- Biotic and abiotic controls on whole-ecosystem carbon balance in forest ecosystems
- Dynamics and age of nonstructural carbohydrate reserves in forest trees;
- Controls on spatial and temporal variability in vegetation phenology
- Model-data fusion and multiple constraints
Model-data fusion and multiple constraints
In model-data fusion, various data streams are combined and synthesized in a modeling framework, generally using some sort of Monte Carlo optimization algorithm that allows model parameters to be estimated conditional on the observational data (J090). The basic philosophy of model-data fusion is that we can use models to scale up and integrate across time and space, but that by merging models and observational data constraints we can improve upon model-only or data-only analyses (J055).
We use statistically rigorous model-data fusion approaches to constrain estimates of model parameters and states (J103, J090, J085, J064, J036), propagate uncertainties (B004, J085, J040, J038, J025, J024, J019), and evaluate competing process representations (J003, J111, J029, J021). We make extensive use of model-data fusion tools in each of the lab’s core areas of research (ecosystem carbon cycling, nonstructural carbohydrate reserve dynamics, and vegetation phenology).
The model-data fusion approach is being widely adopted by the global change community for a number of reasons. First, it enables us to test models and improve their performance. This is essential if we want to run models forward in a prognostic mode and try to predict the effects of global change on terrestrial ecosystems. Second, it enables us to obtain process-level understanding of the relationships between ecosystem functions and environmental drivers. This is especially valuable, since many processes cannot be observed or measured directly. Third, uncertainty analysis is a central feature in model-data fusion, and the confidence intervals on model states and predictions, provided directly by Monte Carlo methods, are desperately needed to inform management and policy decision-making, e.g. with respect to carbon budgets and carbon accounting.
Effective use of model-data fusion techniques can identify key areas where models can be improved, and help explore uncertainty in projections of future carbon cycling. (Figure taken from Keenan et al., 2012 (J103)