Abstract: Abstract: Evapotranspiration (ET) is a key link in the hydrological and energy cycle in agricultural irrigation and water resource management. An accurate and quantitative estimation of regional ET is crucial to clarify the land-air interaction and global climate change. MOD16 model has been one of the most commonly-used remote sensing models to obtain the regional ET using the Penman-Monteith (P-M) equation. However, the MOD16 model cannot directly consider the soil moisture using the Relative Humidity (RH), Vapor Pressure Deficit (VPD), and Leaf Area Index (LAI), leading to some uncertainties in the ET estimation. In this study, an attempt was made to improve the MOD16 model using the normalized water index (NDWI) as an additional item of soil moisture. As such, the modified surface impedance of the MOD16 model (the improved model named MOD16-sm) was achieved to verify and apply in the arid oasis located in northwest China. Model validation included the comparative (between ET estimates and measurements) and error analyses. The comparative analyses between ET estimates and measurements showed that the scatter distribution of ET observations and estimates by the MOD16-sm model was very close to the 1:1 straight line. The accuracy of ET estimates acquired by the MOD16-sm model increased, with a higher Coefficient of Determination (R2) of 0.77, a lower Root Mean Square Error (RMSE) of 0.8 mm/d, and a lower Mean Absolute Deviation (MAE) of 0.46 mm/d, while the accuracy of ET estimates obtained by the MOD16 model was low, with a lower R2 of 0.64, a higher RMSE of 0.93 mm/d, and a higher MAE of 0.7 mm/d, indicating the convincible MOD16-sm model. Error analyses showed that the percentage error ranges of ET estimates by the two models were significantly different, and the maximum percentage error of ET estimates by the MOD16 model was higher than that by the MOD16-sm model, indicating the better performance of the MOD16-sm model on the ET estimates. The comparative analyses between ET estimates and measurements indicated that the MOD16-sm model corrected some overestimations of the MOD16 model, particularly with the better performance for the influence of soil moisture on ET estimations. The model application was conducted to analyze the spatial distribution of ET estimates, and the frequency distribution histograms on ET estimates. Correspondingly, the spatial distribution of ET estimates was closely related to the land use types, with the high vegetation cover regions (such as farmland, and woodland) in the high ET values, while the sparse vegetation cover areas (such as villages) showing the low ET values, which was in line with the objective facts. The ET frequency distribution demonstrated that the MOD16-sm model can be expected to serve as the heterogeneity of ET flux in the different vegetation cover areas. Therefore, it is feasible and reasonable for the improved MOD16 model by the Normalized Difference Water Index (NDWI) as an additional item of soil moisture. The finding can also provide a strong reference for the higher accuracy of regional ET estimations.