Examination of model uncertainty and parameter interaction in a global carbon cycling model (GLOCO)

Simulation models play an important role in understanding the causes and consequences of climate change. In order to make full use of these models, it is necessary to establish the magnitude and sources of uncertainty associated with their predictions. This information can be used to achieve a better understanding of the simulated systems, to increase the reliability of model predictions, to guide field surveys and laboratory experiments, and to define realistic values that should be used in scientific, economic, and political discussions of future conditions. In this paper, a new tree-structured density estimation technique that extends the ability of Monte Carlo-based analyses to explore parameter interactions and uncertainty in complex environmental models was applied. The application of the technique is demonstrated using the GLOCO global carbon cycle model. The paper demonstrates that there are numerous distinct parameter combinations that can meet fairly stringent calibration criteria, and they are concentrated in relatively small subsets of the parameter space. These different subsets can be viewed as representing different ecological systems that achieve the same calibration or performance goals in fundamentally different ways. It is also shown that the simulated responses of these systems to future environmental change can lead to different conclusions regarding the interaction between factors affecting environmental processes, such as the growth of vegetation. Together, these results show how the tree-structured density estimation technique can be applied to gain a broader understanding of model performance and of ecosystem responses to change.