investment in this paper is the proportion of regional environmental pollution control investment in fiscal general budget expenditure, which only accounts for a small part of fiscal general budget expenditure. This makes the insignificant impact of environmental regulation on SO₂. Although the impact of the coal consumption structure and environmental regulation investment on SO² emissions is insignificant, to prevent the rebound of SO² emissions, it is still necessary to pay attention to control coal consumption and promote the investment in environmental regulation.

4. Conclusions

In recent years, China has made remarkable achievements in SO₂ emissions reduction, but the emissions in the northern regions of China are still significant. As the energy structure is dominated by fossil energy, the SO₂ emissions of the 14 provinces and regions in north China still need attention. To scientifically guide the coordinated treatment of SO₂, in this paper, spatial econometric methods are used to analyze the economic data and energy consumption data of 14 provinces and regions in northern China. Based on the results of spatial econometric model solving and discussion, the following conclusions are drawn:

• Spatial measurements based on the 0-1 matrix show that there are significant spatial clustering characteristics of SO₂ emissions in 14 provinces and regions in northern China, most of which belong to the high-high and low-low categories. National SO₂ control should focus on the overall situation, fully consider the spatial dependency, and systematically address the air quality problems in the northern region and take spatial characteristics into account in the specific policy measures.
• The consumption of electricity has a negative effect on the SO₂ emissions in the local region, but it has a positive effect on the adjacent regions, which indicated that the SO₂ emissions can be transferred by the electricity transmission. Based on the characteristics of large-scale power transmission across regions in China, SO₂ emissions reduction should focus on power production regions. It is suggested developing new energy power generation, strengthening coal power desulfurization treatment, and other methods to reduce SO₂ emissions in power production region.
• The scale of industrial enterprises is an important factor in the growth of SO₂ emissions in locations and adjacent regions. Therefore, the northern region should still focus on SO₂ treatment in industrial industries. To reduce SO₂ emissions from industrial enterprises, it is suggested to optimize industrial structure and increase desulfurization facilities.