In the Yellow River basin, China, ecosystems suffer from the overexploitation and utilization of water resources, resulting in various environmental impacts. Consideration must be given to both human and ecosystem water requirements in water resources management. A water footprint (WF) is a tool for estimating industrial, agricultural, commercial and household water requirements and for examining the impact of consumption on water resources. The study attempts to establish an approach to analyse the dynamic processes and driving forces that result in certain WFs. Using input–output tables for provinces and municipalities, we calculate water use coefficients, the total WF and the net external WF of consumption in China’s Yellow River Basin. A biproportional algorithm is employed to revise the input–output tables for analysing the temporal dynamics of the WF. The factor analysis and linear regression were used to analyse the main influencing factors of WF. Results indicate that the coefficient for water use by primary industries is highest and that coefficients for provincial water use differ significantly. Second, household consumption and residuals from capital accumulation constituted approximately half of the total WF of the Yellow River basin in 2002 and also differed significantly among provinces. Third, the ratio of the net external WF to the total WF increased, and the ratio of final consumption to the total WF declined during the period examined. Fourth, output by secondary industries correlated most strongly with the WF, followed by area under irrigation, per capita meat consumption, water consumption per 10,000-yuan increase in added value and population.

Hydrological models and remote sensing evapotranspiration (ET) models usually are used to estimate regional ET. This study aims to integrate the advantages of both the models to simulate the daily ET processes. A compromise between these two methodologies is represented by improving the optimization of the hydrological model on the basis of a new probability optimal ET series, which is produced by a data assimilation scheme combining sparse remote estimates and continuous modeling of regional ETs. The distributed time-variant gain hydrological model (DTVGM) and a two-layer remote sensing ET model are chosen. First, the DTVGM is optimized by maximizing the Nash–Sutcliffe efficiency of daily streamflow in the Shahe River basin, and simulates the daily hydrological processes of 1999–2007. For improving the accuracy of continuous ET simulation, the DTVGM is further optimized by dual objective functions composed of the assimilated ETs and observed outlet discharge. The results show that the accuracy of the DTVGM-based daily ETs is improved after the dual optimization, and the mean absolute percentage error between the DTVGM-based ETs and the measured ETs in the study area is reduced by 5.84%. The integrated method is proved better, and improves the hydrology modeling accuracy.