Carbon-phosphorus cycle models overestimate carbon dioxide response in a mature Eucalyptus forest

The importance of phosphorus in regulating ecosystem responses to climate change has fostered phosphorus-cycle implementation in land surface models, but their carbon dioxide effects predictions have not been evaluated against measurements. In a new study in Science Advances, simulations of eight widely used phosphorus-enabled models were confronted with observations from a long-term free-air carbon dioxide enrichment experiment (FACE) in a mature, phosphorus-limited Eucalyptus forest. Large rings have been installed in the forest to artificially increase carbon dioxide concentrations within the rings by 150 part per million. By comparing the part of the ecosystem within the rings to the outer parts, the effect of elevated carbon dioxide can be assessed.

The analysis shows that most models predicted the correct sign and magnitude of the carbon dioxide effect on ecosystem carbon sequestration, but they generally overestimated the effects on plant carbon uptake and growth. Leaf-to-canopy scaling of photosynthesis, plant tissue stoichiometry, plant below-ground carbon allocation, and the subsequent consequences for plant-microbial interaction were identified as key areas in which models of ecosystem carbon-phosphorus interaction can be improved. Together, this data-model inter-comparison reveals data-driven insights into the performance and functionality of P-enabled models and adds to the existing evidence that the global carbon dioxide-driven carbon sink is overestimated by models.

A novelty of the study are two land surface models which deployed a new generation of microbial explicit soil organic matter modules. The comparison with predictions from pre-cursors of the two models which deployed classical soil organic modules allowed to assess the implications of the increased realism in resolved soil processes on model predictions. However, the land surface models with the advanced microbial explicit soil modules simulated a decrease in soil phosphorus release from organic matter under elevated carbon dioxide, in the opposite direction of their classical models and disagree with the observations. 'This is not surprising' says Daniel Goll, the lead developer of one of the two models, ORCHIDEE-CNP. 'Resolving new processes in models comes with an increase in model complexity which poses a challenging for model calibration as these models are already highly parametric. Ecosystem manipulation experiments like this one are critical to test & identify deficiency in our models.'