Abstract: With economic development and accelerating urbanization, water shortage situation is becoming more and more serious in China. Water resources assessment has been of concern for many researchers and policy makers. This paper details an application of the Driving forces-Pressure-State-Impact-Response(DPSIR)model, a holistic tool concerning with multiple systems such as water resources, economy, society and environment, to develop indicators of water poverty in China. This is different from traditional Water Poverty Index(WPI)theory and integrates various indicators by structuring the cause-effect relationships. Furthermore, we chose the partial least squares approach to structural equation modeling(PLS-SEM)to calculate data and test model. This PLS-SEM give the weights based on the model validation, which has advantages compared with the traditional methods such as the analytic hierarchy process, principal component analysis, and entropy value method. While using the DPSIR-PLS, all hypotheses of this study passed the test, indicating that there was causality among the components of water poverty assessment model, and the index system was suitable for evaluating water poverty in China. The results show that holistic scores of water poverty presented an increase trend, indicating the situation of water poverty in China was improving gradually. The transformation of nuclear density distribution curve was changing obviously from double peak to single peak. At the same time, the distribution curve gave certain "club convergence" characteristic, and nuclear density values corresponding to severe water poverty decreased. In addition, ISOQATA clustering method and Markov chains were used to analyze types of water poverty driving system and dynamic evolution of water poverty. Results show as follows:(1)Areas with multiple types to drive water poverty were mainly distributed in the southern and eastern coastal regions, and areas with double types to drive water poverty were mainly distributed in the northern inland and western arid or semi-arid regions.(2)The type of water poverty in China had strong stability. Markov chains analysis demonstrates that a region of micro water poverty or serious water poverty had the probability of 62.5% and 71.43% respectively to maintain the same type of water poverty; while a region of both heavy and moderate water poverty had the probability of 50%. (3)The adjacent areas usually had the similar water poverty types with the region,and the clustering phenomenon of severe water poverty, heavy water poverty, moderate water and micro water poverty always emerged. By introducing the DPSIR-PLS model, This paper developed an indicator system and measured water poverty in China, which provided a new approach to evaluate water poverty in the future. The results can well reflect the actual situation and are helpful for developing countermeasures to alleviate water poverty in China.

Key words: DPSIR-PLS model, water poverty assessment, structural equation, Kernel density estimation, Markov chains