Abstract: The southern Horqin sandy land，about 70 km away from Tongliao City，Inner Mongolia，China，is a typical sandy land in North China with limited water resources and fragile ecosystems. Quantification of evaporation from water surfaces over this region is critical to developing an understanding of eco-hydrological processes and therefore could be helpful to ecological restoration and water resources management over the arid region. In this study，processes and factors of water surface evaporation in the sand-oasis ecotone of the semiarid region were analyzed using data collected at the Agula Eco-hydrological Experimental Station. Then the stepwise multivariate regression analysis was applied to identify meteorological factors that primarily determine water surface evaporation. Furthermore，multivariate nonlinear stepwise regression forecasting models between water surface evaporation and the selected factors were built. Both the Dalton equation and Penman equation were used in this study to compute evaporation from water surfaces，with their wind functions being modified based on the observed wind velocity at 2 m height above the water surface. Finally the multivariate nonlinear stepwise regression models were compared with evaporation estimated from the Penman equation and Dalton equation. By using the multivariate stepwise regression analysis method，factors with varying significance could be ordered，with insignificant factors being removed from the subsequent analysis and factor correlation being reduced. Wind speed，saturated vapor pressure deficit and net radiation were identified as three most important factors determining water surface evaporation in the study area. The fitting of the optimal equation was satisfactory，indicating a coefficient of determination of 0.77 between the original equation and the optimized equation. All these forecasting models were generally consistent with the observations. Results of daily evaporation from the Dalton equation，the Penman equation and the optimal equation were almost the same，with a root mean square error of 1.98 mm，2.21 mm and 1.87 mm between the modeled evaporation and observed evaporation during the calibration period，and 2.63 mm，3.22 mm and 2.93 mm during the validation period for these three equations，respectively. Both the Dalton equation and the Penman equation whose parameters of the wind function were optimized to accommodate characteristics of the study region could perform well in simulating water surface evaporation.

Key words: multivariate nonlinear stepwise regression analysis, Penman model, variation of Dalton model, water surface evaporation, significant factor