| Summary: | Thesis (M.Sc.)--Memorial University of Newfoundland, 2009. Physical Oceanography Includes bibliographical references (leaves 78-81) Earth Systems models that attempt to forecast equilibrium states or make long term predictions are sensitive to the unavoidable approximations they employ. It is therefore important for such models to be parameterized through objective and repeatable methods that quantify the uncertainties associated with the inexactness of these approximations. In this study Ensemble Kalman Filters and Neural Network Bayesian Models are used to investigate parameter sets for the Budyko Energy Balance Model and the more computationally demanding Planet Simulator of the University of Hamburg Meteorological Institute. These calibration methods employ observational data to generate posterior probability distributions for model parameter sets, allowing the determination of high-probability parameter sets and their confidence intervals. Being fully Bayesian, such approaches accurately propagate uncertainties in observational data into the posterior distributions. Comparing calibrated model results permits the two approaches to be assessed under varying levels of model complexity.
|
|---|