Abstract: Using stable isotope technique to determine the sources of plant water is essential to improve our understanding of ecohydrological process and ecological management in arid and semiarid regions. There are many methods for determining plant water sources using stable isotope technique, but few studies have tried to compare outcomes of different methods. Based on empirical data, we used direct comparison method, multivariate linear mixed model (IsoSource), Bayesian mixing models (MixSIR, MixSIAR) and water absorption depth model, analyzed the water sources of vegetation, and compared the merit and demerit of each method. The results showed that Bayesian mixing models (MixSIR, MixSIAR) performed better in water source apportionment than IsoSource. The Bayesian models have higher requirements for data. The smaller the standard deviation of isotopic composition of vegetation xylem water and potential water source, the higher the reliability of model operation results. Bayesian hybrid model (MixSIR) is the optimal solution in this study. When determining plant water source by stable isotopes of hydrogen and oxygen, the potential water source of vegetation can be determined by direct comparison method. Furthermore, the contribution rate and range for each potential water source of plants could be calculated by the multivariate linear mixed model (IsoSource), and Bayesian mixing models (MixSIR, MixSIAR). When needed, the optimal model of quantitative analysis of plant water sources should be selected by evaluating the performance of models. If plant mainly absorbs soil water at different soil depths, the average depth of soil water absorbed by plants can be calculated by combining water absorption depth model. This study provides theoretical basis for the selection of method for identifying plant water sources by isotope techniques in arid and semiarid regions.

Key words: grid unit, land use change, ecosystem service value, spatial pattern, Maduo County