Modeling the spatiotemporal dynamics of industrial sulfur dioxide emissions in China based on DMSP-OLS nighttime stable light data

Due to the rapid economic growth and the heavy reliance on fossil fuels, China has become one of the countries with the highest sulfur dioxide (SO2) emissions, which pose a severe challenge to human health and the sustainable development of social economy. In order to cope with the serious problem of SO2 pollution, this study attempts to explore the spatial temporal variations of industrial SO2 emissions in China utilizing the Defense Meteorological Satellite Program’s Operational Linescan System (DMSP-OLS) nighttime stable light (NSL) data. We first explored the relationship between the NSL data and the statistical industrial SO2 emissions at the provincial level, and confirmed that there was a positive correlation between these two datasets. Consequently, 17 linear regression models were established based on the NSL data and the provincial statistical emissions to model the spatial-temporal dynamics of China’s industrial SO2 emissions from 1997 to 2013. Next, the NSL-based estimated results were evaluated utilizing the prefectural statistical industrial SO2 emissions and emission inventory data, respectively. Finally, the distribution of China’s industrial SO2 emissions at 1 km spatial resolution were estimated, and the temporal and spatial dynamics were explored from multiple scales (national scale, regional scale and scale of urban agglomeration). The results show that: (1) The NSL data can be successfully applied to estimate the dynamic changes of China’s industrial SO2 emissions. The coefficient of determination (R2) values of the NSL-based estimation results in most years were greater than 0.6, and the relative error (RE) values were less than 10%, when validated by the prefectural statistical SO2 emissions. Moreover, compared with the inventory emissions, the adjusted coefficient of determination (Adj.R-Square) reached 0.61, with the significance at the 0.001 level. (2) During the observation period, the temporal and spatial dynamics of industrial SO2 emissions varied greatly in different regions. The high growth type was largely distributed in China’s Western region, Central region, and Shandong Peninsula, while the no-obvious-growth type was concentrated in Western region, Beijing-Tianjin-Tangshan and Middle south of Liaoning. The high grade of industrial SO2 emissions was mostly concentrated in China’s Eastern region, Western region, Shanghai-Nanjing-Hangzhou and Shandong Peninsula, while the low grade mainly concentrated in China’s Western region, Middle south of Liaoning and Beijing-Tianjin-Tangshan. These results of our research can not only enhance the understanding of the spatial-temporal dynamics of industrial SO2 emissions in China, but also offer some scientific references for formulating feasible industrial SO2 emission reduction policies.

Introduction Sulfur dioxide (SO 2 ) is one of the main pollutants in the atmosphere, which is an important indicator to measure whether the atmosphere is polluted [1]. SO 2 is released when burning materials containing sulfur, which is found in all types of coal and oil across the world in varying proportions. As an acidic and toxic gas, SO 2 leads to global acid rain, visibility degradation, destroy of terrestrial and aquatic ecosystems and dangerous impacts on the human health [2][3][4][5][6]. In brief, it causes serious losses to the whole society and the economy. SO 2 in the atmosphere can be emitted from both natural and anthropogenic sources [7]. Natural SO 2 sources mainly come from oxidation of biogenic dimethyl sulfide and volcanic eruption [8]. While, the primary anthropogenic SO 2 emission is from fossil fuel consumption, especially for power generation, and other industrial production activities [9]. It is estimated that anthropogenic emissions are the main source of global SO 2 emissions [10,11].
China is one of the countries with the highest SO 2 emissions [12] and the largest area exposed to acid precipitation in the world [13], because its huge economy relies heavily on fossil fuels as an energy source [14]. In 2015, China emitted 18.591 million tons of SO 2 , of which 83.73% was industrial SO 2 [15]. In order to reduce SO 2 emission to mitigate the adverse impacts, a range of measures has been advanced by the Chinese government, including installing flue gas desulfurization on facilities power plants [16], increasing the proportion of nonfossil fuels [17], and carrying out a plan that the SO 2 emission in 2020 shall reduce by 15% compared with 2015 [18]. Although the Chinese government has made tremendous efforts to control the air quality, the current situation of air pollution is still not optimistic, as for the air quality in many cities still could not meet the national air quality standards [19]. For instance, in 2017, only 99 out of the 338 cities in China met the environmental air quality standards, while 70.7% failed to achieve the national air quality standards [20]. More seriously, SO 2 can affect the atmosphere and environment on a global scale [21,22]. It means that studying on SO 2 emissions in China is also of great significance to improve global environmental performance. Therefore, it is urgently necessary to explore the dynamic characteristics of SO 2 emissions in China, for obtaining a better understanding of the current air pollution situation and formulating appropriate emission reduction policies.
Remote sensing can provide valuable data sources in the research of detecting the spatial and temporal changes of geospatial information. Numerous literatures have proved that DMSP-OLS nighttime light imagery can perform well in the detection of socioeconomic activities, such as, monitoring urban dynamics [23][24][25][26], measuring spatial distribution of population [27][28][29][30], investigating economic development [31][32][33], estimating power energy consumption [34][35][36][37], etc. The significant correlation between the DMSP-OLS NSL data and energy consumption has been verified [38,39], and the NSL data has been successfully applied to fossil fuel related CO 2 emissions [40][41][42][43][44] and pollutant emissions (such as PM 2.5 emissions and nitrogen oxides emissions) [45][46][47][48][49][50]. For instance, Ghosh et al. [41] applied nighttime satellite imagery favorably to map fossil fuel CO 2 emissions. Li et al. [45] utilized nighttime light imagery to PM 2.5 pollution estimation in Beijing using an established model and the average precision of the estimation reached 0.796. Xu et al. [46] taking Shanghai, China as an example, verified the effectiveness of DMSP nighttime light images in predicting urban daily PM 2.5 concentrations. Toenges-Schuller et al. [47] employed DMSP-OLS nighttime light images for detecting the global distribution patterns of anthropogenic nitrogen oxides emission. Jiang et al. [48] quantified the spatial-temporal dynamics of nitrogen oxides emissions in China utilizing a NSL based model. In summary, DMSP-OLS NSL data has been proved to be promising in monitoring fossil fuel related CO 2 emissions and pollutant emissions. As anthropogenic SO 2 emissions are mainly from fossil fuel combustion [51,52], thus, there is a good potential to estimate industrial SO 2 emissions based on NSL data. Moreover, existing researches have also demonstrated that there was a significant correlation between nighttime light data and gross domestic product of secondary industry [53,54]. Therefore, theoretically, the industrial SO 2 emissions can also be estimated utilizing nighttime light data. In addition, different from the bottom-up emission inventories which are usually highly uncertain and not timely updated [55], NSL data can provide spatial explicit images with high resolution in time.
But there are few studies about whether and how the NSL data could be applied to estimate the spatial-temporal dynamics of industrial SO 2 emissions. Hence, we tried to apply the nighttime light data to estimate industrial SO 2 emissions in China.
This study aims to test the utility of modeling the spatiotemporal dynamics changes of China's industrial SO 2 emission based on DMSP-OLS nighttime stable light data. Specifically, the major objectives of our study are: (1) confirming a positive correlation was truly existed between the industrial sulfur dioxide emissions and DMSP-OLS NSL data, (2) building models to investigate China's industrial SO 2 emission utilizing the NSL data, and evaluating the estimation accuracy using statistical emissions and emission inventory data, respectively, (3) exploring the spatiotemporal distribution characteristics of China's industrial SO 2 emission from three different scales on the basis of the NSL-based estimation results, and putting forward some suggestions on industrial SO 2 emission mitigation, accordingly.

Study area
In order to learn more about the spatiotemporal variations and changes of industrial SO 2 emissions in China for 1997-2013, the research area was determined by three different administrative levels (Fig 1). National scale is the first administrative level. In recent years, a vast volume of SO 2 emitted from industrial production has brought about serious atmospheric pollution. It is necessary to explore the overall situation of industrial SO 2 emissions in the whole country firstly. Considering that the statistical industrial SO 2 emissions data was absent in some areas (Hong Kong, Macao and Taiwan), the first level was limited to mainland China. Then, regional scale is the second level. Because of the imbalanced socioeconomic development in our country, great disparities of industrial SO 2 emissions within different economic regions have been formed. In order to reveal the differences among regions, we divided the research area into four regions according to its geographical position and socioeconomic development level. So, Eastern region, Central region, Western region and Northeastern region were studied separately. Finally, the scale of urban agglomeration is the third administrative level. As far as we know, the population and economic growth of China were mainly concentrated in urban agglomerations, these areas contributed more to industrial SO 2 emissions in the whole country. Consequently, it is of great significance to explore the characteristic of industrial SO 2 emissions in urban agglomerations for reducing SO 2 pollution. Therefore, six representative urban agglomerations were finally chosen as the third level, namely Middle south of Liaoning, Beijing-Tianjin-Tangshan, Shandong Peninsula, Pearl River Delta, Sichuan-Chongqing and Shanghai-Nanjing-Hangzhou.

Data
There are mainly two kinds of data sets utilized in this research, namely, DMSP-OLS NSL data and statistical industrial SO 2 emissions data. NSL data from 1997 to 2013 were derived from the National Oceanic and Atmospheric Administration's National Geophysical Data Center (NOAA/NGDC) website (http://www.ngdc.noaa.gov/eog/dmsp/downloadV4composites. html). These NSL images measured lights on the Earth's surface from human settlements, road networks and other sites with continuous lighting. The digital number (DN) value of the NSL imagery ranges from 0 to 63. And, the spatial resolution of the NSL images is 0.0083˚(about 1 km). Due to the following two shortcomings of DMSP-OLS NSL data: (1) pixel saturation effect and (2) discontinuity and incomparability phenomenon, it is necessary to intercalibrate the data before using it. In order to intercalibrate the NSL data, Shi et al. [37] developed a modified invariant region (MIR) method, consisting of reduction of saturation effect and

PLOS ONE
correction of discontinuity effect. In our study, the time series NSL data were directly intercalibrated based on the specific equations and parameters reported in the article of Shi et al.

Methodology
For investigating the spatial-temporal dynamics of China's industrial SO 2 emissions, main research procedures include: (1) correlation analysis between nighttime light images and statistical industrial SO 2 emissions at the provincial level; (2) estimation of industrial SO 2 emissions at 1 km resolution based on NSL data and provincial statistics emissions; (3) accuracy assessment of industrial SO 2 emissions estimation by prefectural statistical emission data and emission inventory data; (4) multiscale analysis of the spatiotemporal characteristics of industrial SO 2 emissions.

Correlation analysis
Correlation analysis was employed to analyze whether there exists a significant correlation between DMSP nighttime light images and statistical industrial SO 2 emissions or not. The formulas we used for correlation analysis can be represented as follows: ðx k À xÞðy k À yÞ ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi X n k¼1 ðx k À xÞ 2 X n k¼1 ðy k À yÞ 2 s ð1Þ Where, r xy is expressed as the degree of correlation between variable x and variable y, whose value ranges from -1 to 1. And the closer absolute value of r xy is to 1, the stronger correlation between x and y is. In this paper, variables x and y represent DN values of DMSP nighttime light images and industrial SO 2 emissions, respectively.

Estimation of industrial SO 2 emissions
Once the significant positive correlation between NSL data and statistical industrial SO 2 emissions data is confirmed, industrial SO 2 emission at the provincial level can be simulated using NSL data. Then, the linear regression model was performed to estimate industrial SO 2 emissions, and the equation can be described as the following: Where S P stands for the provincial industrial SO 2 emissions, TNSL represents the total nighttime stable light values of each province, a is the regression coefficient, and b stands for the intercept.
Considering the absence of industrial SO 2 emissions at the pixel level, the positive correlation between NSL data and industrial SO 2 emissions is assumed to be constant within the same province. Additionally, provincial statistical industrial SO 2 emissions data were employed to correct the estimation models to reduce the error within a provincial unit. The formula is: Where CS i indicates the corrected industrial SO 2 emissions of the i pixel; SS p stands for the statistical industrial SO 2 emissions of the p province; S i is the estimated industrial SO 2 emissions of the i pixel; TS p represents the estimated industrial SO 2 emissions of the p province.

Accuracy assessment of industrial SO 2 emissions estimation
It is necessary and crucial to assess the accuracy of industrial SO 2 emissions estimation. Two indicators, the coefficient of determination (R 2 ), and the relative error (RE) were often used to assess the accuracies of simulated results [56,57]. For instance, Shi et al. [37] modeled the spatiotemporal dynamics of global electric power consumption by DMSP-OLS NSL data, and 128 samples of country-level statistical electric power consumption data from 1992 to 2012 were collected to calculate the R 2 and RE which were employed to evaluate the estimation accuracy.
To validate the accuracy of industrial SO 2 estimation models, R 2 , and RE are calculated: Where, m is the total number of validation regions which was set to 263 in our research, SS c stands for the statistical industrial SO 2 emissions of the c city, SS c indicates the average of SS c , S c is the estimated industrial SO 2 emissions of the c city. In the above three parameters, the higher R 2 value and the lower absolute RE values indicate a higher simulation accuracy.

Evaluation of spatiotemporal dynamics of industrial SO 2 emissions
First of all, the average industrial SO 2 emissions from 1997 to 2013 was calculated using the following equation to analyze the spatial pattern of industrial SO 2 emissions: Where S i stands for the average industrial SO 2 emissions in pixel i for 1997-2013, and t indicates the total number of years which was set to 17 in our research. Next, the temporal variation of industrial SO 2 emissions between 1997 and 2013 can be described by the following formula:

PLOS ONE
Where S tem i stands for the temporal variation of industrial SO 2 emissions in pixel i from 1997 to 2013.
Then, the Natural Break method, which can maximize the differences between classes with no effect of human factors [58], was used for investigating the spatial and temporal changes of industrial SO 2 emissions in China. In detail, the spatial variation map of industrial SO 2 emissions was divided into five grades: low (< 3 t), relatively low (3-15 t), medium (14-37 t), relatively-high (37-70 t) and high (> 70 t). And, the temporal variation of China's industrial SO 2 emissions was sorted into 4 types: no-obvious-growth (< 2t), low growth (2-9 t), moderategrowth (9-22 t), and high growth (> 22 t).

Correlation analysis results
The relationship between the DMSP nighttime light imagery and industrial SO 2 emissions for 1997-2013 was confirmed utilizing the formulas as described in Section 3.1. And the correlation coefficients between the two from 1997 to 2013 were listed in Table 1. Obviously, the correlation coefficients over the observation period are all greater than r 0.001 = 0.5974, which demonstrated that the correlation between these two datasets was significant at the level of α = 0.001. Based on this, 17 liner regression models were constructed to estimate industrial SO 2 emissions. And the F values of these models are all greater than F 0.005 (1, 29) = 9.23, revealing a statistical significance at the level of α = 0.005.
Spatiotemporal dynamics of industrial SO 2 emissions for 1997-2013. The spatial-temporal variations of China's industrial SO 2 emissions during the period of 1997-2013 was mapped in Fig 2. In terms of spatial distribution, industrial SO 2 emissions in China were mainly concentrated in the eastern half of the country. Specifically, the high industrial SO 2 emissions were clearly identified in some economically developed cities like Yangtze River Delta, Sichuan-Chongqing, and Pearl River Delta and Huang-Huai-Hai region. While, the low industrial SO 2 emissions were largely distributed in the western and northeastern China. In terms of the temporal variations, China's industrial SO 2 emissions increased significantly in the initial period from 1997 to 2013 and then slowly decreased. Overall, the temporal and spatial changes of industrial SO 2 emissions in China are significant.  (Fig 4a) and five grades (Fig 4b) in China. On the whole, we can find that the growth of industrial SO 2 emissions was mostly distributed in 6.32% of China' total areas (Figs 3a and 4a). Specifically, two types (including high-growth

PLOS ONE
type and moderate-growth type) with relatively rapid growth, accounting for 1.35% of the national area, were mainly distributed in coastal areas and some metropolitan areas, including Chongqing and provincial capital cities. While, the other two types reflecting time variations of industrial SO 2 emissions, including no-obvious-growth type and low-growth type, which occupy 93.68% and 4.97% of the total national areas respectively, were located in the Western and Northeastern regions mainly. As for spatial variations, there existed a similar variations pattern. The low and relatively-low grade of industrial SO 2 emissions were mainly distributed in the Western region, covering 88.97% and 8.60% of the total national land, respectively (Figs  3b and 4b). Moreover, the high, relatively-high and medium grades accounted for 0.20%, 0.64%, 1.59% of the total national land, respectively, mainly distributing in the coastal and Central areas and Sichuan-Chongqing.
Spatiotemporal dynamics of industrial SO 2 emissions at regional scale.  Fig 5a, the high-growth type was mostly distributed in the Western region and Central region, occupying 62.47% and 25.17% of the total areas of that type, respectively. The moderate-growth and low-growth types were comparatively evenly located in China's Western region, Central region and Eastern region. While, the no-obvious-growth type was concentrated in the Western region mainly, covering 73.31% of the total areas of that type. Summarily, the high-growth of the industrial SO 2 emissions was mostly located in the Western region, followed by Central region, and the no-obvious-growth was mainly located in the Western region. Both the highgrowth and no-obvious-growth of the industrial SO 2 emissions accounted for a large share in the western region, partly due to the large size of this region. Additionally, it is notable that the distribution of these four types in the Northeastern region was relatively small. This probably resulted from its relatively small size. As for the areal percentage of each grade, 38.57% of the high grade was distributed in Eastern region, 30.82% was located in Western region, 23.46% in Central region and the rest was located in Northeastern region (Fig 5b). The relatively-high grade was largely distributed in Eastern region where accounted for 52.82% of the total areas of this grade. While, 76.12% of the low grade was located in the Western region, with a fairly small proportion distributed in the Eastern, Central, and Northeastern regions. Similar to the four types, the proportions of the five grades of industrial SO 2 emissions in the Northeastern region were all relatively low. In summary, the high grade of industrial SO 2 emissions was mostly concentrated in the Eastern and Western regions, while the low grade was mainly located in the Western region.
Spatiotemporal dynamics of industrial SO 2 emissions at urban agglomeration scale. The representative six urban agglomerations covered 7.68% land area of this country, but contributed 35.33% of China's industrial SO 2 emissions for 1997-2013. In terms of the percentage of the total areas, 88.55% of Beijing-Tianjin-Tangshan and 86.17% of Middle south of Liaoning showed a no-obvious-growth type (Fig 6a). The growth of industrial SO 2 emissions was 21.32% in Pearl River Delta and 17.01% in Shanghai-Nanjing-Hangzhou presented a low-growth type, whereas 7.86% in Shanghai-Nanjing-Hangzhou and 6.41% in Pearl River Delta described a moderate-growth type. Besides, Shandong Peninsula should be paid much more attention to, because the areal percentage of high-growth type in this urban agglomeration was the biggest, reaching 1.27 percent. To sum up, the high-growth type of industrial SO 2 emissions was mainly located in Shandong Peninsula, while Beijing-Tianjin-Tangshan and Middle south of Liaoning showed a no-obvious-growth variation. Moreover, the low grade of industrial SO 2 emissions in Middle south of Liaoning and Beijing-Tianjin-Tangshan were 66.79% and 58.10%, respectively (Fig 6b). In addition, 3.27% in the Shanghai-Nanjing-Hangzhou showed a high grade and 10.76% presented a relatively-high grade. 2.43% of Shandong Peninsula presented a high grade and 4.54% showed a relatively-high grade. Summarily, the high grade of industrial SO 2 emissions was concentrated in Shanghai-Nanjing-Hangzhou and Shandong Peninsula, while Middle south of Liaoning and Beijing-Tianjin-Tangshan presented a low grade.

Accuracy evaluation of industrial SO 2 emissions estimation
Since we used the provincial statistical data to estimate the industrial SO 2 emissions, it is reasonable and reliable to assess the accuracy of the estimation models utilizing the industrial SO 2  (Fig 7).
It can be found that the minimum coefficient of determination (R 2 ), was approximately 0.58, and all the other R 2 values were higher than 0.6. Additionally, the figure also reflected

PLOS ONE
China's PM 10 emissions using DMSP-OLS data, the estimation accuracy was validated by citylevel statistical PM 10 emissions for 1995, 2000 and 2005, and the R 2 were 0.5217, 0.5437, and 0.5158, respectively. Zhao et al. [57] employed nighttime light datasets to simulate urban residential CO 2 emissions in China, and the maximum RE is 23.084%, the average absolute value of RE is 12.84%.
Moreover, in order to further evaluate the reliability of the NSL-based estimation results, we compared the estimated industrial SO 2 emissions with the Multi-resolution Emission Inventory for China (MEIC) database, which is a bottom-up emission inventory framework developed and maintained by Tsinghua University [60]. The emission inventory utilized in this article was downloaded from http://www.meicmodel.org/dataset-mix.html, where it is freely available for non-commercial purposes. Based on the availability of data, we compared the emissions of 356 prefecture level cities in 2008 and 2010. And the comparison results are shown in Fig 8. It can be discovered that the Adj.R-Square reached 0.61, with the significance at the 0.001 level. In other words, the DMSP-OLS NSL-based estimation results are acceptable compared with the emissions estimated by other approach.
Therefore, the accuracy evaluation results show that the industrial SO 2 emissions in China can be modeled by NSL data. In addition, it is worth noting that the inconsistency of statistical

PLOS ONE
criterion between provincial and municipal statistical data affects the simulation accuracy to a certain extent. We believe that the simulation accuracy will be improved if the availability and reliability of statistical data was enhanced.

Suggestions for industrial SO 2 emissions reduction
In order to reduce China's industrial SO 2 emissions to achieve the sustainable development of social economy, more efforts should be made to adjust, optimize and upgrade the industrial structures and enhance the energy utilizing efficiency. Considering the great differences in economic development among regions, the Chinese government should adopt differentiated mitigation strategies for different regions. For the Eastern and Central regions with higher levels of economic development, the reduction strategies of industrial SO 2 emissions should focus on adjusting and optimizing the industrial structure. High energy-consumed industries, such as the manufactures of chemical materials and products, metal smelting and calendering, production and supply of electric power and hot power should be close down or reformed through improving the manufacturing technology, technologic process and production equipment. At the same time, the government should vigorously develop the low-energy-consuming industries such as information services, financial insurance, Internet and tourism, actively promote industrial upgrading and change the situation of heavy industrial structure. Due to the Western and Northeastern regions are mainly dominated by energy-related and heavy industries, the reduction strategies of industrial SO 2 emissions should put more emphasis on the energy structure optimizations and energy efficiency improvement in these two regions. Since these two regions are all at the initial stages of China's economic development, it seems to be unfeasible and unrealistic for them to alter the coal-based energy consumption structure at the present stage. But, reducing industrial SO 2 emissions through improving energy efficiency and the performance of flue gas desulfurization facilities seems to be more feasible and effective, in a short-term period. In a long run, the Western and Northeastern regions with abundant wind and solar energy resources, can gradually develop and utilize these renewable energy sources to replace the coal-dominated energy structure. Besides, related laws and policies should be timely formulated for facilitating the industrial SO 2 emissions reduction in the whole country. For example, extra taxation should be imposed on the industries with high industrial SO 2 emissions in the Eastern and Central regions. While, in the Western and Northeast regions, tax breaks, loan concessions and fiscal subsidy could be granted to the industries using renewable energies or developing advanced technologies for lower industrial SO 2 emissions. Additionally, for the six representative urban agglomerations, industrial SO 2 emissions in Shandong Peninsula was not only higher in grade, but also more obvious in increasing trend. Therefore, more attention should be paid to industrial SO 2 emission reduction of this urban agglomerations.

Conclusions
In order to cope with the industrial SO 2 pollution problems in China, this study explored the spatiotemporal dynamics of industrial SO 2 emissions from 1997 to 2013, and put forward relevant suggestions on industrial SO 2 emission mitigation. On the basis of proving that there was a positive correlation between the DMSP-OLS stable lights and industrial SO 2 emissions, we tried to utilize NSL data to simulate industrial SO 2 emissions. By building linear regression models, we estimated China's industrial SO 2 emissions at 1 km resolution from 1997 to 2013, and evaluated the NSL-based estimation results. The accuracy evaluation results showed that the NSL-based estimation results were acceptable. Eventually, we investigated the spatiotemporal dynamic changes of China's industrial SO 2 emissions from three different scales, and proposed corresponding reduction suggestions for industrial SO 2 emissions. The estimation results apparently exhibited that the distribution of industrial SO 2 emissions differed greatly during the investigation period. Specifically, the high growth type of industrial SO 2 emissions was mainly distributed in the Western region, Central region, and Shandong Peninsula, while the no-obvious-growth type was concentrated in Western region, Beijing-Tianjin-Tangshan and Middle south of Liaoning. And, the high grade was concentrated in Eastern China, Western region, Shanghai-Nanjing-Hangzhou, and Shandong Peninsula, while the low grade mostly located in Western region, Middle south of Liaoning and Beijing-Tianjin-Tangshan. Seeing that the spatio-temporal changes of industrial SO 2 emissions in different regions varied greatly, reduction strategies in Eastern China and Central China should put emphasis on industrial restructuring, while in Western China and Northeastern China, more attentions should be paid to optimize the regional energy structure and improve the energy utilization efficiency.
The findings of our research can not only contribute to comprehensively comprehend the regional differences of spatiotemporal industrial SO 2 emission dynamics at the multiple scales, but also give some scientific references for formulating feasible industrial SO 2 emission reduction policies. But, there are limitations that are worth mentioning. First, the provincial statistical industrial SO 2 emissions which were used to model the distributions of China's industrial SO 2 emissions, may be distorted due to the inconsistent statistical caliber and artificial error. Second, although the linear regression model established in this study based on NSL data, was proved to be an effective means to estimate industrial SO 2 emissions, there indeed exist errors with NSL data as the only index in the simulation models. In order to improve the simulation accuracy, other indicators (such as economic development, industrial structure, land use data, etc.) should also be taken into account when establishing estimation models. In addition, other simulation methods (like panel data analysis, exponential model, and logarithmic model, etc.) should also be tried and compared to determine which model will be better. Third, as for the spatiotemporal dynamics of industrial SO 2 emissions varied greatly from different regions, the driving mechanism of industrial SO 2 emissions in China should be detected in a follow-up study.