Summary: We analysed the responses of 11 ecosystem models to elevated atmospheric [CO₂] (eCO₂) at two temperate forest ecosystems (Duke and Oak Ridge National Laboratory (ORNL) Free‐Air CO₂ Enrichment (FACE) experiments) to test alternative representations of carbon (C)–nitrogen (N) cycle processes. We decomposed the model responses into component processes affecting the response to eCO₂ and confronted these with observations from the FACE experiments. Most of the models reproduced the observed initial enhancement of net primary production (NPP) at both sites, but none was able to simulate both the sustained 10‐yr enhancement at Duke and the declining response at ORNL: models generally showed signs of progressive N limitation as a result of lower than observed plant N uptake. Nonetheless, many models showed qualitative agreement with observed component processes. The results suggest that improved representation of above‐ground–below‐ground interactions and better constraints on plant stoichiometry are important for a predictive understanding of eCO₂ effects. Improved accuracy of soil organic matter inventories is pivotal to reduce uncertainty in the observed C–N budgets. The two FACE experiments are insufficient to fully constrain terrestrial responses to eCO₂, given the complexity of factors leading to the observed diverging trends, and the consequential inability of the models to explain these trends. Nevertheless, the ecosystem models were able to capture important features of the experiments, lending some support to their projections.

Ecosystem responses to rising CO₂ concentrations are a major source of uncertainty in climate change projections. Data from ecosystem-scale Free-Air CO₂ Enrichment (FACE) experiments provide a unique opportunity to reduce this uncertainty. The recent FACE Model–Data Synthesis project aimed to use the information gathered in two forest FACE experiments to assess and improve land ecosystem models. A new 'assumption-centred' model intercomparison approach was used, in which participating models were evaluated against experimental data based on the ways in which they represent key ecological processes. By identifying and evaluating the main assumptions causing differences among models, the assumption-centred approach produced a clear roadmap for reducing model uncertainty. Here, we explain this approach and summarize the resulting research agenda. We encourage the application of this approach in other model intercomparison projects to fundamentally improve predictive understanding of the Earth system.7 page(s

Summary: Elevated atmospheric CO₂ concentration (eCO₂) has the potential to increase vegetation carbon storage if increased net primary production causes increased long‐lived biomass. Model predictions of eCO₂ effects on vegetation carbon storage depend on how allocation and turnover processes are represented. We used data from two temperate forest free‐air CO₂ enrichment (FACE) experiments to evaluate representations of allocation and turnover in 11 ecosystem models. Observed eCO₂ effects on allocation were dynamic. Allocation schemes based on functional relationships among biomass fractions that vary with resource availability were best able to capture the general features of the observations. Allocation schemes based on constant fractions or resource limitations performed less well, with some models having unintended outcomes. Few models represent turnover processes mechanistically and there was wide variation in predictions of tissue lifespan. Consequently, models did not perform well at predicting eCO₂ effects on vegetation carbon storage. Our recommendations to reduce uncertainty include: use of allocation schemes constrained by biomass fractions; careful testing of allocation schemes; and synthesis of allocation and turnover data in terms of model parameters. Data from intensively studied ecosystem manipulation experiments are invaluable for constraining models and we recommend that such experiments should attempt to fully quantify carbon, water and nutrient budgets.

Models represent our primary method for integration of small‐scale, process‐level phenomena into a comprehensive description of forest‐stand or ecosystem function. They also represent a key method for testing hypotheses about the response of forest ecosystems to multiple changing environmental conditions. This paper describes the evaluation of 13 stand‐level models varying in their spatial, mechanistic, and temporal complexity for their ability to capture intra‐ and interannual components of the water and carbon cycle for an upland, oak‐dominated forest of eastern Tennessee. Comparisons between model simulations and observations were conducted for hourly, daily, and annual time steps. Data for the comparisons were obtained from a wide range of methods including: eddy covariance, sapflow, chamber‐based soil respiration, biometric estimates of stand‐level net primary production and growth, and soil water content by time or frequency domain reflectometry. Response surfaces of carbon and water flux as a function of environmental drivers, and a variety of goodness‐of‐fit statistics (bias, absolute bias, and model efficiency) were used to judge model performance.A single model did not consistently perform the best at all time steps or for all variables considered. Intermodel comparisons showed good agreement for water cycle fluxes, but considerable disagreement among models for predicted carbon fluxes. The mean of all model outputs, however, was nearly always the best fit to the observations. Not surprisingly, models missing key forest components or processes, such as roots or modeled soil water content, were unable to provide accurate predictions of ecosystem responses to short‐term drought phenomenon. Nevertheless, an inability to correctly capture short‐term physiological processes under drought was not necessarily an indicator of poor annual water and carbon budget simulations. This is possible because droughts in the subject ecosystem were of short duration and therefore had a small cumulative impact. Models using hourly time steps and detailed mechanistic processes, and having a realistic spatial representation of the forest ecosystem provided the best predictions of observed data. Predictive ability of all models deteriorated under drought conditions, suggesting that further work is needed to evaluate and improve ecosystem model performance under unusual conditions, such as drought, that are a common focus of environmental change discussions.