Using measurements at various scales to explore uncertainty in model predictions of carbon budgets and cycling through time and space

Process-based models of atmosphere-biosphere interactions, along with empirical data mining techniques, are the primary tools used for scaling disparate observations through space and time. In the past few decades they have been developed in tandem with our understanding of ecological theory, resulting in models of various levels of complexity and detail. Model intercomparisons, however, show a large range in model performance, with no clear consensus as to whether model structural error (process mis-representation) or mis-parameterization is to blame. One potential reason for this lies in difficulties in using data sources at different scales to adequately test model performance. Another is the common reliance on uni-variate model ‘validations', where only one aspect of model performance is tested, most commonly at a single site.

We propose that a stronger reliance on advanced methodologies, spanning both within- and between-site model performance, in the present climate and under future climate change, will lead to more complete model evaluation and development. Model-data fusion, for example, is a powerful framework by which to combine models with various data streams (including observations at different spatial or temporal scales), and account for associated uncertainties. Model benchmarking tools, such as empirical data mining techniques, also provide a strong alternative model evaluation. To illustrate the potential benefits of such an approach, we assess the performance of 17 process-based models of atmosphere-biosphere interactions, and two data mining tools, across 11 long-term eddy covariance forest sites. The results highlight details of model performance often overlooked by conventional model-data comparisons, and quantify the degree of coupling of terrestrial carbon sequestration to climate anomalies at multiple sites and time scales.