Abstract: The correlation coefficient (R), the average deviation between (ME), root mean square error (RMSE), and coefficient of Nash (NSE) were calculated between the three land surface model from the land data assimilation system of the China Meteorological Administration (CLDAS CLM, CLDAS Noah and CLDAS-Noah-MP) and global land surface assimilation system (GLDAS-Noah) and the flux tower standing observation data, and the accuracy evaluation in terms of different time scales and the different underlying surface were given. The results show that the diurnal variation and annual variation trends and peak time of the single peak can be simulated according to the simulation results of the four models. The peak of diurnal variation generally occurs at 14：00, the annual variation peak occurs in summer, and the numerical simulation effect is slightly worse in irrigated or freezing-thawing farmland and wetland in spring. On the scale of the hour, day, and month, the simulation of models driven by CLDAS is generally better than that of GLDAS. The mean R-value of simulation by models driven by CLDAS is higher than that of GLDAS-NOAH, which is 0.07, 0.08, and 0.02, respectively. The mean value of RMSE is lower than that of GLDAS -NOAH, and the errors are reduced by 6.6, 5.5, and 2.3W·m−2, respectively. The simulation of the model will change along with the time scale and the underlying surface properties. From the hour to the day to the month scale, the simulation goes through a process of first getting worse and then getting better. The simulation on the month scale is the best, in which Noah-MP performs well, with R value of 0.88, RMSE of 20.8 W·m−2, and NSE of 0.58. Four model simulation results under different performance have the same certain commonality that simulation in mixed forest and coniferous forest were overestimated and that in the rest underlying surface are underestimated. Although no models work well for all underlying surfaces, CLM performs best in stations covered by Gobi, mixed forest, and coniferous forest, Noah performs best in desert and cropland, and Noah-MP performs best in meadow, grassland, and wetland.

Key words: Latent heat flux, CLDAS2.0, Land surface model data, Accuracy evaluation