Evaluating water resource carrying capacity using the deep learning method: a case study of Yunnan, Southwest China

Water resource carrying capacity (WRCC) is an important index for measuring the relations between water resource systems and socio-economic-environmental development. In view of the difficulty in describing the complex and nonlinear relationships between the WRCC and indicators using traditional methods, this study introduces deep learning theory and proposes a novel deep neural network named WRCC-Net for WRCC assessment. Unlike typical network structures, we constructed a hierarchical structure that can indicate the index system in WRCC evaluation. Furthermore, we utilized a residual learning technique to increase the network depth for fitting the complex relationship between the WRCC state and indicators. The proposed deep learning method was applied to solve the real-world WRCC problem by taking the Yunnan province (Southwest China) as the case area. The WRCC was assessed from the following five dimensions: the water resources, social, economic, ecological environment, and coordination subsystems. Performance evaluation shows the advantages of the proposed WRCC-Net over the typical deep feed-forward network and shallow methods. Therefore, the proposed method provides a new way of evaluating the WRCC state and has potential for WRCC research. Overall, the WRCC evaluation using the WRCC-Net shows that the state of the WRCC in Yunnan constantly decreased from 2008 to 2018. These central-eastern areas in the Yunnan province, such as Kunming, Qujing, and Yuxi, are under an unfavorable capacity state. Measures, such as improving water resources management and increasing water utilization efficiency, should be considered in water resource planning in Yunnan province for the sustainable development of water resources.