Uncertainty Quantification of Groundwater Reactive Transport and Coastal Morphological Modeling

Different sources of uncertainties have been inevitably induced into the environmental modeling due to different reasons such as the variability in the future climate state, incomplete knowledge and complexity of the nature system, and randomness in the system properties. These uncertainties make the model predictions inherently uncertain, and uncertainty becomes an important obstacle in environmental modeling. This dissertation presents a general framework for purpose of uncertainty quantification and it provides quantitative measures for relative importance of different uncertain factors to model outputs. The framework includes two parts: uncertainty analysis which implements variance decomposition technique to decompose and quantify different types of input uncertainty sources (i.e., scenario, model and parametric uncertainties); global sensitivity analysis which develops a new set of variance-based global sensitivity indices for measuring importance of model parameters with considering multiple future climate scenarios and plausible models. To demonstrate the usage and compatibility of the uncertainty quantification framework with different types of models, it was applied into two distinct cases: a synthetic groundwater reactive transport case and a barrier island morphological case. In the groundwater case, a Bayesian network integrated groundwater reactive transport model was built and studied for a synthetic case. Different uncertainty sources are described as uncertain nodes in the Bayesian network. All the nodes are characterized by multiple states, representing their uncertainty, in the form of continuous or discrete probability distributions that are propagated to the model endpoint, which is the spatial distribution of contaminant concentrations. In the barrier island case, a new Barrier Island Profile (BIP) model which simulates the barrier island cross-section morphological evolution was developed and studied. For a series of barrier island cross-sections derived from the characteristics of Santa Rosa Island, Florida, BIP was used to evaluate their responses to random storm events and five potential accelerated rates of sea-level rise projected over a century. Monte Carlo simulation is used to decompose and quantify the predictive uncertainties for uncertainty analysis of both cases. In the global sensitivity analysis, besides quasi-Monte Carlo simulation, sparse grid collocation method was also implemented to estimate the global sensitivity index to save the computational cost in the groundwater case. The study of BIP model demonstrates that BIP is capable of simulating realistic patterns of barrier island profile evolution over the span of a century using relatively simple representations of time- and space-averaged processes with consideration of uncertainty of future climate impacts. The results of uncertainty quantification for both cases demonstrate different types of model input uncertainty sources and the relative importance of model parameters can be quantified using the developed uncertainty quantification framework. And the global sensitivity indices may vary substantially between different models and scenarios. Not considering the model and scenario uncertainties, may result biased identification of important model parameters. The framework will be very useful for environmental modelers to prioritize different uncertainties and optimize expanse of limited resources to more efficiently decrease predictive uncertainty. Although only two applications are demonstrated, this uncertainty quantification framework is mathematically general and it can be applied to a wider range of hydrologic and environmental problems.