The accurate modeling of forest ecosystems is crucial to the understanding of the response and feedbacks of forest systems to global climate change and how climate and land-use change can impact carbon storage. With the vast quantities of carbon stored, forests are a key target in the understanding of the storage and flux of carbon in a changing climate. The Ecosys model is a process-based, 3-dimensional mathematical model that has been extensively used to predict ecosystem behavior under various environmental conditions and serves as a robust model to predict the future carbon dynamics.

An accurate Ecosys model prediction is dependent on a set of carefully calibrated model parameters. In addition, parametric uncertainty is inherent in the model, and should be taken into consideration in the analysis of model predictions. To reduce the impact of the uncertainty in input data and model parameters and enhance the accuracy of Ecosys predictions, we combine observational data from the Duke Forest, located in North Carolina, USA with a vegetation type of evergreen needleleaf forests, to calibrate the Ecosys parameters. The calibration is performed in a Bayesian framework, so that probabilistic distributions of the model parameters are obtained. We propose implementing a state-of-the-art Bayesian parameter estimation method, namely the implicit particle filters, along with parallelized Ecosys simulations. The uncertainty of Ecosys predictions for the Duke Forest can then be quantified by propagating the parameter uncertainty through the Ecosys model, which provides a confidence level about the model forecast.