The parameterization of an ecosystem model can be greatly improved by the assimilation of multiple observatory datasets.But different observations can make the constraints of parameters and model projections vary wildly.In this study, we analyzed parameter constraints, correlations, and information contributions of data to model predictions at the assimilation of eight sets of observations with two weighting schemes of observations to explore how the data assimilation procedures compromise conflicting information from different observations.The eight sets of observations were collected at Duke Forest Free Atmospheric CO2 Enhancement (FACE) experimental site, consisting of seven carbon (C) pool datasets with low observation frequencies (foliage biomass, wood biomass, fine root biomass, microbial biomass, litter fall, litter, and soil carbon content), and a C flux dataset, soil respiration, with a high frequency.Two types of cost functions were used in data assimilation procedures, representing different weighting schemes.The cost function 1 (CF1) is the summation of the squared differences between simulations and observations, which highly weighted the soil respiration data.The cost function 2 (CF2) is the product of the sum of the squared differences between simulations and observations of each dataset, which treated all the datasets equally.The results showed that the parameters accepted with the CF1 were well constrained but weakly correlated, while those with the CF2 were loosely constrained but highly correlated.The C pool and soil respiration data contributed different information to the model predictions.Assimilation of the C pool data led to much lower predictions of ecosystem C than that of soil respiration data.The information contribution of the eight datasets decreased during the period of predicting at the two cost functions.The high weight of the soil respiration data led to rapid decreases in the information contributed by the eight datasets.These results indicate the importance of weighting schemes in multi-objective conditioning of ecosystem models.