Testing simulations of intra- and inter-annual variation in the plant production response to elevated CO₂ against measurements from an 11-year FACE experiment on grazed pasture.

摘要

Ecosystem models play a crucial role in understanding and evaluating the combined impacts of rising atmospheric CO₂ concentration and changing climate on terrestrial ecosystems. However, we are not aware of any studies where the capacity of models to simulate intra- and inter-annual variation in responses to elevated CO₂ has been tested against long-term experimental data. Here we tested how well the ecosystem model APSIM/AgPasture was able to simulate the results from a free air carbon dioxide enrichment (FACE) experiment on grazed pasture. At this FACE site, during 11 years of CO₂ enrichment, a wide range in annual plant production response to CO₂ (-6 to +28%) was observed. As well as running the full model, which includes three plant CO₂ response functions (plant photosynthesis, nitrogen (N) demand and stomatal conductance), we also tested the influence of these three functions on model predictions. Model/data comparisons showed that: (i) overall the model over-predicted the mean annual plant production response to CO₂ (18.5% cf 13.1%) largely because years with small or negative responses to CO₂ were not well simulated; (ii) in general seasonal and inter-annual variation in plant production responses to elevated CO₂ were well represented by the model; (iii) the observed CO₂ enhancement in overall mean legume content was well simulated but year-to-year variation in legume content was poorly captured by the model; (iv) the best fit of the model to the data required all three CO₂ response functions to be invoked; (v) using actual legume content and reduced N fixation rate under elevated CO₂ in the model provided the best fit to the experimental data. We conclude that in temperate grasslands the N dynamics (particularly the legume content and N fixation activity) play a critical role in pasture production responses to elevated CO₂, and are processes for model improvement.