Advance the utility of environmental models by improving how models are tested against data and how they are used to understand simulated processes, predictions and prediction uncertainty.
This includes ways of making models more transparent and refutable. Making a model transparent means that tests of model adequacy are clearly defined and conducted and the importance of different aspects of the model to predictions of interest are readily apparent. Thus, in more transparent models it is easier to determine what data and simulated processes dominate model development, predictions, and measures of prediction uncertainty. I consider sensitivity analysis to be a primary way of making models more transparent. Making a model refutable means that the model is designed such that data can be used to test specific aspects of model construction. In a more refutable model, it is more likely that model inadequacies can be readily identified. This has implications for identifying models that likely have greater predictive ability.
The ability to analyze models depends on model characteristics. More nonlinear models are more difficult to analyze. Models that are more nonlinear than the systems they intend to represent produce make it difficult to understand the actual system with its real nonlinearities and are more difficult to analyze than they need to be. An important aspect of improving models is identifying and, where possible, correcting these unrealistic nonlinearities.