Identification of urban drinking water supply patterns across 627 cities in China based on supervised and unsupervised statistical learning

Urbanization, one of the predominant trends of the 21st century, places great stress on urban water supply networks. This paper aimed to identify the most important variables driving urban water supply patterns in China, a region which has seen rapid urban growth in the last few decades. In addition, a principal component analysis-informed urban water sustainability index was developed in order to benchmark cities. The research involved applying statistical learning and other analytical methods to 12 years of urban water supply data for 627 cities across China. The findings were as follows: (1) PCA showed that approximately 46.8% of variability in the data could be explained by two principal components. Component 1 (37.26%) was more closely associated with variables related to water supply and sale, supply pipelines, and water supply finance. C2 (9.51%) was clearly related to urban water prices and average per capita water use. (2) Random forest and XGBoost algorithms were effective in classifying cities according to their region, with model testing accuracies of 87.69% and 88.32% respectively. (3) Chinese cities have consistently suffered water loss/leakage rates above 20% since 2001, and water prices are closely associated with leakage. (4) China's urban water sustainability has increased by just 3.56% between 2001 and 2013; Southwest China saw the highest growth rate in urban water supply sustainability. The implications of our research effort will be useful for decision makers in water-stressed urban areas around the world who are seeking novel insights in how to leverage statistical learning techniques to gain insights into urban drinking water supply patterns.