Identifying determinants of urban water use using data mining approach

Abstract

This study develops a new approach to quantitatively identify the most important determinants of urban water use. The approach is based on a data mining model called genetic programming (GP), which automatically optimizes the structure of the function and parameters simultaneously. With historical urban water use as the target, the GP model identifies the most relevant factors for 47 cities in northern China. Compared with conventional regressive models, the GP model performs better than the double-log model. The Nash–Sutcliffe model efficiency coefficient (NSE) of the GP model is 0.87, while the NSE of the double-log model is 0.79. According to the results of the case study, urban water use is determined by both socio-economic and natural variables. Total population, service industry indicators, green land area, housing area, water price, and rainfall are the most significant determinants of urban water use. Among them, total population, service industry indicators, and green land area clearly have positive contributions to urban water use, whereas rainfall has a negative impact on urban water use. The impacts of housing area and water price are complex, which implies that these determinants may have different impacts on urban water use in different conditions. The new model and new insights developed in this study could be helpful for urban water management, especially for cities that experience water scarcity.

Keywords::

• genetic programming
• machine learning
• nonlinear regression
• water price

Additional information

Funding

This research was supported by the National Natural Science Foundation of China [grant numbers 91125018 and 51179085] the Ministry of Science and Technology of China [grant numbers 2011BAC09B07 and 2013BAB05B03].