Coupling and Coordination Degrees of the Core Water–Energy–Food Nexus in China
Abstract
:1. Introduction
2. Methods
2.1. Evaluation Index System for Measuring Coupling and Coordination Degrees of the Core Water–Energy–Food (WEF) Nexus
2.2. Coupling and Coordination Degrees of the Core WEF Nexus
2.2.1. Weighting Index
- Index selection—suppose there were subsystems, years, regions, and indicators, then was the th indicator of th region of the th subsystem in the th year.
- Data standardization—because the indicators had different dimensions and units, they needed to be standardized. At the same time, in order to avoid meaningless logarithms in entropy calculations, non-zero processing was carried out on the data.If is a positive indicator:If is a negative indicator:
- The proportion of the th index is indicated by:
- The entropy value of the th index is indicated by:
- The information utility value of the index is given by:
- The weight of the th index is given by:
2.2.2. Coupling Degree of the Core WEF Nexus
2.2.3. Coordination Degree of the Core WEF Nexus
2.3. Study Area and Data
3. Results and Discussion
3.1. The Development Level of the Core WEF Nexus
3.2. The Coupling Degree of the Core WEF Nexus
3.3. The Coordination Degree of the Core WEF Nexus
4. Conclusions
Author Contributions
Funding
Conflicts of Interest
References
- World Economic Forum. Global Risks 2011 Report, 6th ed.; World Economic Forum: Cologne, Germany, 2011. [Google Scholar]
- Hoff, H. Understanding the Nexus. In Proceedings of the Bonn 2011 Nexus Conference: The Water, Energy and Food Security Nexus, Bonn, Germany, 16–18 November 2011; Stockholm Environment Institute: Stockholm, Sweden, 2011. [Google Scholar]
- Uen, T.S.; Chang, F.J.; Zhou, Y.; Tsai, W.P. Exploring synergetic benefits of water-food-energy nexus through multi-objective reservoir optimization schemes. Sci. Total Environ. 2018, 633, 341–351. [Google Scholar] [CrossRef]
- Yang, H.; Wang, L.; Abbaspour, K.C.; Zehnder, A.J.B. Virtual water trade: An assessment of water use efficiency in the international food trade. Hydrol. Earth Syst. Sci. 2006, 10, 443–454. [Google Scholar] [CrossRef]
- Li, G.J.; Huang, D.H.; Li, Y.L. Water–Energy-Food Nexus (WEF-Nexus): New Perspective on Regional Sustainable Development. J. Cent. Univ. Financ. Econ. 2016, 12, 76–90. [Google Scholar]
- Wicaksono, A.; Jeong, G.; Kang, D. Water, energy, and food nexus: review of global implementation and simulation model development. Water Policy 2017, 19, 440–462. [Google Scholar] [CrossRef] [Green Version]
- O’Brien, E. Biological Relativity to Water-Energy Dynamics. J. Biogeogr. 2010, 33, 1868–1888. [Google Scholar] [CrossRef]
- Degefu, D.M.; Weijun, H.; Zaiyi, L.; Zhengwei, H.; Min, A. Mapping Monthly Water Scarcity in Global Transboundary Basins at Country-Basin Mesh Based Spatial Resolution. Sci. Rep. 2018, 8, 2144. [Google Scholar] [CrossRef] [Green Version]
- Hardberger, A.; Stillwell, A.S.; King, C.W.; Webber, M.; Duncan, I.J. Energy-Water Nexus in Texas. Soc. Sci. Electr. Publ. 2011, 16, 209–225. [Google Scholar] [CrossRef]
- Siddiqi, A.; Anadon, L.D. Quantifying the Water-Energy Nexus in Middle East and North Africa. Energy Policy 2011, 8, 4529–4540. [Google Scholar] [CrossRef]
- Perrone, D.; Murphy, J.; Hornberger, G.M. Gaining Perspective on the Water-Energy Nexus at the Community Scale. Environ. Sci. Technol. 2011, 10, 4228–4234. [Google Scholar] [CrossRef] [PubMed]
- Kenway, S.J.; Lant, P.A.; Priestley, A.; Daniels, P. The connection between water and energy in cities: A review. Water Sci. Technol. 2011, 9, 1983–1990. [Google Scholar] [CrossRef]
- Hardy, L.; Garrido, A.; Juana, L. Evaluation of Spain’s water-energy nexus. Int. J. Water Resour. Dev. 2012, 1, 151–170. [Google Scholar] [CrossRef]
- Benzie, M.; Davis, M.; Hoff, H. Climate Change, Water and Energy in the MENA Region: Why a “Nexus” Approach is Crucial for Mitigation and Adaptation. Available online: https://www.sei.org/publications/climate-change-water-and-energy-in-the-mena-region-why-a-nexus-approach-is-crucial-for-mitigation-and-adaptation/ (accessed on 11 May 2019).
- Yang, H.; Zehnder, A.J. Water Scarcity and Food Import: A Case Study for Southern Mediterranean Countries. World Dev. 2002, 8, 1413–1430. [Google Scholar] [CrossRef]
- Rosegrant, M.W.; Cai, X.M.; Cline, S.A. World Water and Food to 2025: Dealing with Scarcity. Available online: https://reliefweb.int/sites/reliefweb.int/files/resources/1B0BBA6D1080C010C1256C6E002E1D17-ifpri-water2025-16oct.pdf (accessed on 11 May 2019).
- Kirby, R.M.; Bartram, J.; Carr, R. Water in food production and processing: Quantity and quality concerns. Food Control 2003, 5, 283–299. [Google Scholar] [CrossRef]
- Konar, M.; Dalin, C.; Suweis, S.; Hanasaki, N.; Rinaldo, A.; Rodriguez, I. Water for food: The global virtual water trade network. Water Resour. Res. 2011, 5, 143–158. [Google Scholar] [CrossRef]
- FAO. Agriculture, Food and Water; Edward Elgar: Cheltenham, UK, 2014. [Google Scholar]
- Grafton, Q. Food and Water Gaps to 2050. In Proceedings of the AGU Fall Meeting, San Francisco, CA, USA, 15–19 December 2014. [Google Scholar]
- Pereira, L.S. Water, Agriculture and Food: Challenges and Issues. Water Resour. Manag. 2017, 10, 2985–2999. [Google Scholar] [CrossRef]
- Haber, W. Energy, food, and land—The ecological traps of humankind. Environ. Sci. Pollut. Res. Int. 2007, 6, 359. [Google Scholar] [CrossRef]
- Selya, R.M. China’s Past, China’s Future: Energy, Food, Environment. Prof. Geogr. 2010, 3, 492–493. [Google Scholar] [CrossRef]
- Bogdanski, A. Integrated food-energy systems for climate-smart agriculture. Agric. Food Secur. 2012, 1, 9. [Google Scholar] [CrossRef]
- Yuan, L.; He, W.; Liao, Z.; Degefu, D.M.; An, M.; Zhang, Z.; Wu, X. Allocating water in the Mekong river basin during the dry season. Water 2019, 11, 400. [Google Scholar] [CrossRef]
- Owen, A.; Scott, K.; Barrett, J. Identifying critical supply chains and final products: An input-output approach to exploring the energy-water-food nexus. Appl. Energy 2017, S0306261917313466. [Google Scholar] [CrossRef]
- Lee, S.H.; Choi, J.Y.; Yoo, S.H.; Mohtar, R.H. Water footprint for Korean rice products and virtual water trade in a water-energy-food nexus. Water Int. 2018, 43, 871–886. [Google Scholar] [CrossRef]
- Hellegers, P.; Zilberman, D.; Steduto, P.; Priscoli, J.D. Interactions between Water, Energy, Food and Environment: Evolving Perspectives and Policy Issues. Water Policy 2018, 10, 1. [Google Scholar] [CrossRef]
- Flammini, A.; Puri, M.; Pluschke, L.; Dubois, O. Walking the nexus talk: Assessing the water-energy-food nexus in the context of the sustainable energy for all initiative. J. Phys. Chem. A 2014, 115, 7869–7870. [Google Scholar] [CrossRef]
- Conway, D.; Garderen EA, V.; Deryng, D.; Dorling, S.; Krueger, T.; Landman, W.; Lankford, B.; Lebek, K.; Osborn, T.; Ringler, C. Climate and southern Africa’s water-energy-food nexus. LSE Res. Online Doc. Econ. 2015, 5, 837–846. [Google Scholar] [CrossRef]
- Martinez-Hernandez, E.; Leach, M.; Yang, A. Understanding water-energy-food and ecosystem interactions using the nexus simulation tool NexSym. Appl. Energy 2017, 206, 1009–1021. [Google Scholar] [CrossRef]
- Leung Pah Hang, M.Y.; Martinez-Hernandez, E.; Leach, M.; Yang, A. Insight-based approach for the design of integrated local food-energy-water systems. Environ. Sci. Technol. 2017, 51, 8643–8653. [Google Scholar] [CrossRef]
- White, D.J.; Hubacek, K.; Feng, K.; Sun, L.; Meng, B. The Water-Energy-Food Nexus in East Asia: A tele-connected value chain analysis using inter-regional input-output analysis. Appl. Energy 2018, 210, 550–567. [Google Scholar] [CrossRef]
- Karnib, A. A quantitative assessment framework for water, energy and food nexus. Comput. Water Energy Environ. Eng. 2017, 6, 11–23. [Google Scholar] [CrossRef]
- Garcia, D.J.; You, F. The water-energy-food nexus and process systems engineering: A new focus. Comput. Chem. Eng. 2016, 91, 49–67. [Google Scholar] [CrossRef]
- Degefu, D.M.; Weijun, H.; Liang, Y.; Jian, H.Z. Water Allocation in Transboundary River Basins under Water Scarcity: A Cooperative Bargaining Approach. Water Resour. Manag. 2016, 30, 4451–4466. [Google Scholar] [CrossRef]
- Bazilian, M.; Rogner, H.; Howells, M.; Hermann, S.; Arent, D.; Gielen, D.; Steduto, P.; Mueller, A.; Komor, P.; Tol, R.S.J. Considering the energy, water and food nexus: Towards an integrated modelling approach. Energy Policy 2011, 39, 7896–7906. [Google Scholar] [CrossRef]
- Yaobin, L.; Rendong, L.; Xuefeng, S. Analysis of coupling degrees of urbanization and ecological environment in China. J. Nat. Resour. 2005, 1, 105–112. [Google Scholar]
- Sui, Z.; Mei, Z.; Guotai, C. The scinence and technology evaluation model based on tntropy weight and empirical research during the 10th five-Year of China. Chin. J. Manag. 2010, 7, 34–42. [Google Scholar]
- Chongbin, L. Quantitative judgement and classification system for coordinated development of environment and economy—A case study of the city group in the Pearl River Delta. Trop. Geogr. 1999, 2, 171–177. [Google Scholar] [CrossRef]
- Fengying, H.; Yi, Z.; Zhenglong, Z. Interaction study between tourism industry and regional economic of Guangxi based on the coupling coordinative degree. J. Cent. China Normal Univ. 2015, 4, 640–646. [Google Scholar] [CrossRef]
- Cheng, Z.; Chuan, J.; Biao, Z.; Feng, Z. The provincial difference of coupling coordinative development of regional economy-ecology-tourism. J. Arid Land Resour. Environ. 2016, 7, 203–208. [Google Scholar]
- Li, Q.; Zhongwei, C.; Renqiao, X. China regional industrialization, urbanization and agricultural modernization coupling coordination degree and influencing factors of research. Inq. Econ. Iss. 2012, 11, 10–17. [Google Scholar]
- Haifeng, X. Research on the coordinated development of tourism industry and regional economy from the perspective of system coupling—A case study of Zhejiang Province. J. Commer. Econ. 2019, 1, 166–168. [Google Scholar]
- Guijun, L.; Yulong, L.; Xiaojing, J.; Lei, D.; Daohan, H. Establishment and simulation study of system dynamic model on sustainable development of Water-Energy-Food nexus in Beijing. Manag. Rev. 2016, 10, 11–26. [Google Scholar] [CrossRef]
- Shaoming, P.; Xiaokang, Z.; Yu, W.; Guiqin, J. Study on water-energy-food collaboration optimization for Yellow River basin. Adv. Water Sci. 2017, 5, 681–690. [Google Scholar] [CrossRef]
- Junchao, J.; Liangshi, Z.; Defeng, Z.; Zheng, Z. Analysis of food security and its provincial contribution in mainland China from 1998 to 2016. J. Liaoning Norm. Univ. 2018, 3, 395–402. [Google Scholar]
- Ting, X.; Qi, A. Analysis of energy supply and consumption pattern and energy efficiency in China. Res. Appr. 2018, 7, 40–44. [Google Scholar]
- BP Statistical Review of World Energy (67th edition). Available online: https://www.bp.com/en/global/corporate/energy-economics/statistical-review-of-world-energy.html (accessed on 12 February 2019).
- Liming, H. China’s grain import and risk prevention of grain industry chain. J. Xihua Univ. 2019, 1, 90–97. [Google Scholar] [CrossRef]
- State Statistical Bureau. Available online: http://www.stats.gov.cn/tjsj/ndsj/ (accessed on 12 October 2018).
- National Bureau of Statistics Beijing Survey Team Website. Available online: http://www.bjstats.gov.cn/tjsj/ (accessed on 12 October 2018).
- Tianjin Statistical Yearbook—Website of Tianjin Bureau of Statistics. Available online: http://stats.tj.gov.cn/Category_29/Index.aspx (accessed on 12 October 2018).
- Hebei Provincial Bureau of Statistics. Available online: http://www.hetj.gov.cn/hetj/tjsj/jjnj/ (accessed on 12 October 2018).
- Data Release—Tibet Statistical Survey Information Network. Available online: http://www.xzqtjdcw.gov.cn/sjfb/index.jhtml (accessed on 12 October 2018).
- Heilongjiang Provincial Bureau of Statistics. Available online: http://www.hlj.stats.gov.cn/tjsj/ (accessed on 12 October 2018).
- Statistical Yearbook. Available online: http://www.stats-sx.gov.cn/tjsj/tjnj/ (accessed on 12 October 2018).
- Annual Data—Data Query—Inner Mongolia Autonomous Region Statistics Bureau. Available online: http://www.nmgtj.gov.cn/channel/nmg_tjj/col10471f.html (accessed on 12 October 2018).
- Liaoning Provincial Bureau of Statistics. Available online: http://www.ln.stats.gov.cn/tjsj/sjcx/ndsj/ (accessed on 12 October 2018).
- Jilin Provincial Statistics Bureau—Statistical Yearbook. Available online: http://tjj.jl.gov.cn/tjsj/tjnj/ (accessed on 12 October 2018).
- 2018—Statistical Yearbook—Shanghai Statistics. Available online: http://www.stats-sh.gov.cn/html/sjfb/201901/1003014.html (accessed on 12 October 2018).
- Jiangsu Statistics Bureau—Jiangsu Statistical Yearbook. Available online: http://tj.jiangsu.gov.cn/col/col4009/index.html (accessed on 12 October 2018).
- Zhejiang Statistical Information Network. Available online: http://tjj.zj.gov.cn/col/col1525563/index.html (accessed on 12 October 2018).
- Anhui Province Bureau of Statistics (NBS). Available online: http://www.ahtjj.gov.cn/tjjweb/web/tjnj_view.jsp?strColId=13787135717978521&_index=1 (accessed on 12 October 2018).
- Annual Data—Information Disclosure—Fujian Provincial Bureau of Statistics. Available online: http://tjj.fujian.gov.cn/xxgk/ndsj/ (accessed on 12 October 2018).
- Jiangxi Provincial Bureau of Statistics. Available online: http://www.jxstj.gov.cn/ (accessed on 12 October 2018).
- Annual Data from Shandong Provincial Bureau of Statistics. Available online: http://www.stats-sd.gov.cn/col/col6279/index.html (accessed on 12 October 2018).
- Statistical Yearbook. Available online: http://www.ha.stats.gov.cn/sitesources/hntj/page_pc/tjfw/tjcbw/tjnj/list1.html (accessed on 12 October 2018).
- Hunan Provincial Bureau of Statistics. Available online: http://tjj.hunan.gov.cn/tjsj/tjnj/ (accessed on 12 October 2018).
- The Data of Hubei Province. Available online: http://data.hb.stats.cn/CityData.aspx?DataType=67&ReportType=3 (accessed on 12 October 2018).
- Welcome to Guangdong Statistical Information Network. Available online: http://www.gdstats.gov.cn/tjsj/gdtjnj/ (accessed on 12 October 2018).
- Guangxi Statistical Information Network. Available online: http://www.gxtj.gov.cn/tjsj/tjnj/ (accessed on 12 October 2018).
- Annual Data—Statistical Data—Hainan Provincial Bureau of Statistics. Available online: http://stats.hainan.gov.cn/tjj/tjsu/ndsj/ (accessed on 12 October 2018).
- Chongqing Statistical Information Network. Available online: http://www.cqtj.gov.cn/tjsj/shuju/tjnj/ (accessed on 12 October 2018).
- Sichuan Statistics. Available online: http://tjj.sc.gov.cn/tjcbw/tjnj/ (accessed on 12 October 2018).
- Website of Guizhou Provincial Statistics Bureau. Available online: http://www.gz.stats.gov.cn/tjsj_35719/sjcx_35720/gztjnj_40112/tjnj2018/ (accessed on 12 October 2018).
- Statistical Yearbook—Yunnan Provincial Bureau of Statistics. Available online: http://www.stats.yn.gov.cn/tjsj/tjnj/ (accessed on 12 October 2018).
- Statistics Bureau of Shaanxi Province. Available online: http://www.shaanxitj.gov.cn/site/1/html/126/127/233/list.htm (accessed on 12 October 2018).
- Statistical Yearbook—Gansu Provincial Bureau of Statistics. Available online: http://www.gstj.gov.cn/HdApp/HdBas/HdClsContentMain.asp?ClassId=70 (accessed on 12 October 2018).
- Qinghai Statistical Yearbook. Available online: http://www.qhtjj.gov.cn/tjData/qhtjnj/ (accessed on 12 October 2018).
- Ningxia Data. Available online: http://nxdata.gov.cn/publish.htm?cn=G01 (accessed on 12 October 2018).
- Bureau of Statistics of Xinjiang Uygur Autonomous Region. Available online: http://www.xjtj.gov.cn/sjcx/tjnj_3415/ (accessed on 12 October 2018).
- Jie, G.; Yujie, L.; Zhiming, F.; Tao, P.; Yanzhao, Y.; Hao, Z. Research on monitoring and early warning of land and water resources carrying capacity in the Tibet Autonomous Region. Resour. Sci. 2018, 6, 1209–1221. [Google Scholar]
- Qi, L. The situation and the contradictory transformation of China’s energy security. Int. Petrol. Econ. 2018, 4, 18–26. [Google Scholar]
- Rui, L. Study on the Performance of Irrigation and Water Conservancy Infrastructure Investment. Master’s Thesis, Hunan Agricultural University, Hunan, China, 2015. [Google Scholar]
- Zonghui, Y.; Jinkai, L.; Chenxue, H.; Heguang, L. The evolution of China’s grain production base and the influencing factors. Res. Agric. Modern. 2019, 1, 36–43. [Google Scholar]
- Maoshan, C.; Nongdi, W.; Sihui, L. A deep understanding of the grim situation in which new and old problems are interwoven in China’s water security. Water Resour. Dev. Res. 2018, 9, 2–7. [Google Scholar]
Unit | Attribute of Index | ||
---|---|---|---|
Water Subsystem (f1) | Total water resources () | 108 m3 | positive |
Total water supply () | 108 m3 | positive | |
Amount of precipitation () | mm | positive | |
Average per capita water resources (f14) | m3 | positive | |
Agricultural water consumption () | 108 m3 | positive | |
Irrigation of farmland uses water per mu () | m3 | positive | |
Industrial water consumption () | 108 m3 | negative | |
Residential water consumption () | 108 m3 | negative | |
Energy Subsystem (f2) | Total energy production () | 104 tce | positive |
Total energy consumption () | 104 tce | positive | |
Total energy consumption in primary industry () | 104 tce | positive | |
Total energy consumption in water production and supply industries () | 104 tce | positive | |
Total industrial energy consumption () | 104 tce | negative | |
Food Subsystem (f3) | Food acreage () | 103 hm2 | positive |
Food total output () | 10 kiloton | positive | |
Food output per unit area () | Kg/hectare | positive | |
Per capita output of food () | Kg | positive | |
Total power of agricultural machinery () | GW | positive | |
Irrigable area of arable land () | 103 hm2 | positive | |
Amount of fertilizer applied to agriculture () | 104 ton | positive |
Coupling Stage | Coupling Range | |
---|---|---|
1 | Very Low | (0, 0.3] |
2 | Low | (0.3, 0.5] |
3 | High | (0.5, 0.8] |
4 | Very High | (0.8, 1] |
The Type of Coordination Level | Coordination Stage | Coordination Range |
---|---|---|
1 | Low | (0, 0.4] |
2 | middle | (0.4, 0.5] |
3 | High | (0.5, 0.8] |
4 | Extreme high | (0.8, 1) |
Water Subsystem () | Total water capital ( | 0.159 | 0.188 |
Total water supply ( | 0.117 | 0.215 | |
Amount of precipitation ( | 0.067 | 0.147 | |
Average per capita water availability ( | 0.498 | 0.153 | |
Agricultural water consumption ( | 0.122 | 0.155 | |
Irrigation of farmland uses water per mu ( | 0.021 | 0.061 | |
Industrial water consumption ( | 0.007 | 0.037 | |
Residential water consumption ( | 0.009 | 0.044 | |
Energy Subsystem () | Total energy production ( | 0.362 | 0.252 |
Total energy consumption ( | 0.147 | 0.222 | |
Total energy consumption in primary industry ( | 0.184 | 0.257 | |
Total energy consumption in water production and supply industries ( | 0.269 | 0.144 | |
Total industrial energy consumption ( | 0.038 | 0.124 | |
Food Subsystem () | Food acreage ( | 0.167 | 0.178 |
Food total output ( | 0.181 | 0.187 | |
Food output per unit area ( | 0.010 | 0.025 | |
Per capita output of food ( | 0.117 | 0.104 | |
The total power of agricultural machinery ( | 0.191 | 0.158 | |
Irrigable area of arable land ( | 0.174 | 0.202 | |
Amount of fertilizer applied to agriculture ( | 0.160 | 0.146 |
Year | 2007 | 2008 | 2009 | 2010 | 2011 | 2012 | 2013 | 2014 | 2015 | 2016 | |
---|---|---|---|---|---|---|---|---|---|---|---|
Regions | |||||||||||
Beijing | 0.602 | 0.619 | 0.563 | 0.490 | 0.534 | 0.518 | 0.423 | 0.354 | 0.353 | 0.331 | |
Tianjin | 0.877 | 0.876 | 0.881 | 0.756 | 0.773 | 0.765 | 0.703 | 0.688 | 0.688 | 0.711 | |
Hebei | 0.582 | 0.585 | 0.587 | 0.550 | 0.547 | 0.598 | 0.555 | 0.549 | 0.549 | 0.603 | |
Shanxi | 0.506 | 0.505 | 0.509 | 0.413 | 0.373 | 0.374 | 0.420 | 0.421 | 0.413 | 0.415 | |
Inner Mongolia | 0.793 | 0.762 | 0.835 | 0.723 | 0.686 | 0.684 | 0.744 | 0.697 | 0.716 | 0.702 | |
Liaoning | 0.860 | 0.849 | 0.857 | 0.873 | 0.830 | 0.868 | 0.856 | 0.810 | 0.800 | 0.783 | |
Jilin | 0.508 | 0.558 | 0.507 | 0.545 | 0.440 | 0.483 | 0.462 | 0.387 | 0.407 | 0.399 | |
Heilongjiang | 0.639 | 0.558 | 0.595 | 0.556 | 0.532 | 0.600 | 0.619 | 0.563 | 0.533 | 0.572 | |
Shanghai | 0.599 | 0.624 | 0.607 | 0.570 | 0.525 | 0.556 | 0.510 | 0.528 | 0.548 | 0.518 | |
Jiangsu | 0.822 | 0.820 | 0.833 | 0.836 | 0.854 | 0.833 | 0.847 | 0.858 | 0.864 | 0.896 | |
Zhejiang | 0.999 | 0.999 | 0.998 | 0.991 | 0.986 | 0.989 | 0.979 | 0.986 | 0.981 | 0.969 | |
Anhui | 0.667 | 0.608 | 0.633 | 0.650 | 0.622 | 0.622 | 0.578 | 0.565 | 0.555 | 0.630 | |
Fujian | 0.953 | 0.959 | 0.982 | 0.899 | 0.990 | 0.940 | 0.977 | 0.954 | 0.943 | 0.916 | |
Jiangxi | 0.859 | 0.868 | 0.857 | 0.849 | 0.826 | 0.828 | 0.871 | 0.870 | 0.875 | 0.867 | |
Shandong | 0.501 | 0.497 | 0.495 | 0.503 | 0.522 | 0.519 | 0.521 | 0.498 | 0.500 | 0.531 | |
Henan | 0.442 | 0.441 | 0.449 | 0.455 | 0.415 | 0.404 | 0.404 | 0.392 | 0.406 | 0.416 | |
Hubei | 0.890 | 0.882 | 0.860 | 0.875 | 0.829 | 0.836 | 0.828 | 0.833 | 0.830 | 0.856 | |
Hunan | 0.925 | 0.928 | 0.914 | 0.950 | 0.900 | 0.949 | 0.924 | 0.912 | 0.913 | 0.924 | |
Guangdong | 0.883 | 0.909 | 0.916 | 0.763 | 0.870 | 0.883 | 0.790 | 0.735 | 0.743 | 0.725 | |
Guangxi | 0.877 | 0.827 | 0.867 | 0.879 | 0.860 | 0.882 | 0.913 | 0.925 | 0.906 | 0.949 | |
Hainan | 0.840 | 0.812 | 0.797 | 0.814 | 0.820 | 0.884 | 0.812 | 0.859 | 0.899 | 0.823 | |
Chongqing | 0.979 | 0.947 | 0.948 | 0.928 | 0.931 | 0.927 | 0.934 | 0.966 | 0.955 | 0.960 | |
Sichuan | 0.867 | 0.887 | 0.858 | 0.872 | 0.863 | 0.888 | 0.856 | 0.870 | 0.854 | 0.868 | |
Guizhou | 0.988 | 0.983 | 0.965 | 0.967 | 0.942 | 0.953 | 0.948 | 0.978 | 0.969 | 0.969 | |
Yunnan | 0.949 | 0.940 | 0.922 | 0.915 | 0.905 | 0.902 | 0.907 | 0.891 | 0.913 | 0.918 | |
Tibet | 0.199 | 0.194 | 0.235 | 0.238 | 0.274 | 0.309 | 0.304 | 0.309 | 0.357 | 0.295 | |
Shaanxi | 0.857 | 0.824 | 0.822 | 0.790 | 0.795 | 0.733 | 0.709 | 0.704 | 0.683 | 0.685 | |
Gansu | 0.938 | 0.900 | 0.909 | 0.896 | 0.894 | 0.887 | 0.885 | 0.849 | 0.841 | 0.850 | |
Qinghai | 0.764 | 0.780 | 0.685 | 0.749 | 0.742 | 0.714 | 0.791 | 0.757 | 0.819 | 0.820 | |
Ningxia | 0.923 | 0.918 | 0.882 | 0.852 | 0.821 | 0.800 | 0.790 | 0.790 | 0.753 | 0.752 | |
Xinjiang | 0.957 | 0.973 | 0.982 | 0.985 | 0.997 | 0.998 | 0.998 | 0.995 | 0.993 | 0.993 |
Year | 2007 | 2008 | 2009 | 2010 | 2011 | 2012 | 2013 | 2014 | 2015 | 2016 | |
---|---|---|---|---|---|---|---|---|---|---|---|
Regions | |||||||||||
Beijing | 0.182 | 0.192 | 0.183 | 0.172 | 0.184 | 0.181 | 0.162 | 0.146 | 0.145 | 0.145 | |
Tianjin | 0.22 | 0.224 | 0.227 | 0.210 | 0.219 | 0.221 | 0.208 | 0.206 | 0.209 | 0.214 | |
Hebei | 0.447 | 0.452 | 0.456 | 0.440 | 0.444 | 0.474 | 0.451 | 0.450 | 0.452 | 0.460 | |
Shanxi | 0.317 | 0.318 | 0.323 | 0.310 | 0.308 | 0.311 | 0.327 | 0.322 | 0.323 | 0.312 | |
Inner Mongolia | 0.449 | 0.445 | 0.504 | 0.477 | 0.480 | 0.490 | 0.510 | 0.496 | 0.508 | 0.496 | |
Liaoning | 0.397 | 0.395 | 0.397 | 0.409 | 0.413 | 0.427 | 0.422 | 0.392 | 0.397 | 0.389 | |
Jilin | 0.313 | 0.345 | 0.317 | 0.347 | 0.316 | 0.342 | 0.337 | 0.307 | 0.323 | 0.321 | |
Heilongjiang | 0.453 | 0.435 | 0.466 | 0.468 | 0.472 | 0.520 | 0.545 | 0.519 | 0.507 | 0.530 | |
Shanghai | 0.210 | 0.220 | 0.218 | 0.210 | 0.198 | 0.209 | 0.197 | 0.202 | 0.210 | 0.205 | |
Jiangsu | 0.516 | 0.517 | 0.526 | 0.527 | 0.540 | 0.532 | 0.538 | 0.550 | 0.557 | 0.575 | |
Zhejiang | 0.394 | 0.397 | 0.399 | 0.412 | 0.397 | 0.415 | 0.399 | 0.405 | 0.411 | 0.407 | |
Anhui | 0.446 | 0.426 | 0.442 | 0.453 | 0.444 | 0.449 | 0.439 | 0.437 | 0.436 | 0.473 | |
Fujian | 0.364 | 0.370 | 0.372 | 0.373 | 0.376 | 0.386 | 0.388 | 0.378 | 0.381 | 0.399 | |
Jiangxi | 0.421 | 0.442 | 0.436 | 0.450 | 0.430 | 0.453 | 0.436 | 0.445 | 0.454 | 0.452 | |
Shandong | 0.448 | 0.447 | 0.451 | 0.460 | 0.474 | 0.476 | 0.476 | 0.467 | 0.472 | 0.474 | |
Henan | 0.439 | 0.446 | 0.455 | 0.464 | 0.443 | 0.440 | 0.442 | 0.438 | 0.452 | 0.451 | |
Hubei | 0.468 | 0.477 | 0.474 | 0.487 | 0.472 | 0.481 | 0.486 | 0.494 | 0.497 | 0.502 | |
Hunan | 0.517 | 0.527 | 0.529 | 0.556 | 0.537 | 0.571 | 0.561 | 0.553 | 0.555 | 0.562 | |
Guangdong | 0.488 | 0.517 | 0.507 | 0.505 | 0.501 | 0.517 | 0.508 | 0.489 | 0.497 | 0.504 | |
Guangxi | 0.424 | 0.424 | 0.428 | 0.438 | 0.429 | 0.456 | 0.470 | 0.475 | 0.477 | 0.486 | |
Hainan | 0.256 | 0.267 | 0.270 | 0.273 | 0.278 | 0.279 | 0.280 | 0.280 | 0.275 | 0.285 | |
Chongqing | 0.357 | 0.350 | 0.353 | 0.352 | 0.358 | 0.361 | 0.350 | 0.362 | 0.360 | 0.364 | |
Sichuan | 0.495 | 0.510 | 0.501 | 0.512 | 0.512 | 0.531 | 0.517 | 0.526 | 0.522 | 0.530 | |
Guizhou | 0.378 | 0.387 | 0.381 | 0.387 | 0.372 | 0.388 | 0.385 | 0.410 | 0.412 | 0.407 | |
Yunnan | 0.441 | 0.444 | 0.431 | 0.436 | 0.439 | 0.447 | 0.456 | 0.454 | 0.467 | 0.474 | |
Tibet | 0.234 | 0.236 | 0.252 | 0.261 | 0.276 | 0.289 | 0.292 | 0.293 | 0.299 | 0.288 | |
Shaanxi | 0.367 | 0.368 | 0.379 | 0.384 | 0.396 | 0.386 | 0.383 | 0.385 | 0.381 | 0.375 | |
Gansu | 0.351 | 0.349 | 0.356 | 0.358 | 0.364 | 0.370 | 0.373 | 0.364 | 0.363 | 0.353 | |
Qinghai | 0.224 | 0.228 | 0.225 | 0.231 | 0.232 | 0.235 | 0.234 | 0.234 | 0.235 | 0.235 | |
Ningxia | 0.296 | 0.298 | 0.294 | 0.294 | 0.294 | 0.294 | 0.294 | 0.292 | 0.282 | 0.281 | |
Xinjiang | 0.478 | 0.493 | 0.506 | 0.520 | 0.533 | 0.553 | 0.571 | 0.577 | 0.583 | 0.585 |
© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Share and Cite
Xu, S.; He, W.; Shen, J.; Degefu, D.M.; Yuan, L.; Kong, Y. Coupling and Coordination Degrees of the Core Water–Energy–Food Nexus in China. Int. J. Environ. Res. Public Health 2019, 16, 1648. https://doi.org/10.3390/ijerph16091648
Xu S, He W, Shen J, Degefu DM, Yuan L, Kong Y. Coupling and Coordination Degrees of the Core Water–Energy–Food Nexus in China. International Journal of Environmental Research and Public Health. 2019; 16(9):1648. https://doi.org/10.3390/ijerph16091648
Chicago/Turabian StyleXu, Shasha, Weijun He, Juqin Shen, Dagmawi Mulugeta Degefu, Liang Yuan, and Yang Kong. 2019. "Coupling and Coordination Degrees of the Core Water–Energy–Food Nexus in China" International Journal of Environmental Research and Public Health 16, no. 9: 1648. https://doi.org/10.3390/ijerph16091648