Investigating the spatial variability of water security risk and its driving mechanisms in China using machine learning. (15th August 2022)

• Record Type: Journal Article
• Authors: Xu, Zhicheng; Cheng, Lei; Liu, Pan; Hou, Qinyao; Cheng, Shujie; Qin, Shujing; Liu, Liu; Xia, Jun
• Is Part Of: Journal of cleaner production. Volume 362(2022)
• Publication Date: 2022-08-15
• Subjects: China -- Comprehensive water security risk -- Driving mechanism -- Spatial variability -- Machine learning -- Catchment unit; Factory and trade waste -- Management -- Periodicals; Manufactures -- Environmental aspects -- Periodicals; Déchets industriels -- Gestion -- Périodiques; Usines -- Aspect de l'environnement -- Périodiques; 628.5
• DOI: 10.1016/j.jclepro.2022.132303
• Languages: English
• ISSNs: 0959-6526

Abstract

Abstract: Comprehensively analyzing the present condition of water security and its driving factors at a countrywide scale is of prime importance for systematically and spatially managing water resource in China. In this study, a countrywide and comprehensive water security risk (CWSR) was proposed and calculated as the average of water shortage degree, water pollution level, and water-related hazard risk at catchment unit. Considering the limitations associated with traditional regression-based frameworks in coping with collinearity that might produce regionally different relationships with multiple driving factors, a machine learning method (boost regression tree, BRT) and three machine-learning interpretable techniques were used to determine the mechanisms underlying the spatial variability of CWSR. Results showed that catchments in northern China generally have a higher CWSR than catchments in southern China. The Hai River Basin and its surrounding regions were found to face the most serious water security pressure. Such spatial distributions of CWSR were well predicted by the BRT model with six explanatory variables. Mean annual precipitation ( MAP ) had the largest influence on CWSR (relative importance of 40.3±2.8%), followed by human influence ( Hi, 21.8±2.7%), average slope ( S, 15.2±2.7%), river density ( Rd, 9.3±1.7%), proportion of land covered by water ( Pw, 7.5±1.6%), and river network connectivity ( Rc, 5.9±1.1%). However, the relative order of importance of these six driving factors varied between regions, suggesting that the drivers of CWSR varied spatially. Three of the six factors had monotonic negative effects on CWSR including MAP, S, and Pw . CWSR significantly increased with Hi . The relationships of Rd and Rc with CWSR were relatively complex and non-monotonic. By jointly making use of machine learning and further interpretation methods, the spatial variability of water security risks in China and their underlying driving factors were determined, and a new framework for quantitatively understanding the complex mechanisms of regional water security was provided. The results of this study will be helpful in making recommendations for reducing CWSR in China. Highlights: A comprehensive water security risk (CWSR) was proposed and calculated at catchment scale in China. The complex spatial relationship between CWSR and its driving factors was studied using a machine learning method. Further interpretations of machine learning relationships indicated spatially variable mechanisms of CWSR.