Two of the most pressing questions in global change research are to what extent terrestrial ecosystems will continue to take up CO2 and how much their growth is stimulated by elevated CO2. The Free Air CO2 Experiments (FACE) were a multi-million dollar, decadal time-scale experiment aimed at addressing these questions experimentally. While the FACE experiments have proven invaluable, the ability to scale up these results over space and time requires the use of models, and it is not clear whether the current generation of ecosystem models are able to capture the details of the responses observed at the FACE sites. The FACE/model inter-comparison project is a multi-year effort aimed at assessing the performance of a dozen ecosystem models at the Duke Forest and Oak Ridge National Lab FACE sites and determining how models can be improved by taking a detailed look at the full suite data collected at these sites. The first round assessed model performance blind to data. The second round allowed calibration to the early years of data and then predicted both the later years and the results of nitrogen addition experiments.
The twelve models in the inter-comparison represent a diverse set of modeling approaches that captures the range of models in use at the regional and global scale. Models vary in their temporal resolution (subdaily, daily, monthly), the structural complexity of the canopy and the rooting zone, their representation of different biogeochemical pathways, and the inclusion of community/successional dynamics. All models showed marked improvement between the first and second round of modeling. Models without a dynamic C:N ratio tended to overpredict the CO2 fertilization response, while those with dynamic C:N showed the potential to generate a “runaway” progressive nitrogen limitation feedback. Consistent with observations, models generally predicted a greater growth response for the Duke Loblolly pine stand than the ORNL sweetgum stand and greater nutrient limitation and belowground allocation at ORNL. Models were generally better at capturing the overall magnitude of the CO2 response than the inter-annual variability at each site. Overall biogeochemical complexity appears to be more important to model performance than temporal resolution, structural complexity, or community feedbacks.