A spatial autocorrelation analysis method was employed to process the spatial change of rural water supply over the past 19 years in the People's Republic of China. Statistical analyses indicate great achievements in rural water supply construction. Two main indices describing rural drinking water supply status, the Rural Popularization Rate of Tap Water and the Rural Popularization Rate of Water Improvement Beneficiaries, were found to be spatially auto-correlated. The Global Moran's I of the latter decreased generally, and local spatial autocorrelation analysis showed that the regional gap of rural water supply infrastructure is declining. The main factors affecting the spatial pattern of rural water supply were analyzed through the mean centre method. Our research shows that the spatial pattern of economic development and government investment has had a decisive role in the formation and evolution of rural water supply.
Access to safe drinking water is a basic human need and diseases related to unsafe water constitute a major public health concern in developing countries. One of the primary targets of the Millennium Development Goals (MDGs) is to halve by 2015 (from 1990 levels) the proportion of the population without sustainable access to safe drinking water and basic sanitation (United Nations (UN), 2000). Substantial progress has been made toward meeting this goal, and the People's Republic of China (PR China) has always been focused on the water supply of rural areas which is home to half of the population (Yang et al., 2012). The Chinese government has prioritized this issue and a series of engineering and management measures have been instituted (Liu, 2008). Rural water improvement in PR China can be divided into three stages (Zhang, 2011a, b) as follows:
The development and improvement of decentralized water supply (in force from the founding of the Republic to the early 1980s).
The construction of a centralized water supply (in force since the end of the first stage to the first years of this century and supported by the Rural People/Livestock Drinking Water Program (RPLDWP) with funds from the central government as well as from international cooperation involving foreign governments and the World Bank).
The rapid development of rural drinking water sources (currently in force via investments in the rural drinking water sector by the central and local governments).
During the ‘11th Five-Year Plan’ (2006 to 2010), 105.3 billion (109) Renminbi (RMB) was invested in building 225,000 centralized water supply plants that provided drinking water for 21.2 million people (Yan et al., 2011). In the ‘12th Five-Year Plan’ (2011 to 2015), investment in rural drinking water was increased (NDRC et al., 2012). Over 520,000 centralized water supply projects have been built in rural areas and by the end of the year 2010, 58% of the total rural population had access to safe water, thereby providing social and economic as well as environmental benefits. However, many problems remain in this sector (Gong & Chen, 2009). A prominent problem is the unbalanced infrastructure between the eastern and western parts of the country with regard to rural water supply (Zhou & Zhou, 2009). This imbalance has contributed to the shortage of public services in the western regions, which constrains consumption and counteracts the establishment of sustainable economic development. Hence, safe drinking water is not only a public health problem but a more comprehensive problem related to basic public service equalization, sustainable economic development, social justice and social stability.
Much research effort has gone into the regional imbalance of rural water supply in PR China, but few have noticed that the regional differences in rural water supply infrastructure vary positively and negatively over time. It has been shown that regional economic differences lead to differences with regard to rural health services (Zhao et al., 2008). However, the rural water supply was not treated as part of the rural health service in that study. Progress towards MDG target 7C is monitored by two indicators: the proportion of the population using improved sources of drinking water and the proportion of the population using improved sanitation facilities (WHO & UNICEF, 2010; Onda et al., 2012). The Rural Popularization Rate of Tap Water (RPRTW) and the Rural Popularization Rate of Water Improvement Beneficiaries (RPRWIB) are frequently used as statistical indicators to assess the improvement of regional rural drinking water in PR China. Details regarding the two indices are given in Section 2.2. This study applies spatial analysis to assess the annual variation of these two indices and attempts to analyze the reasons for this variation to provide support for governmental policy formulation.
Materials and methods
The two indices (RPRTW and RPRWIB) were collected at the provincial level for the period 1993 to 2011 from the China Health Year Book (MOH, 2009, 2010–2012), and the gross domestic product (GDP) per capita data were collected from the China Year Book (NBS, 1994–2012). Because the Chongqing Municipality was separated from Sichuan Province and established as a municipality directly under the Central Government in 1997, the Chongqing data before 1997 were included as Sichuan's in this study. The reason for choosing 2011 as the end year is due to the lack of available relevant data for 2012. Hong Kong, Macao and Taiwan were not included due to statistical reasons.
Operationalization of the indices
RPRTW indicates the proportion of the population drinking tap water in rural areas via measuring the number of people in rural areas with access to tap water per 100 residents, and tap water here refers to water provided by rural waterworks. The rural waterworks include facilities where source water is stored, purified and distributed to the local rural community (generally speaking these sites have been gradually established in recent years) as well as facilities that pump ground water and directly supply water to local users. The facilities storing unpolluted spring water and sending the water via networked pipes are also included in our conception of waterworks. RPRWIB measures the number of people with access to improved rural drinking water supply per 100 residents. The improved water supplies include all the supplies mentioned above in RPRTW as well as hand-pump wells, cisterns and rainwater collectors. The RPRTW is always larger than the RPRWIB.
Spatial autocorrelation analyses
A global spatial autocorrelation statistic method is used to measure the correlation among neighbouring observations and to find patterns and levels of spatial clustering among neighbouring districts (Cliff & Ord, 1970). The value of the Moran's I statistic ranges between −1 and 1. Values closer to 1 indicate positive spatial autocorrelation, while values closer to −1 indicate negative spatial autocorrelation. When there is no autocorrelation, the expected value approximates zero. However, the Global Moran's I cannot indicate where clusters are located or the type of spatial autocorrelation found (Anselin, 1995). A local indicator of spatial autocorrelation (LISA) was therefore applied to investigate these patterns. The local Moran detects local spatial autocorrelation (Anselin, 1995) and can be used to identify local clusters and spatial outliers. Anselin defined LISA statistics as having the following two properties: (i) they must indicate the extent of significant spatial clustering of similar values around each observation; and (ii) the sum of LISAs for all observations is proportional to a global indicator of spatial association (Anselin, 1995). The Moran scatter plot (Anselin, 1993) is a useful visual tool for exploratory analysis, because it enables the assessment of how similar an observed value is compared to neighbouring observations. Its horizontal axis (also known as the response axis) is based on the values of the observations. The vertical axis is based on the weighted average or spatial lag of the corresponding observation on the horizontal axis. The Moran scatter plot has four quadrants which indicate four types of local spatial patterns. The first quadrant indicates ‘High-High’, that is, high values surrounded by neighbouring units which are also high. The second quadrant indicates ‘Low-High’ signifying low values adjacent to neighbouring units with higher values. The third quadrant indicates low values surrounded by neighbouring units of equally low values. The fourth quadrant ‘High-Low’ indicates high values adjacent to neighbouring lower values. Although a Moran scatter plot can show local spatial patterns, it is not a LISA because it cannot provide the significance index of local clustering. We set up a LISA significance map with both local spatial patterns and significance information. On this map ‘Not Significant’ values are also given, which indicate a lack of spatial autocorrelation. The high-high and low-low locations (positive local spatial autocorrelation) are typically referred to as spatial clusters, while the high-low and low-high locations (negative local spatial autocorrelation) are termed spatial outliers.
Spatial centre statistics
Spatial centre statistics are a basic method to describe the spatial distribution of target variables. Mean centre, median centre and standard distance are commonly used indices to describe the spatial tendency of dispersion (Goodchild, 1987). These values can be used to track changes in distribution over time or to compare the distributions of different features (Environmental Systems Research Institute, 2004). In this study, the mean centre index is used to describe the mean spatial variation over time.
In the above equation, xi and yi are the coordinates for feature i and n is the total number of features.
The weighted mean centre is calculated from giving different weights to spatial object geometry. The advantage of this index is that multiple features of the object can also be included by choosing different attribute variables as weights.
In the above equation, wi is the weight at feature i.
To analyze the cause of the spatial pattern, the weighted mean centre of RPRWIB, RPRTW, GDP per capita and PIPR were calculated yearly, and scatter plots were made to describe the trajectories of these indices.
All the data were cleaned in dBase format (DBF) and were connected with the China region border shapefile (SHP file extension) via the province ID. The latitude and longitude of the provinces in this map were calculated from a centroid. There are two methods for spatial weight: the simple neighbouring weight and distance neighbouring. Most analysis studies that use spatial autocorrelation observe the common definitions of spatial neighbouring relations such as rook and bishop or Queen and King (Elobaid et al., 2012). Because the borders of the provinces in China are irregular, the neighbouring weights can be more accurate measures of spatial relationships as compared to simple distance-based weights. The emphasis of this study is how the neighbouring relation affects the two indices of rural drinking water infrastructure, so the first order Queen Contiguity was selected as the rule for spatial weights and the weights were generated in the GeoDa software program (Anselin et al., 2006).
A Moran scatter plot and LISA significance map were created using GeoDa (Anselin et al., 2006), a free spatial statistics software. Global Moran's I, local Moran's I and LISA calculations were performed using the R software program (Ihaka & Gentleman, 1996) and a package called ‘spdep’ (Bivand, 2006).
Results and discussion
Changes over time
In the past few years, the rural water supply of China has had significant enhancements. Many rural water supply facilities have been built, and RPRTW rose from 82.9% in 1993 to 94.2% in 2011. Meanwhile, RPRWIB rose from 38.1% in 1993 to 72.1% in 2011 (Figure 1). The great progress made in rural water supply is due to China's huge investments in this field. The Government of China first included the target of improving rural drinking water beneficiaries in the country's Social and Economic Five-Year Plans in 1986, and there were significant funds allocated to support rural water supply infrastructure in each subsequent Five-Year Plan.
The RPRWIB of China varied among provinces from 1993 to 2011, and the distribution of rates for the first year and last year is shown (Figure 2). The western and eastern provinces have obvious differences in 1993 RPRWIB values. The rates of most western provinces were lower than 60%, and the rates of most central and eastern provinces were higher than 80%. Since great progress was made between 1993 and 2011, all the provincial data for 2011 are reported higher than 70% and data of most of provinces are approaching 100%. The increase in the value of the percentage rate at the national level was 34. The average increase in value of the percentage rate of the western region (28.92) was obviously higher than those of the eastern (5.09) and middle regions (8.82). For example, some of the increased rates for the western region were: Yunnan (from 49.8 to 88.1%), Qinghai (from 59.5 to 86.6%), Ningxia (from 65.6 to 96.3%) and Xinjiang (from 56.6 to 87.8%).
The provincial RPRTW witnessed more obvious changes from 1993 to 2011 (Figure 3). In 1993, only a few developed municipalities had rates higher than 80%, such as Beijing, Tianjin and Shanghai. As great progress was made in tap water supply, the rates of most of the provinces were higher than 50% as of the end of 2011. Nevertheless, the rates in the eastern region were still higher than in the western and middle regions. For example, the rates of eastern coastal regions such as Shandong, Zhejiang, Jiangsu and Shanghai were more than 90% while western inland areas such as Sichuan, Guangxi and Shaanxi had rates lower than 70%. There is a spatial aggregation possibility due to neighbouring provinces with high or low values. This is regarded as only a possibility because there is no statistical support.
Global spatial autocorrelation
The Moran's I statistics of RPRWIB and RPRTW were calculated for each year. The significance test for Moran's I showed that the two yearly indices have a low p value (p < 0.05). Figure 4 shows the Global Moran's I values of the two indices and indicates the degree of spatial autocorrelation, which reflects their general spatiotemporal trends. It can be seen that the Global Moran's I of RPRWIB is almost negative (Figure 4). On the other hand, the Global Moran's I of GDP per capita is positive in this period, showing the expanding spatial agglomeration phenomenon of China's economy (Meng et al., 2005).
By the last year of the ‘8th Five-Year’ period (1995), the Global Moran's I was 0.56, and it continues to decrease from 0.5456 in 1996 to 0.3409 in 2000 after the ‘9th Five-Year’ period (1996–2000). During the ‘10th Five-Year’ period (2000–2005), the Global Moran's I showed a dramatic downward trend from 0.3664 in 2000 to 0.1787 in 2005. There was also a downward trend in the ‘11th Five-Year’ period. The lines in Figure 4 show discontinuous change and the 5-year cycle. The figure also shows that the Global Moran's I of RPRTW had no significant change from 1993 to 2011. The positive Moran's I values show a positive spatial autocorrelation with a declining tendency.
Local spatial autocorrelation
The LISA cluster maps can help find the spatial phenomena characters that vary over time (Figures 5 and 6). The plots give LISA cluster maps for 1995, 2000, 2005 and 2010, which are the end years of each ‘Five-Year Plan’. In these plots, HH represents a High-High spatial pattern; HL signifies a High-Low spatial pattern; LH means a Low-High spatial pattern and LL is a Low-Low spatial pattern.
There is one agglomeration region in the west with a High-High spatial pattern and one agglomeration region in the east with a Low-Low spatial pattern in 1995 (Figure 5). This spatial agglomeration phenomenon shows the imbalance in rural water supply infrastructure in 1995. Although this phenomenon persists in 2000, 2005 and 2010, the agglomeration regions are changing. The Low-Low regions (in light grey/blue in online version at http://www.iwaponline.com/wp/toc.htm) are expanding and the High-High regions (in dark grey/red in online version) are narrowing. Some of the western provinces have been developing faster in rural water supply infrastructure such as Sichuan, Chongqing and Gansu, as shown in some of the HL regions. The western provinces, such as Xinjiang, Qinghai, Sichuan and Gansu, were defined as Low-Low spatial aggregation areas, and Beijing and Tianjin were designated High-High areas (Figure 6). Heilongjiang Province is in the Low-Low area, which indicates it has low RPRTW values and that its adjacent provinces also have low values. The spatial patterns of rates in 2000, 2005 and 2010 show similar agglomeration. The Low-Low regions have continued to move from the west to the middle. The Gansu, Shaanxi, Ningxia and Hubei provinces form a Low-Low agglomeration. Chongqing and Shanxi are labelled as a High-Low pattern with a higher tap water rate than neighbouring provinces. The extent of the High-High areas expanded and a new High-High area was formed including Shanghai and Jiangsu.
Causal analysis of spatial pattern
Major factors affecting water supply infrastructure spatial variations
Domestic sources account for 90% of water supply infrastructure investments in developing countries, and these are primarily from the public sector (Briscoe, 1999). In China, the government has become a leading investor in the rural water supply sector and there are also some private capital investments in this field. Participation of private capital investments in rural water supply depends on government support and marketization levels of rural water supply in China, which are related to economic growth. Research shows that the global-scale changes in population and economic development have much more influence on the water supply and its demand, compared to the effects of global climate change (Vörösmarty et al., 2000). We can conclude that spatial and temporal distributions of RPRTW and RPRWIB depend on the comprehensive effects of society, economy, politics and technology. This research focuses on the relationships between government investment and RPRTW, RPRWIB and GDP per capita.
Our research analyzes the movement of mean centres of RPRWIB and RPRTW, and the relationship between the movement and GDP per capita and PIPR, respectively (Figure 7). GDP per capita is one of the indicators reflecting regional difference in economic development, and PIPR is an indicator reflecting regional difference in investment. It can be seen from Figure 7 that the mean centres of the four indicators are approaching each other both in longitude and latitude but do not coincide. The mean centre of GDP per capita is east of that of RPRWIB and RPRTW, and the PIPR centre is west of the mean centre of RPRWIB and RPRTW. This indicates that the mean centres of RPRWIB and RPRTW are co-influenced by GDP per capita and PIPR.
GDP per capita
Water and economic and human development have been discussed in depth in recent research. This research indicates that investing in water infrastructure could lead to improved access to water and can support progress in sustainable economic growth (Kumar et al., 2008). China's rapid economic development is the most likely driving factor accounting for the temporal and spatial distributions of RPRTW and RPRWIB. Also, the rapid increases of RPRTW and RPRWIB were caused by rapid economic development. As shown in Figure 7 which analyzes the temporal movement trend of the four indicators, the extent of the mean centres of RPRWIB and RPRTW were similar and both centres tended to move toward the west. The tendency of RPRWIB and RPRTW corresponds with that of the Moran's I, which illustrates the weakening of regional difference between the east and the west (Figures 7(a) and 7(b)). Although the temporal movement of centres of GDP per capita is very slow, the general movement trend is to the west (Figure 7(c)). Other researchers have also revealed similar results. Strong patterns of both global and local spatial autocorrelation in regional economic income are found in the USA and in China (Rey & Montouri, 1993; Meng et al., 2005). Spatial analysis shows the mean centre of GDP per capita in China is moving from east to west, although LISA analyses show that the regional gap in GDP is expanding (Meng et al., 2005). This research shows that the mean centres of GDP per capita are moving in the same direction as the mean centres of RPRTW and RPRWIB. This suggests that the moving centre of GDP per capita drives the moving of the mean centres of RPRTW and RPRWIB.
On the other hand, rural drinking water safety improvement can also spur socio-economic development. Supplying safe drinking water can not only reduce disease, improve life quality and promote public health but can also reduce the consumption of medical resources, reduce environmental pollution, improve regional environment, narrow the economic gap between urban and rural, as well as promote regional economic and social development. Related research that analyzed the statistical data of 30 provinces in China revealed a positive correlation between RPRTW and GDP (Zhou et al., 2010). The regions with an RPRTW close to or above 90% at the end of 2007, such as Shanghai, Beijing, Jiangsu, Zhejiang and Tianjin, had GDPs per capita that were 30,000 RMB higher on average compared to study areas with lower RPRTW values. Shanghai had the highest RPRTW and the highest GDP per capita. In the provinces with RPRTW values less than 50%, GDP per capita was approximately 10,000 RMB, which is significantly less than the national average level. Although the above data cannot confirm a causal relationship between the development of safe drinking water and economic condition, there is a clear correlation.
Besides local economic factors, central investments are also a crucial factor. The Chinese central government has prioritized rural drinking water safety and consistently increases investment in this field. In Figure 7, the mean centre of PIPR is 3° west of that of RPRWIB and RPRTW. This implies that the west attracts more investment. The centre has remained in the west despite its movement towards the middle and the east. Because low-income regions are more reliant on outside funds from central governments and international organizations (Zhang, 2011a, b) and the central government tends to invest more in the western regions (Meng et al., 2005), the regional difference of RPRWIB and RPRTW in east and west has been decreasing gradually. Therefore, regional investment difference is the main reason for the imbalance of rural water supply, but investments from the central government are changing this fact.
In China, urbanization has progressed at a significant rate in the last 30 years (the proportion of the population living in urban areas was approximately 28% in 1993 and increased to ∼53% by 2012). The focus of our study is whether the spatial variability in urbanization at the provincial level has any effect on the two indices (RPRWIB and RPRTW). The provincial differentiation analysis of urbanization in China shows notable discrepancy in urbanization (Li, 2006), and urbanization level is highly correlated with GDP per capita (Li, 2007). In light of this, urbanization not only directly impacts water supply via heavy water demands (WHO & UNICEF, 2010) but also indirectly effects the relation between economy and water supply. A population shift away from rural areas would reduce demand on rural water supply coverage. ‘Unification of the urban and rural water supply’ in China also boosts the rural water supply coverage and improves water quality (Zhang et al., 2006). Hence, urbanization may have a positive effect on rural water supply in China.
Our study shows the spatial pattern of rural water supply infrastructure in China and reveals the main factors related to the pattern. The spatial distribution of rural water supply infrastructure coverage is imbalanced, but over the past 20 years, with the rapid development of rural water supply infrastructures in China, this imbalance is reducing. Although we do not deny the involvement of other factors, the spatial pattern of economic development and government investment has had a decisive role in the formation and evolution of rural water supply.
This research was supported by the Key Technology Study of Drinking Water Safety Testing, Monitoring, Risk evaluation, Alarming and Forecasting (No. 201302004) project, a part of China's Non-profit Health Sector Scientific Research Program of 2013.