Obtaining reliable estimates of net ecosystem exchange (NEE) for particular regions is essential for the estimation of regional carbon balances. The Community Land Model (CLM4.5-BGC) was applied to the Rur catchment in western Germany to foster the understanding of spatial carbon flux patterns and the predictability of climate-ecosystem feedbacks. We evaluated the effect estimated values of ecological key parameters had on modelled carbon fluxes and LAI compared to global default values. For three of the four most widespread PFTs in the catchment (C3-grass, coniferous forest, deciduous forest), successfully estimated and validated parameter values were adopted from a previous study. Only for C3-crop, new parameter values were estimated herein. The Markov Chain Monte Carlo (MCMC) approach DREAM<sub>(zs)</sub> was used to constrain the CLM parameters with eddy covariance (EC) NEE data. We evaluated the model performance with estimated parameter values using (i) measured NEE from four EC crop sites located inside the catchment, and (ii) LAI data from the RapidEye satellite of about 18 days. The difference between the measured and simulated NEE sum for the evaluation period was reduced ~ 40 % on average, if estimated parameters instead of default parameters were used as input. For all PFTs, estimated parameter values had a strong effect on the predicted NEE sum. This would notably alter regional carbon balance estimates, since the catchment scale NEE sum was strongly positive with default parameter values and strongly negative with the estimated values. To obtain a more comprehensive picture of model uncertainty, additional CLM ensembles were setup where perturbed meteorological input data and uncertain initial states were applied in addition to the uncertain parameters. We found that C3-grass and C3-crop were particularly sensitive to the perturbed meteorological input data, which resulted in a strong increase of the standard deviation of the NEE sum (σ<sub>∑NEE</sub>) for the different ensemble members by a factor of ~ 14 and 28. Thus, model uncertainty for these PFTs is clearly underestimated if uncertain forcings and initial states are not taken into account. The increase of σ<sub>∑NEE</sub> for needleleaf evergreen temperate tree and broadleaf deciduous temperate tree was below factor 3.5. We conclude that that the uncertainty of regional CLM NEE predictions was considerably reduced via parameter estimation. However, in order to further enhance the reliability of regional scale NEE predictions, it is considered essential to treat different crop types separately. Thus, more EC tower stations and data are needed.