Abstract: Abstract: Net primary productivity (NPP) can be one of the most important indicators to evaluate the ecosystem structure and function in grassland protection. Therefore, it is necessary to rapidly and accurately evaluate the NPP under the carbon cycle of the ecosystem. The land surface process model can be used to realize the time series simulation of NPP using the physical mechanism. However, the default parameters of the models cannot be suitable for the target area, leading to the simulation deviation. There is also limited cognition of the physiological and biochemical characteristics of vegetation. Therefore, it is a high demand to optimize the parameters, when applied to the target area. The CESM framework has recently released the land surface model CLM5.0, indicating the most promising land surface process model for the evaluation of grassland productivity. In this study, the spatiotemporal simulation of grassland productivity was carried out in the Hulunbuir grassland, Inner Mongolia, China. Firstly, the local sensitivity analysis was used to choose the ten sensitive parameters of the CLM5.0_BGC module. Taking the global daily NPP data products as the reference, the Differential Evolution Adaptive Metropolis (DREAM) was then utilized to optimize ten sensitive parameters after sensitivity analysis. The accuracy of optimization was verified using global daily NPP data and GLASS annual net primary productivity products. Finally, the parameter setting was established for the CLM5.0 model for the grassland productivity simulation. The results were as follows: 1) Local sensitivity analysis was conducted to determine 74 parameters of the CLM5.0_BGC module. The most sensitive parameters were obtained to simulate the grassland productivity, including respiration, followed by carbon transfer. 2) Similar to the leaf C/N ratio, the Gaussian distribution was found in the posterior probability distribution in the slope parameter of the relationship between the stomatal conductance and photosynthesis. The optimization was observed in the benign constraint. The effectiveness of the carbon/nitrogen correlation and respiration parameters were verified to simulate the optimized model. 3) Parameter optimization effectively improved the simulation performance of CLM5.0 on the productivity of the grassland. The parameter optimization of the model shared a great improvement for the targeted pixel, where the relative error decreased from 33.82% to 10.97%, a relative improvement of 67%, and root mean square error decreased by 0.12 g/m2. The optimization effect of the model in summer was also better than that in autumn. 4) The grassland productivity in the next year was evaluated on the simulation with the optimized parameter, where the relative error decreased from 8.74% to 2.51%, and the absolute difference in annual mean index decreased from 15.64 g/m2 to 13.63 g/m2. Spatially, the accuracy of the optimized model was improved in the different types of grassland. Among them, the relative errors of meadow grassland, lowland meadow, and typical grassland decreased from 43.43% to 37.81%, from 39.23% to 31.17%, and from 42.03% to 33.00%, respectively. The root mean square error was reduced from 0.58 g/m2 to 0.52 g/m2 in the meadow grassland, from 0.62 g/m2 to 0.57 g/m2 in the lowland meadow, and from 0.51 g/m2 to 0.44 g/m2 in the typical grassland. The finding can provide a strong reference to simulate the grassland productivity in the Hulunbuir region using CLM5.0. There can be offered a positive role in the reasonable assessment of the grassland ecosystem under the carbon cycle.