Evapotranspiration (ET) plays a crucial role in water resource management in arid and semi-arid regions. However, accurate estimation of ET is subject to significant uncertainties due to discrepancies among various ET methods. This study utilized the generalized Three-Cornered Hat (TCH) method to quantify ET uncertainty from four datasets: ERA5L, GLEAM, GLDAS, and MEP, spanning the period from 1980 to 2018. In-situ observations and the water balance approach were employed for validation at both site and basin scales. Results indicated substantial discrepancies in the spatial patterns of uncertainty among the four ET products, with GLEAM demonstrating the best performance. The uncertainty of the ET methods was ranked as follows: GLEAM < GLDAS < MEP < ERA5L. Based on these results, we applied the Bayesian Three-Cornered Hat (BTCH) method to produce an integrated ET product, achieving the highest ET accuracy. The BTCH method reduced the root mean square error (RMSE) from 80 mm/year to 16 mm/year and increased the Kling-Gupta Efficiency (KGE) value from 0.33 to 0.75 at the basin scale, outperforming all individual ET products. Overall, TCH effectively identifies uncertainty across various ecosystems, and BTCH significantly enhances ET reliability. This study provides a new benchmark for ET estimation, which may serve as a reference for water resource management in arid and semi-arid regions.