Accurate quantification of land surface evapotranspiration (ET) is desperately crucial for agricultural irrigation strategy, drought monitoring, and water resource management in semiarid and arid regions. The two-source energy balance (TSEB) model is an appropriate method for estimating ET over the mixed surface of vegetation and soil since it considers both energy contributions from soil and vegetation. The key parameters of the TSEB model, additional heat transfer resistance (kB^-1) and soil resistance, are vital to the estimation of water vapor and heat fluxes over the heterogeneous surface. In this study, the kB^-1 parameterization and soil resistance formulation with different coefficients were integrated into the TSEB model to improve the original model performance over the heterogeneous surface in the midstream area of the Heihe River Basin (HRB). The effect of kB^-1 on the TSEB model performance was greater than the soil resistance coefficient by validating the model output using the Eddy Covariance (EC) measurement installed at vegetables, maize, residential, orchard, Gobi Desert, sandy desert, and desert steppe surface. The optimized TSEB model would significantly reduce the uncertainty of the original model and improve the consistency between the simulated and observed sensible heat flux (H) and latent heat flux (LE) over the heterogeneous surfaces; RMSE values of H and LE decreased to 53 and 78 W m^-2 from 67 and 104 W m^-2 for seven-sites-average in 2012. The daily ET of this study area ranged from 0.69 to 6.48 mm day^-1 during the growing period; it agreed well with EC flux observations for the vegetated area, especially irrigated cropland with MBE, RMSE, and MAPE values of 0.21 mm d^-1, 1.02 mm d^-1, and 14.62 % at maize site, 0.39 mm d^-1, 1.14 mm d^-1 and 20.25 % at vegetables site. This research enhances the utility of TSEB in flux estimation over heterogeneous landscapes and thus is of great help in formulating sustainable water resource management schemes.