The input-output (IO) model, which is based on the Leontief model, is recognized as an analysis tool to explain the relationships between different sectors from an economic perspective. However, the traditional single region input-output table can explain trade flow only in a single region. To elaborate interregional flows of goods and services, a multiregional input-output (MRIO) model is developed, which can fully reflect the economic connections between sectors in the same and different regions. Typically, the raw balance in the MRIO model can be expressed as: x r = A r r x r + y r r + ∑ s ≠ r A r s x s ∑ s ≠ r y r s (1) where x r and x s are the total output of region r and region s, respectively; A r r represents the local intermediate inputs in region r; A r s represents the direct input coefficient from region r to region s; and y r r and y r s express the final demand from region r to itself and to region s, respectively. We merge different regions in the same expression, and the following matrix equation can be obtained: X = A X + Y (2)

To solve formula 2, we can obtain the following: X = ( I − A ) − 1 Y = L Y (3) where I, A, L, and Y represent the identity matrix, technical coefficient matrix (intermediate use divided by the total output, representing the direct input in sectors required by unit of output), Leontief inverse matrix (the production structure, representing the direct and indirect input in sectors required by unit of output), and final demand (including household consumption, government expenditures, investment, and net export), respectively.

An environmentally extended MRIO (EE-MRIO) model extends the MRIO by calculating the environmental flows of each sector. A water use extended MRIO model can comprehensively characterize the water footprint relationship between different regional economies. It can be used to quantify the water footprint of each region from the consumption of specific goods or services, as well as the transfer of the water footprint through interregional trade. With further extension by environmental factors, the following equation is obtained: W = d ( I − A ) − 1 Y (4) where W represents each province’s sectoral water consumption and d is regarded as the direct water intensity, namely, water withdrawal per unit output). Here, we focus on the effect of urbanization on total water use. Therefore, rural and urban household consumption is considered in Eq. 3: H w = d ( I − A ) − 1 Y h = d L Y h (5) where H w is the total water withdrawal vector attributable to household consumption Y h .

We propose a two-stage structural decomposition model to reveal the drivers of the household water footprint in China. Several methods are usually contained in SDA, such as traditional ad hoc decomposition, the D&L method (Dietzenbacher and Los, 1998), the S/S method (Shapley, 1953), and the MRCI method (Chung and Rhee, 2001). Additive decomposition is the most common form of SDA. The result of D&L is cumbersome when the number of main factors is larger than 5. S/S and MRIC leave a residual term in the results and fail the factor-reversal test (Wei et al., 2020).To solve the problem of factor irreversibility, Sun, (1998) proposed a complete structural decomposition method that distributed the residuals evenly among the factors. Based on our previous studies (Wang et al., 2019; Wei et al., 2020), we developed a two-stage complete decomposition method to identify drivers of the household water footprint in China.

In the first stage, the household water footprint over the period [0,t] can be decomposed as follows: Δ H w =   H w t − H w 0 = d t L t Y h t −   d 0 L 0 Y h 0 =   ( d t − d 0 ) L 0 Y h 0 + ( L t − L 0 ) d 0 Y h 0 + ( Y h t − Y h 0 ) d 0 L 0 +   ( d t − d 0 ) ( L t − L 0 ) Y h 0 + ( L t − L 0 ) ( Y h t − Y h 0 ) d 0 + ( Y h t − Y h 0 )   ( d t − d 0 ) L 0 +   ( d t − d 0 ) ( L t − L 0 ) ( Y h t − Y h 0 ) =   Δ d L 0 Y h 0 + Δ L d 0 Y h 0 +   Δ Y h d 0 L 0 +   Δ d Δ L Y h 0 +   Δ L Δ Y h d 0 +   Δ Y h Δ d L 0 + Δ d Δ L Δ Y h (6)

Then, the contributions of changes in d, L and Y h   to the total change in H w   are approximated as follows: d w = Δ d L 0 Y h 0 + Δ d ( Δ L Y h 0 + L 0 Δ Y h ) 2 + Δ d Δ L Δ Y h 3 (7) L w = Δ L d 0 Y h 0 + Δ L ( Δ d Y h 0 + d 0 Δ Y h ) 2 + Δ d Δ L Δ Y h 3 (8) Y h w = Δ Y h d 0 L 0 + Δ Y h ( Δ d L 0 + d 0 Δ L ) 2 + Δ d Δ L Δ Y h 3 (9)

However, the effects of urbanization and consumption patterns remain unclear. Then, in the second stage, we further decompose Y h   as follows: Δ Y h = Y h t − Y h 0 = C u t P u t + C r t P r t − C u 0 P u 0 − C r 0 P r 0 =   C u t ( P u 0 + U n + M ) + C r t ( P r 0 + R n − M ) − C u 0 P u 0 − C r 0 P r 0 = C u t P u 0 + C u t U n + C u t M + C r t P r 0 + C r t R n − C r t M − C u 0 P u 0 − C r 0 P r 0 = ( C u t − C u 0 ) P u 0 + ( C r t − C r 0 ) P r 0 + C u t U n + C r t R n + ( C u t − C r t ) M (10) where C r and C u are the per capita rural and urban consumption, respectively;   P u , P r ,   U n , M ,   R n are the urban population, rural population, natural population increase in the local urban area, population migrated from rural areas to the urban area, and natural population increase from local rural areas, respectively.

Thus, using Eqs 5–9, the increment in the household water footprint can be further decomposed as follows: Δ H w = d w + L w + h u w + h r w + P u w + P r w + r w (11) d w = Δ d L 0 Y h 0 + Δ d ( Δ L Y h 0 + L 0 Δ Y h ) 2 + Δ d Δ L Δ Y h 3 (12) L w = d 0 Δ L Y h 0 + Δ L ( Δ d Y h 0 + d 0 Δ Y h ) 2 + Δ d Δ L Δ Y h 3 (13) h u w = k [ ( C u t − C u 0 ) P u 0 ] (14) h r w = k [ ( C r t − C r 0 ) P r 0 ] (15) P u w = k [ C u t U n ] (16) P r w = k [ C r t R n ] (17) r w = k [ ( C u t − C r t ) M ] (18) where k = d 0 L 0 + ( Δ d L 0 + d 0 Δ L ) 2 + Δ d Δ L 3 ; d w , L w ,   h u w ,   h r w , P u w ,   P r w , and r w denote the contributions to the household water footprint made by water intensity, production structural change, urban consumption pattern, rural consumption pattern, urban population, rural population and urbanization, respectively. The decomposition is reversible.

Three types of data are needed in this study: MRIO tables in 2012 and 2017, the annual water footprint of the sectors of each province, and the rural-to-urban migration data of each province.

The 2012 and 2017 MRIO tables were collected from the CEADs dataset (CEADs, 2021). We adjusted the 2017 MRIO table to constant 2012 prices to eliminate the impact of inflation. Price indexes from 2012 to 2017 were collected from provincial statistical yearbooks (National Bureau of Statistics of China, 2018).

Water withdrawal data and water use data were collected from various sources. In this study, we focused on “blue water” as defined in Yang et al. (2015). For agriculture, water use data of each province were obtained from the China Statistical Yearbook (Ministry of Water Resources: National Bureau of Statistics, 2013; National Bureau of Statistics of China, 2018). Provincial water withdrawals in the industry with detailed sectoral data for 2012 and 2017 were estimated from China Economic Census Yearbook (National Bureau of Statistics of China, 2008) in proportion to sectoral water withdrawal data, considering the economic effect on the sector structure by multiplying GDP rates and checking for accuracy with the gross industrial water use data in the current year. Water use data in the construction and service industries were collected from the Bulletin of the First National Census for Water conducted by the Ministry of Water Resources of China in 2011 and estimated by allocating the water use in a production activity based on the proportions of intermediate inputs from the “water production and supply” sector to different economic sectors in the input-output tables. Sector abbreviations can be found in the Supplementary Table S1.

It is worth noting that rural-to-urban migration cannot be directly obtained from statistical yearbooks. Therefore, we need to estimate the natural birth rates of the urban and rural populations and then extrapolate rural-to-urban migration from the statistical data. Limited by the availability of relevant data, we hypothesize that the rural birth rate is 0.4% higher than the urban birth rate of reference, according to the study of (Cui, 2011). The migrating population can be calculated by the difference between natural population data and statistical data. In addition, the natural birth rate, death rate and total population of urban and rural areas were obtained from the Statistical Yearbook.

Due to data availability, we targeted our study at the provincial scale. The study area includes 22 provinces, five autonomous regions, four municipalities directly under the Central Government, Hong Kong, Macau, and Taiwan. For convenience, we refer to all of these as provinces in the rest of the paper.

We calculated the changes in the water footprint in the household sector from 2012 to 2017. Overall, the household water footprint in mainland China dropped from 292.33 billion m³ in 2012 to 291.18 billion m³ in 2017, showing a slight decrement of approximately 1.15 billion m³. However, Figure 1 reveals that different provinces represented different household water footprint changes. Most provinces displayed an increasing household water footprint trend during the study period, in which Jiangsu and Hubei ranked at the top, with increments of 5.6 and 5.5 billion m³ of household water, respectively. In addition, 12 provinces showed a decreasing trend, in which Guangdong, Shanghai and Shandong obtained the largest water footprint decrements of approximately 14.5, 13.5, and 9.1 billion m³, respectively. In general, household water footprint values showed a downward trend in the eastern coastal region and an upward trend in the central and western regions. In addition, household water footprint of sectors in each province can be found in Supplementary Table S2.

Over the studied period, China’s urbanization rate rose from 53% in 2012 to 59% in 2017, with a 1.2% annual growth rate, which was significantly higher than 0.37%, the value of the world average growth rate (Department of economic and social affairs of United nation, 2018). A total of 101.65 million people moved from rural to urban areas from 2012 to 2017. Figure 2 shows the urbanization rates among provinces, and the dots represent the annual gradient of the urbanization rate. Most provinces exhibited increasing urbanization rates, except for Shanghai, which exhibited a “counter-urbanization” phenomenon, with the urbanization rate dropping from 89.3 to 87.7%. The urbanization rates of provinces exhibited obvious polarization. Eastern coastal provinces showed a higher urbanization rate than inland provinces. Nevertheless, inland provinces had a more rapid urbanization process than coastal provinces, which meant that the disparities in the urbanization process among provinces were diminishing, and China’s economic development was becoming more balanced.

Consumption patterns change when people migrate from rural to urban areas because of the significant differences in infrastructure, income distribution, material basis and lifestyle. Taking the consumption level as an example, Figure 3 shows that obvious growth occurred in both rural and urban areas from 2012 to 2017, while the per capita rural consumption level increased from 1,159 dollars in 2012 to 1,927 dollars in 2017, with an annual growth rate of 13.2%, and the per capita urban consumption level increased from 2,837 dollars to 3,993 dollars, with an annual growth rate of 8.1%. In addition, the gaps between rural and urban consumption levels (calculated by urban per capital consumption level minus the rural per capital consumption level) exhibited obvious regional differences. The value of the gaps among provinces ranged from −83.4 to 55.9%, with an average of −28.6% during the study period. Overall, the gaps were enlarging in most provinces, except for developed provinces (such as Shanghai and Guangdong) and less-developed provinces (such as Jilin), where obvious declines can be observed in Figure 3C.

Figure 4 shows the contributions of the various driving factors to the household water footprints of the provinces; the dots represent the total changes. Here, we divide the driving factors into three categories: technological level (water use intensity, production structure), consumption patterns (rural consumption pattern, urban consumption pattern), and population (rural population, urban population, and urbanized population). The contributions of these three categories were ranked as technological level (−75.6 billion m³), consumption patterns (52.6 billion m³), and population (21.8 billion m³).

At the technological level, the water use intensity of each province showed a downward trend, indicating that the water efficiency of each province improved during the study period. The greatest reductions can be seen in several relatively developed provinces, such as Guangdong (−8.6 billion m3), Shandong (−5.3 billion m³), Jiangsu (−5.2 billion m³) and Zhejiang (−3.5 billion m³), and agricultural provinces, such as Xinjiang (−7.8 billion m³) and Guangxi (−3.9 billion m³). On the whole, water intensity had a greater impact on the south than on the north because the rich water resources in the south led to a more water-intensive lifestyle and a greater potential for water-saving. The production structure contributed to a relatively slight impact but showed marked regional differences. The negative effect basically played a role in the northern provinces, such as Shandong (−5.2 billion m³), Inner Mongolia (−0.63 billion m³), Henan (−0.50 billion m³), and Beijing (−0.37 billion m³). That is, the water-saving performance of the industrial structure of northern provinces is better than that of southern provinces, possibly because the high water shortage rate in these provinces forces industries to make structural changes that support water saving. Shanghai also demonstrated a considerable reduction of −0.78 billion m³. Overall, there is a high water-saving capacity at the technological level.

Consumption patterns are identified to have positive effects. It is interesting to find that the household consumption patterns of developed provinces tend to be more water-friendly while developing provinces tend to be more water-intensive as the economy grows. Therefore, the consumption patterns had different impacts on the water footprints in regions at different stages of economic development. The consumption patterns in western regions, where developing provinces were concentrated, had positive impacts on the changes in the household water footprint. The eastern coastal developed regions, such as Shanghai (−11.1 billion m³), Guangdong (−9.9 billion m³), and Tianjin (−0.02 billion m³), demonstrated slight declines, which implied that the developed provinces might have crossed the threshold of the environmental Kuznets curve and that continued economic growth might help save water and reduce pollution. Hence, there probably exists a decoupling relationship between the water footprint and economic growth in the case of consumption pattern transformation. From the perspective of rural and urban consumption patterns the water use increments due to changes in the urban consumption pattern were approximately 1.4 times the increments in rural areas. The obvious distinction between urban and rural consumption impacts indicates that a high water demand arises when people move from rural areas to urban areas.

Population is another vital driving force of the water footprint, and additional virtual water demand is required when the population grows. The influence of the total population ranked third but is still worthy of attention. Urbanization (namely, the migrated population) presented a positive impact in almost all provinces, except for Shanghai (−0.04 billion m³), due to local counter-urbanization. The contribution of urban population change was relatively small but represented a positive impact in most provinces, except for some north-eastern provinces, such as Heilongjiang (−0.06 billion m³), Liaoning (−0.05 billion m³) and Jilin (−0.02 billion m³), whose results hinted that urban population outflow existed. The rural population change has an effect similar to the urban population change, i.e., it positively increased the water footprint in most provinces except for north-eastern provinces.

Overall, the technological level can bring considerable changes through policy guidance and technological innovation, which the government can control or guide easily. Generally, consumption patterns and population change contributed continuously to the household water footprint and were not easy to control; thus, more attention to these factors is needed, especially in the context of China’s rapid urbanization and transformation to a consumption-driven economy. In this regard, it is essential to investigate the impact of urbanization and lifestyle change on cross-regional water use to formulate more water-efficient policies to tackle regional water shortages.

The consumption patterns had a positive impact overall and increased the cross-regional water footprint by 15.4 billion m³ during the study period. Figure 5 highlights the changes in the cross-regional water footprint driven by consumption patterns. The horizontal axes show the water footprint caused by changes in consumption patterns of commodities transferred from other provinces (influencing provinces) to the focal provinces, and the vertical axes show the water footprint transferred from the focal provinces to the other provinces (influenced provinces).

From the perspective of influencing provinces, Shanghai and Guangdong exhibited obvious water decrements for other provinces of 0.49 billion m³ and 0.55 billion m³, respectively. To further determine the reason, we present the per capita urban consumption in the sectors of the two provinces in Figure 6, which display similar trends regarding the water-use structure change. Namely, both provinces spend less on water-intensive sectors, such as agriculture (AGR), food processing and tobacco (FPT), the chemical industry (CHE), metal products (MEP), and electrical equipment (EIE), while more expenditures are made in water-friendly sectors, such as real estate (TRE), education (EDU), and health and social work (HSW). According to our results, the appropriate consumption pattern can help provinces save water as the economy grows.

From the perspective of influenced provinces, Xinjiang was the most influenced province, where agriculture was the largest product-exporting sector and contributed to an increase in the water footprint of 6.83 billion m³, accounting for 95.5% of the total water footprint increment. Cai et al. (2019) also recognized that agriculture was the most water-intensive sector and contributed approximately 70% of the water footprint. For both rural and urban consumption patterns, Guangdong, Shanghai and Shandong exerted obvious negative impacts on Xinjiang, while Jiangsu, Fujian and Chongqing had obvious positive impacts, which implies that Xinjiang’s main export destination gradually changed from developed coastal provinces to provinces with faster urbanization.

Our results indicate that owing to the larger proportion of agriculture in their industrial structure, inland regions usually face considerable water use stress due to their contributions to national supply chains, especially in the trade with developed regions. We also found that the most obvious water footprint transfers occurred between neighbouring provinces, and this phenomenon partly confirms a more frequent commercial intercourse between neighbouring provinces from an environmental perspective. Generally, the eastern and coastal regions, usually with prosperous tertiary industries, were able to save their water footprint from their positions in national supply chains by trading of goods and services with inland regions.

Urbanization, the largest driving factor of the population, plays an important role in the water footprint. Figure 7 highlights the changes in the cross-regional water footprint driven by urbanization, adding up to an increment of 4.6 billion m³. The effect of urbanization should be the combined result of changes in the consumption level and the migrant population between rural and urban areas, which distinguishes this study from other studies. Figure 8A denotes the change in the consumption pattern driven by urbanization, and Figure 8B shows the migrant population in each province.

Urbanization mainly led to positive cross-regional water flow, except for Shanghai. In addition, the changes in the water footprints in Heilongjiang and Xinjiang were affected by the urbanization process. What is evident in Figure 8 is that the eastern region had a larger migrant population than other regions; however, Shanghai is the only province showing negative migration because of local counter-urbanization. We find that the provinces most affected by urbanization often showed prominent gaps in consumption levels between urban and rural areas, high migration, or both (Figure 8), as usually observed in developed areas. Hence, we deduced that the cross-regional impact of urbanization may be closely related to local development. Less-developed provinces, such as Jilin, Tibet, Qinghai, and Xinjiang, displayed relatively small changes in their cross-regional water footprints, with slight gaps between local urban and rural areas and less migration. Developed provinces, such as Jiangsu and Shandong, exhibited growing water footprint trends with urbanization mainly because of the local rapid urbanization process, as shown in Figure 2. From the perspective of affected provinces, Xinjiang was the most affected province, which indicates that the rest of China had an increasing demand for water-intensive products from Xinjiang to satisfy the needs of urbanization. Similarly, Heilongjiang, as the “granary of China”, also showed a significant increase in its water footprint. Generally, agriculture-dominated provinces are more affected by other provinces in the process of urbanization because urbanization leads to an increase in food demand, which is transmitted through a cross-regional supply chain to agricultural provinces and results in an increase in the local water footprint.