Accelerating the connection between experiments and models: The FACE-MDS experience
Abstract:The mandate is clear for improving communication between models and experiments to better evaluate terrestrial responses to atmospheric and climatic change. Unfortunately, progress in linking experimental and modeling approaches has been slow and sometimes frustrating. Recent successes in linking results from the Duke and Oak Ridge free-air CO2 enrichment (FACE) experiments with ecosystem and land surface models – the FACE Model-Data Synthesis (FACE-MDS) project – came only after a period of slow progress, but the experience points the way to future model-experiment interactions.
As the FACE experiments were approaching their termination, the FACE research community made an explicit attempt to work together with the modeling community to synthesize and deliver experimental data to benchmark models and to use models to supply appropriate context for the experimental results. Initial problems that impeded progress were: measurement protocols were not consistent across different experiments; data were not well organized for model input; and parameterizing and spinning up models that were not designed for simulating a specific site was difficult. Once these problems were worked out, the FACE-MDS project has been very successful in using data from the Duke and ORNL FACE experiment to test critical assumptions in the models. The project showed, for example, that the stomatal conductance model most widely used in models was supported by experimental data, but models did not capture important responses such as increased leaf mass per unit area in elevated CO2, and did not appropriately represent foliar nitrogen allocation.
We now have an opportunity to learn from this experience. New FACE experiments that have recently been initiated, or are about to be initiated, include a eucalyptus forest in Australia; the AmazonFACE experiment in a primary, tropical forest in Brazil; and a mature oak woodland in England. Cross-site science questions are being developed that will have a strong modeling framework, and modelers and experimentalists will work to establish common measurement protocols and data format. By starting the model-experiment connection early and learning from our past experiences, we expect to significantly shorten the time lags between advances in process-oriented studies and large-scale models.