Peak of SO 2 Emissions Embodied in International Trade: Patterns, Drivers and Implications

: International trade links countries consuming goods and services to those where products and related SO 2 pollution are produced, thereby affecting national mitigation responsibilities. This study combined accounting and decomposition techniques to investigate the patterns and drivers of SO 2 emissions embodied in international trade from 1995 to 2015 and quantiﬁed the contribution of each country or region on the production and consumption sides. The global embodied emissions increased at an accelerated rate before the global ﬁnancial crisis and peaked at 51.3 Mt in 2008, followed by a ﬂuctuating decline from 2008 to 2015. Spatially, the transfers of SO 2 emissions tended to ﬂow from developed countries to less developed ones, but the trend has weakened after the ﬁnancial crisis. Our decomposition analysis suggests that the energy and production system transitions and the slowdown in international trade jointly accounted for the peak and decline in emissions. Our contribution analysis indicates that developing economies have contributed to decreased emissions due to their recent efforts in production technology upgrading, energy efﬁciency improvement and energy structure optimization. The inﬂuence of developed economies on emissions decreased due to their reduced dependency on imports. Targeted policy methods are provided from the production and consumption perspectives for developing and developed economies, respectively.


Introduction
SO X (including sulfur monoxide and sulfur dioxide) are the world's second largest GHG and have maintained an increasing trend, posing a serious threat to regional ecological, environmental and human health. Sulfur dioxide (SO 2 ) is known to cause serious environmental issues, such as acid rain and fog-haze, and frequent SO 2 exposure can induce cardiovascular and respiratory diseases, causing serious damage to human health [1]. Moreover, due to the growing population and accelerated urbanization in developing countries, as well as the accompanying improvement of living standards, the resulting pollution emissions (e.g., carbon dioxides, SO 2 and nitrogen oxides) from these regions have also grown at an alarming rate [2]. However, in the era of economic globalization, production in a country is not only to meet its own needs but also to meet the demand of the export market [3]. Thus, international trade separates countries consuming products from countries where products and related environmental pollutants are made. Actually, not only the knowledge where the pollution is produced but also the understanding where the products related to the pollution are eventually consumed would affect the effectiveness of pollution reduction efforts [4]. Ignoring the impacts of trade on regional emissions could result in a misunderstanding of the determinants behind regional pollution and misleading regional emission reduction policies [5]. Therefore, it is an urgent issue to investigate the patterns and drivers of the SO 2 emissions embodied in international trade so as to provide

Literature Review
The relationship between trade and the environment has been viewed largely through the lens of comparative advantage [6]. Pollution embodied in trade (PET) is an environmental outcome of the spatial mismatch between production and final consumption in the context of trade liberalization [7]. The interaction between inter-industry trade liberalization and environmental quality has been examined extensively in the literature [8]. A strand of the literature maps the flows of PET, offering strong evidence on the imbalance of PET between developed and developing economies [9]. On the other hand, intraindustry trade rejects that products are produced under identical technical conditions due to greater varieties of intermediate and final goods [10]. Since pollution emissions are subject to the technical conditions of production, the intra-industry trade will trigger the flows of PET naturally. However, Leitão et al. found that agricultural intra-industry trade was negatively correlated with CO 2 emissions in European Union [11], and trade intensity contributed to environmental improvements in Portugal [12], confirming the less pollutant hypothesis. Thus, it plays an essential part in environment protection and trade promotion to properly deal with the relationship between trade and the environment.
In this regard, the multi-regional input-output analysis (MRIO; Leontief [13]) is a popular tool that has been widely applied to estimate emissions embodied in trade [14]. More specifically, the existing studies mainly focus on international transfers between countries [11,15], intranational transfers between provinces [16,17] or industries [18,19]. For example, Feng et al. [20] firstly employed the MRIO method to estimate the carbon emissions embodied in goods and services traded between provincial regions in China and internationally. Moreover, it was revealed that considerable amounts of pollutant emissions were produced for products traded domestically [21] or internationally [22,23]. According to the findings of Arce et al. [24], more than 20% of the global CO 2 emissions were from products that were traded across country borders. Moreover, Oita et al. [25] revealed that the amount of nitrogen emissions embodied in international trade accounted for nearly 25% of the global emissions.
As for the SO 2 emissions embodied in trade, the existing studies can be mainly divided into three categories. The first category focuses on the inter-regional SO 2 transfers within the borders of a country [17,26]. In these studies, a common finding was revealed that about 20-30% of China's SO 2 emissions were embodied in the commodities trade between the provinces and SO 2 pollution tended to transfer from the coastal rich provinces to inner-land undeveloped regions [27,28]. The second category focuses on the embodied SO 2 emissions in the export of a country or region [29,30]. For example, Chuai et al. [30] employed the Eora input-output model to calculate the SO 2 emissions embodied in China's imports and exports and found that the exports generated more SO 2 emissions than the imports. The third category focuses on the SO 2 transfers between countries via international trade [31]. Zhong et al. [32] estimated the sulfur oxides emissions embodied in international trade from 1995 to 2011 and found that the embodied emissions tended to flow from developed economies, such as the USA, to undeveloped ones, such as China. However, to date, there have been no studies on the SO 2 emissions embodied in international trade covering the years beyond 2012, which results in the failure to present the longer-term trends in embodied SO 2 emissions and reveal some interesting findings (e.g., the peak and decline in global embodied SO 2 emissions).
The existing studies have unveiled the invisible transfers of SO 2 emissions embodied in interregional trade. However, it is not sufficient for laying an informational foundation for the guidance of designing fair and reasonable mitigation policies, which requires not only the understanding of the SO 2 flows themselves but also their underlying determinants. In this regard, both index decomposition analysis (IDA) and structural decomposition analysis (SDA) are popular techniques in quantifying the driving factors of a dependent variable. The two methods have their own advantages. More specifically, IDA has a lower requirement for data and has more ease of application, while SDA enables us to distinguish the effects from production and consumption [33].
SDA has been widely applied to identify the determinants behind changes in energy use [34][35][36] and carbon emissions over time [37,38]. For example, using the SDA approach, Feng et al. [39] employed the SDA model to investigate the factors influencing US emissions in the period 1997-2013 and found that the decreases in emissions were largely attributed to the economic recession. Moreover, the driving forces of embodied SO 2 emissions have been identified by using SDA approaches at different scales [17,40]. For example, Liu et al. [26] examined the determinants behind the changes in SO 2 emissions of Chinese provinces and found that decreasing emissions were mainly attributed to end-of-pipe treatment and cleaner production. By using SDA, Yuan et al. [41] investigated the determinants behind the changes in China's industrial SO 2 emissions and found that the changes were dominated by emission intensity and economic scale effects.
On the other hand, IDA has also been widely applied to decomposition analysis on SO 2 emissions or intensity. For example, Yang et al. [42] employed LMDI to analyze the impacts of treatment technology, energy consumption and energy structure on China's industrial SO 2 emissions between 1995 and 2014. Wang et al. [43] identified the factors influencing SO 2 emissions in Jiangsu and its 13 cities from the perspective of a whole process treatment. Applying the whole process decomposition approach, Hang et al. [44] decomposed the industrial SO 2 emissions change in China into six specific driving factors. Zhang [45] used the LMDI approach to decompose the changes in the province-level industrial SO 2 emission intensity into the contributions of energy structure, energy intensity and emission coefficient. Xing et al. [1] employed LMDI to decompose the decline in China's SO 2 emission intensity of thermal power generation into SO 2 treatment, SO 2 emission factor of coal, coal intensity and geographical patterns effects.
Obviously, previous studies have applied the SDA and IDA approaches to examine the determinants behind changes in regional production-based and consumption-based SO 2 emissions, respectively. However, to the best knowledge of the authors, insufficient attention has been paid to the driving factors of bilateral SO 2 transfers embodied in trade. Moreover, the existing studies focusing on factors from either the production or consumption side failed to take into account the impacts from both sides, which prevents the provision of sufficient supporting information for policy makers to make decisions and derive differentiated mitigation policies for different countries from production and consumption perspectives simultaneously. To this end, this study combined the IDA and SDA techniques to conduct an integrated decomposition of the SO 2 emissions embodied in international trade from both the production and consumption perspectives and further quantify the contributions of each region to the global SO 2 transfers from the production and consumption sides, respectively.
The main innovations of this study are as follows: first, we extended the research period of the SO 2 emissions embodied in international trade right up to 2015 with the latest released Eora data, which allows us to observe the longer-term trends in the embodied SO 2 emissions and reveal some interesting findings (e.g., the peak and decline in global embodied SO 2 emissions). Second, as few studies have decomposed the SO 2 emissions embodied in international trade from both the production and consumption perspectives, we combined the IDA and SDA techniques to fill this gap, which could identify the driving factors of global SO 2 transfers and quantify the contributions of each region on both the production and consumption sides. The rest of this paper is structured as follows. Section 3 introduces the methods and materials, including MRIO, SDA, IDA and data sources. Section 4 presents the accounting, decomposition and contribution results of the emissions embodied in international trade between 1995 and 2015. Section 5 discusses the patterns and drivers of the trends in emissions. Section 6 concludes the study and provides corresponding policy implications.

Materials and Methods
Following the research route of accounting-decomposition-contribution analysis, this study combined MRIO, IDA and SDA techniques to investigate the patterns and drivers of SO 2 emissions embodied in international trade and quantified the contribution of each country or region on production and consumption sides. This section will introduce the MRIO, IDA and SDA methods and show how to use them to calculate embodied SO 2 emissions and identify the driving factors. The research framework is shown in Figure 1. patterns and drivers of the trends in emissions. Section 6 concludes the study and provides corresponding policy implications.

Materials and Methods
Following the research route of accounting-decomposition-contribution analysis, this study combined MRIO, IDA and SDA techniques to investigate the patterns and drivers of SO2 emissions embodied in international trade and quantified the contribution of each country or region on production and consumption sides. This section will introduce the MRIO, IDA and SDA methods and show how to use them to calculate embodied SO2 emissions and identify the driving factors. The research framework is shown in Figure 1.

Embodied Emissions in Trade
The environmentally extended multiregional input-output model was used to quantify the SO2 emissions embodied in international trade [13]. The bilateral trade in MRIO consists of intermediate input and final use, which builds the complete production and consumption linkages between regions. Consumption-based embodied emissions are accounted based on final use, so one-unit emission may go through many regions before it eventually arrives at the final consumer.
Supposing an MRIO table with m regions and n sectors, the total output of sector i in region r ( r i x ) can be calculated by the following formula on the basis of the horizontal accounting balance:

Embodied Emissions in Trade
The environmentally extended multiregional input-output model was used to quantify the SO 2 emissions embodied in international trade [13]. The bilateral trade in MRIO consists of intermediate input and final use, which builds the complete production and consumption linkages between regions. Consumption-based embodied emissions are accounted based on final use, so one-unit emission may go through many regions before it eventually arrives at the final consumer.
Supposing an MRIO table with m regions and n sectors, the total output of sector i in region r (x r i ) can be calculated by the following formula on the basis of the horizontal accounting balance: where z rs ij denotes the intermediate input from sector i in region r to sector j in region s (r, s = 1, 2, · · · , m; i, j = 1, 2, · · · , n); y rs i stands for the final use of region s for commodities in sector i from region r.
Defining a rs ij = z rs ij /x s j as the MRIO technical coefficient [46], Equation (1) can be transferred into: Using matrix notations, Equation (2) can be transformed into the following matrix form: where A denotes the direct consumption matrix. Solving for X yields: where L = (I − A) −1 denotes the Leontief inverse matrix, which indicates the total output required to satisfy one unit of final use. The SO 2 extended MRIO model is constructed by adding the row vector of SO 2 emission coefficient, which describes the direct emissions per unit of total output. The total amount of embodied emissions in final demand of region s can be computed by: where D r = d r i n×1 = k r i /x r i n×1 denotes the column vector of SO 2 emission coefficient in region r; k r i denotes the direct SO 2 emissions of sector i in region r. The embodied emissions in region r driven by final demand in region s is: where ek r Y s denotes the flow of SO 2 emissions from region s to region r; L rt is the submatrix in the Leontief inverse matrix (t = 1, 2, · · · , m).

Structural Decomposition Analysis
The transfer of SO 2 emissions from region s to region r embodied in international trade is driven by both production-side and consumption-side factors. The production technology factors at the production side and the final demand factors at the consumption side jointly affect the SO 2 flow between them. SDA based on IO model is a popular technique in measuring the contributions of different forces to changes in energy use and carbon emissions over time.
Expanding Equation (6) in matrix form yields: where L r = L r1 L r2 · · · L rm denotes the input coefficient matrix of region r to all the regions, which is also defined as production structure of region r; Y s denotes the column vector of final demand in region s. Equation (7) shows that the inter-regional SO 2 transfer is related to three factors (the SO 2 emission intensity, the production structure, and the consumption-side final demand). Among them, the first two factors are the driving factors in the production side, and the third factor is the driving factor in the consumption side, which together constitute the driving factors of SO 2 transfer. To distinguish the contributions of different components in final demand, Y is further decomposed into five components as follows [3]: where the • symbol denotes the pointwise multiplication of the corresponding elements of two matrices. Y d is a four-dimensional column vector reflecting the type patterns of the final demand. Its elements α i (i = 1, 2, 3, 4) denote the proportion of household consumption, government investment, fixed assets formation and changes in inventories in the total final demand. Y p is a matrix composed of four column vectors corresponding to the product patterns of four kinds of final demand. Each column vector is stacked with m same column vectors β(β 1 = β 2 = · · · = β m ), and its element is the share of the final products provided by each department. Y g is a matrix composed of four column vectors corresponding to the geographic patterns of four kinds of final demand. Each column vector is stacked with m different column vectors η(η 1 = η 2 = · · · = η m ), and its element η r (j) denotes the share of the final demand for j product provided by region r, reflecting the regional source distribution of various final products. Y v denotes per capita consumption volume. p is population. Therefore, Equation (7) can be transformed to: Equation (9) provides the computation of SO 2 transfer from region s to region r, which is related to seven factors that include emission intensity, production structure, consumption patterns (i.e., product patterns, geographic patterns, type patterns), consumption volume and population. A total difference of Equation (9) generates Equation (10): where ∆ is the difference operator. Equation (10) converts seven multiplicative terms in Equation (9) into seven additive terms and fully accounts for the changes in embodied SO 2 emissions in trade. Each additive term in Equation (10) denotes the contribution of a factor assuming all other factors are constant. Due to a unique solution for the decomposition in Equation (10) not being available [47], we followed the methods of previous studies and used the average of the so-called polar decompositions as an approximation of the average of all decompositions [48]. Solving Equation (10) with the polar decomposition method yields: where ∆F r Y s denotes the changes in emission transfers between two time points; ∆D, ∆L, ∆S p , ∆S g , ∆S d , ∆V and ∆P denote changes in emission intensity, production structure, product patterns, geographic patterns, type patterns, consumption volume and population, respectively. The first two effects are related to the production side, while the last five effects are associated with the consumption side.

Index Decomposition Analysis
Equation (11) has decomposed changes in embodied emissions into seven factors, of which five factors are related to consumption side. By contrast, production-side factors seem not to be investigated enough, especially for emission intensity effect ∆D, which could be decomposed further by using index decomposition analysis. Following the spirit of Wang et al. [43], the global SO 2 emission intensity (SEI) is typically expressed in Equation (12): where SE denotes global SO 2 emissions; X denotes the total output; SE i denotes the SO 2 emissions of sector i; E i denotes the energy consumption of sector i; X i denotes the output of sector i. The global SEI for time t − 1 and t can be described as Equations (13) and (14), respectively. where , respectively, denote SO 2 emissions of unit energy consumption for sector i at time t and t − 1, which could be used to represent energy structure under the assumption that the SO 2 emission factor for each kind of energy source is constant.
, respectively, denote energy consumption of unit output (i.e., energy intensity) for sector i at time t and t − 1, indicating energy efficiency level. IS t i = X t i /X t and IS t−1 i = X t−1 i /X t−1 , respectively, denote the share of output of sector i at time t and t − 1, which could be used to represent industrial structure.
Suppose that the global SEI varies from time t − 1 to t (i.e., V t−1,t tot = SEI t − SEI t−1 ). Such a change can be expressed in the following additive form [49] as Equation (15), which indicates that global SEI change is related to three factors: energy structure, energy intensity and industrial structure.
, respectively, measure the effects of energy structure, energy intensity and industrial structure over the period [t − 1, t]; ω i denotes the weight of sector i, and L(a, b) = (b − a)/(ln b − ln a) is the logarithmic mean function.
Equation (15) describes the single-period decomposition results of SEI change. In the case of multi-period decomposition, the accumulative effect V 0,T tot from time 0 to T can be calculated by Equation (16): where V 0,T ES , V 0,T EI and V 0,T IS are the corresponding cumulative sum of single-period decomposed indexes.

Data Sources
Data used in this study include monetary MRIO tables and SO 2 emissions data of countries or regions. The related data were gathered from the Eora global supply chain database [50], which provides a time series of high-resolution IO tables with matching environmental and social satellite accounts for 189 countries or regions (https://worldmrio. com/ (accessed on 12 May 2018)). Eora has been applied in some excellent works, such as Peters et al. [51], Yang et al. [52] and Lin et al. [53], so the data credibility can be guaranteed. In order to eliminate the impact of price change, MRIO tables in 1996 to 2015 were deflated to 1995 constant price.

Results
This section will present the accounting, decomposition and contribution results to reveal the patterns and drivers of change in the embodied SO 2 emissions and compare the contributions of different economies on the production and consumption sides.

Patterns of Emissions Embodied in International Trade
The  Figure  2, orange bar), presenting significant temporal variation characteristics that trended upward between 1995 and 2008 but downward between 2008 and 2015. The finding that the SO2 emissions embodied in international trade decreased after 2008 has been revealed in existing literature [32]. Nevertheless, our analysis of more recent data, up to 2015, permits the more confident conclusion that the global embodied emissions peaked in 2008.  . This tendency has also been revealed by other studies regarding carbon emissions [22,55], pollutant emissions [3] and metal use [56]. As shown in Figure 2 Spatially, Figure 4, plotted by ArcGIS 10.2.2, maps the inter-regional SO 2 transfers via international trade and resulting SO 2 trade balance (i.e., the difference between SO 2 inflows and outflows) in different regions. The shading of regions in Figure 4 denotes the magnitude of net inflows (red) or net outflows (blue) embodied in international trade in 1995, 2008 and 2015. Arrows in the figure denote the ten largest inter-regional SO 2 flows. The number and arrow thickness indicate emissions embodied in flows.
emissions growth in the same period. Actually, China's accession into the WTO stimulated the rapid growth of global trade and exacerbated the geospatial separation between the regions consuming commodities and regions where products and related SO2 pollution are produced, thereby resulting in the dramatic growth in embodied emissions. From a sectoral point of view, a large share of embodied SO2 emissions growth in this period was attributed to the machinery and equipment manufacturing industry (Figure 3, grey and dark orange shading). For instance, emissions embodied in the final use of electrical and machinery products grew significantly, contributing 2.8 Mt SO2 to the global embodied emissions between 2002 and 2008, which accounted for 25.2% of the total embodied emissions growth over the same period (Figure 3, dark orange shading). Spatially, Figure 4, plotted by ArcGIS 10.2.2, maps the inter-regional SO2 transfers via international trade and resulting SO2 trade balance (i.e., the difference between SO2  From the perspective of SO2 trade balance, the largest net SO2 exporters primarily were developed economies, such as the USA, Japan, Germany, UK, France and Chinese Hong Kong, while developing economies (e.g., Chinese mainland, Kazakhstan, India, From the perspective of SO 2 trade balance, the largest net SO 2 exporters primarily were developed economies, such as the USA, Japan, Germany, UK, France and Chinese Hong Kong, while developing economies (e.g., Chinese mainland, Kazakhstan, India, Russia) and resource-abundant economies (e.g., Australia) were the highest net SO 2 importers. Especially, the USA and Chinese mainland served as the largest source and sink of global SO 2 transfers, respectively. The net SO 2 inflows of the USA in 1995, 2008 and 2015 were -4.5, -9.3 and -6.6 Mt, respectively, while those of the Chinese mainland were 4.8, 9.8 and 4.9 Mt, respectively.
Looking at the inter-regional SO 2 flows, the largest one was the flow from the USA to Chinese mainland, which transferred emissions of 1190.4, 3342.0 and 2137.1 Kt in 1995, 2008 and 2015, respectively. Clearly, it also increased firstly and then decreased, corresponding to the inversed "U" trend of the total embodied emissions. The following notable flows included Japan→Chinese mainland, USA→Canada, Chinese Hong Kong→ Chinese mainland. Generally, the global embodied SO 2 emissions were mainly transferred from west to east and from developing to developed economies (Figure 4). Figure 5a indicates the accumulative SO 2 flows between developed and developing economies from 1995 to 2015. It is found that developed economies transferred 415.2 Mt SO 2 to developing economies via international trade between 1995 and 2015, accounting for 44.4% of the total embodied SO 2 emissions (Figure 5a, purple belt). In contrast, developing economies only transferred about one-in-five (85.4 Mt) back to developed economies (Figure 5a, yellow belt). As a result, the pollution emissions grew rapidly in some developing economies but stabilized or even decreased in many developed economies in recent years. Actually, due to the pollution intensive energy mix, less efficient production technologies as well as an inferior position in the global value chain, developing economies contributed a large amount of energy-and emission-intensive final and intermediate products to support the production and consumption of finished goods in developed economies [57]. Additionally, notable variations in the flow patterns between developed and developing economies were observed over this period. As shown in Figure 5b, the share of developed → developing economies decreased to 38.9%, while that of developing → developed economies increased to 12.3%. Notably, the share of developing → developing economies witnessed considerable growth from 18.3% in 1995 to 26.8% in 2015 (Figure 5b, orange shading), which may be attributed to the rapid rise of south-south trade since the global financial crisis [48].  (Figure 5a, yellow belt). As a result, the pollution emissions grew rapidly in some developing economies but stabilized or even decreased in many developed economies in recent years. Actually, due to the pollution intensive energy mix, less efficient production technologies as well as an inferior position in the global value chain, developing economies contributed a large amount of energy-and emission-intensive final and intermediate products to support the production and consumption of finished goods in developed economies [57]. Additionally, notable variations in the flow patterns between developed and developing economies were observed over this period. As shown in Figure 5b, the share of developed → developing economies decreased to 38.9%, while that of developing → developed economies increased to 12.3%. Notably, the share of developing → developing economies witnessed considerable growth from 18.3% in 1995 to 26.8% in 2015 (Figure 5b, orange shading), which may be attributed to the rapid rise of south-south trade since the global financial crisis [48].

Determinants of Changes in Emissions Embodied in International Trade
Equation (11) was used to decompose the changes in embodied emissions from 1995 to 2015. As shown in Figure 6, global embodied SO2 emissions increased by 15.4% in this period (Figure 6, black curve). The increase was dominated by the expansion in consump-

Determinants of Changes in Emissions Embodied in International Trade
Equation (11) was used to decompose the changes in embodied emissions from 1995 to 2015. As shown in Figure 6, global embodied SO 2 emissions increased by 15.4% in this period (Figure 6, black curve). The increase was dominated by the expansion in consumption volume, which-in the absence of other factors-would have caused embodied emissions to increase by 101.2% over the two decades ( Figure 6, red curve). The next most important driver during this time frame was production structure, which contributed to a 45.3% increase during this period (Figure 6,   In order to reveal more production-related factors, Equation (15) was used to decompose the changes in SO2 emission intensity from 1995 to 2015. As shown in Figure 7, global emission intensity decreased by 62.7% in this period (Figure 7, black curve). Our analysis shows that the main factor behind this decrease was the decrease in energy intensity, which corresponds to a contribution of a 43.7% decrease in emission intensity over this period (Figure 7, green curve). The next most important factor was energy structure, which contributed a 24.8% decrease in emission intensity (Figure 7, cyan curve). However, the effect of industrial structure exerted an upward influence of 5.8% (Figure 7, dark yellow curve). In order to reveal more production-related factors, Equation (15) was used to decompose the changes in SO 2 emission intensity from 1995 to 2015. As shown in Figure 7, global emission intensity decreased by 62.7% in this period (Figure 7, black curve). Our analysis shows that the main factor behind this decrease was the decrease in energy intensity, which corresponds to a contribution of a 43.7% decrease in emission intensity over this period (Figure 7, green curve). The next most important factor was energy structure, which contributed a 24.8% decrease in emission intensity (Figure 7, cyan curve). However, the effect of industrial structure exerted an upward influence of 5.8% (Figure 7, dark yellow curve). emission intensity decreased by 62.7% in this period (Figure 7, black curve). Our analysis shows that the main factor behind this decrease was the decrease in energy intensity, which corresponds to a contribution of a 43.7% decrease in emission intensity over this period (Figure 7, green curve). The next most important factor was energy structure, which contributed a 24.8% decrease in emission intensity (Figure 7, cyan curve). However, the effect of industrial structure exerted an upward influence of 5.8% (Figure 7, dark yellow curve).  In light of the global embodied emissions peaking in 2008, we divided the whole period into two subperiods, 1995-2008 and 2008-2015, so as to investigate the drivers behind the peaking emissions.
Growing emissions from 1995 to 2008. In the period from 1995 to 2008, the global embodied SO 2 emissions increased by 29.9% (Figure 8, black bar). The increase was dominated by the expansion in consumption volume (83.6%; Figure 8, red bar), followed by production structure (61.3%, Figure 8, red bar). The effects of consumption patterns-related factors, such as geographic patterns, product patterns and type patterns, contributed 16.5%, 4.3% and 0.5% increases to emissions, respectively (Figure 8, yellow, blue and dark blue bars). Other promoting factors, such as population and industrial structure, exerted modest upward influences of 11.9% and 11.6% on emissions, respectively (Figure 8, purple and dark yellow bars). However, the upward influence of the above seven factors was largely offset by the downward influence of energy intensity and energy structure, which contributed 106.5% and 53.2% decreases to emissions, respectively (Figure 8, green and cyan bars). Growing emissions from 1995 to 2008. In the period from 1995 to 2008, the global embodied SO2 emissions increased by 29.9% (Figure 8, black bar). The increase was dominated by the expansion in consumption volume (83.6%; Figure 8, red bar), followed by production structure (61.3%, Figure 8, red bar). The effects of consumption patternsrelated factors, such as geographic patterns, product patterns and type patterns, contributed 16.5%, 4.3% and 0.5% increases to emissions, respectively (Figure 8, yellow, blue and dark blue bars). Other promoting factors, such as population and industrial structure, exerted modest upward influences of 11.9% and 11.6% on emissions, respectively (Figure 8, purple and dark yellow bars). However, the upward influence of the above seven factors was largely offset by the downward influence of energy intensity and energy structure, which contributed 106.5% and 53.2% decreases to emissions, respectively ( Figure 8, green and cyan bars).  (Figure 8, black bar). During this time period, the effects of consumption volume, population, industrial structure, product patterns and type patterns still acted as the drivers increasing embodied emissions by 13.6%, 5.2%, 3.4%, 1.3% and 0.3%, respectively (Figure 8, red, purple, dark yellow, blue  (Figure 8, black bar). During this time period, the effects of consumption volume, population, industrial structure, product patterns and type patterns still acted as the drivers increasing embodied emissions by 13.6%, 5.2%, 3.4%, 1.3% and 0.3%, respectively (Figure 8, red, purple, dark yellow, blue and dark blue bars). While the consumption volume still caused an upward influence on emissions, the magnitude underwent a sharp decline from 83.6% to 13.6%. As a result, the upward influence of the five factors was overwhelmed by the downward influence of production structure (-12.3%), energy structure (-11.5%), energy intensity (-10.6%) and geographic patterns (-0.5%; Figure 8, orange, cyan, green and yellow bars). Notably, shifts in production structure switched from a significant promoting factor between 1995 and 2008 (61.3%) to the dominant inhabiting factor between 2008 and 2015 (-12.3%; Figure 8, orange bars). This may be because the volume and type of intermediate products consumed by various industry sectors at home and abroad has evolved and become more efficient [39]. Taken together, production-side effects, such as energy and production system transitions and the slowdown in international trade, were primarily responsible for the peak in embodied emissions in international trade.

Production-and Consumption-Side Contributions of Regions
We further calculated the contribution of each region to the changes in global embodied emissions for the periods 1995-2008 and 2008-2015 on the production and consumption sides. As shown in Figure 9, the horizontal and vertical axes denote the contribution rates on production and consumption sides, respectively. The red line in the figure is a straight line with a slope of −1 passing through the origin. Figure 9a,b are the contribution results for period 1995-2008, and Figure 9b is the enlarged version of the inner part of the green ellipse in Figure 9a so as to present the distribution of each point more clearly. Similarly, Figure 9a,b correspond to the period 2008-2015. Blue triangle points and orange round points stand for developed and developing economies, respectively. Due to the limited space, only 30 major economies with the largest contributions were presented in Figure 9.
As can be seen from Figure 9, all points fell in the second quadrant, indicating that all the 30 main economies, in both periods 1995-2008 and 2008-2015, contributed negatively to the change in embodied emissions on the production side, while they made a positive contribution to emissions on the consumption side. However, the relative size of production-and consumption-side contributions varied greatly in different regions, resulting in significant differences in the total contribution of regions. The point below the red line indicates that the total contribution of the region was negative; that is, the upward influence on the consumption side was offset by the downward influence on the production side and vice versa. In general, developed economies are mainly located above the red line ( Figure 9, blue triangle points), while developing economies are below the red line ( Figure 9, orange round points).
In the period 1995-2008, as shown in Figure 9a,b, Kazakhstan had the largest negative contribution (−2.3%), which was attributable to its great efforts toward a clean energy structure through the use of policy and technology [58], thereby causing a significant downward influence on the production side (−3.0%). On the contrary, the USA was found to have the largest positive contribution (19.8%), which was mainly due to its huge upward influence on the consumption side (24.7%). Notably, the Chinese mainland had the second largest production-side contribution (8.2%) resulting from its huge population base and corresponding final demand growth [59]. However, the significant upward influence was largely offset by its huge downward influence on the production side (−7.6%), thereby exerting a minor upward influence on global emissions.
In the period 2008-2015, as shown in Figure 9a,b, the Chinese mainland became the largest contributor to reduce global embodied SO 2 emissions (−5.9%). This may be attributed to its recent great efforts in energy-saving and emission reduction [60], which largely removed SO 2 emissions from the production process, thereby amounting to the remarkable negative contribution on the production side (−11.6%). Notably, the consumption-side positive contribution of the USA also met a sharp decline from 24.7% between 1995 and 2008 to 1.1% between 2008 and 2015. A similar condition was also observed for the UK, Germany and Chinese Hong Kong. As a result, some developed economies, such as the UK, Australia, Canada and Singapore, fell below the red line and the whole developed economies switched to exert a downward influence on global emissions. This is because developed economies became less dependent on imports after the financial crisis, hence slowing down the emissions embodied in international trade on the consumption side [61].

Production-and Consumption-Side Contributions of Regions
We further calculated the contribution of each region to the changes in global embodied emissions for the periods 1995-2008 and 2008-2015 on the production and consumption sides. As shown in Figure 9, the horizontal and vertical axes denote the contribution rates on production and consumption sides, respectively. The red line in the figure is a straight line with a slope of -1 passing through the origin. Figure 9a,b are the contribution results for period 1995-2008, and Figure 9b is the enlarged version of the inner part of the green ellipse in Figure 9a so as to present the distribution of each point more clearly. Similarly, Figure 9a,b correspond to the period 2008-2015. Blue triangle points and orange round points stand for developed and developing economies, respectively. Due to the limited space, only 30 major economies with the largest contributions were presented in Figure 9. As can be seen from Figure 9, all points fell in the second quadrant, indicating that all the 30 main economies, in both periods 1995-2008 and 2008-2015, contributed negatively to the change in embodied emissions on the production side, while they made a positive contribution to emissions on the consumption side. However, the relative size of production-and consumption-side contributions varied greatly in different regions, resulting in significant differences in the total contribution of regions. The point below the red line indicates that the total contribution of the region was negative; that is, the upward influence on the consumption side was offset by the downward influence on the production side and vice versa. In general, developed economies are mainly located above the

Discussion
The SO 2 emissions embodied in international trade experienced an inverted U-shape changing trend. They grew dramatically before the global financial crisis, which was mainly attributable to the expansion in the consumption volume and shifts in the production structure, while the dramatic shift after 2008 was mainly due to production-side factors (i.e., shifts in production structure, decline in energy intensity and shifts in energy structure).
Developing economies, such as the Chinese mainland and Kazakhstan, contributed a significant decrease to global emissions. This may be attributed to their great efforts toward clean energy structures through the use of policy and technology [58]. Additionally, the Chinese government has made great efforts to reduce the SO 2 emissions from the production side through implementing a mandatory emissions control system [62][63][64].
With the implementation of measures that include shutting down small thermal power units and installing desulfurization facilities, considerable emissions have been removed since 2008 [1]. Meanwhile, the volume and type of intermediate products consumed by various industry sectors at home and abroad has evolved and become more efficient [39], thereby bringing about the significant downward influence of the production structure between 2008 and 2015. The deglobalization has resulted in a significant decrease in the upward influence of the consumption volume on emissions between 1995 and 2008 and 2008 and 2015, which is also an important reason for the peak and decline in emissions.
The transfers of SO 2 emissions tended to flow from developed economies, such as the USA, to less developed ones, such as the Chinese mainland. Actually, developed countries imported high energy-and emission-intensive products from developing countries to meet their final demands instead of producing them by themselves, thereby resulting in large SO 2 transfers from developed to developing economies. However, there has been some decrease in the amount after the financial crisis because developed economies became less dependent on imports [61]. Moreover, the decline in their final demands for products abroad reduced their upward influence on the consumption side, and some, such as the UK, Australia, Canada and Singapore, even switched to contribute a decrease in the global emissions. On the contrary, the SO 2 flowing between developing economies witnessed a significant increase, which may be attributed to the rapid rise of south-south trade since the global financial crisis [48].

Conclusions
This study investigated the patterns and drivers of the trends in the SO 2 emissions embodied in international trade from 1995 to 2015. Some interesting findings have been revealed: First, the SO 2 emissions embodied in international trade accounted for about 24%-30% of the global emissions between 1995 and 2015. The global embodied emissions increased at an accelerated rate before the global financial crisis and peaked at 51.3 Mt in 2008, followed by a fluctuating decline from 2008 to 2015. Spatially, the transfers of SO 2 emissions tended to flow from developed countries to developing countries, but the trend has weakened after the financial crisis.
Second, the increase before 2008 was dominated by the expansion in the consumption volume, especially for the products of the machinery and equipment manufacturing industry, but its upward influence significantly decreased between 2008 and 2015 due to deglobalization. Even more, shifts in the production structure switched from a significant promoting factor between 1995 and 2008 to the dominant inhabiting factor between 2008 and 2015.
Third, more developing economies, including the Chinese mainland, contributed negatively to the global embodied emissions growth, while more developed economies, such as the USA, were found to contribute an increase to emissions, but the upward influence has decreased significantly after the crisis.
Fourth, the negative contribution of developing economies on the production side offsets their positive contribution on the consumption side, while developed economies are just the opposite. Nevertheless, the consumption-side contribution of developed economies has decreased after the crisis, thereby resulting in the weakening upward influence on emissions. Taken together, energy and production technology upgrading and energy structure and declining final demands for products abroad in developed countries both accounted for the peak and decline in the global embodied SO 2 emissions.

Policy Implications
According to the above findings, several policy implications can be summarized as follows.
Since the SO 2 emissions embodied in international trade accounted for about a quarter of the global emissions, the adverse impact of international trade on the world's eco-environment cannot be ignored. Meanwhile, in the context of large amounts of SO 2 emissions transferred from developed regions to developing regions, the allocation of pollutant emission reduction responsibilities based on production-based accounting is obviously lacking justice for developing countries. Therefore, the pollution mitigation tasks should be reasonably distributed among the countries to reduce SO 2 leakage in international trade, and attention should be paid to the impact of international trade to prevent developed countries from outsourcing more pollution to others. Developed and developing countries need to be treated differently when it comes to formulating a global pollutant emission reduction policy. Developed countries such as the USA have removed the production of resources abroad or imported substitutes to circumvent their own emission reduction responsibilities. To this end, the international community should urge developed countries to take on more reduction responsibilities and push technology transfer or offer financial aid to improve other countries' technology for dealing with pollutants.
Since developing countries' production-side factors, such as energy and production system transitions, significantly contributed to decreasing emissions, policymakers in developing countries should continue to maintain their current policies on energy production, industrial technology and environmental regulation. Additionally, environmental subsidies would help fund firms to upgrade their desulfurization technologies, eliminate outdated production technologies and apply clean energy resources. Greater international cooperation and support are strongly recommended to promote technology transfer, and effective supervision should also occur regarding the implementation and operation of these imported technologies. On the other hand, as developed countries' consumption-side factors, such as consumption volume and patterns, contributed to increasing the embodied emissions throughout the whole period, developed countries should correctly guide the consumption concept of residents to help curb luxury consumption and promote green consumption habits.