Given the scarcity of streamflow observations, existing studies commonly rely on evapotranspiration (ET) data as supplementary or alternative data for hydrological model calibration. However, the uncertainty in different ET products can introduce additional errors into runoff simulations, thereby impeding water resource management in data-scarce regions. In this study, six ET products from the Ili River Basin, Xinjiang, were evaluated using station-observed ET and water balance-based estimates at both the point and basin scales. The three-cornered hat (TCH) method was used to quantify ET uncertainty, and the Bayesian three-cornered hat (BTCH) method was applied for the first time in this basin to develop a monthly ET dataset with increased accuracy. Finally, we explored the applicability of integrating ET products to calibrate the VIC-glacier model for runoff simulations. Results showed that ETMonitor performs best at the point scale, while GLEAM ET aligns most closely with basin-scale water balance ET (ETWB), outperforming other products in all metrics. Uncertainty was highest in summer because the higher temperatures and rainfall increase ET variability, whereas the BTCH method significantly decreased uncertainty, with BTCH3 performing best (uncertainty = 5.83 mm/month). Among the various calibration strategies for VIC-glacier model simulations, integrating ET products yielded the most reliable streamflow simulations, increasing the validation-period NSE to 0.78 and R-2 to 0.86 compared with using single ET products. This research supports optimal ET dataset selection and highlights the value of ET data fusion for enhancing hydrological modelling, offering guidance for water resource management in regions with limited streamflow observations.