Carbon exchange between the atmosphere and terrestrial ecosystem is a key component affecting climate changes. Because the in situ measurements are not dense enough to resolve CO2 exchange spatial variation on various scales, the variation has been mainly simulated by numerical ecosystem models. These models contain large uncertainties in estimating CO2 exchange owing to incorporating a number of empirical parameters on different scales. This study applied a global optimization algorithm and ensemble approach to a surface CO2 flux scheme to (1) identify sensitive photosynthetic and respirational parameters, and (2) optimize the sensitive parameters in the modeling sense and improve the model skills. The photosynthetic and respirational parameters of corn (C4 species) and soybean (C3 species) in NCAR land surface model (LSM) are calibrated against observations from AmeriFlux site at Bondville, IL during 1999 and 2000 growing seasons. Results showed that the most sensitive parameters are maximum carboxylation rate at 25°C and its temperature sensitivity parameter (Vcmax25 and avc), quantum efficiency at 25°C (Qe25), temperature sensitivity parameter for maintenance respiration (arm), and temperature sensitivity parameter for microbial respiration (amr). After adopting calibrated parameter values, simulated seasonal averaged CO2 fluxes were improved for both the C4 and the C3 crops (relative bias reduced from 0.09 to −0.02 for the C4 case and from 0.28 to −0.01 for the C3 case). An updated scheme incorporating new parameters and a revised flux-integration treatment is also proposed.