The Impacts of the Geographic Distribution of Manufacturing Plants on Groundwater Withdrawal in China

The overexploitation of groundwater in China has raised concern as it has caused a series of environmental and ecological problems. However, far too little attention has been paid to the relationship between groundwater use and the spatial distribution of water users, especially that of manufacturing factories. This study proposed a factory scatter index (FSI) that incorporates the latitude and longitude of each plant and calculates the distance between factories to characterize the degree to which manufacturing plants are scattered in China. It is found that counties and border areas between neighboring provinces registered the highest FSI increase. It seems that the degree of scattering of manufacturing plants is closely related to land planning and management of local governments. Further non-spatial and spatial regression models using 205 provincial-level secondary river basins in China from 2016 show that the scattered distribution of manufacturing plants played a key role in groundwater withdrawal in China, especially in fragile ecological-environment areas. The scattered distribution of manufacturing plants raises the cost of tap water transmission, makes monitoring and supervision more diﬃcult, and increases the possibility of surface water pollution, thereby intensifying groundwater withdrawal. A reasonable spatial adjustment of manufacturing industry through planning and management can reduce groundwater withdrawal and realize the protection of groundwater. Our study may provide a basis for water-demand management through spatial adjustment in areas with high water scarcity and fragile ecological environment.


Introduction
Groundwater is the world's largest freshwater resource and accounts for 33% of the annual global freshwater withdrawal.Globally, groundwater supplies drinking water to more than 2 billion people and provides more than half of the irrigation water (Giordano, 2009;de Graaf et al., 2019;Olea-Olea et al., 2020).In recent years, the increase in the global population, urbanization, and rising demands from the industrial and agricultural sectors have led to the excessive abstraction of groundwater, which in turn has led to an extreme lowering of water tables.The overexploitation of groundwater has caused a series of environmental and ecological problems, such as ground subsidence, seawater intrusion, and groundwater pollution (Braadbaart, O. & Braadbaart, F., 1997;Koncagül, 2015;de Graaf et al., 2019;Shah et al., 2000).Water demand must be managed to reduce groundwater consumption and hence to control ecological and environmental risks caused by the overexploitation of groundwater.
As well as generic water-demand management measures such as developmental and technical measures, market-based measures have been reported in the broader literature (Hamdy et al., 2003;Yang et al., 2003;Gilg & Barr, 2006;Chang et al., 2017).However, the spatial distribution of water users is rarely incorporated into water-demand management measures.
By using GIS and statistical models to analyze single-family residential water withdrawal, Chang et al. (2010) found that the water withdrawal of communities with a high degree of aggregation was less than that of scattered communities.Shandas and Parandvash (2010) studied the relationship between land-withdrawal zoning and development-induced water withdrawal in Portland, Oregon, USA.They argued that the coordination between land-withdrawal planning and water demand management should be improved.Additionally, Sanchez et al. (2018) found that the agglomeration patterns of water users have the potential to improve water withdrawal efficiency.These authors showed that the spatial distribution of water users has an important impact on water consumption.However, far too little attention has been paid to the relationship between groundwater use and the spatial distribution of water users.
Groundwater is a vital source of industrial water.In North China, 50% of industrial water consumption is supplied by groundwater (Chinese Ministry of Environmental Protection, 2011).China's manufacturing industries are considered to be characterized by scattered distribution, which is mostly based on qualitatively studies of its status, forming mechanisms, or background of political institutions (Zhu & Guo, 2014;Zhu, 2017;Zhang et al., 2018).As has been found in this study, this scattered distribution often lead to more usage of groundwater, not only because of the difficulties to lay water pipelines, but also because it can lead to severe contamination of surface water and people have to shift to using more groundwater (Brown & Halweil, 1998).However, what is the degree of scattering of China's manufacturing industry?This issue has rarely been quantitatively measured, which limits our ability to study the impacts of manufacturing plants distribution on groundwater.Zheng et al. (2019) investigated the relationship between the dispersion of manufacturing factories and groundwater withdrawal in Hebei Province in the North China Plain.They revealed that, in Hebei Province, the manufacturing industry is relatively dispersed, and the greater the dispersion of the manufacturing industry the greater the groundwater withdrawal.However, it is unclear whether the same relationship exists at the national level, and this research gap limits the planning of groundwater resource demand management.
In this paper, we quantitatively characterize the distribution of manufacturing plants and examine the relationship between the spatial distribution of the manufacturing industry and groundwater withdrawal in 205 provincial-level secondary river basins in China.The remainder of this paper is organized as follows: Section 2 describes the methodology; Section 3 presents the spatial distribution of manufacturing plants in China and empirically analyzes the relationship between the distribution of manufacturing plants and groundwater withdrawal; Section 4 discusses the results; and Section 5 presents the conclusions.

Factory Scatter Index (FSI)
Based on the address of each manufacturing factory in China derived from the Chinese Industrial Enterprises Database, the address resolution method was used to determine the latitude and longitude of each factory (Zheng et al., 2018).Then, using the spatial location of the manufacturing factories, an index named the factory scatter index (FSI) was designed to measure the degree of scattering of the manufacturing factories (Zheng et al., 2019).The FSI was calculated as follows: using the geographic location information for each manufacturing plant, grids were created with cell size d around factory j at the center of a square, as shown in Figure 1.The average distances between factories in each unit were calculated as follows to quantify the extent to which factories were scattered.
where   is the average distance between factory j and all other factories in a study unit; j represents a factory; ∆  and ∆  are the differences in latitude and longitude between factory j and factory s, respectively; and   represents the number of factories that satisfy the conditions j ∈ i and -6 km ≤∆  , ∆  ≤ +6 km.
The FSI of a study unit can then be defined as the average value of dj as follows: where   is the average distance between factories, i denotes a study unit, and   is the number of factories in study unit i.
According to the calculation of Zheng et al. (2019), the optimal value of d that most accurately characterizes the degree to which plants are scattered is 6 km.Areas within 6 km of development are assumed to be hotspots for settlement (Zhang et al., 2009).Therefore, considering that it is difficult to obtain a uniform d value nationwide, we also chose 6 km as the cell size for this study.

Models and Variables
We used the following three models to study the influencing factors of groundwater withdrawal.The first model was the ordinary least squares (OLS), which uses all variables to fit a single linear regression.Its expression formula is: where ln   is the dependent variable,   is the independent variables, and   is the random error.
It was found that there were areas where the groundwater withdrawal was 0. So Tobit model was used to eliminate the influence of zero value.Its expression formula is the same as that of the OLS regression.
To further examine the relationship between independent variables and the dependent variable with the consideration of their spatial variations, Geographically Weighted Regression (GWR) model was also used to integrate the geographic coordinates of each observation into the linear regression model.The expression formula of the GWR model is: where ln   is the dependent variable,  0 is a constant term, (  ,   ) is the spatial position of the sampling point i,   (  ,  ) is the correlation coefficient between variables at point (  ,   ),   is the independent variables, and   is the random error.
In all models, the logarithm of groundwater withdrawal was taken as the dependent variable and the FSI of the same basin as the target independent variable.Since the groundwater withdrawal data at the district and county level is not available, we take 205 provincial-level secondary river basins in China as samples.
Based on related literature, we considered both the natural and social-economic factors as independent variables to examine their influence on groundwater withdrawal.In China, the main types of water use are agricultural water, industrial water and residential water.Among them, agriculture is the largest user of groundwater in China (Zhang et al., 2013).Since the amount of water used in agriculture is closely related to the area irrigated, we characterized it by the area of actual irrigated land.According to the results of a national water conservancy census in 2011, in that year, high-water-consumption industries accounted for 3/4 of China's total industrial water consumption.So we used the number of high-water-consumption factories and the proportion of high-water-consumption factories to the total number of factories to represent the industrial water usage.Residential water use is closely related to population size and level of urbanization, so we adopted total population and urbanization rate.
At the same time, economic efficiency and water use efficiency can also affect water consumption, so we chose GDP per capita and water withdrawal per GDP as indicators.In addition, for natural factors, the average rainfall per year was used to control for the effect of possible increased surface water, and the average temperature per year was used to control for the effect of possible increased water demand.Moreover, considering the differences in factors such as hydrology and climate between North China and South China, provinces were divided into northern provinces and southern provinces according to the perspective of economic geography of Sheng et al. (2018) Table 1 reports the basic characteristics of each variable.Due to the differences in the units of each variable, we normalized all the original data.Table S1 reports the correlation coefficients between the independent variables (after normalization).Although there was a high correlation between temperature and rainfall, they did not affect the target variable.
Additionally, the calculated variance inflation factor was less than 5. Therefore, there were no serious multicollinearity existing among independent variables.798 Note.we used the factories above the designated size in China in the thermal power industry and other high-water-consumption industries as high-water-consumption factories.Among them, industrial enterprises above a designated size in China refer to the enterprises whose annual revenue of main business is more than 20 million yuan according to the National Bureau of Statistics of China.

General Characteristics of the Distribution of Manufacturing Plants in China
In order to determine the general characteristics of the distribution of manufacturing plants in China, we calculated the number of manufacturing plants and the FSIs in the four geographic regions of East China, Central China, West China, and Northeast China, respectively, and subdivided areas in the four regions into districts and counties-which were classified based on the administrative division of China in 2015-in order to compare the differences between urban and non-urban areas.
As shown in Table S2, from 2000 to 2010, the number of manufacturing plants in China increased by nearly two times.Among the four regions, East China had the largest number of manufacturing plants and the fastest growth rate of 223%.From the comparison of counties and districts, in East China and North China, the number of manufacturing plants in districts was found to be more than that in counties, both in 2000 and 2010; meanwhile, in Central China and West China, the number of manufacturing plants in districts was always less than that in counties.Additionally, in East, Central, and Northeast China, the growth rate of the number of manufacturing plants in counties was higher than that in districts.Note.due to the incompleteness of the information about the plants in districts and counties of the provinces of Xinjiang, Tibet, Inner Mongolia, Gansu, and Qinghai, these areas are not discussed.

Empirical Results
Taking groundwater withdrawal as the dependent variable and FSIs as the target independent variable, OLS regression, Tobit regression, and GWR were carried out in turn to examine the relationship between the degree of scattering of manufacturing plants and groundwater withdrawal.
As shown in Table 3, of all the OLS regression results, column (4) performed best, explaining 52.2% of the groundwater withdrawal.Among them, the FSI showed a relatively high importance in the model, accounting for 17.85% of the groundwater withdrawal, ranking third among all the influencing factors (see Figure 5).Generally, the coefficients of FSI in all models were significantly positive, indicating that the degree of scattering of manufacturing plants had a significant impact on groundwater withdrawal.Additionally, the coefficients of the total population and the area of actual irrigated land were also significantly positive in all the models, which is consistent with reality.Furthermore, the urbanization rate was found to be significantly positively correlated with the groundwater withdrawal, meaning that the increase of the urbanization rate will aggravate groundwater withdrawal.Finally, there was a positive relationship between groundwater consumption and water withdrawal per GDP, which indicates that the lower the water withdrawal efficiency, the more groundwater is used.
Furthermore, the regression results of the Tobit model were basically consistent with those of the OLS model (see Table 3, column ( 5)-( 8)); therefore, no further explanation of the results of this model were given.Table 4 shows the results of the GWR.The coefficient of FSI was stable to positive in all results of the GWR, with a minimum value of 0.1043 and a maximum value of 0.3782, confirming that the more scattered the spatial distribution of manufacturing plants, the greater the groundwater withdrawal.
Across all models, GWR outperformed OLS, as indicated by lower AIC values and higher global R-squared values (Table S3).The GWR model explained 60% of the variation in groundwater withdrawal.The important improvement in performance of the GWR relative to the OLS regression indicates spatial non-stationarity in statistical relationships across the study area.Therefore, we provided an in-depth analysis of spatial heterogeneity as represented by the GWR model.As shown in Figure A4, local R-squared values varied from 0.474 to 0.7267 (Figure 6(a)) and the standard error varied from 0.0107 to 0.1471 (Figure 6(b)).We can also see that GWR coefficients varied significantly between different regions of China (Figure 7).
Specifically, the coefficients of the FSI were relatively small in Hebei, Tianjin, Beijing, and Inner Mongolia, while large in West China which is ecologically fragile the most.

Table 4
The Results of Geographically Weighted Regression (GWR).

Discussion
4.1.Regional Differences in the Degree of Scattering of Manufacturing Plants As shown in Figure 2, the FSIs in districts were higher than those in counties and the average FSIs in districts in different regions of China were similar, suggesting that factories within a district are generally farther apart from each other.High land rents owing to high levels of urban services and infrastructure in the districts often push manufacturing to the fringes, where the distance between manufacturing plants are often far apart.
Additionally, from 2000 to 2010, the average FSIs in counties increased more than those in districts.So recently, it was the counties, with relatively low land prices and weak environmental management regulations, that have taken over most of the manufacturing plants.In China, counties usually have fierce competition in attracting investment.Therefore, local governments, especially those in less developed areas, are more supportive than regulated to manufacturing plants.For example, when Foxconn moved to Jincheng, Shanxi province, the local government provided the most favorable policies for land, labor recruitment, water and electricity supply, and tax breaks (Geng & Lin, 2014).As a result, the lack of planning in site selection often leads to a spatially dispersed distribution of regional manufacturing (Fan et al., 2009).
Some researchers may argue that an increase in the distance between plants (i.e.FSI value) is inevitable, especially when a large number of manufacturing plants enter with rapid industrial development.However, our results found that the districts or counties with the largest increase in the number of manufacturing plants are not necessarily those with the largest increase in FSI values.In Jiangsu Province, the number of manufacturing plants increased greatly between 2000 and 2010, but the growth rate of FSI during this period was small, which means that the average distance between factories did not increase much.It seems that the spatial pattern of plants can be adjusted by local government's planning and management (Fan, 1996).The FSI index can reflect the extent to which local government's planning and management plays a role in formatting an appropriate spatial structure.
Interestingly, districts or counties with the largest increase in the scattering degree of manufacturing plants appeared at the boundary between neighboring provinces.Plants located there can often easily escape punishment for polluting because their emissions often affect neighboring provinces.Disputes over pollution need to be reported to higher-level government, which makes management more difficult.So, people there often choose to close an eye on and local governments tend to implement loose land planning and management in border areas (Duvivier & Xiong, 2013).Therefore, the plants there often located according to their own requirements, e.g.large areas of single-story plants, which lead to a dispersed distribution of manufacturing plants there.

Effects of the Scattering of Manufacturing Plants on Groundwater Withdrawal
The results of this study suggest that the degree of scattering of manufacturing plants has a significant impact on groundwater withdrawal, that is, the more scattered the manufacturing plants are, the larger the groundwater withdrawal.In China, areas of enterprise clusters (such as industrial parks) are usually equipped with complete municipal waterworks and facilities (Zhao et al., 2013).So, it is convenient to monitor and charge for the water consumption of manufacturing plants, and it is also easier for local water resources department to supervise water use in the cluster area.Since strict management can lead to the increase of cost, manufacturing plants have to reduce their cost by improving the resource utilization efficiency, such as saving water or upgrading technology (Wang et al., 2018).Therefore, the manufacturing plants in the cluster area tend to reduce the use of groundwater.
However, scattered distribution of manufacturing plants increases the cost of pipeline laying, making municipal works difficult.In addition, due to advances in water drilling technology, scattered factories usually choose to drill on-site, especially when China did not restrict well drilling in previous years (e.g., there were 58 counties with more than 10,000 wells and 6 counties with more than 100,000 wells in Hebei Province in 2011) (Zheng et al., 2019).
Manufacturing plants scattered in rural areas may also share wells with villagers.In such cases, the amount of water used by factories cannot be assessed quantitatively, and strict monitoring cannot be performed (Zhang et al., 2014).The cost of water for scattered factories is relatively low; factories that share wells with villagers usually pay only a small fee to the local village.
Therefore, with convenient well drilling and low water costs, manufacturing plants have weak awareness of water conservation, and groundwater over-extraction and waste occur frequently.
The discharge of sewage into local rivers leads to surface water contamination, leaving no available clean surface water, which in turn causes the entire region to rely on groundwater (Brown & Halweil, 1998).The discharge of wastewater has an important indirect but non-negligible impact on the increase of groundwater use throughout China.
It is clear that the scattering of manufacturing plants and the corresponding water management play an exceptionally significant role in groundwater withdrawal, which deserves much attention.
4.3.Regional Differences in the Effects of the Scattering of Manufacturing Plants on

Groundwater Withdrawal
The regression results of the GWR model (Figure 7) show that the impact of the degree of scattering of manufacturing plants on groundwater withdrawal varies in different regions of China.As such, when planning efficient water resources development, it may be more useful to adopt different water conservation strategies in different regions according to the spatially varying trends in groundwater withdrawal than to adopt a "one size fits all" strategy.
It is noted that the spatial scattering of manufacturing plants is not the most important factor affecting groundwater withdrawal in North China (Figure 7(a)).The actual irrigated agricultural area is the main variable affecting groundwater consumption there (Figure 7(b)).This is consistent with the results of Tian et al. (2016), who found that agricultural irrigation is the main factor affecting groundwater withdrawal in the North China Plain, and the greater the dependence of agricultural irrigation water on groundwater, the more serious the groundwater withdrawal is.
However, the spatial scattering of manufacturing plants has a great impact on groundwater withdrawal in West China, especially in fragile ecological-environment areas (Figure 7(a)).The reasons for this are as follows: First of all, groundwater is an important source for industrial, agricultural, and domestic usage in West China due to severe shortage of available surface water (Wu et al., 2020).Secondly, compared with East China, West China is geographically vast and sparsely populated, so the cost of tap water transmission caused by the scattered distribution of manufacturing plants is much higher.Additionally, the broad jurisdiction of district and county governments in West China makes it more difficult to regulate the use of water by scattered manufacturing plants.Therefore, in West China, factories often choose to use groundwater, which is more convenient and available, to save costs.
Thirdly, due to the lack of unified water withdrawal planning in West China, the structure of water use and industry is irrational, resulting in the low efficiency of water withdrawal (Liu et al., 2016).The ecological environment of West China is extremely fragile, and groundwater withdrawal will aggravate this vulnerability, thus affecting local sustainable development (Zhou, 2015;Wang & Shao, 2016).However, given the data availability, the empirical part of this paper used only one year of provincial-level secondary river basin data.The lack of accurate data made it impossible for us to continue to measure the impact of the scattered distribution of manufacturing plants on groundwater withdrawal at the district and county scale.With the availability of various resource data in the future, we believe that we will be able to measure the impact of the FSI on resource consumption and environmental pollution in a more detailed way, which is our next research direction.

Figures S1 to S2
Tables S1 to S3 Introduction This supporting information reports additional results of spatial and empirical analyses.Table S1 reports the correlation coefficients between the independent variables (after normalization).
Table S2 shows the number of manufacturing plants in different regions of China.
Table S3 is a comparison of the regression results of the OLS and GWR models, which shows that GWR outperformed OLS across all models, as indicated by lower AIC values and higher global R-squared values.

Figure 1 .
Figure 1.The calculation of the average distance between factory j and factory s.

Figure 2
Figure 2 displays the FSIs in different regions of China in 2000 and 2010, and further Figure 2. a Values of the FSI in four regions of China in 2000.b Values of the FSI in four

Figure 3
Figure 3 shows the spatial distribution of FSIs in China in 2000 and 2010, respectively.

Figure 3 .
Figure 3. Values of the FSIs in districts and counties of China in 2000 and 2010.

Figure 4 Figure 4 .
Figure 4 shows the spatial distribution of the change rate of the number of manufacturing

Figure 5 .
Figure 5. Measures of relative importance for ordinary least squares (OLS) regression

Figure 6 .
Figure 6.The spatial distribution of the local R-squared (a) and standard error (b) in the

Figure 7 .
Figure 7.The coefficients of some variables in the GWR model.

Figure
FigureS1shows the FSIs by province and displays the distribution of manufacturing plants in the four provinces with the highest FSIs in 2000 and 2010, respectively.

Figure
Figure S2 displays the number of manufacturing plants in districts and counties of China in 2000 and 2010.

Figure S2 .
Figure S2.The number of manufacturing plants in districts and counties of China in 2000 and 2010.

Table 1
Variable Summary Statistics

Table 2 251
Table2).As shown in Table2, the national average change rate of FSIs from 2000 to 2010 was 33%.Among the 26 studied provinces, 11 had FSI change rates that were higher than the national average.By comparing the change rate of the number of manufacturing plants and the change rate of FSIs, it was found that the provinces with the largest increase in the number of manufacturing plants were not necessarily those with the largest change rate of FSIs.In other words, the increase in the number of manufacturing plants does not necessarily lead to the spatial scattering of manufacturing 249 plants, which is consistent with the above conclusion.250TheChange Rate of the Number of Manufacturing Plants and the Change Rate of FSIs in 252

Table 3
The Results of Ordinary Least Squares (OLS) Regression and Tobit Regression

Table S1 .
Correlation matrix between the independent variables after normalization.

Table S2 .
The number of manufacturing plants in different regions of China.