Considering the large bias of the simulated surface heat fluxes with Noah-MP model at the Semi-Arid Climate and Environment Observatory of Lanzhou University (SACOL) in August 2009, an analysis of the physical processes and a series of sensitivity tests are carried out to investigate the main reasons for the large bias and select appropriate parameterization schemes. Results show that the Chen97 parameterization scheme for the surface sensible heat exchange coefficient can improve the sensible heat flux simulation. The Jarvis stomatal resistance parameterization scheme can increase the transpiration of plants and thereby improves the latent heat flux simulation and produces a more reasonable distribution ratio of sensible and latent heat fluxes. The LP92 evaporation parameterization scheme for bare soil can significantly increase the evaporation compared with the original scheme and gives a better simulation of the latent heat flux. Finally, according to the results with different combinations of the above three parameterization schemes, it is found that using two or three new parameterization schemes simultaneously can further reduce the RMSE (Root Mean Square Error) of the simulated surface sensible and latent heat fluxes, but the simulated soil moisture and soil temperature are not improved accordingly.