Trends and Consumption Structures of China’s Blue and Grey Water Footprint

: Water footprint has become a common method to study the water resources utilization in recent years. By using input–output analysis and dilution theory, the internal water footprint, blue water footprint and grey water footprint of China from 2002 to 2012 were estimated, and the consumption structure of water footprint and virtual water trade were analyzed. The results show: (1) From 2002 to 2012, the average annual internal water footprint was 3.83 trillion m 3 in China, of which the blue water footprint was 0.25 trillion m 3 , and the grey water footprint was 3.58 trillion m 3 (with Grade III water standard accounting); both the internal water footprint and grey water footprint experienced decreasing trends from 2002 to 2012, except for a dramatic increase in 2010; (2) Average annual virtual blue water footprint was the greatest in agriculture (39.2%), while tertiary industry (27.5%) and food and tobacco processing (23.7%) were the top two highest for average annual virtual grey water footprint; (3) Virtual blue water footprint in most sectors showed increasing trends due to the increase of ﬁnal demand, while virtual grey water footprint in most sectors showed decreasing trends due to the decreases of total return water coefﬁcients and conversion coefﬁcients of virtual grey water footprint; (4) For water resources, China was self-reliant: the water used for producing the products and services to meet domestic consumption was taken domestically; meanwhile, China exported virtual water to other countries, which aggravated the water stress in China.


Introduction
Water footprint refers to the water consumption for producing products and services for a certain population (individual, city or country) under certain material living standards [1]. This part of water resources includes not only the actual water used in daily life, water used for industrial and agricultural goods (services) production, and municipal water, but also the water used for processing sewage and waste water that is produced during the life and production activities. The concept of water footprint is actually a combination of physical and virtual water consumption, a combination of the consumption of "blue water" (surface water and groundwater) and "green water" (precipitation which does not form surface water, groundwater and is reserved in soil) [2], and a combination of changes in water quantity and quality. It shows the impact of human consumption on water resources from a broader view.
Raised by A. Y. Hoekstra in 2002, water footprint has been estimated at varies scales [3][4][5][6]. At a national level, Hoekstra and Chapagain carried out a systematic estimation of water footprint in 2007-2008 [7]. A. Y. Hoekstra et al. centralized years of research to compile the "Water Footprint Assessment Manual" [8], which was a comprehensive summary for water footprint analysis and could provide samples for water footprint research. Methods for calculating water footprint Figure 1. Calculation flow chart. WF: Water footprint; NVBWI/NVGWI: Net virtual blue/grey water import; TNVBWI/TNVGWI: Total net virtual blue/grey water import; k: Conversion coefficient of grey water footprint; V c /V r : Water consumption/Return water; r c /r r : Total water consumption coefficient/Total return water coefficient; P: Expenditure. Table   In the input-output tables (IO table), the row model is shown as:

Input-Output
indicates a certain sector in rows and j indicates a certain sector in columns; x ij is the input from sector i to sector j; Y i is the final demand in sector i; and X i is the total output in sector i. a ij = x ij /X i is defined as the direct consumption coefficient, and A = a ij n×n is the direct consumption coefficient matrix correspondingly; therefore, the row model can be expressed in matrix form as: AX + Y = X, where X is the column vector of total output, Y is the column vector of final demand. Therefore, we can get X = (I − A) −1 Y, and then the complete demand matrix is defined as B = (I − A) −1 , indicating the increase of output to meet one monetary unit increase of final use.
In this study, the input-output tables were aggregated into 26 sectors in order to match the water consumption and return water data. The 26 sectors are shown in Table 1.  21 Scrap waste and other manufacturing products 9 Timber processing and furniture manufacturing 22 Electronic and heating power production and supply 10 Papermaking and cultural articles 23 Gas production and supply 11 Petroleum processing and coking 24 Water production and supply 12 Chemicals 25 Construction 13 Non-metal mineral products 26 Tertiary industry

Water Use, Water Consumption and Return Water
Water use refers to the amount of fresh water taken by water users. In the process of water transfer and water use, through evapotranspiration, soil absorption, product consumption, people and livestock drinking and other forms water consumption, some water cannot return to surface water or aquifer water, which is called water consumption. In the concept of water footprint, the amount of water should exclude the part of water returning to surface water and groundwater. Therefore, water consumption data should be used instead of water use data. The relationship among water use, water consumption and return water is as follows: where V u is the row vector of water use and V u,j is the water use in sector j (m 3 ); V c is the row vector of water consumption and V c,j is the water consumption in sector j (m 3 ); V r,j is the row vector of return water and V r,j is the return water in sector j (m 3 ).

Water Consumption Coefficient and Return Water Coefficient
Referring to the concept of direct water use coefficient and total water use coefficient [6], we proposed direct water consumption coefficient, total water consumption coefficient, direct return water coefficient and total return water coefficient.
Direct water consumption coefficient refers to the water resources consumed in natural forms to increase one monetary unit output: where f c is the row vector of direct water consumption coefficient and f c,j is the direct water consumption coefficient in sector j (m 3 /10 4 Yuan); X j is the total output in sector j (10 4 Yuan). Direct return water coefficient refers to the amount of sewage and waste water discharged directly to produce one monetary unit output: where f r,j is the row vector of direct return water coefficient and f r,j is the direct return water coefficient in sector j (m 3 /10 4 Yuan). The total water consumption coefficient refers to the amount of water consumption in the whole production chain to increase one monetary output, consisting of direct water consumption coefficient ( f c ) and indirect water consumption coefficient ( f c,indirect ): where r c is the row vector of total water consumption coefficient (m 3 /10 4 Yuan). Similarly, the total return water coefficient refers to the amount of total return water in the whole production chain to increase one monetary unit output, consisting of direct return water coefficient ( f r ) and indirect return water consumption coefficient ( f r,indirect ): where r r is the row vector of total return water coefficient (m 3 /10 4 Yuan).

Internal Water Footprint, Blue Water Footprint and Grey Water Footprint
Water footprint includes green water footprint, blue water footprint (WF blue ) and grey water footprint (WF grey ). Green water footprint and blue water footprint refer to the water consumption of green water and blue water during the production, respectively; grey water is the volume of water needed to dilute the pollutants [22,23]. Therefore, green water footprint and blue water footprint are related to the quantity of water resources, while grey water is related to the water quality.
This study focused on blue water footprint and grey water, but excluded green water footprint, because input-output analysis is sector-based, but green water footprint is always calculated based on the evapotranspiration of a specific crop; therefore, it is hard to determine which crops should be included in the calculation of agriculture (sector 1).
Internal water footprint (WF internal ) denotes the domestic water needed for producing services and products consumed by domestic residents and consists of WF blue and WF grey in this study.
Blue water footprint consists of actual blue water footprint (WF blue,actual , million m 3 ) and virtual blue water footprint (WF blue,virtual , million m 3 ), where WF blue,actual refers to the living water consumption, while WF blue,virtual refers to the water used for producing the goods and services for consumption: where WF blue,virtual is the row vector of virtual blue water footprint and WF blue,virtual,j is the virtual blue water footprint in sector j; P is the row vector of expenditure and P j is the expenditure in sector j (10 4 Yuan). Similar with the blue water footprint, return water also includes actual return water (WF return,actual , million m 3 ) and virtual return water (WF return,virtual , million m 3 ), which are caused by living return water and the return water generated in producing the goods and services for domestic consumption, respectively: where WF return,virtual is the row vector of virtual return water footprint and WF return,virtual is the virtual return water footprint in sector j.
In calculating the grey water footprint, we need to consider both the volume and pollutant concentration of the waste water, and the water quality standard and the concentration of background pollutants. Since different pollutants can be diluted in the meantime, the pollutants chosen for calculation should be those that can cause the maximum grey water footprint. In this study, two biggest pollutants, COD (Chemical Oxygen Demand) and NH 3 -N (Ammonia Nitrogen), were chosen as the particular pollutants to analyze the grey water footprint.
Define k as the row vector of conversion coefficient of grey water footprint, where k j for second and tertiary industrial sector j and k actual for living return water. The grey water footprint can be calculated as: WF grey,virtual = [WF grey,virtual,j ], WF grey,virtual,j = k j × WF return,virtual,j , WF grey,actual = k actual × WF return,actual , where WF grey,virtual is the row vector of virtual grey water footprint and WF grey,virtual,j refers to the virtual grey water footprint in sector j (million m 3 ); WF grey,actual is the grey water footprint of domestic living, (million m 3 ); c COD,j and c NH 3 -N,j refer to the COD and NH 3 -N concentrations in sector j, respectively (mg/L); c max,COD and c max,NH 3 -N are the maximum allowable pollutants concentrations (mg/L) for COD and NH 3 -N of Grade III water, which are 20 mg/L and 1 mg/L, respectively, according to the China Environmental Quality Standard for Surface Water (GB3838-2002); c nat is the concentration of background pollutants in natural condition (mg/L), and is assumed to be zero as the concentration is very low under natural condition.
For agriculture, the grey water footprint was calculated as: where α denotes the nitrogen leaching loss rate, which is 7% on average in China [24]; Appl denotes the quantity of nitrogen fertilizer (10 4 ton); c max are the maximum allowable pollutants concentration (mg/L) for nitrate, which is 10 mg/L according to the China Environmental Quality Standard for Surface Water (GB3838-2002).

External Water Footprint and Water Footprint Export
When there is goods and services trade among countries, there is virtual blue water and virtual return water trade. The net virtual blue water import (NVBW I, million m 3 ) and net virtual grey water import (NVGW I, million m 3 ) can indicate the underlying blue water and grey water transfer with trade activity, respectively. This study assumes imported goods had the same total water consumption coefficient and total return water coefficient with the domestic one: where NVBW I is the row vector of net virtual blue water import and NVBW I j refers to the net virtual blue water import in sector j; NVGW I is the row vector of net virtual grey water import and NVGW I j refers to the net virtual grey water import in sector j; P net,imp,j (10 4 Yuan) is the net import of sector j. The positive NVBW I and NVGW I indicate a net virtual blue/grey water import, implying a water stress alleviation of China by importing virtual water through trade activities; while the negative NVBW I and NVGW I indicate a net virtual blue/grey water export, implying that the water stress in China is aggravated by exporting virtual water to other countries through trade activities.
If the total net virtual water import (TNVBW I) or total net virtual return water import (TNVGW I) is positive, it is defined as the external water footprint (WF external , million m 3 ), indicating the total net virtual water import for domestic consumption. It consists of external blue water footprint (WF external,blue , million m 3 ) and external grey water footprint (WF external,grey , million m 3 ) [25].
When TNVBW I or TNVGW I is negative, its opposite number is defined as the water footprint export (WF export , million m 3 ), indicating the total net virtual water export for abroad consumption. It consists of blue water footprint export (WF export,blue , million m 3 ) and grey water footprint export (WF export,grey , million m 3 ): WF export,blue = max{−TNVBW I, 0}, WF export,grey = max{−TNVGW I, 0}.
Based on the conceptions of external water footprint and water footprint export, two indices, water self-reliance (WSR) and water export fraction (WEF), were proposed. The level of WSR denotes how much the local consumption relies on local water resources. WSR is one hundred percent if a country was a net virtual water export country; it approaches zero if a country heavily relied on virtual water import.
WEF reflects how much a country exports virtual water abroad. If WEF is zero, the country exports no virtual water abroad, i.e., the country is a net virtual water import country. The larger WEF is, the more the country exports virtual water to other countries.
2.6. Materials Therein, the input-output tables of China were obtained from Chinese Input-Output Association (http://www.stats.gov.cn/ztjc/tjzdgg/trccxh/). Agricultural water consumption is the product of agricultural water use and agricultural water consumption rate obtained from China Water Resources Bulletin. For the second industry, water consumption is the difference of water use and return water obtained from the China Environment Yearbook and China Statistical Yearbook on Environment in 2002 and 2005; especially, water use in 2007, 2010 and 2012 were calculated using return water and the corresponding water consumption rates which were estimated as the average value of those in the last two years. Living water consumption and tertiary industrial water consumption were identified by living water, public water and the corresponding water consumption rates obtained from the China Water Resources Bulletin. On the balance of the total volume, water consumption and return flow in each industrial sector was adjusted in equal proportion to meet the total volume provided by China Water Resources Bulletin. It should be noted that the return water in the China Water Resources Bulletin does not include the discharge of thermal power dc cooling water and mine return water, which can make the grey water footprint in this study a bit underestimated. NH 3 -N and COD concentrations in sectors were determined according to the statistical data in China Environment Yearbook and China Statistical Yearbook on Environment. In particular, the NH 3 -N and COD concentrations in tertiary industry referred to that in living return water for substitution. Quantity of nitrogen fertilizer was obtained from China Statistical Yearbook. All input data are available in Supplementary Materials.

Internal Water Footprint
From 2002 to 2012, the average annual internal water footprint was 3.83 trillion m 3 in China, of which the blue water footprint was 0.25 trillion m 3 and the grey water footprint was 3.58 trillion m 3 (with Grade III water standard accounting). Grey water footprint greatly dominated the internal water footprint and accounted for 93% on average of the internal water footprint, ranging from 91% to 96% during the study period. The internal water footprint overall showed a decreasing trend from 2002 to 2012, except for a dramatic increase in 2010 due to the increase of grey water footprint, blue water footprint did not change significantly, while grey water footprint showed consistent trend with internal water footprint. It is because since 2008, volume of return water in sector 25 experienced a dominant decreasing trend; however, COD and NH 3 -N emissions kept a relatively stable level according to the China Statistical Yearbook on Environment, which resulted in a larger conversion coefficient that caused a dramatic increase of grey water footprint in 2010. Changes of internal water footprint and its compositions are shown in Figure 2.

The Consumption Structure of Blue Water Footprint
From 2002 to 2012, the annual average of WF blue,actual was 0.03 trillion m 3 , and the annual average of WF blue,virtual was 0.22 trillion m 3 ; virtual blue water footprint occupied a large proportion (89%) of the blue water footprint, while the proportion of actual water consumption was almost negligible, indicating that the consumption of goods and services for final use was the major way to produce blue water footprint. Figure 3 shows the inter-annual variations of WF blue and its compositions from 2002 to 2012, it is obvious that WF blue,virtual first decreased from 2002 to 2005, and then increased continuously and slightly from 2005 to 2012. WF blue,actual was relatively stable.  Table 2 shows the sectoral WF blue,virtual of China from 2002 to 2012. Virtual blue water footprint in agriculture (sector 1) was the largest, with an average value of 87 billion m 3 , accounting for 39.2% of the total virtual blue water footprint, followed by that in sector 6 (food and tobacco processing) which was 46.6 billion m 3 , accounting for 21% of the total virtual blue water footprint. Nineteen out of twenty-four industrial sectors had the proportions of WF blue,virtual over total virtual blue water footprint less than one percent. WF blue,virtual of tertiary industry (sector 26) was 33.5 billion m 3 , accounting for 15.1% of the total virtual blue water footprint. Change trends of WF blue,virtual varied among different sectors. Seventeen out of twenty-six sectors had WF blue,virtual increased, among which WF blue,virtual in sector 11 (petroleum processing and coking) increased the fastest during the study period, with an average increase rate of 9.6%; followed by that in sector 10 (papermaking and cultural articles) and sector 17 (transport equipment), of which the increase rates were 9.3% and 8.3%, respectively. WF blue,virtual in sector 2 (coal mining and processing) decreased the fastest, with an average decrease rate of −14.7%, followed by that in sector 5 (non-metallic and other minerals mining) whose decrease rate was −7.9%. WF blue,virtual in sector 19 (electronic and telecommunications equipment) and sector 23 (gas production and supply) had very small changes, with the change rates less than one percent. Virtual blue water footprint was caused by rural residents' consumption (23%), urban residents' consumption (42%), government consumption (7%), gross fixed capital formation and inventory investment (28%).
For the WF blue,virtual caused by rural residents' consumption (Table 3), agriculture (sector 1) contributed the most, of which the WF blue,virtual on average was as high as 32.3 billion m 3 , accounting for 62.8% of the total WF blue,virtual caused by rural residents' consumption, followed by sector 6 (food and tobacco processing), of which the WF blue,virtual on average was 12.4 billion m 3 , accounting for 24.2%. Thirteen out of twenty-six sectors had very small WF blue,virtual caused by rural residents' consumption, all accounting for less than 0.1%. As for change trends, WF blue,virtual caused by rural residents' consumption in sector 14 (metal smelting and processing) and sector 5 (non-metallic and other minerals mining) presented the greatest and second greatest decreasing trends, with the decrease rates of −16% and −15.8%, respectively; while WF blue,virtual caused by rural residents' consumption in sector 23 (electronic and heating power production and supply) and sector 10 (papermaking and cultural articles) presented the greatest and second greatest increasing trends, with the increase rates of 10.7% and 9%, respectively.  Among the WF blue,virtual caused by urban residents' consumption, agriculture (sector 1) also contributed the most, of which the WF blue,virtual on average was 38.5 billion m 3 , accounting for 41% of the total WF blue,virtual caused by urban residents' consumption, followed by sector 6, of which the WF blue,virtual on average was 30.3 billion m 3 , accounting for 32.3%. Seven out of twenty-six sectors had very small WF blue,virtual caused by urban residents' consumption, all accounting for less than 0.1%. As for change trends, eleven out of twenty-six sectors showed decreasing trends. WF blue,virtual caused by urban residents' consumption in sector 3 (crude mining and processing), sector 5 (non-metallic and other minerals mining) and sector 14 (metal smelting and processing) showed the top three greatest decreasing trends, with the decrease rates of −20%, −15.9% and −15.9%, respectively; while WF blue,virtual caused by urban residents' consumption in sector 11 (petroleum processing and coking) and sector 17 (transport equipment) showed the greatest and second greatest increasing trends, with the increase rates of 16.1% and 12.2%, respectively.
For WF blue,virtual caused by government consumption, it concentrated in tertiary industry (sector 26) and agriculture (sector 1), which were 13.1 billion m 3 and 18.8 billion m 3 , accounting for 87.4% and 12.6% of the total WF blue,virtual caused by government consumption, respectively.
Virtual blue water footprint is determined by total water consumption coefficient and the corresponding expenditure. The expenditure reflects the consumption structure, while the total water consumption coefficient is related to the direct water consumption coefficient and industrial chain, thus reflecting the technological level and industrial structure. From 2002 to 2012, all the sectors had their total water consumption coefficients decreased, with the change rates ranging from −8.6% to −0.9% (Figure 4). Therein, the total water consumption coefficient in sector 25 (construction) decreased the most, of which the decrease rate was −8.6%. Therefore, the increase of WF blue,virtual in most sectors was caused by the increase of domestic consumption. Especially, rural residents' demands for sector 23 (gas production and supply) and urban residents' demands for sector 11 (petroleum processing and coking) and sector 17 (transport equipment) expanded obviously during the study period. Taking the total water consumption coefficient in 2010 for example (Figure 4), we find that agriculture (sector 1) had the highest total water consumption coefficient, followed by food and tobacco processing (sector 6).
The compositions of total water consumption coefficients varied among sectors. Taking 2010 for example ( Figure 5), the direct water consumption coefficient in agriculture accounted for the largest proportion of total water consumption coefficient, which reached 81%, followed by that in sector 22 (electronic and heating power production and supply) and that in sector 11 (petroleum processing and coking), of which the proportions were 51% and 44%, respectively. Sectors with the indirect water consumption coefficients higher than 50% reached 24, accounting for 92.3% of the total sectors; those higher than 90% reached 10, accounting for 38% of the total sectors, indicating that most of the virtual blue water footprint was produced by the intermediate virtual water input, i.e., the virtual water embedded in the raw materials, other than the fresh water input. In other words, the accumulated effect of industrial chain had very significant impacts on virtual blue water footprint.

The Consumption Structure of Grey Water Footprint
From 2002 to 2012, the average grey water caused by living return water was 0.72 trillion m 3 , and that caused by virtual return water was 2.89 trillion m 3 (conversion with the Grade III water standard). Grey water footprint caused by virtual return water took a substantial proportion (80%) of grey water footprint, indicating that the consumption of goods and services for domestic consumption was the major way to produce grey water footprint. Table 4 shows the sectoral virtual grey water footprint of China from 2002 to 2012. Virtual grey water footprint in tertiary industry (sector 26) was the highest (sector 25 was not taken into discussion since it does not indicate a specific industry), which was as high as 787.1 billion m 3 , accounting for 27.5% of the total WF grey,virtual , followed by that in sector 6 (food tobacco and processing), which was 676 billion m 3 , accounting for 23.7% of the total WF grey,virtual . Nineteen sectors had the proportions of WF grey,virtual over total virtual grey water footprint less than one percent. As for change trends, seventeen sectors showed overall decreasing trends, with sector 2 (coal mining and processing) decreased the most, with the decrease rate of −18.4%. Sector 15 (metal products) had WF grey,virtual increased the most, with the increase rate of 13.5%.
Among the WF grey,virtual caused by rural residents' consumption (Table 3), sector 6 (food and tobacco processing) contributed the most, which was 181.7 billion m 3 on average, accounting for 54.6% of the total WF grey,virtual caused by rural residents' consumption, followed by tertiary industry (sector 26), which contributed 88.1 billion m 3 on average, accounting for 26.5%; twenty-one out of twenty-six sectors had very small WF grey,virtual caused by rural residents' consumption, accounting for less than one percent. As for change trends, sixteen out of twenty-six sectors showed decreasing trends. WF grey, virtual caused by rural residents' consumption in sector 5 (non-metallic and other minerals mining) and sector 14 (metal smelting and processing) presented the greatest and second greatest decreasing trends, with the decrease rates of −15.7% and −15.4%, respectively; while WF grey,virtual caused by rural residents' consumption in sector 23 (electronic and heating power production and supply) presented the greatest increasing trend, with the increase rate of 5.2%. For the WF grey,virtual caused by urban residents' consumption, sector 6 (food and tobacco processing) also contributed the most, which was 416.8 billion m 3 on average, accounting for 44.8% of the total WF grey,virtual caused by urban residents' consumption, followed by tertiary industry (sector 26), which contributed 312.8 billion m 3 on average, accounting for 33.6%. As for change trends, seventeen out of twenty-six sectors showed decreasing trends. WF grey,virtual caused by urban residents' consumption in sector 3 (crude mining and processing), sector 5 (non-metallic and other minerals mining) and sector 14 (metal smelting and processing) showed the top three greatest decreasing trends, with the decrease rates of −20%, −15.8% and −15.4%, respectively; while WF grey,virtual caused by urban residents' consumption in sector 17 (transport equipment) and sector 11 (petroleum processing and coking) showed the greatest and second greatest increasing trends, with the increase rates of 10.2% and 9.9%, respectively.
For WF grey,virtual caused by government consumption, it concentrated in tertiary industry (sector 26) and agriculture (sector 1), which were 307.5 billion m 3 and 1.3 billion m 3 , accounting for 99.4% and 0.4% of the total WF return,virtual caused by government consumption, respectively.
Virtual grey water footprint was determined by the conversion coefficient of grey water footprint, total return water coefficient and the corresponding expenditure. The expenditure reflects the consumption structure; total return water coefficient is related to the sectoral direct return water coefficient and industrial chain, thus reflecting the technological level and industrial structure; the conversion coefficient of grey water footprint is related to the pollutant concentration, water quality standard and background pollutant concentration, which reflects the production process, and the demands and status of environment quality.
From 2002 to 2012, all sectors had the total return water coefficients decreased, with the decrease rates ranging from −8.8% to −1.3% (Figure 6). The contributions of direct return water coefficient to total return water coefficient varied among sectors. Taking total return water coefficient in 2010 for example (Figure 7), direct return water coefficient in agriculture (sector 1) contributed the greatest, accounting for 80%, followed by that in sector 24 (water production and supply), of which the proportion reached 76%. Sectors that had the proportions of indirect return water coefficient over total return water coefficient over 50% reached 24, accounting for 92% of the total sectors, those over 90% reached 11, accounting for 42% of the total sectors, indicating that most of the virtual return water was produced by the intermediate virtual return water input during the production, i.e., the virtual return water embedded in the raw materials, other than the direct return water. In other words, the accumulated effect of production chain had great impacts on grey water footprint.   Table 5 shows the change trends of conversion coefficients of sectoral virtual grey water footprint under Grade III water quality standard from 2002 to 2012. It is evident that from 2002 to 2012, most sectors had the conversion coefficients decreased, indicating a water quality improvement of return water during the study period. Therein, conversion coefficient in sector 3 decreased most, followed by that in sector 16 (general and specialized machinery). Since both total return water coefficient and conversion coefficient of virtual grey water footprint mostly presented decreasing trends, while final demand exhibited an increasing trend as we discussed in Section 3.2, the conclusion is obvious that the decrease effect caused by total return water coefficients and conversion coefficients of virtual grey water footprint is more significant than the increase effect caused by final demand during the study period. Of particular importance, rural residents' demands for sector 23 (gas production and supply) and urban residents' demands for sector 11 (petroleum processing and coking) and sector 17 (transport equipment) expanded obviously during the study period.

External Water Footprint and Water Footprint Export
From 2002 to 2012, the average annual net virtual blue water export was 17.6 billion m 3 , and the average annual net virtual grey water export was 251.6 billion m 3 , implying China was a typical net virtual water export country, which was consistent with the result by Zhang et al. [26]. Figure 8 shows the compositions of water footprint export in China from 2002 to 2012. It is obvious that from 2007 to 2010, WF export experienced a dramatic decrease due to the sharp decrease of WF export,grey . WF export,grey dominated WF export significantly. WSR and WEF from 2002 to 2012 (Table 6) indicate that, in terms of virtual water trade, China was self-reliant, the water used for producing the products and services to meet domestic consumption was taken domestically; meanwhile, China exported virtual water to other countries, which aggravated the water stress in China. Overall, China's water stress is mainly from the water footprint produced by local consumption; water footprint export, especially after 2010, was relatively small.

Discussion
The average annual internal water footprint in China was 3.83 trillion m 3 in this study, where the blue water footprint was 0.25 trillion m 3 , and the grey water footprint was 3.58 trillion m 3 . The internal water footprint is significantly higher and the blue water footprint is significantly lower than the 0.88 trillion m 3 estimated by Hoekstra [27]. This is because on the one hand, we considered grey water footprint and distinguished water use and water consumption which Hoekstra did not; on the other hand, we did not consider the green water footprint that Hoekstra considered in calculating their agricultural water consumption. All these differences in data processing mean that the two estimates varied.
Different from blue water footprint and green water footprint, grey water footprint primarily focuses on the water quality variation. Many chemical substances can affect the water quality, so indicators for assessing water quality were various, and types and concentrations of pollutants also differed among different sectors. In this study, we choose NH 3 -N and COD as the indexes to assess water quality because they were the most representative index in China's water quality research. Concentration of background pollutants may also vary among different regions and can affect the estimation of grey water footprint, here we neglect the impact by considering the background pollution as zero. Besides, it is also subjective when we set the maximum allowable pollutant concentration. In this study, we refer to the maximum allowable pollutant concentration for Grade III of water quality defined by China Environmental Quality Standard for Surface Water (GB3838−2002). Based on the above analysis, our estimation of grey water footprint should be lower than the actual value.

Conclusions
By using input-output analysis and dilution theory, the internal water footprint, blue water footprint and grey water footprint of China from 2002 to 2012 were estimated, and the consumption structure of water footprint and virtual water trade were analyzed. Conclusions are as follows: (1) From 2002 to 2012, the average annual internal water footprint was 3.83 trillion m 3 in China, in which grey water footprint took a large proportion of 93% on average. Both internal water footprint and grey water footprint experienced a decreasing trend from 2002 to 2012, except for a dramatic increase in 2010; while blue water footprint did not change significantly. (2) The annual average of blue water footprint was 0.25 trillion m 3 , in which the annual average of WF blue,virtual was 0.22 trillion m 3 . WF blue,virtual in agriculture (sector 1) was the largest among sectors, accounting for 39.2% of the total virtual blue water footprint. WF blue,virtual in most sectors showed increased trends due to the increase of final demand. (3) The average grey water footprint caused by living return water was 0.72 trillion m 3 , and that caused by virtual return water was 2.14 trillion m 3 (conversion with the Grade III water standard). Among sectors, annual WF grey,virtual in tertiary industry (sector 26) was the highest, which accounted for 27.5% of the total WF grey,virtual , followed by that in food and tobacco processing (sector 6). WF grey,virtual in most sectors showed decreasing trends due to the decreases of total return water coefficients and conversion coefficients of virtual grey water footprint. (4) For water resources, China was self-reliant, the water used for producing the products and services to meet domestic consumption was taken domestically; meanwhile, China exported virtual water to other countries, which aggravated the water stress in China.  Table S3: Quantity of nitrogen fertilizer; Table S4: Water use and return water in industrial sectors; Table S5: Water use, water consumption and return water.