Tiantian Zhou

Accurate and continuous estimates of evapotranspiration (ET) are crucial for comprehending associated hydrological processes, especially in dryland watersheds. However, there remains a scarcity of comprehensive evaluations of multi‐source ET estimates' performance and uncertainties. To further assist ET estimation improvement and water resources management, we incorporated diffuse radiation partitioning and water‐carbon coupling processes into the satellite‐driven Vegetation Interface Processes (VIP) model for estimating ET in the Heihe River Basin, Northwest China, and compared it with eddy covariance (EC) observations, water budget estimates, and five global satellite‐based products from Breathing Earth System Simulator (BESS), MODerate‐resolution Imaging Spectroradiometer (MODIS), Penman‐Monteith‐Leuning (PML), Operational Simplified Surface Energy Balance (SSEBop), and Global LAnd Surface Satellite (GLASS). Sobol' sensitivity analysis was applied to the VIP model. Uncertainties in ET estimates were identified by the three‐cornered hat method. Results indicate that stomatal conductance coefficient, quantum efficiency, and clear sky index were dominant influencing parameters for VIP ET. Moreover, VIP and GLASS ET were in strong agreement with the EC‐based ET, although SSEBop ET was not, and MODIS ET could not capture seasonal variations. VIP could best represent ET dynamics from water budgets on the basin scale, followed by PML. During the period 2003–2015, GLASS, PML, and VIP provided reasonable ET estimates in this region, with an annual ET of 159.3, 153.2, and 173.5 mm, respectively, across the basin. Furthermore, ET algorithm benefits were presented: BESS and VIP, including diffuse radiation partitioning, performed better in cropland; PML and VIP that address water‐carbon coupling were stable and independent of the aridity index, and GLASS could considerably reduce uncertainty. The findings contribute to the collective knowledge of ET estimates in similar dryland areas and are relevant worldwide.