Water demand predictions for megacities: system dynamics modeling and implications

A megacity is usually defined as a metropolitan area with a total population in excess of 10 million. As of 2016, there are 36 megacities in the world and 8 in the developing world. Megacities are increasingly a phenomenon of the developing world that will affect the future prosperity and stability of the entire world (Bugliarello, 1999); meanwhile, the developed world also needs to pay attention to this phenomenon because of globalization. Characteristics of megacities are their size, their complexity in terms of administration and infrastructure, and their environmental impact. Most megacities around the world are experiencing water shortages driven by population, land use, urbanization and industrial development. Water shortages and their associated impacts will extend well beyond the administrative boundaries of megacities because of cascading effects on regional water supplies and the profound influence that these cities have globally.

To manage the water shortages and ensure water sustainability in megacities, it is important and urgent that we determine the amount of water that a megacity needs over a given period of time in the future, based on which water conservation measures and policies can be developed to deal with the possible water shortages via optimal planning of investment for expanding and/or updating water supply systems and institutional reform for water demand management. Reasonable predictions of future water demands in megacities will play a decisive role in this planning.

Water demand predictions in megacities involve complex economic, social, technological and engineering factors, such as population, structure of the economy, residential water users' behaviors, water conservation technologies, and others. Interaction and feedback relationships generally exist between these factors. An integrated and systemic approach is needed to assess the interactions and feedback among the various factors. Reasonable water demand predictions cannot be obtained based on only one of these factors or a linear superposition of the results from individual analyses. These factors should be systematically considered in an integrated framework.

However, most water demand studies have either been fragmented or have not effectively captured the interactive and synergetic effects of the coupled socioeconomic and physical processes (Yang et al., 2016). The methods commonly used in water demand predictions include statistical methods (e.g. time series methods and structural analysis methods (Zhang et al., 2003; Msiza et al., 2008; Zhai et al., 2012; Tiwari & Adamowski, 2013) and systems simulation methods (He, 2009)). Statistical methods generally rely on historical data to establish relationships between the water demand and relevant factors (Shao et al., 2012). However, statistical methods often neglect the interactions and interconnectedness among various factors, thereby failing to capture the systematic behavior of water demands in megacities.

System simulation methods are supposed to overcome the deficiency of statistical methods. In this study, we develop a simulation model using the system dynamic (SD) approach. This choice is based on several justifications. First, multiple studies have proved that the SD approach can effectively address water shortage problems using both interactive and non-linear relationships in megacities (Winz et al., 2009; Mirchi et al., 2012). The complex factors associated with water demand predictions in megacities can be integrated into a holistic model to conduct water demand analysis of multiple sectors in a region (Sehlke & Jacobson, 2005; Qin et al., 2012). Second, the SD approach allows users to easily understand the relationships and feedback among the various factors associated with water demand predictions in megacities. Finally, the SD approach allows the incorporation of hydrologic, agronomic and behavioral factors and processes into the simulation of water demand predictions.

We develop a SD model in this study for water demand prediction in megacities. The Beijing municipality in China is selected as a case study to illustrate the functionality of the proposed SD model. Beijing is a typical megacity that has faced severe water shortages over the past several decades. The study on water demand in Beijing can provide an example for other megacities in the world to deal with existing or potential water shortage.

SD was first introduced by Forrester in the 1950s (Forrester, 1961). The method was originally used to analyze complex industrial processes, such as production management and inventory management (Qin et al., 2012). However, SD gradually became a modeling framework in systems theory and has been applied to various disciplines to understand the dynamic behaviors of diverse complex systems. SD models are causal mathematical models (Barlas, 1996) based on the fundamental premise that the structure of a system causes observable and predictable behavior. SD modeling starts with determining the structure of the system. This process consists of identifying the interactions and relationships among different components of the system. These relationships are represented in both qualitative (i.e. causal loop diagrams) and quantitative (i.e. stock-and-flow models) forms and are followed by mathematical formulation. Ultimately, scenario analyses are conducted to project the future state of the system. SD models are typically developed as user-friendly softwares with graphical user interfaces, such as PD-plus, VENSIM, STELLA and POWERSIM.

The water demand calculation is the main component of the methods used in this study. Total water demand (TWD) is the sum of domestic water demand (DWD), industrial water demand (IWD), agricultural water demand (AWD) and environmental water use (EWU). AWD can be further divided into irrigation water demand (IrriWD), livestock water demand (LWD) and forestry and fishery water demand (FFWD). FFWD and EWU are given as inputs while the other terms are intermediate variables that are calculated based on the methods presented in the following sections. The calculation methods involve socioeconomic, hydrologic and agronomic factors that affect the water demands.

WDI is a quantitative indicator determined by the total water supply and demand. It considers both the supply and demand aspects of the complex water utilization system. Larger values of WDI correspond to more serious water shortage problems in a region.

IWD is calculated based on several factors, such as the gross domestic product (GDP), income (GDP per capita, GDPC), water use technology improvement and industrial water price (P_IW). Improvements to water use technologies can promote water conservation in the industrial sector and decrease the water demand. Water price plays a modulatory role in the industrial sector. We estimate an overall IWD rather than specific water demands in different industrial sectors.

The city of Beijing is located on the northernmost portion of the North China Plain (NCP) (Figure 1). Specifically, the city is situated in the transition zone between the Mongolian Plateau and the NCP. The city is located within Hebei Province and neighbored by the city of Tianjin. Beijing is the center of China's political, economic and cultural development. The total population of the city in 2014 was approximately 21.5 million (population density was 1,311 persons/km² and urbanization rate was 86.3%). The transient population became an important driving factor in the increase in urbanization rate in Beijing. Agricultural land use accounted for about 67.5% of the total land area in 2014. In the past 30 years, the land use and land cover of Beijing have undergone huge change. A large area of agricultural land, such as cultivated land, forest land, grassland, wetland and other land cover types, has gradually become urban land, forming an obvious transition structure from the urban core zone, rural–urban fringe zone to suburb area zone. Since the 11th five-year plan, the energy development situation in Beijing has been good. In 2009, the energy consumption per 10⁴ RMB GDP fell by 3.8% on a year-on-year basis; the ratio of energy growth and economic growth is 1:2.65, supporting rapid economic development with a relatively low energy growth rate (Guo, 2011). However, there are four challenges in the energy development of Beijing. First, the general situation of clean energy supply remains constrained by the prominent supply–demand contradiction in natural gas and high-quality coal. Second, heating is facing restrictions from available resources, the environment and economy. Third, the difficulty of energy conservation and emissions' reduction increases with time. Fourth, regulating energy supply and utilization is harder than before because of the complex operating environment.

Beijing belongs to the warm, temperate and semi-humid continental monsoon climate zone. The long-term annual average rainfall is 603 mm (1956–2003); however, rainfall often exhibits significant inter-annual variations. Recently, Beijing experienced a severe drought event that lasted five years (1999–2003). Furthermore, the seasonal rainfall distribution is uneven, with approximately 60–80% of the rainfall occurring throughout the summer flood season (i.e. from June to September) (Sun et al., 2007). The surface water evaporation is 1,500–2,200 mm and has increased over time. As part of the NCP, Beijing relies heavily on groundwater, especially within the agricultural water consumption sector. The water deficit problem in Beijing is closely related to the dependence upon groundwater supply and the large AWD (Yang et al., 2016). Given a limited total water supply, an increasing water demand may worsen the water shortage problem.

Beijing has required additional water to support the expanding economic development, thereby worsening the water resources situation in the city. The ever-growing demand for water poses a major challenge to sustainable development in megacities such as Beijing. The annual average TWD in Beijing reached 3.91 × 10⁹ m³ during the period 1988–2012 (Figure 2), with large inter-annual variations during these years. However, the multi-year (1956–2000) average water availability (i.e. maximum water supply) of Beijing is 3.73 × 10⁹ m³ (Zhang et al., 2012). In individual years, the total water availability fluctuates and is often significantly less than the TWD. This situation results in groundwater overexploitation and the depletion of rivers and reservoir storage. Beijing currently faces a severe water shortage problem. The water resource availability per capita in Beijing is approximately 300 m³ per year, which is significantly less than national and world averages and is far below the internationally recognized water shortage standard (i.e. 1,000 m³ per capita) (Zhang, 2002). Agricultural and domestic demands accounted for 45% and 31% of water consumption in 2001 and 26% and 44% of consumption in 2012, respectively.

Since 1980, the water price has been used as an economic tool to adjust the balance between water supply and demand in Beijing. The water price in Beijing includes water resource fees, water engineering fees and sewage disposal fees. Beijing has implemented nine water price adjustments since 1992. Subsequently, the water prices for different industries have increased between threefold and eightfold (Shen et al., 2009). The domestic water price is 4 RMB/m³, while the non-domestic water price ranges from 5.8 RMB/m³ for administrative purposes to 81.68 RMB/m³ in the leisure spa industry. The Beijing Government has announced that it will gradually increase the domestic water price and reasonably adjust the urban domestic water price. In addition, the government will enforce over-utilization surcharges associated with water use in the industry and service sectors and will continue to strictly implement different water price policies in high water-consumption industries to promote adjustments within the industrial and water use structures.

Figure 3 illustrates the causal loop diagram (CLD) (i.e. conceptual model) of the water demand simulation model in Beijing. This diagram provides valuable system information, including feedback loops, loop dominance and time delays (Mirchi et al., 2012). CLD development is the first step in developing the SD model. This step identifies the elements of the model and the causal relationships among them. The water demand component includes DWD, IWD, AWD (IrriWD, LWD and FFWD) and EWU. Furthermore, the water supply component accounts for groundwater, surface water, water from the South-to-North Water Diversion Project (SNWDP) (a multi-decade infrastructure mega-project in China, which ultimately seeks to channel 4.48 × 10¹⁰ m³ of fresh water annually from the Yangtze River in southern China to the more arid and industrialized north using three canal systems: the eastern route, central route, and western route; Feng et al., 2007) and reused water. According to different water consumption sectors and environmental and supply principles, the SD model is divided into five sub-systems: socioeconomic (population growth), agricultural, industrial, water reuse, and water resources. The name of each sub-system describes the main component of that sub-system. The main socioeconomic, technological, hydrological and agronomic factors that influence each sub-system are considered and presented in the CLD. Each sub-system is also interrelated and mutually influenced by other sub- systems (Qin et al., 2012). The WDI plays an important role in the connections between sub-systems by reinforcing (positive) effects or balancing (negative) effects associated with stable sub-system variables. The CLD demonstrates the qualitative relationships among various factors and serves as the framework for the quantitative construction of the model.

The second step in SD model construction develops a stock-flow diagram (SFD) based on the CLD. The SFD describes the quantitative relationships among variables using equations and input data (Figure 4) and represents the system in terms of stocks and flows. Stocks (levels) are calculated at one specific time and represent any variable that accumulates or depletes over time. Flows (rates) are calculated over an interval of time and indicate activities or variables that cause the stock to change (Mirchi et al., 2012). The SFD can characterize the system processes graphically using a series of quantitative relationships and the associated equations, providing the foundation for model calibration, validation and prediction under different scenarios.

Socioeconomic (population growth) sub-system. The state variables in this sub-system include the total population and the DWD. The natural population growth rate has an impact on the total population (Figure 4(a)), while the DWD is calculated using Equation (1), which considers various socioeconomic factors that affect the DWD. The population serves as a driver of the DWD. The more rapid the population growth is, the higher the DWD is.

Industrial sub-system. The state variable in this sub-system is GDP, which is determined by the GDP growth rate and the WDI (Figure 4(b)). The IWD intensity is calculated using Equation (2), which considers the economic and technological factors that affect the IWD. GDP is the main driver exerting a positive effect on the IWD.

Agricultural sub-system. Four LWDs and the IrriWDs of 12 crops are the main components of this sub-system (Figures 4(c) and 4(d)). The FFWD is treated as an input variable in this model. The IrriWD is calculated using Equation (3), which considers the hydrological and agronomic factors affecting the IrriWD. The LWDs are calculated using the water quota method in Equation (4). The drivers of IrriWD include the crop characteristics, irrigation areas and meteorological conditions, while the drivers of LWDs include the livestock quantities and associated quotas.

Water reuse sub-system. This sub-system considers wastewater reuse (Figure 4(e)). Wastewater is generated by domestic and industrial activities. However, these wastewaters have been increasingly reused as a result of technological improvements, including water treatment and reuse technologies. The reused wastewaters are added to the water supply sector and help to reduce the water shortages in megacities.

Water resources sub-system. The water supply sector in this sub-system includes the groundwater supply, surface water supply and water supply from SNWDP. The water demand sector includes the aforementioned water demands as well as EWU, which is treated as an input variable in this SD model (Figure 4(f)). The balance of the water supply and demand determines the sustainability of water resources in megacities. In return, the WDI affects the water demands. This impact is generally negative.

Three types of input data are used in the simulation model: table functions, constant values, and initial values. A table function is a time series variable used to express the non-linear relationship between variables in a model, which cannot be expressed in the form of an equation. A constant value does not change over time or within a time interval. An initial value refers to the initial value of a level variable, which reflects the cumulative effects of that variable. The input data used in the model are listed in Table 1, while the constant values used in the model are listed in Table 2, excluding the crop coefficients (kc; listed in appendix Table A1, available with the online version of this paper). These constant values were obtained using calibration processes. They fall within a reasonable range based on literature values.

Future water demand predictions were conducted for the period spanning 2012 to 2030 at a one-year time step. Data from the period 2000–2011 are used to calibrate the model. The goal of the calibration process is to provide a satisfactory match between simulation results and historical observations. Four different water demand variables in the model are used in the verification process. Historical annual water demands data were collected from the Beijing Water Resources Bulletin, which is released yearly by the Beijing Water Authority (BWA, 2001–2012).

Scenarios can be used to assess the rationalities and uncertainties associated with potential events during a given time period, which is often a future period. A scenario analysis provides an interesting and practical method for comparing the future states of a system based on different projections. A scenario analysis method is used in this study to select a set of scenarios that can help to solve or diminish the water scarcity problems in Beijing. Such an analysis can potentially be used by policy makers to design sustainable policies for the future.

Ten scenarios are designed to predict the future water demand of Beijing under different climatic and socioeconomic conditions (Table 3). These scenarios include business-as-usual (BAU) scenario, climate change (CC) scenario, high economic development (HED) scenario, SNWDP scenario, livestock water conservation (LWC) scenario and combined scenario. The climate change scenario includes three scenarios that represent wet climate (CC_wet), normal climate (CC_normal) and dry climate (CC_dry). Additionally, the combined scenario includes three scenarios. The HED_SNWDP scenario considers both economic development and SNWDP. Additionally, the HED_LWC scenario considers both economic development and LWC. Finally, the HED_SNWDP_LWC scenario considers economic development, SNWDP and LWC simultaneously.

The BAU scenario, which serves as our baseline, assumes that the system structure and development policies do not markedly vary during the prediction period, maintaining the present development trends into the future. Historical climate data from 2001 to 2011 are used to generate a 22-year time series and to represent climate conditions from 2012 to 2030 (a total of 19 years). Therefore, we obtain the climate data required for this scenario and other non-climate-change scenarios. According to the real-world scenario and the authors' knowledge, the economic development rate will be 8% in 2020 and 10% in 2030. Table 4 provides a summary of the table function parameters used in the BAU scenario.

Appendix Table A3 (available with the online version of this paper) shows the annual rainfall and reference ET values for the prediction period under three different climate change scenarios.

The HED scenario is designed based on the BAU scenario (Table 3). However, it differs in assuming higher economic development based on development rates (GDP growth rate) of 15% and 20% in 2020 and 2030, respectively. The other parameters are similar to those of the BAU scenario.

The SNWDP scenario considers the impact of the SNWDP on the water resources balance and the water demand projection of Beijing. The water from the SNWDP is considered a surface water supply in the SD model. This supply will increase the total water supply to Beijing and potentially alleviate or solve water shortage problems in the future. According to the SNWDP plan, water was first supplied to Beijing from 27 December 2014 at a diversion rate of 12 × 10⁸ m³ per year. This diversion process is implemented in the model using ‘IF THEN ELSE’ statement. The model structure and the other parameters are similar to those of the BAU scenario.

The LWC scenario attempts to save water associated with the livestock sector during the prediction period. Because the LWD is calculated using the water use quota method, reducing the water use quota directly would likely provide a practical water conservation strategy. Specifically, this scenario reduces the water use quota of all four livestock groups by 30%. The other parameters are similar to those of the BAU scenario.

Three scenarios (HED_SNWDP, HED_LWC and HED_SNWDP_LWC) are included in this scenario, which considers both economic development and the water diversion project (Table 3). Each combined scenario is the combination of two or three scenarios; thus, the parameters of these scenarios are used in this combined scenario.

This section provides the results of the water demand simulations for the period of 2012–2030 under different climatic, socioeconomic and development scenarios. Comparing the results under different scenarios can provide useful insights regarding Beijing's social and water resources development in the future. This comparison can benefit local governments and policy makers.

Figure 5 compares the simulation model results and historical data. Overall, the DWD, IWD, AWD and TWD simulations all adequately reflect historical trends. The domestic and industrial water demands generally agree with the historical data, while discrepancies exist between the agricultural and total water demands and the associated historical data. These discrepancies are mainly due to the IrriWD simulation. In reality, farmers may use substantially more water for irrigation than what is simulated in this model. Overwatering is a common practice of many farmers, as they believe that increasing irrigation will increase crop yields. This relationship is difficult to simulate because the model cannot accurately reflect or simulate the irrigation behaviors of all the farmers in a megacity such as Beijing.

Table 5 shows the total water demand, total water supply and water deficit values in 2012, 2020 and 2030, respectively, under different scenarios. Due to the increased reuse of wastewater, the total water supply continually increases in all scenarios, including those without SNWDP and those with dry climates. By 2030, the TWD is likely to increase by at least 36.1% (up to 62.5%) compared with the historical TWD in 2011. However, different prediction scenarios suggest different reasons for these increases. The TWDs increase dramatically under climate change scenarios due to vastly increased IrriWDs. Population growth and industrial development also contribute to demand increases in scenarios related to high economic development, water conservation and the water diversion project. Scenarios that consider economic development require more water than the BAU scenario. CC_dry and LWC exhibit the largest and smallest TWDs, respectively, in 2030. The results imply that further economic development and climate change would increase the water demand and water deficit. If economic development continues to be the top priority in Beijing (scenario HED), water users will require 5.18 × 10⁹ m³ of water in 2030; however, the water demand will be approximately 5 × 10⁹ m³ of water in 2030 if economic development, water conservation and water diversion are all considered in the future (scenario HED_ SNWDP_LWC). Therefore, water deficit problems will continue to exist for all scenarios except SNWDP, HED_SNWDP and HED_SNWDP_LWC (Table 5).

Figure 6 shows the maximum, minimum and mean values of TWD under all scenarios. The maximum value is approximately 6 × 10⁹ m³, while the minimum value is approximately 3.7 × 10⁹ m³. The values vary widely based on the scenario. The scenarios that consider economic development exhibit the largest maximum TWDs, while the scenarios that include water diversion or water conservation exhibit the smallest maximum TWDs.

The water demands predicted by our model are comparable to predictions in the literature (Table 6). Most of these studies conducted water demand predictions under different scenarios, which allow them to be compared with our study. Although these models adopted different methods and prediction periods, they obtained similar prediction results given these inherent differences. For the year 2020, the TWD predicted by Zhai et al. (2012) and Huang et al. (2009) are 34.24–46.53 × 10⁸m³ and 28.38–52.83 × 10⁸m³, respectively, which are similar to the prediction results of this study (45.89–50.89 × 10⁸m³), while the result of Cheng et al. (2013) (36.32 × 10⁸m³) is smaller and Fan et al. (2006) (48.72–56.41 × 10⁸ m³) is larger than those of this study. As the method, model structure, assumptions and designed scenarios are very different among these studies, the prediction results cannot be exactly the same. This comparison demonstrates that the SD methodology and model developed in this study can be used to predict water demands in Beijing. Note that not only macroeconomic factors but also hydrological and agronomic factors are included in water demand simulations in our model, while other studies mainly considered macroeconomic factors alone. This is the biggest difference between our study and other studies. This is also the reason why there are studies that have predicted larger or smaller water demands than ours. However, the addition of non-economic factors likely reduces the uncertainty of the model and provides more realistic water demand predictions, as more realistic factors that affect the water demands are considered and modeled in this study.

WDI is an indicator of the degree of water shortage in a system. A positive WDI value indicates that a water shortage problem exists in that system. Figure 7 illustrates the WDI of Beijing during the prediction period. For scenarios BAU, CC_wet, CC_normal, CC_dry, HED, LWC and HED_LWC, the WDI values are always positive, suggesting that water shortage problems exist throughout the entire prediction period. The other three scenarios (i.e. scenarios including SNWDP) include WDI values of zero in many years, implying that no water shortage problems exist in those years. Although the water demand will increase due to societal development, the water diversion project can potentially solve water shortage problems by increasing the water supply in Beijing. However, water shortages appear towards the end of the prediction period under these scenarios due to societal development and crop irrigation. Figure 7 shows that water conservation in the livestock sector can partially help to solve water shortage problems in Beijing. Conversely, climate change may worsen the water shortage issue, which may require proper adaptation strategies. Predictably, drier climates resulted in larger WDI values and more serious water shortage problems.

Figure 8 shows the variations in water demand shares in each sector compared with the TWD under different scenarios. The shares of different sectors under different scenarios do not exhibit substantial variations. The AWD percentages are relatively constant and exhibit decreasing trends under all scenarios. This is due to the stable IrriWD. The DWD percentages also display decreasing trends under all scenarios. The IWD and EWU percentages both show increasing trends during the prediction period because economic development and environmental protection are considered very important in Beijing. Among different sectors, DWD contributes the most to TWD. This is mainly because of the dense population and the high urbanization rate in the city.

Wastewater drainage is a water quality variable in this study that indicates the water environment and water recycling of the study site. The reuse of wastewater has a positive effect on the water utilization of Beijing. Figure 9 shows the average volume of wastewater drainage and wastewater reuse for all ten scenarios. From this figure we can see that the wastewater drainage increases with the development of the economy, which means that the water environment becomes worse. On average, the wastewater reuse rate is about 0.5 for all scenarios. This indicates that there is huge potential for Beijing to increase the wastewater reuse rate through the advances in technology, in order to increase the water supply and hence to alleviate the water deficit problems. Meanwhile, the water quality will be better with more wastewater being reused.

According to the simulation results, scenarios HED_SNWDP and HED_ SNWDP_LWC represent the best options for alleviating water shortage problems in Beijing. Scenario SNWDP can also effectively solve water shortage problems by increasing the water supply through the water diversion project. Climate change may largely impact the water balance in the future due to the direct influence on the water supply and AWD.

As demonstrated by the modeling scenarios, the SD model can be used as a tool for ‘what-if’ analysis under the various conditions in the future, and thus provide recommendations for effective solutions to address existing or potential water shortage problems. The model can be used as a communication tool between water resources modelers and managers or planners to estimate the range of water demand in the future under both natural and socioeconomic conditions and explore strategies to mitigate water shortage risk in a megacity. The managers' and planners' knowledge and justification are often helpful to quantify model parameters, justify the results, and provide feedback to modify the model.

The SD model can also be used together with other models for more detailed and comprehensive analysis. For example, the IWR-MAIN (Institute for Water Resources-Municipal And Industrial Needs) model is a decision-support package for projecting municipal and industrial water demands at a finer spatial and temporal scale (Sellers & Hatcher, 1991; Mohamed & Al-Mualla, 2010a, 2010b). However, IWR-MAIN does not relate water demand to hydroclimatic, socioeconomic and technological factors. Using both the SD model proposed in this paper and IWR-MAIN can provide complementary information for city managers and planners.

In addition, the SD software can provide graphical user interfaces for model users to view model inputs and outputs under a modeling scenario, conduct sensitivity analysis of model parameters, and design multiple scenarios for ‘what-if’ analysis. This feature makes it convenient for modeler–user communications.

There were a total of 36 megacities in the world by 2016. Many megacities face water shortage threats due to high population growth and rapid economic development. The water demand prediction model based on systems dynamics may also be extended to other megacities for water demand prediction and water shortage assessment for the following reasons. First, the water demand calculation methods (Equations (1)–(5)) take into account the various hydroclimatic, agronomic, socioeconomic and technological factors and the interactions among these factors to calculate domestic, industrial, irrigation and livestock water demands. The methods are general and can be used for other megacities as long as the data are available for the calibration of the parameters.

Second, the SD model components and the connection between them (i.e. the model structure) developed for Beijing (Figure 4) can also be transformed to other megacities though the model components may need some modification according to data availability and specific water use conditions (e.g. some cities have negligible irrigation water use; some have large water use for recreation, etc.).

Additionally, the scenario design and analysis procedures for Beijing can be followed when the model is applied to other megacities. The SD model provides flexible interfaces to account for any change and impact of the various factors.

Finally, the trend of water demand change in Beijing that is identified from the scenario analysis in this study might occur in other megacities and thus the implications for Beijing can be informative for other cities. However, it should be noted that the SD model designed is empirical because the relationship and parameter quantifications more or less involve our knowledge, estimation and justification based on our long-term experiences of water resources management in the NCP and Beijing. When the SD model is extended to another city, the experiences of local modelers and water resources managers will also be important for the successful application of the model. Further extension and improvement of the model include using additional water use observations to develop more appropriate empirical relationships and quantify parameters, assessing the impact of new technologies such as those for water treatment, recycling and water storage (e.g. building a ‘sponge’ city; Yu et al., 2015) and considering the nexus relationship (trade-off and synergy) between the water and energy sectors.

This study adopts an integrated and systematic approach that takes into account many socioeconomic, agronomic, hydrological and technological factors and their feedback relationships associated with water demands in megacities in a consistent predictive model. The water demand calculations are undertaken for different water use sectors, including domestic, industrial and agricultural sectors. Beijing was selected as a case study to demonstrate the developed SD approach. Data from 2001 to 2011 were used to calibrate the model. Ten scenarios are designed to predict annual water demands from 2012 to 2030 using the calibrated model, considering economic development, climate change, inter-basin water transfer and water conservation technology. The predicted water demands by our model are comparable to those in the literature. The TWD is likely to increase by at least 36.1% (up to 62.5%) from 2011 to 2030. The corresponding water deficits range from 0.80 to 1.80 km³. However, several scenarios end with water surplus ranging from 0.26 to 0.36 km³ for the year 2030 (Table 5). Climate change may largely impact the water balance in Beijing, as indicated by the large water demands and water deficits under the three climate change scenarios. The inter-basin water transfer project (SNWDP) and water conservation technology play an important role in alleviating water shortage in the future, and three scenarios, HED_SNWDP, HED_SNWDP_LWC and SNWDP, are recommended for implementation. Beyond the case study of Beijing demonstrated in this paper, the proposed SD model is applicable to other megacities because of its generality in water demand calculation methods by sector and scenario design and analysis procedures. The model results are useful for city planners for infrastructure and technology investment and water demand management policy reform considering the various hydroclimatic and socioeconomic conditions.

This work was supported by the National Natural Science Foundation of China (Grants No. 41330632 and No. 91225301) and China Postdoctoral Science Foundation (Grant No. 2015M580913). Additional support was provided by the State Key Laboratory Breeding Base of Nuclear Resources and Environment (Grant No. NRE1516), and the Doctoral Start-up Fund of the East China Institute of Technology (Grant No. DHBK2016104).