Abstract: Water use efficiency (WUE) of vegetation reflects the amount of its dry matter through consuming per unit amount of water, a comprehensive indicator for assessing its growth conditions. However, contributions of multiple forcing factors to WUE are unclear due to the complicated influencing mechanism. Combining a structural equation model (SEM) with the artificial neural network (ANN), this paper develops a hybrid SEM-ANN model for analysis of the direct and indirect influences of WUE multiple factors to achieve an improvement on the simulations. It determines the structural relationship among the factors and their degrees of influence by using SEM, and then constructs the topology of ANN. The results show that in the Hotan region, various vegetation types have different WUE factors at different levels. We divide them into direct factors and intermediate variables that impact WUE indirectly－with the former including temperature (T), precipitation (P), vapor pressure deficit (VPD), and wind speed (WS); the latter including an enhanced vegetation index (EVI) for grassland and cropland and a standardized precipitation evapotranspiration index (SPEI) for shrub land and evergreen needle leaved forest. The SEM-optimized structure of ANN fits better, and the SEM-ANN model has high explanatory capacity and higher accuracy in the ecosystem’s environmental control and simulations of WUE, thus providing a theoretical basis and simulation method that can improve efficient water use and predict future WUE responses to climate changes in Xinjiang.

Key words: vegetation ecosystem, water use efficiency, structural equation model, artificial neural network, Xinjiang