Due to the spatial resolution of evapotranspiration (ET) derived by the thermal infrared images from satellite data is coarse, higher spatial resolution ET retrieval relies on downscaling strategies (input/output downscaling strategy: based on downscaled input data/downscaling low resolution ET). This study used remote sensing surface flux equilibrium-non-parametric (RS-SFE-NP) ET model and enhanced spatiotemporal adaptive reflectance fusion model (ESTARFM) to globally evaluate two strategies at different latitudes, vegetation coverage, and humidity gradients. Overall, the performance of the input downscaling strategy is better than the output downscaling, especially in low latitude areas, humid areas, and low vegetation coverage areas, with RMSE is reduced by 68W/m², 25 W/m², and 11 W/m², respectively. Further analysis was conducted on the sources error of the ET obtained from the input downscaling strategy, and the results showed that neglecting the downscaling surface temperature (LST) and broadband reflectance (BBR) can result in significant errors in ET estimation. Specifically, neglecting the downscaling of BBR in high latitude and middle vegetation coverage fraction regions can result in an increase bias of approximately 60 W/m² and 200 W/m² in ET estimation, respectively. Neglecting the downscaling of LST in arid regions can lead to a relative error increase of nearly 50%. This study provides valuable insights into the selection of spatial downscaling strategies for obtaining global higher spatial resolution ET.