Abstract: The prediction of agricultural water consumption is of great significance for regional water resource planning, management, and utilization. Multivariate statistical analysis methods were used to research water use structure and trends of agricultural water consumption in Tianjin in this paper. Results indicated that from 2001 to 2020, agricultural water consumption in Tianjin showed a slow decline, but still higher than that of other industries. In 2021—2022, the ecological water consumption（accounting for about 35%）exceeded the agricultural water consumption（accounting for about 30%）. Fourteen influencing factors related to agricultural water consumption were screened out, and based on correlation analysis, three highly correlated influencing factors of agricultural water consumption in Tianjin were determined: domestic water consumption, ecological water consumption, and disposable income of rural residents. Based on factor analysis, three main driving factors affecting agricultural water consumption in Tianjin were identified: social and irrigation area, water resources, and industrial water consumption. Based on multiple linear and multiple nonlinear regression, three prediction models for agricultural water consumption in Tianjin were constructed, namely MLR model（R²=0.722）, MNLR model（R²=0.729）and FA-MNLR model（R²=0.849）; These three models have mean relative error（MRE）≤11.861% and root mean square error（RMSE）≤1.787, all of which can effectively predict agricultural water consumption in Tianjin. Among them, FA-MNLR model has the highest fit, integrating 14 influencing factors and having logical relationships with agricultural water consumption. Factor analysis and multiple nonlinear regression are combined, which provides a new modeling method. Results of this paper can provide reference for agricultural water consumption prediction, water resource allocation and management.

Key words: agricultural water, analytical model, factor analysis, correlation analysis, nonlinear